diff options
Diffstat (limited to 'media/libopus')
301 files changed, 70841 insertions, 0 deletions
diff --git a/media/libopus/COPYING b/media/libopus/COPYING new file mode 100644 index 0000000000..9c739c34a3 --- /dev/null +++ b/media/libopus/COPYING @@ -0,0 +1,44 @@ +Copyright 2001-2011 Xiph.Org, Skype Limited, Octasic, + Jean-Marc Valin, Timothy B. Terriberry, + CSIRO, Gregory Maxwell, Mark Borgerding, + Erik de Castro Lopo + +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: + +- Redistributions of source code must retain the above copyright +notice, this list of conditions and the following disclaimer. + +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. + +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. + +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER +OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, +EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, +PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR +PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF +LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING +NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS +SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +Opus is subject to the royalty-free patent licenses which are +specified at: + +Xiph.Org Foundation: +https://datatracker.ietf.org/ipr/1524/ + +Microsoft Corporation: +https://datatracker.ietf.org/ipr/1914/ + +Broadcom Corporation: +https://datatracker.ietf.org/ipr/1526/ diff --git a/media/libopus/Makefile.in b/media/libopus/Makefile.in new file mode 100644 index 0000000000..4fa77570f6 --- /dev/null +++ b/media/libopus/Makefile.in @@ -0,0 +1,21 @@ +# This Source Code Form is subject to the terms of the Mozilla Public +# License, v. 2.0. If a copy of the MPL was not distributed with this +# file, You can obtain one at http://mozilla.org/MPL/2.0/. + +include $(topsrcdir)/config/rules.mk + +ifdef GNU_AS +ifeq ($(CPU_ARCH),arm) + +celt_pitch_xcorr_arm-gnu.s: celt/arm/armopts-gnu.S + +# armopts needs a specific rule, because arm2gnu.pl will always add the .S +# suffix when translating the files that include it. +celt/arm/armopts-gnu.S: $(srcdir)/celt/arm/armopts.s $(call mkdir_deps,celt/arm) $(GLOBAL_DEPS) + $(PERL) $(srcdir)/celt/arm/arm2gnu.pl < $< > $@ +# For all others, we can use an implicit rule +%-gnu.s: $(srcdir)/celt/arm/%.s $(GLOBAL_DEPS) + $(PERL) $(srcdir)/celt/arm/arm2gnu.pl < $< > $@ + +endif +endif diff --git a/media/libopus/README_MOZILLA b/media/libopus/README_MOZILLA new file mode 100644 index 0000000000..90a1c81f2d --- /dev/null +++ b/media/libopus/README_MOZILLA @@ -0,0 +1,11 @@ +IETF Opus audio codec reference implementation. + +The source in this directory was copied from an opus +repository checkout by running the ./update.sh script. +Any changes made to this version of the source should +be reflected in that script, e.g. by applying patch +files after the copy step. + +The upstream repository is https://gitlab.xiph.org/xiph/opus + +Refer to moz.yaml for the current in-tree revision. diff --git a/media/libopus/celt/_kiss_fft_guts.h b/media/libopus/celt/_kiss_fft_guts.h new file mode 100644 index 0000000000..17392b3e90 --- /dev/null +++ b/media/libopus/celt/_kiss_fft_guts.h @@ -0,0 +1,182 @@ +/*Copyright (c) 2003-2004, Mark Borgerding + + All rights reserved. + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions are met: + + * Redistributions of source code must retain the above copyright notice, + this list of conditions and the following disclaimer. + * Redistributions in binary form must reproduce the above copyright notice, + this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" + AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE + LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR + CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF + SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS + INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN + CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) + ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE + POSSIBILITY OF SUCH DAMAGE.*/ + +#ifndef KISS_FFT_GUTS_H +#define KISS_FFT_GUTS_H + +#define MIN(a,b) ((a)<(b) ? (a):(b)) +#define MAX(a,b) ((a)>(b) ? (a):(b)) + +/* kiss_fft.h + defines kiss_fft_scalar as either short or a float type + and defines + typedef struct { kiss_fft_scalar r; kiss_fft_scalar i; }kiss_fft_cpx; */ +#include "kiss_fft.h" + +/* + Explanation of macros dealing with complex math: + + C_MUL(m,a,b) : m = a*b + C_FIXDIV( c , div ) : if a fixed point impl., c /= div. noop otherwise + C_SUB( res, a,b) : res = a - b + C_SUBFROM( res , a) : res -= a + C_ADDTO( res , a) : res += a + * */ +#ifdef FIXED_POINT +#include "arch.h" + + +#define SAMP_MAX 2147483647 +#define TWID_MAX 32767 +#define TRIG_UPSCALE 1 + +#define SAMP_MIN -SAMP_MAX + + +# define S_MUL(a,b) MULT16_32_Q15(b, a) + +# define C_MUL(m,a,b) \ + do{ (m).r = SUB32_ovflw(S_MUL((a).r,(b).r) , S_MUL((a).i,(b).i)); \ + (m).i = ADD32_ovflw(S_MUL((a).r,(b).i) , S_MUL((a).i,(b).r)); }while(0) + +# define C_MULC(m,a,b) \ + do{ (m).r = ADD32_ovflw(S_MUL((a).r,(b).r) , S_MUL((a).i,(b).i)); \ + (m).i = SUB32_ovflw(S_MUL((a).i,(b).r) , S_MUL((a).r,(b).i)); }while(0) + +# define C_MULBYSCALAR( c, s ) \ + do{ (c).r = S_MUL( (c).r , s ) ;\ + (c).i = S_MUL( (c).i , s ) ; }while(0) + +# define DIVSCALAR(x,k) \ + (x) = S_MUL( x, (TWID_MAX-((k)>>1))/(k)+1 ) + +# define C_FIXDIV(c,div) \ + do { DIVSCALAR( (c).r , div); \ + DIVSCALAR( (c).i , div); }while (0) + +#define C_ADD( res, a,b)\ + do {(res).r=ADD32_ovflw((a).r,(b).r); (res).i=ADD32_ovflw((a).i,(b).i); \ + }while(0) +#define C_SUB( res, a,b)\ + do {(res).r=SUB32_ovflw((a).r,(b).r); (res).i=SUB32_ovflw((a).i,(b).i); \ + }while(0) +#define C_ADDTO( res , a)\ + do {(res).r = ADD32_ovflw((res).r, (a).r); (res).i = ADD32_ovflw((res).i,(a).i);\ + }while(0) + +#define C_SUBFROM( res , a)\ + do {(res).r = ADD32_ovflw((res).r,(a).r); (res).i = SUB32_ovflw((res).i,(a).i); \ + }while(0) + +#if defined(OPUS_ARM_INLINE_ASM) +#include "arm/kiss_fft_armv4.h" +#endif + +#if defined(OPUS_ARM_INLINE_EDSP) +#include "arm/kiss_fft_armv5e.h" +#endif +#if defined(MIPSr1_ASM) +#include "mips/kiss_fft_mipsr1.h" +#endif + +#else /* not FIXED_POINT*/ + +# define S_MUL(a,b) ( (a)*(b) ) +#define C_MUL(m,a,b) \ + do{ (m).r = (a).r*(b).r - (a).i*(b).i;\ + (m).i = (a).r*(b).i + (a).i*(b).r; }while(0) +#define C_MULC(m,a,b) \ + do{ (m).r = (a).r*(b).r + (a).i*(b).i;\ + (m).i = (a).i*(b).r - (a).r*(b).i; }while(0) + +#define C_MUL4(m,a,b) C_MUL(m,a,b) + +# define C_FIXDIV(c,div) /* NOOP */ +# define C_MULBYSCALAR( c, s ) \ + do{ (c).r *= (s);\ + (c).i *= (s); }while(0) +#endif + +#ifndef CHECK_OVERFLOW_OP +# define CHECK_OVERFLOW_OP(a,op,b) /* noop */ +#endif + +#ifndef C_ADD +#define C_ADD( res, a,b)\ + do { \ + CHECK_OVERFLOW_OP((a).r,+,(b).r)\ + CHECK_OVERFLOW_OP((a).i,+,(b).i)\ + (res).r=(a).r+(b).r; (res).i=(a).i+(b).i; \ + }while(0) +#define C_SUB( res, a,b)\ + do { \ + CHECK_OVERFLOW_OP((a).r,-,(b).r)\ + CHECK_OVERFLOW_OP((a).i,-,(b).i)\ + (res).r=(a).r-(b).r; (res).i=(a).i-(b).i; \ + }while(0) +#define C_ADDTO( res , a)\ + do { \ + CHECK_OVERFLOW_OP((res).r,+,(a).r)\ + CHECK_OVERFLOW_OP((res).i,+,(a).i)\ + (res).r += (a).r; (res).i += (a).i;\ + }while(0) + +#define C_SUBFROM( res , a)\ + do {\ + CHECK_OVERFLOW_OP((res).r,-,(a).r)\ + CHECK_OVERFLOW_OP((res).i,-,(a).i)\ + (res).r -= (a).r; (res).i -= (a).i; \ + }while(0) +#endif /* C_ADD defined */ + +#ifdef FIXED_POINT +/*# define KISS_FFT_COS(phase) TRIG_UPSCALE*floor(MIN(32767,MAX(-32767,.5+32768 * cos (phase)))) +# define KISS_FFT_SIN(phase) TRIG_UPSCALE*floor(MIN(32767,MAX(-32767,.5+32768 * sin (phase))))*/ +# define KISS_FFT_COS(phase) floor(.5+TWID_MAX*cos (phase)) +# define KISS_FFT_SIN(phase) floor(.5+TWID_MAX*sin (phase)) +# define HALF_OF(x) ((x)>>1) +#elif defined(USE_SIMD) +# define KISS_FFT_COS(phase) _mm_set1_ps( cos(phase) ) +# define KISS_FFT_SIN(phase) _mm_set1_ps( sin(phase) ) +# define HALF_OF(x) ((x)*_mm_set1_ps(.5f)) +#else +# define KISS_FFT_COS(phase) (kiss_fft_scalar) cos(phase) +# define KISS_FFT_SIN(phase) (kiss_fft_scalar) sin(phase) +# define HALF_OF(x) ((x)*.5f) +#endif + +#define kf_cexp(x,phase) \ + do{ \ + (x)->r = KISS_FFT_COS(phase);\ + (x)->i = KISS_FFT_SIN(phase);\ + }while(0) + +#define kf_cexp2(x,phase) \ + do{ \ + (x)->r = TRIG_UPSCALE*celt_cos_norm((phase));\ + (x)->i = TRIG_UPSCALE*celt_cos_norm((phase)-32768);\ +}while(0) + +#endif /* KISS_FFT_GUTS_H */ diff --git a/media/libopus/celt/arch.h b/media/libopus/celt/arch.h new file mode 100644 index 0000000000..3845c3a083 --- /dev/null +++ b/media/libopus/celt/arch.h @@ -0,0 +1,291 @@ +/* Copyright (c) 2003-2008 Jean-Marc Valin + Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/** + @file arch.h + @brief Various architecture definitions for CELT +*/ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef ARCH_H +#define ARCH_H + +#include "opus_types.h" +#include "opus_defines.h" + +# if !defined(__GNUC_PREREQ) +# if defined(__GNUC__)&&defined(__GNUC_MINOR__) +# define __GNUC_PREREQ(_maj,_min) \ + ((__GNUC__<<16)+__GNUC_MINOR__>=((_maj)<<16)+(_min)) +# else +# define __GNUC_PREREQ(_maj,_min) 0 +# endif +# endif + +#if OPUS_GNUC_PREREQ(3, 0) +#define opus_likely(x) (__builtin_expect(!!(x), 1)) +#define opus_unlikely(x) (__builtin_expect(!!(x), 0)) +#else +#define opus_likely(x) (!!(x)) +#define opus_unlikely(x) (!!(x)) +#endif + +#define CELT_SIG_SCALE 32768.f + +#define CELT_FATAL(str) celt_fatal(str, __FILE__, __LINE__); + +#if defined(ENABLE_ASSERTIONS) || defined(ENABLE_HARDENING) +#ifdef __GNUC__ +__attribute__((noreturn)) +#endif +void celt_fatal(const char *str, const char *file, int line); + +#if defined(CELT_C) && !defined(OVERRIDE_celt_fatal) +#include <stdio.h> +#include <stdlib.h> +#ifdef __GNUC__ +__attribute__((noreturn)) +#endif +void celt_fatal(const char *str, const char *file, int line) +{ + fprintf (stderr, "Fatal (internal) error in %s, line %d: %s\n", file, line, str); +#if defined(_MSC_VER) + _set_abort_behavior( 0, _WRITE_ABORT_MSG); +#endif + abort(); +} +#endif + +#define celt_assert(cond) {if (!(cond)) {CELT_FATAL("assertion failed: " #cond);}} +#define celt_assert2(cond, message) {if (!(cond)) {CELT_FATAL("assertion failed: " #cond "\n" message);}} +#define MUST_SUCCEED(call) celt_assert((call) == OPUS_OK) +#else +#define celt_assert(cond) +#define celt_assert2(cond, message) +#define MUST_SUCCEED(call) do {if((call) != OPUS_OK) {RESTORE_STACK; return OPUS_INTERNAL_ERROR;} } while (0) +#endif + +#if defined(ENABLE_ASSERTIONS) +#define celt_sig_assert(cond) {if (!(cond)) {CELT_FATAL("signal assertion failed: " #cond);}} +#else +#define celt_sig_assert(cond) +#endif + +#define IMUL32(a,b) ((a)*(b)) + +#define MIN16(a,b) ((a) < (b) ? (a) : (b)) /**< Minimum 16-bit value. */ +#define MAX16(a,b) ((a) > (b) ? (a) : (b)) /**< Maximum 16-bit value. */ +#define MIN32(a,b) ((a) < (b) ? (a) : (b)) /**< Minimum 32-bit value. */ +#define MAX32(a,b) ((a) > (b) ? (a) : (b)) /**< Maximum 32-bit value. */ +#define IMIN(a,b) ((a) < (b) ? (a) : (b)) /**< Minimum int value. */ +#define IMAX(a,b) ((a) > (b) ? (a) : (b)) /**< Maximum int value. */ +#define UADD32(a,b) ((a)+(b)) +#define USUB32(a,b) ((a)-(b)) + +/* Set this if opus_int64 is a native type of the CPU. */ +/* Assume that all LP64 architectures have fast 64-bit types; also x86_64 + (which can be ILP32 for x32) and Win64 (which is LLP64). */ +#if defined(__x86_64__) || defined(__LP64__) || defined(_WIN64) +#define OPUS_FAST_INT64 1 +#else +#define OPUS_FAST_INT64 0 +#endif + +#define PRINT_MIPS(file) + +#ifdef FIXED_POINT + +typedef opus_int16 opus_val16; +typedef opus_int32 opus_val32; +typedef opus_int64 opus_val64; + +typedef opus_val32 celt_sig; +typedef opus_val16 celt_norm; +typedef opus_val32 celt_ener; + +#define celt_isnan(x) 0 + +#define Q15ONE 32767 + +#define SIG_SHIFT 12 +/* Safe saturation value for 32-bit signals. Should be less than + 2^31*(1-0.85) to avoid blowing up on DC at deemphasis.*/ +#define SIG_SAT (300000000) + +#define NORM_SCALING 16384 + +#define DB_SHIFT 10 + +#define EPSILON 1 +#define VERY_SMALL 0 +#define VERY_LARGE16 ((opus_val16)32767) +#define Q15_ONE ((opus_val16)32767) + +#define SCALEIN(a) (a) +#define SCALEOUT(a) (a) + +#define ABS16(x) ((x) < 0 ? (-(x)) : (x)) +#define ABS32(x) ((x) < 0 ? (-(x)) : (x)) + +static OPUS_INLINE opus_int16 SAT16(opus_int32 x) { + return x > 32767 ? 32767 : x < -32768 ? -32768 : (opus_int16)x; +} + +#ifdef FIXED_DEBUG +#include "fixed_debug.h" +#else + +#include "fixed_generic.h" + +#ifdef OPUS_ARM_PRESUME_AARCH64_NEON_INTR +#include "arm/fixed_arm64.h" +#elif defined (OPUS_ARM_INLINE_EDSP) +#include "arm/fixed_armv5e.h" +#elif defined (OPUS_ARM_INLINE_ASM) +#include "arm/fixed_armv4.h" +#elif defined (BFIN_ASM) +#include "fixed_bfin.h" +#elif defined (TI_C5X_ASM) +#include "fixed_c5x.h" +#elif defined (TI_C6X_ASM) +#include "fixed_c6x.h" +#endif + +#endif + +#else /* FIXED_POINT */ + +typedef float opus_val16; +typedef float opus_val32; +typedef float opus_val64; + +typedef float celt_sig; +typedef float celt_norm; +typedef float celt_ener; + +#ifdef FLOAT_APPROX +/* This code should reliably detect NaN/inf even when -ffast-math is used. + Assumes IEEE 754 format. */ +static OPUS_INLINE int celt_isnan(float x) +{ + union {float f; opus_uint32 i;} in; + in.f = x; + return ((in.i>>23)&0xFF)==0xFF && (in.i&0x007FFFFF)!=0; +} +#else +#ifdef __FAST_MATH__ +#error Cannot build libopus with -ffast-math unless FLOAT_APPROX is defined. This could result in crashes on extreme (e.g. NaN) input +#endif +#define celt_isnan(x) ((x)!=(x)) +#endif + +#define Q15ONE 1.0f + +#define NORM_SCALING 1.f + +#define EPSILON 1e-15f +#define VERY_SMALL 1e-30f +#define VERY_LARGE16 1e15f +#define Q15_ONE ((opus_val16)1.f) + +/* This appears to be the same speed as C99's fabsf() but it's more portable. */ +#define ABS16(x) ((float)fabs(x)) +#define ABS32(x) ((float)fabs(x)) + +#define QCONST16(x,bits) (x) +#define QCONST32(x,bits) (x) + +#define NEG16(x) (-(x)) +#define NEG32(x) (-(x)) +#define NEG32_ovflw(x) (-(x)) +#define EXTRACT16(x) (x) +#define EXTEND32(x) (x) +#define SHR16(a,shift) (a) +#define SHL16(a,shift) (a) +#define SHR32(a,shift) (a) +#define SHL32(a,shift) (a) +#define PSHR32(a,shift) (a) +#define VSHR32(a,shift) (a) + +#define PSHR(a,shift) (a) +#define SHR(a,shift) (a) +#define SHL(a,shift) (a) +#define SATURATE(x,a) (x) +#define SATURATE16(x) (x) + +#define ROUND16(a,shift) (a) +#define SROUND16(a,shift) (a) +#define HALF16(x) (.5f*(x)) +#define HALF32(x) (.5f*(x)) + +#define ADD16(a,b) ((a)+(b)) +#define SUB16(a,b) ((a)-(b)) +#define ADD32(a,b) ((a)+(b)) +#define SUB32(a,b) ((a)-(b)) +#define ADD32_ovflw(a,b) ((a)+(b)) +#define SUB32_ovflw(a,b) ((a)-(b)) +#define MULT16_16_16(a,b) ((a)*(b)) +#define MULT16_16(a,b) ((opus_val32)(a)*(opus_val32)(b)) +#define MAC16_16(c,a,b) ((c)+(opus_val32)(a)*(opus_val32)(b)) + +#define MULT16_32_Q15(a,b) ((a)*(b)) +#define MULT16_32_Q16(a,b) ((a)*(b)) + +#define MULT32_32_Q31(a,b) ((a)*(b)) + +#define MAC16_32_Q15(c,a,b) ((c)+(a)*(b)) +#define MAC16_32_Q16(c,a,b) ((c)+(a)*(b)) + +#define MULT16_16_Q11_32(a,b) ((a)*(b)) +#define MULT16_16_Q11(a,b) ((a)*(b)) +#define MULT16_16_Q13(a,b) ((a)*(b)) +#define MULT16_16_Q14(a,b) ((a)*(b)) +#define MULT16_16_Q15(a,b) ((a)*(b)) +#define MULT16_16_P15(a,b) ((a)*(b)) +#define MULT16_16_P13(a,b) ((a)*(b)) +#define MULT16_16_P14(a,b) ((a)*(b)) +#define MULT16_32_P16(a,b) ((a)*(b)) + +#define DIV32_16(a,b) (((opus_val32)(a))/(opus_val16)(b)) +#define DIV32(a,b) (((opus_val32)(a))/(opus_val32)(b)) + +#define SCALEIN(a) ((a)*CELT_SIG_SCALE) +#define SCALEOUT(a) ((a)*(1/CELT_SIG_SCALE)) + +#define SIG2WORD16(x) (x) + +#endif /* !FIXED_POINT */ + +#ifndef GLOBAL_STACK_SIZE +#ifdef FIXED_POINT +#define GLOBAL_STACK_SIZE 120000 +#else +#define GLOBAL_STACK_SIZE 120000 +#endif +#endif + +#endif /* ARCH_H */ diff --git a/media/libopus/celt/arm/arm2gnu.pl b/media/libopus/celt/arm/arm2gnu.pl new file mode 100755 index 0000000000..a2895f7445 --- /dev/null +++ b/media/libopus/celt/arm/arm2gnu.pl @@ -0,0 +1,353 @@ +#!/usr/bin/perl +# Copyright (C) 2002-2013 Xiph.org Foundation +# +# Redistribution and use in source and binary forms, with or without +# modification, are permitted provided that the following conditions +# are met: +# +# - Redistributions of source code must retain the above copyright +# notice, this list of conditions and the following disclaimer. +# +# - Redistributions in binary form must reproduce the above copyright +# notice, this list of conditions and the following disclaimer in the +# documentation and/or other materials provided with the distribution. +# +# THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +# ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +# LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +# A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER +# OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, +# EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, +# PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR +# PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF +# LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING +# NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS +# SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + +my $bigend; # little/big endian +my $nxstack; +my $apple = 0; +my $symprefix = ""; + +$nxstack = 0; + +eval 'exec /usr/local/bin/perl -S $0 ${1+"$@"}' + if $running_under_some_shell; + +while ($ARGV[0] =~ /^-/) { + $_ = shift; + last if /^--$/; + if (/^-n$/) { + $nflag++; + next; + } + if (/^--apple$/) { + $apple = 1; + $symprefix = "_"; + next; + } + die "I don't recognize this switch: $_\\n"; +} +$printit++ unless $nflag; + +$\ = "\n"; # automatically add newline on print +$n=0; + +$thumb = 0; # ARM mode by default, not Thumb. +@proc_stack = (); + +printf (" .syntax unified\n"); + +LINE: +while (<>) { + + # For ADRLs we need to add a new line after the substituted one. + $addPadding = 0; + + # First, we do not dare to touch *anything* inside double quotes, do we? + # Second, if you want a dollar character in the string, + # insert two of them -- that's how ARM C and assembler treat strings. + s/^([A-Za-z_]\w*)[ \t]+DCB[ \t]*\"/$1: .ascii \"/ && do { s/\$\$/\$/g; next }; + s/\bDCB\b[ \t]*\"/.ascii \"/ && do { s/\$\$/\$/g; next }; + s/^(\S+)\s+RN\s+(\S+)/$1 .req r$2/ && do { s/\$\$/\$/g; next }; + # If there's nothing on a line but a comment, don't try to apply any further + # substitutions (this is a cheap hack to avoid mucking up the license header) + s/^([ \t]*);/$1@/ && do { s/\$\$/\$/g; next }; + # If substituted -- leave immediately ! + + s/@/,:/; + s/;/@/; + while ( /@.*'/ ) { + s/(@.*)'/$1/g; + } + s/\{FALSE\}/0/g; + s/\{TRUE\}/1/g; + s/\{(\w\w\w\w+)\}/$1/g; + s/\bINCLUDE[ \t]*([^ \t\n]+)/.include \"$1\"/; + s/\bGET[ \t]*([^ \t\n]+)/.include \"${ my $x=$1; $x =~ s|\.s|-gnu.S|; \$x }\"/; + s/\bIMPORT\b/.extern/; + s/\bEXPORT\b\s*/.global $symprefix/; + s/^(\s+)\[/$1IF/; + s/^(\s+)\|/$1ELSE/; + s/^(\s+)\]/$1ENDIF/; + s/IF *:DEF:/ .ifdef/; + s/IF *:LNOT: *:DEF:/ .ifndef/; + s/ELSE/ .else/; + s/ENDIF/ .endif/; + + if( /\bIF\b/ ) { + s/\bIF\b/ .if/; + s/=/==/; + } + if ( $n == 2) { + s/\$/\\/g; + } + if ($n == 1) { + s/\$//g; + s/label//g; + $n = 2; + } + if ( /MACRO/ ) { + s/MACRO *\n/.macro/; + $n=1; + } + if ( /\bMEND\b/ ) { + s/\bMEND\b/.endm/; + $n=0; + } + + # ".rdata" doesn't work in 'as' version 2.13.2, as it is ".rodata" there. + # + if ( /\bAREA\b/ ) { + my $align; + $align = "2"; + if ( /ALIGN=(\d+)/ ) { + $align = $1; + } + if ( /CODE/ ) { + $nxstack = 1; + } + s/^(.+)CODE(.+)READONLY(.*)/ .text/; + s/^(.+)DATA(.+)READONLY(.*)/ .section .rdata/; + s/^(.+)\|\|\.data\|\|(.+)/ .data/; + s/^(.+)\|\|\.bss\|\|(.+)/ .bss/; + s/$/; .p2align $align/; + # Enable NEON instructions but don't produce a binary that requires + # ARMv7. RVCT does not have equivalent directives, so we just do this + # for all CODE areas. + if ( /.text/ ) { + # Separating .arch, .fpu, etc., by semicolons does not work (gas + # thinks the semicolon is part of the arch name, even when there's + # whitespace separating them). Sadly this means our line numbers + # won't match the original source file (we could use the .line + # directive, which is documented to be obsolete, but then gdb will + # show the wrong line in the translated source file). + s/$/; .arch armv7-a\n .fpu neon\n .object_arch armv4t/ unless ($apple); + } + } + + s/\|\|\.constdata\$(\d+)\|\|/.L_CONST$1/; # ||.constdata$3|| + s/\|\|\.bss\$(\d+)\|\|/.L_BSS$1/; # ||.bss$2|| + s/\|\|\.data\$(\d+)\|\|/.L_DATA$1/; # ||.data$2|| + s/\|\|([a-zA-Z0-9_]+)\@([a-zA-Z0-9_]+)\|\|/@ $&/; + s/^(\s+)\%(\s)/ .space $1/; + + s/\|(.+)\.(\d+)\|/\.$1_$2/; # |L80.123| -> .L80_123 + s/\bCODE32\b/.code 32/ && do {$thumb = 0}; + s/\bCODE16\b/.code 16/ && do {$thumb = 1}; + if (/\bPROC\b/) + { + my $prefix; + my $proc; + /^([A-Za-z_\.]\w+)\b/; + $proc = $1; + $prefix = ""; + if ($proc) + { + $prefix = $prefix.sprintf("\t.type\t%s, %%function", $proc) unless ($apple); + # Make sure we $prefix isn't empty here (for the $apple case). + # We handle mangling the label here, make sure it doesn't match + # the label handling below (if $prefix would be empty). + $prefix = $prefix."; "; + push(@proc_stack, $proc); + s/^[A-Za-z_\.]\w+/$symprefix$&:/; + } + $prefix = $prefix."\t.thumb_func; " if ($thumb); + s/\bPROC\b/@ $&/; + $_ = $prefix.$_; + } + s/^(\s*)(S|Q|SH|U|UQ|UH)ASX\b/$1$2ADDSUBX/; + s/^(\s*)(S|Q|SH|U|UQ|UH)SAX\b/$1$2SUBADDX/; + if (/\bENDP\b/) + { + my $proc; + s/\bENDP\b/@ $&/; + $proc = pop(@proc_stack); + $_ = "\t.size $proc, .-$proc".$_ if ($proc && !$apple); + } + s/\bSUBT\b/@ $&/; + s/\bDATA\b/@ $&/; # DATA directive is deprecated -- Asm guide, p.7-25 + s/\bKEEP\b/@ $&/; + s/\bEXPORTAS\b/@ $&/; + s/\|\|(.)+\bEQU\b/@ $&/; + s/\|\|([\w\$]+)\|\|/$1/; + s/\bENTRY\b/@ $&/; + s/\bASSERT\b/@ $&/; + s/\bGBLL\b/@ $&/; + s/\bGBLA\b/@ $&/; + s/^\W+OPT\b/@ $&/; + s/:OR:/|/g; + s/:SHL:/<</g; + s/:SHR:/>>/g; + s/:AND:/&/g; + s/:LAND:/&&/g; + s/CPSR/cpsr/; + s/SPSR/spsr/; + s/ALIGN$/.balign 4/; + s/ALIGN\s+([0-9x]+)$/.balign $1/; + s/psr_cxsf/psr_all/; + s/LTORG/.ltorg/; + s/^([A-Za-z_]\w*)[ \t]+EQU/ .set $1,/; + s/^([A-Za-z_]\w*)[ \t]+SETL/ .set $1,/; + s/^([A-Za-z_]\w*)[ \t]+SETA/ .set $1,/; + s/^([A-Za-z_]\w*)[ \t]+\*/ .set $1,/; + + # {PC} + 0xdeadfeed --> . + 0xdeadfeed + s/\{PC\} \+/ \. +/; + + # Single hex constant on the line ! + # + # >>> NOTE <<< + # Double-precision floats in gcc are always mixed-endian, which means + # bytes in two words are little-endian, but words are big-endian. + # So, 0x0000deadfeed0000 would be stored as 0x0000dead at low address + # and 0xfeed0000 at high address. + # + s/\bDCFD\b[ \t]+0x([a-fA-F0-9]{8})([a-fA-F0-9]{8})/.long 0x$1, 0x$2/; + # Only decimal constants on the line, no hex ! + s/\bDCFD\b[ \t]+([0-9\.\-]+)/.double $1/; + + # Single hex constant on the line ! +# s/\bDCFS\b[ \t]+0x([a-f0-9]{8})([a-f0-9]{8})/.long 0x$1, 0x$2/; + # Only decimal constants on the line, no hex ! +# s/\bDCFS\b[ \t]+([0-9\.\-]+)/.double $1/; + s/\bDCFS[ \t]+0x/.word 0x/; + s/\bDCFS\b/.float/; + + s/^([A-Za-z_]\w*)[ \t]+DCD/$1 .word/; + s/\bDCD\b/.word/; + s/^([A-Za-z_]\w*)[ \t]+DCW/$1 .short/; + s/\bDCW\b/.short/; + s/^([A-Za-z_]\w*)[ \t]+DCB/$1 .byte/; + s/\bDCB\b/.byte/; + s/^([A-Za-z_]\w*)[ \t]+\%/.comm $1,/; + s/^[A-Za-z_\.]\w+/$&:/; + s/^(\d+)/$1:/; + s/\%(\d+)/$1b_or_f/; + s/\%[Bb](\d+)/$1b/; + s/\%[Ff](\d+)/$1f/; + s/\%[Ff][Tt](\d+)/$1f/; + s/&([\dA-Fa-f]+)/0x$1/; + if ( /\b2_[01]+\b/ ) { + s/\b2_([01]+)\b/conv$1&&&&/g; + while ( /[01][01][01][01]&&&&/ ) { + s/0000&&&&/&&&&0/g; + s/0001&&&&/&&&&1/g; + s/0010&&&&/&&&&2/g; + s/0011&&&&/&&&&3/g; + s/0100&&&&/&&&&4/g; + s/0101&&&&/&&&&5/g; + s/0110&&&&/&&&&6/g; + s/0111&&&&/&&&&7/g; + s/1000&&&&/&&&&8/g; + s/1001&&&&/&&&&9/g; + s/1010&&&&/&&&&A/g; + s/1011&&&&/&&&&B/g; + s/1100&&&&/&&&&C/g; + s/1101&&&&/&&&&D/g; + s/1110&&&&/&&&&E/g; + s/1111&&&&/&&&&F/g; + } + s/000&&&&/&&&&0/g; + s/001&&&&/&&&&1/g; + s/010&&&&/&&&&2/g; + s/011&&&&/&&&&3/g; + s/100&&&&/&&&&4/g; + s/101&&&&/&&&&5/g; + s/110&&&&/&&&&6/g; + s/111&&&&/&&&&7/g; + s/00&&&&/&&&&0/g; + s/01&&&&/&&&&1/g; + s/10&&&&/&&&&2/g; + s/11&&&&/&&&&3/g; + s/0&&&&/&&&&0/g; + s/1&&&&/&&&&1/g; + s/conv&&&&/0x/g; + } + + if ( /commandline/) + { + if( /-bigend/) + { + $bigend=1; + } + } + + if ( /\bDCDU\b/ ) + { + my $cmd=$_; + my $value; + my $prefix; + my $w1; + my $w2; + my $w3; + my $w4; + + s/\s+DCDU\b/@ $&/; + + $cmd =~ /\bDCDU\b\s+0x(\d+)/; + $value = $1; + $value =~ /(\w\w)(\w\w)(\w\w)(\w\w)/; + $w1 = $1; + $w2 = $2; + $w3 = $3; + $w4 = $4; + + if( $bigend ne "") + { + # big endian + $prefix = "\t.byte\t0x".$w1.";". + "\t.byte\t0x".$w2.";". + "\t.byte\t0x".$w3.";". + "\t.byte\t0x".$w4."; "; + } + else + { + # little endian + $prefix = "\t.byte\t0x".$w4.";". + "\t.byte\t0x".$w3.";". + "\t.byte\t0x".$w2.";". + "\t.byte\t0x".$w1."; "; + } + $_=$prefix.$_; + } + + if ( /\badrl\b/i ) + { + s/\badrl\s+(\w+)\s*,\s*(\w+)/ldr $1,=$2/i; + $addPadding = 1; + } + s/\bEND\b/@ END/; +} continue { + printf ("%s", $_) if $printit; + if ($addPadding != 0) + { + printf (" mov r0,r0\n"); + $addPadding = 0; + } +} +#If we had a code section, mark that this object doesn't need an executable +# stack. +if ($nxstack && !$apple) { + printf (" .section\t.note.GNU-stack,\"\",\%\%progbits\n"); +} diff --git a/media/libopus/celt/arm/arm_celt_map.c b/media/libopus/celt/arm/arm_celt_map.c new file mode 100644 index 0000000000..ca988b66f5 --- /dev/null +++ b/media/libopus/celt/arm/arm_celt_map.c @@ -0,0 +1,160 @@ +/* Copyright (c) 2010 Xiph.Org Foundation + * Copyright (c) 2013 Parrot */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "pitch.h" +#include "kiss_fft.h" +#include "mdct.h" + +#if defined(OPUS_HAVE_RTCD) + +# if defined(OPUS_ARM_MAY_HAVE_NEON_INTR) && !defined(OPUS_ARM_PRESUME_NEON_INTR) +opus_val32 (*const CELT_INNER_PROD_IMPL[OPUS_ARCHMASK+1])(const opus_val16 *x, const opus_val16 *y, int N) = { + celt_inner_prod_c, /* ARMv4 */ + celt_inner_prod_c, /* EDSP */ + celt_inner_prod_c, /* Media */ + celt_inner_prod_neon /* NEON */ +}; + +void (*const DUAL_INNER_PROD_IMPL[OPUS_ARCHMASK+1])(const opus_val16 *x, const opus_val16 *y01, const opus_val16 *y02, + int N, opus_val32 *xy1, opus_val32 *xy2) = { + dual_inner_prod_c, /* ARMv4 */ + dual_inner_prod_c, /* EDSP */ + dual_inner_prod_c, /* Media */ + dual_inner_prod_neon /* NEON */ +}; +# endif + +# if defined(FIXED_POINT) +# if ((defined(OPUS_ARM_MAY_HAVE_NEON) && !defined(OPUS_ARM_PRESUME_NEON)) || \ + (defined(OPUS_ARM_MAY_HAVE_MEDIA) && !defined(OPUS_ARM_PRESUME_MEDIA)) || \ + (defined(OPUS_ARM_MAY_HAVE_EDSP) && !defined(OPUS_ARM_PRESUME_EDSP))) +opus_val32 (*const CELT_PITCH_XCORR_IMPL[OPUS_ARCHMASK+1])(const opus_val16 *, + const opus_val16 *, opus_val32 *, int, int, int) = { + celt_pitch_xcorr_c, /* ARMv4 */ + MAY_HAVE_EDSP(celt_pitch_xcorr), /* EDSP */ + MAY_HAVE_MEDIA(celt_pitch_xcorr), /* Media */ + MAY_HAVE_NEON(celt_pitch_xcorr) /* NEON */ +}; + +# endif +# else /* !FIXED_POINT */ +# if defined(OPUS_ARM_MAY_HAVE_NEON_INTR) && !defined(OPUS_ARM_PRESUME_NEON_INTR) +void (*const CELT_PITCH_XCORR_IMPL[OPUS_ARCHMASK+1])(const opus_val16 *, + const opus_val16 *, opus_val32 *, int, int, int) = { + celt_pitch_xcorr_c, /* ARMv4 */ + celt_pitch_xcorr_c, /* EDSP */ + celt_pitch_xcorr_c, /* Media */ + celt_pitch_xcorr_float_neon /* Neon */ +}; +# endif +# endif /* FIXED_POINT */ + +#if defined(FIXED_POINT) && defined(OPUS_HAVE_RTCD) && \ + defined(OPUS_ARM_MAY_HAVE_NEON_INTR) && !defined(OPUS_ARM_PRESUME_NEON_INTR) + +void (*const XCORR_KERNEL_IMPL[OPUS_ARCHMASK + 1])( + const opus_val16 *x, + const opus_val16 *y, + opus_val32 sum[4], + int len +) = { + xcorr_kernel_c, /* ARMv4 */ + xcorr_kernel_c, /* EDSP */ + xcorr_kernel_c, /* Media */ + xcorr_kernel_neon_fixed, /* Neon */ +}; + +#endif + +# if defined(OPUS_ARM_MAY_HAVE_NEON_INTR) +# if defined(HAVE_ARM_NE10) +# if defined(CUSTOM_MODES) +int (*const OPUS_FFT_ALLOC_ARCH_IMPL[OPUS_ARCHMASK+1])(kiss_fft_state *st) = { + opus_fft_alloc_arch_c, /* ARMv4 */ + opus_fft_alloc_arch_c, /* EDSP */ + opus_fft_alloc_arch_c, /* Media */ + opus_fft_alloc_arm_neon /* Neon with NE10 library support */ +}; + +void (*const OPUS_FFT_FREE_ARCH_IMPL[OPUS_ARCHMASK+1])(kiss_fft_state *st) = { + opus_fft_free_arch_c, /* ARMv4 */ + opus_fft_free_arch_c, /* EDSP */ + opus_fft_free_arch_c, /* Media */ + opus_fft_free_arm_neon /* Neon with NE10 */ +}; +# endif /* CUSTOM_MODES */ + +void (*const OPUS_FFT[OPUS_ARCHMASK+1])(const kiss_fft_state *cfg, + const kiss_fft_cpx *fin, + kiss_fft_cpx *fout) = { + opus_fft_c, /* ARMv4 */ + opus_fft_c, /* EDSP */ + opus_fft_c, /* Media */ + opus_fft_neon /* Neon with NE10 */ +}; + +void (*const OPUS_IFFT[OPUS_ARCHMASK+1])(const kiss_fft_state *cfg, + const kiss_fft_cpx *fin, + kiss_fft_cpx *fout) = { + opus_ifft_c, /* ARMv4 */ + opus_ifft_c, /* EDSP */ + opus_ifft_c, /* Media */ + opus_ifft_neon /* Neon with NE10 */ +}; + +void (*const CLT_MDCT_FORWARD_IMPL[OPUS_ARCHMASK+1])(const mdct_lookup *l, + kiss_fft_scalar *in, + kiss_fft_scalar * OPUS_RESTRICT out, + const opus_val16 *window, + int overlap, int shift, + int stride, int arch) = { + clt_mdct_forward_c, /* ARMv4 */ + clt_mdct_forward_c, /* EDSP */ + clt_mdct_forward_c, /* Media */ + clt_mdct_forward_neon /* Neon with NE10 */ +}; + +void (*const CLT_MDCT_BACKWARD_IMPL[OPUS_ARCHMASK+1])(const mdct_lookup *l, + kiss_fft_scalar *in, + kiss_fft_scalar * OPUS_RESTRICT out, + const opus_val16 *window, + int overlap, int shift, + int stride, int arch) = { + clt_mdct_backward_c, /* ARMv4 */ + clt_mdct_backward_c, /* EDSP */ + clt_mdct_backward_c, /* Media */ + clt_mdct_backward_neon /* Neon with NE10 */ +}; + +# endif /* HAVE_ARM_NE10 */ +# endif /* OPUS_ARM_MAY_HAVE_NEON_INTR */ + +#endif /* OPUS_HAVE_RTCD */ diff --git a/media/libopus/celt/arm/armcpu.c b/media/libopus/celt/arm/armcpu.c new file mode 100644 index 0000000000..c7d16e6d61 --- /dev/null +++ b/media/libopus/celt/arm/armcpu.c @@ -0,0 +1,194 @@ +/* Copyright (c) 2010 Xiph.Org Foundation + * Copyright (c) 2013 Parrot */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +/* Original code from libtheora modified to suit to Opus */ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#ifdef OPUS_HAVE_RTCD + +#include "armcpu.h" +#include "cpu_support.h" +#include "os_support.h" +#include "opus_types.h" +#include "arch.h" + +#define OPUS_CPU_ARM_V4_FLAG (1<<OPUS_ARCH_ARM_V4) +#define OPUS_CPU_ARM_EDSP_FLAG (1<<OPUS_ARCH_ARM_EDSP) +#define OPUS_CPU_ARM_MEDIA_FLAG (1<<OPUS_ARCH_ARM_MEDIA) +#define OPUS_CPU_ARM_NEON_FLAG (1<<OPUS_ARCH_ARM_NEON) + +#if defined(_MSC_VER) +/*For GetExceptionCode() and EXCEPTION_ILLEGAL_INSTRUCTION.*/ +# define WIN32_LEAN_AND_MEAN +# define WIN32_EXTRA_LEAN +# include <windows.h> + +static OPUS_INLINE opus_uint32 opus_cpu_capabilities(void){ + opus_uint32 flags; + flags=0; + /* MSVC has no OPUS_INLINE __asm support for ARM, but it does let you __emit + * instructions via their assembled hex code. + * All of these instructions should be essentially nops. */ +# if defined(OPUS_ARM_MAY_HAVE_EDSP) || defined(OPUS_ARM_MAY_HAVE_MEDIA) \ + || defined(OPUS_ARM_MAY_HAVE_NEON) || defined(OPUS_ARM_MAY_HAVE_NEON_INTR) + __try{ + /*PLD [r13]*/ + __emit(0xF5DDF000); + flags|=OPUS_CPU_ARM_EDSP_FLAG; + } + __except(GetExceptionCode()==EXCEPTION_ILLEGAL_INSTRUCTION){ + /*Ignore exception.*/ + } +# if defined(OPUS_ARM_MAY_HAVE_MEDIA) \ + || defined(OPUS_ARM_MAY_HAVE_NEON) || defined(OPUS_ARM_MAY_HAVE_NEON_INTR) + __try{ + /*SHADD8 r3,r3,r3*/ + __emit(0xE6333F93); + flags|=OPUS_CPU_ARM_MEDIA_FLAG; + } + __except(GetExceptionCode()==EXCEPTION_ILLEGAL_INSTRUCTION){ + /*Ignore exception.*/ + } +# if defined(OPUS_ARM_MAY_HAVE_NEON) || defined(OPUS_ARM_MAY_HAVE_NEON_INTR) + __try{ + /*VORR q0,q0,q0*/ + __emit(0xF2200150); + flags|=OPUS_CPU_ARM_NEON_FLAG; + } + __except(GetExceptionCode()==EXCEPTION_ILLEGAL_INSTRUCTION){ + /*Ignore exception.*/ + } +# endif +# endif +# endif + return flags; +} + +#elif defined(__linux__) +/* Linux based */ +#include <stdio.h> + +opus_uint32 opus_cpu_capabilities(void) +{ + opus_uint32 flags = 0; + FILE *cpuinfo; + + /* Reading /proc/self/auxv would be easier, but that doesn't work reliably on + * Android */ + cpuinfo = fopen("/proc/cpuinfo", "r"); + + if(cpuinfo != NULL) + { + /* 512 should be enough for anybody (it's even enough for all the flags that + * x86 has accumulated... so far). */ + char buf[512]; + + while(fgets(buf, 512, cpuinfo) != NULL) + { +# if defined(OPUS_ARM_MAY_HAVE_EDSP) || defined(OPUS_ARM_MAY_HAVE_MEDIA) \ + || defined(OPUS_ARM_MAY_HAVE_NEON) || defined(OPUS_ARM_MAY_HAVE_NEON_INTR) + /* Search for edsp and neon flag */ + if(memcmp(buf, "Features", 8) == 0) + { + char *p; + p = strstr(buf, " edsp"); + if(p != NULL && (p[5] == ' ' || p[5] == '\n')) + flags |= OPUS_CPU_ARM_EDSP_FLAG; + +# if defined(OPUS_ARM_MAY_HAVE_NEON) || defined(OPUS_ARM_MAY_HAVE_NEON_INTR) + p = strstr(buf, " neon"); + if(p != NULL && (p[5] == ' ' || p[5] == '\n')) + flags |= OPUS_CPU_ARM_NEON_FLAG; +# endif + } +# endif + +# if defined(OPUS_ARM_MAY_HAVE_MEDIA) \ + || defined(OPUS_ARM_MAY_HAVE_NEON) || defined(OPUS_ARM_MAY_HAVE_NEON_INTR) + /* Search for media capabilities (>= ARMv6) */ + if(memcmp(buf, "CPU architecture:", 17) == 0) + { + int version; + version = atoi(buf+17); + + if(version >= 6) + flags |= OPUS_CPU_ARM_MEDIA_FLAG; + } +# endif + } + + fclose(cpuinfo); + } + return flags; +} +#else +/* The feature registers which can tell us what the processor supports are + * accessible in priveleged modes only, so we can't have a general user-space + * detection method like on x86.*/ +# error "Configured to use ARM asm but no CPU detection method available for " \ + "your platform. Reconfigure with --disable-rtcd (or send patches)." +#endif + +static int opus_select_arch_impl(void) +{ + opus_uint32 flags = opus_cpu_capabilities(); + int arch = 0; + + if(!(flags & OPUS_CPU_ARM_EDSP_FLAG)) { + /* Asserts ensure arch values are sequential */ + celt_assert(arch == OPUS_ARCH_ARM_V4); + return arch; + } + arch++; + + if(!(flags & OPUS_CPU_ARM_MEDIA_FLAG)) { + celt_assert(arch == OPUS_ARCH_ARM_EDSP); + return arch; + } + arch++; + + if(!(flags & OPUS_CPU_ARM_NEON_FLAG)) { + celt_assert(arch == OPUS_ARCH_ARM_MEDIA); + return arch; + } + arch++; + + celt_assert(arch == OPUS_ARCH_ARM_NEON); + return arch; +} + +int opus_select_arch(void) { + int arch = opus_select_arch_impl(); +#ifdef FUZZING + arch = rand()%(arch+1); +#endif + return arch; +} +#endif diff --git a/media/libopus/celt/arm/armcpu.h b/media/libopus/celt/arm/armcpu.h new file mode 100644 index 0000000000..820262ff5f --- /dev/null +++ b/media/libopus/celt/arm/armcpu.h @@ -0,0 +1,77 @@ +/* Copyright (c) 2010 Xiph.Org Foundation + * Copyright (c) 2013 Parrot */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#if !defined(ARMCPU_H) +# define ARMCPU_H + +# if defined(OPUS_ARM_MAY_HAVE_EDSP) +# define MAY_HAVE_EDSP(name) name ## _edsp +# else +# define MAY_HAVE_EDSP(name) name ## _c +# endif + +# if defined(OPUS_ARM_MAY_HAVE_MEDIA) +# define MAY_HAVE_MEDIA(name) name ## _media +# else +# define MAY_HAVE_MEDIA(name) MAY_HAVE_EDSP(name) +# endif + +# if defined(OPUS_ARM_MAY_HAVE_NEON) +# define MAY_HAVE_NEON(name) name ## _neon +# else +# define MAY_HAVE_NEON(name) MAY_HAVE_MEDIA(name) +# endif + +# if defined(OPUS_ARM_PRESUME_EDSP) +# define PRESUME_EDSP(name) name ## _edsp +# else +# define PRESUME_EDSP(name) name ## _c +# endif + +# if defined(OPUS_ARM_PRESUME_MEDIA) +# define PRESUME_MEDIA(name) name ## _media +# else +# define PRESUME_MEDIA(name) PRESUME_EDSP(name) +# endif + +# if defined(OPUS_ARM_PRESUME_NEON) +# define PRESUME_NEON(name) name ## _neon +# else +# define PRESUME_NEON(name) PRESUME_MEDIA(name) +# endif + +# if defined(OPUS_HAVE_RTCD) +int opus_select_arch(void); + +#define OPUS_ARCH_ARM_V4 (0) +#define OPUS_ARCH_ARM_EDSP (1) +#define OPUS_ARCH_ARM_MEDIA (2) +#define OPUS_ARCH_ARM_NEON (3) + +# endif + +#endif diff --git a/media/libopus/celt/arm/armopts.s b/media/libopus/celt/arm/armopts.s new file mode 100644 index 0000000000..7e270e83b1 --- /dev/null +++ b/media/libopus/celt/arm/armopts.s @@ -0,0 +1,37 @@ +/* Copyright (C) 2013 Mozilla Corporation */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +; Set the following to 1 if we have EDSP instructions +; (LDRD/STRD, etc., ARMv5E and later). +OPUS_ARM_MAY_HAVE_EDSP * 1 + +; Set the following to 1 if we have ARMv6 media instructions. +OPUS_ARM_MAY_HAVE_MEDIA * 1 + +; Set the following to 1 if we have NEON (some ARMv7) +OPUS_ARM_MAY_HAVE_NEON * 1 + +END diff --git a/media/libopus/celt/arm/armopts.s.in b/media/libopus/celt/arm/armopts.s.in new file mode 100644 index 0000000000..3d8aaf2754 --- /dev/null +++ b/media/libopus/celt/arm/armopts.s.in @@ -0,0 +1,37 @@ +/* Copyright (C) 2013 Mozilla Corporation */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +; Set the following to 1 if we have EDSP instructions +; (LDRD/STRD, etc., ARMv5E and later). +OPUS_ARM_MAY_HAVE_EDSP * @OPUS_ARM_MAY_HAVE_EDSP@ + +; Set the following to 1 if we have ARMv6 media instructions. +OPUS_ARM_MAY_HAVE_MEDIA * @OPUS_ARM_MAY_HAVE_MEDIA@ + +; Set the following to 1 if we have NEON (some ARMv7) +OPUS_ARM_MAY_HAVE_NEON * @OPUS_ARM_MAY_HAVE_NEON@ + +END diff --git a/media/libopus/celt/arm/celt_fft_ne10.c b/media/libopus/celt/arm/celt_fft_ne10.c new file mode 100644 index 0000000000..ea5fd7808b --- /dev/null +++ b/media/libopus/celt/arm/celt_fft_ne10.c @@ -0,0 +1,173 @@ +/* Copyright (c) 2015 Xiph.Org Foundation + Written by Viswanath Puttagunta */ +/** + @file celt_fft_ne10.c + @brief ARM Neon optimizations for fft using NE10 library + */ + +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef SKIP_CONFIG_H +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif +#endif + +#include <NE10_dsp.h> +#include "os_support.h" +#include "kiss_fft.h" +#include "stack_alloc.h" + +#if !defined(FIXED_POINT) +# define NE10_FFT_ALLOC_C2C_TYPE_NEON ne10_fft_alloc_c2c_float32_neon +# define NE10_FFT_CFG_TYPE_T ne10_fft_cfg_float32_t +# define NE10_FFT_STATE_TYPE_T ne10_fft_state_float32_t +# define NE10_FFT_DESTROY_C2C_TYPE ne10_fft_destroy_c2c_float32 +# define NE10_FFT_CPX_TYPE_T ne10_fft_cpx_float32_t +# define NE10_FFT_C2C_1D_TYPE_NEON ne10_fft_c2c_1d_float32_neon +#else +# define NE10_FFT_ALLOC_C2C_TYPE_NEON(nfft) ne10_fft_alloc_c2c_int32_neon(nfft) +# define NE10_FFT_CFG_TYPE_T ne10_fft_cfg_int32_t +# define NE10_FFT_STATE_TYPE_T ne10_fft_state_int32_t +# define NE10_FFT_DESTROY_C2C_TYPE ne10_fft_destroy_c2c_int32 +# define NE10_FFT_DESTROY_C2C_TYPE ne10_fft_destroy_c2c_int32 +# define NE10_FFT_CPX_TYPE_T ne10_fft_cpx_int32_t +# define NE10_FFT_C2C_1D_TYPE_NEON ne10_fft_c2c_1d_int32_neon +#endif + +#if defined(CUSTOM_MODES) + +/* nfft lengths in NE10 that support scaled fft */ +# define NE10_FFTSCALED_SUPPORT_MAX 4 +static const int ne10_fft_scaled_support[NE10_FFTSCALED_SUPPORT_MAX] = { + 480, 240, 120, 60 +}; + +int opus_fft_alloc_arm_neon(kiss_fft_state *st) +{ + int i; + size_t memneeded = sizeof(struct arch_fft_state); + + st->arch_fft = (arch_fft_state *)opus_alloc(memneeded); + if (!st->arch_fft) + return -1; + + for (i = 0; i < NE10_FFTSCALED_SUPPORT_MAX; i++) { + if(st->nfft == ne10_fft_scaled_support[i]) + break; + } + if (i == NE10_FFTSCALED_SUPPORT_MAX) { + /* This nfft length (scaled fft) is not supported in NE10 */ + st->arch_fft->is_supported = 0; + st->arch_fft->priv = NULL; + } + else { + st->arch_fft->is_supported = 1; + st->arch_fft->priv = (void *)NE10_FFT_ALLOC_C2C_TYPE_NEON(st->nfft); + if (st->arch_fft->priv == NULL) { + return -1; + } + } + return 0; +} + +void opus_fft_free_arm_neon(kiss_fft_state *st) +{ + NE10_FFT_CFG_TYPE_T cfg; + + if (!st->arch_fft) + return; + + cfg = (NE10_FFT_CFG_TYPE_T)st->arch_fft->priv; + if (cfg) + NE10_FFT_DESTROY_C2C_TYPE(cfg); + opus_free(st->arch_fft); +} +#endif + +void opus_fft_neon(const kiss_fft_state *st, + const kiss_fft_cpx *fin, + kiss_fft_cpx *fout) +{ + NE10_FFT_STATE_TYPE_T state; + NE10_FFT_CFG_TYPE_T cfg = &state; + VARDECL(NE10_FFT_CPX_TYPE_T, buffer); + SAVE_STACK; + ALLOC(buffer, st->nfft, NE10_FFT_CPX_TYPE_T); + + if (!st->arch_fft->is_supported) { + /* This nfft length (scaled fft) not supported in NE10 */ + opus_fft_c(st, fin, fout); + } + else { + memcpy((void *)cfg, st->arch_fft->priv, sizeof(NE10_FFT_STATE_TYPE_T)); + state.buffer = (NE10_FFT_CPX_TYPE_T *)&buffer[0]; +#if !defined(FIXED_POINT) + state.is_forward_scaled = 1; + + NE10_FFT_C2C_1D_TYPE_NEON((NE10_FFT_CPX_TYPE_T *)fout, + (NE10_FFT_CPX_TYPE_T *)fin, + cfg, 0); +#else + NE10_FFT_C2C_1D_TYPE_NEON((NE10_FFT_CPX_TYPE_T *)fout, + (NE10_FFT_CPX_TYPE_T *)fin, + cfg, 0, 1); +#endif + } + RESTORE_STACK; +} + +void opus_ifft_neon(const kiss_fft_state *st, + const kiss_fft_cpx *fin, + kiss_fft_cpx *fout) +{ + NE10_FFT_STATE_TYPE_T state; + NE10_FFT_CFG_TYPE_T cfg = &state; + VARDECL(NE10_FFT_CPX_TYPE_T, buffer); + SAVE_STACK; + ALLOC(buffer, st->nfft, NE10_FFT_CPX_TYPE_T); + + if (!st->arch_fft->is_supported) { + /* This nfft length (scaled fft) not supported in NE10 */ + opus_ifft_c(st, fin, fout); + } + else { + memcpy((void *)cfg, st->arch_fft->priv, sizeof(NE10_FFT_STATE_TYPE_T)); + state.buffer = (NE10_FFT_CPX_TYPE_T *)&buffer[0]; +#if !defined(FIXED_POINT) + state.is_backward_scaled = 0; + + NE10_FFT_C2C_1D_TYPE_NEON((NE10_FFT_CPX_TYPE_T *)fout, + (NE10_FFT_CPX_TYPE_T *)fin, + cfg, 1); +#else + NE10_FFT_C2C_1D_TYPE_NEON((NE10_FFT_CPX_TYPE_T *)fout, + (NE10_FFT_CPX_TYPE_T *)fin, + cfg, 1, 0); +#endif + } + RESTORE_STACK; +} diff --git a/media/libopus/celt/arm/celt_mdct_ne10.c b/media/libopus/celt/arm/celt_mdct_ne10.c new file mode 100644 index 0000000000..3531d02d10 --- /dev/null +++ b/media/libopus/celt/arm/celt_mdct_ne10.c @@ -0,0 +1,258 @@ +/* Copyright (c) 2015 Xiph.Org Foundation + Written by Viswanath Puttagunta */ +/** + @file celt_mdct_ne10.c + @brief ARM Neon optimizations for mdct using NE10 library + */ + +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef SKIP_CONFIG_H +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif +#endif + +#include "kiss_fft.h" +#include "_kiss_fft_guts.h" +#include "mdct.h" +#include "stack_alloc.h" + +void clt_mdct_forward_neon(const mdct_lookup *l, + kiss_fft_scalar *in, + kiss_fft_scalar * OPUS_RESTRICT out, + const opus_val16 *window, + int overlap, int shift, int stride, int arch) +{ + int i; + int N, N2, N4; + VARDECL(kiss_fft_scalar, f); + VARDECL(kiss_fft_cpx, f2); + const kiss_fft_state *st = l->kfft[shift]; + const kiss_twiddle_scalar *trig; + + SAVE_STACK; + + N = l->n; + trig = l->trig; + for (i=0;i<shift;i++) + { + N >>= 1; + trig += N; + } + N2 = N>>1; + N4 = N>>2; + + ALLOC(f, N2, kiss_fft_scalar); + ALLOC(f2, N4, kiss_fft_cpx); + + /* Consider the input to be composed of four blocks: [a, b, c, d] */ + /* Window, shuffle, fold */ + { + /* Temp pointers to make it really clear to the compiler what we're doing */ + const kiss_fft_scalar * OPUS_RESTRICT xp1 = in+(overlap>>1); + const kiss_fft_scalar * OPUS_RESTRICT xp2 = in+N2-1+(overlap>>1); + kiss_fft_scalar * OPUS_RESTRICT yp = f; + const opus_val16 * OPUS_RESTRICT wp1 = window+(overlap>>1); + const opus_val16 * OPUS_RESTRICT wp2 = window+(overlap>>1)-1; + for(i=0;i<((overlap+3)>>2);i++) + { + /* Real part arranged as -d-cR, Imag part arranged as -b+aR*/ + *yp++ = MULT16_32_Q15(*wp2, xp1[N2]) + MULT16_32_Q15(*wp1,*xp2); + *yp++ = MULT16_32_Q15(*wp1, *xp1) - MULT16_32_Q15(*wp2, xp2[-N2]); + xp1+=2; + xp2-=2; + wp1+=2; + wp2-=2; + } + wp1 = window; + wp2 = window+overlap-1; + for(;i<N4-((overlap+3)>>2);i++) + { + /* Real part arranged as a-bR, Imag part arranged as -c-dR */ + *yp++ = *xp2; + *yp++ = *xp1; + xp1+=2; + xp2-=2; + } + for(;i<N4;i++) + { + /* Real part arranged as a-bR, Imag part arranged as -c-dR */ + *yp++ = -MULT16_32_Q15(*wp1, xp1[-N2]) + MULT16_32_Q15(*wp2, *xp2); + *yp++ = MULT16_32_Q15(*wp2, *xp1) + MULT16_32_Q15(*wp1, xp2[N2]); + xp1+=2; + xp2-=2; + wp1+=2; + wp2-=2; + } + } + /* Pre-rotation */ + { + kiss_fft_scalar * OPUS_RESTRICT yp = f; + const kiss_twiddle_scalar *t = &trig[0]; + for(i=0;i<N4;i++) + { + kiss_fft_cpx yc; + kiss_twiddle_scalar t0, t1; + kiss_fft_scalar re, im, yr, yi; + t0 = t[i]; + t1 = t[N4+i]; + re = *yp++; + im = *yp++; + yr = S_MUL(re,t0) - S_MUL(im,t1); + yi = S_MUL(im,t0) + S_MUL(re,t1); + yc.r = yr; + yc.i = yi; + f2[i] = yc; + } + } + + opus_fft(st, f2, (kiss_fft_cpx *)f, arch); + + /* Post-rotate */ + { + /* Temp pointers to make it really clear to the compiler what we're doing */ + const kiss_fft_cpx * OPUS_RESTRICT fp = (kiss_fft_cpx *)f; + kiss_fft_scalar * OPUS_RESTRICT yp1 = out; + kiss_fft_scalar * OPUS_RESTRICT yp2 = out+stride*(N2-1); + const kiss_twiddle_scalar *t = &trig[0]; + /* Temp pointers to make it really clear to the compiler what we're doing */ + for(i=0;i<N4;i++) + { + kiss_fft_scalar yr, yi; + yr = S_MUL(fp->i,t[N4+i]) - S_MUL(fp->r,t[i]); + yi = S_MUL(fp->r,t[N4+i]) + S_MUL(fp->i,t[i]); + *yp1 = yr; + *yp2 = yi; + fp++; + yp1 += 2*stride; + yp2 -= 2*stride; + } + } + RESTORE_STACK; +} + +void clt_mdct_backward_neon(const mdct_lookup *l, + kiss_fft_scalar *in, + kiss_fft_scalar * OPUS_RESTRICT out, + const opus_val16 * OPUS_RESTRICT window, + int overlap, int shift, int stride, int arch) +{ + int i; + int N, N2, N4; + VARDECL(kiss_fft_scalar, f); + const kiss_twiddle_scalar *trig; + const kiss_fft_state *st = l->kfft[shift]; + + N = l->n; + trig = l->trig; + for (i=0;i<shift;i++) + { + N >>= 1; + trig += N; + } + N2 = N>>1; + N4 = N>>2; + + ALLOC(f, N2, kiss_fft_scalar); + + /* Pre-rotate */ + { + /* Temp pointers to make it really clear to the compiler what we're doing */ + const kiss_fft_scalar * OPUS_RESTRICT xp1 = in; + const kiss_fft_scalar * OPUS_RESTRICT xp2 = in+stride*(N2-1); + kiss_fft_scalar * OPUS_RESTRICT yp = f; + const kiss_twiddle_scalar * OPUS_RESTRICT t = &trig[0]; + for(i=0;i<N4;i++) + { + kiss_fft_scalar yr, yi; + yr = S_MUL(*xp2, t[i]) + S_MUL(*xp1, t[N4+i]); + yi = S_MUL(*xp1, t[i]) - S_MUL(*xp2, t[N4+i]); + yp[2*i] = yr; + yp[2*i+1] = yi; + xp1+=2*stride; + xp2-=2*stride; + } + } + + opus_ifft(st, (kiss_fft_cpx *)f, (kiss_fft_cpx*)(out+(overlap>>1)), arch); + + /* Post-rotate and de-shuffle from both ends of the buffer at once to make + it in-place. */ + { + kiss_fft_scalar * yp0 = out+(overlap>>1); + kiss_fft_scalar * yp1 = out+(overlap>>1)+N2-2; + const kiss_twiddle_scalar *t = &trig[0]; + /* Loop to (N4+1)>>1 to handle odd N4. When N4 is odd, the + middle pair will be computed twice. */ + for(i=0;i<(N4+1)>>1;i++) + { + kiss_fft_scalar re, im, yr, yi; + kiss_twiddle_scalar t0, t1; + re = yp0[0]; + im = yp0[1]; + t0 = t[i]; + t1 = t[N4+i]; + /* We'd scale up by 2 here, but instead it's done when mixing the windows */ + yr = S_MUL(re,t0) + S_MUL(im,t1); + yi = S_MUL(re,t1) - S_MUL(im,t0); + re = yp1[0]; + im = yp1[1]; + yp0[0] = yr; + yp1[1] = yi; + + t0 = t[(N4-i-1)]; + t1 = t[(N2-i-1)]; + /* We'd scale up by 2 here, but instead it's done when mixing the windows */ + yr = S_MUL(re,t0) + S_MUL(im,t1); + yi = S_MUL(re,t1) - S_MUL(im,t0); + yp1[0] = yr; + yp0[1] = yi; + yp0 += 2; + yp1 -= 2; + } + } + + /* Mirror on both sides for TDAC */ + { + kiss_fft_scalar * OPUS_RESTRICT xp1 = out+overlap-1; + kiss_fft_scalar * OPUS_RESTRICT yp1 = out; + const opus_val16 * OPUS_RESTRICT wp1 = window; + const opus_val16 * OPUS_RESTRICT wp2 = window+overlap-1; + + for(i = 0; i < overlap/2; i++) + { + kiss_fft_scalar x1, x2; + x1 = *xp1; + x2 = *yp1; + *yp1++ = MULT16_32_Q15(*wp2, x2) - MULT16_32_Q15(*wp1, x1); + *xp1-- = MULT16_32_Q15(*wp1, x2) + MULT16_32_Q15(*wp2, x1); + wp1++; + wp2--; + } + } + RESTORE_STACK; +} diff --git a/media/libopus/celt/arm/celt_neon_intr.c b/media/libopus/celt/arm/celt_neon_intr.c new file mode 100644 index 0000000000..effda769d0 --- /dev/null +++ b/media/libopus/celt/arm/celt_neon_intr.c @@ -0,0 +1,211 @@ +/* Copyright (c) 2014-2015 Xiph.Org Foundation + Written by Viswanath Puttagunta */ +/** + @file celt_neon_intr.c + @brief ARM Neon Intrinsic optimizations for celt + */ + +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include <arm_neon.h> +#include "../pitch.h" + +#if defined(FIXED_POINT) +void xcorr_kernel_neon_fixed(const opus_val16 * x, const opus_val16 * y, opus_val32 sum[4], int len) +{ + int j; + int32x4_t a = vld1q_s32(sum); + /* Load y[0...3] */ + /* This requires len>0 to always be valid (which we assert in the C code). */ + int16x4_t y0 = vld1_s16(y); + y += 4; + + for (j = 0; j + 8 <= len; j += 8) + { + /* Load x[0...7] */ + int16x8_t xx = vld1q_s16(x); + int16x4_t x0 = vget_low_s16(xx); + int16x4_t x4 = vget_high_s16(xx); + /* Load y[4...11] */ + int16x8_t yy = vld1q_s16(y); + int16x4_t y4 = vget_low_s16(yy); + int16x4_t y8 = vget_high_s16(yy); + int32x4_t a0 = vmlal_lane_s16(a, y0, x0, 0); + int32x4_t a1 = vmlal_lane_s16(a0, y4, x4, 0); + + int16x4_t y1 = vext_s16(y0, y4, 1); + int16x4_t y5 = vext_s16(y4, y8, 1); + int32x4_t a2 = vmlal_lane_s16(a1, y1, x0, 1); + int32x4_t a3 = vmlal_lane_s16(a2, y5, x4, 1); + + int16x4_t y2 = vext_s16(y0, y4, 2); + int16x4_t y6 = vext_s16(y4, y8, 2); + int32x4_t a4 = vmlal_lane_s16(a3, y2, x0, 2); + int32x4_t a5 = vmlal_lane_s16(a4, y6, x4, 2); + + int16x4_t y3 = vext_s16(y0, y4, 3); + int16x4_t y7 = vext_s16(y4, y8, 3); + int32x4_t a6 = vmlal_lane_s16(a5, y3, x0, 3); + int32x4_t a7 = vmlal_lane_s16(a6, y7, x4, 3); + + y0 = y8; + a = a7; + x += 8; + y += 8; + } + + for (; j < len; j++) + { + int16x4_t x0 = vld1_dup_s16(x); /* load next x */ + int32x4_t a0 = vmlal_s16(a, y0, x0); + + int16x4_t y4 = vld1_dup_s16(y); /* load next y */ + y0 = vext_s16(y0, y4, 1); + a = a0; + x++; + y++; + } + + vst1q_s32(sum, a); +} + +#else +/* + * Function: xcorr_kernel_neon_float + * --------------------------------- + * Computes 4 correlation values and stores them in sum[4] + */ +static void xcorr_kernel_neon_float(const float32_t *x, const float32_t *y, + float32_t sum[4], int len) { + float32x4_t YY[3]; + float32x4_t YEXT[3]; + float32x4_t XX[2]; + float32x2_t XX_2; + float32x4_t SUMM; + const float32_t *xi = x; + const float32_t *yi = y; + + celt_assert(len>0); + + YY[0] = vld1q_f32(yi); + SUMM = vdupq_n_f32(0); + + /* Consume 8 elements in x vector and 12 elements in y + * vector. However, the 12'th element never really gets + * touched in this loop. So, if len == 8, then we only + * must access y[0] to y[10]. y[11] must not be accessed + * hence make sure len > 8 and not len >= 8 + */ + while (len > 8) { + yi += 4; + YY[1] = vld1q_f32(yi); + yi += 4; + YY[2] = vld1q_f32(yi); + + XX[0] = vld1q_f32(xi); + xi += 4; + XX[1] = vld1q_f32(xi); + xi += 4; + + SUMM = vmlaq_lane_f32(SUMM, YY[0], vget_low_f32(XX[0]), 0); + YEXT[0] = vextq_f32(YY[0], YY[1], 1); + SUMM = vmlaq_lane_f32(SUMM, YEXT[0], vget_low_f32(XX[0]), 1); + YEXT[1] = vextq_f32(YY[0], YY[1], 2); + SUMM = vmlaq_lane_f32(SUMM, YEXT[1], vget_high_f32(XX[0]), 0); + YEXT[2] = vextq_f32(YY[0], YY[1], 3); + SUMM = vmlaq_lane_f32(SUMM, YEXT[2], vget_high_f32(XX[0]), 1); + + SUMM = vmlaq_lane_f32(SUMM, YY[1], vget_low_f32(XX[1]), 0); + YEXT[0] = vextq_f32(YY[1], YY[2], 1); + SUMM = vmlaq_lane_f32(SUMM, YEXT[0], vget_low_f32(XX[1]), 1); + YEXT[1] = vextq_f32(YY[1], YY[2], 2); + SUMM = vmlaq_lane_f32(SUMM, YEXT[1], vget_high_f32(XX[1]), 0); + YEXT[2] = vextq_f32(YY[1], YY[2], 3); + SUMM = vmlaq_lane_f32(SUMM, YEXT[2], vget_high_f32(XX[1]), 1); + + YY[0] = YY[2]; + len -= 8; + } + + /* Consume 4 elements in x vector and 8 elements in y + * vector. However, the 8'th element in y never really gets + * touched in this loop. So, if len == 4, then we only + * must access y[0] to y[6]. y[7] must not be accessed + * hence make sure len>4 and not len>=4 + */ + if (len > 4) { + yi += 4; + YY[1] = vld1q_f32(yi); + + XX[0] = vld1q_f32(xi); + xi += 4; + + SUMM = vmlaq_lane_f32(SUMM, YY[0], vget_low_f32(XX[0]), 0); + YEXT[0] = vextq_f32(YY[0], YY[1], 1); + SUMM = vmlaq_lane_f32(SUMM, YEXT[0], vget_low_f32(XX[0]), 1); + YEXT[1] = vextq_f32(YY[0], YY[1], 2); + SUMM = vmlaq_lane_f32(SUMM, YEXT[1], vget_high_f32(XX[0]), 0); + YEXT[2] = vextq_f32(YY[0], YY[1], 3); + SUMM = vmlaq_lane_f32(SUMM, YEXT[2], vget_high_f32(XX[0]), 1); + + YY[0] = YY[1]; + len -= 4; + } + + while (--len > 0) { + XX_2 = vld1_dup_f32(xi++); + SUMM = vmlaq_lane_f32(SUMM, YY[0], XX_2, 0); + YY[0]= vld1q_f32(++yi); + } + + XX_2 = vld1_dup_f32(xi); + SUMM = vmlaq_lane_f32(SUMM, YY[0], XX_2, 0); + + vst1q_f32(sum, SUMM); +} + +void celt_pitch_xcorr_float_neon(const opus_val16 *_x, const opus_val16 *_y, + opus_val32 *xcorr, int len, int max_pitch, int arch) { + int i; + (void)arch; + celt_assert(max_pitch > 0); + celt_sig_assert((((unsigned char *)_x-(unsigned char *)NULL)&3)==0); + + for (i = 0; i < (max_pitch-3); i += 4) { + xcorr_kernel_neon_float((const float32_t *)_x, (const float32_t *)_y+i, + (float32_t *)xcorr+i, len); + } + + /* In case max_pitch isn't a multiple of 4, do non-unrolled version. */ + for (; i < max_pitch; i++) { + xcorr[i] = celt_inner_prod_neon(_x, _y+i, len); + } +} +#endif diff --git a/media/libopus/celt/arm/celt_pitch_xcorr_arm.s b/media/libopus/celt/arm/celt_pitch_xcorr_arm.s new file mode 100644 index 0000000000..6e873afc37 --- /dev/null +++ b/media/libopus/celt/arm/celt_pitch_xcorr_arm.s @@ -0,0 +1,551 @@ +; Copyright (c) 2007-2008 CSIRO +; Copyright (c) 2007-2009 Xiph.Org Foundation +; Copyright (c) 2013 Parrot +; Written by Aurélien Zanelli +; +; Redistribution and use in source and binary forms, with or without +; modification, are permitted provided that the following conditions +; are met: +; +; - Redistributions of source code must retain the above copyright +; notice, this list of conditions and the following disclaimer. +; +; - Redistributions in binary form must reproduce the above copyright +; notice, this list of conditions and the following disclaimer in the +; documentation and/or other materials provided with the distribution. +; +; THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS +; ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT +; LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR +; A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER +; OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, +; EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, +; PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR +; PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF +; LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING +; NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS +; SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + + AREA |.text|, CODE, READONLY + + GET celt/arm/armopts.s + +IF OPUS_ARM_MAY_HAVE_EDSP + EXPORT celt_pitch_xcorr_edsp +ENDIF + +IF OPUS_ARM_MAY_HAVE_NEON + EXPORT celt_pitch_xcorr_neon +ENDIF + +IF OPUS_ARM_MAY_HAVE_NEON + +; Compute sum[k]=sum(x[j]*y[j+k],j=0...len-1), k=0...3 +xcorr_kernel_neon PROC +xcorr_kernel_neon_start + ; input: + ; r3 = int len + ; r4 = opus_val16 *x + ; r5 = opus_val16 *y + ; q0 = opus_val32 sum[4] + ; output: + ; q0 = opus_val32 sum[4] + ; preserved: r0-r3, r6-r11, d2, q4-q7, q9-q15 + ; internal usage: + ; r12 = int j + ; d3 = y_3|y_2|y_1|y_0 + ; q2 = y_B|y_A|y_9|y_8|y_7|y_6|y_5|y_4 + ; q3 = x_7|x_6|x_5|x_4|x_3|x_2|x_1|x_0 + ; q8 = scratch + ; + ; Load y[0...3] + ; This requires len>0 to always be valid (which we assert in the C code). + VLD1.16 {d5}, [r5]! + SUBS r12, r3, #8 + BLE xcorr_kernel_neon_process4 +; Process 8 samples at a time. +; This loop loads one y value more than we actually need. Therefore we have to +; stop as soon as there are 8 or fewer samples left (instead of 7), to avoid +; reading past the end of the array. +xcorr_kernel_neon_process8 + ; This loop has 19 total instructions (10 cycles to issue, minimum), with + ; - 2 cycles of ARM insrtuctions, + ; - 10 cycles of load/store/byte permute instructions, and + ; - 9 cycles of data processing instructions. + ; On a Cortex A8, we dual-issue the maximum amount (9 cycles) between the + ; latter two categories, meaning the whole loop should run in 10 cycles per + ; iteration, barring cache misses. + ; + ; Load x[0...7] + VLD1.16 {d6, d7}, [r4]! + ; Unlike VMOV, VAND is a data processsing instruction (and doesn't get + ; assembled to VMOV, like VORR would), so it dual-issues with the prior VLD1. + VAND d3, d5, d5 + SUBS r12, r12, #8 + ; Load y[4...11] + VLD1.16 {d4, d5}, [r5]! + VMLAL.S16 q0, d3, d6[0] + VEXT.16 d16, d3, d4, #1 + VMLAL.S16 q0, d4, d7[0] + VEXT.16 d17, d4, d5, #1 + VMLAL.S16 q0, d16, d6[1] + VEXT.16 d16, d3, d4, #2 + VMLAL.S16 q0, d17, d7[1] + VEXT.16 d17, d4, d5, #2 + VMLAL.S16 q0, d16, d6[2] + VEXT.16 d16, d3, d4, #3 + VMLAL.S16 q0, d17, d7[2] + VEXT.16 d17, d4, d5, #3 + VMLAL.S16 q0, d16, d6[3] + VMLAL.S16 q0, d17, d7[3] + BGT xcorr_kernel_neon_process8 +; Process 4 samples here if we have > 4 left (still reading one extra y value). +xcorr_kernel_neon_process4 + ADDS r12, r12, #4 + BLE xcorr_kernel_neon_process2 + ; Load x[0...3] + VLD1.16 d6, [r4]! + ; Use VAND since it's a data processing instruction again. + VAND d4, d5, d5 + SUB r12, r12, #4 + ; Load y[4...7] + VLD1.16 d5, [r5]! + VMLAL.S16 q0, d4, d6[0] + VEXT.16 d16, d4, d5, #1 + VMLAL.S16 q0, d16, d6[1] + VEXT.16 d16, d4, d5, #2 + VMLAL.S16 q0, d16, d6[2] + VEXT.16 d16, d4, d5, #3 + VMLAL.S16 q0, d16, d6[3] +; Process 2 samples here if we have > 2 left (still reading one extra y value). +xcorr_kernel_neon_process2 + ADDS r12, r12, #2 + BLE xcorr_kernel_neon_process1 + ; Load x[0...1] + VLD2.16 {d6[],d7[]}, [r4]! + ; Use VAND since it's a data processing instruction again. + VAND d4, d5, d5 + SUB r12, r12, #2 + ; Load y[4...5] + VLD1.32 {d5[]}, [r5]! + VMLAL.S16 q0, d4, d6 + VEXT.16 d16, d4, d5, #1 + ; Replace bottom copy of {y5,y4} in d5 with {y3,y2} from d4, using VSRI + ; instead of VEXT, since it's a data-processing instruction. + VSRI.64 d5, d4, #32 + VMLAL.S16 q0, d16, d7 +; Process 1 sample using the extra y value we loaded above. +xcorr_kernel_neon_process1 + ; Load next *x + VLD1.16 {d6[]}, [r4]! + ADDS r12, r12, #1 + ; y[0...3] are left in d5 from prior iteration(s) (if any) + VMLAL.S16 q0, d5, d6 + MOVLE pc, lr +; Now process 1 last sample, not reading ahead. + ; Load last *y + VLD1.16 {d4[]}, [r5]! + VSRI.64 d4, d5, #16 + ; Load last *x + VLD1.16 {d6[]}, [r4]! + VMLAL.S16 q0, d4, d6 + MOV pc, lr + ENDP + +; opus_val32 celt_pitch_xcorr_neon(opus_val16 *_x, opus_val16 *_y, +; opus_val32 *xcorr, int len, int max_pitch, int arch) +celt_pitch_xcorr_neon PROC + ; input: + ; r0 = opus_val16 *_x + ; r1 = opus_val16 *_y + ; r2 = opus_val32 *xcorr + ; r3 = int len + ; output: + ; r0 = int maxcorr + ; internal usage: + ; r4 = opus_val16 *x (for xcorr_kernel_neon()) + ; r5 = opus_val16 *y (for xcorr_kernel_neon()) + ; r6 = int max_pitch + ; r12 = int j + ; q15 = int maxcorr[4] (q15 is not used by xcorr_kernel_neon()) + ; ignored: + ; int arch + STMFD sp!, {r4-r6, lr} + LDR r6, [sp, #16] + VMOV.S32 q15, #1 + ; if (max_pitch < 4) goto celt_pitch_xcorr_neon_process4_done + SUBS r6, r6, #4 + BLT celt_pitch_xcorr_neon_process4_done +celt_pitch_xcorr_neon_process4 + ; xcorr_kernel_neon parameters: + ; r3 = len, r4 = _x, r5 = _y, q0 = {0, 0, 0, 0} + MOV r4, r0 + MOV r5, r1 + VEOR q0, q0, q0 + ; xcorr_kernel_neon only modifies r4, r5, r12, and q0...q3. + ; So we don't save/restore any other registers. + BL xcorr_kernel_neon_start + SUBS r6, r6, #4 + VST1.32 {q0}, [r2]! + ; _y += 4 + ADD r1, r1, #8 + VMAX.S32 q15, q15, q0 + ; if (max_pitch < 4) goto celt_pitch_xcorr_neon_process4_done + BGE celt_pitch_xcorr_neon_process4 +; We have less than 4 sums left to compute. +celt_pitch_xcorr_neon_process4_done + ADDS r6, r6, #4 + ; Reduce maxcorr to a single value + VMAX.S32 d30, d30, d31 + VPMAX.S32 d30, d30, d30 + ; if (max_pitch <= 0) goto celt_pitch_xcorr_neon_done + BLE celt_pitch_xcorr_neon_done +; Now compute each remaining sum one at a time. +celt_pitch_xcorr_neon_process_remaining + MOV r4, r0 + MOV r5, r1 + VMOV.I32 q0, #0 + SUBS r12, r3, #8 + BLT celt_pitch_xcorr_neon_process_remaining4 +; Sum terms 8 at a time. +celt_pitch_xcorr_neon_process_remaining_loop8 + ; Load x[0...7] + VLD1.16 {q1}, [r4]! + ; Load y[0...7] + VLD1.16 {q2}, [r5]! + SUBS r12, r12, #8 + VMLAL.S16 q0, d4, d2 + VMLAL.S16 q0, d5, d3 + BGE celt_pitch_xcorr_neon_process_remaining_loop8 +; Sum terms 4 at a time. +celt_pitch_xcorr_neon_process_remaining4 + ADDS r12, r12, #4 + BLT celt_pitch_xcorr_neon_process_remaining4_done + ; Load x[0...3] + VLD1.16 {d2}, [r4]! + ; Load y[0...3] + VLD1.16 {d3}, [r5]! + SUB r12, r12, #4 + VMLAL.S16 q0, d3, d2 +celt_pitch_xcorr_neon_process_remaining4_done + ; Reduce the sum to a single value. + VADD.S32 d0, d0, d1 + VPADDL.S32 d0, d0 + ADDS r12, r12, #4 + BLE celt_pitch_xcorr_neon_process_remaining_loop_done +; Sum terms 1 at a time. +celt_pitch_xcorr_neon_process_remaining_loop1 + VLD1.16 {d2[]}, [r4]! + VLD1.16 {d3[]}, [r5]! + SUBS r12, r12, #1 + VMLAL.S16 q0, d2, d3 + BGT celt_pitch_xcorr_neon_process_remaining_loop1 +celt_pitch_xcorr_neon_process_remaining_loop_done + VST1.32 {d0[0]}, [r2]! + VMAX.S32 d30, d30, d0 + SUBS r6, r6, #1 + ; _y++ + ADD r1, r1, #2 + ; if (--max_pitch > 0) goto celt_pitch_xcorr_neon_process_remaining + BGT celt_pitch_xcorr_neon_process_remaining +celt_pitch_xcorr_neon_done + VMOV.32 r0, d30[0] + LDMFD sp!, {r4-r6, pc} + ENDP + +ENDIF + +IF OPUS_ARM_MAY_HAVE_EDSP + +; This will get used on ARMv7 devices without NEON, so it has been optimized +; to take advantage of dual-issuing where possible. +xcorr_kernel_edsp PROC +xcorr_kernel_edsp_start + ; input: + ; r3 = int len + ; r4 = opus_val16 *_x (must be 32-bit aligned) + ; r5 = opus_val16 *_y (must be 32-bit aligned) + ; r6...r9 = opus_val32 sum[4] + ; output: + ; r6...r9 = opus_val32 sum[4] + ; preserved: r0-r5 + ; internal usage + ; r2 = int j + ; r12,r14 = opus_val16 x[4] + ; r10,r11 = opus_val16 y[4] + STMFD sp!, {r2,r4,r5,lr} + LDR r10, [r5], #4 ; Load y[0...1] + SUBS r2, r3, #4 ; j = len-4 + LDR r11, [r5], #4 ; Load y[2...3] + BLE xcorr_kernel_edsp_process4_done + LDR r12, [r4], #4 ; Load x[0...1] + ; Stall +xcorr_kernel_edsp_process4 + ; The multiplies must issue from pipeline 0, and can't dual-issue with each + ; other. Every other instruction here dual-issues with a multiply, and is + ; thus "free". There should be no stalls in the body of the loop. + SMLABB r6, r12, r10, r6 ; sum[0] = MAC16_16(sum[0],x_0,y_0) + LDR r14, [r4], #4 ; Load x[2...3] + SMLABT r7, r12, r10, r7 ; sum[1] = MAC16_16(sum[1],x_0,y_1) + SUBS r2, r2, #4 ; j-=4 + SMLABB r8, r12, r11, r8 ; sum[2] = MAC16_16(sum[2],x_0,y_2) + SMLABT r9, r12, r11, r9 ; sum[3] = MAC16_16(sum[3],x_0,y_3) + SMLATT r6, r12, r10, r6 ; sum[0] = MAC16_16(sum[0],x_1,y_1) + LDR r10, [r5], #4 ; Load y[4...5] + SMLATB r7, r12, r11, r7 ; sum[1] = MAC16_16(sum[1],x_1,y_2) + SMLATT r8, r12, r11, r8 ; sum[2] = MAC16_16(sum[2],x_1,y_3) + SMLATB r9, r12, r10, r9 ; sum[3] = MAC16_16(sum[3],x_1,y_4) + LDRGT r12, [r4], #4 ; Load x[0...1] + SMLABB r6, r14, r11, r6 ; sum[0] = MAC16_16(sum[0],x_2,y_2) + SMLABT r7, r14, r11, r7 ; sum[1] = MAC16_16(sum[1],x_2,y_3) + SMLABB r8, r14, r10, r8 ; sum[2] = MAC16_16(sum[2],x_2,y_4) + SMLABT r9, r14, r10, r9 ; sum[3] = MAC16_16(sum[3],x_2,y_5) + SMLATT r6, r14, r11, r6 ; sum[0] = MAC16_16(sum[0],x_3,y_3) + LDR r11, [r5], #4 ; Load y[6...7] + SMLATB r7, r14, r10, r7 ; sum[1] = MAC16_16(sum[1],x_3,y_4) + SMLATT r8, r14, r10, r8 ; sum[2] = MAC16_16(sum[2],x_3,y_5) + SMLATB r9, r14, r11, r9 ; sum[3] = MAC16_16(sum[3],x_3,y_6) + BGT xcorr_kernel_edsp_process4 +xcorr_kernel_edsp_process4_done + ADDS r2, r2, #4 + BLE xcorr_kernel_edsp_done + LDRH r12, [r4], #2 ; r12 = *x++ + SUBS r2, r2, #1 ; j-- + ; Stall + SMLABB r6, r12, r10, r6 ; sum[0] = MAC16_16(sum[0],x,y_0) + LDRHGT r14, [r4], #2 ; r14 = *x++ + SMLABT r7, r12, r10, r7 ; sum[1] = MAC16_16(sum[1],x,y_1) + SMLABB r8, r12, r11, r8 ; sum[2] = MAC16_16(sum[2],x,y_2) + SMLABT r9, r12, r11, r9 ; sum[3] = MAC16_16(sum[3],x,y_3) + BLE xcorr_kernel_edsp_done + SMLABT r6, r14, r10, r6 ; sum[0] = MAC16_16(sum[0],x,y_1) + SUBS r2, r2, #1 ; j-- + SMLABB r7, r14, r11, r7 ; sum[1] = MAC16_16(sum[1],x,y_2) + LDRH r10, [r5], #2 ; r10 = y_4 = *y++ + SMLABT r8, r14, r11, r8 ; sum[2] = MAC16_16(sum[2],x,y_3) + LDRHGT r12, [r4], #2 ; r12 = *x++ + SMLABB r9, r14, r10, r9 ; sum[3] = MAC16_16(sum[3],x,y_4) + BLE xcorr_kernel_edsp_done + SMLABB r6, r12, r11, r6 ; sum[0] = MAC16_16(sum[0],tmp,y_2) + CMP r2, #1 ; j-- + SMLABT r7, r12, r11, r7 ; sum[1] = MAC16_16(sum[1],tmp,y_3) + LDRH r2, [r5], #2 ; r2 = y_5 = *y++ + SMLABB r8, r12, r10, r8 ; sum[2] = MAC16_16(sum[2],tmp,y_4) + LDRHGT r14, [r4] ; r14 = *x + SMLABB r9, r12, r2, r9 ; sum[3] = MAC16_16(sum[3],tmp,y_5) + BLE xcorr_kernel_edsp_done + SMLABT r6, r14, r11, r6 ; sum[0] = MAC16_16(sum[0],tmp,y_3) + LDRH r11, [r5] ; r11 = y_6 = *y + SMLABB r7, r14, r10, r7 ; sum[1] = MAC16_16(sum[1],tmp,y_4) + SMLABB r8, r14, r2, r8 ; sum[2] = MAC16_16(sum[2],tmp,y_5) + SMLABB r9, r14, r11, r9 ; sum[3] = MAC16_16(sum[3],tmp,y_6) +xcorr_kernel_edsp_done + LDMFD sp!, {r2,r4,r5,pc} + ENDP + +celt_pitch_xcorr_edsp PROC + ; input: + ; r0 = opus_val16 *_x (must be 32-bit aligned) + ; r1 = opus_val16 *_y (only needs to be 16-bit aligned) + ; r2 = opus_val32 *xcorr + ; r3 = int len + ; output: + ; r0 = maxcorr + ; internal usage + ; r4 = opus_val16 *x + ; r5 = opus_val16 *y + ; r6 = opus_val32 sum0 + ; r7 = opus_val32 sum1 + ; r8 = opus_val32 sum2 + ; r9 = opus_val32 sum3 + ; r1 = int max_pitch + ; r12 = int j + ; ignored: + ; int arch + STMFD sp!, {r4-r11, lr} + MOV r5, r1 + LDR r1, [sp, #36] + MOV r4, r0 + TST r5, #3 + ; maxcorr = 1 + MOV r0, #1 + BEQ celt_pitch_xcorr_edsp_process1u_done +; Compute one sum at the start to make y 32-bit aligned. + SUBS r12, r3, #4 + ; r14 = sum = 0 + MOV r14, #0 + LDRH r8, [r5], #2 + BLE celt_pitch_xcorr_edsp_process1u_loop4_done + LDR r6, [r4], #4 + MOV r8, r8, LSL #16 +celt_pitch_xcorr_edsp_process1u_loop4 + LDR r9, [r5], #4 + SMLABT r14, r6, r8, r14 ; sum = MAC16_16(sum, x_0, y_0) + LDR r7, [r4], #4 + SMLATB r14, r6, r9, r14 ; sum = MAC16_16(sum, x_1, y_1) + LDR r8, [r5], #4 + SMLABT r14, r7, r9, r14 ; sum = MAC16_16(sum, x_2, y_2) + SUBS r12, r12, #4 ; j-=4 + SMLATB r14, r7, r8, r14 ; sum = MAC16_16(sum, x_3, y_3) + LDRGT r6, [r4], #4 + BGT celt_pitch_xcorr_edsp_process1u_loop4 + MOV r8, r8, LSR #16 +celt_pitch_xcorr_edsp_process1u_loop4_done + ADDS r12, r12, #4 +celt_pitch_xcorr_edsp_process1u_loop1 + LDRHGE r6, [r4], #2 + ; Stall + SMLABBGE r14, r6, r8, r14 ; sum = MAC16_16(sum, *x, *y) + SUBSGE r12, r12, #1 + LDRHGT r8, [r5], #2 + BGT celt_pitch_xcorr_edsp_process1u_loop1 + ; Restore _x + SUB r4, r4, r3, LSL #1 + ; Restore and advance _y + SUB r5, r5, r3, LSL #1 + ; maxcorr = max(maxcorr, sum) + CMP r0, r14 + ADD r5, r5, #2 + MOVLT r0, r14 + SUBS r1, r1, #1 + ; xcorr[i] = sum + STR r14, [r2], #4 + BLE celt_pitch_xcorr_edsp_done +celt_pitch_xcorr_edsp_process1u_done + ; if (max_pitch < 4) goto celt_pitch_xcorr_edsp_process2 + SUBS r1, r1, #4 + BLT celt_pitch_xcorr_edsp_process2 +celt_pitch_xcorr_edsp_process4 + ; xcorr_kernel_edsp parameters: + ; r3 = len, r4 = _x, r5 = _y, r6...r9 = sum[4] = {0, 0, 0, 0} + MOV r6, #0 + MOV r7, #0 + MOV r8, #0 + MOV r9, #0 + BL xcorr_kernel_edsp_start ; xcorr_kernel_edsp(_x, _y+i, xcorr+i, len) + ; maxcorr = max(maxcorr, sum0, sum1, sum2, sum3) + CMP r0, r6 + ; _y+=4 + ADD r5, r5, #8 + MOVLT r0, r6 + CMP r0, r7 + MOVLT r0, r7 + CMP r0, r8 + MOVLT r0, r8 + CMP r0, r9 + MOVLT r0, r9 + STMIA r2!, {r6-r9} + SUBS r1, r1, #4 + BGE celt_pitch_xcorr_edsp_process4 +celt_pitch_xcorr_edsp_process2 + ADDS r1, r1, #2 + BLT celt_pitch_xcorr_edsp_process1a + SUBS r12, r3, #4 + ; {r10, r11} = {sum0, sum1} = {0, 0} + MOV r10, #0 + MOV r11, #0 + LDR r8, [r5], #4 + BLE celt_pitch_xcorr_edsp_process2_loop_done + LDR r6, [r4], #4 + LDR r9, [r5], #4 +celt_pitch_xcorr_edsp_process2_loop4 + SMLABB r10, r6, r8, r10 ; sum0 = MAC16_16(sum0, x_0, y_0) + LDR r7, [r4], #4 + SMLABT r11, r6, r8, r11 ; sum1 = MAC16_16(sum1, x_0, y_1) + SUBS r12, r12, #4 ; j-=4 + SMLATT r10, r6, r8, r10 ; sum0 = MAC16_16(sum0, x_1, y_1) + LDR r8, [r5], #4 + SMLATB r11, r6, r9, r11 ; sum1 = MAC16_16(sum1, x_1, y_2) + LDRGT r6, [r4], #4 + SMLABB r10, r7, r9, r10 ; sum0 = MAC16_16(sum0, x_2, y_2) + SMLABT r11, r7, r9, r11 ; sum1 = MAC16_16(sum1, x_2, y_3) + SMLATT r10, r7, r9, r10 ; sum0 = MAC16_16(sum0, x_3, y_3) + LDRGT r9, [r5], #4 + SMLATB r11, r7, r8, r11 ; sum1 = MAC16_16(sum1, x_3, y_4) + BGT celt_pitch_xcorr_edsp_process2_loop4 +celt_pitch_xcorr_edsp_process2_loop_done + ADDS r12, r12, #2 + BLE celt_pitch_xcorr_edsp_process2_1 + LDR r6, [r4], #4 + ; Stall + SMLABB r10, r6, r8, r10 ; sum0 = MAC16_16(sum0, x_0, y_0) + LDR r9, [r5], #4 + SMLABT r11, r6, r8, r11 ; sum1 = MAC16_16(sum1, x_0, y_1) + SUB r12, r12, #2 + SMLATT r10, r6, r8, r10 ; sum0 = MAC16_16(sum0, x_1, y_1) + MOV r8, r9 + SMLATB r11, r6, r9, r11 ; sum1 = MAC16_16(sum1, x_1, y_2) +celt_pitch_xcorr_edsp_process2_1 + LDRH r6, [r4], #2 + ADDS r12, r12, #1 + ; Stall + SMLABB r10, r6, r8, r10 ; sum0 = MAC16_16(sum0, x_0, y_0) + LDRHGT r7, [r4], #2 + SMLABT r11, r6, r8, r11 ; sum1 = MAC16_16(sum1, x_0, y_1) + BLE celt_pitch_xcorr_edsp_process2_done + LDRH r9, [r5], #2 + SMLABT r10, r7, r8, r10 ; sum0 = MAC16_16(sum0, x_0, y_1) + SMLABB r11, r7, r9, r11 ; sum1 = MAC16_16(sum1, x_0, y_2) +celt_pitch_xcorr_edsp_process2_done + ; Restore _x + SUB r4, r4, r3, LSL #1 + ; Restore and advance _y + SUB r5, r5, r3, LSL #1 + ; maxcorr = max(maxcorr, sum0) + CMP r0, r10 + ADD r5, r5, #2 + MOVLT r0, r10 + SUB r1, r1, #2 + ; maxcorr = max(maxcorr, sum1) + CMP r0, r11 + ; xcorr[i] = sum + STR r10, [r2], #4 + MOVLT r0, r11 + STR r11, [r2], #4 +celt_pitch_xcorr_edsp_process1a + ADDS r1, r1, #1 + BLT celt_pitch_xcorr_edsp_done + SUBS r12, r3, #4 + ; r14 = sum = 0 + MOV r14, #0 + BLT celt_pitch_xcorr_edsp_process1a_loop_done + LDR r6, [r4], #4 + LDR r8, [r5], #4 + LDR r7, [r4], #4 + LDR r9, [r5], #4 +celt_pitch_xcorr_edsp_process1a_loop4 + SMLABB r14, r6, r8, r14 ; sum = MAC16_16(sum, x_0, y_0) + SUBS r12, r12, #4 ; j-=4 + SMLATT r14, r6, r8, r14 ; sum = MAC16_16(sum, x_1, y_1) + LDRGE r6, [r4], #4 + SMLABB r14, r7, r9, r14 ; sum = MAC16_16(sum, x_2, y_2) + LDRGE r8, [r5], #4 + SMLATT r14, r7, r9, r14 ; sum = MAC16_16(sum, x_3, y_3) + LDRGE r7, [r4], #4 + LDRGE r9, [r5], #4 + BGE celt_pitch_xcorr_edsp_process1a_loop4 +celt_pitch_xcorr_edsp_process1a_loop_done + ADDS r12, r12, #2 + LDRGE r6, [r4], #4 + LDRGE r8, [r5], #4 + ; Stall + SMLABBGE r14, r6, r8, r14 ; sum = MAC16_16(sum, x_0, y_0) + SUBGE r12, r12, #2 + SMLATTGE r14, r6, r8, r14 ; sum = MAC16_16(sum, x_1, y_1) + ADDS r12, r12, #1 + LDRHGE r6, [r4], #2 + LDRHGE r8, [r5], #2 + ; Stall + SMLABBGE r14, r6, r8, r14 ; sum = MAC16_16(sum, *x, *y) + ; maxcorr = max(maxcorr, sum) + CMP r0, r14 + ; xcorr[i] = sum + STR r14, [r2], #4 + MOVLT r0, r14 +celt_pitch_xcorr_edsp_done + LDMFD sp!, {r4-r11, pc} + ENDP + +ENDIF + +END diff --git a/media/libopus/celt/arm/fft_arm.h b/media/libopus/celt/arm/fft_arm.h new file mode 100644 index 0000000000..0b78175f3a --- /dev/null +++ b/media/libopus/celt/arm/fft_arm.h @@ -0,0 +1,71 @@ +/* Copyright (c) 2015 Xiph.Org Foundation + Written by Viswanath Puttagunta */ +/** + @file fft_arm.h + @brief ARM Neon Intrinsic optimizations for fft using NE10 library + */ + +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + + +#if !defined(FFT_ARM_H) +#define FFT_ARM_H + +#include "kiss_fft.h" + +#if defined(HAVE_ARM_NE10) + +int opus_fft_alloc_arm_neon(kiss_fft_state *st); +void opus_fft_free_arm_neon(kiss_fft_state *st); + +void opus_fft_neon(const kiss_fft_state *st, + const kiss_fft_cpx *fin, + kiss_fft_cpx *fout); + +void opus_ifft_neon(const kiss_fft_state *st, + const kiss_fft_cpx *fin, + kiss_fft_cpx *fout); + +#if !defined(OPUS_HAVE_RTCD) +#define OVERRIDE_OPUS_FFT (1) + +#define opus_fft_alloc_arch(_st, arch) \ + ((void)(arch), opus_fft_alloc_arm_neon(_st)) + +#define opus_fft_free_arch(_st, arch) \ + ((void)(arch), opus_fft_free_arm_neon(_st)) + +#define opus_fft(_st, _fin, _fout, arch) \ + ((void)(arch), opus_fft_neon(_st, _fin, _fout)) + +#define opus_ifft(_st, _fin, _fout, arch) \ + ((void)(arch), opus_ifft_neon(_st, _fin, _fout)) + +#endif /* OPUS_HAVE_RTCD */ + +#endif /* HAVE_ARM_NE10 */ + +#endif diff --git a/media/libopus/celt/arm/fixed_arm64.h b/media/libopus/celt/arm/fixed_arm64.h new file mode 100644 index 0000000000..c6fbd3db2c --- /dev/null +++ b/media/libopus/celt/arm/fixed_arm64.h @@ -0,0 +1,35 @@ +/* Copyright (C) 2015 Vidyo */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef FIXED_ARM64_H +#define FIXED_ARM64_H + +#include <arm_neon.h> + +#undef SIG2WORD16 +#define SIG2WORD16(x) (vqmovns_s32(PSHR32((x), SIG_SHIFT))) + +#endif diff --git a/media/libopus/celt/arm/fixed_armv4.h b/media/libopus/celt/arm/fixed_armv4.h new file mode 100644 index 0000000000..d84888a772 --- /dev/null +++ b/media/libopus/celt/arm/fixed_armv4.h @@ -0,0 +1,80 @@ +/* Copyright (C) 2013 Xiph.Org Foundation and contributors */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef FIXED_ARMv4_H +#define FIXED_ARMv4_H + +/** 16x32 multiplication, followed by a 16-bit shift right. Results fits in 32 bits */ +#undef MULT16_32_Q16 +static OPUS_INLINE opus_val32 MULT16_32_Q16_armv4(opus_val16 a, opus_val32 b) +{ + unsigned rd_lo; + int rd_hi; + __asm__( + "#MULT16_32_Q16\n\t" + "smull %0, %1, %2, %3\n\t" + : "=&r"(rd_lo), "=&r"(rd_hi) + : "%r"(b),"r"(SHL32(a,16)) + ); + return rd_hi; +} +#define MULT16_32_Q16(a, b) (MULT16_32_Q16_armv4(a, b)) + + +/** 16x32 multiplication, followed by a 15-bit shift right. Results fits in 32 bits */ +#undef MULT16_32_Q15 +static OPUS_INLINE opus_val32 MULT16_32_Q15_armv4(opus_val16 a, opus_val32 b) +{ + unsigned rd_lo; + int rd_hi; + __asm__( + "#MULT16_32_Q15\n\t" + "smull %0, %1, %2, %3\n\t" + : "=&r"(rd_lo), "=&r"(rd_hi) + : "%r"(b), "r"(SHL32(a,16)) + ); + /*We intentionally don't OR in the high bit of rd_lo for speed.*/ + return SHL32(rd_hi,1); +} +#define MULT16_32_Q15(a, b) (MULT16_32_Q15_armv4(a, b)) + + +/** 16x32 multiply, followed by a 15-bit shift right and 32-bit add. + b must fit in 31 bits. + Result fits in 32 bits. */ +#undef MAC16_32_Q15 +#define MAC16_32_Q15(c, a, b) ADD32(c, MULT16_32_Q15(a, b)) + +/** 16x32 multiply, followed by a 16-bit shift right and 32-bit add. + Result fits in 32 bits. */ +#undef MAC16_32_Q16 +#define MAC16_32_Q16(c, a, b) ADD32(c, MULT16_32_Q16(a, b)) + +/** 32x32 multiplication, followed by a 31-bit shift right. Results fits in 32 bits */ +#undef MULT32_32_Q31 +#define MULT32_32_Q31(a,b) (opus_val32)((((opus_int64)(a)) * ((opus_int64)(b)))>>31) + +#endif diff --git a/media/libopus/celt/arm/fixed_armv5e.h b/media/libopus/celt/arm/fixed_armv5e.h new file mode 100644 index 0000000000..6bf73cbace --- /dev/null +++ b/media/libopus/celt/arm/fixed_armv5e.h @@ -0,0 +1,151 @@ +/* Copyright (C) 2007-2009 Xiph.Org Foundation + Copyright (C) 2003-2008 Jean-Marc Valin + Copyright (C) 2007-2008 CSIRO + Copyright (C) 2013 Parrot */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef FIXED_ARMv5E_H +#define FIXED_ARMv5E_H + +#include "fixed_armv4.h" + +/** 16x32 multiplication, followed by a 16-bit shift right. Results fits in 32 bits */ +#undef MULT16_32_Q16 +static OPUS_INLINE opus_val32 MULT16_32_Q16_armv5e(opus_val16 a, opus_val32 b) +{ + int res; + __asm__( + "#MULT16_32_Q16\n\t" + "smulwb %0, %1, %2\n\t" + : "=r"(res) + : "r"(b),"r"(a) + ); + return res; +} +#define MULT16_32_Q16(a, b) (MULT16_32_Q16_armv5e(a, b)) + + +/** 16x32 multiplication, followed by a 15-bit shift right. Results fits in 32 bits */ +#undef MULT16_32_Q15 +static OPUS_INLINE opus_val32 MULT16_32_Q15_armv5e(opus_val16 a, opus_val32 b) +{ + int res; + __asm__( + "#MULT16_32_Q15\n\t" + "smulwb %0, %1, %2\n\t" + : "=r"(res) + : "r"(b), "r"(a) + ); + return SHL32(res,1); +} +#define MULT16_32_Q15(a, b) (MULT16_32_Q15_armv5e(a, b)) + + +/** 16x32 multiply, followed by a 15-bit shift right and 32-bit add. + b must fit in 31 bits. + Result fits in 32 bits. */ +#undef MAC16_32_Q15 +static OPUS_INLINE opus_val32 MAC16_32_Q15_armv5e(opus_val32 c, opus_val16 a, + opus_val32 b) +{ + int res; + __asm__( + "#MAC16_32_Q15\n\t" + "smlawb %0, %1, %2, %3;\n" + : "=r"(res) + : "r"(SHL32(b,1)), "r"(a), "r"(c) + ); + return res; +} +#define MAC16_32_Q15(c, a, b) (MAC16_32_Q15_armv5e(c, a, b)) + +/** 16x32 multiply, followed by a 16-bit shift right and 32-bit add. + Result fits in 32 bits. */ +#undef MAC16_32_Q16 +static OPUS_INLINE opus_val32 MAC16_32_Q16_armv5e(opus_val32 c, opus_val16 a, + opus_val32 b) +{ + int res; + __asm__( + "#MAC16_32_Q16\n\t" + "smlawb %0, %1, %2, %3;\n" + : "=r"(res) + : "r"(b), "r"(a), "r"(c) + ); + return res; +} +#define MAC16_32_Q16(c, a, b) (MAC16_32_Q16_armv5e(c, a, b)) + +/** 16x16 multiply-add where the result fits in 32 bits */ +#undef MAC16_16 +static OPUS_INLINE opus_val32 MAC16_16_armv5e(opus_val32 c, opus_val16 a, + opus_val16 b) +{ + int res; + __asm__( + "#MAC16_16\n\t" + "smlabb %0, %1, %2, %3;\n" + : "=r"(res) + : "r"(a), "r"(b), "r"(c) + ); + return res; +} +#define MAC16_16(c, a, b) (MAC16_16_armv5e(c, a, b)) + +/** 16x16 multiplication where the result fits in 32 bits */ +#undef MULT16_16 +static OPUS_INLINE opus_val32 MULT16_16_armv5e(opus_val16 a, opus_val16 b) +{ + int res; + __asm__( + "#MULT16_16\n\t" + "smulbb %0, %1, %2;\n" + : "=r"(res) + : "r"(a), "r"(b) + ); + return res; +} +#define MULT16_16(a, b) (MULT16_16_armv5e(a, b)) + +#ifdef OPUS_ARM_INLINE_MEDIA + +#undef SIG2WORD16 +static OPUS_INLINE opus_val16 SIG2WORD16_armv6(opus_val32 x) +{ + celt_sig res; + __asm__( + "#SIG2WORD16\n\t" + "ssat %0, #16, %1, ASR #12\n\t" + : "=r"(res) + : "r"(x+2048) + ); + return EXTRACT16(res); +} +#define SIG2WORD16(x) (SIG2WORD16_armv6(x)) + +#endif /* OPUS_ARM_INLINE_MEDIA */ + +#endif diff --git a/media/libopus/celt/arm/kiss_fft_armv4.h b/media/libopus/celt/arm/kiss_fft_armv4.h new file mode 100644 index 0000000000..e4faad6f2b --- /dev/null +++ b/media/libopus/celt/arm/kiss_fft_armv4.h @@ -0,0 +1,121 @@ +/*Copyright (c) 2013, Xiph.Org Foundation and contributors. + + All rights reserved. + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions are met: + + * Redistributions of source code must retain the above copyright notice, + this list of conditions and the following disclaimer. + * Redistributions in binary form must reproduce the above copyright notice, + this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" + AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE + LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR + CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF + SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS + INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN + CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) + ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE + POSSIBILITY OF SUCH DAMAGE.*/ + +#ifndef KISS_FFT_ARMv4_H +#define KISS_FFT_ARMv4_H + +#if !defined(KISS_FFT_GUTS_H) +#error "This file should only be included from _kiss_fft_guts.h" +#endif + +#ifdef FIXED_POINT + +#undef C_MUL +#define C_MUL(m,a,b) \ + do{ \ + int br__; \ + int bi__; \ + int tt__; \ + __asm__ __volatile__( \ + "#C_MUL\n\t" \ + "ldrsh %[br], [%[bp], #0]\n\t" \ + "ldm %[ap], {r0,r1}\n\t" \ + "ldrsh %[bi], [%[bp], #2]\n\t" \ + "smull %[tt], %[mi], r1, %[br]\n\t" \ + "smlal %[tt], %[mi], r0, %[bi]\n\t" \ + "rsb %[bi], %[bi], #0\n\t" \ + "smull %[br], %[mr], r0, %[br]\n\t" \ + "mov %[tt], %[tt], lsr #15\n\t" \ + "smlal %[br], %[mr], r1, %[bi]\n\t" \ + "orr %[mi], %[tt], %[mi], lsl #17\n\t" \ + "mov %[br], %[br], lsr #15\n\t" \ + "orr %[mr], %[br], %[mr], lsl #17\n\t" \ + : [mr]"=r"((m).r), [mi]"=r"((m).i), \ + [br]"=&r"(br__), [bi]"=r"(bi__), [tt]"=r"(tt__) \ + : [ap]"r"(&(a)), [bp]"r"(&(b)) \ + : "r0", "r1" \ + ); \ + } \ + while(0) + +#undef C_MUL4 +#define C_MUL4(m,a,b) \ + do{ \ + int br__; \ + int bi__; \ + int tt__; \ + __asm__ __volatile__( \ + "#C_MUL4\n\t" \ + "ldrsh %[br], [%[bp], #0]\n\t" \ + "ldm %[ap], {r0,r1}\n\t" \ + "ldrsh %[bi], [%[bp], #2]\n\t" \ + "smull %[tt], %[mi], r1, %[br]\n\t" \ + "smlal %[tt], %[mi], r0, %[bi]\n\t" \ + "rsb %[bi], %[bi], #0\n\t" \ + "smull %[br], %[mr], r0, %[br]\n\t" \ + "mov %[tt], %[tt], lsr #17\n\t" \ + "smlal %[br], %[mr], r1, %[bi]\n\t" \ + "orr %[mi], %[tt], %[mi], lsl #15\n\t" \ + "mov %[br], %[br], lsr #17\n\t" \ + "orr %[mr], %[br], %[mr], lsl #15\n\t" \ + : [mr]"=r"((m).r), [mi]"=r"((m).i), \ + [br]"=&r"(br__), [bi]"=r"(bi__), [tt]"=r"(tt__) \ + : [ap]"r"(&(a)), [bp]"r"(&(b)) \ + : "r0", "r1" \ + ); \ + } \ + while(0) + +#undef C_MULC +#define C_MULC(m,a,b) \ + do{ \ + int br__; \ + int bi__; \ + int tt__; \ + __asm__ __volatile__( \ + "#C_MULC\n\t" \ + "ldrsh %[br], [%[bp], #0]\n\t" \ + "ldm %[ap], {r0,r1}\n\t" \ + "ldrsh %[bi], [%[bp], #2]\n\t" \ + "smull %[tt], %[mr], r0, %[br]\n\t" \ + "smlal %[tt], %[mr], r1, %[bi]\n\t" \ + "rsb %[bi], %[bi], #0\n\t" \ + "smull %[br], %[mi], r1, %[br]\n\t" \ + "mov %[tt], %[tt], lsr #15\n\t" \ + "smlal %[br], %[mi], r0, %[bi]\n\t" \ + "orr %[mr], %[tt], %[mr], lsl #17\n\t" \ + "mov %[br], %[br], lsr #15\n\t" \ + "orr %[mi], %[br], %[mi], lsl #17\n\t" \ + : [mr]"=r"((m).r), [mi]"=r"((m).i), \ + [br]"=&r"(br__), [bi]"=r"(bi__), [tt]"=r"(tt__) \ + : [ap]"r"(&(a)), [bp]"r"(&(b)) \ + : "r0", "r1" \ + ); \ + } \ + while(0) + +#endif /* FIXED_POINT */ + +#endif /* KISS_FFT_ARMv4_H */ diff --git a/media/libopus/celt/arm/kiss_fft_armv5e.h b/media/libopus/celt/arm/kiss_fft_armv5e.h new file mode 100644 index 0000000000..9eca183d77 --- /dev/null +++ b/media/libopus/celt/arm/kiss_fft_armv5e.h @@ -0,0 +1,118 @@ +/*Copyright (c) 2013, Xiph.Org Foundation and contributors. + + All rights reserved. + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions are met: + + * Redistributions of source code must retain the above copyright notice, + this list of conditions and the following disclaimer. + * Redistributions in binary form must reproduce the above copyright notice, + this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" + AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE + LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR + CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF + SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS + INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN + CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) + ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE + POSSIBILITY OF SUCH DAMAGE.*/ + +#ifndef KISS_FFT_ARMv5E_H +#define KISS_FFT_ARMv5E_H + +#if !defined(KISS_FFT_GUTS_H) +#error "This file should only be included from _kiss_fft_guts.h" +#endif + +#ifdef FIXED_POINT + +#if defined(__thumb__)||defined(__thumb2__) +#define LDRD_CONS "Q" +#else +#define LDRD_CONS "Uq" +#endif + +#undef C_MUL +#define C_MUL(m,a,b) \ + do{ \ + int mr1__; \ + int mr2__; \ + int mi__; \ + long long aval__; \ + int bval__; \ + __asm__( \ + "#C_MUL\n\t" \ + "ldrd %[aval], %H[aval], %[ap]\n\t" \ + "ldr %[bval], %[bp]\n\t" \ + "smulwb %[mi], %H[aval], %[bval]\n\t" \ + "smulwb %[mr1], %[aval], %[bval]\n\t" \ + "smulwt %[mr2], %H[aval], %[bval]\n\t" \ + "smlawt %[mi], %[aval], %[bval], %[mi]\n\t" \ + : [mr1]"=r"(mr1__), [mr2]"=r"(mr2__), [mi]"=r"(mi__), \ + [aval]"=&r"(aval__), [bval]"=r"(bval__) \ + : [ap]LDRD_CONS(a), [bp]"m"(b) \ + ); \ + (m).r = SHL32(SUB32(mr1__, mr2__), 1); \ + (m).i = SHL32(mi__, 1); \ + } \ + while(0) + +#undef C_MUL4 +#define C_MUL4(m,a,b) \ + do{ \ + int mr1__; \ + int mr2__; \ + int mi__; \ + long long aval__; \ + int bval__; \ + __asm__( \ + "#C_MUL4\n\t" \ + "ldrd %[aval], %H[aval], %[ap]\n\t" \ + "ldr %[bval], %[bp]\n\t" \ + "smulwb %[mi], %H[aval], %[bval]\n\t" \ + "smulwb %[mr1], %[aval], %[bval]\n\t" \ + "smulwt %[mr2], %H[aval], %[bval]\n\t" \ + "smlawt %[mi], %[aval], %[bval], %[mi]\n\t" \ + : [mr1]"=r"(mr1__), [mr2]"=r"(mr2__), [mi]"=r"(mi__), \ + [aval]"=&r"(aval__), [bval]"=r"(bval__) \ + : [ap]LDRD_CONS(a), [bp]"m"(b) \ + ); \ + (m).r = SHR32(SUB32(mr1__, mr2__), 1); \ + (m).i = SHR32(mi__, 1); \ + } \ + while(0) + +#undef C_MULC +#define C_MULC(m,a,b) \ + do{ \ + int mr__; \ + int mi1__; \ + int mi2__; \ + long long aval__; \ + int bval__; \ + __asm__( \ + "#C_MULC\n\t" \ + "ldrd %[aval], %H[aval], %[ap]\n\t" \ + "ldr %[bval], %[bp]\n\t" \ + "smulwb %[mr], %[aval], %[bval]\n\t" \ + "smulwb %[mi1], %H[aval], %[bval]\n\t" \ + "smulwt %[mi2], %[aval], %[bval]\n\t" \ + "smlawt %[mr], %H[aval], %[bval], %[mr]\n\t" \ + : [mr]"=r"(mr__), [mi1]"=r"(mi1__), [mi2]"=r"(mi2__), \ + [aval]"=&r"(aval__), [bval]"=r"(bval__) \ + : [ap]LDRD_CONS(a), [bp]"m"(b) \ + ); \ + (m).r = SHL32(mr__, 1); \ + (m).i = SHL32(SUB32(mi1__, mi2__), 1); \ + } \ + while(0) + +#endif /* FIXED_POINT */ + +#endif /* KISS_FFT_GUTS_H */ diff --git a/media/libopus/celt/arm/mdct_arm.h b/media/libopus/celt/arm/mdct_arm.h new file mode 100644 index 0000000000..14200bac4b --- /dev/null +++ b/media/libopus/celt/arm/mdct_arm.h @@ -0,0 +1,59 @@ +/* Copyright (c) 2015 Xiph.Org Foundation + Written by Viswanath Puttagunta */ +/** + @file arm_mdct.h + @brief ARM Neon Intrinsic optimizations for mdct using NE10 library + */ + +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#if !defined(MDCT_ARM_H) +#define MDCT_ARM_H + +#include "mdct.h" + +#if defined(HAVE_ARM_NE10) +/** Compute a forward MDCT and scale by 4/N, trashes the input array */ +void clt_mdct_forward_neon(const mdct_lookup *l, kiss_fft_scalar *in, + kiss_fft_scalar * OPUS_RESTRICT out, + const opus_val16 *window, int overlap, + int shift, int stride, int arch); + +void clt_mdct_backward_neon(const mdct_lookup *l, kiss_fft_scalar *in, + kiss_fft_scalar * OPUS_RESTRICT out, + const opus_val16 *window, int overlap, + int shift, int stride, int arch); + +#if !defined(OPUS_HAVE_RTCD) +#define OVERRIDE_OPUS_MDCT (1) +#define clt_mdct_forward(_l, _in, _out, _window, _int, _shift, _stride, _arch) \ + clt_mdct_forward_neon(_l, _in, _out, _window, _int, _shift, _stride, _arch) +#define clt_mdct_backward(_l, _in, _out, _window, _int, _shift, _stride, _arch) \ + clt_mdct_backward_neon(_l, _in, _out, _window, _int, _shift, _stride, _arch) +#endif /* OPUS_HAVE_RTCD */ +#endif /* HAVE_ARM_NE10 */ + +#endif diff --git a/media/libopus/celt/arm/pitch_arm.h b/media/libopus/celt/arm/pitch_arm.h new file mode 100644 index 0000000000..bed8b04eac --- /dev/null +++ b/media/libopus/celt/arm/pitch_arm.h @@ -0,0 +1,160 @@ +/* Copyright (c) 2010 Xiph.Org Foundation + * Copyright (c) 2013 Parrot */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#if !defined(PITCH_ARM_H) +# define PITCH_ARM_H + +# include "armcpu.h" + +# if defined(OPUS_ARM_MAY_HAVE_NEON_INTR) +opus_val32 celt_inner_prod_neon(const opus_val16 *x, const opus_val16 *y, int N); +void dual_inner_prod_neon(const opus_val16 *x, const opus_val16 *y01, + const opus_val16 *y02, int N, opus_val32 *xy1, opus_val32 *xy2); + +# if !defined(OPUS_HAVE_RTCD) && defined(OPUS_ARM_PRESUME_NEON) +# define OVERRIDE_CELT_INNER_PROD (1) +# define OVERRIDE_DUAL_INNER_PROD (1) +# define celt_inner_prod(x, y, N, arch) ((void)(arch), PRESUME_NEON(celt_inner_prod)(x, y, N)) +# define dual_inner_prod(x, y01, y02, N, xy1, xy2, arch) ((void)(arch), PRESUME_NEON(dual_inner_prod)(x, y01, y02, N, xy1, xy2)) +# endif +# endif + +# if !defined(OVERRIDE_CELT_INNER_PROD) +# if defined(OPUS_HAVE_RTCD) && (defined(OPUS_ARM_MAY_HAVE_NEON_INTR) && !defined(OPUS_ARM_PRESUME_NEON_INTR)) +extern opus_val32 (*const CELT_INNER_PROD_IMPL[OPUS_ARCHMASK+1])(const opus_val16 *x, const opus_val16 *y, int N); +# define OVERRIDE_CELT_INNER_PROD (1) +# define celt_inner_prod(x, y, N, arch) ((*CELT_INNER_PROD_IMPL[(arch)&OPUS_ARCHMASK])(x, y, N)) +# elif defined(OPUS_ARM_PRESUME_NEON_INTR) +# define OVERRIDE_CELT_INNER_PROD (1) +# define celt_inner_prod(x, y, N, arch) ((void)(arch), celt_inner_prod_neon(x, y, N)) +# endif +# endif + +# if !defined(OVERRIDE_DUAL_INNER_PROD) +# if defined(OPUS_HAVE_RTCD) && (defined(OPUS_ARM_MAY_HAVE_NEON_INTR) && !defined(OPUS_ARM_PRESUME_NEON_INTR)) +extern void (*const DUAL_INNER_PROD_IMPL[OPUS_ARCHMASK+1])(const opus_val16 *x, + const opus_val16 *y01, const opus_val16 *y02, int N, opus_val32 *xy1, opus_val32 *xy2); +# define OVERRIDE_DUAL_INNER_PROD (1) +# define dual_inner_prod(x, y01, y02, N, xy1, xy2, arch) ((*DUAL_INNER_PROD_IMPL[(arch)&OPUS_ARCHMASK])(x, y01, y02, N, xy1, xy2)) +# elif defined(OPUS_ARM_PRESUME_NEON_INTR) +# define OVERRIDE_DUAL_INNER_PROD (1) +# define dual_inner_prod(x, y01, y02, N, xy1, xy2, arch) ((void)(arch), dual_inner_prod_neon(x, y01, y02, N, xy1, xy2)) +# endif +# endif + +# if defined(FIXED_POINT) + +# if defined(OPUS_ARM_MAY_HAVE_NEON) +opus_val32 celt_pitch_xcorr_neon(const opus_val16 *_x, const opus_val16 *_y, + opus_val32 *xcorr, int len, int max_pitch, int arch); +# endif + +# if defined(OPUS_ARM_MAY_HAVE_MEDIA) +# define celt_pitch_xcorr_media MAY_HAVE_EDSP(celt_pitch_xcorr) +# endif + +# if defined(OPUS_ARM_MAY_HAVE_EDSP) +opus_val32 celt_pitch_xcorr_edsp(const opus_val16 *_x, const opus_val16 *_y, + opus_val32 *xcorr, int len, int max_pitch, int arch); +# endif + +# if defined(OPUS_HAVE_RTCD) && \ + ((defined(OPUS_ARM_MAY_HAVE_NEON) && !defined(OPUS_ARM_PRESUME_NEON)) || \ + (defined(OPUS_ARM_MAY_HAVE_MEDIA) && !defined(OPUS_ARM_PRESUME_MEDIA)) || \ + (defined(OPUS_ARM_MAY_HAVE_EDSP) && !defined(OPUS_ARM_PRESUME_EDSP))) +extern opus_val32 +(*const CELT_PITCH_XCORR_IMPL[OPUS_ARCHMASK+1])(const opus_val16 *, + const opus_val16 *, opus_val32 *, int, int, int); +# define OVERRIDE_PITCH_XCORR (1) +# define celt_pitch_xcorr(_x, _y, xcorr, len, max_pitch, arch) \ + ((*CELT_PITCH_XCORR_IMPL[(arch)&OPUS_ARCHMASK])(_x, _y, \ + xcorr, len, max_pitch, arch)) + +# elif defined(OPUS_ARM_PRESUME_EDSP) || \ + defined(OPUS_ARM_PRESUME_MEDIA) || \ + defined(OPUS_ARM_PRESUME_NEON) +# define OVERRIDE_PITCH_XCORR (1) +# define celt_pitch_xcorr (PRESUME_NEON(celt_pitch_xcorr)) + +# endif + +# if defined(OPUS_ARM_MAY_HAVE_NEON_INTR) +void xcorr_kernel_neon_fixed( + const opus_val16 *x, + const opus_val16 *y, + opus_val32 sum[4], + int len); +# endif + +# if defined(OPUS_HAVE_RTCD) && \ + (defined(OPUS_ARM_MAY_HAVE_NEON_INTR) && !defined(OPUS_ARM_PRESUME_NEON_INTR)) + +extern void (*const XCORR_KERNEL_IMPL[OPUS_ARCHMASK + 1])( + const opus_val16 *x, + const opus_val16 *y, + opus_val32 sum[4], + int len); + +# define OVERRIDE_XCORR_KERNEL (1) +# define xcorr_kernel(x, y, sum, len, arch) \ + ((*XCORR_KERNEL_IMPL[(arch) & OPUS_ARCHMASK])(x, y, sum, len)) + +# elif defined(OPUS_ARM_PRESUME_NEON_INTR) +# define OVERRIDE_XCORR_KERNEL (1) +# define xcorr_kernel(x, y, sum, len, arch) \ + ((void)arch, xcorr_kernel_neon_fixed(x, y, sum, len)) + +# endif + +#else /* Start !FIXED_POINT */ +/* Float case */ +#if defined(OPUS_ARM_MAY_HAVE_NEON_INTR) +void celt_pitch_xcorr_float_neon(const opus_val16 *_x, const opus_val16 *_y, + opus_val32 *xcorr, int len, int max_pitch, int arch); +#endif + +# if defined(OPUS_HAVE_RTCD) && \ + (defined(OPUS_ARM_MAY_HAVE_NEON_INTR) && !defined(OPUS_ARM_PRESUME_NEON_INTR)) +extern void +(*const CELT_PITCH_XCORR_IMPL[OPUS_ARCHMASK+1])(const opus_val16 *, + const opus_val16 *, opus_val32 *, int, int, int); + +# define OVERRIDE_PITCH_XCORR (1) +# define celt_pitch_xcorr(_x, _y, xcorr, len, max_pitch, arch) \ + ((*CELT_PITCH_XCORR_IMPL[(arch)&OPUS_ARCHMASK])(_x, _y, \ + xcorr, len, max_pitch, arch)) + +# elif defined(OPUS_ARM_PRESUME_NEON_INTR) + +# define OVERRIDE_PITCH_XCORR (1) +# define celt_pitch_xcorr celt_pitch_xcorr_float_neon + +# endif + +#endif /* end !FIXED_POINT */ + +#endif diff --git a/media/libopus/celt/arm/pitch_neon_intr.c b/media/libopus/celt/arm/pitch_neon_intr.c new file mode 100644 index 0000000000..35cc46e2c2 --- /dev/null +++ b/media/libopus/celt/arm/pitch_neon_intr.c @@ -0,0 +1,281 @@ +/*********************************************************************** +Copyright (c) 2017 Google Inc. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include <arm_neon.h> +#include "pitch.h" + +#ifdef FIXED_POINT + +opus_val32 celt_inner_prod_neon(const opus_val16 *x, const opus_val16 *y, int N) +{ + int i; + opus_val32 xy; + int16x8_t x_s16x8, y_s16x8; + int32x4_t xy_s32x4 = vdupq_n_s32(0); + int64x2_t xy_s64x2; + int64x1_t xy_s64x1; + + for (i = 0; i < N - 7; i += 8) { + x_s16x8 = vld1q_s16(&x[i]); + y_s16x8 = vld1q_s16(&y[i]); + xy_s32x4 = vmlal_s16(xy_s32x4, vget_low_s16 (x_s16x8), vget_low_s16 (y_s16x8)); + xy_s32x4 = vmlal_s16(xy_s32x4, vget_high_s16(x_s16x8), vget_high_s16(y_s16x8)); + } + + if (N - i >= 4) { + const int16x4_t x_s16x4 = vld1_s16(&x[i]); + const int16x4_t y_s16x4 = vld1_s16(&y[i]); + xy_s32x4 = vmlal_s16(xy_s32x4, x_s16x4, y_s16x4); + i += 4; + } + + xy_s64x2 = vpaddlq_s32(xy_s32x4); + xy_s64x1 = vadd_s64(vget_low_s64(xy_s64x2), vget_high_s64(xy_s64x2)); + xy = vget_lane_s32(vreinterpret_s32_s64(xy_s64x1), 0); + + for (; i < N; i++) { + xy = MAC16_16(xy, x[i], y[i]); + } + +#ifdef OPUS_CHECK_ASM + celt_assert(celt_inner_prod_c(x, y, N) == xy); +#endif + + return xy; +} + +void dual_inner_prod_neon(const opus_val16 *x, const opus_val16 *y01, const opus_val16 *y02, + int N, opus_val32 *xy1, opus_val32 *xy2) +{ + int i; + opus_val32 xy01, xy02; + int16x8_t x_s16x8, y01_s16x8, y02_s16x8; + int32x4_t xy01_s32x4 = vdupq_n_s32(0); + int32x4_t xy02_s32x4 = vdupq_n_s32(0); + int64x2_t xy01_s64x2, xy02_s64x2; + int64x1_t xy01_s64x1, xy02_s64x1; + + for (i = 0; i < N - 7; i += 8) { + x_s16x8 = vld1q_s16(&x[i]); + y01_s16x8 = vld1q_s16(&y01[i]); + y02_s16x8 = vld1q_s16(&y02[i]); + xy01_s32x4 = vmlal_s16(xy01_s32x4, vget_low_s16 (x_s16x8), vget_low_s16 (y01_s16x8)); + xy02_s32x4 = vmlal_s16(xy02_s32x4, vget_low_s16 (x_s16x8), vget_low_s16 (y02_s16x8)); + xy01_s32x4 = vmlal_s16(xy01_s32x4, vget_high_s16(x_s16x8), vget_high_s16(y01_s16x8)); + xy02_s32x4 = vmlal_s16(xy02_s32x4, vget_high_s16(x_s16x8), vget_high_s16(y02_s16x8)); + } + + if (N - i >= 4) { + const int16x4_t x_s16x4 = vld1_s16(&x[i]); + const int16x4_t y01_s16x4 = vld1_s16(&y01[i]); + const int16x4_t y02_s16x4 = vld1_s16(&y02[i]); + xy01_s32x4 = vmlal_s16(xy01_s32x4, x_s16x4, y01_s16x4); + xy02_s32x4 = vmlal_s16(xy02_s32x4, x_s16x4, y02_s16x4); + i += 4; + } + + xy01_s64x2 = vpaddlq_s32(xy01_s32x4); + xy02_s64x2 = vpaddlq_s32(xy02_s32x4); + xy01_s64x1 = vadd_s64(vget_low_s64(xy01_s64x2), vget_high_s64(xy01_s64x2)); + xy02_s64x1 = vadd_s64(vget_low_s64(xy02_s64x2), vget_high_s64(xy02_s64x2)); + xy01 = vget_lane_s32(vreinterpret_s32_s64(xy01_s64x1), 0); + xy02 = vget_lane_s32(vreinterpret_s32_s64(xy02_s64x1), 0); + + for (; i < N; i++) { + xy01 = MAC16_16(xy01, x[i], y01[i]); + xy02 = MAC16_16(xy02, x[i], y02[i]); + } + *xy1 = xy01; + *xy2 = xy02; + +#ifdef OPUS_CHECK_ASM + { + opus_val32 xy1_c, xy2_c; + dual_inner_prod_c(x, y01, y02, N, &xy1_c, &xy2_c); + celt_assert(xy1_c == *xy1); + celt_assert(xy2_c == *xy2); + } +#endif +} + +#else /* !FIXED_POINT */ + +/* ========================================================================== */ + +#ifdef OPUS_CHECK_ASM + +/* This part of code simulates floating-point NEON operations. */ + +/* celt_inner_prod_neon_float_c_simulation() simulates the floating-point */ +/* operations of celt_inner_prod_neon(), and both functions should have bit */ +/* exact output. */ +static opus_val32 celt_inner_prod_neon_float_c_simulation(const opus_val16 *x, const opus_val16 *y, float *err, int N) +{ + int i; + *err = 0; + opus_val32 xy, xy0 = 0, xy1 = 0, xy2 = 0, xy3 = 0; + for (i = 0; i < N - 3; i += 4) { + xy0 = MAC16_16(xy0, x[i + 0], y[i + 0]); + xy1 = MAC16_16(xy1, x[i + 1], y[i + 1]); + xy2 = MAC16_16(xy2, x[i + 2], y[i + 2]); + xy3 = MAC16_16(xy3, x[i + 3], y[i + 3]); + *err += ABS32(xy0)+ABS32(xy1)+ABS32(xy2)+ABS32(xy3); + } + xy0 += xy2; + xy1 += xy3; + xy = xy0 + xy1; + *err += ABS32(xy1)+ABS32(xy0)+ABS32(xy); + for (; i < N; i++) { + xy = MAC16_16(xy, x[i], y[i]); + *err += ABS32(xy); + } + *err = *err*2e-7 + N*1e-37; + return xy; +} + +/* dual_inner_prod_neon_float_c_simulation() simulates the floating-point */ +/* operations of dual_inner_prod_neon(), and both functions should have bit */ +/* exact output. */ +static void dual_inner_prod_neon_float_c_simulation(const opus_val16 *x, const opus_val16 *y01, const opus_val16 *y02, + int N, opus_val32 *xy1, opus_val32 *xy2, float *err) +{ + *xy1 = celt_inner_prod_neon_float_c_simulation(x, y01, &err[0], N); + *xy2 = celt_inner_prod_neon_float_c_simulation(x, y02, &err[1], N); +} + +#endif /* OPUS_CHECK_ASM */ + +/* ========================================================================== */ + +opus_val32 celt_inner_prod_neon(const opus_val16 *x, const opus_val16 *y, int N) +{ + int i; + opus_val32 xy; + float32x4_t xy_f32x4 = vdupq_n_f32(0); + float32x2_t xy_f32x2; + + for (i = 0; i < N - 7; i += 8) { + float32x4_t x_f32x4, y_f32x4; + x_f32x4 = vld1q_f32(&x[i]); + y_f32x4 = vld1q_f32(&y[i]); + xy_f32x4 = vmlaq_f32(xy_f32x4, x_f32x4, y_f32x4); + x_f32x4 = vld1q_f32(&x[i + 4]); + y_f32x4 = vld1q_f32(&y[i + 4]); + xy_f32x4 = vmlaq_f32(xy_f32x4, x_f32x4, y_f32x4); + } + + if (N - i >= 4) { + const float32x4_t x_f32x4 = vld1q_f32(&x[i]); + const float32x4_t y_f32x4 = vld1q_f32(&y[i]); + xy_f32x4 = vmlaq_f32(xy_f32x4, x_f32x4, y_f32x4); + i += 4; + } + + xy_f32x2 = vadd_f32(vget_low_f32(xy_f32x4), vget_high_f32(xy_f32x4)); + xy_f32x2 = vpadd_f32(xy_f32x2, xy_f32x2); + xy = vget_lane_f32(xy_f32x2, 0); + + for (; i < N; i++) { + xy = MAC16_16(xy, x[i], y[i]); + } + +#ifdef OPUS_CHECK_ASM + { + float err, res; + res = celt_inner_prod_neon_float_c_simulation(x, y, &err, N); + /*if (ABS32(res - xy) > err) fprintf(stderr, "%g %g %g\n", res, xy, err);*/ + celt_assert(ABS32(res - xy) <= err); + } +#endif + + return xy; +} + +void dual_inner_prod_neon(const opus_val16 *x, const opus_val16 *y01, const opus_val16 *y02, + int N, opus_val32 *xy1, opus_val32 *xy2) +{ + int i; + opus_val32 xy01, xy02; + float32x4_t xy01_f32x4 = vdupq_n_f32(0); + float32x4_t xy02_f32x4 = vdupq_n_f32(0); + float32x2_t xy01_f32x2, xy02_f32x2; + + for (i = 0; i < N - 7; i += 8) { + float32x4_t x_f32x4, y01_f32x4, y02_f32x4; + x_f32x4 = vld1q_f32(&x[i]); + y01_f32x4 = vld1q_f32(&y01[i]); + y02_f32x4 = vld1q_f32(&y02[i]); + xy01_f32x4 = vmlaq_f32(xy01_f32x4, x_f32x4, y01_f32x4); + xy02_f32x4 = vmlaq_f32(xy02_f32x4, x_f32x4, y02_f32x4); + x_f32x4 = vld1q_f32(&x[i + 4]); + y01_f32x4 = vld1q_f32(&y01[i + 4]); + y02_f32x4 = vld1q_f32(&y02[i + 4]); + xy01_f32x4 = vmlaq_f32(xy01_f32x4, x_f32x4, y01_f32x4); + xy02_f32x4 = vmlaq_f32(xy02_f32x4, x_f32x4, y02_f32x4); + } + + if (N - i >= 4) { + const float32x4_t x_f32x4 = vld1q_f32(&x[i]); + const float32x4_t y01_f32x4 = vld1q_f32(&y01[i]); + const float32x4_t y02_f32x4 = vld1q_f32(&y02[i]); + xy01_f32x4 = vmlaq_f32(xy01_f32x4, x_f32x4, y01_f32x4); + xy02_f32x4 = vmlaq_f32(xy02_f32x4, x_f32x4, y02_f32x4); + i += 4; + } + + xy01_f32x2 = vadd_f32(vget_low_f32(xy01_f32x4), vget_high_f32(xy01_f32x4)); + xy02_f32x2 = vadd_f32(vget_low_f32(xy02_f32x4), vget_high_f32(xy02_f32x4)); + xy01_f32x2 = vpadd_f32(xy01_f32x2, xy01_f32x2); + xy02_f32x2 = vpadd_f32(xy02_f32x2, xy02_f32x2); + xy01 = vget_lane_f32(xy01_f32x2, 0); + xy02 = vget_lane_f32(xy02_f32x2, 0); + + for (; i < N; i++) { + xy01 = MAC16_16(xy01, x[i], y01[i]); + xy02 = MAC16_16(xy02, x[i], y02[i]); + } + *xy1 = xy01; + *xy2 = xy02; + +#ifdef OPUS_CHECK_ASM + { + opus_val32 xy1_c, xy2_c; + float err[2]; + dual_inner_prod_neon_float_c_simulation(x, y01, y02, N, &xy1_c, &xy2_c, err); + /*if (ABS32(xy1_c - *xy1) > err[0]) fprintf(stderr, "dual1 fail: %g %g %g\n", xy1_c, *xy1, err[0]); + if (ABS32(xy2_c - *xy2) > err[1]) fprintf(stderr, "dual2 fail: %g %g %g\n", xy2_c, *xy2, err[1]);*/ + celt_assert(ABS32(xy1_c - *xy1) <= err[0]); + celt_assert(ABS32(xy2_c - *xy2) <= err[1]); + } +#endif +} + +#endif /* FIXED_POINT */ diff --git a/media/libopus/celt/bands.c b/media/libopus/celt/bands.c new file mode 100644 index 0000000000..5320ffab07 --- /dev/null +++ b/media/libopus/celt/bands.c @@ -0,0 +1,1673 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Copyright (c) 2008-2009 Gregory Maxwell + Written by Jean-Marc Valin and Gregory Maxwell */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include <math.h> +#include "bands.h" +#include "modes.h" +#include "vq.h" +#include "cwrs.h" +#include "stack_alloc.h" +#include "os_support.h" +#include "mathops.h" +#include "rate.h" +#include "quant_bands.h" +#include "pitch.h" + +int hysteresis_decision(opus_val16 val, const opus_val16 *thresholds, const opus_val16 *hysteresis, int N, int prev) +{ + int i; + for (i=0;i<N;i++) + { + if (val < thresholds[i]) + break; + } + if (i>prev && val < thresholds[prev]+hysteresis[prev]) + i=prev; + if (i<prev && val > thresholds[prev-1]-hysteresis[prev-1]) + i=prev; + return i; +} + +opus_uint32 celt_lcg_rand(opus_uint32 seed) +{ + return 1664525 * seed + 1013904223; +} + +/* This is a cos() approximation designed to be bit-exact on any platform. Bit exactness + with this approximation is important because it has an impact on the bit allocation */ +opus_int16 bitexact_cos(opus_int16 x) +{ + opus_int32 tmp; + opus_int16 x2; + tmp = (4096+((opus_int32)(x)*(x)))>>13; + celt_sig_assert(tmp<=32767); + x2 = tmp; + x2 = (32767-x2) + FRAC_MUL16(x2, (-7651 + FRAC_MUL16(x2, (8277 + FRAC_MUL16(-626, x2))))); + celt_sig_assert(x2<=32766); + return 1+x2; +} + +int bitexact_log2tan(int isin,int icos) +{ + int lc; + int ls; + lc=EC_ILOG(icos); + ls=EC_ILOG(isin); + icos<<=15-lc; + isin<<=15-ls; + return (ls-lc)*(1<<11) + +FRAC_MUL16(isin, FRAC_MUL16(isin, -2597) + 7932) + -FRAC_MUL16(icos, FRAC_MUL16(icos, -2597) + 7932); +} + +#ifdef FIXED_POINT +/* Compute the amplitude (sqrt energy) in each of the bands */ +void compute_band_energies(const CELTMode *m, const celt_sig *X, celt_ener *bandE, int end, int C, int LM, int arch) +{ + int i, c, N; + const opus_int16 *eBands = m->eBands; + (void)arch; + N = m->shortMdctSize<<LM; + c=0; do { + for (i=0;i<end;i++) + { + int j; + opus_val32 maxval=0; + opus_val32 sum = 0; + + maxval = celt_maxabs32(&X[c*N+(eBands[i]<<LM)], (eBands[i+1]-eBands[i])<<LM); + if (maxval > 0) + { + int shift = celt_ilog2(maxval) - 14 + (((m->logN[i]>>BITRES)+LM+1)>>1); + j=eBands[i]<<LM; + if (shift>0) + { + do { + sum = MAC16_16(sum, EXTRACT16(SHR32(X[j+c*N],shift)), + EXTRACT16(SHR32(X[j+c*N],shift))); + } while (++j<eBands[i+1]<<LM); + } else { + do { + sum = MAC16_16(sum, EXTRACT16(SHL32(X[j+c*N],-shift)), + EXTRACT16(SHL32(X[j+c*N],-shift))); + } while (++j<eBands[i+1]<<LM); + } + /* We're adding one here to ensure the normalized band isn't larger than unity norm */ + bandE[i+c*m->nbEBands] = EPSILON+VSHR32(EXTEND32(celt_sqrt(sum)),-shift); + } else { + bandE[i+c*m->nbEBands] = EPSILON; + } + /*printf ("%f ", bandE[i+c*m->nbEBands]);*/ + } + } while (++c<C); + /*printf ("\n");*/ +} + +/* Normalise each band such that the energy is one. */ +void normalise_bands(const CELTMode *m, const celt_sig * OPUS_RESTRICT freq, celt_norm * OPUS_RESTRICT X, const celt_ener *bandE, int end, int C, int M) +{ + int i, c, N; + const opus_int16 *eBands = m->eBands; + N = M*m->shortMdctSize; + c=0; do { + i=0; do { + opus_val16 g; + int j,shift; + opus_val16 E; + shift = celt_zlog2(bandE[i+c*m->nbEBands])-13; + E = VSHR32(bandE[i+c*m->nbEBands], shift); + g = EXTRACT16(celt_rcp(SHL32(E,3))); + j=M*eBands[i]; do { + X[j+c*N] = MULT16_16_Q15(VSHR32(freq[j+c*N],shift-1),g); + } while (++j<M*eBands[i+1]); + } while (++i<end); + } while (++c<C); +} + +#else /* FIXED_POINT */ +/* Compute the amplitude (sqrt energy) in each of the bands */ +void compute_band_energies(const CELTMode *m, const celt_sig *X, celt_ener *bandE, int end, int C, int LM, int arch) +{ + int i, c, N; + const opus_int16 *eBands = m->eBands; + N = m->shortMdctSize<<LM; + c=0; do { + for (i=0;i<end;i++) + { + opus_val32 sum; + sum = 1e-27f + celt_inner_prod(&X[c*N+(eBands[i]<<LM)], &X[c*N+(eBands[i]<<LM)], (eBands[i+1]-eBands[i])<<LM, arch); + bandE[i+c*m->nbEBands] = celt_sqrt(sum); + /*printf ("%f ", bandE[i+c*m->nbEBands]);*/ + } + } while (++c<C); + /*printf ("\n");*/ +} + +/* Normalise each band such that the energy is one. */ +void normalise_bands(const CELTMode *m, const celt_sig * OPUS_RESTRICT freq, celt_norm * OPUS_RESTRICT X, const celt_ener *bandE, int end, int C, int M) +{ + int i, c, N; + const opus_int16 *eBands = m->eBands; + N = M*m->shortMdctSize; + c=0; do { + for (i=0;i<end;i++) + { + int j; + opus_val16 g = 1.f/(1e-27f+bandE[i+c*m->nbEBands]); + for (j=M*eBands[i];j<M*eBands[i+1];j++) + X[j+c*N] = freq[j+c*N]*g; + } + } while (++c<C); +} + +#endif /* FIXED_POINT */ + +/* De-normalise the energy to produce the synthesis from the unit-energy bands */ +void denormalise_bands(const CELTMode *m, const celt_norm * OPUS_RESTRICT X, + celt_sig * OPUS_RESTRICT freq, const opus_val16 *bandLogE, int start, + int end, int M, int downsample, int silence) +{ + int i, N; + int bound; + celt_sig * OPUS_RESTRICT f; + const celt_norm * OPUS_RESTRICT x; + const opus_int16 *eBands = m->eBands; + N = M*m->shortMdctSize; + bound = M*eBands[end]; + if (downsample!=1) + bound = IMIN(bound, N/downsample); + if (silence) + { + bound = 0; + start = end = 0; + } + f = freq; + x = X+M*eBands[start]; + for (i=0;i<M*eBands[start];i++) + *f++ = 0; + for (i=start;i<end;i++) + { + int j, band_end; + opus_val16 g; + opus_val16 lg; +#ifdef FIXED_POINT + int shift; +#endif + j=M*eBands[i]; + band_end = M*eBands[i+1]; + lg = SATURATE16(ADD32(bandLogE[i], SHL32((opus_val32)eMeans[i],6))); +#ifndef FIXED_POINT + g = celt_exp2(MIN32(32.f, lg)); +#else + /* Handle the integer part of the log energy */ + shift = 16-(lg>>DB_SHIFT); + if (shift>31) + { + shift=0; + g=0; + } else { + /* Handle the fractional part. */ + g = celt_exp2_frac(lg&((1<<DB_SHIFT)-1)); + } + /* Handle extreme gains with negative shift. */ + if (shift<0) + { + /* For shift <= -2 and g > 16384 we'd be likely to overflow, so we're + capping the gain here, which is equivalent to a cap of 18 on lg. + This shouldn't trigger unless the bitstream is already corrupted. */ + if (shift <= -2) + { + g = 16384; + shift = -2; + } + do { + *f++ = SHL32(MULT16_16(*x++, g), -shift); + } while (++j<band_end); + } else +#endif + /* Be careful of the fixed-point "else" just above when changing this code */ + do { + *f++ = SHR32(MULT16_16(*x++, g), shift); + } while (++j<band_end); + } + celt_assert(start <= end); + OPUS_CLEAR(&freq[bound], N-bound); +} + +/* This prevents energy collapse for transients with multiple short MDCTs */ +void anti_collapse(const CELTMode *m, celt_norm *X_, unsigned char *collapse_masks, int LM, int C, int size, + int start, int end, const opus_val16 *logE, const opus_val16 *prev1logE, + const opus_val16 *prev2logE, const int *pulses, opus_uint32 seed, int arch) +{ + int c, i, j, k; + for (i=start;i<end;i++) + { + int N0; + opus_val16 thresh, sqrt_1; + int depth; +#ifdef FIXED_POINT + int shift; + opus_val32 thresh32; +#endif + + N0 = m->eBands[i+1]-m->eBands[i]; + /* depth in 1/8 bits */ + celt_sig_assert(pulses[i]>=0); + depth = celt_udiv(1+pulses[i], (m->eBands[i+1]-m->eBands[i]))>>LM; + +#ifdef FIXED_POINT + thresh32 = SHR32(celt_exp2(-SHL16(depth, 10-BITRES)),1); + thresh = MULT16_32_Q15(QCONST16(0.5f, 15), MIN32(32767,thresh32)); + { + opus_val32 t; + t = N0<<LM; + shift = celt_ilog2(t)>>1; + t = SHL32(t, (7-shift)<<1); + sqrt_1 = celt_rsqrt_norm(t); + } +#else + thresh = .5f*celt_exp2(-.125f*depth); + sqrt_1 = celt_rsqrt(N0<<LM); +#endif + + c=0; do + { + celt_norm *X; + opus_val16 prev1; + opus_val16 prev2; + opus_val32 Ediff; + opus_val16 r; + int renormalize=0; + prev1 = prev1logE[c*m->nbEBands+i]; + prev2 = prev2logE[c*m->nbEBands+i]; + if (C==1) + { + prev1 = MAX16(prev1,prev1logE[m->nbEBands+i]); + prev2 = MAX16(prev2,prev2logE[m->nbEBands+i]); + } + Ediff = EXTEND32(logE[c*m->nbEBands+i])-EXTEND32(MIN16(prev1,prev2)); + Ediff = MAX32(0, Ediff); + +#ifdef FIXED_POINT + if (Ediff < 16384) + { + opus_val32 r32 = SHR32(celt_exp2(-EXTRACT16(Ediff)),1); + r = 2*MIN16(16383,r32); + } else { + r = 0; + } + if (LM==3) + r = MULT16_16_Q14(23170, MIN32(23169, r)); + r = SHR16(MIN16(thresh, r),1); + r = SHR32(MULT16_16_Q15(sqrt_1, r),shift); +#else + /* r needs to be multiplied by 2 or 2*sqrt(2) depending on LM because + short blocks don't have the same energy as long */ + r = 2.f*celt_exp2(-Ediff); + if (LM==3) + r *= 1.41421356f; + r = MIN16(thresh, r); + r = r*sqrt_1; +#endif + X = X_+c*size+(m->eBands[i]<<LM); + for (k=0;k<1<<LM;k++) + { + /* Detect collapse */ + if (!(collapse_masks[i*C+c]&1<<k)) + { + /* Fill with noise */ + for (j=0;j<N0;j++) + { + seed = celt_lcg_rand(seed); + X[(j<<LM)+k] = (seed&0x8000 ? r : -r); + } + renormalize = 1; + } + } + /* We just added some energy, so we need to renormalise */ + if (renormalize) + renormalise_vector(X, N0<<LM, Q15ONE, arch); + } while (++c<C); + } +} + +/* Compute the weights to use for optimizing normalized distortion across + channels. We use the amplitude to weight square distortion, which means + that we use the square root of the value we would have been using if we + wanted to minimize the MSE in the non-normalized domain. This roughly + corresponds to some quick-and-dirty perceptual experiments I ran to + measure inter-aural masking (there doesn't seem to be any published data + on the topic). */ +static void compute_channel_weights(celt_ener Ex, celt_ener Ey, opus_val16 w[2]) +{ + celt_ener minE; +#ifdef FIXED_POINT + int shift; +#endif + minE = MIN32(Ex, Ey); + /* Adjustment to make the weights a bit more conservative. */ + Ex = ADD32(Ex, minE/3); + Ey = ADD32(Ey, minE/3); +#ifdef FIXED_POINT + shift = celt_ilog2(EPSILON+MAX32(Ex, Ey))-14; +#endif + w[0] = VSHR32(Ex, shift); + w[1] = VSHR32(Ey, shift); +} + +static void intensity_stereo(const CELTMode *m, celt_norm * OPUS_RESTRICT X, const celt_norm * OPUS_RESTRICT Y, const celt_ener *bandE, int bandID, int N) +{ + int i = bandID; + int j; + opus_val16 a1, a2; + opus_val16 left, right; + opus_val16 norm; +#ifdef FIXED_POINT + int shift = celt_zlog2(MAX32(bandE[i], bandE[i+m->nbEBands]))-13; +#endif + left = VSHR32(bandE[i],shift); + right = VSHR32(bandE[i+m->nbEBands],shift); + norm = EPSILON + celt_sqrt(EPSILON+MULT16_16(left,left)+MULT16_16(right,right)); + a1 = DIV32_16(SHL32(EXTEND32(left),14),norm); + a2 = DIV32_16(SHL32(EXTEND32(right),14),norm); + for (j=0;j<N;j++) + { + celt_norm r, l; + l = X[j]; + r = Y[j]; + X[j] = EXTRACT16(SHR32(MAC16_16(MULT16_16(a1, l), a2, r), 14)); + /* Side is not encoded, no need to calculate */ + } +} + +static void stereo_split(celt_norm * OPUS_RESTRICT X, celt_norm * OPUS_RESTRICT Y, int N) +{ + int j; + for (j=0;j<N;j++) + { + opus_val32 r, l; + l = MULT16_16(QCONST16(.70710678f, 15), X[j]); + r = MULT16_16(QCONST16(.70710678f, 15), Y[j]); + X[j] = EXTRACT16(SHR32(ADD32(l, r), 15)); + Y[j] = EXTRACT16(SHR32(SUB32(r, l), 15)); + } +} + +static void stereo_merge(celt_norm * OPUS_RESTRICT X, celt_norm * OPUS_RESTRICT Y, opus_val16 mid, int N, int arch) +{ + int j; + opus_val32 xp=0, side=0; + opus_val32 El, Er; + opus_val16 mid2; +#ifdef FIXED_POINT + int kl, kr; +#endif + opus_val32 t, lgain, rgain; + + /* Compute the norm of X+Y and X-Y as |X|^2 + |Y|^2 +/- sum(xy) */ + dual_inner_prod(Y, X, Y, N, &xp, &side, arch); + /* Compensating for the mid normalization */ + xp = MULT16_32_Q15(mid, xp); + /* mid and side are in Q15, not Q14 like X and Y */ + mid2 = SHR16(mid, 1); + El = MULT16_16(mid2, mid2) + side - 2*xp; + Er = MULT16_16(mid2, mid2) + side + 2*xp; + if (Er < QCONST32(6e-4f, 28) || El < QCONST32(6e-4f, 28)) + { + OPUS_COPY(Y, X, N); + return; + } + +#ifdef FIXED_POINT + kl = celt_ilog2(El)>>1; + kr = celt_ilog2(Er)>>1; +#endif + t = VSHR32(El, (kl-7)<<1); + lgain = celt_rsqrt_norm(t); + t = VSHR32(Er, (kr-7)<<1); + rgain = celt_rsqrt_norm(t); + +#ifdef FIXED_POINT + if (kl < 7) + kl = 7; + if (kr < 7) + kr = 7; +#endif + + for (j=0;j<N;j++) + { + celt_norm r, l; + /* Apply mid scaling (side is already scaled) */ + l = MULT16_16_P15(mid, X[j]); + r = Y[j]; + X[j] = EXTRACT16(PSHR32(MULT16_16(lgain, SUB16(l,r)), kl+1)); + Y[j] = EXTRACT16(PSHR32(MULT16_16(rgain, ADD16(l,r)), kr+1)); + } +} + +/* Decide whether we should spread the pulses in the current frame */ +int spreading_decision(const CELTMode *m, const celt_norm *X, int *average, + int last_decision, int *hf_average, int *tapset_decision, int update_hf, + int end, int C, int M, const int *spread_weight) +{ + int i, c, N0; + int sum = 0, nbBands=0; + const opus_int16 * OPUS_RESTRICT eBands = m->eBands; + int decision; + int hf_sum=0; + + celt_assert(end>0); + + N0 = M*m->shortMdctSize; + + if (M*(eBands[end]-eBands[end-1]) <= 8) + return SPREAD_NONE; + c=0; do { + for (i=0;i<end;i++) + { + int j, N, tmp=0; + int tcount[3] = {0,0,0}; + const celt_norm * OPUS_RESTRICT x = X+M*eBands[i]+c*N0; + N = M*(eBands[i+1]-eBands[i]); + if (N<=8) + continue; + /* Compute rough CDF of |x[j]| */ + for (j=0;j<N;j++) + { + opus_val32 x2N; /* Q13 */ + + x2N = MULT16_16(MULT16_16_Q15(x[j], x[j]), N); + if (x2N < QCONST16(0.25f,13)) + tcount[0]++; + if (x2N < QCONST16(0.0625f,13)) + tcount[1]++; + if (x2N < QCONST16(0.015625f,13)) + tcount[2]++; + } + + /* Only include four last bands (8 kHz and up) */ + if (i>m->nbEBands-4) + hf_sum += celt_udiv(32*(tcount[1]+tcount[0]), N); + tmp = (2*tcount[2] >= N) + (2*tcount[1] >= N) + (2*tcount[0] >= N); + sum += tmp*spread_weight[i]; + nbBands+=spread_weight[i]; + } + } while (++c<C); + + if (update_hf) + { + if (hf_sum) + hf_sum = celt_udiv(hf_sum, C*(4-m->nbEBands+end)); + *hf_average = (*hf_average+hf_sum)>>1; + hf_sum = *hf_average; + if (*tapset_decision==2) + hf_sum += 4; + else if (*tapset_decision==0) + hf_sum -= 4; + if (hf_sum > 22) + *tapset_decision=2; + else if (hf_sum > 18) + *tapset_decision=1; + else + *tapset_decision=0; + } + /*printf("%d %d %d\n", hf_sum, *hf_average, *tapset_decision);*/ + celt_assert(nbBands>0); /* end has to be non-zero */ + celt_assert(sum>=0); + sum = celt_udiv((opus_int32)sum<<8, nbBands); + /* Recursive averaging */ + sum = (sum+*average)>>1; + *average = sum; + /* Hysteresis */ + sum = (3*sum + (((3-last_decision)<<7) + 64) + 2)>>2; + if (sum < 80) + { + decision = SPREAD_AGGRESSIVE; + } else if (sum < 256) + { + decision = SPREAD_NORMAL; + } else if (sum < 384) + { + decision = SPREAD_LIGHT; + } else { + decision = SPREAD_NONE; + } +#ifdef FUZZING + decision = rand()&0x3; + *tapset_decision=rand()%3; +#endif + return decision; +} + +/* Indexing table for converting from natural Hadamard to ordery Hadamard + This is essentially a bit-reversed Gray, on top of which we've added + an inversion of the order because we want the DC at the end rather than + the beginning. The lines are for N=2, 4, 8, 16 */ +static const int ordery_table[] = { + 1, 0, + 3, 0, 2, 1, + 7, 0, 4, 3, 6, 1, 5, 2, + 15, 0, 8, 7, 12, 3, 11, 4, 14, 1, 9, 6, 13, 2, 10, 5, +}; + +static void deinterleave_hadamard(celt_norm *X, int N0, int stride, int hadamard) +{ + int i,j; + VARDECL(celt_norm, tmp); + int N; + SAVE_STACK; + N = N0*stride; + ALLOC(tmp, N, celt_norm); + celt_assert(stride>0); + if (hadamard) + { + const int *ordery = ordery_table+stride-2; + for (i=0;i<stride;i++) + { + for (j=0;j<N0;j++) + tmp[ordery[i]*N0+j] = X[j*stride+i]; + } + } else { + for (i=0;i<stride;i++) + for (j=0;j<N0;j++) + tmp[i*N0+j] = X[j*stride+i]; + } + OPUS_COPY(X, tmp, N); + RESTORE_STACK; +} + +static void interleave_hadamard(celt_norm *X, int N0, int stride, int hadamard) +{ + int i,j; + VARDECL(celt_norm, tmp); + int N; + SAVE_STACK; + N = N0*stride; + ALLOC(tmp, N, celt_norm); + if (hadamard) + { + const int *ordery = ordery_table+stride-2; + for (i=0;i<stride;i++) + for (j=0;j<N0;j++) + tmp[j*stride+i] = X[ordery[i]*N0+j]; + } else { + for (i=0;i<stride;i++) + for (j=0;j<N0;j++) + tmp[j*stride+i] = X[i*N0+j]; + } + OPUS_COPY(X, tmp, N); + RESTORE_STACK; +} + +void haar1(celt_norm *X, int N0, int stride) +{ + int i, j; + N0 >>= 1; + for (i=0;i<stride;i++) + for (j=0;j<N0;j++) + { + opus_val32 tmp1, tmp2; + tmp1 = MULT16_16(QCONST16(.70710678f,15), X[stride*2*j+i]); + tmp2 = MULT16_16(QCONST16(.70710678f,15), X[stride*(2*j+1)+i]); + X[stride*2*j+i] = EXTRACT16(PSHR32(ADD32(tmp1, tmp2), 15)); + X[stride*(2*j+1)+i] = EXTRACT16(PSHR32(SUB32(tmp1, tmp2), 15)); + } +} + +static int compute_qn(int N, int b, int offset, int pulse_cap, int stereo) +{ + static const opus_int16 exp2_table8[8] = + {16384, 17866, 19483, 21247, 23170, 25267, 27554, 30048}; + int qn, qb; + int N2 = 2*N-1; + if (stereo && N==2) + N2--; + /* The upper limit ensures that in a stereo split with itheta==16384, we'll + always have enough bits left over to code at least one pulse in the + side; otherwise it would collapse, since it doesn't get folded. */ + qb = celt_sudiv(b+N2*offset, N2); + qb = IMIN(b-pulse_cap-(4<<BITRES), qb); + + qb = IMIN(8<<BITRES, qb); + + if (qb<(1<<BITRES>>1)) { + qn = 1; + } else { + qn = exp2_table8[qb&0x7]>>(14-(qb>>BITRES)); + qn = (qn+1)>>1<<1; + } + celt_assert(qn <= 256); + return qn; +} + +struct band_ctx { + int encode; + int resynth; + const CELTMode *m; + int i; + int intensity; + int spread; + int tf_change; + ec_ctx *ec; + opus_int32 remaining_bits; + const celt_ener *bandE; + opus_uint32 seed; + int arch; + int theta_round; + int disable_inv; + int avoid_split_noise; +}; + +struct split_ctx { + int inv; + int imid; + int iside; + int delta; + int itheta; + int qalloc; +}; + +static void compute_theta(struct band_ctx *ctx, struct split_ctx *sctx, + celt_norm *X, celt_norm *Y, int N, int *b, int B, int B0, + int LM, + int stereo, int *fill) +{ + int qn; + int itheta=0; + int delta; + int imid, iside; + int qalloc; + int pulse_cap; + int offset; + opus_int32 tell; + int inv=0; + int encode; + const CELTMode *m; + int i; + int intensity; + ec_ctx *ec; + const celt_ener *bandE; + + encode = ctx->encode; + m = ctx->m; + i = ctx->i; + intensity = ctx->intensity; + ec = ctx->ec; + bandE = ctx->bandE; + + /* Decide on the resolution to give to the split parameter theta */ + pulse_cap = m->logN[i]+LM*(1<<BITRES); + offset = (pulse_cap>>1) - (stereo&&N==2 ? QTHETA_OFFSET_TWOPHASE : QTHETA_OFFSET); + qn = compute_qn(N, *b, offset, pulse_cap, stereo); + if (stereo && i>=intensity) + qn = 1; + if (encode) + { + /* theta is the atan() of the ratio between the (normalized) + side and mid. With just that parameter, we can re-scale both + mid and side because we know that 1) they have unit norm and + 2) they are orthogonal. */ + itheta = stereo_itheta(X, Y, stereo, N, ctx->arch); + } + tell = ec_tell_frac(ec); + if (qn!=1) + { + if (encode) + { + if (!stereo || ctx->theta_round == 0) + { + itheta = (itheta*(opus_int32)qn+8192)>>14; + if (!stereo && ctx->avoid_split_noise && itheta > 0 && itheta < qn) + { + /* Check if the selected value of theta will cause the bit allocation + to inject noise on one side. If so, make sure the energy of that side + is zero. */ + int unquantized = celt_udiv((opus_int32)itheta*16384, qn); + imid = bitexact_cos((opus_int16)unquantized); + iside = bitexact_cos((opus_int16)(16384-unquantized)); + delta = FRAC_MUL16((N-1)<<7,bitexact_log2tan(iside,imid)); + if (delta > *b) + itheta = qn; + else if (delta < -*b) + itheta = 0; + } + } else { + int down; + /* Bias quantization towards itheta=0 and itheta=16384. */ + int bias = itheta > 8192 ? 32767/qn : -32767/qn; + down = IMIN(qn-1, IMAX(0, (itheta*(opus_int32)qn + bias)>>14)); + if (ctx->theta_round < 0) + itheta = down; + else + itheta = down+1; + } + } + /* Entropy coding of the angle. We use a uniform pdf for the + time split, a step for stereo, and a triangular one for the rest. */ + if (stereo && N>2) + { + int p0 = 3; + int x = itheta; + int x0 = qn/2; + int ft = p0*(x0+1) + x0; + /* Use a probability of p0 up to itheta=8192 and then use 1 after */ + if (encode) + { + ec_encode(ec,x<=x0?p0*x:(x-1-x0)+(x0+1)*p0,x<=x0?p0*(x+1):(x-x0)+(x0+1)*p0,ft); + } else { + int fs; + fs=ec_decode(ec,ft); + if (fs<(x0+1)*p0) + x=fs/p0; + else + x=x0+1+(fs-(x0+1)*p0); + ec_dec_update(ec,x<=x0?p0*x:(x-1-x0)+(x0+1)*p0,x<=x0?p0*(x+1):(x-x0)+(x0+1)*p0,ft); + itheta = x; + } + } else if (B0>1 || stereo) { + /* Uniform pdf */ + if (encode) + ec_enc_uint(ec, itheta, qn+1); + else + itheta = ec_dec_uint(ec, qn+1); + } else { + int fs=1, ft; + ft = ((qn>>1)+1)*((qn>>1)+1); + if (encode) + { + int fl; + + fs = itheta <= (qn>>1) ? itheta + 1 : qn + 1 - itheta; + fl = itheta <= (qn>>1) ? itheta*(itheta + 1)>>1 : + ft - ((qn + 1 - itheta)*(qn + 2 - itheta)>>1); + + ec_encode(ec, fl, fl+fs, ft); + } else { + /* Triangular pdf */ + int fl=0; + int fm; + fm = ec_decode(ec, ft); + + if (fm < ((qn>>1)*((qn>>1) + 1)>>1)) + { + itheta = (isqrt32(8*(opus_uint32)fm + 1) - 1)>>1; + fs = itheta + 1; + fl = itheta*(itheta + 1)>>1; + } + else + { + itheta = (2*(qn + 1) + - isqrt32(8*(opus_uint32)(ft - fm - 1) + 1))>>1; + fs = qn + 1 - itheta; + fl = ft - ((qn + 1 - itheta)*(qn + 2 - itheta)>>1); + } + + ec_dec_update(ec, fl, fl+fs, ft); + } + } + celt_assert(itheta>=0); + itheta = celt_udiv((opus_int32)itheta*16384, qn); + if (encode && stereo) + { + if (itheta==0) + intensity_stereo(m, X, Y, bandE, i, N); + else + stereo_split(X, Y, N); + } + /* NOTE: Renormalising X and Y *may* help fixed-point a bit at very high rate. + Let's do that at higher complexity */ + } else if (stereo) { + if (encode) + { + inv = itheta > 8192 && !ctx->disable_inv; + if (inv) + { + int j; + for (j=0;j<N;j++) + Y[j] = -Y[j]; + } + intensity_stereo(m, X, Y, bandE, i, N); + } + if (*b>2<<BITRES && ctx->remaining_bits > 2<<BITRES) + { + if (encode) + ec_enc_bit_logp(ec, inv, 2); + else + inv = ec_dec_bit_logp(ec, 2); + } else + inv = 0; + /* inv flag override to avoid problems with downmixing. */ + if (ctx->disable_inv) + inv = 0; + itheta = 0; + } + qalloc = ec_tell_frac(ec) - tell; + *b -= qalloc; + + if (itheta == 0) + { + imid = 32767; + iside = 0; + *fill &= (1<<B)-1; + delta = -16384; + } else if (itheta == 16384) + { + imid = 0; + iside = 32767; + *fill &= ((1<<B)-1)<<B; + delta = 16384; + } else { + imid = bitexact_cos((opus_int16)itheta); + iside = bitexact_cos((opus_int16)(16384-itheta)); + /* This is the mid vs side allocation that minimizes squared error + in that band. */ + delta = FRAC_MUL16((N-1)<<7,bitexact_log2tan(iside,imid)); + } + + sctx->inv = inv; + sctx->imid = imid; + sctx->iside = iside; + sctx->delta = delta; + sctx->itheta = itheta; + sctx->qalloc = qalloc; +} +static unsigned quant_band_n1(struct band_ctx *ctx, celt_norm *X, celt_norm *Y, + celt_norm *lowband_out) +{ + int c; + int stereo; + celt_norm *x = X; + int encode; + ec_ctx *ec; + + encode = ctx->encode; + ec = ctx->ec; + + stereo = Y != NULL; + c=0; do { + int sign=0; + if (ctx->remaining_bits>=1<<BITRES) + { + if (encode) + { + sign = x[0]<0; + ec_enc_bits(ec, sign, 1); + } else { + sign = ec_dec_bits(ec, 1); + } + ctx->remaining_bits -= 1<<BITRES; + } + if (ctx->resynth) + x[0] = sign ? -NORM_SCALING : NORM_SCALING; + x = Y; + } while (++c<1+stereo); + if (lowband_out) + lowband_out[0] = SHR16(X[0],4); + return 1; +} + +/* This function is responsible for encoding and decoding a mono partition. + It can split the band in two and transmit the energy difference with + the two half-bands. It can be called recursively so bands can end up being + split in 8 parts. */ +static unsigned quant_partition(struct band_ctx *ctx, celt_norm *X, + int N, int b, int B, celt_norm *lowband, + int LM, + opus_val16 gain, int fill) +{ + const unsigned char *cache; + int q; + int curr_bits; + int imid=0, iside=0; + int B0=B; + opus_val16 mid=0, side=0; + unsigned cm=0; + celt_norm *Y=NULL; + int encode; + const CELTMode *m; + int i; + int spread; + ec_ctx *ec; + + encode = ctx->encode; + m = ctx->m; + i = ctx->i; + spread = ctx->spread; + ec = ctx->ec; + + /* If we need 1.5 more bit than we can produce, split the band in two. */ + cache = m->cache.bits + m->cache.index[(LM+1)*m->nbEBands+i]; + if (LM != -1 && b > cache[cache[0]]+12 && N>2) + { + int mbits, sbits, delta; + int itheta; + int qalloc; + struct split_ctx sctx; + celt_norm *next_lowband2=NULL; + opus_int32 rebalance; + + N >>= 1; + Y = X+N; + LM -= 1; + if (B==1) + fill = (fill&1)|(fill<<1); + B = (B+1)>>1; + + compute_theta(ctx, &sctx, X, Y, N, &b, B, B0, LM, 0, &fill); + imid = sctx.imid; + iside = sctx.iside; + delta = sctx.delta; + itheta = sctx.itheta; + qalloc = sctx.qalloc; +#ifdef FIXED_POINT + mid = imid; + side = iside; +#else + mid = (1.f/32768)*imid; + side = (1.f/32768)*iside; +#endif + + /* Give more bits to low-energy MDCTs than they would otherwise deserve */ + if (B0>1 && (itheta&0x3fff)) + { + if (itheta > 8192) + /* Rough approximation for pre-echo masking */ + delta -= delta>>(4-LM); + else + /* Corresponds to a forward-masking slope of 1.5 dB per 10 ms */ + delta = IMIN(0, delta + (N<<BITRES>>(5-LM))); + } + mbits = IMAX(0, IMIN(b, (b-delta)/2)); + sbits = b-mbits; + ctx->remaining_bits -= qalloc; + + if (lowband) + next_lowband2 = lowband+N; /* >32-bit split case */ + + rebalance = ctx->remaining_bits; + if (mbits >= sbits) + { + cm = quant_partition(ctx, X, N, mbits, B, lowband, LM, + MULT16_16_P15(gain,mid), fill); + rebalance = mbits - (rebalance-ctx->remaining_bits); + if (rebalance > 3<<BITRES && itheta!=0) + sbits += rebalance - (3<<BITRES); + cm |= quant_partition(ctx, Y, N, sbits, B, next_lowband2, LM, + MULT16_16_P15(gain,side), fill>>B)<<(B0>>1); + } else { + cm = quant_partition(ctx, Y, N, sbits, B, next_lowband2, LM, + MULT16_16_P15(gain,side), fill>>B)<<(B0>>1); + rebalance = sbits - (rebalance-ctx->remaining_bits); + if (rebalance > 3<<BITRES && itheta!=16384) + mbits += rebalance - (3<<BITRES); + cm |= quant_partition(ctx, X, N, mbits, B, lowband, LM, + MULT16_16_P15(gain,mid), fill); + } + } else { + /* This is the basic no-split case */ + q = bits2pulses(m, i, LM, b); + curr_bits = pulses2bits(m, i, LM, q); + ctx->remaining_bits -= curr_bits; + + /* Ensures we can never bust the budget */ + while (ctx->remaining_bits < 0 && q > 0) + { + ctx->remaining_bits += curr_bits; + q--; + curr_bits = pulses2bits(m, i, LM, q); + ctx->remaining_bits -= curr_bits; + } + + if (q!=0) + { + int K = get_pulses(q); + + /* Finally do the actual quantization */ + if (encode) + { + cm = alg_quant(X, N, K, spread, B, ec, gain, ctx->resynth, ctx->arch); + } else { + cm = alg_unquant(X, N, K, spread, B, ec, gain); + } + } else { + /* If there's no pulse, fill the band anyway */ + int j; + if (ctx->resynth) + { + unsigned cm_mask; + /* B can be as large as 16, so this shift might overflow an int on a + 16-bit platform; use a long to get defined behavior.*/ + cm_mask = (unsigned)(1UL<<B)-1; + fill &= cm_mask; + if (!fill) + { + OPUS_CLEAR(X, N); + } else { + if (lowband == NULL) + { + /* Noise */ + for (j=0;j<N;j++) + { + ctx->seed = celt_lcg_rand(ctx->seed); + X[j] = (celt_norm)((opus_int32)ctx->seed>>20); + } + cm = cm_mask; + } else { + /* Folded spectrum */ + for (j=0;j<N;j++) + { + opus_val16 tmp; + ctx->seed = celt_lcg_rand(ctx->seed); + /* About 48 dB below the "normal" folding level */ + tmp = QCONST16(1.0f/256, 10); + tmp = (ctx->seed)&0x8000 ? tmp : -tmp; + X[j] = lowband[j]+tmp; + } + cm = fill; + } + renormalise_vector(X, N, gain, ctx->arch); + } + } + } + } + + return cm; +} + + +/* This function is responsible for encoding and decoding a band for the mono case. */ +static unsigned quant_band(struct band_ctx *ctx, celt_norm *X, + int N, int b, int B, celt_norm *lowband, + int LM, celt_norm *lowband_out, + opus_val16 gain, celt_norm *lowband_scratch, int fill) +{ + int N0=N; + int N_B=N; + int N_B0; + int B0=B; + int time_divide=0; + int recombine=0; + int longBlocks; + unsigned cm=0; + int k; + int encode; + int tf_change; + + encode = ctx->encode; + tf_change = ctx->tf_change; + + longBlocks = B0==1; + + N_B = celt_udiv(N_B, B); + + /* Special case for one sample */ + if (N==1) + { + return quant_band_n1(ctx, X, NULL, lowband_out); + } + + if (tf_change>0) + recombine = tf_change; + /* Band recombining to increase frequency resolution */ + + if (lowband_scratch && lowband && (recombine || ((N_B&1) == 0 && tf_change<0) || B0>1)) + { + OPUS_COPY(lowband_scratch, lowband, N); + lowband = lowband_scratch; + } + + for (k=0;k<recombine;k++) + { + static const unsigned char bit_interleave_table[16]={ + 0,1,1,1,2,3,3,3,2,3,3,3,2,3,3,3 + }; + if (encode) + haar1(X, N>>k, 1<<k); + if (lowband) + haar1(lowband, N>>k, 1<<k); + fill = bit_interleave_table[fill&0xF]|bit_interleave_table[fill>>4]<<2; + } + B>>=recombine; + N_B<<=recombine; + + /* Increasing the time resolution */ + while ((N_B&1) == 0 && tf_change<0) + { + if (encode) + haar1(X, N_B, B); + if (lowband) + haar1(lowband, N_B, B); + fill |= fill<<B; + B <<= 1; + N_B >>= 1; + time_divide++; + tf_change++; + } + B0=B; + N_B0 = N_B; + + /* Reorganize the samples in time order instead of frequency order */ + if (B0>1) + { + if (encode) + deinterleave_hadamard(X, N_B>>recombine, B0<<recombine, longBlocks); + if (lowband) + deinterleave_hadamard(lowband, N_B>>recombine, B0<<recombine, longBlocks); + } + + cm = quant_partition(ctx, X, N, b, B, lowband, LM, gain, fill); + + /* This code is used by the decoder and by the resynthesis-enabled encoder */ + if (ctx->resynth) + { + /* Undo the sample reorganization going from time order to frequency order */ + if (B0>1) + interleave_hadamard(X, N_B>>recombine, B0<<recombine, longBlocks); + + /* Undo time-freq changes that we did earlier */ + N_B = N_B0; + B = B0; + for (k=0;k<time_divide;k++) + { + B >>= 1; + N_B <<= 1; + cm |= cm>>B; + haar1(X, N_B, B); + } + + for (k=0;k<recombine;k++) + { + static const unsigned char bit_deinterleave_table[16]={ + 0x00,0x03,0x0C,0x0F,0x30,0x33,0x3C,0x3F, + 0xC0,0xC3,0xCC,0xCF,0xF0,0xF3,0xFC,0xFF + }; + cm = bit_deinterleave_table[cm]; + haar1(X, N0>>k, 1<<k); + } + B<<=recombine; + + /* Scale output for later folding */ + if (lowband_out) + { + int j; + opus_val16 n; + n = celt_sqrt(SHL32(EXTEND32(N0),22)); + for (j=0;j<N0;j++) + lowband_out[j] = MULT16_16_Q15(n,X[j]); + } + cm &= (1<<B)-1; + } + return cm; +} + + +/* This function is responsible for encoding and decoding a band for the stereo case. */ +static unsigned quant_band_stereo(struct band_ctx *ctx, celt_norm *X, celt_norm *Y, + int N, int b, int B, celt_norm *lowband, + int LM, celt_norm *lowband_out, + celt_norm *lowband_scratch, int fill) +{ + int imid=0, iside=0; + int inv = 0; + opus_val16 mid=0, side=0; + unsigned cm=0; + int mbits, sbits, delta; + int itheta; + int qalloc; + struct split_ctx sctx; + int orig_fill; + int encode; + ec_ctx *ec; + + encode = ctx->encode; + ec = ctx->ec; + + /* Special case for one sample */ + if (N==1) + { + return quant_band_n1(ctx, X, Y, lowband_out); + } + + orig_fill = fill; + + compute_theta(ctx, &sctx, X, Y, N, &b, B, B, LM, 1, &fill); + inv = sctx.inv; + imid = sctx.imid; + iside = sctx.iside; + delta = sctx.delta; + itheta = sctx.itheta; + qalloc = sctx.qalloc; +#ifdef FIXED_POINT + mid = imid; + side = iside; +#else + mid = (1.f/32768)*imid; + side = (1.f/32768)*iside; +#endif + + /* This is a special case for N=2 that only works for stereo and takes + advantage of the fact that mid and side are orthogonal to encode + the side with just one bit. */ + if (N==2) + { + int c; + int sign=0; + celt_norm *x2, *y2; + mbits = b; + sbits = 0; + /* Only need one bit for the side. */ + if (itheta != 0 && itheta != 16384) + sbits = 1<<BITRES; + mbits -= sbits; + c = itheta > 8192; + ctx->remaining_bits -= qalloc+sbits; + + x2 = c ? Y : X; + y2 = c ? X : Y; + if (sbits) + { + if (encode) + { + /* Here we only need to encode a sign for the side. */ + sign = x2[0]*y2[1] - x2[1]*y2[0] < 0; + ec_enc_bits(ec, sign, 1); + } else { + sign = ec_dec_bits(ec, 1); + } + } + sign = 1-2*sign; + /* We use orig_fill here because we want to fold the side, but if + itheta==16384, we'll have cleared the low bits of fill. */ + cm = quant_band(ctx, x2, N, mbits, B, lowband, LM, lowband_out, Q15ONE, + lowband_scratch, orig_fill); + /* We don't split N=2 bands, so cm is either 1 or 0 (for a fold-collapse), + and there's no need to worry about mixing with the other channel. */ + y2[0] = -sign*x2[1]; + y2[1] = sign*x2[0]; + if (ctx->resynth) + { + celt_norm tmp; + X[0] = MULT16_16_Q15(mid, X[0]); + X[1] = MULT16_16_Q15(mid, X[1]); + Y[0] = MULT16_16_Q15(side, Y[0]); + Y[1] = MULT16_16_Q15(side, Y[1]); + tmp = X[0]; + X[0] = SUB16(tmp,Y[0]); + Y[0] = ADD16(tmp,Y[0]); + tmp = X[1]; + X[1] = SUB16(tmp,Y[1]); + Y[1] = ADD16(tmp,Y[1]); + } + } else { + /* "Normal" split code */ + opus_int32 rebalance; + + mbits = IMAX(0, IMIN(b, (b-delta)/2)); + sbits = b-mbits; + ctx->remaining_bits -= qalloc; + + rebalance = ctx->remaining_bits; + if (mbits >= sbits) + { + /* In stereo mode, we do not apply a scaling to the mid because we need the normalized + mid for folding later. */ + cm = quant_band(ctx, X, N, mbits, B, lowband, LM, lowband_out, Q15ONE, + lowband_scratch, fill); + rebalance = mbits - (rebalance-ctx->remaining_bits); + if (rebalance > 3<<BITRES && itheta!=0) + sbits += rebalance - (3<<BITRES); + + /* For a stereo split, the high bits of fill are always zero, so no + folding will be done to the side. */ + cm |= quant_band(ctx, Y, N, sbits, B, NULL, LM, NULL, side, NULL, fill>>B); + } else { + /* For a stereo split, the high bits of fill are always zero, so no + folding will be done to the side. */ + cm = quant_band(ctx, Y, N, sbits, B, NULL, LM, NULL, side, NULL, fill>>B); + rebalance = sbits - (rebalance-ctx->remaining_bits); + if (rebalance > 3<<BITRES && itheta!=16384) + mbits += rebalance - (3<<BITRES); + /* In stereo mode, we do not apply a scaling to the mid because we need the normalized + mid for folding later. */ + cm |= quant_band(ctx, X, N, mbits, B, lowband, LM, lowband_out, Q15ONE, + lowband_scratch, fill); + } + } + + + /* This code is used by the decoder and by the resynthesis-enabled encoder */ + if (ctx->resynth) + { + if (N!=2) + stereo_merge(X, Y, mid, N, ctx->arch); + if (inv) + { + int j; + for (j=0;j<N;j++) + Y[j] = -Y[j]; + } + } + return cm; +} + +#ifndef DISABLE_UPDATE_DRAFT +static void special_hybrid_folding(const CELTMode *m, celt_norm *norm, celt_norm *norm2, int start, int M, int dual_stereo) +{ + int n1, n2; + const opus_int16 * OPUS_RESTRICT eBands = m->eBands; + n1 = M*(eBands[start+1]-eBands[start]); + n2 = M*(eBands[start+2]-eBands[start+1]); + /* Duplicate enough of the first band folding data to be able to fold the second band. + Copies no data for CELT-only mode. */ + OPUS_COPY(&norm[n1], &norm[2*n1 - n2], n2-n1); + if (dual_stereo) + OPUS_COPY(&norm2[n1], &norm2[2*n1 - n2], n2-n1); +} +#endif + +void quant_all_bands(int encode, const CELTMode *m, int start, int end, + celt_norm *X_, celt_norm *Y_, unsigned char *collapse_masks, + const celt_ener *bandE, int *pulses, int shortBlocks, int spread, + int dual_stereo, int intensity, int *tf_res, opus_int32 total_bits, + opus_int32 balance, ec_ctx *ec, int LM, int codedBands, + opus_uint32 *seed, int complexity, int arch, int disable_inv) +{ + int i; + opus_int32 remaining_bits; + const opus_int16 * OPUS_RESTRICT eBands = m->eBands; + celt_norm * OPUS_RESTRICT norm, * OPUS_RESTRICT norm2; + VARDECL(celt_norm, _norm); + VARDECL(celt_norm, _lowband_scratch); + VARDECL(celt_norm, X_save); + VARDECL(celt_norm, Y_save); + VARDECL(celt_norm, X_save2); + VARDECL(celt_norm, Y_save2); + VARDECL(celt_norm, norm_save2); + int resynth_alloc; + celt_norm *lowband_scratch; + int B; + int M; + int lowband_offset; + int update_lowband = 1; + int C = Y_ != NULL ? 2 : 1; + int norm_offset; + int theta_rdo = encode && Y_!=NULL && !dual_stereo && complexity>=8; +#ifdef RESYNTH + int resynth = 1; +#else + int resynth = !encode || theta_rdo; +#endif + struct band_ctx ctx; + SAVE_STACK; + + M = 1<<LM; + B = shortBlocks ? M : 1; + norm_offset = M*eBands[start]; + /* No need to allocate norm for the last band because we don't need an + output in that band. */ + ALLOC(_norm, C*(M*eBands[m->nbEBands-1]-norm_offset), celt_norm); + norm = _norm; + norm2 = norm + M*eBands[m->nbEBands-1]-norm_offset; + + /* For decoding, we can use the last band as scratch space because we don't need that + scratch space for the last band and we don't care about the data there until we're + decoding the last band. */ + if (encode && resynth) + resynth_alloc = M*(eBands[m->nbEBands]-eBands[m->nbEBands-1]); + else + resynth_alloc = ALLOC_NONE; + ALLOC(_lowband_scratch, resynth_alloc, celt_norm); + if (encode && resynth) + lowband_scratch = _lowband_scratch; + else + lowband_scratch = X_+M*eBands[m->nbEBands-1]; + ALLOC(X_save, resynth_alloc, celt_norm); + ALLOC(Y_save, resynth_alloc, celt_norm); + ALLOC(X_save2, resynth_alloc, celt_norm); + ALLOC(Y_save2, resynth_alloc, celt_norm); + ALLOC(norm_save2, resynth_alloc, celt_norm); + + lowband_offset = 0; + ctx.bandE = bandE; + ctx.ec = ec; + ctx.encode = encode; + ctx.intensity = intensity; + ctx.m = m; + ctx.seed = *seed; + ctx.spread = spread; + ctx.arch = arch; + ctx.disable_inv = disable_inv; + ctx.resynth = resynth; + ctx.theta_round = 0; + /* Avoid injecting noise in the first band on transients. */ + ctx.avoid_split_noise = B > 1; + for (i=start;i<end;i++) + { + opus_int32 tell; + int b; + int N; + opus_int32 curr_balance; + int effective_lowband=-1; + celt_norm * OPUS_RESTRICT X, * OPUS_RESTRICT Y; + int tf_change=0; + unsigned x_cm; + unsigned y_cm; + int last; + + ctx.i = i; + last = (i==end-1); + + X = X_+M*eBands[i]; + if (Y_!=NULL) + Y = Y_+M*eBands[i]; + else + Y = NULL; + N = M*eBands[i+1]-M*eBands[i]; + celt_assert(N > 0); + tell = ec_tell_frac(ec); + + /* Compute how many bits we want to allocate to this band */ + if (i != start) + balance -= tell; + remaining_bits = total_bits-tell-1; + ctx.remaining_bits = remaining_bits; + if (i <= codedBands-1) + { + curr_balance = celt_sudiv(balance, IMIN(3, codedBands-i)); + b = IMAX(0, IMIN(16383, IMIN(remaining_bits+1,pulses[i]+curr_balance))); + } else { + b = 0; + } + +#ifndef DISABLE_UPDATE_DRAFT + if (resynth && (M*eBands[i]-N >= M*eBands[start] || i==start+1) && (update_lowband || lowband_offset==0)) + lowband_offset = i; + if (i == start+1) + special_hybrid_folding(m, norm, norm2, start, M, dual_stereo); +#else + if (resynth && M*eBands[i]-N >= M*eBands[start] && (update_lowband || lowband_offset==0)) + lowband_offset = i; +#endif + + tf_change = tf_res[i]; + ctx.tf_change = tf_change; + if (i>=m->effEBands) + { + X=norm; + if (Y_!=NULL) + Y = norm; + lowband_scratch = NULL; + } + if (last && !theta_rdo) + lowband_scratch = NULL; + + /* Get a conservative estimate of the collapse_mask's for the bands we're + going to be folding from. */ + if (lowband_offset != 0 && (spread!=SPREAD_AGGRESSIVE || B>1 || tf_change<0)) + { + int fold_start; + int fold_end; + int fold_i; + /* This ensures we never repeat spectral content within one band */ + effective_lowband = IMAX(0, M*eBands[lowband_offset]-norm_offset-N); + fold_start = lowband_offset; + while(M*eBands[--fold_start] > effective_lowband+norm_offset); + fold_end = lowband_offset-1; +#ifndef DISABLE_UPDATE_DRAFT + while(++fold_end < i && M*eBands[fold_end] < effective_lowband+norm_offset+N); +#else + while(M*eBands[++fold_end] < effective_lowband+norm_offset+N); +#endif + x_cm = y_cm = 0; + fold_i = fold_start; do { + x_cm |= collapse_masks[fold_i*C+0]; + y_cm |= collapse_masks[fold_i*C+C-1]; + } while (++fold_i<fold_end); + } + /* Otherwise, we'll be using the LCG to fold, so all blocks will (almost + always) be non-zero. */ + else + x_cm = y_cm = (1<<B)-1; + + if (dual_stereo && i==intensity) + { + int j; + + /* Switch off dual stereo to do intensity. */ + dual_stereo = 0; + if (resynth) + for (j=0;j<M*eBands[i]-norm_offset;j++) + norm[j] = HALF32(norm[j]+norm2[j]); + } + if (dual_stereo) + { + x_cm = quant_band(&ctx, X, N, b/2, B, + effective_lowband != -1 ? norm+effective_lowband : NULL, LM, + last?NULL:norm+M*eBands[i]-norm_offset, Q15ONE, lowband_scratch, x_cm); + y_cm = quant_band(&ctx, Y, N, b/2, B, + effective_lowband != -1 ? norm2+effective_lowband : NULL, LM, + last?NULL:norm2+M*eBands[i]-norm_offset, Q15ONE, lowband_scratch, y_cm); + } else { + if (Y!=NULL) + { + if (theta_rdo && i < intensity) + { + ec_ctx ec_save, ec_save2; + struct band_ctx ctx_save, ctx_save2; + opus_val32 dist0, dist1; + unsigned cm, cm2; + int nstart_bytes, nend_bytes, save_bytes; + unsigned char *bytes_buf; + unsigned char bytes_save[1275]; + opus_val16 w[2]; + compute_channel_weights(bandE[i], bandE[i+m->nbEBands], w); + /* Make a copy. */ + cm = x_cm|y_cm; + ec_save = *ec; + ctx_save = ctx; + OPUS_COPY(X_save, X, N); + OPUS_COPY(Y_save, Y, N); + /* Encode and round down. */ + ctx.theta_round = -1; + x_cm = quant_band_stereo(&ctx, X, Y, N, b, B, + effective_lowband != -1 ? norm+effective_lowband : NULL, LM, + last?NULL:norm+M*eBands[i]-norm_offset, lowband_scratch, cm); + dist0 = MULT16_32_Q15(w[0], celt_inner_prod(X_save, X, N, arch)) + MULT16_32_Q15(w[1], celt_inner_prod(Y_save, Y, N, arch)); + + /* Save first result. */ + cm2 = x_cm; + ec_save2 = *ec; + ctx_save2 = ctx; + OPUS_COPY(X_save2, X, N); + OPUS_COPY(Y_save2, Y, N); + if (!last) + OPUS_COPY(norm_save2, norm+M*eBands[i]-norm_offset, N); + nstart_bytes = ec_save.offs; + nend_bytes = ec_save.storage; + bytes_buf = ec_save.buf+nstart_bytes; + save_bytes = nend_bytes-nstart_bytes; + OPUS_COPY(bytes_save, bytes_buf, save_bytes); + + /* Restore */ + *ec = ec_save; + ctx = ctx_save; + OPUS_COPY(X, X_save, N); + OPUS_COPY(Y, Y_save, N); +#ifndef DISABLE_UPDATE_DRAFT + if (i == start+1) + special_hybrid_folding(m, norm, norm2, start, M, dual_stereo); +#endif + /* Encode and round up. */ + ctx.theta_round = 1; + x_cm = quant_band_stereo(&ctx, X, Y, N, b, B, + effective_lowband != -1 ? norm+effective_lowband : NULL, LM, + last?NULL:norm+M*eBands[i]-norm_offset, lowband_scratch, cm); + dist1 = MULT16_32_Q15(w[0], celt_inner_prod(X_save, X, N, arch)) + MULT16_32_Q15(w[1], celt_inner_prod(Y_save, Y, N, arch)); + if (dist0 >= dist1) { + x_cm = cm2; + *ec = ec_save2; + ctx = ctx_save2; + OPUS_COPY(X, X_save2, N); + OPUS_COPY(Y, Y_save2, N); + if (!last) + OPUS_COPY(norm+M*eBands[i]-norm_offset, norm_save2, N); + OPUS_COPY(bytes_buf, bytes_save, save_bytes); + } + } else { + ctx.theta_round = 0; + x_cm = quant_band_stereo(&ctx, X, Y, N, b, B, + effective_lowband != -1 ? norm+effective_lowband : NULL, LM, + last?NULL:norm+M*eBands[i]-norm_offset, lowband_scratch, x_cm|y_cm); + } + } else { + x_cm = quant_band(&ctx, X, N, b, B, + effective_lowband != -1 ? norm+effective_lowband : NULL, LM, + last?NULL:norm+M*eBands[i]-norm_offset, Q15ONE, lowband_scratch, x_cm|y_cm); + } + y_cm = x_cm; + } + collapse_masks[i*C+0] = (unsigned char)x_cm; + collapse_masks[i*C+C-1] = (unsigned char)y_cm; + balance += pulses[i] + tell; + + /* Update the folding position only as long as we have 1 bit/sample depth. */ + update_lowband = b>(N<<BITRES); + /* We only need to avoid noise on a split for the first band. After that, we + have folding. */ + ctx.avoid_split_noise = 0; + } + *seed = ctx.seed; + + RESTORE_STACK; +} + diff --git a/media/libopus/celt/bands.h b/media/libopus/celt/bands.h new file mode 100644 index 0000000000..422b32cf75 --- /dev/null +++ b/media/libopus/celt/bands.h @@ -0,0 +1,123 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Copyright (c) 2008-2009 Gregory Maxwell + Written by Jean-Marc Valin and Gregory Maxwell */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef BANDS_H +#define BANDS_H + +#include "arch.h" +#include "modes.h" +#include "entenc.h" +#include "entdec.h" +#include "rate.h" + +opus_int16 bitexact_cos(opus_int16 x); +int bitexact_log2tan(int isin,int icos); + +/** Compute the amplitude (sqrt energy) in each of the bands + * @param m Mode data + * @param X Spectrum + * @param bandE Square root of the energy for each band (returned) + */ +void compute_band_energies(const CELTMode *m, const celt_sig *X, celt_ener *bandE, int end, int C, int LM, int arch); + +/*void compute_noise_energies(const CELTMode *m, const celt_sig *X, const opus_val16 *tonality, celt_ener *bandE);*/ + +/** Normalise each band of X such that the energy in each band is + equal to 1 + * @param m Mode data + * @param X Spectrum (returned normalised) + * @param bandE Square root of the energy for each band + */ +void normalise_bands(const CELTMode *m, const celt_sig * OPUS_RESTRICT freq, celt_norm * OPUS_RESTRICT X, const celt_ener *bandE, int end, int C, int M); + +/** Denormalise each band of X to restore full amplitude + * @param m Mode data + * @param X Spectrum (returned de-normalised) + * @param bandE Square root of the energy for each band + */ +void denormalise_bands(const CELTMode *m, const celt_norm * OPUS_RESTRICT X, + celt_sig * OPUS_RESTRICT freq, const opus_val16 *bandE, int start, + int end, int M, int downsample, int silence); + +#define SPREAD_NONE (0) +#define SPREAD_LIGHT (1) +#define SPREAD_NORMAL (2) +#define SPREAD_AGGRESSIVE (3) + +int spreading_decision(const CELTMode *m, const celt_norm *X, int *average, + int last_decision, int *hf_average, int *tapset_decision, int update_hf, + int end, int C, int M, const int *spread_weight); + +#ifdef MEASURE_NORM_MSE +void measure_norm_mse(const CELTMode *m, float *X, float *X0, float *bandE, float *bandE0, int M, int N, int C); +#endif + +void haar1(celt_norm *X, int N0, int stride); + +/** Quantisation/encoding of the residual spectrum + * @param encode flag that indicates whether we're encoding (1) or decoding (0) + * @param m Mode data + * @param start First band to process + * @param end Last band to process + 1 + * @param X Residual (normalised) + * @param Y Residual (normalised) for second channel (or NULL for mono) + * @param collapse_masks Anti-collapse tracking mask + * @param bandE Square root of the energy for each band + * @param pulses Bit allocation (per band) for PVQ + * @param shortBlocks Zero for long blocks, non-zero for short blocks + * @param spread Amount of spreading to use + * @param dual_stereo Zero for MS stereo, non-zero for dual stereo + * @param intensity First band to use intensity stereo + * @param tf_res Time-frequency resolution change + * @param total_bits Total number of bits that can be used for the frame (including the ones already spent) + * @param balance Number of unallocated bits + * @param en Entropy coder state + * @param LM log2() of the number of 2.5 subframes in the frame + * @param codedBands Last band to receive bits + 1 + * @param seed Random generator seed + * @param arch Run-time architecture (see opus_select_arch()) + */ +void quant_all_bands(int encode, const CELTMode *m, int start, int end, + celt_norm * X, celt_norm * Y, unsigned char *collapse_masks, + const celt_ener *bandE, int *pulses, int shortBlocks, int spread, + int dual_stereo, int intensity, int *tf_res, opus_int32 total_bits, + opus_int32 balance, ec_ctx *ec, int M, int codedBands, opus_uint32 *seed, + int complexity, int arch, int disable_inv); + +void anti_collapse(const CELTMode *m, celt_norm *X_, + unsigned char *collapse_masks, int LM, int C, int size, int start, + int end, const opus_val16 *logE, const opus_val16 *prev1logE, + const opus_val16 *prev2logE, const int *pulses, opus_uint32 seed, + int arch); + +opus_uint32 celt_lcg_rand(opus_uint32 seed); + +int hysteresis_decision(opus_val16 val, const opus_val16 *thresholds, const opus_val16 *hysteresis, int N, int prev); + +#endif /* BANDS_H */ diff --git a/media/libopus/celt/celt.c b/media/libopus/celt/celt.c new file mode 100644 index 0000000000..9ce234695c --- /dev/null +++ b/media/libopus/celt/celt.c @@ -0,0 +1,316 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2010 Xiph.Org Foundation + Copyright (c) 2008 Gregory Maxwell + Written by Jean-Marc Valin and Gregory Maxwell */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#define CELT_C + +#include "os_support.h" +#include "mdct.h" +#include <math.h> +#include "celt.h" +#include "pitch.h" +#include "bands.h" +#include "modes.h" +#include "entcode.h" +#include "quant_bands.h" +#include "rate.h" +#include "stack_alloc.h" +#include "mathops.h" +#include "float_cast.h" +#include <stdarg.h> +#include "celt_lpc.h" +#include "vq.h" + +#ifndef PACKAGE_VERSION +#define PACKAGE_VERSION "unknown" +#endif + +#if defined(MIPSr1_ASM) +#include "mips/celt_mipsr1.h" +#endif + + +int resampling_factor(opus_int32 rate) +{ + int ret; + switch (rate) + { + case 48000: + ret = 1; + break; + case 24000: + ret = 2; + break; + case 16000: + ret = 3; + break; + case 12000: + ret = 4; + break; + case 8000: + ret = 6; + break; + default: +#ifndef CUSTOM_MODES + celt_assert(0); +#endif + ret = 0; + break; + } + return ret; +} + +#if !defined(OVERRIDE_COMB_FILTER_CONST) || defined(NON_STATIC_COMB_FILTER_CONST_C) +/* This version should be faster on ARM */ +#ifdef OPUS_ARM_ASM +#ifndef NON_STATIC_COMB_FILTER_CONST_C +static +#endif +void comb_filter_const_c(opus_val32 *y, opus_val32 *x, int T, int N, + opus_val16 g10, opus_val16 g11, opus_val16 g12) +{ + opus_val32 x0, x1, x2, x3, x4; + int i; + x4 = SHL32(x[-T-2], 1); + x3 = SHL32(x[-T-1], 1); + x2 = SHL32(x[-T], 1); + x1 = SHL32(x[-T+1], 1); + for (i=0;i<N-4;i+=5) + { + opus_val32 t; + x0=SHL32(x[i-T+2],1); + t = MAC16_32_Q16(x[i], g10, x2); + t = MAC16_32_Q16(t, g11, ADD32(x1,x3)); + t = MAC16_32_Q16(t, g12, ADD32(x0,x4)); + t = SATURATE(t, SIG_SAT); + y[i] = t; + x4=SHL32(x[i-T+3],1); + t = MAC16_32_Q16(x[i+1], g10, x1); + t = MAC16_32_Q16(t, g11, ADD32(x0,x2)); + t = MAC16_32_Q16(t, g12, ADD32(x4,x3)); + t = SATURATE(t, SIG_SAT); + y[i+1] = t; + x3=SHL32(x[i-T+4],1); + t = MAC16_32_Q16(x[i+2], g10, x0); + t = MAC16_32_Q16(t, g11, ADD32(x4,x1)); + t = MAC16_32_Q16(t, g12, ADD32(x3,x2)); + t = SATURATE(t, SIG_SAT); + y[i+2] = t; + x2=SHL32(x[i-T+5],1); + t = MAC16_32_Q16(x[i+3], g10, x4); + t = MAC16_32_Q16(t, g11, ADD32(x3,x0)); + t = MAC16_32_Q16(t, g12, ADD32(x2,x1)); + t = SATURATE(t, SIG_SAT); + y[i+3] = t; + x1=SHL32(x[i-T+6],1); + t = MAC16_32_Q16(x[i+4], g10, x3); + t = MAC16_32_Q16(t, g11, ADD32(x2,x4)); + t = MAC16_32_Q16(t, g12, ADD32(x1,x0)); + t = SATURATE(t, SIG_SAT); + y[i+4] = t; + } +#ifdef CUSTOM_MODES + for (;i<N;i++) + { + opus_val32 t; + x0=SHL32(x[i-T+2],1); + t = MAC16_32_Q16(x[i], g10, x2); + t = MAC16_32_Q16(t, g11, ADD32(x1,x3)); + t = MAC16_32_Q16(t, g12, ADD32(x0,x4)); + t = SATURATE(t, SIG_SAT); + y[i] = t; + x4=x3; + x3=x2; + x2=x1; + x1=x0; + } +#endif +} +#else +#ifndef NON_STATIC_COMB_FILTER_CONST_C +static +#endif +void comb_filter_const_c(opus_val32 *y, opus_val32 *x, int T, int N, + opus_val16 g10, opus_val16 g11, opus_val16 g12) +{ + opus_val32 x0, x1, x2, x3, x4; + int i; + x4 = x[-T-2]; + x3 = x[-T-1]; + x2 = x[-T]; + x1 = x[-T+1]; + for (i=0;i<N;i++) + { + x0=x[i-T+2]; + y[i] = x[i] + + MULT16_32_Q15(g10,x2) + + MULT16_32_Q15(g11,ADD32(x1,x3)) + + MULT16_32_Q15(g12,ADD32(x0,x4)); + y[i] = SATURATE(y[i], SIG_SAT); + x4=x3; + x3=x2; + x2=x1; + x1=x0; + } + +} +#endif +#endif + +#ifndef OVERRIDE_comb_filter +void comb_filter(opus_val32 *y, opus_val32 *x, int T0, int T1, int N, + opus_val16 g0, opus_val16 g1, int tapset0, int tapset1, + const opus_val16 *window, int overlap, int arch) +{ + int i; + /* printf ("%d %d %f %f\n", T0, T1, g0, g1); */ + opus_val16 g00, g01, g02, g10, g11, g12; + opus_val32 x0, x1, x2, x3, x4; + static const opus_val16 gains[3][3] = { + {QCONST16(0.3066406250f, 15), QCONST16(0.2170410156f, 15), QCONST16(0.1296386719f, 15)}, + {QCONST16(0.4638671875f, 15), QCONST16(0.2680664062f, 15), QCONST16(0.f, 15)}, + {QCONST16(0.7998046875f, 15), QCONST16(0.1000976562f, 15), QCONST16(0.f, 15)}}; + + if (g0==0 && g1==0) + { + /* OPT: Happens to work without the OPUS_MOVE(), but only because the current encoder already copies x to y */ + if (x!=y) + OPUS_MOVE(y, x, N); + return; + } + /* When the gain is zero, T0 and/or T1 is set to zero. We need + to have then be at least 2 to avoid processing garbage data. */ + T0 = IMAX(T0, COMBFILTER_MINPERIOD); + T1 = IMAX(T1, COMBFILTER_MINPERIOD); + g00 = MULT16_16_P15(g0, gains[tapset0][0]); + g01 = MULT16_16_P15(g0, gains[tapset0][1]); + g02 = MULT16_16_P15(g0, gains[tapset0][2]); + g10 = MULT16_16_P15(g1, gains[tapset1][0]); + g11 = MULT16_16_P15(g1, gains[tapset1][1]); + g12 = MULT16_16_P15(g1, gains[tapset1][2]); + x1 = x[-T1+1]; + x2 = x[-T1 ]; + x3 = x[-T1-1]; + x4 = x[-T1-2]; + /* If the filter didn't change, we don't need the overlap */ + if (g0==g1 && T0==T1 && tapset0==tapset1) + overlap=0; + for (i=0;i<overlap;i++) + { + opus_val16 f; + x0=x[i-T1+2]; + f = MULT16_16_Q15(window[i],window[i]); + y[i] = x[i] + + MULT16_32_Q15(MULT16_16_Q15((Q15ONE-f),g00),x[i-T0]) + + MULT16_32_Q15(MULT16_16_Q15((Q15ONE-f),g01),ADD32(x[i-T0+1],x[i-T0-1])) + + MULT16_32_Q15(MULT16_16_Q15((Q15ONE-f),g02),ADD32(x[i-T0+2],x[i-T0-2])) + + MULT16_32_Q15(MULT16_16_Q15(f,g10),x2) + + MULT16_32_Q15(MULT16_16_Q15(f,g11),ADD32(x1,x3)) + + MULT16_32_Q15(MULT16_16_Q15(f,g12),ADD32(x0,x4)); + y[i] = SATURATE(y[i], SIG_SAT); + x4=x3; + x3=x2; + x2=x1; + x1=x0; + + } + if (g1==0) + { + /* OPT: Happens to work without the OPUS_MOVE(), but only because the current encoder already copies x to y */ + if (x!=y) + OPUS_MOVE(y+overlap, x+overlap, N-overlap); + return; + } + + /* Compute the part with the constant filter. */ + comb_filter_const(y+i, x+i, T1, N-i, g10, g11, g12, arch); +} +#endif /* OVERRIDE_comb_filter */ + +/* TF change table. Positive values mean better frequency resolution (longer + effective window), whereas negative values mean better time resolution + (shorter effective window). The second index is computed as: + 4*isTransient + 2*tf_select + per_band_flag */ +const signed char tf_select_table[4][8] = { + /*isTransient=0 isTransient=1 */ + {0, -1, 0, -1, 0,-1, 0,-1}, /* 2.5 ms */ + {0, -1, 0, -2, 1, 0, 1,-1}, /* 5 ms */ + {0, -2, 0, -3, 2, 0, 1,-1}, /* 10 ms */ + {0, -2, 0, -3, 3, 0, 1,-1}, /* 20 ms */ +}; + + +void init_caps(const CELTMode *m,int *cap,int LM,int C) +{ + int i; + for (i=0;i<m->nbEBands;i++) + { + int N; + N=(m->eBands[i+1]-m->eBands[i])<<LM; + cap[i] = (m->cache.caps[m->nbEBands*(2*LM+C-1)+i]+64)*C*N>>2; + } +} + + + +const char *opus_strerror(int error) +{ + static const char * const error_strings[8] = { + "success", + "invalid argument", + "buffer too small", + "internal error", + "corrupted stream", + "request not implemented", + "invalid state", + "memory allocation failed" + }; + if (error > 0 || error < -7) + return "unknown error"; + else + return error_strings[-error]; +} + +const char *opus_get_version_string(void) +{ + return "libopus " PACKAGE_VERSION + /* Applications may rely on the presence of this substring in the version + string to determine if they have a fixed-point or floating-point build + at runtime. */ +#ifdef FIXED_POINT + "-fixed" +#endif +#ifdef FUZZING + "-fuzzing" +#endif + ; +} diff --git a/media/libopus/celt/celt.h b/media/libopus/celt/celt.h new file mode 100644 index 0000000000..24b6b2b520 --- /dev/null +++ b/media/libopus/celt/celt.h @@ -0,0 +1,251 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Copyright (c) 2008 Gregory Maxwell + Written by Jean-Marc Valin and Gregory Maxwell */ +/** + @file celt.h + @brief Contains all the functions for encoding and decoding audio + */ + +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef CELT_H +#define CELT_H + +#include "opus_types.h" +#include "opus_defines.h" +#include "opus_custom.h" +#include "entenc.h" +#include "entdec.h" +#include "arch.h" + +#ifdef __cplusplus +extern "C" { +#endif + +#define CELTEncoder OpusCustomEncoder +#define CELTDecoder OpusCustomDecoder +#define CELTMode OpusCustomMode + +#define LEAK_BANDS 19 + +typedef struct { + int valid; + float tonality; + float tonality_slope; + float noisiness; + float activity; + float music_prob; + float music_prob_min; + float music_prob_max; + int bandwidth; + float activity_probability; + float max_pitch_ratio; + /* Store as Q6 char to save space. */ + unsigned char leak_boost[LEAK_BANDS]; +} AnalysisInfo; + +typedef struct { + int signalType; + int offset; +} SILKInfo; + +#define __celt_check_mode_ptr_ptr(ptr) ((ptr) + ((ptr) - (const CELTMode**)(ptr))) + +#define __celt_check_analysis_ptr(ptr) ((ptr) + ((ptr) - (const AnalysisInfo*)(ptr))) + +#define __celt_check_silkinfo_ptr(ptr) ((ptr) + ((ptr) - (const SILKInfo*)(ptr))) + +/* Encoder/decoder Requests */ + + +#define CELT_SET_PREDICTION_REQUEST 10002 +/** Controls the use of interframe prediction. + 0=Independent frames + 1=Short term interframe prediction allowed + 2=Long term prediction allowed + */ +#define CELT_SET_PREDICTION(x) CELT_SET_PREDICTION_REQUEST, __opus_check_int(x) + +#define CELT_SET_INPUT_CLIPPING_REQUEST 10004 +#define CELT_SET_INPUT_CLIPPING(x) CELT_SET_INPUT_CLIPPING_REQUEST, __opus_check_int(x) + +#define CELT_GET_AND_CLEAR_ERROR_REQUEST 10007 +#define CELT_GET_AND_CLEAR_ERROR(x) CELT_GET_AND_CLEAR_ERROR_REQUEST, __opus_check_int_ptr(x) + +#define CELT_SET_CHANNELS_REQUEST 10008 +#define CELT_SET_CHANNELS(x) CELT_SET_CHANNELS_REQUEST, __opus_check_int(x) + + +/* Internal */ +#define CELT_SET_START_BAND_REQUEST 10010 +#define CELT_SET_START_BAND(x) CELT_SET_START_BAND_REQUEST, __opus_check_int(x) + +#define CELT_SET_END_BAND_REQUEST 10012 +#define CELT_SET_END_BAND(x) CELT_SET_END_BAND_REQUEST, __opus_check_int(x) + +#define CELT_GET_MODE_REQUEST 10015 +/** Get the CELTMode used by an encoder or decoder */ +#define CELT_GET_MODE(x) CELT_GET_MODE_REQUEST, __celt_check_mode_ptr_ptr(x) + +#define CELT_SET_SIGNALLING_REQUEST 10016 +#define CELT_SET_SIGNALLING(x) CELT_SET_SIGNALLING_REQUEST, __opus_check_int(x) + +#define CELT_SET_TONALITY_REQUEST 10018 +#define CELT_SET_TONALITY(x) CELT_SET_TONALITY_REQUEST, __opus_check_int(x) +#define CELT_SET_TONALITY_SLOPE_REQUEST 10020 +#define CELT_SET_TONALITY_SLOPE(x) CELT_SET_TONALITY_SLOPE_REQUEST, __opus_check_int(x) + +#define CELT_SET_ANALYSIS_REQUEST 10022 +#define CELT_SET_ANALYSIS(x) CELT_SET_ANALYSIS_REQUEST, __celt_check_analysis_ptr(x) + +#define OPUS_SET_LFE_REQUEST 10024 +#define OPUS_SET_LFE(x) OPUS_SET_LFE_REQUEST, __opus_check_int(x) + +#define OPUS_SET_ENERGY_MASK_REQUEST 10026 +#define OPUS_SET_ENERGY_MASK(x) OPUS_SET_ENERGY_MASK_REQUEST, __opus_check_val16_ptr(x) + +#define CELT_SET_SILK_INFO_REQUEST 10028 +#define CELT_SET_SILK_INFO(x) CELT_SET_SILK_INFO_REQUEST, __celt_check_silkinfo_ptr(x) + +/* Encoder stuff */ + +int celt_encoder_get_size(int channels); + +int celt_encode_with_ec(OpusCustomEncoder * OPUS_RESTRICT st, const opus_val16 * pcm, int frame_size, unsigned char *compressed, int nbCompressedBytes, ec_enc *enc); + +int celt_encoder_init(CELTEncoder *st, opus_int32 sampling_rate, int channels, + int arch); + + + +/* Decoder stuff */ + +int celt_decoder_get_size(int channels); + + +int celt_decoder_init(CELTDecoder *st, opus_int32 sampling_rate, int channels); + +int celt_decode_with_ec(OpusCustomDecoder * OPUS_RESTRICT st, const unsigned char *data, + int len, opus_val16 * OPUS_RESTRICT pcm, int frame_size, ec_dec *dec, int accum); + +#define celt_encoder_ctl opus_custom_encoder_ctl +#define celt_decoder_ctl opus_custom_decoder_ctl + + +#ifdef CUSTOM_MODES +#define OPUS_CUSTOM_NOSTATIC +#else +#define OPUS_CUSTOM_NOSTATIC static OPUS_INLINE +#endif + +static const unsigned char trim_icdf[11] = {126, 124, 119, 109, 87, 41, 19, 9, 4, 2, 0}; +/* Probs: NONE: 21.875%, LIGHT: 6.25%, NORMAL: 65.625%, AGGRESSIVE: 6.25% */ +static const unsigned char spread_icdf[4] = {25, 23, 2, 0}; + +static const unsigned char tapset_icdf[3]={2,1,0}; + +#ifdef CUSTOM_MODES +static const unsigned char toOpusTable[20] = { + 0xE0, 0xE8, 0xF0, 0xF8, + 0xC0, 0xC8, 0xD0, 0xD8, + 0xA0, 0xA8, 0xB0, 0xB8, + 0x00, 0x00, 0x00, 0x00, + 0x80, 0x88, 0x90, 0x98, +}; + +static const unsigned char fromOpusTable[16] = { + 0x80, 0x88, 0x90, 0x98, + 0x40, 0x48, 0x50, 0x58, + 0x20, 0x28, 0x30, 0x38, + 0x00, 0x08, 0x10, 0x18 +}; + +static OPUS_INLINE int toOpus(unsigned char c) +{ + int ret=0; + if (c<0xA0) + ret = toOpusTable[c>>3]; + if (ret == 0) + return -1; + else + return ret|(c&0x7); +} + +static OPUS_INLINE int fromOpus(unsigned char c) +{ + if (c<0x80) + return -1; + else + return fromOpusTable[(c>>3)-16] | (c&0x7); +} +#endif /* CUSTOM_MODES */ + +#define COMBFILTER_MAXPERIOD 1024 +#define COMBFILTER_MINPERIOD 15 + +extern const signed char tf_select_table[4][8]; + +#if defined(ENABLE_HARDENING) || defined(ENABLE_ASSERTIONS) +void validate_celt_decoder(CELTDecoder *st); +#define VALIDATE_CELT_DECODER(st) validate_celt_decoder(st) +#else +#define VALIDATE_CELT_DECODER(st) +#endif + +int resampling_factor(opus_int32 rate); + +void celt_preemphasis(const opus_val16 * OPUS_RESTRICT pcmp, celt_sig * OPUS_RESTRICT inp, + int N, int CC, int upsample, const opus_val16 *coef, celt_sig *mem, int clip); + +void comb_filter(opus_val32 *y, opus_val32 *x, int T0, int T1, int N, + opus_val16 g0, opus_val16 g1, int tapset0, int tapset1, + const opus_val16 *window, int overlap, int arch); + +#ifdef NON_STATIC_COMB_FILTER_CONST_C +void comb_filter_const_c(opus_val32 *y, opus_val32 *x, int T, int N, + opus_val16 g10, opus_val16 g11, opus_val16 g12); +#endif + +#ifndef OVERRIDE_COMB_FILTER_CONST +# define comb_filter_const(y, x, T, N, g10, g11, g12, arch) \ + ((void)(arch),comb_filter_const_c(y, x, T, N, g10, g11, g12)) +#endif + +void init_caps(const CELTMode *m,int *cap,int LM,int C); + +#ifdef RESYNTH +void deemphasis(celt_sig *in[], opus_val16 *pcm, int N, int C, int downsample, const opus_val16 *coef, celt_sig *mem); +void celt_synthesis(const CELTMode *mode, celt_norm *X, celt_sig * out_syn[], + opus_val16 *oldBandE, int start, int effEnd, int C, int CC, int isTransient, + int LM, int downsample, int silence); +#endif + +#ifdef __cplusplus +} +#endif + +#endif /* CELT_H */ diff --git a/media/libopus/celt/celt_decoder.c b/media/libopus/celt/celt_decoder.c new file mode 100644 index 0000000000..883dae15d2 --- /dev/null +++ b/media/libopus/celt/celt_decoder.c @@ -0,0 +1,1375 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2010 Xiph.Org Foundation + Copyright (c) 2008 Gregory Maxwell + Written by Jean-Marc Valin and Gregory Maxwell */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#define CELT_DECODER_C + +#include "cpu_support.h" +#include "os_support.h" +#include "mdct.h" +#include <math.h> +#include "celt.h" +#include "pitch.h" +#include "bands.h" +#include "modes.h" +#include "entcode.h" +#include "quant_bands.h" +#include "rate.h" +#include "stack_alloc.h" +#include "mathops.h" +#include "float_cast.h" +#include <stdarg.h> +#include "celt_lpc.h" +#include "vq.h" + +/* The maximum pitch lag to allow in the pitch-based PLC. It's possible to save + CPU time in the PLC pitch search by making this smaller than MAX_PERIOD. The + current value corresponds to a pitch of 66.67 Hz. */ +#define PLC_PITCH_LAG_MAX (720) +/* The minimum pitch lag to allow in the pitch-based PLC. This corresponds to a + pitch of 480 Hz. */ +#define PLC_PITCH_LAG_MIN (100) + +#if defined(SMALL_FOOTPRINT) && defined(FIXED_POINT) +#define NORM_ALIASING_HACK +#endif +/**********************************************************************/ +/* */ +/* DECODER */ +/* */ +/**********************************************************************/ +#define DECODE_BUFFER_SIZE 2048 + +/** Decoder state + @brief Decoder state + */ +struct OpusCustomDecoder { + const OpusCustomMode *mode; + int overlap; + int channels; + int stream_channels; + + int downsample; + int start, end; + int signalling; + int disable_inv; + int arch; + + /* Everything beyond this point gets cleared on a reset */ +#define DECODER_RESET_START rng + + opus_uint32 rng; + int error; + int last_pitch_index; + int loss_duration; + int skip_plc; + int postfilter_period; + int postfilter_period_old; + opus_val16 postfilter_gain; + opus_val16 postfilter_gain_old; + int postfilter_tapset; + int postfilter_tapset_old; + + celt_sig preemph_memD[2]; + + celt_sig _decode_mem[1]; /* Size = channels*(DECODE_BUFFER_SIZE+mode->overlap) */ + /* opus_val16 lpc[], Size = channels*LPC_ORDER */ + /* opus_val16 oldEBands[], Size = 2*mode->nbEBands */ + /* opus_val16 oldLogE[], Size = 2*mode->nbEBands */ + /* opus_val16 oldLogE2[], Size = 2*mode->nbEBands */ + /* opus_val16 backgroundLogE[], Size = 2*mode->nbEBands */ +}; + +#if defined(ENABLE_HARDENING) || defined(ENABLE_ASSERTIONS) +/* Make basic checks on the CELT state to ensure we don't end + up writing all over memory. */ +void validate_celt_decoder(CELTDecoder *st) +{ +#ifndef CUSTOM_MODES + celt_assert(st->mode == opus_custom_mode_create(48000, 960, NULL)); + celt_assert(st->overlap == 120); + celt_assert(st->end <= 21); +#else +/* From Section 4.3 in the spec: "The normal CELT layer uses 21 of those bands, + though Opus Custom (see Section 6.2) may use a different number of bands" + + Check if it's within the maximum number of Bark frequency bands instead */ + celt_assert(st->end <= 25); +#endif + celt_assert(st->channels == 1 || st->channels == 2); + celt_assert(st->stream_channels == 1 || st->stream_channels == 2); + celt_assert(st->downsample > 0); + celt_assert(st->start == 0 || st->start == 17); + celt_assert(st->start < st->end); +#ifdef OPUS_ARCHMASK + celt_assert(st->arch >= 0); + celt_assert(st->arch <= OPUS_ARCHMASK); +#endif + celt_assert(st->last_pitch_index <= PLC_PITCH_LAG_MAX); + celt_assert(st->last_pitch_index >= PLC_PITCH_LAG_MIN || st->last_pitch_index == 0); + celt_assert(st->postfilter_period < MAX_PERIOD); + celt_assert(st->postfilter_period >= COMBFILTER_MINPERIOD || st->postfilter_period == 0); + celt_assert(st->postfilter_period_old < MAX_PERIOD); + celt_assert(st->postfilter_period_old >= COMBFILTER_MINPERIOD || st->postfilter_period_old == 0); + celt_assert(st->postfilter_tapset <= 2); + celt_assert(st->postfilter_tapset >= 0); + celt_assert(st->postfilter_tapset_old <= 2); + celt_assert(st->postfilter_tapset_old >= 0); +} +#endif + +int celt_decoder_get_size(int channels) +{ + const CELTMode *mode = opus_custom_mode_create(48000, 960, NULL); + return opus_custom_decoder_get_size(mode, channels); +} + +OPUS_CUSTOM_NOSTATIC int opus_custom_decoder_get_size(const CELTMode *mode, int channels) +{ + int size = sizeof(struct CELTDecoder) + + (channels*(DECODE_BUFFER_SIZE+mode->overlap)-1)*sizeof(celt_sig) + + channels*LPC_ORDER*sizeof(opus_val16) + + 4*2*mode->nbEBands*sizeof(opus_val16); + return size; +} + +#ifdef CUSTOM_MODES +CELTDecoder *opus_custom_decoder_create(const CELTMode *mode, int channels, int *error) +{ + int ret; + CELTDecoder *st = (CELTDecoder *)opus_alloc(opus_custom_decoder_get_size(mode, channels)); + ret = opus_custom_decoder_init(st, mode, channels); + if (ret != OPUS_OK) + { + opus_custom_decoder_destroy(st); + st = NULL; + } + if (error) + *error = ret; + return st; +} +#endif /* CUSTOM_MODES */ + +int celt_decoder_init(CELTDecoder *st, opus_int32 sampling_rate, int channels) +{ + int ret; + ret = opus_custom_decoder_init(st, opus_custom_mode_create(48000, 960, NULL), channels); + if (ret != OPUS_OK) + return ret; + st->downsample = resampling_factor(sampling_rate); + if (st->downsample==0) + return OPUS_BAD_ARG; + else + return OPUS_OK; +} + +OPUS_CUSTOM_NOSTATIC int opus_custom_decoder_init(CELTDecoder *st, const CELTMode *mode, int channels) +{ + if (channels < 0 || channels > 2) + return OPUS_BAD_ARG; + + if (st==NULL) + return OPUS_ALLOC_FAIL; + + OPUS_CLEAR((char*)st, opus_custom_decoder_get_size(mode, channels)); + + st->mode = mode; + st->overlap = mode->overlap; + st->stream_channels = st->channels = channels; + + st->downsample = 1; + st->start = 0; + st->end = st->mode->effEBands; + st->signalling = 1; +#ifndef DISABLE_UPDATE_DRAFT + st->disable_inv = channels == 1; +#else + st->disable_inv = 0; +#endif + st->arch = opus_select_arch(); + + opus_custom_decoder_ctl(st, OPUS_RESET_STATE); + + return OPUS_OK; +} + +#ifdef CUSTOM_MODES +void opus_custom_decoder_destroy(CELTDecoder *st) +{ + opus_free(st); +} +#endif /* CUSTOM_MODES */ + +#ifndef CUSTOM_MODES +/* Special case for stereo with no downsampling and no accumulation. This is + quite common and we can make it faster by processing both channels in the + same loop, reducing overhead due to the dependency loop in the IIR filter. */ +static void deemphasis_stereo_simple(celt_sig *in[], opus_val16 *pcm, int N, const opus_val16 coef0, + celt_sig *mem) +{ + celt_sig * OPUS_RESTRICT x0; + celt_sig * OPUS_RESTRICT x1; + celt_sig m0, m1; + int j; + x0=in[0]; + x1=in[1]; + m0 = mem[0]; + m1 = mem[1]; + for (j=0;j<N;j++) + { + celt_sig tmp0, tmp1; + /* Add VERY_SMALL to x[] first to reduce dependency chain. */ + tmp0 = x0[j] + VERY_SMALL + m0; + tmp1 = x1[j] + VERY_SMALL + m1; + m0 = MULT16_32_Q15(coef0, tmp0); + m1 = MULT16_32_Q15(coef0, tmp1); + pcm[2*j ] = SCALEOUT(SIG2WORD16(tmp0)); + pcm[2*j+1] = SCALEOUT(SIG2WORD16(tmp1)); + } + mem[0] = m0; + mem[1] = m1; +} +#endif + +#ifndef RESYNTH +static +#endif +void deemphasis(celt_sig *in[], opus_val16 *pcm, int N, int C, int downsample, const opus_val16 *coef, + celt_sig *mem, int accum) +{ + int c; + int Nd; + int apply_downsampling=0; + opus_val16 coef0; + VARDECL(celt_sig, scratch); + SAVE_STACK; +#ifndef CUSTOM_MODES + /* Short version for common case. */ + if (downsample == 1 && C == 2 && !accum) + { + deemphasis_stereo_simple(in, pcm, N, coef[0], mem); + return; + } +#endif +#ifndef FIXED_POINT + (void)accum; + celt_assert(accum==0); +#endif + ALLOC(scratch, N, celt_sig); + coef0 = coef[0]; + Nd = N/downsample; + c=0; do { + int j; + celt_sig * OPUS_RESTRICT x; + opus_val16 * OPUS_RESTRICT y; + celt_sig m = mem[c]; + x =in[c]; + y = pcm+c; +#ifdef CUSTOM_MODES + if (coef[1] != 0) + { + opus_val16 coef1 = coef[1]; + opus_val16 coef3 = coef[3]; + for (j=0;j<N;j++) + { + celt_sig tmp = x[j] + m + VERY_SMALL; + m = MULT16_32_Q15(coef0, tmp) + - MULT16_32_Q15(coef1, x[j]); + tmp = SHL32(MULT16_32_Q15(coef3, tmp), 2); + scratch[j] = tmp; + } + apply_downsampling=1; + } else +#endif + if (downsample>1) + { + /* Shortcut for the standard (non-custom modes) case */ + for (j=0;j<N;j++) + { + celt_sig tmp = x[j] + VERY_SMALL + m; + m = MULT16_32_Q15(coef0, tmp); + scratch[j] = tmp; + } + apply_downsampling=1; + } else { + /* Shortcut for the standard (non-custom modes) case */ +#ifdef FIXED_POINT + if (accum) + { + for (j=0;j<N;j++) + { + celt_sig tmp = x[j] + m + VERY_SMALL; + m = MULT16_32_Q15(coef0, tmp); + y[j*C] = SAT16(ADD32(y[j*C], SCALEOUT(SIG2WORD16(tmp)))); + } + } else +#endif + { + for (j=0;j<N;j++) + { + celt_sig tmp = x[j] + VERY_SMALL + m; + m = MULT16_32_Q15(coef0, tmp); + y[j*C] = SCALEOUT(SIG2WORD16(tmp)); + } + } + } + mem[c] = m; + + if (apply_downsampling) + { + /* Perform down-sampling */ +#ifdef FIXED_POINT + if (accum) + { + for (j=0;j<Nd;j++) + y[j*C] = SAT16(ADD32(y[j*C], SCALEOUT(SIG2WORD16(scratch[j*downsample])))); + } else +#endif + { + for (j=0;j<Nd;j++) + y[j*C] = SCALEOUT(SIG2WORD16(scratch[j*downsample])); + } + } + } while (++c<C); + RESTORE_STACK; +} + +#ifndef RESYNTH +static +#endif +void celt_synthesis(const CELTMode *mode, celt_norm *X, celt_sig * out_syn[], + opus_val16 *oldBandE, int start, int effEnd, int C, int CC, + int isTransient, int LM, int downsample, + int silence, int arch) +{ + int c, i; + int M; + int b; + int B; + int N, NB; + int shift; + int nbEBands; + int overlap; + VARDECL(celt_sig, freq); + SAVE_STACK; + + overlap = mode->overlap; + nbEBands = mode->nbEBands; + N = mode->shortMdctSize<<LM; + ALLOC(freq, N, celt_sig); /**< Interleaved signal MDCTs */ + M = 1<<LM; + + if (isTransient) + { + B = M; + NB = mode->shortMdctSize; + shift = mode->maxLM; + } else { + B = 1; + NB = mode->shortMdctSize<<LM; + shift = mode->maxLM-LM; + } + + if (CC==2&&C==1) + { + /* Copying a mono streams to two channels */ + celt_sig *freq2; + denormalise_bands(mode, X, freq, oldBandE, start, effEnd, M, + downsample, silence); + /* Store a temporary copy in the output buffer because the IMDCT destroys its input. */ + freq2 = out_syn[1]+overlap/2; + OPUS_COPY(freq2, freq, N); + for (b=0;b<B;b++) + clt_mdct_backward(&mode->mdct, &freq2[b], out_syn[0]+NB*b, mode->window, overlap, shift, B, arch); + for (b=0;b<B;b++) + clt_mdct_backward(&mode->mdct, &freq[b], out_syn[1]+NB*b, mode->window, overlap, shift, B, arch); + } else if (CC==1&&C==2) + { + /* Downmixing a stereo stream to mono */ + celt_sig *freq2; + freq2 = out_syn[0]+overlap/2; + denormalise_bands(mode, X, freq, oldBandE, start, effEnd, M, + downsample, silence); + /* Use the output buffer as temp array before downmixing. */ + denormalise_bands(mode, X+N, freq2, oldBandE+nbEBands, start, effEnd, M, + downsample, silence); + for (i=0;i<N;i++) + freq[i] = ADD32(HALF32(freq[i]), HALF32(freq2[i])); + for (b=0;b<B;b++) + clt_mdct_backward(&mode->mdct, &freq[b], out_syn[0]+NB*b, mode->window, overlap, shift, B, arch); + } else { + /* Normal case (mono or stereo) */ + c=0; do { + denormalise_bands(mode, X+c*N, freq, oldBandE+c*nbEBands, start, effEnd, M, + downsample, silence); + for (b=0;b<B;b++) + clt_mdct_backward(&mode->mdct, &freq[b], out_syn[c]+NB*b, mode->window, overlap, shift, B, arch); + } while (++c<CC); + } + /* Saturate IMDCT output so that we can't overflow in the pitch postfilter + or in the */ + c=0; do { + for (i=0;i<N;i++) + out_syn[c][i] = SATURATE(out_syn[c][i], SIG_SAT); + } while (++c<CC); + RESTORE_STACK; +} + +static void tf_decode(int start, int end, int isTransient, int *tf_res, int LM, ec_dec *dec) +{ + int i, curr, tf_select; + int tf_select_rsv; + int tf_changed; + int logp; + opus_uint32 budget; + opus_uint32 tell; + + budget = dec->storage*8; + tell = ec_tell(dec); + logp = isTransient ? 2 : 4; + tf_select_rsv = LM>0 && tell+logp+1<=budget; + budget -= tf_select_rsv; + tf_changed = curr = 0; + for (i=start;i<end;i++) + { + if (tell+logp<=budget) + { + curr ^= ec_dec_bit_logp(dec, logp); + tell = ec_tell(dec); + tf_changed |= curr; + } + tf_res[i] = curr; + logp = isTransient ? 4 : 5; + } + tf_select = 0; + if (tf_select_rsv && + tf_select_table[LM][4*isTransient+0+tf_changed] != + tf_select_table[LM][4*isTransient+2+tf_changed]) + { + tf_select = ec_dec_bit_logp(dec, 1); + } + for (i=start;i<end;i++) + { + tf_res[i] = tf_select_table[LM][4*isTransient+2*tf_select+tf_res[i]]; + } +} + +static int celt_plc_pitch_search(celt_sig *decode_mem[2], int C, int arch) +{ + int pitch_index; + VARDECL( opus_val16, lp_pitch_buf ); + SAVE_STACK; + ALLOC( lp_pitch_buf, DECODE_BUFFER_SIZE>>1, opus_val16 ); + pitch_downsample(decode_mem, lp_pitch_buf, + DECODE_BUFFER_SIZE, C, arch); + pitch_search(lp_pitch_buf+(PLC_PITCH_LAG_MAX>>1), lp_pitch_buf, + DECODE_BUFFER_SIZE-PLC_PITCH_LAG_MAX, + PLC_PITCH_LAG_MAX-PLC_PITCH_LAG_MIN, &pitch_index, arch); + pitch_index = PLC_PITCH_LAG_MAX-pitch_index; + RESTORE_STACK; + return pitch_index; +} + +static void celt_decode_lost(CELTDecoder * OPUS_RESTRICT st, int N, int LM) +{ + int c; + int i; + const int C = st->channels; + celt_sig *decode_mem[2]; + celt_sig *out_syn[2]; + opus_val16 *lpc; + opus_val16 *oldBandE, *oldLogE, *oldLogE2, *backgroundLogE; + const OpusCustomMode *mode; + int nbEBands; + int overlap; + int start; + int loss_duration; + int noise_based; + const opus_int16 *eBands; + SAVE_STACK; + + mode = st->mode; + nbEBands = mode->nbEBands; + overlap = mode->overlap; + eBands = mode->eBands; + + c=0; do { + decode_mem[c] = st->_decode_mem + c*(DECODE_BUFFER_SIZE+overlap); + out_syn[c] = decode_mem[c]+DECODE_BUFFER_SIZE-N; + } while (++c<C); + lpc = (opus_val16*)(st->_decode_mem+(DECODE_BUFFER_SIZE+overlap)*C); + oldBandE = lpc+C*LPC_ORDER; + oldLogE = oldBandE + 2*nbEBands; + oldLogE2 = oldLogE + 2*nbEBands; + backgroundLogE = oldLogE2 + 2*nbEBands; + + loss_duration = st->loss_duration; + start = st->start; + noise_based = loss_duration >= 40 || start != 0 || st->skip_plc; + if (noise_based) + { + /* Noise-based PLC/CNG */ +#ifdef NORM_ALIASING_HACK + celt_norm *X; +#else + VARDECL(celt_norm, X); +#endif + opus_uint32 seed; + int end; + int effEnd; + opus_val16 decay; + end = st->end; + effEnd = IMAX(start, IMIN(end, mode->effEBands)); + +#ifdef NORM_ALIASING_HACK + /* This is an ugly hack that breaks aliasing rules and would be easily broken, + but it saves almost 4kB of stack. */ + X = (celt_norm*)(out_syn[C-1]+overlap/2); +#else + ALLOC(X, C*N, celt_norm); /**< Interleaved normalised MDCTs */ +#endif + c=0; do { + OPUS_MOVE(decode_mem[c], decode_mem[c]+N, + DECODE_BUFFER_SIZE-N+(overlap>>1)); + } while (++c<C); + + /* Energy decay */ + decay = loss_duration==0 ? QCONST16(1.5f, DB_SHIFT) : QCONST16(.5f, DB_SHIFT); + c=0; do + { + for (i=start;i<end;i++) + oldBandE[c*nbEBands+i] = MAX16(backgroundLogE[c*nbEBands+i], oldBandE[c*nbEBands+i] - decay); + } while (++c<C); + seed = st->rng; + for (c=0;c<C;c++) + { + for (i=start;i<effEnd;i++) + { + int j; + int boffs; + int blen; + boffs = N*c+(eBands[i]<<LM); + blen = (eBands[i+1]-eBands[i])<<LM; + for (j=0;j<blen;j++) + { + seed = celt_lcg_rand(seed); + X[boffs+j] = (celt_norm)((opus_int32)seed>>20); + } + renormalise_vector(X+boffs, blen, Q15ONE, st->arch); + } + } + st->rng = seed; + + celt_synthesis(mode, X, out_syn, oldBandE, start, effEnd, C, C, 0, LM, st->downsample, 0, st->arch); + } else { + int exc_length; + /* Pitch-based PLC */ + const opus_val16 *window; + opus_val16 *exc; + opus_val16 fade = Q15ONE; + int pitch_index; + VARDECL(opus_val32, etmp); + VARDECL(opus_val16, _exc); + VARDECL(opus_val16, fir_tmp); + + if (loss_duration == 0) + { + st->last_pitch_index = pitch_index = celt_plc_pitch_search(decode_mem, C, st->arch); + } else { + pitch_index = st->last_pitch_index; + fade = QCONST16(.8f,15); + } + + /* We want the excitation for 2 pitch periods in order to look for a + decaying signal, but we can't get more than MAX_PERIOD. */ + exc_length = IMIN(2*pitch_index, MAX_PERIOD); + + ALLOC(etmp, overlap, opus_val32); + ALLOC(_exc, MAX_PERIOD+LPC_ORDER, opus_val16); + ALLOC(fir_tmp, exc_length, opus_val16); + exc = _exc+LPC_ORDER; + window = mode->window; + c=0; do { + opus_val16 decay; + opus_val16 attenuation; + opus_val32 S1=0; + celt_sig *buf; + int extrapolation_offset; + int extrapolation_len; + int j; + + buf = decode_mem[c]; + for (i=0;i<MAX_PERIOD+LPC_ORDER;i++) + exc[i-LPC_ORDER] = SROUND16(buf[DECODE_BUFFER_SIZE-MAX_PERIOD-LPC_ORDER+i], SIG_SHIFT); + + if (loss_duration == 0) + { + opus_val32 ac[LPC_ORDER+1]; + /* Compute LPC coefficients for the last MAX_PERIOD samples before + the first loss so we can work in the excitation-filter domain. */ + _celt_autocorr(exc, ac, window, overlap, + LPC_ORDER, MAX_PERIOD, st->arch); + /* Add a noise floor of -40 dB. */ +#ifdef FIXED_POINT + ac[0] += SHR32(ac[0],13); +#else + ac[0] *= 1.0001f; +#endif + /* Use lag windowing to stabilize the Levinson-Durbin recursion. */ + for (i=1;i<=LPC_ORDER;i++) + { + /*ac[i] *= exp(-.5*(2*M_PI*.002*i)*(2*M_PI*.002*i));*/ +#ifdef FIXED_POINT + ac[i] -= MULT16_32_Q15(2*i*i, ac[i]); +#else + ac[i] -= ac[i]*(0.008f*0.008f)*i*i; +#endif + } + _celt_lpc(lpc+c*LPC_ORDER, ac, LPC_ORDER); +#ifdef FIXED_POINT + /* For fixed-point, apply bandwidth expansion until we can guarantee that + no overflow can happen in the IIR filter. This means: + 32768*sum(abs(filter)) < 2^31 */ + while (1) { + opus_val16 tmp=Q15ONE; + opus_val32 sum=QCONST16(1., SIG_SHIFT); + for (i=0;i<LPC_ORDER;i++) + sum += ABS16(lpc[c*LPC_ORDER+i]); + if (sum < 65535) break; + for (i=0;i<LPC_ORDER;i++) + { + tmp = MULT16_16_Q15(QCONST16(.99f,15), tmp); + lpc[c*LPC_ORDER+i] = MULT16_16_Q15(lpc[c*LPC_ORDER+i], tmp); + } + } +#endif + } + /* Initialize the LPC history with the samples just before the start + of the region for which we're computing the excitation. */ + { + /* Compute the excitation for exc_length samples before the loss. We need the copy + because celt_fir() cannot filter in-place. */ + celt_fir(exc+MAX_PERIOD-exc_length, lpc+c*LPC_ORDER, + fir_tmp, exc_length, LPC_ORDER, st->arch); + OPUS_COPY(exc+MAX_PERIOD-exc_length, fir_tmp, exc_length); + } + + /* Check if the waveform is decaying, and if so how fast. + We do this to avoid adding energy when concealing in a segment + with decaying energy. */ + { + opus_val32 E1=1, E2=1; + int decay_length; +#ifdef FIXED_POINT + int shift = IMAX(0,2*celt_zlog2(celt_maxabs16(&exc[MAX_PERIOD-exc_length], exc_length))-20); +#endif + decay_length = exc_length>>1; + for (i=0;i<decay_length;i++) + { + opus_val16 e; + e = exc[MAX_PERIOD-decay_length+i]; + E1 += SHR32(MULT16_16(e, e), shift); + e = exc[MAX_PERIOD-2*decay_length+i]; + E2 += SHR32(MULT16_16(e, e), shift); + } + E1 = MIN32(E1, E2); + decay = celt_sqrt(frac_div32(SHR32(E1, 1), E2)); + } + + /* Move the decoder memory one frame to the left to give us room to + add the data for the new frame. We ignore the overlap that extends + past the end of the buffer, because we aren't going to use it. */ + OPUS_MOVE(buf, buf+N, DECODE_BUFFER_SIZE-N); + + /* Extrapolate from the end of the excitation with a period of + "pitch_index", scaling down each period by an additional factor of + "decay". */ + extrapolation_offset = MAX_PERIOD-pitch_index; + /* We need to extrapolate enough samples to cover a complete MDCT + window (including overlap/2 samples on both sides). */ + extrapolation_len = N+overlap; + /* We also apply fading if this is not the first loss. */ + attenuation = MULT16_16_Q15(fade, decay); + for (i=j=0;i<extrapolation_len;i++,j++) + { + opus_val16 tmp; + if (j >= pitch_index) { + j -= pitch_index; + attenuation = MULT16_16_Q15(attenuation, decay); + } + buf[DECODE_BUFFER_SIZE-N+i] = + SHL32(EXTEND32(MULT16_16_Q15(attenuation, + exc[extrapolation_offset+j])), SIG_SHIFT); + /* Compute the energy of the previously decoded signal whose + excitation we're copying. */ + tmp = SROUND16( + buf[DECODE_BUFFER_SIZE-MAX_PERIOD-N+extrapolation_offset+j], + SIG_SHIFT); + S1 += SHR32(MULT16_16(tmp, tmp), 10); + } + { + opus_val16 lpc_mem[LPC_ORDER]; + /* Copy the last decoded samples (prior to the overlap region) to + synthesis filter memory so we can have a continuous signal. */ + for (i=0;i<LPC_ORDER;i++) + lpc_mem[i] = SROUND16(buf[DECODE_BUFFER_SIZE-N-1-i], SIG_SHIFT); + /* Apply the synthesis filter to convert the excitation back into + the signal domain. */ + celt_iir(buf+DECODE_BUFFER_SIZE-N, lpc+c*LPC_ORDER, + buf+DECODE_BUFFER_SIZE-N, extrapolation_len, LPC_ORDER, + lpc_mem, st->arch); +#ifdef FIXED_POINT + for (i=0; i < extrapolation_len; i++) + buf[DECODE_BUFFER_SIZE-N+i] = SATURATE(buf[DECODE_BUFFER_SIZE-N+i], SIG_SAT); +#endif + } + + /* Check if the synthesis energy is higher than expected, which can + happen with the signal changes during our window. If so, + attenuate. */ + { + opus_val32 S2=0; + for (i=0;i<extrapolation_len;i++) + { + opus_val16 tmp = SROUND16(buf[DECODE_BUFFER_SIZE-N+i], SIG_SHIFT); + S2 += SHR32(MULT16_16(tmp, tmp), 10); + } + /* This checks for an "explosion" in the synthesis. */ +#ifdef FIXED_POINT + if (!(S1 > SHR32(S2,2))) +#else + /* The float test is written this way to catch NaNs in the output + of the IIR filter at the same time. */ + if (!(S1 > 0.2f*S2)) +#endif + { + for (i=0;i<extrapolation_len;i++) + buf[DECODE_BUFFER_SIZE-N+i] = 0; + } else if (S1 < S2) + { + opus_val16 ratio = celt_sqrt(frac_div32(SHR32(S1,1)+1,S2+1)); + for (i=0;i<overlap;i++) + { + opus_val16 tmp_g = Q15ONE + - MULT16_16_Q15(window[i], Q15ONE-ratio); + buf[DECODE_BUFFER_SIZE-N+i] = + MULT16_32_Q15(tmp_g, buf[DECODE_BUFFER_SIZE-N+i]); + } + for (i=overlap;i<extrapolation_len;i++) + { + buf[DECODE_BUFFER_SIZE-N+i] = + MULT16_32_Q15(ratio, buf[DECODE_BUFFER_SIZE-N+i]); + } + } + } + + /* Apply the pre-filter to the MDCT overlap for the next frame because + the post-filter will be re-applied in the decoder after the MDCT + overlap. */ + comb_filter(etmp, buf+DECODE_BUFFER_SIZE, + st->postfilter_period, st->postfilter_period, overlap, + -st->postfilter_gain, -st->postfilter_gain, + st->postfilter_tapset, st->postfilter_tapset, NULL, 0, st->arch); + + /* Simulate TDAC on the concealed audio so that it blends with the + MDCT of the next frame. */ + for (i=0;i<overlap/2;i++) + { + buf[DECODE_BUFFER_SIZE+i] = + MULT16_32_Q15(window[i], etmp[overlap-1-i]) + + MULT16_32_Q15(window[overlap-i-1], etmp[i]); + } + } while (++c<C); + } + + /* Saturate to soemthing large to avoid wrap-around. */ + st->loss_duration = IMIN(10000, loss_duration+(1<<LM)); + + RESTORE_STACK; +} + +int celt_decode_with_ec(CELTDecoder * OPUS_RESTRICT st, const unsigned char *data, + int len, opus_val16 * OPUS_RESTRICT pcm, int frame_size, ec_dec *dec, int accum) +{ + int c, i, N; + int spread_decision; + opus_int32 bits; + ec_dec _dec; +#ifdef NORM_ALIASING_HACK + celt_norm *X; +#else + VARDECL(celt_norm, X); +#endif + VARDECL(int, fine_quant); + VARDECL(int, pulses); + VARDECL(int, cap); + VARDECL(int, offsets); + VARDECL(int, fine_priority); + VARDECL(int, tf_res); + VARDECL(unsigned char, collapse_masks); + celt_sig *decode_mem[2]; + celt_sig *out_syn[2]; + opus_val16 *lpc; + opus_val16 *oldBandE, *oldLogE, *oldLogE2, *backgroundLogE; + + int shortBlocks; + int isTransient; + int intra_ener; + const int CC = st->channels; + int LM, M; + int start; + int end; + int effEnd; + int codedBands; + int alloc_trim; + int postfilter_pitch; + opus_val16 postfilter_gain; + int intensity=0; + int dual_stereo=0; + opus_int32 total_bits; + opus_int32 balance; + opus_int32 tell; + int dynalloc_logp; + int postfilter_tapset; + int anti_collapse_rsv; + int anti_collapse_on=0; + int silence; + int C = st->stream_channels; + const OpusCustomMode *mode; + int nbEBands; + int overlap; + const opus_int16 *eBands; + opus_val16 max_background_increase; + ALLOC_STACK; + + VALIDATE_CELT_DECODER(st); + mode = st->mode; + nbEBands = mode->nbEBands; + overlap = mode->overlap; + eBands = mode->eBands; + start = st->start; + end = st->end; + frame_size *= st->downsample; + + lpc = (opus_val16*)(st->_decode_mem+(DECODE_BUFFER_SIZE+overlap)*CC); + oldBandE = lpc+CC*LPC_ORDER; + oldLogE = oldBandE + 2*nbEBands; + oldLogE2 = oldLogE + 2*nbEBands; + backgroundLogE = oldLogE2 + 2*nbEBands; + +#ifdef CUSTOM_MODES + if (st->signalling && data!=NULL) + { + int data0=data[0]; + /* Convert "standard mode" to Opus header */ + if (mode->Fs==48000 && mode->shortMdctSize==120) + { + data0 = fromOpus(data0); + if (data0<0) + return OPUS_INVALID_PACKET; + } + st->end = end = IMAX(1, mode->effEBands-2*(data0>>5)); + LM = (data0>>3)&0x3; + C = 1 + ((data0>>2)&0x1); + data++; + len--; + if (LM>mode->maxLM) + return OPUS_INVALID_PACKET; + if (frame_size < mode->shortMdctSize<<LM) + return OPUS_BUFFER_TOO_SMALL; + else + frame_size = mode->shortMdctSize<<LM; + } else { +#else + { +#endif + for (LM=0;LM<=mode->maxLM;LM++) + if (mode->shortMdctSize<<LM==frame_size) + break; + if (LM>mode->maxLM) + return OPUS_BAD_ARG; + } + M=1<<LM; + + if (len<0 || len>1275 || pcm==NULL) + return OPUS_BAD_ARG; + + N = M*mode->shortMdctSize; + c=0; do { + decode_mem[c] = st->_decode_mem + c*(DECODE_BUFFER_SIZE+overlap); + out_syn[c] = decode_mem[c]+DECODE_BUFFER_SIZE-N; + } while (++c<CC); + + effEnd = end; + if (effEnd > mode->effEBands) + effEnd = mode->effEBands; + + if (data == NULL || len<=1) + { + celt_decode_lost(st, N, LM); + deemphasis(out_syn, pcm, N, CC, st->downsample, mode->preemph, st->preemph_memD, accum); + RESTORE_STACK; + return frame_size/st->downsample; + } + + /* Check if there are at least two packets received consecutively before + * turning on the pitch-based PLC */ + st->skip_plc = st->loss_duration != 0; + + if (dec == NULL) + { + ec_dec_init(&_dec,(unsigned char*)data,len); + dec = &_dec; + } + + if (C==1) + { + for (i=0;i<nbEBands;i++) + oldBandE[i]=MAX16(oldBandE[i],oldBandE[nbEBands+i]); + } + + total_bits = len*8; + tell = ec_tell(dec); + + if (tell >= total_bits) + silence = 1; + else if (tell==1) + silence = ec_dec_bit_logp(dec, 15); + else + silence = 0; + if (silence) + { + /* Pretend we've read all the remaining bits */ + tell = len*8; + dec->nbits_total+=tell-ec_tell(dec); + } + + postfilter_gain = 0; + postfilter_pitch = 0; + postfilter_tapset = 0; + if (start==0 && tell+16 <= total_bits) + { + if(ec_dec_bit_logp(dec, 1)) + { + int qg, octave; + octave = ec_dec_uint(dec, 6); + postfilter_pitch = (16<<octave)+ec_dec_bits(dec, 4+octave)-1; + qg = ec_dec_bits(dec, 3); + if (ec_tell(dec)+2<=total_bits) + postfilter_tapset = ec_dec_icdf(dec, tapset_icdf, 2); + postfilter_gain = QCONST16(.09375f,15)*(qg+1); + } + tell = ec_tell(dec); + } + + if (LM > 0 && tell+3 <= total_bits) + { + isTransient = ec_dec_bit_logp(dec, 3); + tell = ec_tell(dec); + } + else + isTransient = 0; + + if (isTransient) + shortBlocks = M; + else + shortBlocks = 0; + + /* Decode the global flags (first symbols in the stream) */ + intra_ener = tell+3<=total_bits ? ec_dec_bit_logp(dec, 3) : 0; + /* Get band energies */ + unquant_coarse_energy(mode, start, end, oldBandE, + intra_ener, dec, C, LM); + + ALLOC(tf_res, nbEBands, int); + tf_decode(start, end, isTransient, tf_res, LM, dec); + + tell = ec_tell(dec); + spread_decision = SPREAD_NORMAL; + if (tell+4 <= total_bits) + spread_decision = ec_dec_icdf(dec, spread_icdf, 5); + + ALLOC(cap, nbEBands, int); + + init_caps(mode,cap,LM,C); + + ALLOC(offsets, nbEBands, int); + + dynalloc_logp = 6; + total_bits<<=BITRES; + tell = ec_tell_frac(dec); + for (i=start;i<end;i++) + { + int width, quanta; + int dynalloc_loop_logp; + int boost; + width = C*(eBands[i+1]-eBands[i])<<LM; + /* quanta is 6 bits, but no more than 1 bit/sample + and no less than 1/8 bit/sample */ + quanta = IMIN(width<<BITRES, IMAX(6<<BITRES, width)); + dynalloc_loop_logp = dynalloc_logp; + boost = 0; + while (tell+(dynalloc_loop_logp<<BITRES) < total_bits && boost < cap[i]) + { + int flag; + flag = ec_dec_bit_logp(dec, dynalloc_loop_logp); + tell = ec_tell_frac(dec); + if (!flag) + break; + boost += quanta; + total_bits -= quanta; + dynalloc_loop_logp = 1; + } + offsets[i] = boost; + /* Making dynalloc more likely */ + if (boost>0) + dynalloc_logp = IMAX(2, dynalloc_logp-1); + } + + ALLOC(fine_quant, nbEBands, int); + alloc_trim = tell+(6<<BITRES) <= total_bits ? + ec_dec_icdf(dec, trim_icdf, 7) : 5; + + bits = (((opus_int32)len*8)<<BITRES) - ec_tell_frac(dec) - 1; + anti_collapse_rsv = isTransient&&LM>=2&&bits>=((LM+2)<<BITRES) ? (1<<BITRES) : 0; + bits -= anti_collapse_rsv; + + ALLOC(pulses, nbEBands, int); + ALLOC(fine_priority, nbEBands, int); + + codedBands = clt_compute_allocation(mode, start, end, offsets, cap, + alloc_trim, &intensity, &dual_stereo, bits, &balance, pulses, + fine_quant, fine_priority, C, LM, dec, 0, 0, 0); + + unquant_fine_energy(mode, start, end, oldBandE, fine_quant, dec, C); + + c=0; do { + OPUS_MOVE(decode_mem[c], decode_mem[c]+N, DECODE_BUFFER_SIZE-N+overlap/2); + } while (++c<CC); + + /* Decode fixed codebook */ + ALLOC(collapse_masks, C*nbEBands, unsigned char); + +#ifdef NORM_ALIASING_HACK + /* This is an ugly hack that breaks aliasing rules and would be easily broken, + but it saves almost 4kB of stack. */ + X = (celt_norm*)(out_syn[CC-1]+overlap/2); +#else + ALLOC(X, C*N, celt_norm); /**< Interleaved normalised MDCTs */ +#endif + + quant_all_bands(0, mode, start, end, X, C==2 ? X+N : NULL, collapse_masks, + NULL, pulses, shortBlocks, spread_decision, dual_stereo, intensity, tf_res, + len*(8<<BITRES)-anti_collapse_rsv, balance, dec, LM, codedBands, &st->rng, 0, + st->arch, st->disable_inv); + + if (anti_collapse_rsv > 0) + { + anti_collapse_on = ec_dec_bits(dec, 1); + } + + unquant_energy_finalise(mode, start, end, oldBandE, + fine_quant, fine_priority, len*8-ec_tell(dec), dec, C); + + if (anti_collapse_on) + anti_collapse(mode, X, collapse_masks, LM, C, N, + start, end, oldBandE, oldLogE, oldLogE2, pulses, st->rng, st->arch); + + if (silence) + { + for (i=0;i<C*nbEBands;i++) + oldBandE[i] = -QCONST16(28.f,DB_SHIFT); + } + + celt_synthesis(mode, X, out_syn, oldBandE, start, effEnd, + C, CC, isTransient, LM, st->downsample, silence, st->arch); + + c=0; do { + st->postfilter_period=IMAX(st->postfilter_period, COMBFILTER_MINPERIOD); + st->postfilter_period_old=IMAX(st->postfilter_period_old, COMBFILTER_MINPERIOD); + comb_filter(out_syn[c], out_syn[c], st->postfilter_period_old, st->postfilter_period, mode->shortMdctSize, + st->postfilter_gain_old, st->postfilter_gain, st->postfilter_tapset_old, st->postfilter_tapset, + mode->window, overlap, st->arch); + if (LM!=0) + comb_filter(out_syn[c]+mode->shortMdctSize, out_syn[c]+mode->shortMdctSize, st->postfilter_period, postfilter_pitch, N-mode->shortMdctSize, + st->postfilter_gain, postfilter_gain, st->postfilter_tapset, postfilter_tapset, + mode->window, overlap, st->arch); + + } while (++c<CC); + st->postfilter_period_old = st->postfilter_period; + st->postfilter_gain_old = st->postfilter_gain; + st->postfilter_tapset_old = st->postfilter_tapset; + st->postfilter_period = postfilter_pitch; + st->postfilter_gain = postfilter_gain; + st->postfilter_tapset = postfilter_tapset; + if (LM!=0) + { + st->postfilter_period_old = st->postfilter_period; + st->postfilter_gain_old = st->postfilter_gain; + st->postfilter_tapset_old = st->postfilter_tapset; + } + + if (C==1) + OPUS_COPY(&oldBandE[nbEBands], oldBandE, nbEBands); + + if (!isTransient) + { + OPUS_COPY(oldLogE2, oldLogE, 2*nbEBands); + OPUS_COPY(oldLogE, oldBandE, 2*nbEBands); + } else { + for (i=0;i<2*nbEBands;i++) + oldLogE[i] = MIN16(oldLogE[i], oldBandE[i]); + } + /* In normal circumstances, we only allow the noise floor to increase by + up to 2.4 dB/second, but when we're in DTX we give the weight of + all missing packets to the update packet. */ + max_background_increase = IMIN(160, st->loss_duration+M)*QCONST16(0.001f,DB_SHIFT); + for (i=0;i<2*nbEBands;i++) + backgroundLogE[i] = MIN16(backgroundLogE[i] + max_background_increase, oldBandE[i]); + /* In case start or end were to change */ + c=0; do + { + for (i=0;i<start;i++) + { + oldBandE[c*nbEBands+i]=0; + oldLogE[c*nbEBands+i]=oldLogE2[c*nbEBands+i]=-QCONST16(28.f,DB_SHIFT); + } + for (i=end;i<nbEBands;i++) + { + oldBandE[c*nbEBands+i]=0; + oldLogE[c*nbEBands+i]=oldLogE2[c*nbEBands+i]=-QCONST16(28.f,DB_SHIFT); + } + } while (++c<2); + st->rng = dec->rng; + + deemphasis(out_syn, pcm, N, CC, st->downsample, mode->preemph, st->preemph_memD, accum); + st->loss_duration = 0; + RESTORE_STACK; + if (ec_tell(dec) > 8*len) + return OPUS_INTERNAL_ERROR; + if(ec_get_error(dec)) + st->error = 1; + return frame_size/st->downsample; +} + + +#ifdef CUSTOM_MODES + +#ifdef FIXED_POINT +int opus_custom_decode(CELTDecoder * OPUS_RESTRICT st, const unsigned char *data, int len, opus_int16 * OPUS_RESTRICT pcm, int frame_size) +{ + return celt_decode_with_ec(st, data, len, pcm, frame_size, NULL, 0); +} + +#ifndef DISABLE_FLOAT_API +int opus_custom_decode_float(CELTDecoder * OPUS_RESTRICT st, const unsigned char *data, int len, float * OPUS_RESTRICT pcm, int frame_size) +{ + int j, ret, C, N; + VARDECL(opus_int16, out); + ALLOC_STACK; + + if (pcm==NULL) + return OPUS_BAD_ARG; + + C = st->channels; + N = frame_size; + + ALLOC(out, C*N, opus_int16); + ret=celt_decode_with_ec(st, data, len, out, frame_size, NULL, 0); + if (ret>0) + for (j=0;j<C*ret;j++) + pcm[j]=out[j]*(1.f/32768.f); + + RESTORE_STACK; + return ret; +} +#endif /* DISABLE_FLOAT_API */ + +#else + +int opus_custom_decode_float(CELTDecoder * OPUS_RESTRICT st, const unsigned char *data, int len, float * OPUS_RESTRICT pcm, int frame_size) +{ + return celt_decode_with_ec(st, data, len, pcm, frame_size, NULL, 0); +} + +int opus_custom_decode(CELTDecoder * OPUS_RESTRICT st, const unsigned char *data, int len, opus_int16 * OPUS_RESTRICT pcm, int frame_size) +{ + int j, ret, C, N; + VARDECL(celt_sig, out); + ALLOC_STACK; + + if (pcm==NULL) + return OPUS_BAD_ARG; + + C = st->channels; + N = frame_size; + ALLOC(out, C*N, celt_sig); + + ret=celt_decode_with_ec(st, data, len, out, frame_size, NULL, 0); + + if (ret>0) + for (j=0;j<C*ret;j++) + pcm[j] = FLOAT2INT16 (out[j]); + + RESTORE_STACK; + return ret; +} + +#endif +#endif /* CUSTOM_MODES */ + +int opus_custom_decoder_ctl(CELTDecoder * OPUS_RESTRICT st, int request, ...) +{ + va_list ap; + + va_start(ap, request); + switch (request) + { + case CELT_SET_START_BAND_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value<0 || value>=st->mode->nbEBands) + goto bad_arg; + st->start = value; + } + break; + case CELT_SET_END_BAND_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value<1 || value>st->mode->nbEBands) + goto bad_arg; + st->end = value; + } + break; + case CELT_SET_CHANNELS_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value<1 || value>2) + goto bad_arg; + st->stream_channels = value; + } + break; + case CELT_GET_AND_CLEAR_ERROR_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (value==NULL) + goto bad_arg; + *value=st->error; + st->error = 0; + } + break; + case OPUS_GET_LOOKAHEAD_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (value==NULL) + goto bad_arg; + *value = st->overlap/st->downsample; + } + break; + case OPUS_RESET_STATE: + { + int i; + opus_val16 *lpc, *oldBandE, *oldLogE, *oldLogE2; + lpc = (opus_val16*)(st->_decode_mem+(DECODE_BUFFER_SIZE+st->overlap)*st->channels); + oldBandE = lpc+st->channels*LPC_ORDER; + oldLogE = oldBandE + 2*st->mode->nbEBands; + oldLogE2 = oldLogE + 2*st->mode->nbEBands; + OPUS_CLEAR((char*)&st->DECODER_RESET_START, + opus_custom_decoder_get_size(st->mode, st->channels)- + ((char*)&st->DECODER_RESET_START - (char*)st)); + for (i=0;i<2*st->mode->nbEBands;i++) + oldLogE[i]=oldLogE2[i]=-QCONST16(28.f,DB_SHIFT); + st->skip_plc = 1; + } + break; + case OPUS_GET_PITCH_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (value==NULL) + goto bad_arg; + *value = st->postfilter_period; + } + break; + case CELT_GET_MODE_REQUEST: + { + const CELTMode ** value = va_arg(ap, const CELTMode**); + if (value==0) + goto bad_arg; + *value=st->mode; + } + break; + case CELT_SET_SIGNALLING_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + st->signalling = value; + } + break; + case OPUS_GET_FINAL_RANGE_REQUEST: + { + opus_uint32 * value = va_arg(ap, opus_uint32 *); + if (value==0) + goto bad_arg; + *value=st->rng; + } + break; + case OPUS_SET_PHASE_INVERSION_DISABLED_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if(value<0 || value>1) + { + goto bad_arg; + } + st->disable_inv = value; + } + break; + case OPUS_GET_PHASE_INVERSION_DISABLED_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->disable_inv; + } + break; + default: + goto bad_request; + } + va_end(ap); + return OPUS_OK; +bad_arg: + va_end(ap); + return OPUS_BAD_ARG; +bad_request: + va_end(ap); + return OPUS_UNIMPLEMENTED; +} diff --git a/media/libopus/celt/celt_encoder.c b/media/libopus/celt/celt_encoder.c new file mode 100644 index 0000000000..637d442cf7 --- /dev/null +++ b/media/libopus/celt/celt_encoder.c @@ -0,0 +1,2613 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2010 Xiph.Org Foundation + Copyright (c) 2008 Gregory Maxwell + Written by Jean-Marc Valin and Gregory Maxwell */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#define CELT_ENCODER_C + +#include "cpu_support.h" +#include "os_support.h" +#include "mdct.h" +#include <math.h> +#include "celt.h" +#include "pitch.h" +#include "bands.h" +#include "modes.h" +#include "entcode.h" +#include "quant_bands.h" +#include "rate.h" +#include "stack_alloc.h" +#include "mathops.h" +#include "float_cast.h" +#include <stdarg.h> +#include "celt_lpc.h" +#include "vq.h" + + +/** Encoder state + @brief Encoder state + */ +struct OpusCustomEncoder { + const OpusCustomMode *mode; /**< Mode used by the encoder */ + int channels; + int stream_channels; + + int force_intra; + int clip; + int disable_pf; + int complexity; + int upsample; + int start, end; + + opus_int32 bitrate; + int vbr; + int signalling; + int constrained_vbr; /* If zero, VBR can do whatever it likes with the rate */ + int loss_rate; + int lsb_depth; + int lfe; + int disable_inv; + int arch; + + /* Everything beyond this point gets cleared on a reset */ +#define ENCODER_RESET_START rng + + opus_uint32 rng; + int spread_decision; + opus_val32 delayedIntra; + int tonal_average; + int lastCodedBands; + int hf_average; + int tapset_decision; + + int prefilter_period; + opus_val16 prefilter_gain; + int prefilter_tapset; +#ifdef RESYNTH + int prefilter_period_old; + opus_val16 prefilter_gain_old; + int prefilter_tapset_old; +#endif + int consec_transient; + AnalysisInfo analysis; + SILKInfo silk_info; + + opus_val32 preemph_memE[2]; + opus_val32 preemph_memD[2]; + + /* VBR-related parameters */ + opus_int32 vbr_reservoir; + opus_int32 vbr_drift; + opus_int32 vbr_offset; + opus_int32 vbr_count; + opus_val32 overlap_max; + opus_val16 stereo_saving; + int intensity; + opus_val16 *energy_mask; + opus_val16 spec_avg; + +#ifdef RESYNTH + /* +MAX_PERIOD/2 to make space for overlap */ + celt_sig syn_mem[2][2*MAX_PERIOD+MAX_PERIOD/2]; +#endif + + celt_sig in_mem[1]; /* Size = channels*mode->overlap */ + /* celt_sig prefilter_mem[], Size = channels*COMBFILTER_MAXPERIOD */ + /* opus_val16 oldBandE[], Size = channels*mode->nbEBands */ + /* opus_val16 oldLogE[], Size = channels*mode->nbEBands */ + /* opus_val16 oldLogE2[], Size = channels*mode->nbEBands */ + /* opus_val16 energyError[], Size = channels*mode->nbEBands */ +}; + +int celt_encoder_get_size(int channels) +{ + CELTMode *mode = opus_custom_mode_create(48000, 960, NULL); + return opus_custom_encoder_get_size(mode, channels); +} + +OPUS_CUSTOM_NOSTATIC int opus_custom_encoder_get_size(const CELTMode *mode, int channels) +{ + int size = sizeof(struct CELTEncoder) + + (channels*mode->overlap-1)*sizeof(celt_sig) /* celt_sig in_mem[channels*mode->overlap]; */ + + channels*COMBFILTER_MAXPERIOD*sizeof(celt_sig) /* celt_sig prefilter_mem[channels*COMBFILTER_MAXPERIOD]; */ + + 4*channels*mode->nbEBands*sizeof(opus_val16); /* opus_val16 oldBandE[channels*mode->nbEBands]; */ + /* opus_val16 oldLogE[channels*mode->nbEBands]; */ + /* opus_val16 oldLogE2[channels*mode->nbEBands]; */ + /* opus_val16 energyError[channels*mode->nbEBands]; */ + return size; +} + +#ifdef CUSTOM_MODES +CELTEncoder *opus_custom_encoder_create(const CELTMode *mode, int channels, int *error) +{ + int ret; + CELTEncoder *st = (CELTEncoder *)opus_alloc(opus_custom_encoder_get_size(mode, channels)); + /* init will handle the NULL case */ + ret = opus_custom_encoder_init(st, mode, channels); + if (ret != OPUS_OK) + { + opus_custom_encoder_destroy(st); + st = NULL; + } + if (error) + *error = ret; + return st; +} +#endif /* CUSTOM_MODES */ + +static int opus_custom_encoder_init_arch(CELTEncoder *st, const CELTMode *mode, + int channels, int arch) +{ + if (channels < 0 || channels > 2) + return OPUS_BAD_ARG; + + if (st==NULL || mode==NULL) + return OPUS_ALLOC_FAIL; + + OPUS_CLEAR((char*)st, opus_custom_encoder_get_size(mode, channels)); + + st->mode = mode; + st->stream_channels = st->channels = channels; + + st->upsample = 1; + st->start = 0; + st->end = st->mode->effEBands; + st->signalling = 1; + st->arch = arch; + + st->constrained_vbr = 1; + st->clip = 1; + + st->bitrate = OPUS_BITRATE_MAX; + st->vbr = 0; + st->force_intra = 0; + st->complexity = 5; + st->lsb_depth=24; + + opus_custom_encoder_ctl(st, OPUS_RESET_STATE); + + return OPUS_OK; +} + +#ifdef CUSTOM_MODES +int opus_custom_encoder_init(CELTEncoder *st, const CELTMode *mode, int channels) +{ + return opus_custom_encoder_init_arch(st, mode, channels, opus_select_arch()); +} +#endif + +int celt_encoder_init(CELTEncoder *st, opus_int32 sampling_rate, int channels, + int arch) +{ + int ret; + ret = opus_custom_encoder_init_arch(st, + opus_custom_mode_create(48000, 960, NULL), channels, arch); + if (ret != OPUS_OK) + return ret; + st->upsample = resampling_factor(sampling_rate); + return OPUS_OK; +} + +#ifdef CUSTOM_MODES +void opus_custom_encoder_destroy(CELTEncoder *st) +{ + opus_free(st); +} +#endif /* CUSTOM_MODES */ + + +static int transient_analysis(const opus_val32 * OPUS_RESTRICT in, int len, int C, + opus_val16 *tf_estimate, int *tf_chan, int allow_weak_transients, + int *weak_transient) +{ + int i; + VARDECL(opus_val16, tmp); + opus_val32 mem0,mem1; + int is_transient = 0; + opus_int32 mask_metric = 0; + int c; + opus_val16 tf_max; + int len2; + /* Forward masking: 6.7 dB/ms. */ +#ifdef FIXED_POINT + int forward_shift = 4; +#else + opus_val16 forward_decay = QCONST16(.0625f,15); +#endif + /* Table of 6*64/x, trained on real data to minimize the average error */ + static const unsigned char inv_table[128] = { + 255,255,156,110, 86, 70, 59, 51, 45, 40, 37, 33, 31, 28, 26, 25, + 23, 22, 21, 20, 19, 18, 17, 16, 16, 15, 15, 14, 13, 13, 12, 12, + 12, 12, 11, 11, 11, 10, 10, 10, 9, 9, 9, 9, 9, 9, 8, 8, + 8, 8, 8, 7, 7, 7, 7, 7, 7, 6, 6, 6, 6, 6, 6, 6, + 6, 6, 6, 6, 6, 6, 6, 6, 6, 5, 5, 5, 5, 5, 5, 5, + 5, 5, 5, 5, 5, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, + 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 3, 3, + 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 2, + }; + SAVE_STACK; + ALLOC(tmp, len, opus_val16); + + *weak_transient = 0; + /* For lower bitrates, let's be more conservative and have a forward masking + decay of 3.3 dB/ms. This avoids having to code transients at very low + bitrate (mostly for hybrid), which can result in unstable energy and/or + partial collapse. */ + if (allow_weak_transients) + { +#ifdef FIXED_POINT + forward_shift = 5; +#else + forward_decay = QCONST16(.03125f,15); +#endif + } + len2=len/2; + for (c=0;c<C;c++) + { + opus_val32 mean; + opus_int32 unmask=0; + opus_val32 norm; + opus_val16 maxE; + mem0=0; + mem1=0; + /* High-pass filter: (1 - 2*z^-1 + z^-2) / (1 - z^-1 + .5*z^-2) */ + for (i=0;i<len;i++) + { + opus_val32 x,y; + x = SHR32(in[i+c*len],SIG_SHIFT); + y = ADD32(mem0, x); +#ifdef FIXED_POINT + mem0 = mem1 + y - SHL32(x,1); + mem1 = x - SHR32(y,1); +#else + mem0 = mem1 + y - 2*x; + mem1 = x - .5f*y; +#endif + tmp[i] = SROUND16(y, 2); + /*printf("%f ", tmp[i]);*/ + } + /*printf("\n");*/ + /* First few samples are bad because we don't propagate the memory */ + OPUS_CLEAR(tmp, 12); + +#ifdef FIXED_POINT + /* Normalize tmp to max range */ + { + int shift=0; + shift = 14-celt_ilog2(MAX16(1, celt_maxabs16(tmp, len))); + if (shift!=0) + { + for (i=0;i<len;i++) + tmp[i] = SHL16(tmp[i], shift); + } + } +#endif + + mean=0; + mem0=0; + /* Grouping by two to reduce complexity */ + /* Forward pass to compute the post-echo threshold*/ + for (i=0;i<len2;i++) + { + opus_val16 x2 = PSHR32(MULT16_16(tmp[2*i],tmp[2*i]) + MULT16_16(tmp[2*i+1],tmp[2*i+1]),16); + mean += x2; +#ifdef FIXED_POINT + /* FIXME: Use PSHR16() instead */ + tmp[i] = mem0 + PSHR32(x2-mem0,forward_shift); +#else + tmp[i] = mem0 + MULT16_16_P15(forward_decay,x2-mem0); +#endif + mem0 = tmp[i]; + } + + mem0=0; + maxE=0; + /* Backward pass to compute the pre-echo threshold */ + for (i=len2-1;i>=0;i--) + { + /* Backward masking: 13.9 dB/ms. */ +#ifdef FIXED_POINT + /* FIXME: Use PSHR16() instead */ + tmp[i] = mem0 + PSHR32(tmp[i]-mem0,3); +#else + tmp[i] = mem0 + MULT16_16_P15(QCONST16(0.125f,15),tmp[i]-mem0); +#endif + mem0 = tmp[i]; + maxE = MAX16(maxE, mem0); + } + /*for (i=0;i<len2;i++)printf("%f ", tmp[i]/mean);printf("\n");*/ + + /* Compute the ratio of the "frame energy" over the harmonic mean of the energy. + This essentially corresponds to a bitrate-normalized temporal noise-to-mask + ratio */ + + /* As a compromise with the old transient detector, frame energy is the + geometric mean of the energy and half the max */ +#ifdef FIXED_POINT + /* Costs two sqrt() to avoid overflows */ + mean = MULT16_16(celt_sqrt(mean), celt_sqrt(MULT16_16(maxE,len2>>1))); +#else + mean = celt_sqrt(mean * maxE*.5*len2); +#endif + /* Inverse of the mean energy in Q15+6 */ + norm = SHL32(EXTEND32(len2),6+14)/ADD32(EPSILON,SHR32(mean,1)); + /* Compute harmonic mean discarding the unreliable boundaries + The data is smooth, so we only take 1/4th of the samples */ + unmask=0; + /* We should never see NaNs here. If we find any, then something really bad happened and we better abort + before it does any damage later on. If these asserts are disabled (no hardening), then the table + lookup a few lines below (id = ...) is likely to crash dur to an out-of-bounds read. DO NOT FIX + that crash on NaN since it could result in a worse issue later on. */ + celt_assert(!celt_isnan(tmp[0])); + celt_assert(!celt_isnan(norm)); + for (i=12;i<len2-5;i+=4) + { + int id; +#ifdef FIXED_POINT + id = MAX32(0,MIN32(127,MULT16_32_Q15(tmp[i]+EPSILON,norm))); /* Do not round to nearest */ +#else + id = (int)MAX32(0,MIN32(127,floor(64*norm*(tmp[i]+EPSILON)))); /* Do not round to nearest */ +#endif + unmask += inv_table[id]; + } + /*printf("%d\n", unmask);*/ + /* Normalize, compensate for the 1/4th of the sample and the factor of 6 in the inverse table */ + unmask = 64*unmask*4/(6*(len2-17)); + if (unmask>mask_metric) + { + *tf_chan = c; + mask_metric = unmask; + } + } + is_transient = mask_metric>200; + /* For low bitrates, define "weak transients" that need to be + handled differently to avoid partial collapse. */ + if (allow_weak_transients && is_transient && mask_metric<600) { + is_transient = 0; + *weak_transient = 1; + } + /* Arbitrary metric for VBR boost */ + tf_max = MAX16(0,celt_sqrt(27*mask_metric)-42); + /* *tf_estimate = 1 + MIN16(1, sqrt(MAX16(0, tf_max-30))/20); */ + *tf_estimate = celt_sqrt(MAX32(0, SHL32(MULT16_16(QCONST16(0.0069,14),MIN16(163,tf_max)),14)-QCONST32(0.139,28))); + /*printf("%d %f\n", tf_max, mask_metric);*/ + RESTORE_STACK; +#ifdef FUZZING + is_transient = rand()&0x1; +#endif + /*printf("%d %f %d\n", is_transient, (float)*tf_estimate, tf_max);*/ + return is_transient; +} + +/* Looks for sudden increases of energy to decide whether we need to patch + the transient decision */ +static int patch_transient_decision(opus_val16 *newE, opus_val16 *oldE, int nbEBands, + int start, int end, int C) +{ + int i, c; + opus_val32 mean_diff=0; + opus_val16 spread_old[26]; + /* Apply an aggressive (-6 dB/Bark) spreading function to the old frame to + avoid false detection caused by irrelevant bands */ + if (C==1) + { + spread_old[start] = oldE[start]; + for (i=start+1;i<end;i++) + spread_old[i] = MAX16(spread_old[i-1]-QCONST16(1.0f, DB_SHIFT), oldE[i]); + } else { + spread_old[start] = MAX16(oldE[start],oldE[start+nbEBands]); + for (i=start+1;i<end;i++) + spread_old[i] = MAX16(spread_old[i-1]-QCONST16(1.0f, DB_SHIFT), + MAX16(oldE[i],oldE[i+nbEBands])); + } + for (i=end-2;i>=start;i--) + spread_old[i] = MAX16(spread_old[i], spread_old[i+1]-QCONST16(1.0f, DB_SHIFT)); + /* Compute mean increase */ + c=0; do { + for (i=IMAX(2,start);i<end-1;i++) + { + opus_val16 x1, x2; + x1 = MAX16(0, newE[i + c*nbEBands]); + x2 = MAX16(0, spread_old[i]); + mean_diff = ADD32(mean_diff, EXTEND32(MAX16(0, SUB16(x1, x2)))); + } + } while (++c<C); + mean_diff = DIV32(mean_diff, C*(end-1-IMAX(2,start))); + /*printf("%f %f %d\n", mean_diff, max_diff, count);*/ + return mean_diff > QCONST16(1.f, DB_SHIFT); +} + +/** Apply window and compute the MDCT for all sub-frames and + all channels in a frame */ +static void compute_mdcts(const CELTMode *mode, int shortBlocks, celt_sig * OPUS_RESTRICT in, + celt_sig * OPUS_RESTRICT out, int C, int CC, int LM, int upsample, + int arch) +{ + const int overlap = mode->overlap; + int N; + int B; + int shift; + int i, b, c; + if (shortBlocks) + { + B = shortBlocks; + N = mode->shortMdctSize; + shift = mode->maxLM; + } else { + B = 1; + N = mode->shortMdctSize<<LM; + shift = mode->maxLM-LM; + } + c=0; do { + for (b=0;b<B;b++) + { + /* Interleaving the sub-frames while doing the MDCTs */ + clt_mdct_forward(&mode->mdct, in+c*(B*N+overlap)+b*N, + &out[b+c*N*B], mode->window, overlap, shift, B, + arch); + } + } while (++c<CC); + if (CC==2&&C==1) + { + for (i=0;i<B*N;i++) + out[i] = ADD32(HALF32(out[i]), HALF32(out[B*N+i])); + } + if (upsample != 1) + { + c=0; do + { + int bound = B*N/upsample; + for (i=0;i<bound;i++) + out[c*B*N+i] *= upsample; + OPUS_CLEAR(&out[c*B*N+bound], B*N-bound); + } while (++c<C); + } +} + + +void celt_preemphasis(const opus_val16 * OPUS_RESTRICT pcmp, celt_sig * OPUS_RESTRICT inp, + int N, int CC, int upsample, const opus_val16 *coef, celt_sig *mem, int clip) +{ + int i; + opus_val16 coef0; + celt_sig m; + int Nu; + + coef0 = coef[0]; + m = *mem; + + /* Fast path for the normal 48kHz case and no clipping */ + if (coef[1] == 0 && upsample == 1 && !clip) + { + for (i=0;i<N;i++) + { + opus_val16 x; + x = SCALEIN(pcmp[CC*i]); + /* Apply pre-emphasis */ + inp[i] = SHL32(x, SIG_SHIFT) - m; + m = SHR32(MULT16_16(coef0, x), 15-SIG_SHIFT); + } + *mem = m; + return; + } + + Nu = N/upsample; + if (upsample!=1) + { + OPUS_CLEAR(inp, N); + } + for (i=0;i<Nu;i++) + inp[i*upsample] = SCALEIN(pcmp[CC*i]); + +#ifndef FIXED_POINT + if (clip) + { + /* Clip input to avoid encoding non-portable files */ + for (i=0;i<Nu;i++) + inp[i*upsample] = MAX32(-65536.f, MIN32(65536.f,inp[i*upsample])); + } +#else + (void)clip; /* Avoids a warning about clip being unused. */ +#endif +#ifdef CUSTOM_MODES + if (coef[1] != 0) + { + opus_val16 coef1 = coef[1]; + opus_val16 coef2 = coef[2]; + for (i=0;i<N;i++) + { + celt_sig x, tmp; + x = inp[i]; + /* Apply pre-emphasis */ + tmp = MULT16_16(coef2, x); + inp[i] = tmp + m; + m = MULT16_32_Q15(coef1, inp[i]) - MULT16_32_Q15(coef0, tmp); + } + } else +#endif + { + for (i=0;i<N;i++) + { + opus_val16 x; + x = inp[i]; + /* Apply pre-emphasis */ + inp[i] = SHL32(x, SIG_SHIFT) - m; + m = SHR32(MULT16_16(coef0, x), 15-SIG_SHIFT); + } + } + *mem = m; +} + + + +static opus_val32 l1_metric(const celt_norm *tmp, int N, int LM, opus_val16 bias) +{ + int i; + opus_val32 L1; + L1 = 0; + for (i=0;i<N;i++) + L1 += EXTEND32(ABS16(tmp[i])); + /* When in doubt, prefer good freq resolution */ + L1 = MAC16_32_Q15(L1, LM*bias, L1); + return L1; + +} + +static int tf_analysis(const CELTMode *m, int len, int isTransient, + int *tf_res, int lambda, celt_norm *X, int N0, int LM, + opus_val16 tf_estimate, int tf_chan, int *importance) +{ + int i; + VARDECL(int, metric); + int cost0; + int cost1; + VARDECL(int, path0); + VARDECL(int, path1); + VARDECL(celt_norm, tmp); + VARDECL(celt_norm, tmp_1); + int sel; + int selcost[2]; + int tf_select=0; + opus_val16 bias; + + SAVE_STACK; + bias = MULT16_16_Q14(QCONST16(.04f,15), MAX16(-QCONST16(.25f,14), QCONST16(.5f,14)-tf_estimate)); + /*printf("%f ", bias);*/ + + ALLOC(metric, len, int); + ALLOC(tmp, (m->eBands[len]-m->eBands[len-1])<<LM, celt_norm); + ALLOC(tmp_1, (m->eBands[len]-m->eBands[len-1])<<LM, celt_norm); + ALLOC(path0, len, int); + ALLOC(path1, len, int); + + for (i=0;i<len;i++) + { + int k, N; + int narrow; + opus_val32 L1, best_L1; + int best_level=0; + N = (m->eBands[i+1]-m->eBands[i])<<LM; + /* band is too narrow to be split down to LM=-1 */ + narrow = (m->eBands[i+1]-m->eBands[i])==1; + OPUS_COPY(tmp, &X[tf_chan*N0 + (m->eBands[i]<<LM)], N); + /* Just add the right channel if we're in stereo */ + /*if (C==2) + for (j=0;j<N;j++) + tmp[j] = ADD16(SHR16(tmp[j], 1),SHR16(X[N0+j+(m->eBands[i]<<LM)], 1));*/ + L1 = l1_metric(tmp, N, isTransient ? LM : 0, bias); + best_L1 = L1; + /* Check the -1 case for transients */ + if (isTransient && !narrow) + { + OPUS_COPY(tmp_1, tmp, N); + haar1(tmp_1, N>>LM, 1<<LM); + L1 = l1_metric(tmp_1, N, LM+1, bias); + if (L1<best_L1) + { + best_L1 = L1; + best_level = -1; + } + } + /*printf ("%f ", L1);*/ + for (k=0;k<LM+!(isTransient||narrow);k++) + { + int B; + + if (isTransient) + B = (LM-k-1); + else + B = k+1; + + haar1(tmp, N>>k, 1<<k); + + L1 = l1_metric(tmp, N, B, bias); + + if (L1 < best_L1) + { + best_L1 = L1; + best_level = k+1; + } + } + /*printf ("%d ", isTransient ? LM-best_level : best_level);*/ + /* metric is in Q1 to be able to select the mid-point (-0.5) for narrower bands */ + if (isTransient) + metric[i] = 2*best_level; + else + metric[i] = -2*best_level; + /* For bands that can't be split to -1, set the metric to the half-way point to avoid + biasing the decision */ + if (narrow && (metric[i]==0 || metric[i]==-2*LM)) + metric[i]-=1; + /*printf("%d ", metric[i]/2 + (!isTransient)*LM);*/ + } + /*printf("\n");*/ + /* Search for the optimal tf resolution, including tf_select */ + tf_select = 0; + for (sel=0;sel<2;sel++) + { + cost0 = importance[0]*abs(metric[0]-2*tf_select_table[LM][4*isTransient+2*sel+0]); + cost1 = importance[0]*abs(metric[0]-2*tf_select_table[LM][4*isTransient+2*sel+1]) + (isTransient ? 0 : lambda); + for (i=1;i<len;i++) + { + int curr0, curr1; + curr0 = IMIN(cost0, cost1 + lambda); + curr1 = IMIN(cost0 + lambda, cost1); + cost0 = curr0 + importance[i]*abs(metric[i]-2*tf_select_table[LM][4*isTransient+2*sel+0]); + cost1 = curr1 + importance[i]*abs(metric[i]-2*tf_select_table[LM][4*isTransient+2*sel+1]); + } + cost0 = IMIN(cost0, cost1); + selcost[sel]=cost0; + } + /* For now, we're conservative and only allow tf_select=1 for transients. + * If tests confirm it's useful for non-transients, we could allow it. */ + if (selcost[1]<selcost[0] && isTransient) + tf_select=1; + cost0 = importance[0]*abs(metric[0]-2*tf_select_table[LM][4*isTransient+2*tf_select+0]); + cost1 = importance[0]*abs(metric[0]-2*tf_select_table[LM][4*isTransient+2*tf_select+1]) + (isTransient ? 0 : lambda); + /* Viterbi forward pass */ + for (i=1;i<len;i++) + { + int curr0, curr1; + int from0, from1; + + from0 = cost0; + from1 = cost1 + lambda; + if (from0 < from1) + { + curr0 = from0; + path0[i]= 0; + } else { + curr0 = from1; + path0[i]= 1; + } + + from0 = cost0 + lambda; + from1 = cost1; + if (from0 < from1) + { + curr1 = from0; + path1[i]= 0; + } else { + curr1 = from1; + path1[i]= 1; + } + cost0 = curr0 + importance[i]*abs(metric[i]-2*tf_select_table[LM][4*isTransient+2*tf_select+0]); + cost1 = curr1 + importance[i]*abs(metric[i]-2*tf_select_table[LM][4*isTransient+2*tf_select+1]); + } + tf_res[len-1] = cost0 < cost1 ? 0 : 1; + /* Viterbi backward pass to check the decisions */ + for (i=len-2;i>=0;i--) + { + if (tf_res[i+1] == 1) + tf_res[i] = path1[i+1]; + else + tf_res[i] = path0[i+1]; + } + /*printf("%d %f\n", *tf_sum, tf_estimate);*/ + RESTORE_STACK; +#ifdef FUZZING + tf_select = rand()&0x1; + tf_res[0] = rand()&0x1; + for (i=1;i<len;i++) + tf_res[i] = tf_res[i-1] ^ ((rand()&0xF) == 0); +#endif + return tf_select; +} + +static void tf_encode(int start, int end, int isTransient, int *tf_res, int LM, int tf_select, ec_enc *enc) +{ + int curr, i; + int tf_select_rsv; + int tf_changed; + int logp; + opus_uint32 budget; + opus_uint32 tell; + budget = enc->storage*8; + tell = ec_tell(enc); + logp = isTransient ? 2 : 4; + /* Reserve space to code the tf_select decision. */ + tf_select_rsv = LM>0 && tell+logp+1 <= budget; + budget -= tf_select_rsv; + curr = tf_changed = 0; + for (i=start;i<end;i++) + { + if (tell+logp<=budget) + { + ec_enc_bit_logp(enc, tf_res[i] ^ curr, logp); + tell = ec_tell(enc); + curr = tf_res[i]; + tf_changed |= curr; + } + else + tf_res[i] = curr; + logp = isTransient ? 4 : 5; + } + /* Only code tf_select if it would actually make a difference. */ + if (tf_select_rsv && + tf_select_table[LM][4*isTransient+0+tf_changed]!= + tf_select_table[LM][4*isTransient+2+tf_changed]) + ec_enc_bit_logp(enc, tf_select, 1); + else + tf_select = 0; + for (i=start;i<end;i++) + tf_res[i] = tf_select_table[LM][4*isTransient+2*tf_select+tf_res[i]]; + /*for(i=0;i<end;i++)printf("%d ", isTransient ? tf_res[i] : LM+tf_res[i]);printf("\n");*/ +} + + +static int alloc_trim_analysis(const CELTMode *m, const celt_norm *X, + const opus_val16 *bandLogE, int end, int LM, int C, int N0, + AnalysisInfo *analysis, opus_val16 *stereo_saving, opus_val16 tf_estimate, + int intensity, opus_val16 surround_trim, opus_int32 equiv_rate, int arch) +{ + int i; + opus_val32 diff=0; + int c; + int trim_index; + opus_val16 trim = QCONST16(5.f, 8); + opus_val16 logXC, logXC2; + /* At low bitrate, reducing the trim seems to help. At higher bitrates, it's less + clear what's best, so we're keeping it as it was before, at least for now. */ + if (equiv_rate < 64000) { + trim = QCONST16(4.f, 8); + } else if (equiv_rate < 80000) { + opus_int32 frac = (equiv_rate-64000) >> 10; + trim = QCONST16(4.f, 8) + QCONST16(1.f/16.f, 8)*frac; + } + if (C==2) + { + opus_val16 sum = 0; /* Q10 */ + opus_val16 minXC; /* Q10 */ + /* Compute inter-channel correlation for low frequencies */ + for (i=0;i<8;i++) + { + opus_val32 partial; + partial = celt_inner_prod(&X[m->eBands[i]<<LM], &X[N0+(m->eBands[i]<<LM)], + (m->eBands[i+1]-m->eBands[i])<<LM, arch); + sum = ADD16(sum, EXTRACT16(SHR32(partial, 18))); + } + sum = MULT16_16_Q15(QCONST16(1.f/8, 15), sum); + sum = MIN16(QCONST16(1.f, 10), ABS16(sum)); + minXC = sum; + for (i=8;i<intensity;i++) + { + opus_val32 partial; + partial = celt_inner_prod(&X[m->eBands[i]<<LM], &X[N0+(m->eBands[i]<<LM)], + (m->eBands[i+1]-m->eBands[i])<<LM, arch); + minXC = MIN16(minXC, ABS16(EXTRACT16(SHR32(partial, 18)))); + } + minXC = MIN16(QCONST16(1.f, 10), ABS16(minXC)); + /*printf ("%f\n", sum);*/ + /* mid-side savings estimations based on the LF average*/ + logXC = celt_log2(QCONST32(1.001f, 20)-MULT16_16(sum, sum)); + /* mid-side savings estimations based on min correlation */ + logXC2 = MAX16(HALF16(logXC), celt_log2(QCONST32(1.001f, 20)-MULT16_16(minXC, minXC))); +#ifdef FIXED_POINT + /* Compensate for Q20 vs Q14 input and convert output to Q8 */ + logXC = PSHR32(logXC-QCONST16(6.f, DB_SHIFT),DB_SHIFT-8); + logXC2 = PSHR32(logXC2-QCONST16(6.f, DB_SHIFT),DB_SHIFT-8); +#endif + + trim += MAX16(-QCONST16(4.f, 8), MULT16_16_Q15(QCONST16(.75f,15),logXC)); + *stereo_saving = MIN16(*stereo_saving + QCONST16(0.25f, 8), -HALF16(logXC2)); + } + + /* Estimate spectral tilt */ + c=0; do { + for (i=0;i<end-1;i++) + { + diff += bandLogE[i+c*m->nbEBands]*(opus_int32)(2+2*i-end); + } + } while (++c<C); + diff /= C*(end-1); + /*printf("%f\n", diff);*/ + trim -= MAX32(-QCONST16(2.f, 8), MIN32(QCONST16(2.f, 8), SHR32(diff+QCONST16(1.f, DB_SHIFT),DB_SHIFT-8)/6 )); + trim -= SHR16(surround_trim, DB_SHIFT-8); + trim -= 2*SHR16(tf_estimate, 14-8); +#ifndef DISABLE_FLOAT_API + if (analysis->valid) + { + trim -= MAX16(-QCONST16(2.f, 8), MIN16(QCONST16(2.f, 8), + (opus_val16)(QCONST16(2.f, 8)*(analysis->tonality_slope+.05f)))); + } +#else + (void)analysis; +#endif + +#ifdef FIXED_POINT + trim_index = PSHR32(trim, 8); +#else + trim_index = (int)floor(.5f+trim); +#endif + trim_index = IMAX(0, IMIN(10, trim_index)); + /*printf("%d\n", trim_index);*/ +#ifdef FUZZING + trim_index = rand()%11; +#endif + return trim_index; +} + +static int stereo_analysis(const CELTMode *m, const celt_norm *X, + int LM, int N0) +{ + int i; + int thetas; + opus_val32 sumLR = EPSILON, sumMS = EPSILON; + + /* Use the L1 norm to model the entropy of the L/R signal vs the M/S signal */ + for (i=0;i<13;i++) + { + int j; + for (j=m->eBands[i]<<LM;j<m->eBands[i+1]<<LM;j++) + { + opus_val32 L, R, M, S; + /* We cast to 32-bit first because of the -32768 case */ + L = EXTEND32(X[j]); + R = EXTEND32(X[N0+j]); + M = ADD32(L, R); + S = SUB32(L, R); + sumLR = ADD32(sumLR, ADD32(ABS32(L), ABS32(R))); + sumMS = ADD32(sumMS, ADD32(ABS32(M), ABS32(S))); + } + } + sumMS = MULT16_32_Q15(QCONST16(0.707107f, 15), sumMS); + thetas = 13; + /* We don't need thetas for lower bands with LM<=1 */ + if (LM<=1) + thetas -= 8; + return MULT16_32_Q15((m->eBands[13]<<(LM+1))+thetas, sumMS) + > MULT16_32_Q15(m->eBands[13]<<(LM+1), sumLR); +} + +#define MSWAP(a,b) do {opus_val16 tmp = a;a=b;b=tmp;} while(0) +static opus_val16 median_of_5(const opus_val16 *x) +{ + opus_val16 t0, t1, t2, t3, t4; + t2 = x[2]; + if (x[0] > x[1]) + { + t0 = x[1]; + t1 = x[0]; + } else { + t0 = x[0]; + t1 = x[1]; + } + if (x[3] > x[4]) + { + t3 = x[4]; + t4 = x[3]; + } else { + t3 = x[3]; + t4 = x[4]; + } + if (t0 > t3) + { + MSWAP(t0, t3); + MSWAP(t1, t4); + } + if (t2 > t1) + { + if (t1 < t3) + return MIN16(t2, t3); + else + return MIN16(t4, t1); + } else { + if (t2 < t3) + return MIN16(t1, t3); + else + return MIN16(t2, t4); + } +} + +static opus_val16 median_of_3(const opus_val16 *x) +{ + opus_val16 t0, t1, t2; + if (x[0] > x[1]) + { + t0 = x[1]; + t1 = x[0]; + } else { + t0 = x[0]; + t1 = x[1]; + } + t2 = x[2]; + if (t1 < t2) + return t1; + else if (t0 < t2) + return t2; + else + return t0; +} + +static opus_val16 dynalloc_analysis(const opus_val16 *bandLogE, const opus_val16 *bandLogE2, + int nbEBands, int start, int end, int C, int *offsets, int lsb_depth, const opus_int16 *logN, + int isTransient, int vbr, int constrained_vbr, const opus_int16 *eBands, int LM, + int effectiveBytes, opus_int32 *tot_boost_, int lfe, opus_val16 *surround_dynalloc, + AnalysisInfo *analysis, int *importance, int *spread_weight) +{ + int i, c; + opus_int32 tot_boost=0; + opus_val16 maxDepth; + VARDECL(opus_val16, follower); + VARDECL(opus_val16, noise_floor); + SAVE_STACK; + ALLOC(follower, C*nbEBands, opus_val16); + ALLOC(noise_floor, C*nbEBands, opus_val16); + OPUS_CLEAR(offsets, nbEBands); + /* Dynamic allocation code */ + maxDepth=-QCONST16(31.9f, DB_SHIFT); + for (i=0;i<end;i++) + { + /* Noise floor must take into account eMeans, the depth, the width of the bands + and the preemphasis filter (approx. square of bark band ID) */ + noise_floor[i] = MULT16_16(QCONST16(0.0625f, DB_SHIFT),logN[i]) + +QCONST16(.5f,DB_SHIFT)+SHL16(9-lsb_depth,DB_SHIFT)-SHL16(eMeans[i],6) + +MULT16_16(QCONST16(.0062,DB_SHIFT),(i+5)*(i+5)); + } + c=0;do + { + for (i=0;i<end;i++) + maxDepth = MAX16(maxDepth, bandLogE[c*nbEBands+i]-noise_floor[i]); + } while (++c<C); + { + /* Compute a really simple masking model to avoid taking into account completely masked + bands when computing the spreading decision. */ + VARDECL(opus_val16, mask); + VARDECL(opus_val16, sig); + ALLOC(mask, nbEBands, opus_val16); + ALLOC(sig, nbEBands, opus_val16); + for (i=0;i<end;i++) + mask[i] = bandLogE[i]-noise_floor[i]; + if (C==2) + { + for (i=0;i<end;i++) + mask[i] = MAX16(mask[i], bandLogE[nbEBands+i]-noise_floor[i]); + } + OPUS_COPY(sig, mask, end); + for (i=1;i<end;i++) + mask[i] = MAX16(mask[i], mask[i-1] - QCONST16(2.f, DB_SHIFT)); + for (i=end-2;i>=0;i--) + mask[i] = MAX16(mask[i], mask[i+1] - QCONST16(3.f, DB_SHIFT)); + for (i=0;i<end;i++) + { + /* Compute SMR: Mask is never more than 72 dB below the peak and never below the noise floor.*/ + opus_val16 smr = sig[i]-MAX16(MAX16(0, maxDepth-QCONST16(12.f, DB_SHIFT)), mask[i]); + /* Clamp SMR to make sure we're not shifting by something negative or too large. */ +#ifdef FIXED_POINT + /* FIXME: Use PSHR16() instead */ + int shift = -PSHR32(MAX16(-QCONST16(5.f, DB_SHIFT), MIN16(0, smr)), DB_SHIFT); +#else + int shift = IMIN(5, IMAX(0, -(int)floor(.5f + smr))); +#endif + spread_weight[i] = 32 >> shift; + } + /*for (i=0;i<end;i++) + printf("%d ", spread_weight[i]); + printf("\n");*/ + } + /* Make sure that dynamic allocation can't make us bust the budget */ + if (effectiveBytes > 50 && LM>=1 && !lfe) + { + int last=0; + c=0;do + { + opus_val16 offset; + opus_val16 tmp; + opus_val16 *f; + f = &follower[c*nbEBands]; + f[0] = bandLogE2[c*nbEBands]; + for (i=1;i<end;i++) + { + /* The last band to be at least 3 dB higher than the previous one + is the last we'll consider. Otherwise, we run into problems on + bandlimited signals. */ + if (bandLogE2[c*nbEBands+i] > bandLogE2[c*nbEBands+i-1]+QCONST16(.5f,DB_SHIFT)) + last=i; + f[i] = MIN16(f[i-1]+QCONST16(1.5f,DB_SHIFT), bandLogE2[c*nbEBands+i]); + } + for (i=last-1;i>=0;i--) + f[i] = MIN16(f[i], MIN16(f[i+1]+QCONST16(2.f,DB_SHIFT), bandLogE2[c*nbEBands+i])); + + /* Combine with a median filter to avoid dynalloc triggering unnecessarily. + The "offset" value controls how conservative we are -- a higher offset + reduces the impact of the median filter and makes dynalloc use more bits. */ + offset = QCONST16(1.f, DB_SHIFT); + for (i=2;i<end-2;i++) + f[i] = MAX16(f[i], median_of_5(&bandLogE2[c*nbEBands+i-2])-offset); + tmp = median_of_3(&bandLogE2[c*nbEBands])-offset; + f[0] = MAX16(f[0], tmp); + f[1] = MAX16(f[1], tmp); + tmp = median_of_3(&bandLogE2[c*nbEBands+end-3])-offset; + f[end-2] = MAX16(f[end-2], tmp); + f[end-1] = MAX16(f[end-1], tmp); + + for (i=0;i<end;i++) + f[i] = MAX16(f[i], noise_floor[i]); + } while (++c<C); + if (C==2) + { + for (i=start;i<end;i++) + { + /* Consider 24 dB "cross-talk" */ + follower[nbEBands+i] = MAX16(follower[nbEBands+i], follower[ i]-QCONST16(4.f,DB_SHIFT)); + follower[ i] = MAX16(follower[ i], follower[nbEBands+i]-QCONST16(4.f,DB_SHIFT)); + follower[i] = HALF16(MAX16(0, bandLogE[i]-follower[i]) + MAX16(0, bandLogE[nbEBands+i]-follower[nbEBands+i])); + } + } else { + for (i=start;i<end;i++) + { + follower[i] = MAX16(0, bandLogE[i]-follower[i]); + } + } + for (i=start;i<end;i++) + follower[i] = MAX16(follower[i], surround_dynalloc[i]); + for (i=start;i<end;i++) + { +#ifdef FIXED_POINT + importance[i] = PSHR32(13*celt_exp2(MIN16(follower[i], QCONST16(4.f, DB_SHIFT))), 16); +#else + importance[i] = (int)floor(.5f+13*celt_exp2(MIN16(follower[i], QCONST16(4.f, DB_SHIFT)))); +#endif + } + /* For non-transient CBR/CVBR frames, halve the dynalloc contribution */ + if ((!vbr || constrained_vbr)&&!isTransient) + { + for (i=start;i<end;i++) + follower[i] = HALF16(follower[i]); + } + for (i=start;i<end;i++) + { + if (i<8) + follower[i] *= 2; + if (i>=12) + follower[i] = HALF16(follower[i]); + } +#ifdef DISABLE_FLOAT_API + (void)analysis; +#else + if (analysis->valid) + { + for (i=start;i<IMIN(LEAK_BANDS, end);i++) + follower[i] = follower[i] + QCONST16(1.f/64.f, DB_SHIFT)*analysis->leak_boost[i]; + } +#endif + for (i=start;i<end;i++) + { + int width; + int boost; + int boost_bits; + + follower[i] = MIN16(follower[i], QCONST16(4, DB_SHIFT)); + + width = C*(eBands[i+1]-eBands[i])<<LM; + if (width<6) + { + boost = (int)SHR32(EXTEND32(follower[i]),DB_SHIFT); + boost_bits = boost*width<<BITRES; + } else if (width > 48) { + boost = (int)SHR32(EXTEND32(follower[i])*8,DB_SHIFT); + boost_bits = (boost*width<<BITRES)/8; + } else { + boost = (int)SHR32(EXTEND32(follower[i])*width/6,DB_SHIFT); + boost_bits = boost*6<<BITRES; + } + /* For CBR and non-transient CVBR frames, limit dynalloc to 2/3 of the bits */ + if ((!vbr || (constrained_vbr&&!isTransient)) + && (tot_boost+boost_bits)>>BITRES>>3 > 2*effectiveBytes/3) + { + opus_int32 cap = ((2*effectiveBytes/3)<<BITRES<<3); + offsets[i] = cap-tot_boost; + tot_boost = cap; + break; + } else { + offsets[i] = boost; + tot_boost += boost_bits; + } + } + } else { + for (i=start;i<end;i++) + importance[i] = 13; + } + *tot_boost_ = tot_boost; + RESTORE_STACK; + return maxDepth; +} + + +static int run_prefilter(CELTEncoder *st, celt_sig *in, celt_sig *prefilter_mem, int CC, int N, + int prefilter_tapset, int *pitch, opus_val16 *gain, int *qgain, int enabled, int nbAvailableBytes, AnalysisInfo *analysis) +{ + int c; + VARDECL(celt_sig, _pre); + celt_sig *pre[2]; + const CELTMode *mode; + int pitch_index; + opus_val16 gain1; + opus_val16 pf_threshold; + int pf_on; + int qg; + int overlap; + SAVE_STACK; + + mode = st->mode; + overlap = mode->overlap; + ALLOC(_pre, CC*(N+COMBFILTER_MAXPERIOD), celt_sig); + + pre[0] = _pre; + pre[1] = _pre + (N+COMBFILTER_MAXPERIOD); + + + c=0; do { + OPUS_COPY(pre[c], prefilter_mem+c*COMBFILTER_MAXPERIOD, COMBFILTER_MAXPERIOD); + OPUS_COPY(pre[c]+COMBFILTER_MAXPERIOD, in+c*(N+overlap)+overlap, N); + } while (++c<CC); + + if (enabled) + { + VARDECL(opus_val16, pitch_buf); + ALLOC(pitch_buf, (COMBFILTER_MAXPERIOD+N)>>1, opus_val16); + + pitch_downsample(pre, pitch_buf, COMBFILTER_MAXPERIOD+N, CC, st->arch); + /* Don't search for the fir last 1.5 octave of the range because + there's too many false-positives due to short-term correlation */ + pitch_search(pitch_buf+(COMBFILTER_MAXPERIOD>>1), pitch_buf, N, + COMBFILTER_MAXPERIOD-3*COMBFILTER_MINPERIOD, &pitch_index, + st->arch); + pitch_index = COMBFILTER_MAXPERIOD-pitch_index; + + gain1 = remove_doubling(pitch_buf, COMBFILTER_MAXPERIOD, COMBFILTER_MINPERIOD, + N, &pitch_index, st->prefilter_period, st->prefilter_gain, st->arch); + if (pitch_index > COMBFILTER_MAXPERIOD-2) + pitch_index = COMBFILTER_MAXPERIOD-2; + gain1 = MULT16_16_Q15(QCONST16(.7f,15),gain1); + /*printf("%d %d %f %f\n", pitch_change, pitch_index, gain1, st->analysis.tonality);*/ + if (st->loss_rate>2) + gain1 = HALF32(gain1); + if (st->loss_rate>4) + gain1 = HALF32(gain1); + if (st->loss_rate>8) + gain1 = 0; + } else { + gain1 = 0; + pitch_index = COMBFILTER_MINPERIOD; + } +#ifndef DISABLE_FLOAT_API + if (analysis->valid) + gain1 = (opus_val16)(gain1 * analysis->max_pitch_ratio); +#else + (void)analysis; +#endif + /* Gain threshold for enabling the prefilter/postfilter */ + pf_threshold = QCONST16(.2f,15); + + /* Adjusting the threshold based on rate and continuity */ + if (abs(pitch_index-st->prefilter_period)*10>pitch_index) + pf_threshold += QCONST16(.2f,15); + if (nbAvailableBytes<25) + pf_threshold += QCONST16(.1f,15); + if (nbAvailableBytes<35) + pf_threshold += QCONST16(.1f,15); + if (st->prefilter_gain > QCONST16(.4f,15)) + pf_threshold -= QCONST16(.1f,15); + if (st->prefilter_gain > QCONST16(.55f,15)) + pf_threshold -= QCONST16(.1f,15); + + /* Hard threshold at 0.2 */ + pf_threshold = MAX16(pf_threshold, QCONST16(.2f,15)); + if (gain1<pf_threshold) + { + gain1 = 0; + pf_on = 0; + qg = 0; + } else { + /*This block is not gated by a total bits check only because + of the nbAvailableBytes check above.*/ + if (ABS16(gain1-st->prefilter_gain)<QCONST16(.1f,15)) + gain1=st->prefilter_gain; + +#ifdef FIXED_POINT + qg = ((gain1+1536)>>10)/3-1; +#else + qg = (int)floor(.5f+gain1*32/3)-1; +#endif + qg = IMAX(0, IMIN(7, qg)); + gain1 = QCONST16(0.09375f,15)*(qg+1); + pf_on = 1; + } + /*printf("%d %f\n", pitch_index, gain1);*/ + + c=0; do { + int offset = mode->shortMdctSize-overlap; + st->prefilter_period=IMAX(st->prefilter_period, COMBFILTER_MINPERIOD); + OPUS_COPY(in+c*(N+overlap), st->in_mem+c*(overlap), overlap); + if (offset) + comb_filter(in+c*(N+overlap)+overlap, pre[c]+COMBFILTER_MAXPERIOD, + st->prefilter_period, st->prefilter_period, offset, -st->prefilter_gain, -st->prefilter_gain, + st->prefilter_tapset, st->prefilter_tapset, NULL, 0, st->arch); + + comb_filter(in+c*(N+overlap)+overlap+offset, pre[c]+COMBFILTER_MAXPERIOD+offset, + st->prefilter_period, pitch_index, N-offset, -st->prefilter_gain, -gain1, + st->prefilter_tapset, prefilter_tapset, mode->window, overlap, st->arch); + OPUS_COPY(st->in_mem+c*(overlap), in+c*(N+overlap)+N, overlap); + + if (N>COMBFILTER_MAXPERIOD) + { + OPUS_COPY(prefilter_mem+c*COMBFILTER_MAXPERIOD, pre[c]+N, COMBFILTER_MAXPERIOD); + } else { + OPUS_MOVE(prefilter_mem+c*COMBFILTER_MAXPERIOD, prefilter_mem+c*COMBFILTER_MAXPERIOD+N, COMBFILTER_MAXPERIOD-N); + OPUS_COPY(prefilter_mem+c*COMBFILTER_MAXPERIOD+COMBFILTER_MAXPERIOD-N, pre[c]+COMBFILTER_MAXPERIOD, N); + } + } while (++c<CC); + + RESTORE_STACK; + *gain = gain1; + *pitch = pitch_index; + *qgain = qg; + return pf_on; +} + +static int compute_vbr(const CELTMode *mode, AnalysisInfo *analysis, opus_int32 base_target, + int LM, opus_int32 bitrate, int lastCodedBands, int C, int intensity, + int constrained_vbr, opus_val16 stereo_saving, int tot_boost, + opus_val16 tf_estimate, int pitch_change, opus_val16 maxDepth, + int lfe, int has_surround_mask, opus_val16 surround_masking, + opus_val16 temporal_vbr) +{ + /* The target rate in 8th bits per frame */ + opus_int32 target; + int coded_bins; + int coded_bands; + opus_val16 tf_calibration; + int nbEBands; + const opus_int16 *eBands; + + nbEBands = mode->nbEBands; + eBands = mode->eBands; + + coded_bands = lastCodedBands ? lastCodedBands : nbEBands; + coded_bins = eBands[coded_bands]<<LM; + if (C==2) + coded_bins += eBands[IMIN(intensity, coded_bands)]<<LM; + + target = base_target; + + /*printf("%f %f %f %f %d %d ", st->analysis.activity, st->analysis.tonality, tf_estimate, st->stereo_saving, tot_boost, coded_bands);*/ +#ifndef DISABLE_FLOAT_API + if (analysis->valid && analysis->activity<.4) + target -= (opus_int32)((coded_bins<<BITRES)*(.4f-analysis->activity)); +#endif + /* Stereo savings */ + if (C==2) + { + int coded_stereo_bands; + int coded_stereo_dof; + opus_val16 max_frac; + coded_stereo_bands = IMIN(intensity, coded_bands); + coded_stereo_dof = (eBands[coded_stereo_bands]<<LM)-coded_stereo_bands; + /* Maximum fraction of the bits we can save if the signal is mono. */ + max_frac = DIV32_16(MULT16_16(QCONST16(0.8f, 15), coded_stereo_dof), coded_bins); + stereo_saving = MIN16(stereo_saving, QCONST16(1.f, 8)); + /*printf("%d %d %d ", coded_stereo_dof, coded_bins, tot_boost);*/ + target -= (opus_int32)MIN32(MULT16_32_Q15(max_frac,target), + SHR32(MULT16_16(stereo_saving-QCONST16(0.1f,8),(coded_stereo_dof<<BITRES)),8)); + } + /* Boost the rate according to dynalloc (minus the dynalloc average for calibration). */ + target += tot_boost-(19<<LM); + /* Apply transient boost, compensating for average boost. */ + tf_calibration = QCONST16(0.044f,14); + target += (opus_int32)SHL32(MULT16_32_Q15(tf_estimate-tf_calibration, target),1); + +#ifndef DISABLE_FLOAT_API + /* Apply tonality boost */ + if (analysis->valid && !lfe) + { + opus_int32 tonal_target; + float tonal; + + /* Tonality boost (compensating for the average). */ + tonal = MAX16(0.f,analysis->tonality-.15f)-0.12f; + tonal_target = target + (opus_int32)((coded_bins<<BITRES)*1.2f*tonal); + if (pitch_change) + tonal_target += (opus_int32)((coded_bins<<BITRES)*.8f); + /*printf("%f %f ", analysis->tonality, tonal);*/ + target = tonal_target; + } +#else + (void)analysis; + (void)pitch_change; +#endif + + if (has_surround_mask&&!lfe) + { + opus_int32 surround_target = target + (opus_int32)SHR32(MULT16_16(surround_masking,coded_bins<<BITRES), DB_SHIFT); + /*printf("%f %d %d %d %d %d %d ", surround_masking, coded_bins, st->end, st->intensity, surround_target, target, st->bitrate);*/ + target = IMAX(target/4, surround_target); + } + + { + opus_int32 floor_depth; + int bins; + bins = eBands[nbEBands-2]<<LM; + /*floor_depth = SHR32(MULT16_16((C*bins<<BITRES),celt_log2(SHL32(MAX16(1,sample_max),13))), DB_SHIFT);*/ + floor_depth = (opus_int32)SHR32(MULT16_16((C*bins<<BITRES),maxDepth), DB_SHIFT); + floor_depth = IMAX(floor_depth, target>>2); + target = IMIN(target, floor_depth); + /*printf("%f %d\n", maxDepth, floor_depth);*/ + } + + /* Make VBR less aggressive for constrained VBR because we can't keep a higher bitrate + for long. Needs tuning. */ + if ((!has_surround_mask||lfe) && constrained_vbr) + { + target = base_target + (opus_int32)MULT16_32_Q15(QCONST16(0.67f, 15), target-base_target); + } + + if (!has_surround_mask && tf_estimate < QCONST16(.2f, 14)) + { + opus_val16 amount; + opus_val16 tvbr_factor; + amount = MULT16_16_Q15(QCONST16(.0000031f, 30), IMAX(0, IMIN(32000, 96000-bitrate))); + tvbr_factor = SHR32(MULT16_16(temporal_vbr, amount), DB_SHIFT); + target += (opus_int32)MULT16_32_Q15(tvbr_factor, target); + } + + /* Don't allow more than doubling the rate */ + target = IMIN(2*base_target, target); + + return target; +} + +int celt_encode_with_ec(CELTEncoder * OPUS_RESTRICT st, const opus_val16 * pcm, int frame_size, unsigned char *compressed, int nbCompressedBytes, ec_enc *enc) +{ + int i, c, N; + opus_int32 bits; + ec_enc _enc; + VARDECL(celt_sig, in); + VARDECL(celt_sig, freq); + VARDECL(celt_norm, X); + VARDECL(celt_ener, bandE); + VARDECL(opus_val16, bandLogE); + VARDECL(opus_val16, bandLogE2); + VARDECL(int, fine_quant); + VARDECL(opus_val16, error); + VARDECL(int, pulses); + VARDECL(int, cap); + VARDECL(int, offsets); + VARDECL(int, importance); + VARDECL(int, spread_weight); + VARDECL(int, fine_priority); + VARDECL(int, tf_res); + VARDECL(unsigned char, collapse_masks); + celt_sig *prefilter_mem; + opus_val16 *oldBandE, *oldLogE, *oldLogE2, *energyError; + int shortBlocks=0; + int isTransient=0; + const int CC = st->channels; + const int C = st->stream_channels; + int LM, M; + int tf_select; + int nbFilledBytes, nbAvailableBytes; + int start; + int end; + int effEnd; + int codedBands; + int alloc_trim; + int pitch_index=COMBFILTER_MINPERIOD; + opus_val16 gain1 = 0; + int dual_stereo=0; + int effectiveBytes; + int dynalloc_logp; + opus_int32 vbr_rate; + opus_int32 total_bits; + opus_int32 total_boost; + opus_int32 balance; + opus_int32 tell; + opus_int32 tell0_frac; + int prefilter_tapset=0; + int pf_on; + int anti_collapse_rsv; + int anti_collapse_on=0; + int silence=0; + int tf_chan = 0; + opus_val16 tf_estimate; + int pitch_change=0; + opus_int32 tot_boost; + opus_val32 sample_max; + opus_val16 maxDepth; + const OpusCustomMode *mode; + int nbEBands; + int overlap; + const opus_int16 *eBands; + int secondMdct; + int signalBandwidth; + int transient_got_disabled=0; + opus_val16 surround_masking=0; + opus_val16 temporal_vbr=0; + opus_val16 surround_trim = 0; + opus_int32 equiv_rate; + int hybrid; + int weak_transient = 0; + int enable_tf_analysis; + VARDECL(opus_val16, surround_dynalloc); + ALLOC_STACK; + + mode = st->mode; + nbEBands = mode->nbEBands; + overlap = mode->overlap; + eBands = mode->eBands; + start = st->start; + end = st->end; + hybrid = start != 0; + tf_estimate = 0; + if (nbCompressedBytes<2 || pcm==NULL) + { + RESTORE_STACK; + return OPUS_BAD_ARG; + } + + frame_size *= st->upsample; + for (LM=0;LM<=mode->maxLM;LM++) + if (mode->shortMdctSize<<LM==frame_size) + break; + if (LM>mode->maxLM) + { + RESTORE_STACK; + return OPUS_BAD_ARG; + } + M=1<<LM; + N = M*mode->shortMdctSize; + + prefilter_mem = st->in_mem+CC*(overlap); + oldBandE = (opus_val16*)(st->in_mem+CC*(overlap+COMBFILTER_MAXPERIOD)); + oldLogE = oldBandE + CC*nbEBands; + oldLogE2 = oldLogE + CC*nbEBands; + energyError = oldLogE2 + CC*nbEBands; + + if (enc==NULL) + { + tell0_frac=tell=1; + nbFilledBytes=0; + } else { + tell0_frac=ec_tell_frac(enc); + tell=ec_tell(enc); + nbFilledBytes=(tell+4)>>3; + } + +#ifdef CUSTOM_MODES + if (st->signalling && enc==NULL) + { + int tmp = (mode->effEBands-end)>>1; + end = st->end = IMAX(1, mode->effEBands-tmp); + compressed[0] = tmp<<5; + compressed[0] |= LM<<3; + compressed[0] |= (C==2)<<2; + /* Convert "standard mode" to Opus header */ + if (mode->Fs==48000 && mode->shortMdctSize==120) + { + int c0 = toOpus(compressed[0]); + if (c0<0) + { + RESTORE_STACK; + return OPUS_BAD_ARG; + } + compressed[0] = c0; + } + compressed++; + nbCompressedBytes--; + } +#else + celt_assert(st->signalling==0); +#endif + + /* Can't produce more than 1275 output bytes */ + nbCompressedBytes = IMIN(nbCompressedBytes,1275); + nbAvailableBytes = nbCompressedBytes - nbFilledBytes; + + if (st->vbr && st->bitrate!=OPUS_BITRATE_MAX) + { + opus_int32 den=mode->Fs>>BITRES; + vbr_rate=(st->bitrate*frame_size+(den>>1))/den; +#ifdef CUSTOM_MODES + if (st->signalling) + vbr_rate -= 8<<BITRES; +#endif + effectiveBytes = vbr_rate>>(3+BITRES); + } else { + opus_int32 tmp; + vbr_rate = 0; + tmp = st->bitrate*frame_size; + if (tell>1) + tmp += tell; + if (st->bitrate!=OPUS_BITRATE_MAX) + nbCompressedBytes = IMAX(2, IMIN(nbCompressedBytes, + (tmp+4*mode->Fs)/(8*mode->Fs)-!!st->signalling)); + effectiveBytes = nbCompressedBytes - nbFilledBytes; + } + equiv_rate = ((opus_int32)nbCompressedBytes*8*50 << (3-LM)) - (40*C+20)*((400>>LM) - 50); + if (st->bitrate != OPUS_BITRATE_MAX) + equiv_rate = IMIN(equiv_rate, st->bitrate - (40*C+20)*((400>>LM) - 50)); + + if (enc==NULL) + { + ec_enc_init(&_enc, compressed, nbCompressedBytes); + enc = &_enc; + } + + if (vbr_rate>0) + { + /* Computes the max bit-rate allowed in VBR mode to avoid violating the + target rate and buffering. + We must do this up front so that bust-prevention logic triggers + correctly if we don't have enough bits. */ + if (st->constrained_vbr) + { + opus_int32 vbr_bound; + opus_int32 max_allowed; + /* We could use any multiple of vbr_rate as bound (depending on the + delay). + This is clamped to ensure we use at least two bytes if the encoder + was entirely empty, but to allow 0 in hybrid mode. */ + vbr_bound = vbr_rate; + max_allowed = IMIN(IMAX(tell==1?2:0, + (vbr_rate+vbr_bound-st->vbr_reservoir)>>(BITRES+3)), + nbAvailableBytes); + if(max_allowed < nbAvailableBytes) + { + nbCompressedBytes = nbFilledBytes+max_allowed; + nbAvailableBytes = max_allowed; + ec_enc_shrink(enc, nbCompressedBytes); + } + } + } + total_bits = nbCompressedBytes*8; + + effEnd = end; + if (effEnd > mode->effEBands) + effEnd = mode->effEBands; + + ALLOC(in, CC*(N+overlap), celt_sig); + + sample_max=MAX32(st->overlap_max, celt_maxabs16(pcm, C*(N-overlap)/st->upsample)); + st->overlap_max=celt_maxabs16(pcm+C*(N-overlap)/st->upsample, C*overlap/st->upsample); + sample_max=MAX32(sample_max, st->overlap_max); +#ifdef FIXED_POINT + silence = (sample_max==0); +#else + silence = (sample_max <= (opus_val16)1/(1<<st->lsb_depth)); +#endif +#ifdef FUZZING + if ((rand()&0x3F)==0) + silence = 1; +#endif + if (tell==1) + ec_enc_bit_logp(enc, silence, 15); + else + silence=0; + if (silence) + { + /*In VBR mode there is no need to send more than the minimum. */ + if (vbr_rate>0) + { + effectiveBytes=nbCompressedBytes=IMIN(nbCompressedBytes, nbFilledBytes+2); + total_bits=nbCompressedBytes*8; + nbAvailableBytes=2; + ec_enc_shrink(enc, nbCompressedBytes); + } + /* Pretend we've filled all the remaining bits with zeros + (that's what the initialiser did anyway) */ + tell = nbCompressedBytes*8; + enc->nbits_total+=tell-ec_tell(enc); + } + c=0; do { + int need_clip=0; +#ifndef FIXED_POINT + need_clip = st->clip && sample_max>65536.f; +#endif + celt_preemphasis(pcm+c, in+c*(N+overlap)+overlap, N, CC, st->upsample, + mode->preemph, st->preemph_memE+c, need_clip); + } while (++c<CC); + + + + /* Find pitch period and gain */ + { + int enabled; + int qg; + enabled = ((st->lfe&&nbAvailableBytes>3) || nbAvailableBytes>12*C) && !hybrid && !silence && !st->disable_pf + && st->complexity >= 5; + + prefilter_tapset = st->tapset_decision; + pf_on = run_prefilter(st, in, prefilter_mem, CC, N, prefilter_tapset, &pitch_index, &gain1, &qg, enabled, nbAvailableBytes, &st->analysis); + if ((gain1 > QCONST16(.4f,15) || st->prefilter_gain > QCONST16(.4f,15)) && (!st->analysis.valid || st->analysis.tonality > .3) + && (pitch_index > 1.26*st->prefilter_period || pitch_index < .79*st->prefilter_period)) + pitch_change = 1; + if (pf_on==0) + { + if(!hybrid && tell+16<=total_bits) + ec_enc_bit_logp(enc, 0, 1); + } else { + /*This block is not gated by a total bits check only because + of the nbAvailableBytes check above.*/ + int octave; + ec_enc_bit_logp(enc, 1, 1); + pitch_index += 1; + octave = EC_ILOG(pitch_index)-5; + ec_enc_uint(enc, octave, 6); + ec_enc_bits(enc, pitch_index-(16<<octave), 4+octave); + pitch_index -= 1; + ec_enc_bits(enc, qg, 3); + ec_enc_icdf(enc, prefilter_tapset, tapset_icdf, 2); + } + } + + isTransient = 0; + shortBlocks = 0; + if (st->complexity >= 1 && !st->lfe) + { + /* Reduces the likelihood of energy instability on fricatives at low bitrate + in hybrid mode. It seems like we still want to have real transients on vowels + though (small SILK quantization offset value). */ + int allow_weak_transients = hybrid && effectiveBytes<15 && st->silk_info.signalType != 2; + isTransient = transient_analysis(in, N+overlap, CC, + &tf_estimate, &tf_chan, allow_weak_transients, &weak_transient); + } + if (LM>0 && ec_tell(enc)+3<=total_bits) + { + if (isTransient) + shortBlocks = M; + } else { + isTransient = 0; + transient_got_disabled=1; + } + + ALLOC(freq, CC*N, celt_sig); /**< Interleaved signal MDCTs */ + ALLOC(bandE,nbEBands*CC, celt_ener); + ALLOC(bandLogE,nbEBands*CC, opus_val16); + + secondMdct = shortBlocks && st->complexity>=8; + ALLOC(bandLogE2, C*nbEBands, opus_val16); + if (secondMdct) + { + compute_mdcts(mode, 0, in, freq, C, CC, LM, st->upsample, st->arch); + compute_band_energies(mode, freq, bandE, effEnd, C, LM, st->arch); + amp2Log2(mode, effEnd, end, bandE, bandLogE2, C); + for (c=0;c<C;c++) + { + for (i=0;i<end;i++) + bandLogE2[nbEBands*c+i] += HALF16(SHL16(LM, DB_SHIFT)); + } + } + + compute_mdcts(mode, shortBlocks, in, freq, C, CC, LM, st->upsample, st->arch); + /* This should catch any NaN in the CELT input. Since we're not supposed to see any (they're filtered + at the Opus layer), just abort. */ + celt_assert(!celt_isnan(freq[0]) && (C==1 || !celt_isnan(freq[N]))); + if (CC==2&&C==1) + tf_chan = 0; + compute_band_energies(mode, freq, bandE, effEnd, C, LM, st->arch); + + if (st->lfe) + { + for (i=2;i<end;i++) + { + bandE[i] = IMIN(bandE[i], MULT16_32_Q15(QCONST16(1e-4f,15),bandE[0])); + bandE[i] = MAX32(bandE[i], EPSILON); + } + } + amp2Log2(mode, effEnd, end, bandE, bandLogE, C); + + ALLOC(surround_dynalloc, C*nbEBands, opus_val16); + OPUS_CLEAR(surround_dynalloc, end); + /* This computes how much masking takes place between surround channels */ + if (!hybrid&&st->energy_mask&&!st->lfe) + { + int mask_end; + int midband; + int count_dynalloc; + opus_val32 mask_avg=0; + opus_val32 diff=0; + int count=0; + mask_end = IMAX(2,st->lastCodedBands); + for (c=0;c<C;c++) + { + for(i=0;i<mask_end;i++) + { + opus_val16 mask; + mask = MAX16(MIN16(st->energy_mask[nbEBands*c+i], + QCONST16(.25f, DB_SHIFT)), -QCONST16(2.0f, DB_SHIFT)); + if (mask > 0) + mask = HALF16(mask); + mask_avg += MULT16_16(mask, eBands[i+1]-eBands[i]); + count += eBands[i+1]-eBands[i]; + diff += MULT16_16(mask, 1+2*i-mask_end); + } + } + celt_assert(count>0); + mask_avg = DIV32_16(mask_avg,count); + mask_avg += QCONST16(.2f, DB_SHIFT); + diff = diff*6/(C*(mask_end-1)*(mask_end+1)*mask_end); + /* Again, being conservative */ + diff = HALF32(diff); + diff = MAX32(MIN32(diff, QCONST32(.031f, DB_SHIFT)), -QCONST32(.031f, DB_SHIFT)); + /* Find the band that's in the middle of the coded spectrum */ + for (midband=0;eBands[midband+1] < eBands[mask_end]/2;midband++); + count_dynalloc=0; + for(i=0;i<mask_end;i++) + { + opus_val32 lin; + opus_val16 unmask; + lin = mask_avg + diff*(i-midband); + if (C==2) + unmask = MAX16(st->energy_mask[i], st->energy_mask[nbEBands+i]); + else + unmask = st->energy_mask[i]; + unmask = MIN16(unmask, QCONST16(.0f, DB_SHIFT)); + unmask -= lin; + if (unmask > QCONST16(.25f, DB_SHIFT)) + { + surround_dynalloc[i] = unmask - QCONST16(.25f, DB_SHIFT); + count_dynalloc++; + } + } + if (count_dynalloc>=3) + { + /* If we need dynalloc in many bands, it's probably because our + initial masking rate was too low. */ + mask_avg += QCONST16(.25f, DB_SHIFT); + if (mask_avg>0) + { + /* Something went really wrong in the original calculations, + disabling masking. */ + mask_avg = 0; + diff = 0; + OPUS_CLEAR(surround_dynalloc, mask_end); + } else { + for(i=0;i<mask_end;i++) + surround_dynalloc[i] = MAX16(0, surround_dynalloc[i]-QCONST16(.25f, DB_SHIFT)); + } + } + mask_avg += QCONST16(.2f, DB_SHIFT); + /* Convert to 1/64th units used for the trim */ + surround_trim = 64*diff; + /*printf("%d %d ", mask_avg, surround_trim);*/ + surround_masking = mask_avg; + } + /* Temporal VBR (but not for LFE) */ + if (!st->lfe) + { + opus_val16 follow=-QCONST16(10.0f,DB_SHIFT); + opus_val32 frame_avg=0; + opus_val16 offset = shortBlocks?HALF16(SHL16(LM, DB_SHIFT)):0; + for(i=start;i<end;i++) + { + follow = MAX16(follow-QCONST16(1.f, DB_SHIFT), bandLogE[i]-offset); + if (C==2) + follow = MAX16(follow, bandLogE[i+nbEBands]-offset); + frame_avg += follow; + } + frame_avg /= (end-start); + temporal_vbr = SUB16(frame_avg,st->spec_avg); + temporal_vbr = MIN16(QCONST16(3.f, DB_SHIFT), MAX16(-QCONST16(1.5f, DB_SHIFT), temporal_vbr)); + st->spec_avg += MULT16_16_Q15(QCONST16(.02f, 15), temporal_vbr); + } + /*for (i=0;i<21;i++) + printf("%f ", bandLogE[i]); + printf("\n");*/ + + if (!secondMdct) + { + OPUS_COPY(bandLogE2, bandLogE, C*nbEBands); + } + + /* Last chance to catch any transient we might have missed in the + time-domain analysis */ + if (LM>0 && ec_tell(enc)+3<=total_bits && !isTransient && st->complexity>=5 && !st->lfe && !hybrid) + { + if (patch_transient_decision(bandLogE, oldBandE, nbEBands, start, end, C)) + { + isTransient = 1; + shortBlocks = M; + compute_mdcts(mode, shortBlocks, in, freq, C, CC, LM, st->upsample, st->arch); + compute_band_energies(mode, freq, bandE, effEnd, C, LM, st->arch); + amp2Log2(mode, effEnd, end, bandE, bandLogE, C); + /* Compensate for the scaling of short vs long mdcts */ + for (c=0;c<C;c++) + { + for (i=0;i<end;i++) + bandLogE2[nbEBands*c+i] += HALF16(SHL16(LM, DB_SHIFT)); + } + tf_estimate = QCONST16(.2f,14); + } + } + + if (LM>0 && ec_tell(enc)+3<=total_bits) + ec_enc_bit_logp(enc, isTransient, 3); + + ALLOC(X, C*N, celt_norm); /**< Interleaved normalised MDCTs */ + + /* Band normalisation */ + normalise_bands(mode, freq, X, bandE, effEnd, C, M); + + enable_tf_analysis = effectiveBytes>=15*C && !hybrid && st->complexity>=2 && !st->lfe; + + ALLOC(offsets, nbEBands, int); + ALLOC(importance, nbEBands, int); + ALLOC(spread_weight, nbEBands, int); + + maxDepth = dynalloc_analysis(bandLogE, bandLogE2, nbEBands, start, end, C, offsets, + st->lsb_depth, mode->logN, isTransient, st->vbr, st->constrained_vbr, + eBands, LM, effectiveBytes, &tot_boost, st->lfe, surround_dynalloc, &st->analysis, importance, spread_weight); + + ALLOC(tf_res, nbEBands, int); + /* Disable variable tf resolution for hybrid and at very low bitrate */ + if (enable_tf_analysis) + { + int lambda; + lambda = IMAX(80, 20480/effectiveBytes + 2); + tf_select = tf_analysis(mode, effEnd, isTransient, tf_res, lambda, X, N, LM, tf_estimate, tf_chan, importance); + for (i=effEnd;i<end;i++) + tf_res[i] = tf_res[effEnd-1]; + } else if (hybrid && weak_transient) + { + /* For weak transients, we rely on the fact that improving time resolution using + TF on a long window is imperfect and will not result in an energy collapse at + low bitrate. */ + for (i=0;i<end;i++) + tf_res[i] = 1; + tf_select=0; + } else if (hybrid && effectiveBytes<15 && st->silk_info.signalType != 2) + { + /* For low bitrate hybrid, we force temporal resolution to 5 ms rather than 2.5 ms. */ + for (i=0;i<end;i++) + tf_res[i] = 0; + tf_select=isTransient; + } else { + for (i=0;i<end;i++) + tf_res[i] = isTransient; + tf_select=0; + } + + ALLOC(error, C*nbEBands, opus_val16); + c=0; + do { + for (i=start;i<end;i++) + { + /* When the energy is stable, slightly bias energy quantization towards + the previous error to make the gain more stable (a constant offset is + better than fluctuations). */ + if (ABS32(SUB32(bandLogE[i+c*nbEBands], oldBandE[i+c*nbEBands])) < QCONST16(2.f, DB_SHIFT)) + { + bandLogE[i+c*nbEBands] -= MULT16_16_Q15(energyError[i+c*nbEBands], QCONST16(0.25f, 15)); + } + } + } while (++c < C); + quant_coarse_energy(mode, start, end, effEnd, bandLogE, + oldBandE, total_bits, error, enc, + C, LM, nbAvailableBytes, st->force_intra, + &st->delayedIntra, st->complexity >= 4, st->loss_rate, st->lfe); + + tf_encode(start, end, isTransient, tf_res, LM, tf_select, enc); + + if (ec_tell(enc)+4<=total_bits) + { + if (st->lfe) + { + st->tapset_decision = 0; + st->spread_decision = SPREAD_NORMAL; + } else if (hybrid) + { + if (st->complexity == 0) + st->spread_decision = SPREAD_NONE; + else if (isTransient) + st->spread_decision = SPREAD_NORMAL; + else + st->spread_decision = SPREAD_AGGRESSIVE; + } else if (shortBlocks || st->complexity < 3 || nbAvailableBytes < 10*C) + { + if (st->complexity == 0) + st->spread_decision = SPREAD_NONE; + else + st->spread_decision = SPREAD_NORMAL; + } else { + /* Disable new spreading+tapset estimator until we can show it works + better than the old one. So far it seems like spreading_decision() + works best. */ +#if 0 + if (st->analysis.valid) + { + static const opus_val16 spread_thresholds[3] = {-QCONST16(.6f, 15), -QCONST16(.2f, 15), -QCONST16(.07f, 15)}; + static const opus_val16 spread_histeresis[3] = {QCONST16(.15f, 15), QCONST16(.07f, 15), QCONST16(.02f, 15)}; + static const opus_val16 tapset_thresholds[2] = {QCONST16(.0f, 15), QCONST16(.15f, 15)}; + static const opus_val16 tapset_histeresis[2] = {QCONST16(.1f, 15), QCONST16(.05f, 15)}; + st->spread_decision = hysteresis_decision(-st->analysis.tonality, spread_thresholds, spread_histeresis, 3, st->spread_decision); + st->tapset_decision = hysteresis_decision(st->analysis.tonality_slope, tapset_thresholds, tapset_histeresis, 2, st->tapset_decision); + } else +#endif + { + st->spread_decision = spreading_decision(mode, X, + &st->tonal_average, st->spread_decision, &st->hf_average, + &st->tapset_decision, pf_on&&!shortBlocks, effEnd, C, M, spread_weight); + } + /*printf("%d %d\n", st->tapset_decision, st->spread_decision);*/ + /*printf("%f %d %f %d\n\n", st->analysis.tonality, st->spread_decision, st->analysis.tonality_slope, st->tapset_decision);*/ + } + ec_enc_icdf(enc, st->spread_decision, spread_icdf, 5); + } + + /* For LFE, everything interesting is in the first band */ + if (st->lfe) + offsets[0] = IMIN(8, effectiveBytes/3); + ALLOC(cap, nbEBands, int); + init_caps(mode,cap,LM,C); + + dynalloc_logp = 6; + total_bits<<=BITRES; + total_boost = 0; + tell = ec_tell_frac(enc); + for (i=start;i<end;i++) + { + int width, quanta; + int dynalloc_loop_logp; + int boost; + int j; + width = C*(eBands[i+1]-eBands[i])<<LM; + /* quanta is 6 bits, but no more than 1 bit/sample + and no less than 1/8 bit/sample */ + quanta = IMIN(width<<BITRES, IMAX(6<<BITRES, width)); + dynalloc_loop_logp = dynalloc_logp; + boost = 0; + for (j = 0; tell+(dynalloc_loop_logp<<BITRES) < total_bits-total_boost + && boost < cap[i]; j++) + { + int flag; + flag = j<offsets[i]; + ec_enc_bit_logp(enc, flag, dynalloc_loop_logp); + tell = ec_tell_frac(enc); + if (!flag) + break; + boost += quanta; + total_boost += quanta; + dynalloc_loop_logp = 1; + } + /* Making dynalloc more likely */ + if (j) + dynalloc_logp = IMAX(2, dynalloc_logp-1); + offsets[i] = boost; + } + + if (C==2) + { + static const opus_val16 intensity_thresholds[21]= + /* 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 off*/ + { 1, 2, 3, 4, 5, 6, 7, 8,16,24,36,44,50,56,62,67,72,79,88,106,134}; + static const opus_val16 intensity_histeresis[21]= + { 1, 1, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 2, 2, 3, 3, 4, 5, 6, 8, 8}; + + /* Always use MS for 2.5 ms frames until we can do a better analysis */ + if (LM!=0) + dual_stereo = stereo_analysis(mode, X, LM, N); + + st->intensity = hysteresis_decision((opus_val16)(equiv_rate/1000), + intensity_thresholds, intensity_histeresis, 21, st->intensity); + st->intensity = IMIN(end,IMAX(start, st->intensity)); + } + + alloc_trim = 5; + if (tell+(6<<BITRES) <= total_bits - total_boost) + { + if (start > 0 || st->lfe) + { + st->stereo_saving = 0; + alloc_trim = 5; + } else { + alloc_trim = alloc_trim_analysis(mode, X, bandLogE, + end, LM, C, N, &st->analysis, &st->stereo_saving, tf_estimate, + st->intensity, surround_trim, equiv_rate, st->arch); + } + ec_enc_icdf(enc, alloc_trim, trim_icdf, 7); + tell = ec_tell_frac(enc); + } + + /* Variable bitrate */ + if (vbr_rate>0) + { + opus_val16 alpha; + opus_int32 delta; + /* The target rate in 8th bits per frame */ + opus_int32 target, base_target; + opus_int32 min_allowed; + int lm_diff = mode->maxLM - LM; + + /* Don't attempt to use more than 510 kb/s, even for frames smaller than 20 ms. + The CELT allocator will just not be able to use more than that anyway. */ + nbCompressedBytes = IMIN(nbCompressedBytes,1275>>(3-LM)); + if (!hybrid) + { + base_target = vbr_rate - ((40*C+20)<<BITRES); + } else { + base_target = IMAX(0, vbr_rate - ((9*C+4)<<BITRES)); + } + + if (st->constrained_vbr) + base_target += (st->vbr_offset>>lm_diff); + + if (!hybrid) + { + target = compute_vbr(mode, &st->analysis, base_target, LM, equiv_rate, + st->lastCodedBands, C, st->intensity, st->constrained_vbr, + st->stereo_saving, tot_boost, tf_estimate, pitch_change, maxDepth, + st->lfe, st->energy_mask!=NULL, surround_masking, + temporal_vbr); + } else { + target = base_target; + /* Tonal frames (offset<100) need more bits than noisy (offset>100) ones. */ + if (st->silk_info.offset < 100) target += 12 << BITRES >> (3-LM); + if (st->silk_info.offset > 100) target -= 18 << BITRES >> (3-LM); + /* Boosting bitrate on transients and vowels with significant temporal + spikes. */ + target += (opus_int32)MULT16_16_Q14(tf_estimate-QCONST16(.25f,14), (50<<BITRES)); + /* If we have a strong transient, let's make sure it has enough bits to code + the first two bands, so that it can use folding rather than noise. */ + if (tf_estimate > QCONST16(.7f,14)) + target = IMAX(target, 50<<BITRES); + } + /* The current offset is removed from the target and the space used + so far is added*/ + target=target+tell; + /* In VBR mode the frame size must not be reduced so much that it would + result in the encoder running out of bits. + The margin of 2 bytes ensures that none of the bust-prevention logic + in the decoder will have triggered so far. */ + min_allowed = ((tell+total_boost+(1<<(BITRES+3))-1)>>(BITRES+3)) + 2; + /* Take into account the 37 bits we need to have left in the packet to + signal a redundant frame in hybrid mode. Creating a shorter packet would + create an entropy coder desync. */ + if (hybrid) + min_allowed = IMAX(min_allowed, (tell0_frac+(37<<BITRES)+total_boost+(1<<(BITRES+3))-1)>>(BITRES+3)); + + nbAvailableBytes = (target+(1<<(BITRES+2)))>>(BITRES+3); + nbAvailableBytes = IMAX(min_allowed,nbAvailableBytes); + nbAvailableBytes = IMIN(nbCompressedBytes,nbAvailableBytes); + + /* By how much did we "miss" the target on that frame */ + delta = target - vbr_rate; + + target=nbAvailableBytes<<(BITRES+3); + + /*If the frame is silent we don't adjust our drift, otherwise + the encoder will shoot to very high rates after hitting a + span of silence, but we do allow the bitres to refill. + This means that we'll undershoot our target in CVBR/VBR modes + on files with lots of silence. */ + if(silence) + { + nbAvailableBytes = 2; + target = 2*8<<BITRES; + delta = 0; + } + + if (st->vbr_count < 970) + { + st->vbr_count++; + alpha = celt_rcp(SHL32(EXTEND32(st->vbr_count+20),16)); + } else + alpha = QCONST16(.001f,15); + /* How many bits have we used in excess of what we're allowed */ + if (st->constrained_vbr) + st->vbr_reservoir += target - vbr_rate; + /*printf ("%d\n", st->vbr_reservoir);*/ + + /* Compute the offset we need to apply in order to reach the target */ + if (st->constrained_vbr) + { + st->vbr_drift += (opus_int32)MULT16_32_Q15(alpha,(delta*(1<<lm_diff))-st->vbr_offset-st->vbr_drift); + st->vbr_offset = -st->vbr_drift; + } + /*printf ("%d\n", st->vbr_drift);*/ + + if (st->constrained_vbr && st->vbr_reservoir < 0) + { + /* We're under the min value -- increase rate */ + int adjust = (-st->vbr_reservoir)/(8<<BITRES); + /* Unless we're just coding silence */ + nbAvailableBytes += silence?0:adjust; + st->vbr_reservoir = 0; + /*printf ("+%d\n", adjust);*/ + } + nbCompressedBytes = IMIN(nbCompressedBytes,nbAvailableBytes); + /*printf("%d\n", nbCompressedBytes*50*8);*/ + /* This moves the raw bits to take into account the new compressed size */ + ec_enc_shrink(enc, nbCompressedBytes); + } + + /* Bit allocation */ + ALLOC(fine_quant, nbEBands, int); + ALLOC(pulses, nbEBands, int); + ALLOC(fine_priority, nbEBands, int); + + /* bits = packet size - where we are - safety*/ + bits = (((opus_int32)nbCompressedBytes*8)<<BITRES) - ec_tell_frac(enc) - 1; + anti_collapse_rsv = isTransient&&LM>=2&&bits>=((LM+2)<<BITRES) ? (1<<BITRES) : 0; + bits -= anti_collapse_rsv; + signalBandwidth = end-1; +#ifndef DISABLE_FLOAT_API + if (st->analysis.valid) + { + int min_bandwidth; + if (equiv_rate < (opus_int32)32000*C) + min_bandwidth = 13; + else if (equiv_rate < (opus_int32)48000*C) + min_bandwidth = 16; + else if (equiv_rate < (opus_int32)60000*C) + min_bandwidth = 18; + else if (equiv_rate < (opus_int32)80000*C) + min_bandwidth = 19; + else + min_bandwidth = 20; + signalBandwidth = IMAX(st->analysis.bandwidth, min_bandwidth); + } +#endif + if (st->lfe) + signalBandwidth = 1; + codedBands = clt_compute_allocation(mode, start, end, offsets, cap, + alloc_trim, &st->intensity, &dual_stereo, bits, &balance, pulses, + fine_quant, fine_priority, C, LM, enc, 1, st->lastCodedBands, signalBandwidth); + if (st->lastCodedBands) + st->lastCodedBands = IMIN(st->lastCodedBands+1,IMAX(st->lastCodedBands-1,codedBands)); + else + st->lastCodedBands = codedBands; + + quant_fine_energy(mode, start, end, oldBandE, error, fine_quant, enc, C); + + /* Residual quantisation */ + ALLOC(collapse_masks, C*nbEBands, unsigned char); + quant_all_bands(1, mode, start, end, X, C==2 ? X+N : NULL, collapse_masks, + bandE, pulses, shortBlocks, st->spread_decision, + dual_stereo, st->intensity, tf_res, nbCompressedBytes*(8<<BITRES)-anti_collapse_rsv, + balance, enc, LM, codedBands, &st->rng, st->complexity, st->arch, st->disable_inv); + + if (anti_collapse_rsv > 0) + { + anti_collapse_on = st->consec_transient<2; +#ifdef FUZZING + anti_collapse_on = rand()&0x1; +#endif + ec_enc_bits(enc, anti_collapse_on, 1); + } + quant_energy_finalise(mode, start, end, oldBandE, error, fine_quant, fine_priority, nbCompressedBytes*8-ec_tell(enc), enc, C); + OPUS_CLEAR(energyError, nbEBands*CC); + c=0; + do { + for (i=start;i<end;i++) + { + energyError[i+c*nbEBands] = MAX16(-QCONST16(0.5f, 15), MIN16(QCONST16(0.5f, 15), error[i+c*nbEBands])); + } + } while (++c < C); + + if (silence) + { + for (i=0;i<C*nbEBands;i++) + oldBandE[i] = -QCONST16(28.f,DB_SHIFT); + } + +#ifdef RESYNTH + /* Re-synthesis of the coded audio if required */ + { + celt_sig *out_mem[2]; + + if (anti_collapse_on) + { + anti_collapse(mode, X, collapse_masks, LM, C, N, + start, end, oldBandE, oldLogE, oldLogE2, pulses, st->rng); + } + + c=0; do { + OPUS_MOVE(st->syn_mem[c], st->syn_mem[c]+N, 2*MAX_PERIOD-N+overlap/2); + } while (++c<CC); + + c=0; do { + out_mem[c] = st->syn_mem[c]+2*MAX_PERIOD-N; + } while (++c<CC); + + celt_synthesis(mode, X, out_mem, oldBandE, start, effEnd, + C, CC, isTransient, LM, st->upsample, silence, st->arch); + + c=0; do { + st->prefilter_period=IMAX(st->prefilter_period, COMBFILTER_MINPERIOD); + st->prefilter_period_old=IMAX(st->prefilter_period_old, COMBFILTER_MINPERIOD); + comb_filter(out_mem[c], out_mem[c], st->prefilter_period_old, st->prefilter_period, mode->shortMdctSize, + st->prefilter_gain_old, st->prefilter_gain, st->prefilter_tapset_old, st->prefilter_tapset, + mode->window, overlap); + if (LM!=0) + comb_filter(out_mem[c]+mode->shortMdctSize, out_mem[c]+mode->shortMdctSize, st->prefilter_period, pitch_index, N-mode->shortMdctSize, + st->prefilter_gain, gain1, st->prefilter_tapset, prefilter_tapset, + mode->window, overlap); + } while (++c<CC); + + /* We reuse freq[] as scratch space for the de-emphasis */ + deemphasis(out_mem, (opus_val16*)pcm, N, CC, st->upsample, mode->preemph, st->preemph_memD); + st->prefilter_period_old = st->prefilter_period; + st->prefilter_gain_old = st->prefilter_gain; + st->prefilter_tapset_old = st->prefilter_tapset; + } +#endif + + st->prefilter_period = pitch_index; + st->prefilter_gain = gain1; + st->prefilter_tapset = prefilter_tapset; +#ifdef RESYNTH + if (LM!=0) + { + st->prefilter_period_old = st->prefilter_period; + st->prefilter_gain_old = st->prefilter_gain; + st->prefilter_tapset_old = st->prefilter_tapset; + } +#endif + + if (CC==2&&C==1) { + OPUS_COPY(&oldBandE[nbEBands], oldBandE, nbEBands); + } + + if (!isTransient) + { + OPUS_COPY(oldLogE2, oldLogE, CC*nbEBands); + OPUS_COPY(oldLogE, oldBandE, CC*nbEBands); + } else { + for (i=0;i<CC*nbEBands;i++) + oldLogE[i] = MIN16(oldLogE[i], oldBandE[i]); + } + /* In case start or end were to change */ + c=0; do + { + for (i=0;i<start;i++) + { + oldBandE[c*nbEBands+i]=0; + oldLogE[c*nbEBands+i]=oldLogE2[c*nbEBands+i]=-QCONST16(28.f,DB_SHIFT); + } + for (i=end;i<nbEBands;i++) + { + oldBandE[c*nbEBands+i]=0; + oldLogE[c*nbEBands+i]=oldLogE2[c*nbEBands+i]=-QCONST16(28.f,DB_SHIFT); + } + } while (++c<CC); + + if (isTransient || transient_got_disabled) + st->consec_transient++; + else + st->consec_transient=0; + st->rng = enc->rng; + + /* If there's any room left (can only happen for very high rates), + it's already filled with zeros */ + ec_enc_done(enc); + +#ifdef CUSTOM_MODES + if (st->signalling) + nbCompressedBytes++; +#endif + + RESTORE_STACK; + if (ec_get_error(enc)) + return OPUS_INTERNAL_ERROR; + else + return nbCompressedBytes; +} + + +#ifdef CUSTOM_MODES + +#ifdef FIXED_POINT +int opus_custom_encode(CELTEncoder * OPUS_RESTRICT st, const opus_int16 * pcm, int frame_size, unsigned char *compressed, int nbCompressedBytes) +{ + return celt_encode_with_ec(st, pcm, frame_size, compressed, nbCompressedBytes, NULL); +} + +#ifndef DISABLE_FLOAT_API +int opus_custom_encode_float(CELTEncoder * OPUS_RESTRICT st, const float * pcm, int frame_size, unsigned char *compressed, int nbCompressedBytes) +{ + int j, ret, C, N; + VARDECL(opus_int16, in); + ALLOC_STACK; + + if (pcm==NULL) + return OPUS_BAD_ARG; + + C = st->channels; + N = frame_size; + ALLOC(in, C*N, opus_int16); + + for (j=0;j<C*N;j++) + in[j] = FLOAT2INT16(pcm[j]); + + ret=celt_encode_with_ec(st,in,frame_size,compressed,nbCompressedBytes, NULL); +#ifdef RESYNTH + for (j=0;j<C*N;j++) + ((float*)pcm)[j]=in[j]*(1.f/32768.f); +#endif + RESTORE_STACK; + return ret; +} +#endif /* DISABLE_FLOAT_API */ +#else + +int opus_custom_encode(CELTEncoder * OPUS_RESTRICT st, const opus_int16 * pcm, int frame_size, unsigned char *compressed, int nbCompressedBytes) +{ + int j, ret, C, N; + VARDECL(celt_sig, in); + ALLOC_STACK; + + if (pcm==NULL) + return OPUS_BAD_ARG; + + C=st->channels; + N=frame_size; + ALLOC(in, C*N, celt_sig); + for (j=0;j<C*N;j++) { + in[j] = SCALEOUT(pcm[j]); + } + + ret = celt_encode_with_ec(st,in,frame_size,compressed,nbCompressedBytes, NULL); +#ifdef RESYNTH + for (j=0;j<C*N;j++) + ((opus_int16*)pcm)[j] = FLOAT2INT16(in[j]); +#endif + RESTORE_STACK; + return ret; +} + +int opus_custom_encode_float(CELTEncoder * OPUS_RESTRICT st, const float * pcm, int frame_size, unsigned char *compressed, int nbCompressedBytes) +{ + return celt_encode_with_ec(st, pcm, frame_size, compressed, nbCompressedBytes, NULL); +} + +#endif + +#endif /* CUSTOM_MODES */ + +int opus_custom_encoder_ctl(CELTEncoder * OPUS_RESTRICT st, int request, ...) +{ + va_list ap; + + va_start(ap, request); + switch (request) + { + case OPUS_SET_COMPLEXITY_REQUEST: + { + int value = va_arg(ap, opus_int32); + if (value<0 || value>10) + goto bad_arg; + st->complexity = value; + } + break; + case CELT_SET_START_BAND_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value<0 || value>=st->mode->nbEBands) + goto bad_arg; + st->start = value; + } + break; + case CELT_SET_END_BAND_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value<1 || value>st->mode->nbEBands) + goto bad_arg; + st->end = value; + } + break; + case CELT_SET_PREDICTION_REQUEST: + { + int value = va_arg(ap, opus_int32); + if (value<0 || value>2) + goto bad_arg; + st->disable_pf = value<=1; + st->force_intra = value==0; + } + break; + case OPUS_SET_PACKET_LOSS_PERC_REQUEST: + { + int value = va_arg(ap, opus_int32); + if (value<0 || value>100) + goto bad_arg; + st->loss_rate = value; + } + break; + case OPUS_SET_VBR_CONSTRAINT_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + st->constrained_vbr = value; + } + break; + case OPUS_SET_VBR_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + st->vbr = value; + } + break; + case OPUS_SET_BITRATE_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value<=500 && value!=OPUS_BITRATE_MAX) + goto bad_arg; + value = IMIN(value, 260000*st->channels); + st->bitrate = value; + } + break; + case CELT_SET_CHANNELS_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value<1 || value>2) + goto bad_arg; + st->stream_channels = value; + } + break; + case OPUS_SET_LSB_DEPTH_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value<8 || value>24) + goto bad_arg; + st->lsb_depth=value; + } + break; + case OPUS_GET_LSB_DEPTH_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + *value=st->lsb_depth; + } + break; + case OPUS_SET_PHASE_INVERSION_DISABLED_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if(value<0 || value>1) + { + goto bad_arg; + } + st->disable_inv = value; + } + break; + case OPUS_GET_PHASE_INVERSION_DISABLED_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->disable_inv; + } + break; + case OPUS_RESET_STATE: + { + int i; + opus_val16 *oldBandE, *oldLogE, *oldLogE2; + oldBandE = (opus_val16*)(st->in_mem+st->channels*(st->mode->overlap+COMBFILTER_MAXPERIOD)); + oldLogE = oldBandE + st->channels*st->mode->nbEBands; + oldLogE2 = oldLogE + st->channels*st->mode->nbEBands; + OPUS_CLEAR((char*)&st->ENCODER_RESET_START, + opus_custom_encoder_get_size(st->mode, st->channels)- + ((char*)&st->ENCODER_RESET_START - (char*)st)); + for (i=0;i<st->channels*st->mode->nbEBands;i++) + oldLogE[i]=oldLogE2[i]=-QCONST16(28.f,DB_SHIFT); + st->vbr_offset = 0; + st->delayedIntra = 1; + st->spread_decision = SPREAD_NORMAL; + st->tonal_average = 256; + st->hf_average = 0; + st->tapset_decision = 0; + } + break; +#ifdef CUSTOM_MODES + case CELT_SET_INPUT_CLIPPING_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + st->clip = value; + } + break; +#endif + case CELT_SET_SIGNALLING_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + st->signalling = value; + } + break; + case CELT_SET_ANALYSIS_REQUEST: + { + AnalysisInfo *info = va_arg(ap, AnalysisInfo *); + if (info) + OPUS_COPY(&st->analysis, info, 1); + } + break; + case CELT_SET_SILK_INFO_REQUEST: + { + SILKInfo *info = va_arg(ap, SILKInfo *); + if (info) + OPUS_COPY(&st->silk_info, info, 1); + } + break; + case CELT_GET_MODE_REQUEST: + { + const CELTMode ** value = va_arg(ap, const CELTMode**); + if (value==0) + goto bad_arg; + *value=st->mode; + } + break; + case OPUS_GET_FINAL_RANGE_REQUEST: + { + opus_uint32 * value = va_arg(ap, opus_uint32 *); + if (value==0) + goto bad_arg; + *value=st->rng; + } + break; + case OPUS_SET_LFE_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + st->lfe = value; + } + break; + case OPUS_SET_ENERGY_MASK_REQUEST: + { + opus_val16 *value = va_arg(ap, opus_val16*); + st->energy_mask = value; + } + break; + default: + goto bad_request; + } + va_end(ap); + return OPUS_OK; +bad_arg: + va_end(ap); + return OPUS_BAD_ARG; +bad_request: + va_end(ap); + return OPUS_UNIMPLEMENTED; +} diff --git a/media/libopus/celt/celt_lpc.c b/media/libopus/celt/celt_lpc.c new file mode 100644 index 0000000000..f91721bcab --- /dev/null +++ b/media/libopus/celt/celt_lpc.c @@ -0,0 +1,344 @@ +/* Copyright (c) 2009-2010 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "celt_lpc.h" +#include "stack_alloc.h" +#include "mathops.h" +#include "pitch.h" + +void _celt_lpc( + opus_val16 *_lpc, /* out: [0...p-1] LPC coefficients */ +const opus_val32 *ac, /* in: [0...p] autocorrelation values */ +int p +) +{ + int i, j; + opus_val32 r; + opus_val32 error = ac[0]; +#ifdef FIXED_POINT + opus_val32 lpc[LPC_ORDER]; +#else + float *lpc = _lpc; +#endif + + OPUS_CLEAR(lpc, p); +#ifdef FIXED_POINT + if (ac[0] != 0) +#else + if (ac[0] > 1e-10f) +#endif + { + for (i = 0; i < p; i++) { + /* Sum up this iteration's reflection coefficient */ + opus_val32 rr = 0; + for (j = 0; j < i; j++) + rr += MULT32_32_Q31(lpc[j],ac[i - j]); + rr += SHR32(ac[i + 1],6); + r = -frac_div32(SHL32(rr,6), error); + /* Update LPC coefficients and total error */ + lpc[i] = SHR32(r,6); + for (j = 0; j < (i+1)>>1; j++) + { + opus_val32 tmp1, tmp2; + tmp1 = lpc[j]; + tmp2 = lpc[i-1-j]; + lpc[j] = tmp1 + MULT32_32_Q31(r,tmp2); + lpc[i-1-j] = tmp2 + MULT32_32_Q31(r,tmp1); + } + + error = error - MULT32_32_Q31(MULT32_32_Q31(r,r),error); + /* Bail out once we get 30 dB gain */ +#ifdef FIXED_POINT + if (error<=SHR32(ac[0],10)) + break; +#else + if (error<=.001f*ac[0]) + break; +#endif + } + } +#ifdef FIXED_POINT + { + /* Convert the int32 lpcs to int16 and ensure there are no wrap-arounds. + This reuses the logic in silk_LPC_fit() and silk_bwexpander_32(). Any bug + fixes should also be applied there. */ + int iter, idx = 0; + opus_val32 maxabs, absval, chirp_Q16, chirp_minus_one_Q16; + + for (iter = 0; iter < 10; iter++) { + maxabs = 0; + for (i = 0; i < p; i++) { + absval = ABS32(lpc[i]); + if (absval > maxabs) { + maxabs = absval; + idx = i; + } + } + maxabs = PSHR32(maxabs, 13); /* Q25->Q12 */ + + if (maxabs > 32767) { + maxabs = MIN32(maxabs, 163838); + chirp_Q16 = QCONST32(0.999, 16) - DIV32(SHL32(maxabs - 32767, 14), + SHR32(MULT32_32_32(maxabs, idx + 1), 2)); + chirp_minus_one_Q16 = chirp_Q16 - 65536; + + /* Apply bandwidth expansion. */ + for (i = 0; i < p - 1; i++) { + lpc[i] = MULT32_32_Q16(chirp_Q16, lpc[i]); + chirp_Q16 += PSHR32(MULT32_32_32(chirp_Q16, chirp_minus_one_Q16), 16); + } + lpc[p - 1] = MULT32_32_Q16(chirp_Q16, lpc[p - 1]); + } else { + break; + } + } + + if (iter == 10) { + /* If the coeffs still do not fit into the 16 bit range after 10 iterations, + fall back to the A(z)=1 filter. */ + OPUS_CLEAR(lpc, p); + _lpc[0] = 4096; /* Q12 */ + } else { + for (i = 0; i < p; i++) { + _lpc[i] = EXTRACT16(PSHR32(lpc[i], 13)); /* Q25->Q12 */ + } + } + } +#endif +} + + +void celt_fir_c( + const opus_val16 *x, + const opus_val16 *num, + opus_val16 *y, + int N, + int ord, + int arch) +{ + int i,j; + VARDECL(opus_val16, rnum); + SAVE_STACK; + celt_assert(x != y); + ALLOC(rnum, ord, opus_val16); + for(i=0;i<ord;i++) + rnum[i] = num[ord-i-1]; + for (i=0;i<N-3;i+=4) + { + opus_val32 sum[4]; + sum[0] = SHL32(EXTEND32(x[i ]), SIG_SHIFT); + sum[1] = SHL32(EXTEND32(x[i+1]), SIG_SHIFT); + sum[2] = SHL32(EXTEND32(x[i+2]), SIG_SHIFT); + sum[3] = SHL32(EXTEND32(x[i+3]), SIG_SHIFT); + xcorr_kernel(rnum, x+i-ord, sum, ord, arch); + y[i ] = SROUND16(sum[0], SIG_SHIFT); + y[i+1] = SROUND16(sum[1], SIG_SHIFT); + y[i+2] = SROUND16(sum[2], SIG_SHIFT); + y[i+3] = SROUND16(sum[3], SIG_SHIFT); + } + for (;i<N;i++) + { + opus_val32 sum = SHL32(EXTEND32(x[i]), SIG_SHIFT); + for (j=0;j<ord;j++) + sum = MAC16_16(sum,rnum[j],x[i+j-ord]); + y[i] = SROUND16(sum, SIG_SHIFT); + } + RESTORE_STACK; +} + +void celt_iir(const opus_val32 *_x, + const opus_val16 *den, + opus_val32 *_y, + int N, + int ord, + opus_val16 *mem, + int arch) +{ +#ifdef SMALL_FOOTPRINT + int i,j; + (void)arch; + for (i=0;i<N;i++) + { + opus_val32 sum = _x[i]; + for (j=0;j<ord;j++) + { + sum -= MULT16_16(den[j],mem[j]); + } + for (j=ord-1;j>=1;j--) + { + mem[j]=mem[j-1]; + } + mem[0] = SROUND16(sum, SIG_SHIFT); + _y[i] = sum; + } +#else + int i,j; + VARDECL(opus_val16, rden); + VARDECL(opus_val16, y); + SAVE_STACK; + + celt_assert((ord&3)==0); + ALLOC(rden, ord, opus_val16); + ALLOC(y, N+ord, opus_val16); + for(i=0;i<ord;i++) + rden[i] = den[ord-i-1]; + for(i=0;i<ord;i++) + y[i] = -mem[ord-i-1]; + for(;i<N+ord;i++) + y[i]=0; + for (i=0;i<N-3;i+=4) + { + /* Unroll by 4 as if it were an FIR filter */ + opus_val32 sum[4]; + sum[0]=_x[i]; + sum[1]=_x[i+1]; + sum[2]=_x[i+2]; + sum[3]=_x[i+3]; + xcorr_kernel(rden, y+i, sum, ord, arch); + + /* Patch up the result to compensate for the fact that this is an IIR */ + y[i+ord ] = -SROUND16(sum[0],SIG_SHIFT); + _y[i ] = sum[0]; + sum[1] = MAC16_16(sum[1], y[i+ord ], den[0]); + y[i+ord+1] = -SROUND16(sum[1],SIG_SHIFT); + _y[i+1] = sum[1]; + sum[2] = MAC16_16(sum[2], y[i+ord+1], den[0]); + sum[2] = MAC16_16(sum[2], y[i+ord ], den[1]); + y[i+ord+2] = -SROUND16(sum[2],SIG_SHIFT); + _y[i+2] = sum[2]; + + sum[3] = MAC16_16(sum[3], y[i+ord+2], den[0]); + sum[3] = MAC16_16(sum[3], y[i+ord+1], den[1]); + sum[3] = MAC16_16(sum[3], y[i+ord ], den[2]); + y[i+ord+3] = -SROUND16(sum[3],SIG_SHIFT); + _y[i+3] = sum[3]; + } + for (;i<N;i++) + { + opus_val32 sum = _x[i]; + for (j=0;j<ord;j++) + sum -= MULT16_16(rden[j],y[i+j]); + y[i+ord] = SROUND16(sum,SIG_SHIFT); + _y[i] = sum; + } + for(i=0;i<ord;i++) + mem[i] = _y[N-i-1]; + RESTORE_STACK; +#endif +} + +int _celt_autocorr( + const opus_val16 *x, /* in: [0...n-1] samples x */ + opus_val32 *ac, /* out: [0...lag-1] ac values */ + const opus_val16 *window, + int overlap, + int lag, + int n, + int arch + ) +{ + opus_val32 d; + int i, k; + int fastN=n-lag; + int shift; + const opus_val16 *xptr; + VARDECL(opus_val16, xx); + SAVE_STACK; + ALLOC(xx, n, opus_val16); + celt_assert(n>0); + celt_assert(overlap>=0); + if (overlap == 0) + { + xptr = x; + } else { + for (i=0;i<n;i++) + xx[i] = x[i]; + for (i=0;i<overlap;i++) + { + xx[i] = MULT16_16_Q15(x[i],window[i]); + xx[n-i-1] = MULT16_16_Q15(x[n-i-1],window[i]); + } + xptr = xx; + } + shift=0; +#ifdef FIXED_POINT + { + opus_val32 ac0; + ac0 = 1+(n<<7); + if (n&1) ac0 += SHR32(MULT16_16(xptr[0],xptr[0]),9); + for(i=(n&1);i<n;i+=2) + { + ac0 += SHR32(MULT16_16(xptr[i],xptr[i]),9); + ac0 += SHR32(MULT16_16(xptr[i+1],xptr[i+1]),9); + } + + shift = celt_ilog2(ac0)-30+10; + shift = (shift)/2; + if (shift>0) + { + for(i=0;i<n;i++) + xx[i] = PSHR32(xptr[i], shift); + xptr = xx; + } else + shift = 0; + } +#endif + celt_pitch_xcorr(xptr, xptr, ac, fastN, lag+1, arch); + for (k=0;k<=lag;k++) + { + for (i = k+fastN, d = 0; i < n; i++) + d = MAC16_16(d, xptr[i], xptr[i-k]); + ac[k] += d; + } +#ifdef FIXED_POINT + shift = 2*shift; + if (shift<=0) + ac[0] += SHL32((opus_int32)1, -shift); + if (ac[0] < 268435456) + { + int shift2 = 29 - EC_ILOG(ac[0]); + for (i=0;i<=lag;i++) + ac[i] = SHL32(ac[i], shift2); + shift -= shift2; + } else if (ac[0] >= 536870912) + { + int shift2=1; + if (ac[0] >= 1073741824) + shift2++; + for (i=0;i<=lag;i++) + ac[i] = SHR32(ac[i], shift2); + shift += shift2; + } +#endif + + RESTORE_STACK; + return shift; +} diff --git a/media/libopus/celt/celt_lpc.h b/media/libopus/celt/celt_lpc.h new file mode 100644 index 0000000000..a4c5fd6ea5 --- /dev/null +++ b/media/libopus/celt/celt_lpc.h @@ -0,0 +1,66 @@ +/* Copyright (c) 2009-2010 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef PLC_H +#define PLC_H + +#include "arch.h" +#include "cpu_support.h" + +#if defined(OPUS_X86_MAY_HAVE_SSE4_1) +#include "x86/celt_lpc_sse.h" +#endif + +#define LPC_ORDER 24 + +void _celt_lpc(opus_val16 *_lpc, const opus_val32 *ac, int p); + +void celt_fir_c( + const opus_val16 *x, + const opus_val16 *num, + opus_val16 *y, + int N, + int ord, + int arch); + +#if !defined(OVERRIDE_CELT_FIR) +#define celt_fir(x, num, y, N, ord, arch) \ + (celt_fir_c(x, num, y, N, ord, arch)) +#endif + +void celt_iir(const opus_val32 *x, + const opus_val16 *den, + opus_val32 *y, + int N, + int ord, + opus_val16 *mem, + int arch); + +int _celt_autocorr(const opus_val16 *x, opus_val32 *ac, + const opus_val16 *window, int overlap, int lag, int n, int arch); + +#endif /* PLC_H */ diff --git a/media/libopus/celt/cpu_support.h b/media/libopus/celt/cpu_support.h new file mode 100644 index 0000000000..7b5c56ca90 --- /dev/null +++ b/media/libopus/celt/cpu_support.h @@ -0,0 +1,71 @@ +/* Copyright (c) 2010 Xiph.Org Foundation + * Copyright (c) 2013 Parrot */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef CPU_SUPPORT_H +#define CPU_SUPPORT_H + +#include "opus_types.h" +#include "opus_defines.h" + +#if defined(OPUS_HAVE_RTCD) && \ + (defined(OPUS_ARM_ASM) || defined(OPUS_ARM_MAY_HAVE_NEON_INTR)) +#include "arm/armcpu.h" + +/* We currently support 4 ARM variants: + * arch[0] -> ARMv4 + * arch[1] -> ARMv5E + * arch[2] -> ARMv6 + * arch[3] -> NEON + */ +#define OPUS_ARCHMASK 3 + +#elif defined(OPUS_HAVE_RTCD) && \ + ((defined(OPUS_X86_MAY_HAVE_SSE) && !defined(OPUS_X86_PRESUME_SSE)) || \ + (defined(OPUS_X86_MAY_HAVE_SSE2) && !defined(OPUS_X86_PRESUME_SSE2)) || \ + (defined(OPUS_X86_MAY_HAVE_SSE4_1) && !defined(OPUS_X86_PRESUME_SSE4_1)) || \ + (defined(OPUS_X86_MAY_HAVE_AVX) && !defined(OPUS_X86_PRESUME_AVX))) + +#include "x86/x86cpu.h" +/* We currently support 5 x86 variants: + * arch[0] -> non-sse + * arch[1] -> sse + * arch[2] -> sse2 + * arch[3] -> sse4.1 + * arch[4] -> avx + */ +#define OPUS_ARCHMASK 7 +int opus_select_arch(void); + +#else +#define OPUS_ARCHMASK 0 + +static OPUS_INLINE int opus_select_arch(void) +{ + return 0; +} +#endif +#endif diff --git a/media/libopus/celt/cwrs.c b/media/libopus/celt/cwrs.c new file mode 100644 index 0000000000..a552e4f0fb --- /dev/null +++ b/media/libopus/celt/cwrs.c @@ -0,0 +1,715 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Copyright (c) 2007-2009 Timothy B. Terriberry + Written by Timothy B. Terriberry and Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "os_support.h" +#include "cwrs.h" +#include "mathops.h" +#include "arch.h" + +#ifdef CUSTOM_MODES + +/*Guaranteed to return a conservatively large estimate of the binary logarithm + with frac bits of fractional precision. + Tested for all possible 32-bit inputs with frac=4, where the maximum + overestimation is 0.06254243 bits.*/ +int log2_frac(opus_uint32 val, int frac) +{ + int l; + l=EC_ILOG(val); + if(val&(val-1)){ + /*This is (val>>l-16), but guaranteed to round up, even if adding a bias + before the shift would cause overflow (e.g., for 0xFFFFxxxx). + Doesn't work for val=0, but that case fails the test above.*/ + if(l>16)val=((val-1)>>(l-16))+1; + else val<<=16-l; + l=(l-1)<<frac; + /*Note that we always need one iteration, since the rounding up above means + that we might need to adjust the integer part of the logarithm.*/ + do{ + int b; + b=(int)(val>>16); + l+=b<<frac; + val=(val+b)>>b; + val=(val*val+0x7FFF)>>15; + } + while(frac-->0); + /*If val is not exactly 0x8000, then we have to round up the remainder.*/ + return l+(val>0x8000); + } + /*Exact powers of two require no rounding.*/ + else return (l-1)<<frac; +} +#endif + +/*Although derived separately, the pulse vector coding scheme is equivalent to + a Pyramid Vector Quantizer \cite{Fis86}. + Some additional notes about an early version appear at + https://people.xiph.org/~tterribe/notes/cwrs.html, but the codebook ordering + and the definitions of some terms have evolved since that was written. + + The conversion from a pulse vector to an integer index (encoding) and back + (decoding) is governed by two related functions, V(N,K) and U(N,K). + + V(N,K) = the number of combinations, with replacement, of N items, taken K + at a time, when a sign bit is added to each item taken at least once (i.e., + the number of N-dimensional unit pulse vectors with K pulses). + One way to compute this is via + V(N,K) = K>0 ? sum(k=1...K,2**k*choose(N,k)*choose(K-1,k-1)) : 1, + where choose() is the binomial function. + A table of values for N<10 and K<10 looks like: + V[10][10] = { + {1, 0, 0, 0, 0, 0, 0, 0, 0, 0}, + {1, 2, 2, 2, 2, 2, 2, 2, 2, 2}, + {1, 4, 8, 12, 16, 20, 24, 28, 32, 36}, + {1, 6, 18, 38, 66, 102, 146, 198, 258, 326}, + {1, 8, 32, 88, 192, 360, 608, 952, 1408, 1992}, + {1, 10, 50, 170, 450, 1002, 1970, 3530, 5890, 9290}, + {1, 12, 72, 292, 912, 2364, 5336, 10836, 20256, 35436}, + {1, 14, 98, 462, 1666, 4942, 12642, 28814, 59906, 115598}, + {1, 16, 128, 688, 2816, 9424, 27008, 68464, 157184, 332688}, + {1, 18, 162, 978, 4482, 16722, 53154, 148626, 374274, 864146} + }; + + U(N,K) = the number of such combinations wherein N-1 objects are taken at + most K-1 at a time. + This is given by + U(N,K) = sum(k=0...K-1,V(N-1,k)) + = K>0 ? (V(N-1,K-1) + V(N,K-1))/2 : 0. + The latter expression also makes clear that U(N,K) is half the number of such + combinations wherein the first object is taken at least once. + Although it may not be clear from either of these definitions, U(N,K) is the + natural function to work with when enumerating the pulse vector codebooks, + not V(N,K). + U(N,K) is not well-defined for N=0, but with the extension + U(0,K) = K>0 ? 0 : 1, + the function becomes symmetric: U(N,K) = U(K,N), with a similar table: + U[10][10] = { + {1, 0, 0, 0, 0, 0, 0, 0, 0, 0}, + {0, 1, 1, 1, 1, 1, 1, 1, 1, 1}, + {0, 1, 3, 5, 7, 9, 11, 13, 15, 17}, + {0, 1, 5, 13, 25, 41, 61, 85, 113, 145}, + {0, 1, 7, 25, 63, 129, 231, 377, 575, 833}, + {0, 1, 9, 41, 129, 321, 681, 1289, 2241, 3649}, + {0, 1, 11, 61, 231, 681, 1683, 3653, 7183, 13073}, + {0, 1, 13, 85, 377, 1289, 3653, 8989, 19825, 40081}, + {0, 1, 15, 113, 575, 2241, 7183, 19825, 48639, 108545}, + {0, 1, 17, 145, 833, 3649, 13073, 40081, 108545, 265729} + }; + + With this extension, V(N,K) may be written in terms of U(N,K): + V(N,K) = U(N,K) + U(N,K+1) + for all N>=0, K>=0. + Thus U(N,K+1) represents the number of combinations where the first element + is positive or zero, and U(N,K) represents the number of combinations where + it is negative. + With a large enough table of U(N,K) values, we could write O(N) encoding + and O(min(N*log(K),N+K)) decoding routines, but such a table would be + prohibitively large for small embedded devices (K may be as large as 32767 + for small N, and N may be as large as 200). + + Both functions obey the same recurrence relation: + V(N,K) = V(N-1,K) + V(N,K-1) + V(N-1,K-1), + U(N,K) = U(N-1,K) + U(N,K-1) + U(N-1,K-1), + for all N>0, K>0, with different initial conditions at N=0 or K=0. + This allows us to construct a row of one of the tables above given the + previous row or the next row. + Thus we can derive O(NK) encoding and decoding routines with O(K) memory + using only addition and subtraction. + + When encoding, we build up from the U(2,K) row and work our way forwards. + When decoding, we need to start at the U(N,K) row and work our way backwards, + which requires a means of computing U(N,K). + U(N,K) may be computed from two previous values with the same N: + U(N,K) = ((2*N-1)*U(N,K-1) - U(N,K-2))/(K-1) + U(N,K-2) + for all N>1, and since U(N,K) is symmetric, a similar relation holds for two + previous values with the same K: + U(N,K>1) = ((2*K-1)*U(N-1,K) - U(N-2,K))/(N-1) + U(N-2,K) + for all K>1. + This allows us to construct an arbitrary row of the U(N,K) table by starting + with the first two values, which are constants. + This saves roughly 2/3 the work in our O(NK) decoding routine, but costs O(K) + multiplications. + Similar relations can be derived for V(N,K), but are not used here. + + For N>0 and K>0, U(N,K) and V(N,K) take on the form of an (N-1)-degree + polynomial for fixed N. + The first few are + U(1,K) = 1, + U(2,K) = 2*K-1, + U(3,K) = (2*K-2)*K+1, + U(4,K) = (((4*K-6)*K+8)*K-3)/3, + U(5,K) = ((((2*K-4)*K+10)*K-8)*K+3)/3, + and + V(1,K) = 2, + V(2,K) = 4*K, + V(3,K) = 4*K*K+2, + V(4,K) = 8*(K*K+2)*K/3, + V(5,K) = ((4*K*K+20)*K*K+6)/3, + for all K>0. + This allows us to derive O(N) encoding and O(N*log(K)) decoding routines for + small N (and indeed decoding is also O(N) for N<3). + + @ARTICLE{Fis86, + author="Thomas R. Fischer", + title="A Pyramid Vector Quantizer", + journal="IEEE Transactions on Information Theory", + volume="IT-32", + number=4, + pages="568--583", + month=Jul, + year=1986 + }*/ + +#if !defined(SMALL_FOOTPRINT) + +/*U(N,K) = U(K,N) := N>0?K>0?U(N-1,K)+U(N,K-1)+U(N-1,K-1):0:K>0?1:0*/ +# define CELT_PVQ_U(_n,_k) (CELT_PVQ_U_ROW[IMIN(_n,_k)][IMAX(_n,_k)]) +/*V(N,K) := U(N,K)+U(N,K+1) = the number of PVQ codewords for a band of size N + with K pulses allocated to it.*/ +# define CELT_PVQ_V(_n,_k) (CELT_PVQ_U(_n,_k)+CELT_PVQ_U(_n,(_k)+1)) + +/*For each V(N,K) supported, we will access element U(min(N,K+1),max(N,K+1)). + Thus, the number of entries in row I is the larger of the maximum number of + pulses we will ever allocate for a given N=I (K=128, or however many fit in + 32 bits, whichever is smaller), plus one, and the maximum N for which + K=I-1 pulses fit in 32 bits. + The largest band size in an Opus Custom mode is 208. + Otherwise, we can limit things to the set of N which can be achieved by + splitting a band from a standard Opus mode: 176, 144, 96, 88, 72, 64, 48, + 44, 36, 32, 24, 22, 18, 16, 8, 4, 2).*/ +#if defined(CUSTOM_MODES) +static const opus_uint32 CELT_PVQ_U_DATA[1488]={ +#else +static const opus_uint32 CELT_PVQ_U_DATA[1272]={ +#endif + /*N=0, K=0...176:*/ + 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +#if defined(CUSTOM_MODES) + /*...208:*/ + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, +#endif + /*N=1, K=1...176:*/ + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, +#if defined(CUSTOM_MODES) + /*...208:*/ + 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, + 1, 1, 1, 1, 1, 1, +#endif + /*N=2, K=2...176:*/ + 3, 5, 7, 9, 11, 13, 15, 17, 19, 21, 23, 25, 27, 29, 31, 33, 35, 37, 39, 41, + 43, 45, 47, 49, 51, 53, 55, 57, 59, 61, 63, 65, 67, 69, 71, 73, 75, 77, 79, + 81, 83, 85, 87, 89, 91, 93, 95, 97, 99, 101, 103, 105, 107, 109, 111, 113, + 115, 117, 119, 121, 123, 125, 127, 129, 131, 133, 135, 137, 139, 141, 143, + 145, 147, 149, 151, 153, 155, 157, 159, 161, 163, 165, 167, 169, 171, 173, + 175, 177, 179, 181, 183, 185, 187, 189, 191, 193, 195, 197, 199, 201, 203, + 205, 207, 209, 211, 213, 215, 217, 219, 221, 223, 225, 227, 229, 231, 233, + 235, 237, 239, 241, 243, 245, 247, 249, 251, 253, 255, 257, 259, 261, 263, + 265, 267, 269, 271, 273, 275, 277, 279, 281, 283, 285, 287, 289, 291, 293, + 295, 297, 299, 301, 303, 305, 307, 309, 311, 313, 315, 317, 319, 321, 323, + 325, 327, 329, 331, 333, 335, 337, 339, 341, 343, 345, 347, 349, 351, +#if defined(CUSTOM_MODES) + /*...208:*/ + 353, 355, 357, 359, 361, 363, 365, 367, 369, 371, 373, 375, 377, 379, 381, + 383, 385, 387, 389, 391, 393, 395, 397, 399, 401, 403, 405, 407, 409, 411, + 413, 415, +#endif + /*N=3, K=3...176:*/ + 13, 25, 41, 61, 85, 113, 145, 181, 221, 265, 313, 365, 421, 481, 545, 613, + 685, 761, 841, 925, 1013, 1105, 1201, 1301, 1405, 1513, 1625, 1741, 1861, + 1985, 2113, 2245, 2381, 2521, 2665, 2813, 2965, 3121, 3281, 3445, 3613, 3785, + 3961, 4141, 4325, 4513, 4705, 4901, 5101, 5305, 5513, 5725, 5941, 6161, 6385, + 6613, 6845, 7081, 7321, 7565, 7813, 8065, 8321, 8581, 8845, 9113, 9385, 9661, + 9941, 10225, 10513, 10805, 11101, 11401, 11705, 12013, 12325, 12641, 12961, + 13285, 13613, 13945, 14281, 14621, 14965, 15313, 15665, 16021, 16381, 16745, + 17113, 17485, 17861, 18241, 18625, 19013, 19405, 19801, 20201, 20605, 21013, + 21425, 21841, 22261, 22685, 23113, 23545, 23981, 24421, 24865, 25313, 25765, + 26221, 26681, 27145, 27613, 28085, 28561, 29041, 29525, 30013, 30505, 31001, + 31501, 32005, 32513, 33025, 33541, 34061, 34585, 35113, 35645, 36181, 36721, + 37265, 37813, 38365, 38921, 39481, 40045, 40613, 41185, 41761, 42341, 42925, + 43513, 44105, 44701, 45301, 45905, 46513, 47125, 47741, 48361, 48985, 49613, + 50245, 50881, 51521, 52165, 52813, 53465, 54121, 54781, 55445, 56113, 56785, + 57461, 58141, 58825, 59513, 60205, 60901, 61601, +#if defined(CUSTOM_MODES) + /*...208:*/ + 62305, 63013, 63725, 64441, 65161, 65885, 66613, 67345, 68081, 68821, 69565, + 70313, 71065, 71821, 72581, 73345, 74113, 74885, 75661, 76441, 77225, 78013, + 78805, 79601, 80401, 81205, 82013, 82825, 83641, 84461, 85285, 86113, +#endif + /*N=4, K=4...176:*/ + 63, 129, 231, 377, 575, 833, 1159, 1561, 2047, 2625, 3303, 4089, 4991, 6017, + 7175, 8473, 9919, 11521, 13287, 15225, 17343, 19649, 22151, 24857, 27775, + 30913, 34279, 37881, 41727, 45825, 50183, 54809, 59711, 64897, 70375, 76153, + 82239, 88641, 95367, 102425, 109823, 117569, 125671, 134137, 142975, 152193, + 161799, 171801, 182207, 193025, 204263, 215929, 228031, 240577, 253575, + 267033, 280959, 295361, 310247, 325625, 341503, 357889, 374791, 392217, + 410175, 428673, 447719, 467321, 487487, 508225, 529543, 551449, 573951, + 597057, 620775, 645113, 670079, 695681, 721927, 748825, 776383, 804609, + 833511, 863097, 893375, 924353, 956039, 988441, 1021567, 1055425, 1090023, + 1125369, 1161471, 1198337, 1235975, 1274393, 1313599, 1353601, 1394407, + 1436025, 1478463, 1521729, 1565831, 1610777, 1656575, 1703233, 1750759, + 1799161, 1848447, 1898625, 1949703, 2001689, 2054591, 2108417, 2163175, + 2218873, 2275519, 2333121, 2391687, 2451225, 2511743, 2573249, 2635751, + 2699257, 2763775, 2829313, 2895879, 2963481, 3032127, 3101825, 3172583, + 3244409, 3317311, 3391297, 3466375, 3542553, 3619839, 3698241, 3777767, + 3858425, 3940223, 4023169, 4107271, 4192537, 4278975, 4366593, 4455399, + 4545401, 4636607, 4729025, 4822663, 4917529, 5013631, 5110977, 5209575, + 5309433, 5410559, 5512961, 5616647, 5721625, 5827903, 5935489, 6044391, + 6154617, 6266175, 6379073, 6493319, 6608921, 6725887, 6844225, 6963943, + 7085049, 7207551, +#if defined(CUSTOM_MODES) + /*...208:*/ + 7331457, 7456775, 7583513, 7711679, 7841281, 7972327, 8104825, 8238783, + 8374209, 8511111, 8649497, 8789375, 8930753, 9073639, 9218041, 9363967, + 9511425, 9660423, 9810969, 9963071, 10116737, 10271975, 10428793, 10587199, + 10747201, 10908807, 11072025, 11236863, 11403329, 11571431, 11741177, + 11912575, +#endif + /*N=5, K=5...176:*/ + 321, 681, 1289, 2241, 3649, 5641, 8361, 11969, 16641, 22569, 29961, 39041, + 50049, 63241, 78889, 97281, 118721, 143529, 172041, 204609, 241601, 283401, + 330409, 383041, 441729, 506921, 579081, 658689, 746241, 842249, 947241, + 1061761, 1186369, 1321641, 1468169, 1626561, 1797441, 1981449, 2179241, + 2391489, 2618881, 2862121, 3121929, 3399041, 3694209, 4008201, 4341801, + 4695809, 5071041, 5468329, 5888521, 6332481, 6801089, 7295241, 7815849, + 8363841, 8940161, 9545769, 10181641, 10848769, 11548161, 12280841, 13047849, + 13850241, 14689089, 15565481, 16480521, 17435329, 18431041, 19468809, + 20549801, 21675201, 22846209, 24064041, 25329929, 26645121, 28010881, + 29428489, 30899241, 32424449, 34005441, 35643561, 37340169, 39096641, + 40914369, 42794761, 44739241, 46749249, 48826241, 50971689, 53187081, + 55473921, 57833729, 60268041, 62778409, 65366401, 68033601, 70781609, + 73612041, 76526529, 79526721, 82614281, 85790889, 89058241, 92418049, + 95872041, 99421961, 103069569, 106816641, 110664969, 114616361, 118672641, + 122835649, 127107241, 131489289, 135983681, 140592321, 145317129, 150160041, + 155123009, 160208001, 165417001, 170752009, 176215041, 181808129, 187533321, + 193392681, 199388289, 205522241, 211796649, 218213641, 224775361, 231483969, + 238341641, 245350569, 252512961, 259831041, 267307049, 274943241, 282741889, + 290705281, 298835721, 307135529, 315607041, 324252609, 333074601, 342075401, + 351257409, 360623041, 370174729, 379914921, 389846081, 399970689, 410291241, + 420810249, 431530241, 442453761, 453583369, 464921641, 476471169, 488234561, + 500214441, 512413449, 524834241, 537479489, 550351881, 563454121, 576788929, + 590359041, 604167209, 618216201, 632508801, +#if defined(CUSTOM_MODES) + /*...208:*/ + 647047809, 661836041, 676876329, 692171521, 707724481, 723538089, 739615241, + 755958849, 772571841, 789457161, 806617769, 824056641, 841776769, 859781161, + 878072841, 896654849, 915530241, 934702089, 954173481, 973947521, 994027329, + 1014416041, 1035116809, 1056132801, 1077467201, 1099123209, 1121104041, + 1143412929, 1166053121, 1189027881, 1212340489, 1235994241, +#endif + /*N=6, K=6...96:*/ + 1683, 3653, 7183, 13073, 22363, 36365, 56695, 85305, 124515, 177045, 246047, + 335137, 448427, 590557, 766727, 982729, 1244979, 1560549, 1937199, 2383409, + 2908411, 3522221, 4235671, 5060441, 6009091, 7095093, 8332863, 9737793, + 11326283, 13115773, 15124775, 17372905, 19880915, 22670725, 25765455, + 29189457, 32968347, 37129037, 41699767, 46710137, 52191139, 58175189, + 64696159, 71789409, 79491819, 87841821, 96879431, 106646281, 117185651, + 128542501, 140763503, 153897073, 167993403, 183104493, 199284183, 216588185, + 235074115, 254801525, 275831935, 298228865, 322057867, 347386557, 374284647, + 402823977, 433078547, 465124549, 499040399, 534906769, 572806619, 612825229, + 655050231, 699571641, 746481891, 795875861, 847850911, 902506913, 959946283, + 1020274013, 1083597703, 1150027593, 1219676595, 1292660325, 1369097135, + 1449108145, 1532817275, 1620351277, 1711839767, 1807415257, 1907213187, + 2011371957, 2120032959, +#if defined(CUSTOM_MODES) + /*...109:*/ + 2233340609U, 2351442379U, 2474488829U, 2602633639U, 2736033641U, 2874848851U, + 3019242501U, 3169381071U, 3325434321U, 3487575323U, 3655980493U, 3830829623U, + 4012305913U, +#endif + /*N=7, K=7...54*/ + 8989, 19825, 40081, 75517, 134245, 227305, 369305, 579125, 880685, 1303777, + 1884961, 2668525, 3707509, 5064793, 6814249, 9041957, 11847485, 15345233, + 19665841, 24957661, 31388293, 39146185, 48442297, 59511829, 72616013, + 88043969, 106114625, 127178701, 151620757, 179861305, 212358985, 249612805, + 292164445, 340600625, 395555537, 457713341, 527810725, 606639529, 695049433, + 793950709, 904317037, 1027188385, 1163673953, 1314955181, 1482288821, + 1667010073, 1870535785, 2094367717, +#if defined(CUSTOM_MODES) + /*...60:*/ + 2340095869U, 2609401873U, 2904062449U, 3225952925U, 3577050821U, 3959439497U, +#endif + /*N=8, K=8...37*/ + 48639, 108545, 224143, 433905, 795455, 1392065, 2340495, 3800305, 5984767, + 9173505, 13726991, 20103025, 28875327, 40754369, 56610575, 77500017, + 104692735, 139703809, 184327311, 240673265, 311207743, 398796225, 506750351, + 638878193, 799538175, 993696769, 1226990095, 1505789553, 1837271615, + 2229491905U, +#if defined(CUSTOM_MODES) + /*...40:*/ + 2691463695U, 3233240945U, 3866006015U, +#endif + /*N=9, K=9...28:*/ + 265729, 598417, 1256465, 2485825, 4673345, 8405905, 14546705, 24331777, + 39490049, 62390545, 96220561, 145198913, 214828609, 312193553, 446304145, + 628496897, 872893441, 1196924561, 1621925137, 2173806145U, +#if defined(CUSTOM_MODES) + /*...29:*/ + 2883810113U, +#endif + /*N=10, K=10...24:*/ + 1462563, 3317445, 7059735, 14218905, 27298155, 50250765, 89129247, 152951073, + 254831667, 413442773, 654862247, 1014889769, 1541911931, 2300409629U, + 3375210671U, + /*N=11, K=11...19:*/ + 8097453, 18474633, 39753273, 81270333, 158819253, 298199265, 540279585, + 948062325, 1616336765, +#if defined(CUSTOM_MODES) + /*...20:*/ + 2684641785U, +#endif + /*N=12, K=12...18:*/ + 45046719, 103274625, 224298231, 464387817, 921406335, 1759885185, + 3248227095U, + /*N=13, K=13...16:*/ + 251595969, 579168825, 1267854873, 2653649025U, + /*N=14, K=14:*/ + 1409933619 +}; + +#if defined(CUSTOM_MODES) +static const opus_uint32 *const CELT_PVQ_U_ROW[15]={ + CELT_PVQ_U_DATA+ 0,CELT_PVQ_U_DATA+ 208,CELT_PVQ_U_DATA+ 415, + CELT_PVQ_U_DATA+ 621,CELT_PVQ_U_DATA+ 826,CELT_PVQ_U_DATA+1030, + CELT_PVQ_U_DATA+1233,CELT_PVQ_U_DATA+1336,CELT_PVQ_U_DATA+1389, + CELT_PVQ_U_DATA+1421,CELT_PVQ_U_DATA+1441,CELT_PVQ_U_DATA+1455, + CELT_PVQ_U_DATA+1464,CELT_PVQ_U_DATA+1470,CELT_PVQ_U_DATA+1473 +}; +#else +static const opus_uint32 *const CELT_PVQ_U_ROW[15]={ + CELT_PVQ_U_DATA+ 0,CELT_PVQ_U_DATA+ 176,CELT_PVQ_U_DATA+ 351, + CELT_PVQ_U_DATA+ 525,CELT_PVQ_U_DATA+ 698,CELT_PVQ_U_DATA+ 870, + CELT_PVQ_U_DATA+1041,CELT_PVQ_U_DATA+1131,CELT_PVQ_U_DATA+1178, + CELT_PVQ_U_DATA+1207,CELT_PVQ_U_DATA+1226,CELT_PVQ_U_DATA+1240, + CELT_PVQ_U_DATA+1248,CELT_PVQ_U_DATA+1254,CELT_PVQ_U_DATA+1257 +}; +#endif + +#if defined(CUSTOM_MODES) +void get_required_bits(opus_int16 *_bits,int _n,int _maxk,int _frac){ + int k; + /*_maxk==0 => there's nothing to do.*/ + celt_assert(_maxk>0); + _bits[0]=0; + for(k=1;k<=_maxk;k++)_bits[k]=log2_frac(CELT_PVQ_V(_n,k),_frac); +} +#endif + +static opus_uint32 icwrs(int _n,const int *_y){ + opus_uint32 i; + int j; + int k; + celt_assert(_n>=2); + j=_n-1; + i=_y[j]<0; + k=abs(_y[j]); + do{ + j--; + i+=CELT_PVQ_U(_n-j,k); + k+=abs(_y[j]); + if(_y[j]<0)i+=CELT_PVQ_U(_n-j,k+1); + } + while(j>0); + return i; +} + +void encode_pulses(const int *_y,int _n,int _k,ec_enc *_enc){ + celt_assert(_k>0); + ec_enc_uint(_enc,icwrs(_n,_y),CELT_PVQ_V(_n,_k)); +} + +static opus_val32 cwrsi(int _n,int _k,opus_uint32 _i,int *_y){ + opus_uint32 p; + int s; + int k0; + opus_int16 val; + opus_val32 yy=0; + celt_assert(_k>0); + celt_assert(_n>1); + while(_n>2){ + opus_uint32 q; + /*Lots of pulses case:*/ + if(_k>=_n){ + const opus_uint32 *row; + row=CELT_PVQ_U_ROW[_n]; + /*Are the pulses in this dimension negative?*/ + p=row[_k+1]; + s=-(_i>=p); + _i-=p&s; + /*Count how many pulses were placed in this dimension.*/ + k0=_k; + q=row[_n]; + if(q>_i){ + celt_sig_assert(p>q); + _k=_n; + do p=CELT_PVQ_U_ROW[--_k][_n]; + while(p>_i); + } + else for(p=row[_k];p>_i;p=row[_k])_k--; + _i-=p; + val=(k0-_k+s)^s; + *_y++=val; + yy=MAC16_16(yy,val,val); + } + /*Lots of dimensions case:*/ + else{ + /*Are there any pulses in this dimension at all?*/ + p=CELT_PVQ_U_ROW[_k][_n]; + q=CELT_PVQ_U_ROW[_k+1][_n]; + if(p<=_i&&_i<q){ + _i-=p; + *_y++=0; + } + else{ + /*Are the pulses in this dimension negative?*/ + s=-(_i>=q); + _i-=q&s; + /*Count how many pulses were placed in this dimension.*/ + k0=_k; + do p=CELT_PVQ_U_ROW[--_k][_n]; + while(p>_i); + _i-=p; + val=(k0-_k+s)^s; + *_y++=val; + yy=MAC16_16(yy,val,val); + } + } + _n--; + } + /*_n==2*/ + p=2*_k+1; + s=-(_i>=p); + _i-=p&s; + k0=_k; + _k=(_i+1)>>1; + if(_k)_i-=2*_k-1; + val=(k0-_k+s)^s; + *_y++=val; + yy=MAC16_16(yy,val,val); + /*_n==1*/ + s=-(int)_i; + val=(_k+s)^s; + *_y=val; + yy=MAC16_16(yy,val,val); + return yy; +} + +opus_val32 decode_pulses(int *_y,int _n,int _k,ec_dec *_dec){ + return cwrsi(_n,_k,ec_dec_uint(_dec,CELT_PVQ_V(_n,_k)),_y); +} + +#else /* SMALL_FOOTPRINT */ + +/*Computes the next row/column of any recurrence that obeys the relation + u[i][j]=u[i-1][j]+u[i][j-1]+u[i-1][j-1]. + _ui0 is the base case for the new row/column.*/ +static OPUS_INLINE void unext(opus_uint32 *_ui,unsigned _len,opus_uint32 _ui0){ + opus_uint32 ui1; + unsigned j; + /*This do-while will overrun the array if we don't have storage for at least + 2 values.*/ + j=1; do { + ui1=UADD32(UADD32(_ui[j],_ui[j-1]),_ui0); + _ui[j-1]=_ui0; + _ui0=ui1; + } while (++j<_len); + _ui[j-1]=_ui0; +} + +/*Computes the previous row/column of any recurrence that obeys the relation + u[i-1][j]=u[i][j]-u[i][j-1]-u[i-1][j-1]. + _ui0 is the base case for the new row/column.*/ +static OPUS_INLINE void uprev(opus_uint32 *_ui,unsigned _n,opus_uint32 _ui0){ + opus_uint32 ui1; + unsigned j; + /*This do-while will overrun the array if we don't have storage for at least + 2 values.*/ + j=1; do { + ui1=USUB32(USUB32(_ui[j],_ui[j-1]),_ui0); + _ui[j-1]=_ui0; + _ui0=ui1; + } while (++j<_n); + _ui[j-1]=_ui0; +} + +/*Compute V(_n,_k), as well as U(_n,0..._k+1). + _u: On exit, _u[i] contains U(_n,i) for i in [0..._k+1].*/ +static opus_uint32 ncwrs_urow(unsigned _n,unsigned _k,opus_uint32 *_u){ + opus_uint32 um2; + unsigned len; + unsigned k; + len=_k+2; + /*We require storage at least 3 values (e.g., _k>0).*/ + celt_assert(len>=3); + _u[0]=0; + _u[1]=um2=1; + /*If _n==0, _u[0] should be 1 and the rest should be 0.*/ + /*If _n==1, _u[i] should be 1 for i>1.*/ + celt_assert(_n>=2); + /*If _k==0, the following do-while loop will overflow the buffer.*/ + celt_assert(_k>0); + k=2; + do _u[k]=(k<<1)-1; + while(++k<len); + for(k=2;k<_n;k++)unext(_u+1,_k+1,1); + return _u[_k]+_u[_k+1]; +} + +/*Returns the _i'th combination of _k elements chosen from a set of size _n + with associated sign bits. + _y: Returns the vector of pulses. + _u: Must contain entries [0..._k+1] of row _n of U() on input. + Its contents will be destructively modified.*/ +static opus_val32 cwrsi(int _n,int _k,opus_uint32 _i,int *_y,opus_uint32 *_u){ + int j; + opus_int16 val; + opus_val32 yy=0; + celt_assert(_n>0); + j=0; + do{ + opus_uint32 p; + int s; + int yj; + p=_u[_k+1]; + s=-(_i>=p); + _i-=p&s; + yj=_k; + p=_u[_k]; + while(p>_i)p=_u[--_k]; + _i-=p; + yj-=_k; + val=(yj+s)^s; + _y[j]=val; + yy=MAC16_16(yy,val,val); + uprev(_u,_k+2,0); + } + while(++j<_n); + return yy; +} + +/*Returns the index of the given combination of K elements chosen from a set + of size 1 with associated sign bits. + _y: The vector of pulses, whose sum of absolute values is K. + _k: Returns K.*/ +static OPUS_INLINE opus_uint32 icwrs1(const int *_y,int *_k){ + *_k=abs(_y[0]); + return _y[0]<0; +} + +/*Returns the index of the given combination of K elements chosen from a set + of size _n with associated sign bits. + _y: The vector of pulses, whose sum of absolute values must be _k. + _nc: Returns V(_n,_k).*/ +static OPUS_INLINE opus_uint32 icwrs(int _n,int _k,opus_uint32 *_nc,const int *_y, + opus_uint32 *_u){ + opus_uint32 i; + int j; + int k; + /*We can't unroll the first two iterations of the loop unless _n>=2.*/ + celt_assert(_n>=2); + _u[0]=0; + for(k=1;k<=_k+1;k++)_u[k]=(k<<1)-1; + i=icwrs1(_y+_n-1,&k); + j=_n-2; + i+=_u[k]; + k+=abs(_y[j]); + if(_y[j]<0)i+=_u[k+1]; + while(j-->0){ + unext(_u,_k+2,0); + i+=_u[k]; + k+=abs(_y[j]); + if(_y[j]<0)i+=_u[k+1]; + } + *_nc=_u[k]+_u[k+1]; + return i; +} + +#ifdef CUSTOM_MODES +void get_required_bits(opus_int16 *_bits,int _n,int _maxk,int _frac){ + int k; + /*_maxk==0 => there's nothing to do.*/ + celt_assert(_maxk>0); + _bits[0]=0; + if (_n==1) + { + for (k=1;k<=_maxk;k++) + _bits[k] = 1<<_frac; + } + else { + VARDECL(opus_uint32,u); + SAVE_STACK; + ALLOC(u,_maxk+2U,opus_uint32); + ncwrs_urow(_n,_maxk,u); + for(k=1;k<=_maxk;k++) + _bits[k]=log2_frac(u[k]+u[k+1],_frac); + RESTORE_STACK; + } +} +#endif /* CUSTOM_MODES */ + +void encode_pulses(const int *_y,int _n,int _k,ec_enc *_enc){ + opus_uint32 i; + VARDECL(opus_uint32,u); + opus_uint32 nc; + SAVE_STACK; + celt_assert(_k>0); + ALLOC(u,_k+2U,opus_uint32); + i=icwrs(_n,_k,&nc,_y,u); + ec_enc_uint(_enc,i,nc); + RESTORE_STACK; +} + +opus_val32 decode_pulses(int *_y,int _n,int _k,ec_dec *_dec){ + VARDECL(opus_uint32,u); + int ret; + SAVE_STACK; + celt_assert(_k>0); + ALLOC(u,_k+2U,opus_uint32); + ret = cwrsi(_n,_k,ec_dec_uint(_dec,ncwrs_urow(_n,_k,u)),_y,u); + RESTORE_STACK; + return ret; +} + +#endif /* SMALL_FOOTPRINT */ diff --git a/media/libopus/celt/cwrs.h b/media/libopus/celt/cwrs.h new file mode 100644 index 0000000000..7cd4717459 --- /dev/null +++ b/media/libopus/celt/cwrs.h @@ -0,0 +1,48 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Copyright (c) 2007-2009 Timothy B. Terriberry + Written by Timothy B. Terriberry and Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef CWRS_H +#define CWRS_H + +#include "arch.h" +#include "stack_alloc.h" +#include "entenc.h" +#include "entdec.h" + +#ifdef CUSTOM_MODES +int log2_frac(opus_uint32 val, int frac); +#endif + +void get_required_bits(opus_int16 *bits, int N, int K, int frac); + +void encode_pulses(const int *_y, int N, int K, ec_enc *enc); + +opus_val32 decode_pulses(int *_y, int N, int K, ec_dec *dec); + +#endif /* CWRS_H */ diff --git a/media/libopus/celt/ecintrin.h b/media/libopus/celt/ecintrin.h new file mode 100644 index 0000000000..66a4c36ea4 --- /dev/null +++ b/media/libopus/celt/ecintrin.h @@ -0,0 +1,91 @@ +/* Copyright (c) 2003-2008 Timothy B. Terriberry + Copyright (c) 2008 Xiph.Org Foundation */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +/*Some common macros for potential platform-specific optimization.*/ +#include "opus_types.h" +#include <math.h> +#include <limits.h> +#include "arch.h" +#if !defined(_ecintrin_H) +# define _ecintrin_H (1) + +/*Some specific platforms may have optimized intrinsic or OPUS_INLINE assembly + versions of these functions which can substantially improve performance. + We define macros for them to allow easy incorporation of these non-ANSI + features.*/ + +/*Modern gcc (4.x) can compile the naive versions of min and max with cmov if + given an appropriate architecture, but the branchless bit-twiddling versions + are just as fast, and do not require any special target architecture. + Earlier gcc versions (3.x) compiled both code to the same assembly + instructions, because of the way they represented ((_b)>(_a)) internally.*/ +# define EC_MINI(_a,_b) ((_a)+(((_b)-(_a))&-((_b)<(_a)))) + +/*Count leading zeros. + This macro should only be used for implementing ec_ilog(), if it is defined. + All other code should use EC_ILOG() instead.*/ +#if defined(_MSC_VER) && (_MSC_VER >= 1400) +#if defined(_MSC_VER) && (_MSC_VER >= 1910) +# include <intrin0.h> /* Improve compiler throughput. */ +#else +# include <intrin.h> +#endif +/*In _DEBUG mode this is not an intrinsic by default.*/ +# pragma intrinsic(_BitScanReverse) + +static __inline int ec_bsr(unsigned long _x){ + unsigned long ret; + _BitScanReverse(&ret,_x); + return (int)ret; +} +# define EC_CLZ0 (1) +# define EC_CLZ(_x) (-ec_bsr(_x)) +#elif defined(ENABLE_TI_DSPLIB) +# include "dsplib.h" +# define EC_CLZ0 (31) +# define EC_CLZ(_x) (_lnorm(_x)) +#elif __GNUC_PREREQ(3,4) +# if INT_MAX>=2147483647 +# define EC_CLZ0 ((int)sizeof(unsigned)*CHAR_BIT) +# define EC_CLZ(_x) (__builtin_clz(_x)) +# elif LONG_MAX>=2147483647L +# define EC_CLZ0 ((int)sizeof(unsigned long)*CHAR_BIT) +# define EC_CLZ(_x) (__builtin_clzl(_x)) +# endif +#endif + +#if defined(EC_CLZ) +/*Note that __builtin_clz is not defined when _x==0, according to the gcc + documentation (and that of the BSR instruction that implements it on x86). + The majority of the time we can never pass it zero. + When we need to, it can be special cased.*/ +# define EC_ILOG(_x) (EC_CLZ0-EC_CLZ(_x)) +#else +int ec_ilog(opus_uint32 _v); +# define EC_ILOG(_x) (ec_ilog(_x)) +#endif +#endif diff --git a/media/libopus/celt/entcode.c b/media/libopus/celt/entcode.c new file mode 100644 index 0000000000..70f32016ec --- /dev/null +++ b/media/libopus/celt/entcode.c @@ -0,0 +1,153 @@ +/* Copyright (c) 2001-2011 Timothy B. Terriberry +*/ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "entcode.h" +#include "arch.h" + +#if !defined(EC_CLZ) +/*This is a fallback for systems where we don't know how to access + a BSR or CLZ instruction (see ecintrin.h). + If you are optimizing Opus on a new platform and it has a native CLZ or + BZR (e.g. cell, MIPS, x86, etc) then making it available to Opus will be + an easy performance win.*/ +int ec_ilog(opus_uint32 _v){ + /*On a Pentium M, this branchless version tested as the fastest on + 1,000,000,000 random 32-bit integers, edging out a similar version with + branches, and a 256-entry LUT version.*/ + int ret; + int m; + ret=!!_v; + m=!!(_v&0xFFFF0000)<<4; + _v>>=m; + ret|=m; + m=!!(_v&0xFF00)<<3; + _v>>=m; + ret|=m; + m=!!(_v&0xF0)<<2; + _v>>=m; + ret|=m; + m=!!(_v&0xC)<<1; + _v>>=m; + ret|=m; + ret+=!!(_v&0x2); + return ret; +} +#endif + +#if 1 +/* This is a faster version of ec_tell_frac() that takes advantage + of the low (1/8 bit) resolution to use just a linear function + followed by a lookup to determine the exact transition thresholds. */ +opus_uint32 ec_tell_frac(ec_ctx *_this){ + static const unsigned correction[8] = + {35733, 38967, 42495, 46340, + 50535, 55109, 60097, 65535}; + opus_uint32 nbits; + opus_uint32 r; + int l; + unsigned b; + nbits=_this->nbits_total<<BITRES; + l=EC_ILOG(_this->rng); + r=_this->rng>>(l-16); + b = (r>>12)-8; + b += r>correction[b]; + l = (l<<3)+b; + return nbits-l; +} +#else +opus_uint32 ec_tell_frac(ec_ctx *_this){ + opus_uint32 nbits; + opus_uint32 r; + int l; + int i; + /*To handle the non-integral number of bits still left in the encoder/decoder + state, we compute the worst-case number of bits of val that must be + encoded to ensure that the value is inside the range for any possible + subsequent bits. + The computation here is independent of val itself (the decoder does not + even track that value), even though the real number of bits used after + ec_enc_done() may be 1 smaller if rng is a power of two and the + corresponding trailing bits of val are all zeros. + If we did try to track that special case, then coding a value with a + probability of 1/(1<<n) might sometimes appear to use more than n bits. + This may help explain the surprising result that a newly initialized + encoder or decoder claims to have used 1 bit.*/ + nbits=_this->nbits_total<<BITRES; + l=EC_ILOG(_this->rng); + r=_this->rng>>(l-16); + for(i=BITRES;i-->0;){ + int b; + r=r*r>>15; + b=(int)(r>>16); + l=l<<1|b; + r>>=b; + } + return nbits-l; +} +#endif + +#ifdef USE_SMALL_DIV_TABLE +/* Result of 2^32/(2*i+1), except for i=0. */ +const opus_uint32 SMALL_DIV_TABLE[129] = { + 0xFFFFFFFF, 0x55555555, 0x33333333, 0x24924924, + 0x1C71C71C, 0x1745D174, 0x13B13B13, 0x11111111, + 0x0F0F0F0F, 0x0D79435E, 0x0C30C30C, 0x0B21642C, + 0x0A3D70A3, 0x097B425E, 0x08D3DCB0, 0x08421084, + 0x07C1F07C, 0x07507507, 0x06EB3E45, 0x06906906, + 0x063E7063, 0x05F417D0, 0x05B05B05, 0x0572620A, + 0x05397829, 0x05050505, 0x04D4873E, 0x04A7904A, + 0x047DC11F, 0x0456C797, 0x04325C53, 0x04104104, + 0x03F03F03, 0x03D22635, 0x03B5CC0E, 0x039B0AD1, + 0x0381C0E0, 0x0369D036, 0x03531DEC, 0x033D91D2, + 0x0329161F, 0x03159721, 0x03030303, 0x02F14990, + 0x02E05C0B, 0x02D02D02, 0x02C0B02C, 0x02B1DA46, + 0x02A3A0FD, 0x0295FAD4, 0x0288DF0C, 0x027C4597, + 0x02702702, 0x02647C69, 0x02593F69, 0x024E6A17, + 0x0243F6F0, 0x0239E0D5, 0x02302302, 0x0226B902, + 0x021D9EAD, 0x0214D021, 0x020C49BA, 0x02040810, + 0x01FC07F0, 0x01F44659, 0x01ECC07B, 0x01E573AC, + 0x01DE5D6E, 0x01D77B65, 0x01D0CB58, 0x01CA4B30, + 0x01C3F8F0, 0x01BDD2B8, 0x01B7D6C3, 0x01B20364, + 0x01AC5701, 0x01A6D01A, 0x01A16D3F, 0x019C2D14, + 0x01970E4F, 0x01920FB4, 0x018D3018, 0x01886E5F, + 0x0183C977, 0x017F405F, 0x017AD220, 0x01767DCE, + 0x01724287, 0x016E1F76, 0x016A13CD, 0x01661EC6, + 0x01623FA7, 0x015E75BB, 0x015AC056, 0x01571ED3, + 0x01539094, 0x01501501, 0x014CAB88, 0x0149539E, + 0x01460CBC, 0x0142D662, 0x013FB013, 0x013C995A, + 0x013991C2, 0x013698DF, 0x0133AE45, 0x0130D190, + 0x012E025C, 0x012B404A, 0x01288B01, 0x0125E227, + 0x01234567, 0x0120B470, 0x011E2EF3, 0x011BB4A4, + 0x01194538, 0x0116E068, 0x011485F0, 0x0112358E, + 0x010FEF01, 0x010DB20A, 0x010B7E6E, 0x010953F3, + 0x01073260, 0x0105197F, 0x0103091B, 0x01010101 +}; +#endif diff --git a/media/libopus/celt/entcode.h b/media/libopus/celt/entcode.h new file mode 100644 index 0000000000..3763e3f284 --- /dev/null +++ b/media/libopus/celt/entcode.h @@ -0,0 +1,152 @@ +/* Copyright (c) 2001-2011 Timothy B. Terriberry + Copyright (c) 2008-2009 Xiph.Org Foundation */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#include "opus_types.h" +#include "opus_defines.h" + +#if !defined(_entcode_H) +# define _entcode_H (1) +# include <limits.h> +# include <stddef.h> +# include "ecintrin.h" + +extern const opus_uint32 SMALL_DIV_TABLE[129]; + +#ifdef OPUS_ARM_ASM +#define USE_SMALL_DIV_TABLE +#endif + +/*OPT: ec_window must be at least 32 bits, but if you have fast arithmetic on a + larger type, you can speed up the decoder by using it here.*/ +typedef opus_uint32 ec_window; +typedef struct ec_ctx ec_ctx; +typedef struct ec_ctx ec_enc; +typedef struct ec_ctx ec_dec; + +# define EC_WINDOW_SIZE ((int)sizeof(ec_window)*CHAR_BIT) + +/*The number of bits to use for the range-coded part of unsigned integers.*/ +# define EC_UINT_BITS (8) + +/*The resolution of fractional-precision bit usage measurements, i.e., + 3 => 1/8th bits.*/ +# define BITRES 3 + +/*The entropy encoder/decoder context. + We use the same structure for both, so that common functions like ec_tell() + can be used on either one.*/ +struct ec_ctx{ + /*Buffered input/output.*/ + unsigned char *buf; + /*The size of the buffer.*/ + opus_uint32 storage; + /*The offset at which the last byte containing raw bits was read/written.*/ + opus_uint32 end_offs; + /*Bits that will be read from/written at the end.*/ + ec_window end_window; + /*Number of valid bits in end_window.*/ + int nend_bits; + /*The total number of whole bits read/written. + This does not include partial bits currently in the range coder.*/ + int nbits_total; + /*The offset at which the next range coder byte will be read/written.*/ + opus_uint32 offs; + /*The number of values in the current range.*/ + opus_uint32 rng; + /*In the decoder: the difference between the top of the current range and + the input value, minus one. + In the encoder: the low end of the current range.*/ + opus_uint32 val; + /*In the decoder: the saved normalization factor from ec_decode(). + In the encoder: the number of oustanding carry propagating symbols.*/ + opus_uint32 ext; + /*A buffered input/output symbol, awaiting carry propagation.*/ + int rem; + /*Nonzero if an error occurred.*/ + int error; +}; + +static OPUS_INLINE opus_uint32 ec_range_bytes(ec_ctx *_this){ + return _this->offs; +} + +static OPUS_INLINE unsigned char *ec_get_buffer(ec_ctx *_this){ + return _this->buf; +} + +static OPUS_INLINE int ec_get_error(ec_ctx *_this){ + return _this->error; +} + +/*Returns the number of bits "used" by the encoded or decoded symbols so far. + This same number can be computed in either the encoder or the decoder, and is + suitable for making coding decisions. + Return: The number of bits. + This will always be slightly larger than the exact value (e.g., all + rounding error is in the positive direction).*/ +static OPUS_INLINE int ec_tell(ec_ctx *_this){ + return _this->nbits_total-EC_ILOG(_this->rng); +} + +/*Returns the number of bits "used" by the encoded or decoded symbols so far. + This same number can be computed in either the encoder or the decoder, and is + suitable for making coding decisions. + Return: The number of bits scaled by 2**BITRES. + This will always be slightly larger than the exact value (e.g., all + rounding error is in the positive direction).*/ +opus_uint32 ec_tell_frac(ec_ctx *_this); + +/* Tested exhaustively for all n and for 1<=d<=256 */ +static OPUS_INLINE opus_uint32 celt_udiv(opus_uint32 n, opus_uint32 d) { + celt_sig_assert(d>0); +#ifdef USE_SMALL_DIV_TABLE + if (d>256) + return n/d; + else { + opus_uint32 t, q; + t = EC_ILOG(d&-d); + q = (opus_uint64)SMALL_DIV_TABLE[d>>t]*(n>>(t-1))>>32; + return q+(n-q*d >= d); + } +#else + return n/d; +#endif +} + +static OPUS_INLINE opus_int32 celt_sudiv(opus_int32 n, opus_int32 d) { + celt_sig_assert(d>0); +#ifdef USE_SMALL_DIV_TABLE + if (n<0) + return -(opus_int32)celt_udiv(-n, d); + else + return celt_udiv(n, d); +#else + return n/d; +#endif +} + +#endif diff --git a/media/libopus/celt/entdec.c b/media/libopus/celt/entdec.c new file mode 100644 index 0000000000..0b3433ed8b --- /dev/null +++ b/media/libopus/celt/entdec.c @@ -0,0 +1,245 @@ +/* Copyright (c) 2001-2011 Timothy B. Terriberry + Copyright (c) 2008-2009 Xiph.Org Foundation */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include <stddef.h> +#include "os_support.h" +#include "arch.h" +#include "entdec.h" +#include "mfrngcod.h" + +/*A range decoder. + This is an entropy decoder based upon \cite{Mar79}, which is itself a + rediscovery of the FIFO arithmetic code introduced by \cite{Pas76}. + It is very similar to arithmetic encoding, except that encoding is done with + digits in any base, instead of with bits, and so it is faster when using + larger bases (i.e.: a byte). + The author claims an average waste of $\frac{1}{2}\log_b(2b)$ bits, where $b$ + is the base, longer than the theoretical optimum, but to my knowledge there + is no published justification for this claim. + This only seems true when using near-infinite precision arithmetic so that + the process is carried out with no rounding errors. + + An excellent description of implementation details is available at + http://www.arturocampos.com/ac_range.html + A recent work \cite{MNW98} which proposes several changes to arithmetic + encoding for efficiency actually re-discovers many of the principles + behind range encoding, and presents a good theoretical analysis of them. + + End of stream is handled by writing out the smallest number of bits that + ensures that the stream will be correctly decoded regardless of the value of + any subsequent bits. + ec_tell() can be used to determine how many bits were needed to decode + all the symbols thus far; other data can be packed in the remaining bits of + the input buffer. + @PHDTHESIS{Pas76, + author="Richard Clark Pasco", + title="Source coding algorithms for fast data compression", + school="Dept. of Electrical Engineering, Stanford University", + address="Stanford, CA", + month=May, + year=1976 + } + @INPROCEEDINGS{Mar79, + author="Martin, G.N.N.", + title="Range encoding: an algorithm for removing redundancy from a digitised + message", + booktitle="Video & Data Recording Conference", + year=1979, + address="Southampton", + month=Jul + } + @ARTICLE{MNW98, + author="Alistair Moffat and Radford Neal and Ian H. Witten", + title="Arithmetic Coding Revisited", + journal="{ACM} Transactions on Information Systems", + year=1998, + volume=16, + number=3, + pages="256--294", + month=Jul, + URL="http://www.stanford.edu/class/ee398a/handouts/papers/Moffat98ArithmCoding.pdf" + }*/ + +static int ec_read_byte(ec_dec *_this){ + return _this->offs<_this->storage?_this->buf[_this->offs++]:0; +} + +static int ec_read_byte_from_end(ec_dec *_this){ + return _this->end_offs<_this->storage? + _this->buf[_this->storage-++(_this->end_offs)]:0; +} + +/*Normalizes the contents of val and rng so that rng lies entirely in the + high-order symbol.*/ +static void ec_dec_normalize(ec_dec *_this){ + /*If the range is too small, rescale it and input some bits.*/ + while(_this->rng<=EC_CODE_BOT){ + int sym; + _this->nbits_total+=EC_SYM_BITS; + _this->rng<<=EC_SYM_BITS; + /*Use up the remaining bits from our last symbol.*/ + sym=_this->rem; + /*Read the next value from the input.*/ + _this->rem=ec_read_byte(_this); + /*Take the rest of the bits we need from this new symbol.*/ + sym=(sym<<EC_SYM_BITS|_this->rem)>>(EC_SYM_BITS-EC_CODE_EXTRA); + /*And subtract them from val, capped to be less than EC_CODE_TOP.*/ + _this->val=((_this->val<<EC_SYM_BITS)+(EC_SYM_MAX&~sym))&(EC_CODE_TOP-1); + } +} + +void ec_dec_init(ec_dec *_this,unsigned char *_buf,opus_uint32 _storage){ + _this->buf=_buf; + _this->storage=_storage; + _this->end_offs=0; + _this->end_window=0; + _this->nend_bits=0; + /*This is the offset from which ec_tell() will subtract partial bits. + The final value after the ec_dec_normalize() call will be the same as in + the encoder, but we have to compensate for the bits that are added there.*/ + _this->nbits_total=EC_CODE_BITS+1 + -((EC_CODE_BITS-EC_CODE_EXTRA)/EC_SYM_BITS)*EC_SYM_BITS; + _this->offs=0; + _this->rng=1U<<EC_CODE_EXTRA; + _this->rem=ec_read_byte(_this); + _this->val=_this->rng-1-(_this->rem>>(EC_SYM_BITS-EC_CODE_EXTRA)); + _this->error=0; + /*Normalize the interval.*/ + ec_dec_normalize(_this); +} + +unsigned ec_decode(ec_dec *_this,unsigned _ft){ + unsigned s; + _this->ext=celt_udiv(_this->rng,_ft); + s=(unsigned)(_this->val/_this->ext); + return _ft-EC_MINI(s+1,_ft); +} + +unsigned ec_decode_bin(ec_dec *_this,unsigned _bits){ + unsigned s; + _this->ext=_this->rng>>_bits; + s=(unsigned)(_this->val/_this->ext); + return (1U<<_bits)-EC_MINI(s+1U,1U<<_bits); +} + +void ec_dec_update(ec_dec *_this,unsigned _fl,unsigned _fh,unsigned _ft){ + opus_uint32 s; + s=IMUL32(_this->ext,_ft-_fh); + _this->val-=s; + _this->rng=_fl>0?IMUL32(_this->ext,_fh-_fl):_this->rng-s; + ec_dec_normalize(_this); +} + +/*The probability of having a "one" is 1/(1<<_logp).*/ +int ec_dec_bit_logp(ec_dec *_this,unsigned _logp){ + opus_uint32 r; + opus_uint32 d; + opus_uint32 s; + int ret; + r=_this->rng; + d=_this->val; + s=r>>_logp; + ret=d<s; + if(!ret)_this->val=d-s; + _this->rng=ret?s:r-s; + ec_dec_normalize(_this); + return ret; +} + +int ec_dec_icdf(ec_dec *_this,const unsigned char *_icdf,unsigned _ftb){ + opus_uint32 r; + opus_uint32 d; + opus_uint32 s; + opus_uint32 t; + int ret; + s=_this->rng; + d=_this->val; + r=s>>_ftb; + ret=-1; + do{ + t=s; + s=IMUL32(r,_icdf[++ret]); + } + while(d<s); + _this->val=d-s; + _this->rng=t-s; + ec_dec_normalize(_this); + return ret; +} + +opus_uint32 ec_dec_uint(ec_dec *_this,opus_uint32 _ft){ + unsigned ft; + unsigned s; + int ftb; + /*In order to optimize EC_ILOG(), it is undefined for the value 0.*/ + celt_assert(_ft>1); + _ft--; + ftb=EC_ILOG(_ft); + if(ftb>EC_UINT_BITS){ + opus_uint32 t; + ftb-=EC_UINT_BITS; + ft=(unsigned)(_ft>>ftb)+1; + s=ec_decode(_this,ft); + ec_dec_update(_this,s,s+1,ft); + t=(opus_uint32)s<<ftb|ec_dec_bits(_this,ftb); + if(t<=_ft)return t; + _this->error=1; + return _ft; + } + else{ + _ft++; + s=ec_decode(_this,(unsigned)_ft); + ec_dec_update(_this,s,s+1,(unsigned)_ft); + return s; + } +} + +opus_uint32 ec_dec_bits(ec_dec *_this,unsigned _bits){ + ec_window window; + int available; + opus_uint32 ret; + window=_this->end_window; + available=_this->nend_bits; + if((unsigned)available<_bits){ + do{ + window|=(ec_window)ec_read_byte_from_end(_this)<<available; + available+=EC_SYM_BITS; + } + while(available<=EC_WINDOW_SIZE-EC_SYM_BITS); + } + ret=(opus_uint32)window&(((opus_uint32)1<<_bits)-1U); + window>>=_bits; + available-=_bits; + _this->end_window=window; + _this->nend_bits=available; + _this->nbits_total+=_bits; + return ret; +} diff --git a/media/libopus/celt/entdec.h b/media/libopus/celt/entdec.h new file mode 100644 index 0000000000..025fc1870d --- /dev/null +++ b/media/libopus/celt/entdec.h @@ -0,0 +1,100 @@ +/* Copyright (c) 2001-2011 Timothy B. Terriberry + Copyright (c) 2008-2009 Xiph.Org Foundation */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#if !defined(_entdec_H) +# define _entdec_H (1) +# include <limits.h> +# include "entcode.h" + +/*Initializes the decoder. + _buf: The input buffer to use. + Return: 0 on success, or a negative value on error.*/ +void ec_dec_init(ec_dec *_this,unsigned char *_buf,opus_uint32 _storage); + +/*Calculates the cumulative frequency for the next symbol. + This can then be fed into the probability model to determine what that + symbol is, and the additional frequency information required to advance to + the next symbol. + This function cannot be called more than once without a corresponding call to + ec_dec_update(), or decoding will not proceed correctly. + _ft: The total frequency of the symbols in the alphabet the next symbol was + encoded with. + Return: A cumulative frequency representing the encoded symbol. + If the cumulative frequency of all the symbols before the one that + was encoded was fl, and the cumulative frequency of all the symbols + up to and including the one encoded is fh, then the returned value + will fall in the range [fl,fh).*/ +unsigned ec_decode(ec_dec *_this,unsigned _ft); + +/*Equivalent to ec_decode() with _ft==1<<_bits.*/ +unsigned ec_decode_bin(ec_dec *_this,unsigned _bits); + +/*Advance the decoder past the next symbol using the frequency information the + symbol was encoded with. + Exactly one call to ec_decode() must have been made so that all necessary + intermediate calculations are performed. + _fl: The cumulative frequency of all symbols that come before the symbol + decoded. + _fh: The cumulative frequency of all symbols up to and including the symbol + decoded. + Together with _fl, this defines the range [_fl,_fh) in which the value + returned above must fall. + _ft: The total frequency of the symbols in the alphabet the symbol decoded + was encoded in. + This must be the same as passed to the preceding call to ec_decode().*/ +void ec_dec_update(ec_dec *_this,unsigned _fl,unsigned _fh,unsigned _ft); + +/* Decode a bit that has a 1/(1<<_logp) probability of being a one */ +int ec_dec_bit_logp(ec_dec *_this,unsigned _logp); + +/*Decodes a symbol given an "inverse" CDF table. + No call to ec_dec_update() is necessary after this call. + _icdf: The "inverse" CDF, such that symbol s falls in the range + [s>0?ft-_icdf[s-1]:0,ft-_icdf[s]), where ft=1<<_ftb. + The values must be monotonically non-increasing, and the last value + must be 0. + _ftb: The number of bits of precision in the cumulative distribution. + Return: The decoded symbol s.*/ +int ec_dec_icdf(ec_dec *_this,const unsigned char *_icdf,unsigned _ftb); + +/*Extracts a raw unsigned integer with a non-power-of-2 range from the stream. + The bits must have been encoded with ec_enc_uint(). + No call to ec_dec_update() is necessary after this call. + _ft: The number of integers that can be decoded (one more than the max). + This must be at least 2, and no more than 2**32-1. + Return: The decoded bits.*/ +opus_uint32 ec_dec_uint(ec_dec *_this,opus_uint32 _ft); + +/*Extracts a sequence of raw bits from the stream. + The bits must have been encoded with ec_enc_bits(). + No call to ec_dec_update() is necessary after this call. + _ftb: The number of bits to extract. + This must be between 0 and 25, inclusive. + Return: The decoded bits.*/ +opus_uint32 ec_dec_bits(ec_dec *_this,unsigned _ftb); + +#endif diff --git a/media/libopus/celt/entenc.c b/media/libopus/celt/entenc.c new file mode 100644 index 0000000000..f1750d25b8 --- /dev/null +++ b/media/libopus/celt/entenc.c @@ -0,0 +1,294 @@ +/* Copyright (c) 2001-2011 Timothy B. Terriberry + Copyright (c) 2008-2009 Xiph.Org Foundation */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#if defined(HAVE_CONFIG_H) +# include "config.h" +#endif +#include "os_support.h" +#include "arch.h" +#include "entenc.h" +#include "mfrngcod.h" + +/*A range encoder. + See entdec.c and the references for implementation details \cite{Mar79,MNW98}. + + @INPROCEEDINGS{Mar79, + author="Martin, G.N.N.", + title="Range encoding: an algorithm for removing redundancy from a digitised + message", + booktitle="Video \& Data Recording Conference", + year=1979, + address="Southampton", + month=Jul + } + @ARTICLE{MNW98, + author="Alistair Moffat and Radford Neal and Ian H. Witten", + title="Arithmetic Coding Revisited", + journal="{ACM} Transactions on Information Systems", + year=1998, + volume=16, + number=3, + pages="256--294", + month=Jul, + URL="http://www.stanford.edu/class/ee398/handouts/papers/Moffat98ArithmCoding.pdf" + }*/ + +static int ec_write_byte(ec_enc *_this,unsigned _value){ + if(_this->offs+_this->end_offs>=_this->storage)return -1; + _this->buf[_this->offs++]=(unsigned char)_value; + return 0; +} + +static int ec_write_byte_at_end(ec_enc *_this,unsigned _value){ + if(_this->offs+_this->end_offs>=_this->storage)return -1; + _this->buf[_this->storage-++(_this->end_offs)]=(unsigned char)_value; + return 0; +} + +/*Outputs a symbol, with a carry bit. + If there is a potential to propagate a carry over several symbols, they are + buffered until it can be determined whether or not an actual carry will + occur. + If the counter for the buffered symbols overflows, then the stream becomes + undecodable. + This gives a theoretical limit of a few billion symbols in a single packet on + 32-bit systems. + The alternative is to truncate the range in order to force a carry, but + requires similar carry tracking in the decoder, needlessly slowing it down.*/ +static void ec_enc_carry_out(ec_enc *_this,int _c){ + if(_c!=EC_SYM_MAX){ + /*No further carry propagation possible, flush buffer.*/ + int carry; + carry=_c>>EC_SYM_BITS; + /*Don't output a byte on the first write. + This compare should be taken care of by branch-prediction thereafter.*/ + if(_this->rem>=0)_this->error|=ec_write_byte(_this,_this->rem+carry); + if(_this->ext>0){ + unsigned sym; + sym=(EC_SYM_MAX+carry)&EC_SYM_MAX; + do _this->error|=ec_write_byte(_this,sym); + while(--(_this->ext)>0); + } + _this->rem=_c&EC_SYM_MAX; + } + else _this->ext++; +} + +static OPUS_INLINE void ec_enc_normalize(ec_enc *_this){ + /*If the range is too small, output some bits and rescale it.*/ + while(_this->rng<=EC_CODE_BOT){ + ec_enc_carry_out(_this,(int)(_this->val>>EC_CODE_SHIFT)); + /*Move the next-to-high-order symbol into the high-order position.*/ + _this->val=(_this->val<<EC_SYM_BITS)&(EC_CODE_TOP-1); + _this->rng<<=EC_SYM_BITS; + _this->nbits_total+=EC_SYM_BITS; + } +} + +void ec_enc_init(ec_enc *_this,unsigned char *_buf,opus_uint32 _size){ + _this->buf=_buf; + _this->end_offs=0; + _this->end_window=0; + _this->nend_bits=0; + /*This is the offset from which ec_tell() will subtract partial bits.*/ + _this->nbits_total=EC_CODE_BITS+1; + _this->offs=0; + _this->rng=EC_CODE_TOP; + _this->rem=-1; + _this->val=0; + _this->ext=0; + _this->storage=_size; + _this->error=0; +} + +void ec_encode(ec_enc *_this,unsigned _fl,unsigned _fh,unsigned _ft){ + opus_uint32 r; + r=celt_udiv(_this->rng,_ft); + if(_fl>0){ + _this->val+=_this->rng-IMUL32(r,(_ft-_fl)); + _this->rng=IMUL32(r,(_fh-_fl)); + } + else _this->rng-=IMUL32(r,(_ft-_fh)); + ec_enc_normalize(_this); +} + +void ec_encode_bin(ec_enc *_this,unsigned _fl,unsigned _fh,unsigned _bits){ + opus_uint32 r; + r=_this->rng>>_bits; + if(_fl>0){ + _this->val+=_this->rng-IMUL32(r,((1U<<_bits)-_fl)); + _this->rng=IMUL32(r,(_fh-_fl)); + } + else _this->rng-=IMUL32(r,((1U<<_bits)-_fh)); + ec_enc_normalize(_this); +} + +/*The probability of having a "one" is 1/(1<<_logp).*/ +void ec_enc_bit_logp(ec_enc *_this,int _val,unsigned _logp){ + opus_uint32 r; + opus_uint32 s; + opus_uint32 l; + r=_this->rng; + l=_this->val; + s=r>>_logp; + r-=s; + if(_val)_this->val=l+r; + _this->rng=_val?s:r; + ec_enc_normalize(_this); +} + +void ec_enc_icdf(ec_enc *_this,int _s,const unsigned char *_icdf,unsigned _ftb){ + opus_uint32 r; + r=_this->rng>>_ftb; + if(_s>0){ + _this->val+=_this->rng-IMUL32(r,_icdf[_s-1]); + _this->rng=IMUL32(r,_icdf[_s-1]-_icdf[_s]); + } + else _this->rng-=IMUL32(r,_icdf[_s]); + ec_enc_normalize(_this); +} + +void ec_enc_uint(ec_enc *_this,opus_uint32 _fl,opus_uint32 _ft){ + unsigned ft; + unsigned fl; + int ftb; + /*In order to optimize EC_ILOG(), it is undefined for the value 0.*/ + celt_assert(_ft>1); + _ft--; + ftb=EC_ILOG(_ft); + if(ftb>EC_UINT_BITS){ + ftb-=EC_UINT_BITS; + ft=(_ft>>ftb)+1; + fl=(unsigned)(_fl>>ftb); + ec_encode(_this,fl,fl+1,ft); + ec_enc_bits(_this,_fl&(((opus_uint32)1<<ftb)-1U),ftb); + } + else ec_encode(_this,_fl,_fl+1,_ft+1); +} + +void ec_enc_bits(ec_enc *_this,opus_uint32 _fl,unsigned _bits){ + ec_window window; + int used; + window=_this->end_window; + used=_this->nend_bits; + celt_assert(_bits>0); + if(used+_bits>EC_WINDOW_SIZE){ + do{ + _this->error|=ec_write_byte_at_end(_this,(unsigned)window&EC_SYM_MAX); + window>>=EC_SYM_BITS; + used-=EC_SYM_BITS; + } + while(used>=EC_SYM_BITS); + } + window|=(ec_window)_fl<<used; + used+=_bits; + _this->end_window=window; + _this->nend_bits=used; + _this->nbits_total+=_bits; +} + +void ec_enc_patch_initial_bits(ec_enc *_this,unsigned _val,unsigned _nbits){ + int shift; + unsigned mask; + celt_assert(_nbits<=EC_SYM_BITS); + shift=EC_SYM_BITS-_nbits; + mask=((1<<_nbits)-1)<<shift; + if(_this->offs>0){ + /*The first byte has been finalized.*/ + _this->buf[0]=(unsigned char)((_this->buf[0]&~mask)|_val<<shift); + } + else if(_this->rem>=0){ + /*The first byte is still awaiting carry propagation.*/ + _this->rem=(_this->rem&~mask)|_val<<shift; + } + else if(_this->rng<=(EC_CODE_TOP>>_nbits)){ + /*The renormalization loop has never been run.*/ + _this->val=(_this->val&~((opus_uint32)mask<<EC_CODE_SHIFT))| + (opus_uint32)_val<<(EC_CODE_SHIFT+shift); + } + /*The encoder hasn't even encoded _nbits of data yet.*/ + else _this->error=-1; +} + +void ec_enc_shrink(ec_enc *_this,opus_uint32 _size){ + celt_assert(_this->offs+_this->end_offs<=_size); + OPUS_MOVE(_this->buf+_size-_this->end_offs, + _this->buf+_this->storage-_this->end_offs,_this->end_offs); + _this->storage=_size; +} + +void ec_enc_done(ec_enc *_this){ + ec_window window; + int used; + opus_uint32 msk; + opus_uint32 end; + int l; + /*We output the minimum number of bits that ensures that the symbols encoded + thus far will be decoded correctly regardless of the bits that follow.*/ + l=EC_CODE_BITS-EC_ILOG(_this->rng); + msk=(EC_CODE_TOP-1)>>l; + end=(_this->val+msk)&~msk; + if((end|msk)>=_this->val+_this->rng){ + l++; + msk>>=1; + end=(_this->val+msk)&~msk; + } + while(l>0){ + ec_enc_carry_out(_this,(int)(end>>EC_CODE_SHIFT)); + end=(end<<EC_SYM_BITS)&(EC_CODE_TOP-1); + l-=EC_SYM_BITS; + } + /*If we have a buffered byte flush it into the output buffer.*/ + if(_this->rem>=0||_this->ext>0)ec_enc_carry_out(_this,0); + /*If we have buffered extra bits, flush them as well.*/ + window=_this->end_window; + used=_this->nend_bits; + while(used>=EC_SYM_BITS){ + _this->error|=ec_write_byte_at_end(_this,(unsigned)window&EC_SYM_MAX); + window>>=EC_SYM_BITS; + used-=EC_SYM_BITS; + } + /*Clear any excess space and add any remaining extra bits to the last byte.*/ + if(!_this->error){ + OPUS_CLEAR(_this->buf+_this->offs, + _this->storage-_this->offs-_this->end_offs); + if(used>0){ + /*If there's no range coder data at all, give up.*/ + if(_this->end_offs>=_this->storage)_this->error=-1; + else{ + l=-l; + /*If we've busted, don't add too many extra bits to the last byte; it + would corrupt the range coder data, and that's more important.*/ + if(_this->offs+_this->end_offs>=_this->storage&&l<used){ + window&=(1<<l)-1; + _this->error=-1; + } + _this->buf[_this->storage-_this->end_offs-1]|=(unsigned char)window; + } + } + } +} diff --git a/media/libopus/celt/entenc.h b/media/libopus/celt/entenc.h new file mode 100644 index 0000000000..f502eaf662 --- /dev/null +++ b/media/libopus/celt/entenc.h @@ -0,0 +1,110 @@ +/* Copyright (c) 2001-2011 Timothy B. Terriberry + Copyright (c) 2008-2009 Xiph.Org Foundation */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#if !defined(_entenc_H) +# define _entenc_H (1) +# include <stddef.h> +# include "entcode.h" + +/*Initializes the encoder. + _buf: The buffer to store output bytes in. + _size: The size of the buffer, in chars.*/ +void ec_enc_init(ec_enc *_this,unsigned char *_buf,opus_uint32 _size); +/*Encodes a symbol given its frequency information. + The frequency information must be discernable by the decoder, assuming it + has read only the previous symbols from the stream. + It is allowable to change the frequency information, or even the entire + source alphabet, so long as the decoder can tell from the context of the + previously encoded information that it is supposed to do so as well. + _fl: The cumulative frequency of all symbols that come before the one to be + encoded. + _fh: The cumulative frequency of all symbols up to and including the one to + be encoded. + Together with _fl, this defines the range [_fl,_fh) in which the + decoded value will fall. + _ft: The sum of the frequencies of all the symbols*/ +void ec_encode(ec_enc *_this,unsigned _fl,unsigned _fh,unsigned _ft); + +/*Equivalent to ec_encode() with _ft==1<<_bits.*/ +void ec_encode_bin(ec_enc *_this,unsigned _fl,unsigned _fh,unsigned _bits); + +/* Encode a bit that has a 1/(1<<_logp) probability of being a one */ +void ec_enc_bit_logp(ec_enc *_this,int _val,unsigned _logp); + +/*Encodes a symbol given an "inverse" CDF table. + _s: The index of the symbol to encode. + _icdf: The "inverse" CDF, such that symbol _s falls in the range + [_s>0?ft-_icdf[_s-1]:0,ft-_icdf[_s]), where ft=1<<_ftb. + The values must be monotonically non-increasing, and the last value + must be 0. + _ftb: The number of bits of precision in the cumulative distribution.*/ +void ec_enc_icdf(ec_enc *_this,int _s,const unsigned char *_icdf,unsigned _ftb); + +/*Encodes a raw unsigned integer in the stream. + _fl: The integer to encode. + _ft: The number of integers that can be encoded (one more than the max). + This must be at least 2, and no more than 2**32-1.*/ +void ec_enc_uint(ec_enc *_this,opus_uint32 _fl,opus_uint32 _ft); + +/*Encodes a sequence of raw bits in the stream. + _fl: The bits to encode. + _ftb: The number of bits to encode. + This must be between 1 and 25, inclusive.*/ +void ec_enc_bits(ec_enc *_this,opus_uint32 _fl,unsigned _ftb); + +/*Overwrites a few bits at the very start of an existing stream, after they + have already been encoded. + This makes it possible to have a few flags up front, where it is easy for + decoders to access them without parsing the whole stream, even if their + values are not determined until late in the encoding process, without having + to buffer all the intermediate symbols in the encoder. + In order for this to work, at least _nbits bits must have already been + encoded using probabilities that are an exact power of two. + The encoder can verify the number of encoded bits is sufficient, but cannot + check this latter condition. + _val: The bits to encode (in the least _nbits significant bits). + They will be decoded in order from most-significant to least. + _nbits: The number of bits to overwrite. + This must be no more than 8.*/ +void ec_enc_patch_initial_bits(ec_enc *_this,unsigned _val,unsigned _nbits); + +/*Compacts the data to fit in the target size. + This moves up the raw bits at the end of the current buffer so they are at + the end of the new buffer size. + The caller must ensure that the amount of data that's already been written + will fit in the new size. + _size: The number of bytes in the new buffer. + This must be large enough to contain the bits already written, and + must be no larger than the existing size.*/ +void ec_enc_shrink(ec_enc *_this,opus_uint32 _size); + +/*Indicates that there are no more symbols to encode. + All reamining output bytes are flushed to the output buffer. + ec_enc_init() must be called before the encoder can be used again.*/ +void ec_enc_done(ec_enc *_this); + +#endif diff --git a/media/libopus/celt/fixed_debug.h b/media/libopus/celt/fixed_debug.h new file mode 100644 index 0000000000..ef2e5d02fb --- /dev/null +++ b/media/libopus/celt/fixed_debug.h @@ -0,0 +1,836 @@ +/* Copyright (C) 2003-2008 Jean-Marc Valin + Copyright (C) 2007-2012 Xiph.Org Foundation */ +/** + @file fixed_debug.h + @brief Fixed-point operations with debugging +*/ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef FIXED_DEBUG_H +#define FIXED_DEBUG_H + +#include <stdio.h> +#include "opus_defines.h" + +#ifdef CELT_C +OPUS_EXPORT opus_int64 celt_mips=0; +#else +extern opus_int64 celt_mips; +#endif + +#define MULT16_16SU(a,b) ((opus_val32)(opus_val16)(a)*(opus_val32)(opus_uint16)(b)) +#define MULT32_32_Q31(a,b) ADD32(ADD32(SHL32(MULT16_16(SHR32((a),16),SHR((b),16)),1), SHR32(MULT16_16SU(SHR32((a),16),((b)&0x0000ffff)),15)), SHR32(MULT16_16SU(SHR32((b),16),((a)&0x0000ffff)),15)) + +/** 16x32 multiplication, followed by a 16-bit shift right. Results fits in 32 bits */ +#define MULT16_32_Q16(a,b) ADD32(MULT16_16((a),SHR32((b),16)), SHR32(MULT16_16SU((a),((b)&0x0000ffff)),16)) + +#define MULT16_32_P16(a,b) MULT16_32_PX(a,b,16) + +#define QCONST16(x,bits) ((opus_val16)(.5+(x)*(((opus_val32)1)<<(bits)))) +#define QCONST32(x,bits) ((opus_val32)(.5+(x)*(((opus_val32)1)<<(bits)))) + +#define VERIFY_SHORT(x) ((x)<=32767&&(x)>=-32768) +#define VERIFY_INT(x) ((x)<=2147483647LL&&(x)>=-2147483648LL) +#define VERIFY_UINT(x) ((x)<=(2147483647LLU<<1)) + +#define SHR(a,b) SHR32(a,b) +#define PSHR(a,b) PSHR32(a,b) + +/** Add two 32-bit values, ignore any overflows */ +#define ADD32_ovflw(a,b) (celt_mips+=2,(opus_val32)((opus_uint32)(a)+(opus_uint32)(b))) +/** Subtract two 32-bit values, ignore any overflows */ +#define SUB32_ovflw(a,b) (celt_mips+=2,(opus_val32)((opus_uint32)(a)-(opus_uint32)(b))) +/* Avoid MSVC warning C4146: unary minus operator applied to unsigned type */ +/** Negate 32-bit value, ignore any overflows */ +#define NEG32_ovflw(a) (celt_mips+=2,(opus_val32)(0-(opus_uint32)(a))) + +static OPUS_INLINE short NEG16(int x) +{ + int res; + if (!VERIFY_SHORT(x)) + { + fprintf (stderr, "NEG16: input is not short: %d\n", (int)x); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = -x; + if (!VERIFY_SHORT(res)) + { + fprintf (stderr, "NEG16: output is not short: %d\n", (int)res); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips++; + return res; +} +static OPUS_INLINE int NEG32(opus_int64 x) +{ + opus_int64 res; + if (!VERIFY_INT(x)) + { + fprintf (stderr, "NEG16: input is not int: %d\n", (int)x); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = -x; + if (!VERIFY_INT(res)) + { + fprintf (stderr, "NEG16: output is not int: %d\n", (int)res); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=2; + return res; +} + +#define EXTRACT16(x) EXTRACT16_(x, __FILE__, __LINE__) +static OPUS_INLINE short EXTRACT16_(int x, char *file, int line) +{ + int res; + if (!VERIFY_SHORT(x)) + { + fprintf (stderr, "EXTRACT16: input is not short: %d in %s: line %d\n", x, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = x; + celt_mips++; + return res; +} + +#define EXTEND32(x) EXTEND32_(x, __FILE__, __LINE__) +static OPUS_INLINE int EXTEND32_(int x, char *file, int line) +{ + int res; + if (!VERIFY_SHORT(x)) + { + fprintf (stderr, "EXTEND32: input is not short: %d in %s: line %d\n", x, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = x; + celt_mips++; + return res; +} + +#define SHR16(a, shift) SHR16_(a, shift, __FILE__, __LINE__) +static OPUS_INLINE short SHR16_(int a, int shift, char *file, int line) +{ + int res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(shift)) + { + fprintf (stderr, "SHR16: inputs are not short: %d >> %d in %s: line %d\n", a, shift, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = a>>shift; + if (!VERIFY_SHORT(res)) + { + fprintf (stderr, "SHR16: output is not short: %d in %s: line %d\n", res, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips++; + return res; +} +#define SHL16(a, shift) SHL16_(a, shift, __FILE__, __LINE__) +static OPUS_INLINE short SHL16_(int a, int shift, char *file, int line) +{ + opus_int32 res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(shift)) + { + fprintf (stderr, "SHL16: inputs are not short: %d %d in %s: line %d\n", a, shift, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = (opus_int32)((opus_uint32)a<<shift); + if (!VERIFY_SHORT(res)) + { + fprintf (stderr, "SHL16: output is not short: %d in %s: line %d\n", res, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips++; + return res; +} + +static OPUS_INLINE int SHR32(opus_int64 a, int shift) +{ + opus_int64 res; + if (!VERIFY_INT(a) || !VERIFY_SHORT(shift)) + { + fprintf (stderr, "SHR32: inputs are not int: %d %d\n", (int)a, shift); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = a>>shift; + if (!VERIFY_INT(res)) + { + fprintf (stderr, "SHR32: output is not int: %d\n", (int)res); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=2; + return res; +} +#define SHL32(a, shift) SHL32_(a, shift, __FILE__, __LINE__) +static OPUS_INLINE int SHL32_(opus_int64 a, int shift, char *file, int line) +{ + opus_int64 res; + if (!VERIFY_INT(a) || !VERIFY_SHORT(shift)) + { + fprintf (stderr, "SHL32: inputs are not int: %lld %d in %s: line %d\n", (long long)a, shift, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = (opus_int64)((opus_uint64)a<<shift); + if (!VERIFY_INT(res)) + { + fprintf (stderr, "SHL32: output is not int: %lld<<%d = %lld in %s: line %d\n", (long long)a, shift, (long long)res, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=2; + return res; +} + +#define PSHR32(a,shift) (celt_mips--,SHR32(ADD32((a),(((opus_val32)(1)<<((shift))>>1))),shift)) +#define VSHR32(a, shift) (((shift)>0) ? SHR32(a, shift) : SHL32(a, -(shift))) + +#define ROUND16(x,a) (celt_mips--,EXTRACT16(PSHR32((x),(a)))) +#define SROUND16(x,a) (celt_mips--,EXTRACT16(SATURATE(PSHR32(x,a), 32767))); + +#define HALF16(x) (SHR16(x,1)) +#define HALF32(x) (SHR32(x,1)) + +#define ADD16(a, b) ADD16_(a, b, __FILE__, __LINE__) +static OPUS_INLINE short ADD16_(int a, int b, char *file, int line) +{ + int res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "ADD16: inputs are not short: %d %d in %s: line %d\n", a, b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = a+b; + if (!VERIFY_SHORT(res)) + { + fprintf (stderr, "ADD16: output is not short: %d+%d=%d in %s: line %d\n", a,b,res, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips++; + return res; +} + +#define SUB16(a, b) SUB16_(a, b, __FILE__, __LINE__) +static OPUS_INLINE short SUB16_(int a, int b, char *file, int line) +{ + int res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "SUB16: inputs are not short: %d %d in %s: line %d\n", a, b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = a-b; + if (!VERIFY_SHORT(res)) + { + fprintf (stderr, "SUB16: output is not short: %d in %s: line %d\n", res, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips++; + return res; +} + +#define ADD32(a, b) ADD32_(a, b, __FILE__, __LINE__) +static OPUS_INLINE int ADD32_(opus_int64 a, opus_int64 b, char *file, int line) +{ + opus_int64 res; + if (!VERIFY_INT(a) || !VERIFY_INT(b)) + { + fprintf (stderr, "ADD32: inputs are not int: %d %d in %s: line %d\n", (int)a, (int)b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = a+b; + if (!VERIFY_INT(res)) + { + fprintf (stderr, "ADD32: output is not int: %d in %s: line %d\n", (int)res, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=2; + return res; +} + +#define SUB32(a, b) SUB32_(a, b, __FILE__, __LINE__) +static OPUS_INLINE int SUB32_(opus_int64 a, opus_int64 b, char *file, int line) +{ + opus_int64 res; + if (!VERIFY_INT(a) || !VERIFY_INT(b)) + { + fprintf (stderr, "SUB32: inputs are not int: %d %d in %s: line %d\n", (int)a, (int)b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = a-b; + if (!VERIFY_INT(res)) + { + fprintf (stderr, "SUB32: output is not int: %d in %s: line %d\n", (int)res, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=2; + return res; +} + +#undef UADD32 +#define UADD32(a, b) UADD32_(a, b, __FILE__, __LINE__) +static OPUS_INLINE unsigned int UADD32_(opus_uint64 a, opus_uint64 b, char *file, int line) +{ + opus_uint64 res; + if (!VERIFY_UINT(a) || !VERIFY_UINT(b)) + { + fprintf (stderr, "UADD32: inputs are not uint32: %llu %llu in %s: line %d\n", (unsigned long long)a, (unsigned long long)b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = a+b; + if (!VERIFY_UINT(res)) + { + fprintf (stderr, "UADD32: output is not uint32: %llu in %s: line %d\n", (unsigned long long)res, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=2; + return res; +} + +#undef USUB32 +#define USUB32(a, b) USUB32_(a, b, __FILE__, __LINE__) +static OPUS_INLINE unsigned int USUB32_(opus_uint64 a, opus_uint64 b, char *file, int line) +{ + opus_uint64 res; + if (!VERIFY_UINT(a) || !VERIFY_UINT(b)) + { + fprintf (stderr, "USUB32: inputs are not uint32: %llu %llu in %s: line %d\n", (unsigned long long)a, (unsigned long long)b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + if (a<b) + { + fprintf (stderr, "USUB32: inputs underflow: %llu < %llu in %s: line %d\n", (unsigned long long)a, (unsigned long long)b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = a-b; + if (!VERIFY_UINT(res)) + { + fprintf (stderr, "USUB32: output is not uint32: %llu - %llu = %llu in %s: line %d\n", (unsigned long long)a, (unsigned long long)b, (unsigned long long)res, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=2; + return res; +} + +/* result fits in 16 bits */ +static OPUS_INLINE short MULT16_16_16(int a, int b) +{ + int res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "MULT16_16_16: inputs are not short: %d %d\n", a, b); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = a*b; + if (!VERIFY_SHORT(res)) + { + fprintf (stderr, "MULT16_16_16: output is not short: %d\n", res); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips++; + return res; +} + +/* result fits in 32 bits */ +static OPUS_INLINE int MULT32_32_32(opus_int64 a, opus_int64 b) +{ + opus_int64 res; + if (!VERIFY_INT(a) || !VERIFY_INT(b)) + { + fprintf (stderr, "MULT32_32_32: inputs are not int: %lld %lld\n", (long long)a, (long long)b); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = a*b; + if (!VERIFY_INT(res)) + { + fprintf (stderr, "MULT32_32_32: output is not int: %lld\n", (long long)res); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=5; + return res; +} + +static OPUS_INLINE int MULT32_32_Q16(opus_int64 a, opus_int64 b) +{ + opus_int64 res; + if (!VERIFY_INT(a) || !VERIFY_INT(b)) + { + fprintf (stderr, "MULT32_32_Q16: inputs are not int: %lld %lld\n", (long long)a, (long long)b); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = ((opus_int64)(a)*(opus_int64)(b)) >> 16; + if (!VERIFY_INT(res)) + { + fprintf (stderr, "MULT32_32_Q16: output is not int: %lld*%lld=%lld\n", (long long)a, (long long)b, (long long)res); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=5; + return res; +} + +#define MULT16_16(a, b) MULT16_16_(a, b, __FILE__, __LINE__) +static OPUS_INLINE int MULT16_16_(int a, int b, char *file, int line) +{ + opus_int64 res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "MULT16_16: inputs are not short: %d %d in %s: line %d\n", a, b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = ((opus_int64)a)*b; + if (!VERIFY_INT(res)) + { + fprintf (stderr, "MULT16_16: output is not int: %d in %s: line %d\n", (int)res, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips++; + return res; +} + +#define MAC16_16(c,a,b) (celt_mips-=2,ADD32((c),MULT16_16((a),(b)))) + +#define MULT16_32_QX(a, b, Q) MULT16_32_QX_(a, b, Q, __FILE__, __LINE__) +static OPUS_INLINE int MULT16_32_QX_(int a, opus_int64 b, int Q, char *file, int line) +{ + opus_int64 res; + if (!VERIFY_SHORT(a) || !VERIFY_INT(b)) + { + fprintf (stderr, "MULT16_32_Q%d: inputs are not short+int: %d %d in %s: line %d\n", Q, (int)a, (int)b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + if (ABS32(b)>=((opus_int64)(1)<<(16+Q))) + { + fprintf (stderr, "MULT16_32_Q%d: second operand too large: %d %d in %s: line %d\n", Q, (int)a, (int)b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = (((opus_int64)a)*(opus_int64)b) >> Q; + if (!VERIFY_INT(res)) + { + fprintf (stderr, "MULT16_32_Q%d: output is not int: %d*%d=%d in %s: line %d\n", Q, (int)a, (int)b,(int)res, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + if (Q==15) + celt_mips+=3; + else + celt_mips+=4; + return res; +} + +#define MULT16_32_PX(a, b, Q) MULT16_32_PX_(a, b, Q, __FILE__, __LINE__) +static OPUS_INLINE int MULT16_32_PX_(int a, opus_int64 b, int Q, char *file, int line) +{ + opus_int64 res; + if (!VERIFY_SHORT(a) || !VERIFY_INT(b)) + { + fprintf (stderr, "MULT16_32_P%d: inputs are not short+int: %d %d in %s: line %d\n\n", Q, (int)a, (int)b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + if (ABS32(b)>=((opus_int64)(1)<<(16+Q))) + { + fprintf (stderr, "MULT16_32_Q%d: second operand too large: %d %d in %s: line %d\n\n", Q, (int)a, (int)b,file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = ((((opus_int64)a)*(opus_int64)b) + (((opus_val32)(1)<<Q)>>1))>> Q; + if (!VERIFY_INT(res)) + { + fprintf (stderr, "MULT16_32_P%d: output is not int: %d*%d=%d in %s: line %d\n\n", Q, (int)a, (int)b,(int)res, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + if (Q==15) + celt_mips+=4; + else + celt_mips+=5; + return res; +} + +#define MULT16_32_Q15(a,b) MULT16_32_QX(a,b,15) +#define MAC16_32_Q15(c,a,b) (celt_mips-=2,ADD32((c),MULT16_32_Q15((a),(b)))) +#define MAC16_32_Q16(c,a,b) (celt_mips-=2,ADD32((c),MULT16_32_Q16((a),(b)))) + +static OPUS_INLINE int SATURATE(int a, int b) +{ + if (a>b) + a=b; + if (a<-b) + a = -b; + celt_mips+=3; + return a; +} + +static OPUS_INLINE opus_int16 SATURATE16(opus_int32 a) +{ + celt_mips+=3; + if (a>32767) + return 32767; + else if (a<-32768) + return -32768; + else return a; +} + +static OPUS_INLINE int MULT16_16_Q11_32(int a, int b) +{ + opus_int64 res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "MULT16_16_Q11: inputs are not short: %d %d\n", a, b); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = ((opus_int64)a)*b; + res >>= 11; + if (!VERIFY_INT(res)) + { + fprintf (stderr, "MULT16_16_Q11: output is not short: %d*%d=%d\n", (int)a, (int)b, (int)res); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=3; + return res; +} +static OPUS_INLINE short MULT16_16_Q13(int a, int b) +{ + opus_int64 res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "MULT16_16_Q13: inputs are not short: %d %d\n", a, b); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = ((opus_int64)a)*b; + res >>= 13; + if (!VERIFY_SHORT(res)) + { + fprintf (stderr, "MULT16_16_Q13: output is not short: %d*%d=%d\n", a, b, (int)res); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=3; + return res; +} +static OPUS_INLINE short MULT16_16_Q14(int a, int b) +{ + opus_int64 res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "MULT16_16_Q14: inputs are not short: %d %d\n", a, b); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = ((opus_int64)a)*b; + res >>= 14; + if (!VERIFY_SHORT(res)) + { + fprintf (stderr, "MULT16_16_Q14: output is not short: %d\n", (int)res); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=3; + return res; +} + +#define MULT16_16_Q15(a, b) MULT16_16_Q15_(a, b, __FILE__, __LINE__) +static OPUS_INLINE short MULT16_16_Q15_(int a, int b, char *file, int line) +{ + opus_int64 res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "MULT16_16_Q15: inputs are not short: %d %d in %s: line %d\n", a, b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = ((opus_int64)a)*b; + res >>= 15; + if (!VERIFY_SHORT(res)) + { + fprintf (stderr, "MULT16_16_Q15: output is not short: %d in %s: line %d\n", (int)res, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=1; + return res; +} + +static OPUS_INLINE short MULT16_16_P13(int a, int b) +{ + opus_int64 res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "MULT16_16_P13: inputs are not short: %d %d\n", a, b); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = ((opus_int64)a)*b; + res += 4096; + if (!VERIFY_INT(res)) + { + fprintf (stderr, "MULT16_16_P13: overflow: %d*%d=%d\n", a, b, (int)res); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res >>= 13; + if (!VERIFY_SHORT(res)) + { + fprintf (stderr, "MULT16_16_P13: output is not short: %d*%d=%d\n", a, b, (int)res); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=4; + return res; +} +static OPUS_INLINE short MULT16_16_P14(int a, int b) +{ + opus_int64 res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "MULT16_16_P14: inputs are not short: %d %d\n", a, b); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = ((opus_int64)a)*b; + res += 8192; + if (!VERIFY_INT(res)) + { + fprintf (stderr, "MULT16_16_P14: overflow: %d*%d=%d\n", a, b, (int)res); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res >>= 14; + if (!VERIFY_SHORT(res)) + { + fprintf (stderr, "MULT16_16_P14: output is not short: %d*%d=%d\n", a, b, (int)res); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=4; + return res; +} +static OPUS_INLINE short MULT16_16_P15(int a, int b) +{ + opus_int64 res; + if (!VERIFY_SHORT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "MULT16_16_P15: inputs are not short: %d %d\n", a, b); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = ((opus_int64)a)*b; + res += 16384; + if (!VERIFY_INT(res)) + { + fprintf (stderr, "MULT16_16_P15: overflow: %d*%d=%d\n", a, b, (int)res); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res >>= 15; + if (!VERIFY_SHORT(res)) + { + fprintf (stderr, "MULT16_16_P15: output is not short: %d*%d=%d\n", a, b, (int)res); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=2; + return res; +} + +#define DIV32_16(a, b) DIV32_16_(a, b, __FILE__, __LINE__) + +static OPUS_INLINE int DIV32_16_(opus_int64 a, opus_int64 b, char *file, int line) +{ + opus_int64 res; + if (b==0) + { + fprintf(stderr, "DIV32_16: divide by zero: %d/%d in %s: line %d\n", (int)a, (int)b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + return 0; + } + if (!VERIFY_INT(a) || !VERIFY_SHORT(b)) + { + fprintf (stderr, "DIV32_16: inputs are not int/short: %d %d in %s: line %d\n", (int)a, (int)b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = a/b; + if (!VERIFY_SHORT(res)) + { + fprintf (stderr, "DIV32_16: output is not short: %d / %d = %d in %s: line %d\n", (int)a,(int)b,(int)res, file, line); + if (res>32767) + res = 32767; + if (res<-32768) + res = -32768; +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=35; + return res; +} + +#define DIV32(a, b) DIV32_(a, b, __FILE__, __LINE__) +static OPUS_INLINE int DIV32_(opus_int64 a, opus_int64 b, char *file, int line) +{ + opus_int64 res; + if (b==0) + { + fprintf(stderr, "DIV32: divide by zero: %d/%d in %s: line %d\n", (int)a, (int)b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + return 0; + } + + if (!VERIFY_INT(a) || !VERIFY_INT(b)) + { + fprintf (stderr, "DIV32: inputs are not int/short: %d %d in %s: line %d\n", (int)a, (int)b, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + res = a/b; + if (!VERIFY_INT(res)) + { + fprintf (stderr, "DIV32: output is not int: %d in %s: line %d\n", (int)res, file, line); +#ifdef FIXED_DEBUG_ASSERT + celt_assert(0); +#endif + } + celt_mips+=70; + return res; +} + +static OPUS_INLINE opus_val16 SIG2WORD16_generic(celt_sig x) +{ + x = PSHR32(x, SIG_SHIFT); + x = MAX32(x, -32768); + x = MIN32(x, 32767); + return EXTRACT16(x); +} +#define SIG2WORD16(x) (SIG2WORD16_generic(x)) + + +#undef PRINT_MIPS +#define PRINT_MIPS(file) do {fprintf (file, "total complexity = %llu MIPS\n", (unsigned long long)celt_mips);} while (0); + +#endif diff --git a/media/libopus/celt/fixed_generic.h b/media/libopus/celt/fixed_generic.h new file mode 100644 index 0000000000..8f29d46bb7 --- /dev/null +++ b/media/libopus/celt/fixed_generic.h @@ -0,0 +1,188 @@ +/* Copyright (C) 2007-2009 Xiph.Org Foundation + Copyright (C) 2003-2008 Jean-Marc Valin + Copyright (C) 2007-2008 CSIRO */ +/** + @file fixed_generic.h + @brief Generic fixed-point operations +*/ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef FIXED_GENERIC_H +#define FIXED_GENERIC_H + +/** Multiply a 16-bit signed value by a 16-bit unsigned value. The result is a 32-bit signed value */ +#define MULT16_16SU(a,b) ((opus_val32)(opus_val16)(a)*(opus_val32)(opus_uint16)(b)) + +/** 16x32 multiplication, followed by a 16-bit shift right. Results fits in 32 bits */ +#if OPUS_FAST_INT64 +#define MULT16_32_Q16(a,b) ((opus_val32)SHR((opus_int64)((opus_val16)(a))*(b),16)) +#else +#define MULT16_32_Q16(a,b) ADD32(MULT16_16((a),SHR((b),16)), SHR(MULT16_16SU((a),((b)&0x0000ffff)),16)) +#endif + +/** 16x32 multiplication, followed by a 16-bit shift right (round-to-nearest). Results fits in 32 bits */ +#if OPUS_FAST_INT64 +#define MULT16_32_P16(a,b) ((opus_val32)PSHR((opus_int64)((opus_val16)(a))*(b),16)) +#else +#define MULT16_32_P16(a,b) ADD32(MULT16_16((a),SHR((b),16)), PSHR(MULT16_16SU((a),((b)&0x0000ffff)),16)) +#endif + +/** 16x32 multiplication, followed by a 15-bit shift right. Results fits in 32 bits */ +#if OPUS_FAST_INT64 +#define MULT16_32_Q15(a,b) ((opus_val32)SHR((opus_int64)((opus_val16)(a))*(b),15)) +#else +#define MULT16_32_Q15(a,b) ADD32(SHL(MULT16_16((a),SHR((b),16)),1), SHR(MULT16_16SU((a),((b)&0x0000ffff)),15)) +#endif + +/** 32x32 multiplication, followed by a 16-bit shift right. Results fits in 32 bits */ +#if OPUS_FAST_INT64 +#define MULT32_32_Q16(a,b) ((opus_val32)SHR((opus_int64)(a)*(opus_int64)(b),16)) +#else +#define MULT32_32_Q16(a,b) (ADD32(ADD32(ADD32((opus_val32)(SHR32(((opus_uint32)((a)&0x0000ffff)*(opus_uint32)((b)&0x0000ffff)),16)), MULT16_16SU(SHR32(a,16),((b)&0x0000ffff))), MULT16_16SU(SHR32(b,16),((a)&0x0000ffff))), SHL32(MULT16_16(SHR32(a,16),SHR32(b,16)),16))) +#endif + +/** 32x32 multiplication, followed by a 31-bit shift right. Results fits in 32 bits */ +#if OPUS_FAST_INT64 +#define MULT32_32_Q31(a,b) ((opus_val32)SHR((opus_int64)(a)*(opus_int64)(b),31)) +#else +#define MULT32_32_Q31(a,b) ADD32(ADD32(SHL(MULT16_16(SHR((a),16),SHR((b),16)),1), SHR(MULT16_16SU(SHR((a),16),((b)&0x0000ffff)),15)), SHR(MULT16_16SU(SHR((b),16),((a)&0x0000ffff)),15)) +#endif + +/** Compile-time conversion of float constant to 16-bit value */ +#define QCONST16(x,bits) ((opus_val16)(.5+(x)*(((opus_val32)1)<<(bits)))) + +/** Compile-time conversion of float constant to 32-bit value */ +#define QCONST32(x,bits) ((opus_val32)(.5+(x)*(((opus_val32)1)<<(bits)))) + +/** Negate a 16-bit value */ +#define NEG16(x) (-(x)) +/** Negate a 32-bit value */ +#define NEG32(x) (-(x)) + +/** Change a 32-bit value into a 16-bit value. The value is assumed to fit in 16-bit, otherwise the result is undefined */ +#define EXTRACT16(x) ((opus_val16)(x)) +/** Change a 16-bit value into a 32-bit value */ +#define EXTEND32(x) ((opus_val32)(x)) + +/** Arithmetic shift-right of a 16-bit value */ +#define SHR16(a,shift) ((a) >> (shift)) +/** Arithmetic shift-left of a 16-bit value */ +#define SHL16(a,shift) ((opus_int16)((opus_uint16)(a)<<(shift))) +/** Arithmetic shift-right of a 32-bit value */ +#define SHR32(a,shift) ((a) >> (shift)) +/** Arithmetic shift-left of a 32-bit value */ +#define SHL32(a,shift) ((opus_int32)((opus_uint32)(a)<<(shift))) + +/** 32-bit arithmetic shift right with rounding-to-nearest instead of rounding down */ +#define PSHR32(a,shift) (SHR32((a)+((EXTEND32(1)<<((shift))>>1)),shift)) +/** 32-bit arithmetic shift right where the argument can be negative */ +#define VSHR32(a, shift) (((shift)>0) ? SHR32(a, shift) : SHL32(a, -(shift))) + +/** "RAW" macros, should not be used outside of this header file */ +#define SHR(a,shift) ((a) >> (shift)) +#define SHL(a,shift) SHL32(a,shift) +#define PSHR(a,shift) (SHR((a)+((EXTEND32(1)<<((shift))>>1)),shift)) +#define SATURATE(x,a) (((x)>(a) ? (a) : (x)<-(a) ? -(a) : (x))) + +#define SATURATE16(x) (EXTRACT16((x)>32767 ? 32767 : (x)<-32768 ? -32768 : (x))) + +/** Shift by a and round-to-nearest 32-bit value. Result is a 16-bit value */ +#define ROUND16(x,a) (EXTRACT16(PSHR32((x),(a)))) +/** Shift by a and round-to-nearest 32-bit value. Result is a saturated 16-bit value */ +#define SROUND16(x,a) EXTRACT16(SATURATE(PSHR32(x,a), 32767)); + +/** Divide by two */ +#define HALF16(x) (SHR16(x,1)) +#define HALF32(x) (SHR32(x,1)) + +/** Add two 16-bit values */ +#define ADD16(a,b) ((opus_val16)((opus_val16)(a)+(opus_val16)(b))) +/** Subtract two 16-bit values */ +#define SUB16(a,b) ((opus_val16)(a)-(opus_val16)(b)) +/** Add two 32-bit values */ +#define ADD32(a,b) ((opus_val32)(a)+(opus_val32)(b)) +/** Subtract two 32-bit values */ +#define SUB32(a,b) ((opus_val32)(a)-(opus_val32)(b)) + +/** Add two 32-bit values, ignore any overflows */ +#define ADD32_ovflw(a,b) ((opus_val32)((opus_uint32)(a)+(opus_uint32)(b))) +/** Subtract two 32-bit values, ignore any overflows */ +#define SUB32_ovflw(a,b) ((opus_val32)((opus_uint32)(a)-(opus_uint32)(b))) +/* Avoid MSVC warning C4146: unary minus operator applied to unsigned type */ +/** Negate 32-bit value, ignore any overflows */ +#define NEG32_ovflw(a) ((opus_val32)(0-(opus_uint32)(a))) + +/** 16x16 multiplication where the result fits in 16 bits */ +#define MULT16_16_16(a,b) ((((opus_val16)(a))*((opus_val16)(b)))) + +/** 32x32 multiplication where the result fits in 32 bits */ +#define MULT32_32_32(a,b) ((((opus_val32)(a))*((opus_val32)(b)))) + +/* (opus_val32)(opus_val16) gives TI compiler a hint that it's 16x16->32 multiply */ +/** 16x16 multiplication where the result fits in 32 bits */ +#define MULT16_16(a,b) (((opus_val32)(opus_val16)(a))*((opus_val32)(opus_val16)(b))) + +/** 16x16 multiply-add where the result fits in 32 bits */ +#define MAC16_16(c,a,b) (ADD32((c),MULT16_16((a),(b)))) +/** 16x32 multiply, followed by a 15-bit shift right and 32-bit add. + b must fit in 31 bits. + Result fits in 32 bits. */ +#define MAC16_32_Q15(c,a,b) ADD32((c),ADD32(MULT16_16((a),SHR((b),15)), SHR(MULT16_16((a),((b)&0x00007fff)),15))) + +/** 16x32 multiplication, followed by a 16-bit shift right and 32-bit add. + Results fits in 32 bits */ +#define MAC16_32_Q16(c,a,b) ADD32((c),ADD32(MULT16_16((a),SHR((b),16)), SHR(MULT16_16SU((a),((b)&0x0000ffff)),16))) + +#define MULT16_16_Q11_32(a,b) (SHR(MULT16_16((a),(b)),11)) +#define MULT16_16_Q11(a,b) (SHR(MULT16_16((a),(b)),11)) +#define MULT16_16_Q13(a,b) (SHR(MULT16_16((a),(b)),13)) +#define MULT16_16_Q14(a,b) (SHR(MULT16_16((a),(b)),14)) +#define MULT16_16_Q15(a,b) (SHR(MULT16_16((a),(b)),15)) + +#define MULT16_16_P13(a,b) (SHR(ADD32(4096,MULT16_16((a),(b))),13)) +#define MULT16_16_P14(a,b) (SHR(ADD32(8192,MULT16_16((a),(b))),14)) +#define MULT16_16_P15(a,b) (SHR(ADD32(16384,MULT16_16((a),(b))),15)) + +/** Divide a 32-bit value by a 16-bit value. Result fits in 16 bits */ +#define DIV32_16(a,b) ((opus_val16)(((opus_val32)(a))/((opus_val16)(b)))) + +/** Divide a 32-bit value by a 32-bit value. Result fits in 32 bits */ +#define DIV32(a,b) (((opus_val32)(a))/((opus_val32)(b))) + +#if defined(MIPSr1_ASM) +#include "mips/fixed_generic_mipsr1.h" +#endif + +static OPUS_INLINE opus_val16 SIG2WORD16_generic(celt_sig x) +{ + x = PSHR32(x, SIG_SHIFT); + x = MAX32(x, -32768); + x = MIN32(x, 32767); + return EXTRACT16(x); +} +#define SIG2WORD16(x) (SIG2WORD16_generic(x)) + +#endif diff --git a/media/libopus/celt/float_cast.h b/media/libopus/celt/float_cast.h new file mode 100644 index 0000000000..8915a5fd7f --- /dev/null +++ b/media/libopus/celt/float_cast.h @@ -0,0 +1,152 @@ +/* Copyright (C) 2001 Erik de Castro Lopo <erikd AT mega-nerd DOT com> */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +/* Version 1.1 */ + +#ifndef FLOAT_CAST_H +#define FLOAT_CAST_H + + +#include "arch.h" + +/*============================================================================ +** On Intel Pentium processors (especially PIII and probably P4), converting +** from float to int is very slow. To meet the C specs, the code produced by +** most C compilers targeting Pentium needs to change the FPU rounding mode +** before the float to int conversion is performed. +** +** Changing the FPU rounding mode causes the FPU pipeline to be flushed. It +** is this flushing of the pipeline which is so slow. +** +** Fortunately the ISO C99 specifications define the functions lrint, lrintf, +** llrint and llrintf which fix this problem as a side effect. +** +** On Unix-like systems, the configure process should have detected the +** presence of these functions. If they weren't found we have to replace them +** here with a standard C cast. +*/ + +/* +** The C99 prototypes for lrint and lrintf are as follows: +** +** long int lrintf (float x) ; +** long int lrint (double x) ; +*/ + +/* The presence of the required functions are detected during the configure +** process and the values HAVE_LRINT and HAVE_LRINTF are set accordingly in +** the config.h file. +*/ + +/* With GCC, when SSE is available, the fastest conversion is cvtss2si. */ +#if defined(__GNUC__) && defined(__SSE__) + +#include <xmmintrin.h> +static OPUS_INLINE opus_int32 float2int(float x) {return _mm_cvt_ss2si(_mm_set_ss(x));} + +#elif (defined(_MSC_VER) && _MSC_VER >= 1400) && (defined(_M_X64) || (defined(_M_IX86_FP) && _M_IX86_FP >= 1)) + + #include <xmmintrin.h> + static OPUS_INLINE opus_int32 float2int(float value) + { + /* _mm_load_ss will generate same code as _mm_set_ss + ** in _MSC_VER >= 1914 /02 so keep __mm_load__ss + ** for backward compatibility. + */ + return _mm_cvtss_si32(_mm_load_ss(&value)); + } + +#elif (defined(_MSC_VER) && _MSC_VER >= 1400) && defined (_M_IX86) + + #include <math.h> + + /* Win32 doesn't seem to have these functions. + ** Therefore implement OPUS_INLINE versions of these functions here. + */ + + static OPUS_INLINE opus_int32 + float2int (float flt) + { int intgr; + + _asm + { fld flt + fistp intgr + } ; + + return intgr ; + } + +#elif defined(HAVE_LRINTF) && defined(__STDC_VERSION__) && __STDC_VERSION__ >= 199901L + +/* These defines enable functionality introduced with the 1999 ISO C +** standard. They must be defined before the inclusion of math.h to +** engage them. If optimisation is enabled, these functions will be +** inlined. With optimisation switched off, you have to link in the +** maths library using -lm. +*/ + +#define _ISOC9X_SOURCE 1 +#define _ISOC99_SOURCE 1 + +#define __USE_ISOC9X 1 +#define __USE_ISOC99 1 + +#include <math.h> +#define float2int(x) lrintf(x) + +#elif defined(HAVE_LRINT) && defined(__STDC_VERSION__) && __STDC_VERSION__ >= 199901L + +#define _ISOC9X_SOURCE 1 +#define _ISOC99_SOURCE 1 + +#define __USE_ISOC9X 1 +#define __USE_ISOC99 1 + +#include <math.h> +#define float2int(x) lrint(x) + +#else + +#if (defined(__GNUC__) && defined(__STDC__) && __STDC__ && __STDC_VERSION__ >= 199901L) + /* supported by gcc in C99 mode, but not by all other compilers */ + #warning "Don't have the functions lrint() and lrintf ()." + #warning "Replacing these functions with a standard C cast." +#endif /* __STDC_VERSION__ >= 199901L */ + #include <math.h> + #define float2int(flt) ((int)(floor(.5+flt))) +#endif + +#ifndef DISABLE_FLOAT_API +static OPUS_INLINE opus_int16 FLOAT2INT16(float x) +{ + x = x*CELT_SIG_SCALE; + x = MAX32(x, -32768); + x = MIN32(x, 32767); + return (opus_int16)float2int(x); +} +#endif /* DISABLE_FLOAT_API */ + +#endif /* FLOAT_CAST_H */ diff --git a/media/libopus/celt/kiss_fft.c b/media/libopus/celt/kiss_fft.c new file mode 100644 index 0000000000..83775165d8 --- /dev/null +++ b/media/libopus/celt/kiss_fft.c @@ -0,0 +1,604 @@ +/*Copyright (c) 2003-2004, Mark Borgerding + Lots of modifications by Jean-Marc Valin + Copyright (c) 2005-2007, Xiph.Org Foundation + Copyright (c) 2008, Xiph.Org Foundation, CSIRO + + All rights reserved. + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions are met: + + * Redistributions of source code must retain the above copyright notice, + this list of conditions and the following disclaimer. + * Redistributions in binary form must reproduce the above copyright notice, + this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" + AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE + LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR + CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF + SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS + INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN + CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) + ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE + POSSIBILITY OF SUCH DAMAGE.*/ + +/* This code is originally from Mark Borgerding's KISS-FFT but has been + heavily modified to better suit Opus */ + +#ifndef SKIP_CONFIG_H +# ifdef HAVE_CONFIG_H +# include "config.h" +# endif +#endif + +#include "_kiss_fft_guts.h" +#include "arch.h" +#include "os_support.h" +#include "mathops.h" +#include "stack_alloc.h" + +/* The guts header contains all the multiplication and addition macros that are defined for + complex numbers. It also delares the kf_ internal functions. +*/ + +static void kf_bfly2( + kiss_fft_cpx * Fout, + int m, + int N + ) +{ + kiss_fft_cpx * Fout2; + int i; + (void)m; +#ifdef CUSTOM_MODES + if (m==1) + { + celt_assert(m==1); + for (i=0;i<N;i++) + { + kiss_fft_cpx t; + Fout2 = Fout + 1; + t = *Fout2; + C_SUB( *Fout2 , *Fout , t ); + C_ADDTO( *Fout , t ); + Fout += 2; + } + } else +#endif + { + opus_val16 tw; + tw = QCONST16(0.7071067812f, 15); + /* We know that m==4 here because the radix-2 is just after a radix-4 */ + celt_assert(m==4); + for (i=0;i<N;i++) + { + kiss_fft_cpx t; + Fout2 = Fout + 4; + t = Fout2[0]; + C_SUB( Fout2[0] , Fout[0] , t ); + C_ADDTO( Fout[0] , t ); + + t.r = S_MUL(ADD32_ovflw(Fout2[1].r, Fout2[1].i), tw); + t.i = S_MUL(SUB32_ovflw(Fout2[1].i, Fout2[1].r), tw); + C_SUB( Fout2[1] , Fout[1] , t ); + C_ADDTO( Fout[1] , t ); + + t.r = Fout2[2].i; + t.i = -Fout2[2].r; + C_SUB( Fout2[2] , Fout[2] , t ); + C_ADDTO( Fout[2] , t ); + + t.r = S_MUL(SUB32_ovflw(Fout2[3].i, Fout2[3].r), tw); + t.i = S_MUL(NEG32_ovflw(ADD32_ovflw(Fout2[3].i, Fout2[3].r)), tw); + C_SUB( Fout2[3] , Fout[3] , t ); + C_ADDTO( Fout[3] , t ); + Fout += 8; + } + } +} + +static void kf_bfly4( + kiss_fft_cpx * Fout, + const size_t fstride, + const kiss_fft_state *st, + int m, + int N, + int mm + ) +{ + int i; + + if (m==1) + { + /* Degenerate case where all the twiddles are 1. */ + for (i=0;i<N;i++) + { + kiss_fft_cpx scratch0, scratch1; + + C_SUB( scratch0 , *Fout, Fout[2] ); + C_ADDTO(*Fout, Fout[2]); + C_ADD( scratch1 , Fout[1] , Fout[3] ); + C_SUB( Fout[2], *Fout, scratch1 ); + C_ADDTO( *Fout , scratch1 ); + C_SUB( scratch1 , Fout[1] , Fout[3] ); + + Fout[1].r = ADD32_ovflw(scratch0.r, scratch1.i); + Fout[1].i = SUB32_ovflw(scratch0.i, scratch1.r); + Fout[3].r = SUB32_ovflw(scratch0.r, scratch1.i); + Fout[3].i = ADD32_ovflw(scratch0.i, scratch1.r); + Fout+=4; + } + } else { + int j; + kiss_fft_cpx scratch[6]; + const kiss_twiddle_cpx *tw1,*tw2,*tw3; + const int m2=2*m; + const int m3=3*m; + kiss_fft_cpx * Fout_beg = Fout; + for (i=0;i<N;i++) + { + Fout = Fout_beg + i*mm; + tw3 = tw2 = tw1 = st->twiddles; + /* m is guaranteed to be a multiple of 4. */ + for (j=0;j<m;j++) + { + C_MUL(scratch[0],Fout[m] , *tw1 ); + C_MUL(scratch[1],Fout[m2] , *tw2 ); + C_MUL(scratch[2],Fout[m3] , *tw3 ); + + C_SUB( scratch[5] , *Fout, scratch[1] ); + C_ADDTO(*Fout, scratch[1]); + C_ADD( scratch[3] , scratch[0] , scratch[2] ); + C_SUB( scratch[4] , scratch[0] , scratch[2] ); + C_SUB( Fout[m2], *Fout, scratch[3] ); + tw1 += fstride; + tw2 += fstride*2; + tw3 += fstride*3; + C_ADDTO( *Fout , scratch[3] ); + + Fout[m].r = ADD32_ovflw(scratch[5].r, scratch[4].i); + Fout[m].i = SUB32_ovflw(scratch[5].i, scratch[4].r); + Fout[m3].r = SUB32_ovflw(scratch[5].r, scratch[4].i); + Fout[m3].i = ADD32_ovflw(scratch[5].i, scratch[4].r); + ++Fout; + } + } + } +} + + +#ifndef RADIX_TWO_ONLY + +static void kf_bfly3( + kiss_fft_cpx * Fout, + const size_t fstride, + const kiss_fft_state *st, + int m, + int N, + int mm + ) +{ + int i; + size_t k; + const size_t m2 = 2*m; + const kiss_twiddle_cpx *tw1,*tw2; + kiss_fft_cpx scratch[5]; + kiss_twiddle_cpx epi3; + + kiss_fft_cpx * Fout_beg = Fout; +#ifdef FIXED_POINT + /*epi3.r = -16384;*/ /* Unused */ + epi3.i = -28378; +#else + epi3 = st->twiddles[fstride*m]; +#endif + for (i=0;i<N;i++) + { + Fout = Fout_beg + i*mm; + tw1=tw2=st->twiddles; + /* For non-custom modes, m is guaranteed to be a multiple of 4. */ + k=m; + do { + + C_MUL(scratch[1],Fout[m] , *tw1); + C_MUL(scratch[2],Fout[m2] , *tw2); + + C_ADD(scratch[3],scratch[1],scratch[2]); + C_SUB(scratch[0],scratch[1],scratch[2]); + tw1 += fstride; + tw2 += fstride*2; + + Fout[m].r = SUB32_ovflw(Fout->r, HALF_OF(scratch[3].r)); + Fout[m].i = SUB32_ovflw(Fout->i, HALF_OF(scratch[3].i)); + + C_MULBYSCALAR( scratch[0] , epi3.i ); + + C_ADDTO(*Fout,scratch[3]); + + Fout[m2].r = ADD32_ovflw(Fout[m].r, scratch[0].i); + Fout[m2].i = SUB32_ovflw(Fout[m].i, scratch[0].r); + + Fout[m].r = SUB32_ovflw(Fout[m].r, scratch[0].i); + Fout[m].i = ADD32_ovflw(Fout[m].i, scratch[0].r); + + ++Fout; + } while(--k); + } +} + + +#ifndef OVERRIDE_kf_bfly5 +static void kf_bfly5( + kiss_fft_cpx * Fout, + const size_t fstride, + const kiss_fft_state *st, + int m, + int N, + int mm + ) +{ + kiss_fft_cpx *Fout0,*Fout1,*Fout2,*Fout3,*Fout4; + int i, u; + kiss_fft_cpx scratch[13]; + const kiss_twiddle_cpx *tw; + kiss_twiddle_cpx ya,yb; + kiss_fft_cpx * Fout_beg = Fout; + +#ifdef FIXED_POINT + ya.r = 10126; + ya.i = -31164; + yb.r = -26510; + yb.i = -19261; +#else + ya = st->twiddles[fstride*m]; + yb = st->twiddles[fstride*2*m]; +#endif + tw=st->twiddles; + + for (i=0;i<N;i++) + { + Fout = Fout_beg + i*mm; + Fout0=Fout; + Fout1=Fout0+m; + Fout2=Fout0+2*m; + Fout3=Fout0+3*m; + Fout4=Fout0+4*m; + + /* For non-custom modes, m is guaranteed to be a multiple of 4. */ + for ( u=0; u<m; ++u ) { + scratch[0] = *Fout0; + + C_MUL(scratch[1] ,*Fout1, tw[u*fstride]); + C_MUL(scratch[2] ,*Fout2, tw[2*u*fstride]); + C_MUL(scratch[3] ,*Fout3, tw[3*u*fstride]); + C_MUL(scratch[4] ,*Fout4, tw[4*u*fstride]); + + C_ADD( scratch[7],scratch[1],scratch[4]); + C_SUB( scratch[10],scratch[1],scratch[4]); + C_ADD( scratch[8],scratch[2],scratch[3]); + C_SUB( scratch[9],scratch[2],scratch[3]); + + Fout0->r = ADD32_ovflw(Fout0->r, ADD32_ovflw(scratch[7].r, scratch[8].r)); + Fout0->i = ADD32_ovflw(Fout0->i, ADD32_ovflw(scratch[7].i, scratch[8].i)); + + scratch[5].r = ADD32_ovflw(scratch[0].r, ADD32_ovflw(S_MUL(scratch[7].r,ya.r), S_MUL(scratch[8].r,yb.r))); + scratch[5].i = ADD32_ovflw(scratch[0].i, ADD32_ovflw(S_MUL(scratch[7].i,ya.r), S_MUL(scratch[8].i,yb.r))); + + scratch[6].r = ADD32_ovflw(S_MUL(scratch[10].i,ya.i), S_MUL(scratch[9].i,yb.i)); + scratch[6].i = NEG32_ovflw(ADD32_ovflw(S_MUL(scratch[10].r,ya.i), S_MUL(scratch[9].r,yb.i))); + + C_SUB(*Fout1,scratch[5],scratch[6]); + C_ADD(*Fout4,scratch[5],scratch[6]); + + scratch[11].r = ADD32_ovflw(scratch[0].r, ADD32_ovflw(S_MUL(scratch[7].r,yb.r), S_MUL(scratch[8].r,ya.r))); + scratch[11].i = ADD32_ovflw(scratch[0].i, ADD32_ovflw(S_MUL(scratch[7].i,yb.r), S_MUL(scratch[8].i,ya.r))); + scratch[12].r = SUB32_ovflw(S_MUL(scratch[9].i,ya.i), S_MUL(scratch[10].i,yb.i)); + scratch[12].i = SUB32_ovflw(S_MUL(scratch[10].r,yb.i), S_MUL(scratch[9].r,ya.i)); + + C_ADD(*Fout2,scratch[11],scratch[12]); + C_SUB(*Fout3,scratch[11],scratch[12]); + + ++Fout0;++Fout1;++Fout2;++Fout3;++Fout4; + } + } +} +#endif /* OVERRIDE_kf_bfly5 */ + + +#endif + + +#ifdef CUSTOM_MODES + +static +void compute_bitrev_table( + int Fout, + opus_int16 *f, + const size_t fstride, + int in_stride, + opus_int16 * factors, + const kiss_fft_state *st + ) +{ + const int p=*factors++; /* the radix */ + const int m=*factors++; /* stage's fft length/p */ + + /*printf ("fft %d %d %d %d %d %d\n", p*m, m, p, s2, fstride*in_stride, N);*/ + if (m==1) + { + int j; + for (j=0;j<p;j++) + { + *f = Fout+j; + f += fstride*in_stride; + } + } else { + int j; + for (j=0;j<p;j++) + { + compute_bitrev_table( Fout , f, fstride*p, in_stride, factors,st); + f += fstride*in_stride; + Fout += m; + } + } +} + +/* facbuf is populated by p1,m1,p2,m2, ... + where + p[i] * m[i] = m[i-1] + m0 = n */ +static +int kf_factor(int n,opus_int16 * facbuf) +{ + int p=4; + int i; + int stages=0; + int nbak = n; + + /*factor out powers of 4, powers of 2, then any remaining primes */ + do { + while (n % p) { + switch (p) { + case 4: p = 2; break; + case 2: p = 3; break; + default: p += 2; break; + } + if (p>32000 || (opus_int32)p*(opus_int32)p > n) + p = n; /* no more factors, skip to end */ + } + n /= p; +#ifdef RADIX_TWO_ONLY + if (p!=2 && p != 4) +#else + if (p>5) +#endif + { + return 0; + } + facbuf[2*stages] = p; + if (p==2 && stages > 1) + { + facbuf[2*stages] = 4; + facbuf[2] = 2; + } + stages++; + } while (n > 1); + n = nbak; + /* Reverse the order to get the radix 4 at the end, so we can use the + fast degenerate case. It turns out that reversing the order also + improves the noise behaviour. */ + for (i=0;i<stages/2;i++) + { + int tmp; + tmp = facbuf[2*i]; + facbuf[2*i] = facbuf[2*(stages-i-1)]; + facbuf[2*(stages-i-1)] = tmp; + } + for (i=0;i<stages;i++) + { + n /= facbuf[2*i]; + facbuf[2*i+1] = n; + } + return 1; +} + +static void compute_twiddles(kiss_twiddle_cpx *twiddles, int nfft) +{ + int i; +#ifdef FIXED_POINT + for (i=0;i<nfft;++i) { + opus_val32 phase = -i; + kf_cexp2(twiddles+i, DIV32(SHL32(phase,17),nfft)); + } +#else + for (i=0;i<nfft;++i) { + const double pi=3.14159265358979323846264338327; + double phase = ( -2*pi /nfft ) * i; + kf_cexp(twiddles+i, phase ); + } +#endif +} + +int opus_fft_alloc_arch_c(kiss_fft_state *st) { + (void)st; + return 0; +} + +/* + * + * Allocates all necessary storage space for the fft and ifft. + * The return value is a contiguous block of memory. As such, + * It can be freed with free(). + * */ +kiss_fft_state *opus_fft_alloc_twiddles(int nfft,void * mem,size_t * lenmem, + const kiss_fft_state *base, int arch) +{ + kiss_fft_state *st=NULL; + size_t memneeded = sizeof(struct kiss_fft_state); /* twiddle factors*/ + + if ( lenmem==NULL ) { + st = ( kiss_fft_state*)KISS_FFT_MALLOC( memneeded ); + }else{ + if (mem != NULL && *lenmem >= memneeded) + st = (kiss_fft_state*)mem; + *lenmem = memneeded; + } + if (st) { + opus_int16 *bitrev; + kiss_twiddle_cpx *twiddles; + + st->nfft=nfft; +#ifdef FIXED_POINT + st->scale_shift = celt_ilog2(st->nfft); + if (st->nfft == 1<<st->scale_shift) + st->scale = Q15ONE; + else + st->scale = (1073741824+st->nfft/2)/st->nfft>>(15-st->scale_shift); +#else + st->scale = 1.f/nfft; +#endif + if (base != NULL) + { + st->twiddles = base->twiddles; + st->shift = 0; + while (st->shift < 32 && nfft<<st->shift != base->nfft) + st->shift++; + if (st->shift>=32) + goto fail; + } else { + st->twiddles = twiddles = (kiss_twiddle_cpx*)KISS_FFT_MALLOC(sizeof(kiss_twiddle_cpx)*nfft); + compute_twiddles(twiddles, nfft); + st->shift = -1; + } + if (!kf_factor(nfft,st->factors)) + { + goto fail; + } + + /* bitrev */ + st->bitrev = bitrev = (opus_int16*)KISS_FFT_MALLOC(sizeof(opus_int16)*nfft); + if (st->bitrev==NULL) + goto fail; + compute_bitrev_table(0, bitrev, 1,1, st->factors,st); + + /* Initialize architecture specific fft parameters */ + if (opus_fft_alloc_arch(st, arch)) + goto fail; + } + return st; +fail: + opus_fft_free(st, arch); + return NULL; +} + +kiss_fft_state *opus_fft_alloc(int nfft,void * mem,size_t * lenmem, int arch) +{ + return opus_fft_alloc_twiddles(nfft, mem, lenmem, NULL, arch); +} + +void opus_fft_free_arch_c(kiss_fft_state *st) { + (void)st; +} + +void opus_fft_free(const kiss_fft_state *cfg, int arch) +{ + if (cfg) + { + opus_fft_free_arch((kiss_fft_state *)cfg, arch); + opus_free((opus_int16*)cfg->bitrev); + if (cfg->shift < 0) + opus_free((kiss_twiddle_cpx*)cfg->twiddles); + opus_free((kiss_fft_state*)cfg); + } +} + +#endif /* CUSTOM_MODES */ + +void opus_fft_impl(const kiss_fft_state *st,kiss_fft_cpx *fout) +{ + int m2, m; + int p; + int L; + int fstride[MAXFACTORS]; + int i; + int shift; + + /* st->shift can be -1 */ + shift = st->shift>0 ? st->shift : 0; + + fstride[0] = 1; + L=0; + do { + p = st->factors[2*L]; + m = st->factors[2*L+1]; + fstride[L+1] = fstride[L]*p; + L++; + } while(m!=1); + m = st->factors[2*L-1]; + for (i=L-1;i>=0;i--) + { + if (i!=0) + m2 = st->factors[2*i-1]; + else + m2 = 1; + switch (st->factors[2*i]) + { + case 2: + kf_bfly2(fout, m, fstride[i]); + break; + case 4: + kf_bfly4(fout,fstride[i]<<shift,st,m, fstride[i], m2); + break; + #ifndef RADIX_TWO_ONLY + case 3: + kf_bfly3(fout,fstride[i]<<shift,st,m, fstride[i], m2); + break; + case 5: + kf_bfly5(fout,fstride[i]<<shift,st,m, fstride[i], m2); + break; + #endif + } + m = m2; + } +} + +void opus_fft_c(const kiss_fft_state *st,const kiss_fft_cpx *fin,kiss_fft_cpx *fout) +{ + int i; + opus_val16 scale; +#ifdef FIXED_POINT + /* Allows us to scale with MULT16_32_Q16(), which is faster than + MULT16_32_Q15() on ARM. */ + int scale_shift = st->scale_shift-1; +#endif + scale = st->scale; + + celt_assert2 (fin != fout, "In-place FFT not supported"); + /* Bit-reverse the input */ + for (i=0;i<st->nfft;i++) + { + kiss_fft_cpx x = fin[i]; + fout[st->bitrev[i]].r = SHR32(MULT16_32_Q16(scale, x.r), scale_shift); + fout[st->bitrev[i]].i = SHR32(MULT16_32_Q16(scale, x.i), scale_shift); + } + opus_fft_impl(st, fout); +} + + +void opus_ifft_c(const kiss_fft_state *st,const kiss_fft_cpx *fin,kiss_fft_cpx *fout) +{ + int i; + celt_assert2 (fin != fout, "In-place FFT not supported"); + /* Bit-reverse the input */ + for (i=0;i<st->nfft;i++) + fout[st->bitrev[i]] = fin[i]; + for (i=0;i<st->nfft;i++) + fout[i].i = -fout[i].i; + opus_fft_impl(st, fout); + for (i=0;i<st->nfft;i++) + fout[i].i = -fout[i].i; +} diff --git a/media/libopus/celt/kiss_fft.h b/media/libopus/celt/kiss_fft.h new file mode 100644 index 0000000000..267f72f99c --- /dev/null +++ b/media/libopus/celt/kiss_fft.h @@ -0,0 +1,210 @@ +/*Copyright (c) 2003-2004, Mark Borgerding + Lots of modifications by Jean-Marc Valin + Copyright (c) 2005-2007, Xiph.Org Foundation + Copyright (c) 2008, Xiph.Org Foundation, CSIRO + + All rights reserved. + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions are met: + + * Redistributions of source code must retain the above copyright notice, + this list of conditions and the following disclaimer. + * Redistributions in binary form must reproduce the above copyright notice, + this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" + AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE + LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR + CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF + SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS + INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN + CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) + ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE + POSSIBILITY OF SUCH DAMAGE.*/ + +#ifndef KISS_FFT_H +#define KISS_FFT_H + +#include <stdlib.h> +#include <math.h> +#include "arch.h" +#include "cpu_support.h" + +#ifdef __cplusplus +extern "C" { +#endif + +#ifdef USE_SIMD +# include <xmmintrin.h> +# define kiss_fft_scalar __m128 +#define KISS_FFT_MALLOC(nbytes) memalign(16,nbytes) +#else +#define KISS_FFT_MALLOC opus_alloc +#endif + +#ifdef FIXED_POINT +#include "arch.h" + +# define kiss_fft_scalar opus_int32 +# define kiss_twiddle_scalar opus_int16 + +/* Some 32-bit CPUs would load/store a kiss_twiddle_cpx with a single memory + * access, and could benefit from additional alignment. + */ +# define KISS_TWIDDLE_CPX_ALIGNMENT (sizeof(opus_int32)) + +#else +# ifndef kiss_fft_scalar +/* default is float */ +# define kiss_fft_scalar float +# define kiss_twiddle_scalar float +# define KF_SUFFIX _celt_single +# endif +#endif + +#if defined(__GNUC__) && defined(KISS_TWIDDLE_CPX_ALIGNMENT) +#define KISS_TWIDDLE_CPX_ALIGNED __attribute__((aligned(KISS_TWIDDLE_CPX_ALIGNMENT))) +#else +#define KISS_TWIDDLE_CPX_ALIGNED +#endif + +typedef struct { + kiss_fft_scalar r; + kiss_fft_scalar i; +}kiss_fft_cpx; + +typedef struct { + kiss_twiddle_scalar r; + kiss_twiddle_scalar i; +} KISS_TWIDDLE_CPX_ALIGNED kiss_twiddle_cpx; + +#define MAXFACTORS 8 +/* e.g. an fft of length 128 has 4 factors + as far as kissfft is concerned + 4*4*4*2 + */ + +typedef struct arch_fft_state{ + int is_supported; + void *priv; +} arch_fft_state; + +typedef struct kiss_fft_state{ + int nfft; + opus_val16 scale; +#ifdef FIXED_POINT + int scale_shift; +#endif + int shift; + opus_int16 factors[2*MAXFACTORS]; + const opus_int16 *bitrev; + const kiss_twiddle_cpx *twiddles; + arch_fft_state *arch_fft; +} kiss_fft_state; + +#if defined(HAVE_ARM_NE10) +#include "arm/fft_arm.h" +#endif + +/*typedef struct kiss_fft_state* kiss_fft_cfg;*/ + +/** + * opus_fft_alloc + * + * Initialize a FFT (or IFFT) algorithm's cfg/state buffer. + * + * typical usage: kiss_fft_cfg mycfg=opus_fft_alloc(1024,0,NULL,NULL); + * + * The return value from fft_alloc is a cfg buffer used internally + * by the fft routine or NULL. + * + * If lenmem is NULL, then opus_fft_alloc will allocate a cfg buffer using malloc. + * The returned value should be free()d when done to avoid memory leaks. + * + * The state can be placed in a user supplied buffer 'mem': + * If lenmem is not NULL and mem is not NULL and *lenmem is large enough, + * then the function places the cfg in mem and the size used in *lenmem + * and returns mem. + * + * If lenmem is not NULL and ( mem is NULL or *lenmem is not large enough), + * then the function returns NULL and places the minimum cfg + * buffer size in *lenmem. + * */ + +kiss_fft_state *opus_fft_alloc_twiddles(int nfft,void * mem,size_t * lenmem, const kiss_fft_state *base, int arch); + +kiss_fft_state *opus_fft_alloc(int nfft,void * mem,size_t * lenmem, int arch); + +/** + * opus_fft(cfg,in_out_buf) + * + * Perform an FFT on a complex input buffer. + * for a forward FFT, + * fin should be f[0] , f[1] , ... ,f[nfft-1] + * fout will be F[0] , F[1] , ... ,F[nfft-1] + * Note that each element is complex and can be accessed like + f[k].r and f[k].i + * */ +void opus_fft_c(const kiss_fft_state *cfg,const kiss_fft_cpx *fin,kiss_fft_cpx *fout); +void opus_ifft_c(const kiss_fft_state *cfg,const kiss_fft_cpx *fin,kiss_fft_cpx *fout); + +void opus_fft_impl(const kiss_fft_state *st,kiss_fft_cpx *fout); +void opus_ifft_impl(const kiss_fft_state *st,kiss_fft_cpx *fout); + +void opus_fft_free(const kiss_fft_state *cfg, int arch); + + +void opus_fft_free_arch_c(kiss_fft_state *st); +int opus_fft_alloc_arch_c(kiss_fft_state *st); + +#if !defined(OVERRIDE_OPUS_FFT) +/* Is run-time CPU detection enabled on this platform? */ +#if defined(OPUS_HAVE_RTCD) && (defined(HAVE_ARM_NE10)) + +extern int (*const OPUS_FFT_ALLOC_ARCH_IMPL[OPUS_ARCHMASK+1])( + kiss_fft_state *st); + +#define opus_fft_alloc_arch(_st, arch) \ + ((*OPUS_FFT_ALLOC_ARCH_IMPL[(arch)&OPUS_ARCHMASK])(_st)) + +extern void (*const OPUS_FFT_FREE_ARCH_IMPL[OPUS_ARCHMASK+1])( + kiss_fft_state *st); +#define opus_fft_free_arch(_st, arch) \ + ((*OPUS_FFT_FREE_ARCH_IMPL[(arch)&OPUS_ARCHMASK])(_st)) + +extern void (*const OPUS_FFT[OPUS_ARCHMASK+1])(const kiss_fft_state *cfg, + const kiss_fft_cpx *fin, kiss_fft_cpx *fout); +#define opus_fft(_cfg, _fin, _fout, arch) \ + ((*OPUS_FFT[(arch)&OPUS_ARCHMASK])(_cfg, _fin, _fout)) + +extern void (*const OPUS_IFFT[OPUS_ARCHMASK+1])(const kiss_fft_state *cfg, + const kiss_fft_cpx *fin, kiss_fft_cpx *fout); +#define opus_ifft(_cfg, _fin, _fout, arch) \ + ((*OPUS_IFFT[(arch)&OPUS_ARCHMASK])(_cfg, _fin, _fout)) + +#else /* else for if defined(OPUS_HAVE_RTCD) && (defined(HAVE_ARM_NE10)) */ + +#define opus_fft_alloc_arch(_st, arch) \ + ((void)(arch), opus_fft_alloc_arch_c(_st)) + +#define opus_fft_free_arch(_st, arch) \ + ((void)(arch), opus_fft_free_arch_c(_st)) + +#define opus_fft(_cfg, _fin, _fout, arch) \ + ((void)(arch), opus_fft_c(_cfg, _fin, _fout)) + +#define opus_ifft(_cfg, _fin, _fout, arch) \ + ((void)(arch), opus_ifft_c(_cfg, _fin, _fout)) + +#endif /* end if defined(OPUS_HAVE_RTCD) && (defined(HAVE_ARM_NE10)) */ +#endif /* end if !defined(OVERRIDE_OPUS_FFT) */ + +#ifdef __cplusplus +} +#endif + +#endif diff --git a/media/libopus/celt/laplace.c b/media/libopus/celt/laplace.c new file mode 100644 index 0000000000..a7bca874b6 --- /dev/null +++ b/media/libopus/celt/laplace.c @@ -0,0 +1,134 @@ +/* Copyright (c) 2007 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "laplace.h" +#include "mathops.h" + +/* The minimum probability of an energy delta (out of 32768). */ +#define LAPLACE_LOG_MINP (0) +#define LAPLACE_MINP (1<<LAPLACE_LOG_MINP) +/* The minimum number of guaranteed representable energy deltas (in one + direction). */ +#define LAPLACE_NMIN (16) + +/* When called, decay is positive and at most 11456. */ +static unsigned ec_laplace_get_freq1(unsigned fs0, int decay) +{ + unsigned ft; + ft = 32768 - LAPLACE_MINP*(2*LAPLACE_NMIN) - fs0; + return ft*(opus_int32)(16384-decay)>>15; +} + +void ec_laplace_encode(ec_enc *enc, int *value, unsigned fs, int decay) +{ + unsigned fl; + int val = *value; + fl = 0; + if (val) + { + int s; + int i; + s = -(val<0); + val = (val+s)^s; + fl = fs; + fs = ec_laplace_get_freq1(fs, decay); + /* Search the decaying part of the PDF.*/ + for (i=1; fs > 0 && i < val; i++) + { + fs *= 2; + fl += fs+2*LAPLACE_MINP; + fs = (fs*(opus_int32)decay)>>15; + } + /* Everything beyond that has probability LAPLACE_MINP. */ + if (!fs) + { + int di; + int ndi_max; + ndi_max = (32768-fl+LAPLACE_MINP-1)>>LAPLACE_LOG_MINP; + ndi_max = (ndi_max-s)>>1; + di = IMIN(val - i, ndi_max - 1); + fl += (2*di+1+s)*LAPLACE_MINP; + fs = IMIN(LAPLACE_MINP, 32768-fl); + *value = (i+di+s)^s; + } + else + { + fs += LAPLACE_MINP; + fl += fs&~s; + } + celt_assert(fl+fs<=32768); + celt_assert(fs>0); + } + ec_encode_bin(enc, fl, fl+fs, 15); +} + +int ec_laplace_decode(ec_dec *dec, unsigned fs, int decay) +{ + int val=0; + unsigned fl; + unsigned fm; + fm = ec_decode_bin(dec, 15); + fl = 0; + if (fm >= fs) + { + val++; + fl = fs; + fs = ec_laplace_get_freq1(fs, decay)+LAPLACE_MINP; + /* Search the decaying part of the PDF.*/ + while(fs > LAPLACE_MINP && fm >= fl+2*fs) + { + fs *= 2; + fl += fs; + fs = ((fs-2*LAPLACE_MINP)*(opus_int32)decay)>>15; + fs += LAPLACE_MINP; + val++; + } + /* Everything beyond that has probability LAPLACE_MINP. */ + if (fs <= LAPLACE_MINP) + { + int di; + di = (fm-fl)>>(LAPLACE_LOG_MINP+1); + val += di; + fl += 2*di*LAPLACE_MINP; + } + if (fm < fl+fs) + val = -val; + else + fl += fs; + } + celt_assert(fl<32768); + celt_assert(fs>0); + celt_assert(fl<=fm); + celt_assert(fm<IMIN(fl+fs,32768)); + ec_dec_update(dec, fl, IMIN(fl+fs,32768), 32768); + return val; +} diff --git a/media/libopus/celt/laplace.h b/media/libopus/celt/laplace.h new file mode 100644 index 0000000000..46c14b5da5 --- /dev/null +++ b/media/libopus/celt/laplace.h @@ -0,0 +1,48 @@ +/* Copyright (c) 2007 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#include "entenc.h" +#include "entdec.h" + +/** Encode a value that is assumed to be the realisation of a + Laplace-distributed random process + @param enc Entropy encoder state + @param value Value to encode + @param fs Probability of 0, multiplied by 32768 + @param decay Probability of the value +/- 1, multiplied by 16384 +*/ +void ec_laplace_encode(ec_enc *enc, int *value, unsigned fs, int decay); + +/** Decode a value that is assumed to be the realisation of a + Laplace-distributed random process + @param dec Entropy decoder state + @param fs Probability of 0, multiplied by 32768 + @param decay Probability of the value +/- 1, multiplied by 16384 + @return Value decoded + */ +int ec_laplace_decode(ec_dec *dec, unsigned fs, int decay); diff --git a/media/libopus/celt/mathops.c b/media/libopus/celt/mathops.c new file mode 100644 index 0000000000..6ee9b9e101 --- /dev/null +++ b/media/libopus/celt/mathops.c @@ -0,0 +1,209 @@ +/* Copyright (c) 2002-2008 Jean-Marc Valin + Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/** + @file mathops.h + @brief Various math functions +*/ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "mathops.h" + +/*Compute floor(sqrt(_val)) with exact arithmetic. + _val must be greater than 0. + This has been tested on all possible 32-bit inputs greater than 0.*/ +unsigned isqrt32(opus_uint32 _val){ + unsigned b; + unsigned g; + int bshift; + /*Uses the second method from + http://www.azillionmonkeys.com/qed/sqroot.html + The main idea is to search for the largest binary digit b such that + (g+b)*(g+b) <= _val, and add it to the solution g.*/ + g=0; + bshift=(EC_ILOG(_val)-1)>>1; + b=1U<<bshift; + do{ + opus_uint32 t; + t=(((opus_uint32)g<<1)+b)<<bshift; + if(t<=_val){ + g+=b; + _val-=t; + } + b>>=1; + bshift--; + } + while(bshift>=0); + return g; +} + +#ifdef FIXED_POINT + +opus_val32 frac_div32(opus_val32 a, opus_val32 b) +{ + opus_val16 rcp; + opus_val32 result, rem; + int shift = celt_ilog2(b)-29; + a = VSHR32(a,shift); + b = VSHR32(b,shift); + /* 16-bit reciprocal */ + rcp = ROUND16(celt_rcp(ROUND16(b,16)),3); + result = MULT16_32_Q15(rcp, a); + rem = PSHR32(a,2)-MULT32_32_Q31(result, b); + result = ADD32(result, SHL32(MULT16_32_Q15(rcp, rem),2)); + if (result >= 536870912) /* 2^29 */ + return 2147483647; /* 2^31 - 1 */ + else if (result <= -536870912) /* -2^29 */ + return -2147483647; /* -2^31 */ + else + return SHL32(result, 2); +} + +/** Reciprocal sqrt approximation in the range [0.25,1) (Q16 in, Q14 out) */ +opus_val16 celt_rsqrt_norm(opus_val32 x) +{ + opus_val16 n; + opus_val16 r; + opus_val16 r2; + opus_val16 y; + /* Range of n is [-16384,32767] ([-0.5,1) in Q15). */ + n = x-32768; + /* Get a rough initial guess for the root. + The optimal minimax quadratic approximation (using relative error) is + r = 1.437799046117536+n*(-0.823394375837328+n*0.4096419668459485). + Coefficients here, and the final result r, are Q14.*/ + r = ADD16(23557, MULT16_16_Q15(n, ADD16(-13490, MULT16_16_Q15(n, 6713)))); + /* We want y = x*r*r-1 in Q15, but x is 32-bit Q16 and r is Q14. + We can compute the result from n and r using Q15 multiplies with some + adjustment, carefully done to avoid overflow. + Range of y is [-1564,1594]. */ + r2 = MULT16_16_Q15(r, r); + y = SHL16(SUB16(ADD16(MULT16_16_Q15(r2, n), r2), 16384), 1); + /* Apply a 2nd-order Householder iteration: r += r*y*(y*0.375-0.5). + This yields the Q14 reciprocal square root of the Q16 x, with a maximum + relative error of 1.04956E-4, a (relative) RMSE of 2.80979E-5, and a + peak absolute error of 2.26591/16384. */ + return ADD16(r, MULT16_16_Q15(r, MULT16_16_Q15(y, + SUB16(MULT16_16_Q15(y, 12288), 16384)))); +} + +/** Sqrt approximation (QX input, QX/2 output) */ +opus_val32 celt_sqrt(opus_val32 x) +{ + int k; + opus_val16 n; + opus_val32 rt; + static const opus_val16 C[5] = {23175, 11561, -3011, 1699, -664}; + if (x==0) + return 0; + else if (x>=1073741824) + return 32767; + k = (celt_ilog2(x)>>1)-7; + x = VSHR32(x, 2*k); + n = x-32768; + rt = ADD16(C[0], MULT16_16_Q15(n, ADD16(C[1], MULT16_16_Q15(n, ADD16(C[2], + MULT16_16_Q15(n, ADD16(C[3], MULT16_16_Q15(n, (C[4]))))))))); + rt = VSHR32(rt,7-k); + return rt; +} + +#define L1 32767 +#define L2 -7651 +#define L3 8277 +#define L4 -626 + +static OPUS_INLINE opus_val16 _celt_cos_pi_2(opus_val16 x) +{ + opus_val16 x2; + + x2 = MULT16_16_P15(x,x); + return ADD16(1,MIN16(32766,ADD32(SUB16(L1,x2), MULT16_16_P15(x2, ADD32(L2, MULT16_16_P15(x2, ADD32(L3, MULT16_16_P15(L4, x2 + )))))))); +} + +#undef L1 +#undef L2 +#undef L3 +#undef L4 + +opus_val16 celt_cos_norm(opus_val32 x) +{ + x = x&0x0001ffff; + if (x>SHL32(EXTEND32(1), 16)) + x = SUB32(SHL32(EXTEND32(1), 17),x); + if (x&0x00007fff) + { + if (x<SHL32(EXTEND32(1), 15)) + { + return _celt_cos_pi_2(EXTRACT16(x)); + } else { + return NEG16(_celt_cos_pi_2(EXTRACT16(65536-x))); + } + } else { + if (x&0x0000ffff) + return 0; + else if (x&0x0001ffff) + return -32767; + else + return 32767; + } +} + +/** Reciprocal approximation (Q15 input, Q16 output) */ +opus_val32 celt_rcp(opus_val32 x) +{ + int i; + opus_val16 n; + opus_val16 r; + celt_sig_assert(x>0); + i = celt_ilog2(x); + /* n is Q15 with range [0,1). */ + n = VSHR32(x,i-15)-32768; + /* Start with a linear approximation: + r = 1.8823529411764706-0.9411764705882353*n. + The coefficients and the result are Q14 in the range [15420,30840].*/ + r = ADD16(30840, MULT16_16_Q15(-15420, n)); + /* Perform two Newton iterations: + r -= r*((r*n)-1.Q15) + = r*((r*n)+(r-1.Q15)). */ + r = SUB16(r, MULT16_16_Q15(r, + ADD16(MULT16_16_Q15(r, n), ADD16(r, -32768)))); + /* We subtract an extra 1 in the second iteration to avoid overflow; it also + neatly compensates for truncation error in the rest of the process. */ + r = SUB16(r, ADD16(1, MULT16_16_Q15(r, + ADD16(MULT16_16_Q15(r, n), ADD16(r, -32768))))); + /* r is now the Q15 solution to 2/(n+1), with a maximum relative error + of 7.05346E-5, a (relative) RMSE of 2.14418E-5, and a peak absolute + error of 1.24665/32768. */ + return VSHR32(EXTEND32(r),i-16); +} + +#endif diff --git a/media/libopus/celt/mathops.h b/media/libopus/celt/mathops.h new file mode 100644 index 0000000000..478ac9187c --- /dev/null +++ b/media/libopus/celt/mathops.h @@ -0,0 +1,290 @@ +/* Copyright (c) 2002-2008 Jean-Marc Valin + Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/** + @file mathops.h + @brief Various math functions +*/ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef MATHOPS_H +#define MATHOPS_H + +#include "arch.h" +#include "entcode.h" +#include "os_support.h" + +#define PI 3.141592653f + +/* Multiplies two 16-bit fractional values. Bit-exactness of this macro is important */ +#define FRAC_MUL16(a,b) ((16384+((opus_int32)(opus_int16)(a)*(opus_int16)(b)))>>15) + +unsigned isqrt32(opus_uint32 _val); + +/* CELT doesn't need it for fixed-point, by analysis.c does. */ +#if !defined(FIXED_POINT) || defined(ANALYSIS_C) +#define cA 0.43157974f +#define cB 0.67848403f +#define cC 0.08595542f +#define cE ((float)PI/2) +static OPUS_INLINE float fast_atan2f(float y, float x) { + float x2, y2; + x2 = x*x; + y2 = y*y; + /* For very small values, we don't care about the answer, so + we can just return 0. */ + if (x2 + y2 < 1e-18f) + { + return 0; + } + if(x2<y2){ + float den = (y2 + cB*x2) * (y2 + cC*x2); + return -x*y*(y2 + cA*x2) / den + (y<0 ? -cE : cE); + }else{ + float den = (x2 + cB*y2) * (x2 + cC*y2); + return x*y*(x2 + cA*y2) / den + (y<0 ? -cE : cE) - (x*y<0 ? -cE : cE); + } +} +#undef cA +#undef cB +#undef cC +#undef cE +#endif + + +#ifndef OVERRIDE_CELT_MAXABS16 +static OPUS_INLINE opus_val32 celt_maxabs16(const opus_val16 *x, int len) +{ + int i; + opus_val16 maxval = 0; + opus_val16 minval = 0; + for (i=0;i<len;i++) + { + maxval = MAX16(maxval, x[i]); + minval = MIN16(minval, x[i]); + } + return MAX32(EXTEND32(maxval),-EXTEND32(minval)); +} +#endif + +#ifndef OVERRIDE_CELT_MAXABS32 +#ifdef FIXED_POINT +static OPUS_INLINE opus_val32 celt_maxabs32(const opus_val32 *x, int len) +{ + int i; + opus_val32 maxval = 0; + opus_val32 minval = 0; + for (i=0;i<len;i++) + { + maxval = MAX32(maxval, x[i]); + minval = MIN32(minval, x[i]); + } + return MAX32(maxval, -minval); +} +#else +#define celt_maxabs32(x,len) celt_maxabs16(x,len) +#endif +#endif + + +#ifndef FIXED_POINT + +#define celt_sqrt(x) ((float)sqrt(x)) +#define celt_rsqrt(x) (1.f/celt_sqrt(x)) +#define celt_rsqrt_norm(x) (celt_rsqrt(x)) +#define celt_cos_norm(x) ((float)cos((.5f*PI)*(x))) +#define celt_rcp(x) (1.f/(x)) +#define celt_div(a,b) ((a)/(b)) +#define frac_div32(a,b) ((float)(a)/(b)) + +#ifdef FLOAT_APPROX + +/* Note: This assumes radix-2 floating point with the exponent at bits 23..30 and an offset of 127 + denorm, +/- inf and NaN are *not* handled */ + +/** Base-2 log approximation (log2(x)). */ +static OPUS_INLINE float celt_log2(float x) +{ + int integer; + float frac; + union { + float f; + opus_uint32 i; + } in; + in.f = x; + integer = (in.i>>23)-127; + in.i -= (opus_uint32)integer<<23; + frac = in.f - 1.5f; + frac = -0.41445418f + frac*(0.95909232f + + frac*(-0.33951290f + frac*0.16541097f)); + return 1+integer+frac; +} + +/** Base-2 exponential approximation (2^x). */ +static OPUS_INLINE float celt_exp2(float x) +{ + int integer; + float frac; + union { + float f; + opus_uint32 i; + } res; + integer = (int)floor(x); + if (integer < -50) + return 0; + frac = x-integer; + /* K0 = 1, K1 = log(2), K2 = 3-4*log(2), K3 = 3*log(2) - 2 */ + res.f = 0.99992522f + frac * (0.69583354f + + frac * (0.22606716f + 0.078024523f*frac)); + res.i = (res.i + ((opus_uint32)integer<<23)) & 0x7fffffff; + return res.f; +} + +#else +#define celt_log2(x) ((float)(1.442695040888963387*log(x))) +#define celt_exp2(x) ((float)exp(0.6931471805599453094*(x))) +#endif + +#endif + +#ifdef FIXED_POINT + +#include "os_support.h" + +#ifndef OVERRIDE_CELT_ILOG2 +/** Integer log in base2. Undefined for zero and negative numbers */ +static OPUS_INLINE opus_int16 celt_ilog2(opus_int32 x) +{ + celt_sig_assert(x>0); + return EC_ILOG(x)-1; +} +#endif + + +/** Integer log in base2. Defined for zero, but not for negative numbers */ +static OPUS_INLINE opus_int16 celt_zlog2(opus_val32 x) +{ + return x <= 0 ? 0 : celt_ilog2(x); +} + +opus_val16 celt_rsqrt_norm(opus_val32 x); + +opus_val32 celt_sqrt(opus_val32 x); + +opus_val16 celt_cos_norm(opus_val32 x); + +/** Base-2 logarithm approximation (log2(x)). (Q14 input, Q10 output) */ +static OPUS_INLINE opus_val16 celt_log2(opus_val32 x) +{ + int i; + opus_val16 n, frac; + /* -0.41509302963303146, 0.9609890551383969, -0.31836011537636605, + 0.15530808010959576, -0.08556153059057618 */ + static const opus_val16 C[5] = {-6801+(1<<(13-DB_SHIFT)), 15746, -5217, 2545, -1401}; + if (x==0) + return -32767; + i = celt_ilog2(x); + n = VSHR32(x,i-15)-32768-16384; + frac = ADD16(C[0], MULT16_16_Q15(n, ADD16(C[1], MULT16_16_Q15(n, ADD16(C[2], MULT16_16_Q15(n, ADD16(C[3], MULT16_16_Q15(n, C[4])))))))); + return SHL16(i-13,DB_SHIFT)+SHR16(frac,14-DB_SHIFT); +} + +/* + K0 = 1 + K1 = log(2) + K2 = 3-4*log(2) + K3 = 3*log(2) - 2 +*/ +#define D0 16383 +#define D1 22804 +#define D2 14819 +#define D3 10204 + +static OPUS_INLINE opus_val32 celt_exp2_frac(opus_val16 x) +{ + opus_val16 frac; + frac = SHL16(x, 4); + return ADD16(D0, MULT16_16_Q15(frac, ADD16(D1, MULT16_16_Q15(frac, ADD16(D2 , MULT16_16_Q15(D3,frac)))))); +} +/** Base-2 exponential approximation (2^x). (Q10 input, Q16 output) */ +static OPUS_INLINE opus_val32 celt_exp2(opus_val16 x) +{ + int integer; + opus_val16 frac; + integer = SHR16(x,10); + if (integer>14) + return 0x7f000000; + else if (integer < -15) + return 0; + frac = celt_exp2_frac(x-SHL16(integer,10)); + return VSHR32(EXTEND32(frac), -integer-2); +} + +opus_val32 celt_rcp(opus_val32 x); + +#define celt_div(a,b) MULT32_32_Q31((opus_val32)(a),celt_rcp(b)) + +opus_val32 frac_div32(opus_val32 a, opus_val32 b); + +#define M1 32767 +#define M2 -21 +#define M3 -11943 +#define M4 4936 + +/* Atan approximation using a 4th order polynomial. Input is in Q15 format + and normalized by pi/4. Output is in Q15 format */ +static OPUS_INLINE opus_val16 celt_atan01(opus_val16 x) +{ + return MULT16_16_P15(x, ADD32(M1, MULT16_16_P15(x, ADD32(M2, MULT16_16_P15(x, ADD32(M3, MULT16_16_P15(M4, x))))))); +} + +#undef M1 +#undef M2 +#undef M3 +#undef M4 + +/* atan2() approximation valid for positive input values */ +static OPUS_INLINE opus_val16 celt_atan2p(opus_val16 y, opus_val16 x) +{ + if (y < x) + { + opus_val32 arg; + arg = celt_div(SHL32(EXTEND32(y),15),x); + if (arg >= 32767) + arg = 32767; + return SHR16(celt_atan01(EXTRACT16(arg)),1); + } else { + opus_val32 arg; + arg = celt_div(SHL32(EXTEND32(x),15),y); + if (arg >= 32767) + arg = 32767; + return 25736-SHR16(celt_atan01(EXTRACT16(arg)),1); + } +} + +#endif /* FIXED_POINT */ +#endif /* MATHOPS_H */ diff --git a/media/libopus/celt/mdct.c b/media/libopus/celt/mdct.c new file mode 100644 index 0000000000..5c6dab5b75 --- /dev/null +++ b/media/libopus/celt/mdct.c @@ -0,0 +1,343 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2008 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +/* This is a simple MDCT implementation that uses a N/4 complex FFT + to do most of the work. It should be relatively straightforward to + plug in pretty much and FFT here. + + This replaces the Vorbis FFT (and uses the exact same API), which + was a bit too messy and that was ending up duplicating code + (might as well use the same FFT everywhere). + + The algorithm is similar to (and inspired from) Fabrice Bellard's + MDCT implementation in FFMPEG, but has differences in signs, ordering + and scaling in many places. +*/ + +#ifndef SKIP_CONFIG_H +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif +#endif + +#include "mdct.h" +#include "kiss_fft.h" +#include "_kiss_fft_guts.h" +#include <math.h> +#include "os_support.h" +#include "mathops.h" +#include "stack_alloc.h" + +#if defined(MIPSr1_ASM) +#include "mips/mdct_mipsr1.h" +#endif + + +#ifdef CUSTOM_MODES + +int clt_mdct_init(mdct_lookup *l,int N, int maxshift, int arch) +{ + int i; + kiss_twiddle_scalar *trig; + int shift; + int N2=N>>1; + l->n = N; + l->maxshift = maxshift; + for (i=0;i<=maxshift;i++) + { + if (i==0) + l->kfft[i] = opus_fft_alloc(N>>2>>i, 0, 0, arch); + else + l->kfft[i] = opus_fft_alloc_twiddles(N>>2>>i, 0, 0, l->kfft[0], arch); +#ifndef ENABLE_TI_DSPLIB55 + if (l->kfft[i]==NULL) + return 0; +#endif + } + l->trig = trig = (kiss_twiddle_scalar*)opus_alloc((N-(N2>>maxshift))*sizeof(kiss_twiddle_scalar)); + if (l->trig==NULL) + return 0; + for (shift=0;shift<=maxshift;shift++) + { + /* We have enough points that sine isn't necessary */ +#if defined(FIXED_POINT) +#if 1 + for (i=0;i<N2;i++) + trig[i] = TRIG_UPSCALE*celt_cos_norm(DIV32(ADD32(SHL32(EXTEND32(i),17),N2+16384),N)); +#else + for (i=0;i<N2;i++) + trig[i] = (kiss_twiddle_scalar)MAX32(-32767,MIN32(32767,floor(.5+32768*cos(2*M_PI*(i+.125)/N)))); +#endif +#else + for (i=0;i<N2;i++) + trig[i] = (kiss_twiddle_scalar)cos(2*PI*(i+.125)/N); +#endif + trig += N2; + N2 >>= 1; + N >>= 1; + } + return 1; +} + +void clt_mdct_clear(mdct_lookup *l, int arch) +{ + int i; + for (i=0;i<=l->maxshift;i++) + opus_fft_free(l->kfft[i], arch); + opus_free((kiss_twiddle_scalar*)l->trig); +} + +#endif /* CUSTOM_MODES */ + +/* Forward MDCT trashes the input array */ +#ifndef OVERRIDE_clt_mdct_forward +void clt_mdct_forward_c(const mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scalar * OPUS_RESTRICT out, + const opus_val16 *window, int overlap, int shift, int stride, int arch) +{ + int i; + int N, N2, N4; + VARDECL(kiss_fft_scalar, f); + VARDECL(kiss_fft_cpx, f2); + const kiss_fft_state *st = l->kfft[shift]; + const kiss_twiddle_scalar *trig; + opus_val16 scale; +#ifdef FIXED_POINT + /* Allows us to scale with MULT16_32_Q16(), which is faster than + MULT16_32_Q15() on ARM. */ + int scale_shift = st->scale_shift-1; +#endif + SAVE_STACK; + (void)arch; + scale = st->scale; + + N = l->n; + trig = l->trig; + for (i=0;i<shift;i++) + { + N >>= 1; + trig += N; + } + N2 = N>>1; + N4 = N>>2; + + ALLOC(f, N2, kiss_fft_scalar); + ALLOC(f2, N4, kiss_fft_cpx); + + /* Consider the input to be composed of four blocks: [a, b, c, d] */ + /* Window, shuffle, fold */ + { + /* Temp pointers to make it really clear to the compiler what we're doing */ + const kiss_fft_scalar * OPUS_RESTRICT xp1 = in+(overlap>>1); + const kiss_fft_scalar * OPUS_RESTRICT xp2 = in+N2-1+(overlap>>1); + kiss_fft_scalar * OPUS_RESTRICT yp = f; + const opus_val16 * OPUS_RESTRICT wp1 = window+(overlap>>1); + const opus_val16 * OPUS_RESTRICT wp2 = window+(overlap>>1)-1; + for(i=0;i<((overlap+3)>>2);i++) + { + /* Real part arranged as -d-cR, Imag part arranged as -b+aR*/ + *yp++ = MULT16_32_Q15(*wp2, xp1[N2]) + MULT16_32_Q15(*wp1,*xp2); + *yp++ = MULT16_32_Q15(*wp1, *xp1) - MULT16_32_Q15(*wp2, xp2[-N2]); + xp1+=2; + xp2-=2; + wp1+=2; + wp2-=2; + } + wp1 = window; + wp2 = window+overlap-1; + for(;i<N4-((overlap+3)>>2);i++) + { + /* Real part arranged as a-bR, Imag part arranged as -c-dR */ + *yp++ = *xp2; + *yp++ = *xp1; + xp1+=2; + xp2-=2; + } + for(;i<N4;i++) + { + /* Real part arranged as a-bR, Imag part arranged as -c-dR */ + *yp++ = -MULT16_32_Q15(*wp1, xp1[-N2]) + MULT16_32_Q15(*wp2, *xp2); + *yp++ = MULT16_32_Q15(*wp2, *xp1) + MULT16_32_Q15(*wp1, xp2[N2]); + xp1+=2; + xp2-=2; + wp1+=2; + wp2-=2; + } + } + /* Pre-rotation */ + { + kiss_fft_scalar * OPUS_RESTRICT yp = f; + const kiss_twiddle_scalar *t = &trig[0]; + for(i=0;i<N4;i++) + { + kiss_fft_cpx yc; + kiss_twiddle_scalar t0, t1; + kiss_fft_scalar re, im, yr, yi; + t0 = t[i]; + t1 = t[N4+i]; + re = *yp++; + im = *yp++; + yr = S_MUL(re,t0) - S_MUL(im,t1); + yi = S_MUL(im,t0) + S_MUL(re,t1); + yc.r = yr; + yc.i = yi; + yc.r = PSHR32(MULT16_32_Q16(scale, yc.r), scale_shift); + yc.i = PSHR32(MULT16_32_Q16(scale, yc.i), scale_shift); + f2[st->bitrev[i]] = yc; + } + } + + /* N/4 complex FFT, does not downscale anymore */ + opus_fft_impl(st, f2); + + /* Post-rotate */ + { + /* Temp pointers to make it really clear to the compiler what we're doing */ + const kiss_fft_cpx * OPUS_RESTRICT fp = f2; + kiss_fft_scalar * OPUS_RESTRICT yp1 = out; + kiss_fft_scalar * OPUS_RESTRICT yp2 = out+stride*(N2-1); + const kiss_twiddle_scalar *t = &trig[0]; + /* Temp pointers to make it really clear to the compiler what we're doing */ + for(i=0;i<N4;i++) + { + kiss_fft_scalar yr, yi; + yr = S_MUL(fp->i,t[N4+i]) - S_MUL(fp->r,t[i]); + yi = S_MUL(fp->r,t[N4+i]) + S_MUL(fp->i,t[i]); + *yp1 = yr; + *yp2 = yi; + fp++; + yp1 += 2*stride; + yp2 -= 2*stride; + } + } + RESTORE_STACK; +} +#endif /* OVERRIDE_clt_mdct_forward */ + +#ifndef OVERRIDE_clt_mdct_backward +void clt_mdct_backward_c(const mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scalar * OPUS_RESTRICT out, + const opus_val16 * OPUS_RESTRICT window, int overlap, int shift, int stride, int arch) +{ + int i; + int N, N2, N4; + const kiss_twiddle_scalar *trig; + (void) arch; + + N = l->n; + trig = l->trig; + for (i=0;i<shift;i++) + { + N >>= 1; + trig += N; + } + N2 = N>>1; + N4 = N>>2; + + /* Pre-rotate */ + { + /* Temp pointers to make it really clear to the compiler what we're doing */ + const kiss_fft_scalar * OPUS_RESTRICT xp1 = in; + const kiss_fft_scalar * OPUS_RESTRICT xp2 = in+stride*(N2-1); + kiss_fft_scalar * OPUS_RESTRICT yp = out+(overlap>>1); + const kiss_twiddle_scalar * OPUS_RESTRICT t = &trig[0]; + const opus_int16 * OPUS_RESTRICT bitrev = l->kfft[shift]->bitrev; + for(i=0;i<N4;i++) + { + int rev; + kiss_fft_scalar yr, yi; + rev = *bitrev++; + yr = ADD32_ovflw(S_MUL(*xp2, t[i]), S_MUL(*xp1, t[N4+i])); + yi = SUB32_ovflw(S_MUL(*xp1, t[i]), S_MUL(*xp2, t[N4+i])); + /* We swap real and imag because we use an FFT instead of an IFFT. */ + yp[2*rev+1] = yr; + yp[2*rev] = yi; + /* Storing the pre-rotation directly in the bitrev order. */ + xp1+=2*stride; + xp2-=2*stride; + } + } + + opus_fft_impl(l->kfft[shift], (kiss_fft_cpx*)(out+(overlap>>1))); + + /* Post-rotate and de-shuffle from both ends of the buffer at once to make + it in-place. */ + { + kiss_fft_scalar * yp0 = out+(overlap>>1); + kiss_fft_scalar * yp1 = out+(overlap>>1)+N2-2; + const kiss_twiddle_scalar *t = &trig[0]; + /* Loop to (N4+1)>>1 to handle odd N4. When N4 is odd, the + middle pair will be computed twice. */ + for(i=0;i<(N4+1)>>1;i++) + { + kiss_fft_scalar re, im, yr, yi; + kiss_twiddle_scalar t0, t1; + /* We swap real and imag because we're using an FFT instead of an IFFT. */ + re = yp0[1]; + im = yp0[0]; + t0 = t[i]; + t1 = t[N4+i]; + /* We'd scale up by 2 here, but instead it's done when mixing the windows */ + yr = ADD32_ovflw(S_MUL(re,t0), S_MUL(im,t1)); + yi = SUB32_ovflw(S_MUL(re,t1), S_MUL(im,t0)); + /* We swap real and imag because we're using an FFT instead of an IFFT. */ + re = yp1[1]; + im = yp1[0]; + yp0[0] = yr; + yp1[1] = yi; + + t0 = t[(N4-i-1)]; + t1 = t[(N2-i-1)]; + /* We'd scale up by 2 here, but instead it's done when mixing the windows */ + yr = ADD32_ovflw(S_MUL(re,t0), S_MUL(im,t1)); + yi = SUB32_ovflw(S_MUL(re,t1), S_MUL(im,t0)); + yp1[0] = yr; + yp0[1] = yi; + yp0 += 2; + yp1 -= 2; + } + } + + /* Mirror on both sides for TDAC */ + { + kiss_fft_scalar * OPUS_RESTRICT xp1 = out+overlap-1; + kiss_fft_scalar * OPUS_RESTRICT yp1 = out; + const opus_val16 * OPUS_RESTRICT wp1 = window; + const opus_val16 * OPUS_RESTRICT wp2 = window+overlap-1; + + for(i = 0; i < overlap/2; i++) + { + kiss_fft_scalar x1, x2; + x1 = *xp1; + x2 = *yp1; + *yp1++ = SUB32_ovflw(MULT16_32_Q15(*wp2, x2), MULT16_32_Q15(*wp1, x1)); + *xp1-- = ADD32_ovflw(MULT16_32_Q15(*wp1, x2), MULT16_32_Q15(*wp2, x1)); + wp1++; + wp2--; + } + } +} +#endif /* OVERRIDE_clt_mdct_backward */ diff --git a/media/libopus/celt/mdct.h b/media/libopus/celt/mdct.h new file mode 100644 index 0000000000..160ae4e0f3 --- /dev/null +++ b/media/libopus/celt/mdct.h @@ -0,0 +1,112 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2008 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +/* This is a simple MDCT implementation that uses a N/4 complex FFT + to do most of the work. It should be relatively straightforward to + plug in pretty much and FFT here. + + This replaces the Vorbis FFT (and uses the exact same API), which + was a bit too messy and that was ending up duplicating code + (might as well use the same FFT everywhere). + + The algorithm is similar to (and inspired from) Fabrice Bellard's + MDCT implementation in FFMPEG, but has differences in signs, ordering + and scaling in many places. +*/ + +#ifndef MDCT_H +#define MDCT_H + +#include "opus_defines.h" +#include "kiss_fft.h" +#include "arch.h" + +typedef struct { + int n; + int maxshift; + const kiss_fft_state *kfft[4]; + const kiss_twiddle_scalar * OPUS_RESTRICT trig; +} mdct_lookup; + +#if defined(HAVE_ARM_NE10) +#include "arm/mdct_arm.h" +#endif + + +int clt_mdct_init(mdct_lookup *l,int N, int maxshift, int arch); +void clt_mdct_clear(mdct_lookup *l, int arch); + +/** Compute a forward MDCT and scale by 4/N, trashes the input array */ +void clt_mdct_forward_c(const mdct_lookup *l, kiss_fft_scalar *in, + kiss_fft_scalar * OPUS_RESTRICT out, + const opus_val16 *window, int overlap, + int shift, int stride, int arch); + +/** Compute a backward MDCT (no scaling) and performs weighted overlap-add + (scales implicitly by 1/2) */ +void clt_mdct_backward_c(const mdct_lookup *l, kiss_fft_scalar *in, + kiss_fft_scalar * OPUS_RESTRICT out, + const opus_val16 * OPUS_RESTRICT window, + int overlap, int shift, int stride, int arch); + +#if !defined(OVERRIDE_OPUS_MDCT) +/* Is run-time CPU detection enabled on this platform? */ +#if defined(OPUS_HAVE_RTCD) && defined(HAVE_ARM_NE10) + +extern void (*const CLT_MDCT_FORWARD_IMPL[OPUS_ARCHMASK+1])( + const mdct_lookup *l, kiss_fft_scalar *in, + kiss_fft_scalar * OPUS_RESTRICT out, const opus_val16 *window, + int overlap, int shift, int stride, int arch); + +#define clt_mdct_forward(_l, _in, _out, _window, _overlap, _shift, _stride, _arch) \ + ((*CLT_MDCT_FORWARD_IMPL[(arch)&OPUS_ARCHMASK])(_l, _in, _out, \ + _window, _overlap, _shift, \ + _stride, _arch)) + +extern void (*const CLT_MDCT_BACKWARD_IMPL[OPUS_ARCHMASK+1])( + const mdct_lookup *l, kiss_fft_scalar *in, + kiss_fft_scalar * OPUS_RESTRICT out, const opus_val16 *window, + int overlap, int shift, int stride, int arch); + +#define clt_mdct_backward(_l, _in, _out, _window, _overlap, _shift, _stride, _arch) \ + (*CLT_MDCT_BACKWARD_IMPL[(arch)&OPUS_ARCHMASK])(_l, _in, _out, \ + _window, _overlap, _shift, \ + _stride, _arch) + +#else /* if defined(OPUS_HAVE_RTCD) && defined(HAVE_ARM_NE10) */ + +#define clt_mdct_forward(_l, _in, _out, _window, _overlap, _shift, _stride, _arch) \ + clt_mdct_forward_c(_l, _in, _out, _window, _overlap, _shift, _stride, _arch) + +#define clt_mdct_backward(_l, _in, _out, _window, _overlap, _shift, _stride, _arch) \ + clt_mdct_backward_c(_l, _in, _out, _window, _overlap, _shift, _stride, _arch) + +#endif /* end if defined(OPUS_HAVE_RTCD) && defined(HAVE_ARM_NE10) && !defined(FIXED_POINT) */ +#endif /* end if !defined(OVERRIDE_OPUS_MDCT) */ + +#endif diff --git a/media/libopus/celt/mfrngcod.h b/media/libopus/celt/mfrngcod.h new file mode 100644 index 0000000000..809152a59a --- /dev/null +++ b/media/libopus/celt/mfrngcod.h @@ -0,0 +1,48 @@ +/* Copyright (c) 2001-2008 Timothy B. Terriberry + Copyright (c) 2008-2009 Xiph.Org Foundation */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#if !defined(_mfrngcode_H) +# define _mfrngcode_H (1) +# include "entcode.h" + +/*Constants used by the entropy encoder/decoder.*/ + +/*The number of bits to output at a time.*/ +# define EC_SYM_BITS (8) +/*The total number of bits in each of the state registers.*/ +# define EC_CODE_BITS (32) +/*The maximum symbol value.*/ +# define EC_SYM_MAX ((1U<<EC_SYM_BITS)-1) +/*Bits to shift by to move a symbol into the high-order position.*/ +# define EC_CODE_SHIFT (EC_CODE_BITS-EC_SYM_BITS-1) +/*Carry bit of the high-order range symbol.*/ +# define EC_CODE_TOP (((opus_uint32)1U)<<(EC_CODE_BITS-1)) +/*Low-order bit of the high-order range symbol.*/ +# define EC_CODE_BOT (EC_CODE_TOP>>EC_SYM_BITS) +/*The number of bits available for the last, partial symbol in the code field.*/ +# define EC_CODE_EXTRA ((EC_CODE_BITS-2)%EC_SYM_BITS+1) +#endif diff --git a/media/libopus/celt/mips/celt_mipsr1.h b/media/libopus/celt/mips/celt_mipsr1.h new file mode 100644 index 0000000000..c332fe0471 --- /dev/null +++ b/media/libopus/celt/mips/celt_mipsr1.h @@ -0,0 +1,152 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2010 Xiph.Org Foundation + Copyright (c) 2008 Gregory Maxwell + Written by Jean-Marc Valin and Gregory Maxwell */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef __CELT_MIPSR1_H__ +#define __CELT_MIPSR1_H__ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#define CELT_C + +#include "os_support.h" +#include "mdct.h" +#include <math.h> +#include "celt.h" +#include "pitch.h" +#include "bands.h" +#include "modes.h" +#include "entcode.h" +#include "quant_bands.h" +#include "rate.h" +#include "stack_alloc.h" +#include "mathops.h" +#include "float_cast.h" +#include <stdarg.h> +#include "celt_lpc.h" +#include "vq.h" + +#define OVERRIDE_COMB_FILTER_CONST +#define OVERRIDE_comb_filter +void comb_filter(opus_val32 *y, opus_val32 *x, int T0, int T1, int N, + opus_val16 g0, opus_val16 g1, int tapset0, int tapset1, + const opus_val16 *window, int overlap, int arch) +{ + int i; + opus_val32 x0, x1, x2, x3, x4; + + (void)arch; + + /* printf ("%d %d %f %f\n", T0, T1, g0, g1); */ + opus_val16 g00, g01, g02, g10, g11, g12; + static const opus_val16 gains[3][3] = { + {QCONST16(0.3066406250f, 15), QCONST16(0.2170410156f, 15), QCONST16(0.1296386719f, 15)}, + {QCONST16(0.4638671875f, 15), QCONST16(0.2680664062f, 15), QCONST16(0.f, 15)}, + {QCONST16(0.7998046875f, 15), QCONST16(0.1000976562f, 15), QCONST16(0.f, 15)}}; + + if (g0==0 && g1==0) + { + /* OPT: Happens to work without the OPUS_MOVE(), but only because the current encoder already copies x to y */ + if (x!=y) + OPUS_MOVE(y, x, N); + return; + } + + g00 = MULT16_16_P15(g0, gains[tapset0][0]); + g01 = MULT16_16_P15(g0, gains[tapset0][1]); + g02 = MULT16_16_P15(g0, gains[tapset0][2]); + g10 = MULT16_16_P15(g1, gains[tapset1][0]); + g11 = MULT16_16_P15(g1, gains[tapset1][1]); + g12 = MULT16_16_P15(g1, gains[tapset1][2]); + x1 = x[-T1+1]; + x2 = x[-T1 ]; + x3 = x[-T1-1]; + x4 = x[-T1-2]; + /* If the filter didn't change, we don't need the overlap */ + if (g0==g1 && T0==T1 && tapset0==tapset1) + overlap=0; + + for (i=0;i<overlap;i++) + { + opus_val16 f; + opus_val32 res; + f = MULT16_16_Q15(window[i],window[i]); + x0= x[i-T1+2]; + + asm volatile("MULT $ac1, %0, %1" : : "r" ((int)MULT16_16_Q15((Q15ONE-f),g00)), "r" ((int)x[i-T0])); + + asm volatile("MADD $ac1, %0, %1" : : "r" ((int)MULT16_16_Q15((Q15ONE-f),g01)), "r" ((int)ADD32(x[i-T0-1],x[i-T0+1]))); + asm volatile("MADD $ac1, %0, %1" : : "r" ((int)MULT16_16_Q15((Q15ONE-f),g02)), "r" ((int)ADD32(x[i-T0-2],x[i-T0+2]))); + asm volatile("MADD $ac1, %0, %1" : : "r" ((int)MULT16_16_Q15(f,g10)), "r" ((int)x2)); + asm volatile("MADD $ac1, %0, %1" : : "r" ((int)MULT16_16_Q15(f,g11)), "r" ((int)ADD32(x3,x1))); + asm volatile("MADD $ac1, %0, %1" : : "r" ((int)MULT16_16_Q15(f,g12)), "r" ((int)ADD32(x4,x0))); + + asm volatile("EXTR.W %0,$ac1, %1" : "=r" (res): "i" (15)); + + y[i] = x[i] + res; + + x4=x3; + x3=x2; + x2=x1; + x1=x0; + } + + x4 = x[i-T1-2]; + x3 = x[i-T1-1]; + x2 = x[i-T1]; + x1 = x[i-T1+1]; + + if (g1==0) + { + /* OPT: Happens to work without the OPUS_MOVE(), but only because the current encoder already copies x to y */ + if (x!=y) + OPUS_MOVE(y+overlap, x+overlap, N-overlap); + return; + } + + for (i=overlap;i<N;i++) + { + opus_val32 res; + x0=x[i-T1+2]; + + asm volatile("MULT $ac1, %0, %1" : : "r" ((int)g10), "r" ((int)x2)); + + asm volatile("MADD $ac1, %0, %1" : : "r" ((int)g11), "r" ((int)ADD32(x3,x1))); + asm volatile("MADD $ac1, %0, %1" : : "r" ((int)g12), "r" ((int)ADD32(x4,x0))); + asm volatile("EXTR.W %0,$ac1, %1" : "=r" (res): "i" (15)); + y[i] = x[i] + res; + x4=x3; + x3=x2; + x2=x1; + x1=x0; + } +} + +#endif /* __CELT_MIPSR1_H__ */ diff --git a/media/libopus/celt/mips/fixed_generic_mipsr1.h b/media/libopus/celt/mips/fixed_generic_mipsr1.h new file mode 100644 index 0000000000..4a05efbf85 --- /dev/null +++ b/media/libopus/celt/mips/fixed_generic_mipsr1.h @@ -0,0 +1,126 @@ +/* Copyright (C) 2007-2009 Xiph.Org Foundation + Copyright (C) 2003-2008 Jean-Marc Valin + Copyright (C) 2007-2008 CSIRO */ +/** + @file fixed_generic.h + @brief Generic fixed-point operations +*/ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef CELT_FIXED_GENERIC_MIPSR1_H +#define CELT_FIXED_GENERIC_MIPSR1_H + +#undef MULT16_32_Q15_ADD +static inline int MULT16_32_Q15_ADD(int a, int b, int c, int d) { + int m; + asm volatile("MULT $ac1, %0, %1" : : "r" ((int)a), "r" ((int)b)); + asm volatile("madd $ac1, %0, %1" : : "r" ((int)c), "r" ((int)d)); + asm volatile("EXTR.W %0,$ac1, %1" : "=r" (m): "i" (15)); + return m; +} + +#undef MULT16_32_Q15_SUB +static inline int MULT16_32_Q15_SUB(int a, int b, int c, int d) { + int m; + asm volatile("MULT $ac1, %0, %1" : : "r" ((int)a), "r" ((int)b)); + asm volatile("msub $ac1, %0, %1" : : "r" ((int)c), "r" ((int)d)); + asm volatile("EXTR.W %0,$ac1, %1" : "=r" (m): "i" (15)); + return m; +} + +#undef MULT16_16_Q15_ADD +static inline int MULT16_16_Q15_ADD(int a, int b, int c, int d) { + int m; + asm volatile("MULT $ac1, %0, %1" : : "r" ((int)a), "r" ((int)b)); + asm volatile("madd $ac1, %0, %1" : : "r" ((int)c), "r" ((int)d)); + asm volatile("EXTR.W %0,$ac1, %1" : "=r" (m): "i" (15)); + return m; +} + +#undef MULT16_16_Q15_SUB +static inline int MULT16_16_Q15_SUB(int a, int b, int c, int d) { + int m; + asm volatile("MULT $ac1, %0, %1" : : "r" ((int)a), "r" ((int)b)); + asm volatile("msub $ac1, %0, %1" : : "r" ((int)c), "r" ((int)d)); + asm volatile("EXTR.W %0,$ac1, %1" : "=r" (m): "i" (15)); + return m; +} + + +#undef MULT16_32_Q16 +static inline int MULT16_32_Q16(int a, int b) +{ + int c; + asm volatile("MULT $ac1,%0, %1" : : "r" (a), "r" (b)); + asm volatile("EXTR.W %0,$ac1, %1" : "=r" (c): "i" (16)); + return c; +} + +#undef MULT16_32_P16 +static inline int MULT16_32_P16(int a, int b) +{ + int c; + asm volatile("MULT $ac1, %0, %1" : : "r" (a), "r" (b)); + asm volatile("EXTR_R.W %0,$ac1, %1" : "=r" (c): "i" (16)); + return c; +} + +#undef MULT16_32_Q15 +static inline int MULT16_32_Q15(int a, int b) +{ + int c; + asm volatile("MULT $ac1, %0, %1" : : "r" (a), "r" (b)); + asm volatile("EXTR.W %0,$ac1, %1" : "=r" (c): "i" (15)); + return c; +} + +#undef MULT32_32_Q31 +static inline int MULT32_32_Q31(int a, int b) +{ + int r; + asm volatile("MULT $ac1, %0, %1" : : "r" (a), "r" (b)); + asm volatile("EXTR.W %0,$ac1, %1" : "=r" (r): "i" (31)); + return r; +} + +#undef PSHR32 +static inline int PSHR32(int a, int shift) +{ + int r; + asm volatile ("SHRAV_R.W %0, %1, %2" :"=r" (r): "r" (a), "r" (shift)); + return r; +} + +#undef MULT16_16_P15 +static inline int MULT16_16_P15(int a, int b) +{ + int r; + asm volatile ("mul %0, %1, %2" :"=r" (r): "r" (a), "r" (b)); + asm volatile ("SHRA_R.W %0, %1, %2" : "+r" (r): "0" (r), "i"(15)); + return r; +} + +#endif /* CELT_FIXED_GENERIC_MIPSR1_H */ diff --git a/media/libopus/celt/mips/kiss_fft_mipsr1.h b/media/libopus/celt/mips/kiss_fft_mipsr1.h new file mode 100644 index 0000000000..400ca4de9c --- /dev/null +++ b/media/libopus/celt/mips/kiss_fft_mipsr1.h @@ -0,0 +1,167 @@ +/*Copyright (c) 2013, Xiph.Org Foundation and contributors. + + All rights reserved. + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions are met: + + * Redistributions of source code must retain the above copyright notice, + this list of conditions and the following disclaimer. + * Redistributions in binary form must reproduce the above copyright notice, + this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" + AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE + LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR + CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF + SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS + INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN + CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) + ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE + POSSIBILITY OF SUCH DAMAGE.*/ + +#ifndef KISS_FFT_MIPSR1_H +#define KISS_FFT_MIPSR1_H + +#if !defined(KISS_FFT_GUTS_H) +#error "This file should only be included from _kiss_fft_guts.h" +#endif + +#ifdef FIXED_POINT + +#define S_MUL_ADD(a, b, c, d) (S_MUL(a,b)+S_MUL(c,d)) +#define S_MUL_SUB(a, b, c, d) (S_MUL(a,b)-S_MUL(c,d)) + +#undef S_MUL_ADD +static inline int S_MUL_ADD(int a, int b, int c, int d) { + int m; + asm volatile("MULT $ac1, %0, %1" : : "r" ((int)a), "r" ((int)b)); + asm volatile("madd $ac1, %0, %1" : : "r" ((int)c), "r" ((int)d)); + asm volatile("EXTR.W %0,$ac1, %1" : "=r" (m): "i" (15)); + return m; +} + +#undef S_MUL_SUB +static inline int S_MUL_SUB(int a, int b, int c, int d) { + int m; + asm volatile("MULT $ac1, %0, %1" : : "r" ((int)a), "r" ((int)b)); + asm volatile("msub $ac1, %0, %1" : : "r" ((int)c), "r" ((int)d)); + asm volatile("EXTR.W %0,$ac1, %1" : "=r" (m): "i" (15)); + return m; +} + +#undef C_MUL +# define C_MUL(m,a,b) (m=C_MUL_fun(a,b)) +static inline kiss_fft_cpx C_MUL_fun(kiss_fft_cpx a, kiss_twiddle_cpx b) { + kiss_fft_cpx m; + + asm volatile("MULT $ac1, %0, %1" : : "r" ((int)a.r), "r" ((int)b.r)); + asm volatile("msub $ac1, %0, %1" : : "r" ((int)a.i), "r" ((int)b.i)); + asm volatile("EXTR.W %0,$ac1, %1" : "=r" (m.r): "i" (15)); + asm volatile("MULT $ac1, %0, %1" : : "r" ((int)a.r), "r" ((int)b.i)); + asm volatile("madd $ac1, %0, %1" : : "r" ((int)a.i), "r" ((int)b.r)); + asm volatile("EXTR.W %0,$ac1, %1" : "=r" (m.i): "i" (15)); + + return m; +} +#undef C_MULC +# define C_MULC(m,a,b) (m=C_MULC_fun(a,b)) +static inline kiss_fft_cpx C_MULC_fun(kiss_fft_cpx a, kiss_twiddle_cpx b) { + kiss_fft_cpx m; + + asm volatile("MULT $ac1, %0, %1" : : "r" ((int)a.r), "r" ((int)b.r)); + asm volatile("madd $ac1, %0, %1" : : "r" ((int)a.i), "r" ((int)b.i)); + asm volatile("EXTR.W %0,$ac1, %1" : "=r" (m.r): "i" (15)); + asm volatile("MULT $ac1, %0, %1" : : "r" ((int)a.i), "r" ((int)b.r)); + asm volatile("msub $ac1, %0, %1" : : "r" ((int)a.r), "r" ((int)b.i)); + asm volatile("EXTR.W %0,$ac1, %1" : "=r" (m.i): "i" (15)); + + return m; +} + +#endif /* FIXED_POINT */ + +#define OVERRIDE_kf_bfly5 +static void kf_bfly5( + kiss_fft_cpx * Fout, + const size_t fstride, + const kiss_fft_state *st, + int m, + int N, + int mm + ) +{ + kiss_fft_cpx *Fout0,*Fout1,*Fout2,*Fout3,*Fout4; + int i, u; + kiss_fft_cpx scratch[13]; + + const kiss_twiddle_cpx *tw; + kiss_twiddle_cpx ya,yb; + kiss_fft_cpx * Fout_beg = Fout; + +#ifdef FIXED_POINT + ya.r = 10126; + ya.i = -31164; + yb.r = -26510; + yb.i = -19261; +#else + ya = st->twiddles[fstride*m]; + yb = st->twiddles[fstride*2*m]; +#endif + + tw=st->twiddles; + + for (i=0;i<N;i++) + { + Fout = Fout_beg + i*mm; + Fout0=Fout; + Fout1=Fout0+m; + Fout2=Fout0+2*m; + Fout3=Fout0+3*m; + Fout4=Fout0+4*m; + + /* For non-custom modes, m is guaranteed to be a multiple of 4. */ + for ( u=0; u<m; ++u ) { + scratch[0] = *Fout0; + + + C_MUL(scratch[1] ,*Fout1, tw[u*fstride]); + C_MUL(scratch[2] ,*Fout2, tw[2*u*fstride]); + C_MUL(scratch[3] ,*Fout3, tw[3*u*fstride]); + C_MUL(scratch[4] ,*Fout4, tw[4*u*fstride]); + + C_ADD( scratch[7],scratch[1],scratch[4]); + C_SUB( scratch[10],scratch[1],scratch[4]); + C_ADD( scratch[8],scratch[2],scratch[3]); + C_SUB( scratch[9],scratch[2],scratch[3]); + + Fout0->r += scratch[7].r + scratch[8].r; + Fout0->i += scratch[7].i + scratch[8].i; + scratch[5].r = scratch[0].r + S_MUL_ADD(scratch[7].r,ya.r,scratch[8].r,yb.r); + scratch[5].i = scratch[0].i + S_MUL_ADD(scratch[7].i,ya.r,scratch[8].i,yb.r); + + scratch[6].r = S_MUL_ADD(scratch[10].i,ya.i,scratch[9].i,yb.i); + scratch[6].i = -S_MUL_ADD(scratch[10].r,ya.i,scratch[9].r,yb.i); + + C_SUB(*Fout1,scratch[5],scratch[6]); + C_ADD(*Fout4,scratch[5],scratch[6]); + + scratch[11].r = scratch[0].r + S_MUL_ADD(scratch[7].r,yb.r,scratch[8].r,ya.r); + scratch[11].i = scratch[0].i + S_MUL_ADD(scratch[7].i,yb.r,scratch[8].i,ya.r); + + scratch[12].r = S_MUL_SUB(scratch[9].i,ya.i,scratch[10].i,yb.i); + scratch[12].i = S_MUL_SUB(scratch[10].r,yb.i,scratch[9].r,ya.i); + + C_ADD(*Fout2,scratch[11],scratch[12]); + C_SUB(*Fout3,scratch[11],scratch[12]); + + ++Fout0;++Fout1;++Fout2;++Fout3;++Fout4; + } + } +} + + +#endif /* KISS_FFT_MIPSR1_H */ diff --git a/media/libopus/celt/mips/mdct_mipsr1.h b/media/libopus/celt/mips/mdct_mipsr1.h new file mode 100644 index 0000000000..2934dab776 --- /dev/null +++ b/media/libopus/celt/mips/mdct_mipsr1.h @@ -0,0 +1,288 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2008 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +/* This is a simple MDCT implementation that uses a N/4 complex FFT + to do most of the work. It should be relatively straightforward to + plug in pretty much and FFT here. + + This replaces the Vorbis FFT (and uses the exact same API), which + was a bit too messy and that was ending up duplicating code + (might as well use the same FFT everywhere). + + The algorithm is similar to (and inspired from) Fabrice Bellard's + MDCT implementation in FFMPEG, but has differences in signs, ordering + and scaling in many places. +*/ +#ifndef __MDCT_MIPSR1_H__ +#define __MDCT_MIPSR1_H__ + +#ifndef SKIP_CONFIG_H +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif +#endif + +#include "mdct.h" +#include "kiss_fft.h" +#include "_kiss_fft_guts.h" +#include <math.h> +#include "os_support.h" +#include "mathops.h" +#include "stack_alloc.h" + +/* Forward MDCT trashes the input array */ +#define OVERRIDE_clt_mdct_forward +void clt_mdct_forward(const mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scalar * OPUS_RESTRICT out, + const opus_val16 *window, int overlap, int shift, int stride, int arch) +{ + int i; + int N, N2, N4; + VARDECL(kiss_fft_scalar, f); + VARDECL(kiss_fft_cpx, f2); + const kiss_fft_state *st = l->kfft[shift]; + const kiss_twiddle_scalar *trig; + opus_val16 scale; +#ifdef FIXED_POINT + /* Allows us to scale with MULT16_32_Q16(), which is faster than + MULT16_32_Q15() on ARM. */ + int scale_shift = st->scale_shift-1; +#endif + + (void)arch; + + SAVE_STACK; + scale = st->scale; + + N = l->n; + trig = l->trig; + for (i=0;i<shift;i++) + { + N >>= 1; + trig += N; + } + N2 = N>>1; + N4 = N>>2; + + ALLOC(f, N2, kiss_fft_scalar); + ALLOC(f2, N4, kiss_fft_cpx); + + /* Consider the input to be composed of four blocks: [a, b, c, d] */ + /* Window, shuffle, fold */ + { + /* Temp pointers to make it really clear to the compiler what we're doing */ + const kiss_fft_scalar * OPUS_RESTRICT xp1 = in+(overlap>>1); + const kiss_fft_scalar * OPUS_RESTRICT xp2 = in+N2-1+(overlap>>1); + kiss_fft_scalar * OPUS_RESTRICT yp = f; + const opus_val16 * OPUS_RESTRICT wp1 = window+(overlap>>1); + const opus_val16 * OPUS_RESTRICT wp2 = window+(overlap>>1)-1; + for(i=0;i<((overlap+3)>>2);i++) + { + /* Real part arranged as -d-cR, Imag part arranged as -b+aR*/ + *yp++ = S_MUL_ADD(*wp2, xp1[N2],*wp1,*xp2); + *yp++ = S_MUL_SUB(*wp1, *xp1,*wp2, xp2[-N2]); + xp1+=2; + xp2-=2; + wp1+=2; + wp2-=2; + } + wp1 = window; + wp2 = window+overlap-1; + for(;i<N4-((overlap+3)>>2);i++) + { + /* Real part arranged as a-bR, Imag part arranged as -c-dR */ + *yp++ = *xp2; + *yp++ = *xp1; + xp1+=2; + xp2-=2; + } + for(;i<N4;i++) + { + /* Real part arranged as a-bR, Imag part arranged as -c-dR */ + *yp++ = S_MUL_SUB(*wp2, *xp2, *wp1, xp1[-N2]); + *yp++ = S_MUL_ADD(*wp2, *xp1, *wp1, xp2[N2]); + xp1+=2; + xp2-=2; + wp1+=2; + wp2-=2; + } + } + /* Pre-rotation */ + { + kiss_fft_scalar * OPUS_RESTRICT yp = f; + const kiss_twiddle_scalar *t = &trig[0]; + for(i=0;i<N4;i++) + { + kiss_fft_cpx yc; + kiss_twiddle_scalar t0, t1; + kiss_fft_scalar re, im, yr, yi; + t0 = t[i]; + t1 = t[N4+i]; + re = *yp++; + im = *yp++; + + yr = S_MUL_SUB(re,t0,im,t1); + yi = S_MUL_ADD(im,t0,re,t1); + + yc.r = yr; + yc.i = yi; + yc.r = PSHR32(MULT16_32_Q16(scale, yc.r), scale_shift); + yc.i = PSHR32(MULT16_32_Q16(scale, yc.i), scale_shift); + f2[st->bitrev[i]] = yc; + } + } + + /* N/4 complex FFT, does not downscale anymore */ + opus_fft_impl(st, f2); + + /* Post-rotate */ + { + /* Temp pointers to make it really clear to the compiler what we're doing */ + const kiss_fft_cpx * OPUS_RESTRICT fp = f2; + kiss_fft_scalar * OPUS_RESTRICT yp1 = out; + kiss_fft_scalar * OPUS_RESTRICT yp2 = out+stride*(N2-1); + const kiss_twiddle_scalar *t = &trig[0]; + /* Temp pointers to make it really clear to the compiler what we're doing */ + for(i=0;i<N4;i++) + { + kiss_fft_scalar yr, yi; + yr = S_MUL_SUB(fp->i,t[N4+i] , fp->r,t[i]); + yi = S_MUL_ADD(fp->r,t[N4+i] ,fp->i,t[i]); + *yp1 = yr; + *yp2 = yi; + fp++; + yp1 += 2*stride; + yp2 -= 2*stride; + } + } + RESTORE_STACK; +} + +#define OVERRIDE_clt_mdct_backward +void clt_mdct_backward(const mdct_lookup *l, kiss_fft_scalar *in, kiss_fft_scalar * OPUS_RESTRICT out, + const opus_val16 * OPUS_RESTRICT window, int overlap, int shift, int stride, int arch) +{ + int i; + int N, N2, N4; + const kiss_twiddle_scalar *trig; + + (void)arch; + + N = l->n; + trig = l->trig; + for (i=0;i<shift;i++) + { + N >>= 1; + trig += N; + } + N2 = N>>1; + N4 = N>>2; + + /* Pre-rotate */ + { + /* Temp pointers to make it really clear to the compiler what we're doing */ + const kiss_fft_scalar * OPUS_RESTRICT xp1 = in; + const kiss_fft_scalar * OPUS_RESTRICT xp2 = in+stride*(N2-1); + kiss_fft_scalar * OPUS_RESTRICT yp = out+(overlap>>1); + const kiss_twiddle_scalar * OPUS_RESTRICT t = &trig[0]; + const opus_int16 * OPUS_RESTRICT bitrev = l->kfft[shift]->bitrev; + for(i=0;i<N4;i++) + { + int rev; + kiss_fft_scalar yr, yi; + rev = *bitrev++; + yr = S_MUL_ADD(*xp2, t[i] , *xp1, t[N4+i]); + yi = S_MUL_SUB(*xp1, t[i] , *xp2, t[N4+i]); + /* We swap real and imag because we use an FFT instead of an IFFT. */ + yp[2*rev+1] = yr; + yp[2*rev] = yi; + /* Storing the pre-rotation directly in the bitrev order. */ + xp1+=2*stride; + xp2-=2*stride; + } + } + + opus_fft_impl(l->kfft[shift], (kiss_fft_cpx*)(out+(overlap>>1))); + + /* Post-rotate and de-shuffle from both ends of the buffer at once to make + it in-place. */ + { + kiss_fft_scalar * OPUS_RESTRICT yp0 = out+(overlap>>1); + kiss_fft_scalar * OPUS_RESTRICT yp1 = out+(overlap>>1)+N2-2; + const kiss_twiddle_scalar *t = &trig[0]; + /* Loop to (N4+1)>>1 to handle odd N4. When N4 is odd, the + middle pair will be computed twice. */ + for(i=0;i<(N4+1)>>1;i++) + { + kiss_fft_scalar re, im, yr, yi; + kiss_twiddle_scalar t0, t1; + /* We swap real and imag because we're using an FFT instead of an IFFT. */ + re = yp0[1]; + im = yp0[0]; + t0 = t[i]; + t1 = t[N4+i]; + /* We'd scale up by 2 here, but instead it's done when mixing the windows */ + yr = S_MUL_ADD(re,t0 , im,t1); + yi = S_MUL_SUB(re,t1 , im,t0); + /* We swap real and imag because we're using an FFT instead of an IFFT. */ + re = yp1[1]; + im = yp1[0]; + yp0[0] = yr; + yp1[1] = yi; + + t0 = t[(N4-i-1)]; + t1 = t[(N2-i-1)]; + /* We'd scale up by 2 here, but instead it's done when mixing the windows */ + yr = S_MUL_ADD(re,t0,im,t1); + yi = S_MUL_SUB(re,t1,im,t0); + yp1[0] = yr; + yp0[1] = yi; + yp0 += 2; + yp1 -= 2; + } + } + + /* Mirror on both sides for TDAC */ + { + kiss_fft_scalar * OPUS_RESTRICT xp1 = out+overlap-1; + kiss_fft_scalar * OPUS_RESTRICT yp1 = out; + const opus_val16 * OPUS_RESTRICT wp1 = window; + const opus_val16 * OPUS_RESTRICT wp2 = window+overlap-1; + + for(i = 0; i < overlap/2; i++) + { + kiss_fft_scalar x1, x2; + x1 = *xp1; + x2 = *yp1; + *yp1++ = MULT16_32_Q15(*wp2, x2) - MULT16_32_Q15(*wp1, x1); + *xp1-- = MULT16_32_Q15(*wp1, x2) + MULT16_32_Q15(*wp2, x1); + wp1++; + wp2--; + } + } +} +#endif /* __MDCT_MIPSR1_H__ */ diff --git a/media/libopus/celt/mips/pitch_mipsr1.h b/media/libopus/celt/mips/pitch_mipsr1.h new file mode 100644 index 0000000000..a9500aff58 --- /dev/null +++ b/media/libopus/celt/mips/pitch_mipsr1.h @@ -0,0 +1,161 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/** + @file pitch.h + @brief Pitch analysis + */ + +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef PITCH_MIPSR1_H +#define PITCH_MIPSR1_H + +#define OVERRIDE_DUAL_INNER_PROD +static inline void dual_inner_prod(const opus_val16 *x, const opus_val16 *y01, const opus_val16 *y02, + int N, opus_val32 *xy1, opus_val32 *xy2, int arch) +{ + int j; + opus_val32 xy01=0; + opus_val32 xy02=0; + + (void)arch; + + asm volatile("MULT $ac1, $0, $0"); + asm volatile("MULT $ac2, $0, $0"); + /* Compute the norm of X+Y and X-Y as |X|^2 + |Y|^2 +/- sum(xy) */ + for (j=0;j<N;j++) + { + asm volatile("MADD $ac1, %0, %1" : : "r" ((int)x[j]), "r" ((int)y01[j])); + asm volatile("MADD $ac2, %0, %1" : : "r" ((int)x[j]), "r" ((int)y02[j])); + ++j; + asm volatile("MADD $ac1, %0, %1" : : "r" ((int)x[j]), "r" ((int)y01[j])); + asm volatile("MADD $ac2, %0, %1" : : "r" ((int)x[j]), "r" ((int)y02[j])); + } + asm volatile ("mflo %0, $ac1": "=r"(xy01)); + asm volatile ("mflo %0, $ac2": "=r"(xy02)); + *xy1 = xy01; + *xy2 = xy02; +} + +static inline void xcorr_kernel_mips(const opus_val16 * x, + const opus_val16 * y, opus_val32 sum[4], int len) +{ + int j; + opus_val16 y_0, y_1, y_2, y_3; + + opus_int64 sum_0, sum_1, sum_2, sum_3; + sum_0 = (opus_int64)sum[0]; + sum_1 = (opus_int64)sum[1]; + sum_2 = (opus_int64)sum[2]; + sum_3 = (opus_int64)sum[3]; + + y_3=0; /* gcc doesn't realize that y_3 can't be used uninitialized */ + y_0=*y++; + y_1=*y++; + y_2=*y++; + for (j=0;j<len-3;j+=4) + { + opus_val16 tmp; + tmp = *x++; + y_3=*y++; + + sum_0 = __builtin_mips_madd( sum_0, tmp, y_0); + sum_1 = __builtin_mips_madd( sum_1, tmp, y_1); + sum_2 = __builtin_mips_madd( sum_2, tmp, y_2); + sum_3 = __builtin_mips_madd( sum_3, tmp, y_3); + + tmp=*x++; + y_0=*y++; + + sum_0 = __builtin_mips_madd( sum_0, tmp, y_1 ); + sum_1 = __builtin_mips_madd( sum_1, tmp, y_2 ); + sum_2 = __builtin_mips_madd( sum_2, tmp, y_3); + sum_3 = __builtin_mips_madd( sum_3, tmp, y_0); + + tmp=*x++; + y_1=*y++; + + sum_0 = __builtin_mips_madd( sum_0, tmp, y_2 ); + sum_1 = __builtin_mips_madd( sum_1, tmp, y_3 ); + sum_2 = __builtin_mips_madd( sum_2, tmp, y_0); + sum_3 = __builtin_mips_madd( sum_3, tmp, y_1); + + + tmp=*x++; + y_2=*y++; + + sum_0 = __builtin_mips_madd( sum_0, tmp, y_3 ); + sum_1 = __builtin_mips_madd( sum_1, tmp, y_0 ); + sum_2 = __builtin_mips_madd( sum_2, tmp, y_1); + sum_3 = __builtin_mips_madd( sum_3, tmp, y_2); + + } + if (j++<len) + { + opus_val16 tmp = *x++; + y_3=*y++; + + sum_0 = __builtin_mips_madd( sum_0, tmp, y_0 ); + sum_1 = __builtin_mips_madd( sum_1, tmp, y_1 ); + sum_2 = __builtin_mips_madd( sum_2, tmp, y_2); + sum_3 = __builtin_mips_madd( sum_3, tmp, y_3); + } + + if (j++<len) + { + opus_val16 tmp=*x++; + y_0=*y++; + + sum_0 = __builtin_mips_madd( sum_0, tmp, y_1 ); + sum_1 = __builtin_mips_madd( sum_1, tmp, y_2 ); + sum_2 = __builtin_mips_madd( sum_2, tmp, y_3); + sum_3 = __builtin_mips_madd( sum_3, tmp, y_0); + } + + if (j<len) + { + opus_val16 tmp=*x++; + y_1=*y++; + + sum_0 = __builtin_mips_madd( sum_0, tmp, y_2 ); + sum_1 = __builtin_mips_madd( sum_1, tmp, y_3 ); + sum_2 = __builtin_mips_madd( sum_2, tmp, y_0); + sum_3 = __builtin_mips_madd( sum_3, tmp, y_1); + + } + + sum[0] = (opus_val32)sum_0; + sum[1] = (opus_val32)sum_1; + sum[2] = (opus_val32)sum_2; + sum[3] = (opus_val32)sum_3; +} + +#define OVERRIDE_XCORR_KERNEL +#define xcorr_kernel(x, y, sum, len, arch) \ + ((void)(arch), xcorr_kernel_mips(x, y, sum, len)) + +#endif /* PITCH_MIPSR1_H */ diff --git a/media/libopus/celt/mips/vq_mipsr1.h b/media/libopus/celt/mips/vq_mipsr1.h new file mode 100644 index 0000000000..f26a33e755 --- /dev/null +++ b/media/libopus/celt/mips/vq_mipsr1.h @@ -0,0 +1,116 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef __VQ_MIPSR1_H__ +#define __VQ_MIPSR1_H__ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "mathops.h" +#include "arch.h" + +#define OVERRIDE_vq_exp_rotation1 +static void exp_rotation1(celt_norm *X, int len, int stride, opus_val16 c, opus_val16 s) +{ + int i; + opus_val16 ms; + celt_norm *Xptr; + Xptr = X; + ms = NEG16(s); + for (i=0;i<len-stride;i++) + { + celt_norm x1, x2; + x1 = Xptr[0]; + x2 = Xptr[stride]; + Xptr[stride] = EXTRACT16(PSHR32(MAC16_16(MULT16_16(c, x2), s, x1), 15)); + *Xptr++ = EXTRACT16(PSHR32(MAC16_16(MULT16_16(c, x1), ms, x2), 15)); + } + Xptr = &X[len-2*stride-1]; + for (i=len-2*stride-1;i>=0;i--) + { + celt_norm x1, x2; + x1 = Xptr[0]; + x2 = Xptr[stride]; + Xptr[stride] = EXTRACT16(PSHR32(MAC16_16(MULT16_16(c, x2), s, x1), 15)); + *Xptr-- = EXTRACT16(PSHR32(MAC16_16(MULT16_16(c, x1), ms, x2), 15)); + } +} + +#define OVERRIDE_renormalise_vector +void renormalise_vector(celt_norm *X, int N, opus_val16 gain, int arch) +{ + int i; +#ifdef FIXED_POINT + int k; +#endif + opus_val32 E = EPSILON; + opus_val16 g; + opus_val32 t; + celt_norm *xptr = X; + int X0, X1; + + (void)arch; + + asm volatile("mult $ac1, $0, $0"); + asm volatile("MTLO %0, $ac1" : :"r" (E)); + /*if(N %4) + printf("error");*/ + for (i=0;i<N-2;i+=2) + { + X0 = (int)*xptr++; + asm volatile("MADD $ac1, %0, %1" : : "r" (X0), "r" (X0)); + + X1 = (int)*xptr++; + asm volatile("MADD $ac1, %0, %1" : : "r" (X1), "r" (X1)); + } + + for (;i<N;i++) + { + X0 = (int)*xptr++; + asm volatile("MADD $ac1, %0, %1" : : "r" (X0), "r" (X0)); + } + + asm volatile("MFLO %0, $ac1" : "=r" (E)); +#ifdef FIXED_POINT + k = celt_ilog2(E)>>1; +#endif + t = VSHR32(E, 2*(k-7)); + g = MULT16_16_P15(celt_rsqrt_norm(t),gain); + + xptr = X; + for (i=0;i<N;i++) + { + *xptr = EXTRACT16(PSHR32(MULT16_16(g, *xptr), k+1)); + xptr++; + } + /*return celt_sqrt(E);*/ +} + +#endif /* __VQ_MIPSR1_H__ */ diff --git a/media/libopus/celt/modes.c b/media/libopus/celt/modes.c new file mode 100644 index 0000000000..23f7cde662 --- /dev/null +++ b/media/libopus/celt/modes.c @@ -0,0 +1,445 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Copyright (c) 2008 Gregory Maxwell + Written by Jean-Marc Valin and Gregory Maxwell */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "celt.h" +#include "modes.h" +#include "rate.h" +#include "os_support.h" +#include "stack_alloc.h" +#include "quant_bands.h" +#include "cpu_support.h" + +static const opus_int16 eband5ms[] = { +/*0 200 400 600 800 1k 1.2 1.4 1.6 2k 2.4 2.8 3.2 4k 4.8 5.6 6.8 8k 9.6 12k 15.6 */ + 0, 1, 2, 3, 4, 5, 6, 7, 8, 10, 12, 14, 16, 20, 24, 28, 34, 40, 48, 60, 78, 100 +}; + +/* Alternate tuning (partially derived from Vorbis) */ +#define BITALLOC_SIZE 11 +/* Bit allocation table in units of 1/32 bit/sample (0.1875 dB SNR) */ +static const unsigned char band_allocation[] = { +/*0 200 400 600 800 1k 1.2 1.4 1.6 2k 2.4 2.8 3.2 4k 4.8 5.6 6.8 8k 9.6 12k 15.6 */ + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, + 90, 80, 75, 69, 63, 56, 49, 40, 34, 29, 20, 18, 10, 0, 0, 0, 0, 0, 0, 0, 0, +110,100, 90, 84, 78, 71, 65, 58, 51, 45, 39, 32, 26, 20, 12, 0, 0, 0, 0, 0, 0, +118,110,103, 93, 86, 80, 75, 70, 65, 59, 53, 47, 40, 31, 23, 15, 4, 0, 0, 0, 0, +126,119,112,104, 95, 89, 83, 78, 72, 66, 60, 54, 47, 39, 32, 25, 17, 12, 1, 0, 0, +134,127,120,114,103, 97, 91, 85, 78, 72, 66, 60, 54, 47, 41, 35, 29, 23, 16, 10, 1, +144,137,130,124,113,107,101, 95, 88, 82, 76, 70, 64, 57, 51, 45, 39, 33, 26, 15, 1, +152,145,138,132,123,117,111,105, 98, 92, 86, 80, 74, 67, 61, 55, 49, 43, 36, 20, 1, +162,155,148,142,133,127,121,115,108,102, 96, 90, 84, 77, 71, 65, 59, 53, 46, 30, 1, +172,165,158,152,143,137,131,125,118,112,106,100, 94, 87, 81, 75, 69, 63, 56, 45, 20, +200,200,200,200,200,200,200,200,198,193,188,183,178,173,168,163,158,153,148,129,104, +}; + +#ifndef CUSTOM_MODES_ONLY + #ifdef FIXED_POINT + #include "static_modes_fixed.h" + #else + #include "static_modes_float.h" + #endif +#endif /* CUSTOM_MODES_ONLY */ + +#ifndef M_PI +#define M_PI 3.141592653 +#endif + +#ifdef CUSTOM_MODES + +/* Defining 25 critical bands for the full 0-20 kHz audio bandwidth + Taken from http://ccrma.stanford.edu/~jos/bbt/Bark_Frequency_Scale.html */ +#define BARK_BANDS 25 +static const opus_int16 bark_freq[BARK_BANDS+1] = { + 0, 100, 200, 300, 400, + 510, 630, 770, 920, 1080, + 1270, 1480, 1720, 2000, 2320, + 2700, 3150, 3700, 4400, 5300, + 6400, 7700, 9500, 12000, 15500, + 20000}; + +static opus_int16 *compute_ebands(opus_int32 Fs, int frame_size, int res, int *nbEBands) +{ + opus_int16 *eBands; + int i, j, lin, low, high, nBark, offset=0; + + /* All modes that have 2.5 ms short blocks use the same definition */ + if (Fs == 400*(opus_int32)frame_size) + { + *nbEBands = sizeof(eband5ms)/sizeof(eband5ms[0])-1; + eBands = opus_alloc(sizeof(opus_int16)*(*nbEBands+1)); + for (i=0;i<*nbEBands+1;i++) + eBands[i] = eband5ms[i]; + return eBands; + } + /* Find the number of critical bands supported by our sampling rate */ + for (nBark=1;nBark<BARK_BANDS;nBark++) + if (bark_freq[nBark+1]*2 >= Fs) + break; + + /* Find where the linear part ends (i.e. where the spacing is more than min_width */ + for (lin=0;lin<nBark;lin++) + if (bark_freq[lin+1]-bark_freq[lin] >= res) + break; + + low = (bark_freq[lin]+res/2)/res; + high = nBark-lin; + *nbEBands = low+high; + eBands = opus_alloc(sizeof(opus_int16)*(*nbEBands+2)); + + if (eBands==NULL) + return NULL; + + /* Linear spacing (min_width) */ + for (i=0;i<low;i++) + eBands[i] = i; + if (low>0) + offset = eBands[low-1]*res - bark_freq[lin-1]; + /* Spacing follows critical bands */ + for (i=0;i<high;i++) + { + int target = bark_freq[lin+i]; + /* Round to an even value */ + eBands[i+low] = (target+offset/2+res)/(2*res)*2; + offset = eBands[i+low]*res - target; + } + /* Enforce the minimum spacing at the boundary */ + for (i=0;i<*nbEBands;i++) + if (eBands[i] < i) + eBands[i] = i; + /* Round to an even value */ + eBands[*nbEBands] = (bark_freq[nBark]+res)/(2*res)*2; + if (eBands[*nbEBands] > frame_size) + eBands[*nbEBands] = frame_size; + for (i=1;i<*nbEBands-1;i++) + { + if (eBands[i+1]-eBands[i] < eBands[i]-eBands[i-1]) + { + eBands[i] -= (2*eBands[i]-eBands[i-1]-eBands[i+1])/2; + } + } + /* Remove any empty bands. */ + for (i=j=0;i<*nbEBands;i++) + if(eBands[i+1]>eBands[j]) + eBands[++j]=eBands[i+1]; + *nbEBands=j; + + for (i=1;i<*nbEBands;i++) + { + /* Every band must be smaller than the last band. */ + celt_assert(eBands[i]-eBands[i-1]<=eBands[*nbEBands]-eBands[*nbEBands-1]); + /* Each band must be no larger than twice the size of the previous one. */ + celt_assert(eBands[i+1]-eBands[i]<=2*(eBands[i]-eBands[i-1])); + } + + return eBands; +} + +static void compute_allocation_table(CELTMode *mode) +{ + int i, j; + unsigned char *allocVectors; + int maxBands = sizeof(eband5ms)/sizeof(eband5ms[0])-1; + + mode->nbAllocVectors = BITALLOC_SIZE; + allocVectors = opus_alloc(sizeof(unsigned char)*(BITALLOC_SIZE*mode->nbEBands)); + if (allocVectors==NULL) + { + mode->allocVectors = NULL; + return; + } + + /* Check for standard mode */ + if (mode->Fs == 400*(opus_int32)mode->shortMdctSize) + { + for (i=0;i<BITALLOC_SIZE*mode->nbEBands;i++) + allocVectors[i] = band_allocation[i]; + mode->allocVectors = allocVectors; + return; + } + /* If not the standard mode, interpolate */ + /* Compute per-codec-band allocation from per-critical-band matrix */ + for (i=0;i<BITALLOC_SIZE;i++) + { + for (j=0;j<mode->nbEBands;j++) + { + int k; + for (k=0;k<maxBands;k++) + { + if (400*(opus_int32)eband5ms[k] > mode->eBands[j]*(opus_int32)mode->Fs/mode->shortMdctSize) + break; + } + if (k>maxBands-1) + allocVectors[i*mode->nbEBands+j] = band_allocation[i*maxBands + maxBands-1]; + else { + opus_int32 a0, a1; + a1 = mode->eBands[j]*(opus_int32)mode->Fs/mode->shortMdctSize - 400*(opus_int32)eband5ms[k-1]; + a0 = 400*(opus_int32)eband5ms[k] - mode->eBands[j]*(opus_int32)mode->Fs/mode->shortMdctSize; + allocVectors[i*mode->nbEBands+j] = (a0*band_allocation[i*maxBands+k-1] + + a1*band_allocation[i*maxBands+k])/(a0+a1); + } + } + } + + /*printf ("\n"); + for (i=0;i<BITALLOC_SIZE;i++) + { + for (j=0;j<mode->nbEBands;j++) + printf ("%d ", allocVectors[i*mode->nbEBands+j]); + printf ("\n"); + } + exit(0);*/ + + mode->allocVectors = allocVectors; +} + +#endif /* CUSTOM_MODES */ + +CELTMode *opus_custom_mode_create(opus_int32 Fs, int frame_size, int *error) +{ + int i; +#ifdef CUSTOM_MODES + CELTMode *mode=NULL; + int res; + opus_val16 *window; + opus_int16 *logN; + int LM; + int arch = opus_select_arch(); + ALLOC_STACK; +#if !defined(VAR_ARRAYS) && !defined(USE_ALLOCA) + if (global_stack==NULL) + goto failure; +#endif +#endif + +#ifndef CUSTOM_MODES_ONLY + for (i=0;i<TOTAL_MODES;i++) + { + int j; + for (j=0;j<4;j++) + { + if (Fs == static_mode_list[i]->Fs && + (frame_size<<j) == static_mode_list[i]->shortMdctSize*static_mode_list[i]->nbShortMdcts) + { + if (error) + *error = OPUS_OK; + return (CELTMode*)static_mode_list[i]; + } + } + } +#endif /* CUSTOM_MODES_ONLY */ + +#ifndef CUSTOM_MODES + if (error) + *error = OPUS_BAD_ARG; + return NULL; +#else + + /* The good thing here is that permutation of the arguments will automatically be invalid */ + + if (Fs < 8000 || Fs > 96000) + { + if (error) + *error = OPUS_BAD_ARG; + return NULL; + } + if (frame_size < 40 || frame_size > 1024 || frame_size%2!=0) + { + if (error) + *error = OPUS_BAD_ARG; + return NULL; + } + /* Frames of less than 1ms are not supported. */ + if ((opus_int32)frame_size*1000 < Fs) + { + if (error) + *error = OPUS_BAD_ARG; + return NULL; + } + + if ((opus_int32)frame_size*75 >= Fs && (frame_size%16)==0) + { + LM = 3; + } else if ((opus_int32)frame_size*150 >= Fs && (frame_size%8)==0) + { + LM = 2; + } else if ((opus_int32)frame_size*300 >= Fs && (frame_size%4)==0) + { + LM = 1; + } else + { + LM = 0; + } + + /* Shorts longer than 3.3ms are not supported. */ + if ((opus_int32)(frame_size>>LM)*300 > Fs) + { + if (error) + *error = OPUS_BAD_ARG; + return NULL; + } + + mode = opus_alloc(sizeof(CELTMode)); + if (mode==NULL) + goto failure; + mode->Fs = Fs; + + /* Pre/de-emphasis depends on sampling rate. The "standard" pre-emphasis + is defined as A(z) = 1 - 0.85*z^-1 at 48 kHz. Other rates should + approximate that. */ + if(Fs < 12000) /* 8 kHz */ + { + mode->preemph[0] = QCONST16(0.3500061035f, 15); + mode->preemph[1] = -QCONST16(0.1799926758f, 15); + mode->preemph[2] = QCONST16(0.2719968125f, SIG_SHIFT); /* exact 1/preemph[3] */ + mode->preemph[3] = QCONST16(3.6765136719f, 13); + } else if(Fs < 24000) /* 16 kHz */ + { + mode->preemph[0] = QCONST16(0.6000061035f, 15); + mode->preemph[1] = -QCONST16(0.1799926758f, 15); + mode->preemph[2] = QCONST16(0.4424998650f, SIG_SHIFT); /* exact 1/preemph[3] */ + mode->preemph[3] = QCONST16(2.2598876953f, 13); + } else if(Fs < 40000) /* 32 kHz */ + { + mode->preemph[0] = QCONST16(0.7799987793f, 15); + mode->preemph[1] = -QCONST16(0.1000061035f, 15); + mode->preemph[2] = QCONST16(0.7499771125f, SIG_SHIFT); /* exact 1/preemph[3] */ + mode->preemph[3] = QCONST16(1.3333740234f, 13); + } else /* 48 kHz */ + { + mode->preemph[0] = QCONST16(0.8500061035f, 15); + mode->preemph[1] = QCONST16(0.0f, 15); + mode->preemph[2] = QCONST16(1.f, SIG_SHIFT); + mode->preemph[3] = QCONST16(1.f, 13); + } + + mode->maxLM = LM; + mode->nbShortMdcts = 1<<LM; + mode->shortMdctSize = frame_size/mode->nbShortMdcts; + res = (mode->Fs+mode->shortMdctSize)/(2*mode->shortMdctSize); + + mode->eBands = compute_ebands(Fs, mode->shortMdctSize, res, &mode->nbEBands); + if (mode->eBands==NULL) + goto failure; +#if !defined(SMALL_FOOTPRINT) + /* Make sure we don't allocate a band larger than our PVQ table. + 208 should be enough, but let's be paranoid. */ + if ((mode->eBands[mode->nbEBands] - mode->eBands[mode->nbEBands-1])<<LM > + 208) { + goto failure; + } +#endif + + mode->effEBands = mode->nbEBands; + while (mode->eBands[mode->effEBands] > mode->shortMdctSize) + mode->effEBands--; + + /* Overlap must be divisible by 4 */ + mode->overlap = ((mode->shortMdctSize>>2)<<2); + + compute_allocation_table(mode); + if (mode->allocVectors==NULL) + goto failure; + + window = (opus_val16*)opus_alloc(mode->overlap*sizeof(opus_val16)); + if (window==NULL) + goto failure; + +#ifndef FIXED_POINT + for (i=0;i<mode->overlap;i++) + window[i] = Q15ONE*sin(.5*M_PI* sin(.5*M_PI*(i+.5)/mode->overlap) * sin(.5*M_PI*(i+.5)/mode->overlap)); +#else + for (i=0;i<mode->overlap;i++) + window[i] = MIN32(32767,floor(.5+32768.*sin(.5*M_PI* sin(.5*M_PI*(i+.5)/mode->overlap) * sin(.5*M_PI*(i+.5)/mode->overlap)))); +#endif + mode->window = window; + + logN = (opus_int16*)opus_alloc(mode->nbEBands*sizeof(opus_int16)); + if (logN==NULL) + goto failure; + + for (i=0;i<mode->nbEBands;i++) + logN[i] = log2_frac(mode->eBands[i+1]-mode->eBands[i], BITRES); + mode->logN = logN; + + compute_pulse_cache(mode, mode->maxLM); + + if (clt_mdct_init(&mode->mdct, 2*mode->shortMdctSize*mode->nbShortMdcts, + mode->maxLM, arch) == 0) + goto failure; + + if (error) + *error = OPUS_OK; + + return mode; +failure: + if (error) + *error = OPUS_ALLOC_FAIL; + if (mode!=NULL) + opus_custom_mode_destroy(mode); + return NULL; +#endif /* !CUSTOM_MODES */ +} + +#ifdef CUSTOM_MODES +void opus_custom_mode_destroy(CELTMode *mode) +{ + int arch = opus_select_arch(); + + if (mode == NULL) + return; +#ifndef CUSTOM_MODES_ONLY + { + int i; + for (i=0;i<TOTAL_MODES;i++) + { + if (mode == static_mode_list[i]) + { + return; + } + } + } +#endif /* CUSTOM_MODES_ONLY */ + opus_free((opus_int16*)mode->eBands); + opus_free((unsigned char*)mode->allocVectors); + + opus_free((opus_val16*)mode->window); + opus_free((opus_int16*)mode->logN); + + opus_free((opus_int16*)mode->cache.index); + opus_free((unsigned char*)mode->cache.bits); + opus_free((unsigned char*)mode->cache.caps); + clt_mdct_clear(&mode->mdct, arch); + + opus_free((CELTMode *)mode); +} +#endif diff --git a/media/libopus/celt/modes.h b/media/libopus/celt/modes.h new file mode 100644 index 0000000000..be813ccc8b --- /dev/null +++ b/media/libopus/celt/modes.h @@ -0,0 +1,75 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Copyright (c) 2008 Gregory Maxwell + Written by Jean-Marc Valin and Gregory Maxwell */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef MODES_H +#define MODES_H + +#include "opus_types.h" +#include "celt.h" +#include "arch.h" +#include "mdct.h" +#include "entenc.h" +#include "entdec.h" + +#define MAX_PERIOD 1024 + +typedef struct { + int size; + const opus_int16 *index; + const unsigned char *bits; + const unsigned char *caps; +} PulseCache; + +/** Mode definition (opaque) + @brief Mode definition + */ +struct OpusCustomMode { + opus_int32 Fs; + int overlap; + + int nbEBands; + int effEBands; + opus_val16 preemph[4]; + const opus_int16 *eBands; /**< Definition for each "pseudo-critical band" */ + + int maxLM; + int nbShortMdcts; + int shortMdctSize; + + int nbAllocVectors; /**< Number of lines in the matrix below */ + const unsigned char *allocVectors; /**< Number of bits in each band for several rates */ + const opus_int16 *logN; + + const opus_val16 *window; + mdct_lookup mdct; + PulseCache cache; +}; + + +#endif diff --git a/media/libopus/celt/os_support.h b/media/libopus/celt/os_support.h new file mode 100644 index 0000000000..009bf861da --- /dev/null +++ b/media/libopus/celt/os_support.h @@ -0,0 +1,91 @@ +/* Copyright (C) 2007 Jean-Marc Valin + + File: os_support.h + This is the (tiny) OS abstraction layer. Aside from math.h, this is the + only place where system headers are allowed. + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions are + met: + + 1. Redistributions of source code must retain the above copyright notice, + this list of conditions and the following disclaimer. + + 2. Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR + IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES + OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE + DISCLAIMED. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, + INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR + SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, + STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN + ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE + POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef OS_SUPPORT_H +#define OS_SUPPORT_H + +#ifdef CUSTOM_SUPPORT +# include "custom_support.h" +#endif + +#include "opus_types.h" +#include "opus_defines.h" + +#include <string.h> +#include <stdlib.h> + +/** Opus wrapper for malloc(). To do your own dynamic allocation, all you need to do is replace this function and opus_free */ +#ifndef OVERRIDE_OPUS_ALLOC +static OPUS_INLINE void *opus_alloc (size_t size) +{ + return malloc(size); +} +#endif + +/** Same as celt_alloc(), except that the area is only needed inside a CELT call (might cause problem with wideband though) */ +#ifndef OVERRIDE_OPUS_ALLOC_SCRATCH +static OPUS_INLINE void *opus_alloc_scratch (size_t size) +{ + /* Scratch space doesn't need to be cleared */ + return opus_alloc(size); +} +#endif + +/** Opus wrapper for free(). To do your own dynamic allocation, all you need to do is replace this function and opus_alloc */ +#ifndef OVERRIDE_OPUS_FREE +static OPUS_INLINE void opus_free (void *ptr) +{ + free(ptr); +} +#endif + +/** Copy n elements from src to dst. The 0* term provides compile-time type checking */ +#ifndef OVERRIDE_OPUS_COPY +#define OPUS_COPY(dst, src, n) (memcpy((dst), (src), (n)*sizeof(*(dst)) + 0*((dst)-(src)) )) +#endif + +/** Copy n elements from src to dst, allowing overlapping regions. The 0* term + provides compile-time type checking */ +#ifndef OVERRIDE_OPUS_MOVE +#define OPUS_MOVE(dst, src, n) (memmove((dst), (src), (n)*sizeof(*(dst)) + 0*((dst)-(src)) )) +#endif + +/** Set n elements of dst to zero */ +#ifndef OVERRIDE_OPUS_CLEAR +#define OPUS_CLEAR(dst, n) (memset((dst), 0, (n)*sizeof(*(dst)))) +#endif + +/*#ifdef __GNUC__ +#pragma GCC poison printf sprintf +#pragma GCC poison malloc free realloc calloc +#endif*/ + +#endif /* OS_SUPPORT_H */ + diff --git a/media/libopus/celt/pitch.c b/media/libopus/celt/pitch.c new file mode 100644 index 0000000000..7998db4164 --- /dev/null +++ b/media/libopus/celt/pitch.c @@ -0,0 +1,546 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/** + @file pitch.c + @brief Pitch analysis + */ + +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "pitch.h" +#include "os_support.h" +#include "modes.h" +#include "stack_alloc.h" +#include "mathops.h" +#include "celt_lpc.h" + +static void find_best_pitch(opus_val32 *xcorr, opus_val16 *y, int len, + int max_pitch, int *best_pitch +#ifdef FIXED_POINT + , int yshift, opus_val32 maxcorr +#endif + ) +{ + int i, j; + opus_val32 Syy=1; + opus_val16 best_num[2]; + opus_val32 best_den[2]; +#ifdef FIXED_POINT + int xshift; + + xshift = celt_ilog2(maxcorr)-14; +#endif + + best_num[0] = -1; + best_num[1] = -1; + best_den[0] = 0; + best_den[1] = 0; + best_pitch[0] = 0; + best_pitch[1] = 1; + for (j=0;j<len;j++) + Syy = ADD32(Syy, SHR32(MULT16_16(y[j],y[j]), yshift)); + for (i=0;i<max_pitch;i++) + { + if (xcorr[i]>0) + { + opus_val16 num; + opus_val32 xcorr16; + xcorr16 = EXTRACT16(VSHR32(xcorr[i], xshift)); +#ifndef FIXED_POINT + /* Considering the range of xcorr16, this should avoid both underflows + and overflows (inf) when squaring xcorr16 */ + xcorr16 *= 1e-12f; +#endif + num = MULT16_16_Q15(xcorr16,xcorr16); + if (MULT16_32_Q15(num,best_den[1]) > MULT16_32_Q15(best_num[1],Syy)) + { + if (MULT16_32_Q15(num,best_den[0]) > MULT16_32_Q15(best_num[0],Syy)) + { + best_num[1] = best_num[0]; + best_den[1] = best_den[0]; + best_pitch[1] = best_pitch[0]; + best_num[0] = num; + best_den[0] = Syy; + best_pitch[0] = i; + } else { + best_num[1] = num; + best_den[1] = Syy; + best_pitch[1] = i; + } + } + } + Syy += SHR32(MULT16_16(y[i+len],y[i+len]),yshift) - SHR32(MULT16_16(y[i],y[i]),yshift); + Syy = MAX32(1, Syy); + } +} + +static void celt_fir5(opus_val16 *x, + const opus_val16 *num, + int N) +{ + int i; + opus_val16 num0, num1, num2, num3, num4; + opus_val32 mem0, mem1, mem2, mem3, mem4; + num0=num[0]; + num1=num[1]; + num2=num[2]; + num3=num[3]; + num4=num[4]; + mem0=0; + mem1=0; + mem2=0; + mem3=0; + mem4=0; + for (i=0;i<N;i++) + { + opus_val32 sum = SHL32(EXTEND32(x[i]), SIG_SHIFT); + sum = MAC16_16(sum,num0,mem0); + sum = MAC16_16(sum,num1,mem1); + sum = MAC16_16(sum,num2,mem2); + sum = MAC16_16(sum,num3,mem3); + sum = MAC16_16(sum,num4,mem4); + mem4 = mem3; + mem3 = mem2; + mem2 = mem1; + mem1 = mem0; + mem0 = x[i]; + x[i] = ROUND16(sum, SIG_SHIFT); + } +} + + +void pitch_downsample(celt_sig * OPUS_RESTRICT x[], opus_val16 * OPUS_RESTRICT x_lp, + int len, int C, int arch) +{ + int i; + opus_val32 ac[5]; + opus_val16 tmp=Q15ONE; + opus_val16 lpc[4]; + opus_val16 lpc2[5]; + opus_val16 c1 = QCONST16(.8f,15); +#ifdef FIXED_POINT + int shift; + opus_val32 maxabs = celt_maxabs32(x[0], len); + if (C==2) + { + opus_val32 maxabs_1 = celt_maxabs32(x[1], len); + maxabs = MAX32(maxabs, maxabs_1); + } + if (maxabs<1) + maxabs=1; + shift = celt_ilog2(maxabs)-10; + if (shift<0) + shift=0; + if (C==2) + shift++; + for (i=1;i<len>>1;i++) + x_lp[i] = SHR32(x[0][(2*i-1)], shift+2) + SHR32(x[0][(2*i+1)], shift+2) + SHR32(x[0][2*i], shift+1); + x_lp[0] = SHR32(x[0][1], shift+2) + SHR32(x[0][0], shift+1); + if (C==2) + { + for (i=1;i<len>>1;i++) + x_lp[i] += SHR32(x[1][(2*i-1)], shift+2) + SHR32(x[1][(2*i+1)], shift+2) + SHR32(x[1][2*i], shift+1); + x_lp[0] += SHR32(x[1][1], shift+2) + SHR32(x[1][0], shift+1); + } +#else + for (i=1;i<len>>1;i++) + x_lp[i] = .25f*x[0][(2*i-1)] + .25f*x[0][(2*i+1)] + .5f*x[0][2*i]; + x_lp[0] = .25f*x[0][1] + .5f*x[0][0]; + if (C==2) + { + for (i=1;i<len>>1;i++) + x_lp[i] += .25f*x[1][(2*i-1)] + .25f*x[1][(2*i+1)] + .5f*x[1][2*i]; + x_lp[0] += .25f*x[1][1] + .5f*x[1][0]; + } +#endif + _celt_autocorr(x_lp, ac, NULL, 0, + 4, len>>1, arch); + + /* Noise floor -40 dB */ +#ifdef FIXED_POINT + ac[0] += SHR32(ac[0],13); +#else + ac[0] *= 1.0001f; +#endif + /* Lag windowing */ + for (i=1;i<=4;i++) + { + /*ac[i] *= exp(-.5*(2*M_PI*.002*i)*(2*M_PI*.002*i));*/ +#ifdef FIXED_POINT + ac[i] -= MULT16_32_Q15(2*i*i, ac[i]); +#else + ac[i] -= ac[i]*(.008f*i)*(.008f*i); +#endif + } + + _celt_lpc(lpc, ac, 4); + for (i=0;i<4;i++) + { + tmp = MULT16_16_Q15(QCONST16(.9f,15), tmp); + lpc[i] = MULT16_16_Q15(lpc[i], tmp); + } + /* Add a zero */ + lpc2[0] = lpc[0] + QCONST16(.8f,SIG_SHIFT); + lpc2[1] = lpc[1] + MULT16_16_Q15(c1,lpc[0]); + lpc2[2] = lpc[2] + MULT16_16_Q15(c1,lpc[1]); + lpc2[3] = lpc[3] + MULT16_16_Q15(c1,lpc[2]); + lpc2[4] = MULT16_16_Q15(c1,lpc[3]); + celt_fir5(x_lp, lpc2, len>>1); +} + +/* Pure C implementation. */ +#ifdef FIXED_POINT +opus_val32 +#else +void +#endif +celt_pitch_xcorr_c(const opus_val16 *_x, const opus_val16 *_y, + opus_val32 *xcorr, int len, int max_pitch, int arch) +{ + +#if 0 /* This is a simple version of the pitch correlation that should work + well on DSPs like Blackfin and TI C5x/C6x */ + int i, j; +#ifdef FIXED_POINT + opus_val32 maxcorr=1; +#endif +#if !defined(OVERRIDE_PITCH_XCORR) + (void)arch; +#endif + for (i=0;i<max_pitch;i++) + { + opus_val32 sum = 0; + for (j=0;j<len;j++) + sum = MAC16_16(sum, _x[j], _y[i+j]); + xcorr[i] = sum; +#ifdef FIXED_POINT + maxcorr = MAX32(maxcorr, sum); +#endif + } +#ifdef FIXED_POINT + return maxcorr; +#endif + +#else /* Unrolled version of the pitch correlation -- runs faster on x86 and ARM */ + int i; + /*The EDSP version requires that max_pitch is at least 1, and that _x is + 32-bit aligned. + Since it's hard to put asserts in assembly, put them here.*/ +#ifdef FIXED_POINT + opus_val32 maxcorr=1; +#endif + celt_assert(max_pitch>0); + celt_sig_assert(((size_t)_x&3)==0); + for (i=0;i<max_pitch-3;i+=4) + { + opus_val32 sum[4]={0,0,0,0}; + xcorr_kernel(_x, _y+i, sum, len, arch); + xcorr[i]=sum[0]; + xcorr[i+1]=sum[1]; + xcorr[i+2]=sum[2]; + xcorr[i+3]=sum[3]; +#ifdef FIXED_POINT + sum[0] = MAX32(sum[0], sum[1]); + sum[2] = MAX32(sum[2], sum[3]); + sum[0] = MAX32(sum[0], sum[2]); + maxcorr = MAX32(maxcorr, sum[0]); +#endif + } + /* In case max_pitch isn't a multiple of 4, do non-unrolled version. */ + for (;i<max_pitch;i++) + { + opus_val32 sum; + sum = celt_inner_prod(_x, _y+i, len, arch); + xcorr[i] = sum; +#ifdef FIXED_POINT + maxcorr = MAX32(maxcorr, sum); +#endif + } +#ifdef FIXED_POINT + return maxcorr; +#endif +#endif +} + +void pitch_search(const opus_val16 * OPUS_RESTRICT x_lp, opus_val16 * OPUS_RESTRICT y, + int len, int max_pitch, int *pitch, int arch) +{ + int i, j; + int lag; + int best_pitch[2]={0,0}; + VARDECL(opus_val16, x_lp4); + VARDECL(opus_val16, y_lp4); + VARDECL(opus_val32, xcorr); +#ifdef FIXED_POINT + opus_val32 maxcorr; + opus_val32 xmax, ymax; + int shift=0; +#endif + int offset; + + SAVE_STACK; + + celt_assert(len>0); + celt_assert(max_pitch>0); + lag = len+max_pitch; + + ALLOC(x_lp4, len>>2, opus_val16); + ALLOC(y_lp4, lag>>2, opus_val16); + ALLOC(xcorr, max_pitch>>1, opus_val32); + + /* Downsample by 2 again */ + for (j=0;j<len>>2;j++) + x_lp4[j] = x_lp[2*j]; + for (j=0;j<lag>>2;j++) + y_lp4[j] = y[2*j]; + +#ifdef FIXED_POINT + xmax = celt_maxabs16(x_lp4, len>>2); + ymax = celt_maxabs16(y_lp4, lag>>2); + shift = celt_ilog2(MAX32(1, MAX32(xmax, ymax)))-11; + if (shift>0) + { + for (j=0;j<len>>2;j++) + x_lp4[j] = SHR16(x_lp4[j], shift); + for (j=0;j<lag>>2;j++) + y_lp4[j] = SHR16(y_lp4[j], shift); + /* Use double the shift for a MAC */ + shift *= 2; + } else { + shift = 0; + } +#endif + + /* Coarse search with 4x decimation */ + +#ifdef FIXED_POINT + maxcorr = +#endif + celt_pitch_xcorr(x_lp4, y_lp4, xcorr, len>>2, max_pitch>>2, arch); + + find_best_pitch(xcorr, y_lp4, len>>2, max_pitch>>2, best_pitch +#ifdef FIXED_POINT + , 0, maxcorr +#endif + ); + + /* Finer search with 2x decimation */ +#ifdef FIXED_POINT + maxcorr=1; +#endif + for (i=0;i<max_pitch>>1;i++) + { + opus_val32 sum; + xcorr[i] = 0; + if (abs(i-2*best_pitch[0])>2 && abs(i-2*best_pitch[1])>2) + continue; +#ifdef FIXED_POINT + sum = 0; + for (j=0;j<len>>1;j++) + sum += SHR32(MULT16_16(x_lp[j],y[i+j]), shift); +#else + sum = celt_inner_prod(x_lp, y+i, len>>1, arch); +#endif + xcorr[i] = MAX32(-1, sum); +#ifdef FIXED_POINT + maxcorr = MAX32(maxcorr, sum); +#endif + } + find_best_pitch(xcorr, y, len>>1, max_pitch>>1, best_pitch +#ifdef FIXED_POINT + , shift+1, maxcorr +#endif + ); + + /* Refine by pseudo-interpolation */ + if (best_pitch[0]>0 && best_pitch[0]<(max_pitch>>1)-1) + { + opus_val32 a, b, c; + a = xcorr[best_pitch[0]-1]; + b = xcorr[best_pitch[0]]; + c = xcorr[best_pitch[0]+1]; + if ((c-a) > MULT16_32_Q15(QCONST16(.7f,15),b-a)) + offset = 1; + else if ((a-c) > MULT16_32_Q15(QCONST16(.7f,15),b-c)) + offset = -1; + else + offset = 0; + } else { + offset = 0; + } + *pitch = 2*best_pitch[0]-offset; + + RESTORE_STACK; +} + +#ifdef FIXED_POINT +static opus_val16 compute_pitch_gain(opus_val32 xy, opus_val32 xx, opus_val32 yy) +{ + opus_val32 x2y2; + int sx, sy, shift; + opus_val32 g; + opus_val16 den; + if (xy == 0 || xx == 0 || yy == 0) + return 0; + sx = celt_ilog2(xx)-14; + sy = celt_ilog2(yy)-14; + shift = sx + sy; + x2y2 = SHR32(MULT16_16(VSHR32(xx, sx), VSHR32(yy, sy)), 14); + if (shift & 1) { + if (x2y2 < 32768) + { + x2y2 <<= 1; + shift--; + } else { + x2y2 >>= 1; + shift++; + } + } + den = celt_rsqrt_norm(x2y2); + g = MULT16_32_Q15(den, xy); + g = VSHR32(g, (shift>>1)-1); + return EXTRACT16(MIN32(g, Q15ONE)); +} +#else +static opus_val16 compute_pitch_gain(opus_val32 xy, opus_val32 xx, opus_val32 yy) +{ + return xy/celt_sqrt(1+xx*yy); +} +#endif + +static const int second_check[16] = {0, 0, 3, 2, 3, 2, 5, 2, 3, 2, 3, 2, 5, 2, 3, 2}; +opus_val16 remove_doubling(opus_val16 *x, int maxperiod, int minperiod, + int N, int *T0_, int prev_period, opus_val16 prev_gain, int arch) +{ + int k, i, T, T0; + opus_val16 g, g0; + opus_val16 pg; + opus_val32 xy,xx,yy,xy2; + opus_val32 xcorr[3]; + opus_val32 best_xy, best_yy; + int offset; + int minperiod0; + VARDECL(opus_val32, yy_lookup); + SAVE_STACK; + + minperiod0 = minperiod; + maxperiod /= 2; + minperiod /= 2; + *T0_ /= 2; + prev_period /= 2; + N /= 2; + x += maxperiod; + if (*T0_>=maxperiod) + *T0_=maxperiod-1; + + T = T0 = *T0_; + ALLOC(yy_lookup, maxperiod+1, opus_val32); + dual_inner_prod(x, x, x-T0, N, &xx, &xy, arch); + yy_lookup[0] = xx; + yy=xx; + for (i=1;i<=maxperiod;i++) + { + yy = yy+MULT16_16(x[-i],x[-i])-MULT16_16(x[N-i],x[N-i]); + yy_lookup[i] = MAX32(0, yy); + } + yy = yy_lookup[T0]; + best_xy = xy; + best_yy = yy; + g = g0 = compute_pitch_gain(xy, xx, yy); + /* Look for any pitch at T/k */ + for (k=2;k<=15;k++) + { + int T1, T1b; + opus_val16 g1; + opus_val16 cont=0; + opus_val16 thresh; + T1 = celt_udiv(2*T0+k, 2*k); + if (T1 < minperiod) + break; + /* Look for another strong correlation at T1b */ + if (k==2) + { + if (T1+T0>maxperiod) + T1b = T0; + else + T1b = T0+T1; + } else + { + T1b = celt_udiv(2*second_check[k]*T0+k, 2*k); + } + dual_inner_prod(x, &x[-T1], &x[-T1b], N, &xy, &xy2, arch); + xy = HALF32(xy + xy2); + yy = HALF32(yy_lookup[T1] + yy_lookup[T1b]); + g1 = compute_pitch_gain(xy, xx, yy); + if (abs(T1-prev_period)<=1) + cont = prev_gain; + else if (abs(T1-prev_period)<=2 && 5*k*k < T0) + cont = HALF16(prev_gain); + else + cont = 0; + thresh = MAX16(QCONST16(.3f,15), MULT16_16_Q15(QCONST16(.7f,15),g0)-cont); + /* Bias against very high pitch (very short period) to avoid false-positives + due to short-term correlation */ + if (T1<3*minperiod) + thresh = MAX16(QCONST16(.4f,15), MULT16_16_Q15(QCONST16(.85f,15),g0)-cont); + else if (T1<2*minperiod) + thresh = MAX16(QCONST16(.5f,15), MULT16_16_Q15(QCONST16(.9f,15),g0)-cont); + if (g1 > thresh) + { + best_xy = xy; + best_yy = yy; + T = T1; + g = g1; + } + } + best_xy = MAX32(0, best_xy); + if (best_yy <= best_xy) + pg = Q15ONE; + else + pg = SHR32(frac_div32(best_xy,best_yy+1),16); + + for (k=0;k<3;k++) + xcorr[k] = celt_inner_prod(x, x-(T+k-1), N, arch); + if ((xcorr[2]-xcorr[0]) > MULT16_32_Q15(QCONST16(.7f,15),xcorr[1]-xcorr[0])) + offset = 1; + else if ((xcorr[0]-xcorr[2]) > MULT16_32_Q15(QCONST16(.7f,15),xcorr[1]-xcorr[2])) + offset = -1; + else + offset = 0; + if (pg > g) + pg = g; + *T0_ = 2*T+offset; + + if (*T0_<minperiod0) + *T0_=minperiod0; + RESTORE_STACK; + return pg; +} diff --git a/media/libopus/celt/pitch.h b/media/libopus/celt/pitch.h new file mode 100644 index 0000000000..e425f56aea --- /dev/null +++ b/media/libopus/celt/pitch.h @@ -0,0 +1,192 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/** + @file pitch.h + @brief Pitch analysis + */ + +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef PITCH_H +#define PITCH_H + +#include "modes.h" +#include "cpu_support.h" + +#if (defined(OPUS_X86_MAY_HAVE_SSE) && !defined(FIXED_POINT)) \ + || ((defined(OPUS_X86_MAY_HAVE_SSE4_1) || defined(OPUS_X86_MAY_HAVE_SSE2)) && defined(FIXED_POINT)) +#include "x86/pitch_sse.h" +#endif + +#if defined(MIPSr1_ASM) +#include "mips/pitch_mipsr1.h" +#endif + +#if (defined(OPUS_ARM_ASM) || defined(OPUS_ARM_MAY_HAVE_NEON_INTR)) +# include "arm/pitch_arm.h" +#endif + +void pitch_downsample(celt_sig * OPUS_RESTRICT x[], opus_val16 * OPUS_RESTRICT x_lp, + int len, int C, int arch); + +void pitch_search(const opus_val16 * OPUS_RESTRICT x_lp, opus_val16 * OPUS_RESTRICT y, + int len, int max_pitch, int *pitch, int arch); + +opus_val16 remove_doubling(opus_val16 *x, int maxperiod, int minperiod, + int N, int *T0, int prev_period, opus_val16 prev_gain, int arch); + + +/* OPT: This is the kernel you really want to optimize. It gets used a lot + by the prefilter and by the PLC. */ +static OPUS_INLINE void xcorr_kernel_c(const opus_val16 * x, const opus_val16 * y, opus_val32 sum[4], int len) +{ + int j; + opus_val16 y_0, y_1, y_2, y_3; + celt_assert(len>=3); + y_3=0; /* gcc doesn't realize that y_3 can't be used uninitialized */ + y_0=*y++; + y_1=*y++; + y_2=*y++; + for (j=0;j<len-3;j+=4) + { + opus_val16 tmp; + tmp = *x++; + y_3=*y++; + sum[0] = MAC16_16(sum[0],tmp,y_0); + sum[1] = MAC16_16(sum[1],tmp,y_1); + sum[2] = MAC16_16(sum[2],tmp,y_2); + sum[3] = MAC16_16(sum[3],tmp,y_3); + tmp=*x++; + y_0=*y++; + sum[0] = MAC16_16(sum[0],tmp,y_1); + sum[1] = MAC16_16(sum[1],tmp,y_2); + sum[2] = MAC16_16(sum[2],tmp,y_3); + sum[3] = MAC16_16(sum[3],tmp,y_0); + tmp=*x++; + y_1=*y++; + sum[0] = MAC16_16(sum[0],tmp,y_2); + sum[1] = MAC16_16(sum[1],tmp,y_3); + sum[2] = MAC16_16(sum[2],tmp,y_0); + sum[3] = MAC16_16(sum[3],tmp,y_1); + tmp=*x++; + y_2=*y++; + sum[0] = MAC16_16(sum[0],tmp,y_3); + sum[1] = MAC16_16(sum[1],tmp,y_0); + sum[2] = MAC16_16(sum[2],tmp,y_1); + sum[3] = MAC16_16(sum[3],tmp,y_2); + } + if (j++<len) + { + opus_val16 tmp = *x++; + y_3=*y++; + sum[0] = MAC16_16(sum[0],tmp,y_0); + sum[1] = MAC16_16(sum[1],tmp,y_1); + sum[2] = MAC16_16(sum[2],tmp,y_2); + sum[3] = MAC16_16(sum[3],tmp,y_3); + } + if (j++<len) + { + opus_val16 tmp=*x++; + y_0=*y++; + sum[0] = MAC16_16(sum[0],tmp,y_1); + sum[1] = MAC16_16(sum[1],tmp,y_2); + sum[2] = MAC16_16(sum[2],tmp,y_3); + sum[3] = MAC16_16(sum[3],tmp,y_0); + } + if (j<len) + { + opus_val16 tmp=*x++; + y_1=*y++; + sum[0] = MAC16_16(sum[0],tmp,y_2); + sum[1] = MAC16_16(sum[1],tmp,y_3); + sum[2] = MAC16_16(sum[2],tmp,y_0); + sum[3] = MAC16_16(sum[3],tmp,y_1); + } +} + +#ifndef OVERRIDE_XCORR_KERNEL +#define xcorr_kernel(x, y, sum, len, arch) \ + ((void)(arch),xcorr_kernel_c(x, y, sum, len)) +#endif /* OVERRIDE_XCORR_KERNEL */ + + +static OPUS_INLINE void dual_inner_prod_c(const opus_val16 *x, const opus_val16 *y01, const opus_val16 *y02, + int N, opus_val32 *xy1, opus_val32 *xy2) +{ + int i; + opus_val32 xy01=0; + opus_val32 xy02=0; + for (i=0;i<N;i++) + { + xy01 = MAC16_16(xy01, x[i], y01[i]); + xy02 = MAC16_16(xy02, x[i], y02[i]); + } + *xy1 = xy01; + *xy2 = xy02; +} + +#ifndef OVERRIDE_DUAL_INNER_PROD +# define dual_inner_prod(x, y01, y02, N, xy1, xy2, arch) \ + ((void)(arch),dual_inner_prod_c(x, y01, y02, N, xy1, xy2)) +#endif + +/*We make sure a C version is always available for cases where the overhead of + vectorization and passing around an arch flag aren't worth it.*/ +static OPUS_INLINE opus_val32 celt_inner_prod_c(const opus_val16 *x, + const opus_val16 *y, int N) +{ + int i; + opus_val32 xy=0; + for (i=0;i<N;i++) + xy = MAC16_16(xy, x[i], y[i]); + return xy; +} + +#if !defined(OVERRIDE_CELT_INNER_PROD) +# define celt_inner_prod(x, y, N, arch) \ + ((void)(arch),celt_inner_prod_c(x, y, N)) +#endif + +#ifdef NON_STATIC_COMB_FILTER_CONST_C +void comb_filter_const_c(opus_val32 *y, opus_val32 *x, int T, int N, + opus_val16 g10, opus_val16 g11, opus_val16 g12); +#endif + + +#ifdef FIXED_POINT +opus_val32 +#else +void +#endif +celt_pitch_xcorr_c(const opus_val16 *_x, const opus_val16 *_y, + opus_val32 *xcorr, int len, int max_pitch, int arch); + +#ifndef OVERRIDE_PITCH_XCORR +# define celt_pitch_xcorr celt_pitch_xcorr_c +#endif + +#endif diff --git a/media/libopus/celt/quant_bands.c b/media/libopus/celt/quant_bands.c new file mode 100644 index 0000000000..39a221eda5 --- /dev/null +++ b/media/libopus/celt/quant_bands.c @@ -0,0 +1,563 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "quant_bands.h" +#include "laplace.h" +#include <math.h> +#include "os_support.h" +#include "arch.h" +#include "mathops.h" +#include "stack_alloc.h" +#include "rate.h" + +#ifdef FIXED_POINT +/* Mean energy in each band quantized in Q4 */ +const signed char eMeans[25] = { + 103,100, 92, 85, 81, + 77, 72, 70, 78, 75, + 73, 71, 78, 74, 69, + 72, 70, 74, 76, 71, + 60, 60, 60, 60, 60 +}; +#else +/* Mean energy in each band quantized in Q4 and converted back to float */ +const opus_val16 eMeans[25] = { + 6.437500f, 6.250000f, 5.750000f, 5.312500f, 5.062500f, + 4.812500f, 4.500000f, 4.375000f, 4.875000f, 4.687500f, + 4.562500f, 4.437500f, 4.875000f, 4.625000f, 4.312500f, + 4.500000f, 4.375000f, 4.625000f, 4.750000f, 4.437500f, + 3.750000f, 3.750000f, 3.750000f, 3.750000f, 3.750000f +}; +#endif +/* prediction coefficients: 0.9, 0.8, 0.65, 0.5 */ +#ifdef FIXED_POINT +static const opus_val16 pred_coef[4] = {29440, 26112, 21248, 16384}; +static const opus_val16 beta_coef[4] = {30147, 22282, 12124, 6554}; +static const opus_val16 beta_intra = 4915; +#else +static const opus_val16 pred_coef[4] = {29440/32768., 26112/32768., 21248/32768., 16384/32768.}; +static const opus_val16 beta_coef[4] = {30147/32768., 22282/32768., 12124/32768., 6554/32768.}; +static const opus_val16 beta_intra = 4915/32768.; +#endif + +/*Parameters of the Laplace-like probability models used for the coarse energy. + There is one pair of parameters for each frame size, prediction type + (inter/intra), and band number. + The first number of each pair is the probability of 0, and the second is the + decay rate, both in Q8 precision.*/ +static const unsigned char e_prob_model[4][2][42] = { + /*120 sample frames.*/ + { + /*Inter*/ + { + 72, 127, 65, 129, 66, 128, 65, 128, 64, 128, 62, 128, 64, 128, + 64, 128, 92, 78, 92, 79, 92, 78, 90, 79, 116, 41, 115, 40, + 114, 40, 132, 26, 132, 26, 145, 17, 161, 12, 176, 10, 177, 11 + }, + /*Intra*/ + { + 24, 179, 48, 138, 54, 135, 54, 132, 53, 134, 56, 133, 55, 132, + 55, 132, 61, 114, 70, 96, 74, 88, 75, 88, 87, 74, 89, 66, + 91, 67, 100, 59, 108, 50, 120, 40, 122, 37, 97, 43, 78, 50 + } + }, + /*240 sample frames.*/ + { + /*Inter*/ + { + 83, 78, 84, 81, 88, 75, 86, 74, 87, 71, 90, 73, 93, 74, + 93, 74, 109, 40, 114, 36, 117, 34, 117, 34, 143, 17, 145, 18, + 146, 19, 162, 12, 165, 10, 178, 7, 189, 6, 190, 8, 177, 9 + }, + /*Intra*/ + { + 23, 178, 54, 115, 63, 102, 66, 98, 69, 99, 74, 89, 71, 91, + 73, 91, 78, 89, 86, 80, 92, 66, 93, 64, 102, 59, 103, 60, + 104, 60, 117, 52, 123, 44, 138, 35, 133, 31, 97, 38, 77, 45 + } + }, + /*480 sample frames.*/ + { + /*Inter*/ + { + 61, 90, 93, 60, 105, 42, 107, 41, 110, 45, 116, 38, 113, 38, + 112, 38, 124, 26, 132, 27, 136, 19, 140, 20, 155, 14, 159, 16, + 158, 18, 170, 13, 177, 10, 187, 8, 192, 6, 175, 9, 159, 10 + }, + /*Intra*/ + { + 21, 178, 59, 110, 71, 86, 75, 85, 84, 83, 91, 66, 88, 73, + 87, 72, 92, 75, 98, 72, 105, 58, 107, 54, 115, 52, 114, 55, + 112, 56, 129, 51, 132, 40, 150, 33, 140, 29, 98, 35, 77, 42 + } + }, + /*960 sample frames.*/ + { + /*Inter*/ + { + 42, 121, 96, 66, 108, 43, 111, 40, 117, 44, 123, 32, 120, 36, + 119, 33, 127, 33, 134, 34, 139, 21, 147, 23, 152, 20, 158, 25, + 154, 26, 166, 21, 173, 16, 184, 13, 184, 10, 150, 13, 139, 15 + }, + /*Intra*/ + { + 22, 178, 63, 114, 74, 82, 84, 83, 92, 82, 103, 62, 96, 72, + 96, 67, 101, 73, 107, 72, 113, 55, 118, 52, 125, 52, 118, 52, + 117, 55, 135, 49, 137, 39, 157, 32, 145, 29, 97, 33, 77, 40 + } + } +}; + +static const unsigned char small_energy_icdf[3]={2,1,0}; + +static opus_val32 loss_distortion(const opus_val16 *eBands, opus_val16 *oldEBands, int start, int end, int len, int C) +{ + int c, i; + opus_val32 dist = 0; + c=0; do { + for (i=start;i<end;i++) + { + opus_val16 d = SUB16(SHR16(eBands[i+c*len], 3), SHR16(oldEBands[i+c*len], 3)); + dist = MAC16_16(dist, d,d); + } + } while (++c<C); + return MIN32(200,SHR32(dist,2*DB_SHIFT-6)); +} + +static int quant_coarse_energy_impl(const CELTMode *m, int start, int end, + const opus_val16 *eBands, opus_val16 *oldEBands, + opus_int32 budget, opus_int32 tell, + const unsigned char *prob_model, opus_val16 *error, ec_enc *enc, + int C, int LM, int intra, opus_val16 max_decay, int lfe) +{ + int i, c; + int badness = 0; + opus_val32 prev[2] = {0,0}; + opus_val16 coef; + opus_val16 beta; + + if (tell+3 <= budget) + ec_enc_bit_logp(enc, intra, 3); + if (intra) + { + coef = 0; + beta = beta_intra; + } else { + beta = beta_coef[LM]; + coef = pred_coef[LM]; + } + + /* Encode at a fixed coarse resolution */ + for (i=start;i<end;i++) + { + c=0; + do { + int bits_left; + int qi, qi0; + opus_val32 q; + opus_val16 x; + opus_val32 f, tmp; + opus_val16 oldE; + opus_val16 decay_bound; + x = eBands[i+c*m->nbEBands]; + oldE = MAX16(-QCONST16(9.f,DB_SHIFT), oldEBands[i+c*m->nbEBands]); +#ifdef FIXED_POINT + f = SHL32(EXTEND32(x),7) - PSHR32(MULT16_16(coef,oldE), 8) - prev[c]; + /* Rounding to nearest integer here is really important! */ + qi = (f+QCONST32(.5f,DB_SHIFT+7))>>(DB_SHIFT+7); + decay_bound = EXTRACT16(MAX32(-QCONST16(28.f,DB_SHIFT), + SUB32((opus_val32)oldEBands[i+c*m->nbEBands],max_decay))); +#else + f = x-coef*oldE-prev[c]; + /* Rounding to nearest integer here is really important! */ + qi = (int)floor(.5f+f); + decay_bound = MAX16(-QCONST16(28.f,DB_SHIFT), oldEBands[i+c*m->nbEBands]) - max_decay; +#endif + /* Prevent the energy from going down too quickly (e.g. for bands + that have just one bin) */ + if (qi < 0 && x < decay_bound) + { + qi += (int)SHR16(SUB16(decay_bound,x), DB_SHIFT); + if (qi > 0) + qi = 0; + } + qi0 = qi; + /* If we don't have enough bits to encode all the energy, just assume + something safe. */ + tell = ec_tell(enc); + bits_left = budget-tell-3*C*(end-i); + if (i!=start && bits_left < 30) + { + if (bits_left < 24) + qi = IMIN(1, qi); + if (bits_left < 16) + qi = IMAX(-1, qi); + } + if (lfe && i>=2) + qi = IMIN(qi, 0); + if (budget-tell >= 15) + { + int pi; + pi = 2*IMIN(i,20); + ec_laplace_encode(enc, &qi, + prob_model[pi]<<7, prob_model[pi+1]<<6); + } + else if(budget-tell >= 2) + { + qi = IMAX(-1, IMIN(qi, 1)); + ec_enc_icdf(enc, 2*qi^-(qi<0), small_energy_icdf, 2); + } + else if(budget-tell >= 1) + { + qi = IMIN(0, qi); + ec_enc_bit_logp(enc, -qi, 1); + } + else + qi = -1; + error[i+c*m->nbEBands] = PSHR32(f,7) - SHL16(qi,DB_SHIFT); + badness += abs(qi0-qi); + q = (opus_val32)SHL32(EXTEND32(qi),DB_SHIFT); + + tmp = PSHR32(MULT16_16(coef,oldE),8) + prev[c] + SHL32(q,7); +#ifdef FIXED_POINT + tmp = MAX32(-QCONST32(28.f, DB_SHIFT+7), tmp); +#endif + oldEBands[i+c*m->nbEBands] = PSHR32(tmp, 7); + prev[c] = prev[c] + SHL32(q,7) - MULT16_16(beta,PSHR32(q,8)); + } while (++c < C); + } + return lfe ? 0 : badness; +} + +void quant_coarse_energy(const CELTMode *m, int start, int end, int effEnd, + const opus_val16 *eBands, opus_val16 *oldEBands, opus_uint32 budget, + opus_val16 *error, ec_enc *enc, int C, int LM, int nbAvailableBytes, + int force_intra, opus_val32 *delayedIntra, int two_pass, int loss_rate, int lfe) +{ + int intra; + opus_val16 max_decay; + VARDECL(opus_val16, oldEBands_intra); + VARDECL(opus_val16, error_intra); + ec_enc enc_start_state; + opus_uint32 tell; + int badness1=0; + opus_int32 intra_bias; + opus_val32 new_distortion; + SAVE_STACK; + + intra = force_intra || (!two_pass && *delayedIntra>2*C*(end-start) && nbAvailableBytes > (end-start)*C); + intra_bias = (opus_int32)((budget**delayedIntra*loss_rate)/(C*512)); + new_distortion = loss_distortion(eBands, oldEBands, start, effEnd, m->nbEBands, C); + + tell = ec_tell(enc); + if (tell+3 > budget) + two_pass = intra = 0; + + max_decay = QCONST16(16.f,DB_SHIFT); + if (end-start>10) + { +#ifdef FIXED_POINT + max_decay = MIN32(max_decay, SHL32(EXTEND32(nbAvailableBytes),DB_SHIFT-3)); +#else + max_decay = MIN32(max_decay, .125f*nbAvailableBytes); +#endif + } + if (lfe) + max_decay = QCONST16(3.f,DB_SHIFT); + enc_start_state = *enc; + + ALLOC(oldEBands_intra, C*m->nbEBands, opus_val16); + ALLOC(error_intra, C*m->nbEBands, opus_val16); + OPUS_COPY(oldEBands_intra, oldEBands, C*m->nbEBands); + + if (two_pass || intra) + { + badness1 = quant_coarse_energy_impl(m, start, end, eBands, oldEBands_intra, budget, + tell, e_prob_model[LM][1], error_intra, enc, C, LM, 1, max_decay, lfe); + } + + if (!intra) + { + unsigned char *intra_buf; + ec_enc enc_intra_state; + opus_int32 tell_intra; + opus_uint32 nstart_bytes; + opus_uint32 nintra_bytes; + opus_uint32 save_bytes; + int badness2; + VARDECL(unsigned char, intra_bits); + + tell_intra = ec_tell_frac(enc); + + enc_intra_state = *enc; + + nstart_bytes = ec_range_bytes(&enc_start_state); + nintra_bytes = ec_range_bytes(&enc_intra_state); + intra_buf = ec_get_buffer(&enc_intra_state) + nstart_bytes; + save_bytes = nintra_bytes-nstart_bytes; + if (save_bytes == 0) + save_bytes = ALLOC_NONE; + ALLOC(intra_bits, save_bytes, unsigned char); + /* Copy bits from intra bit-stream */ + OPUS_COPY(intra_bits, intra_buf, nintra_bytes - nstart_bytes); + + *enc = enc_start_state; + + badness2 = quant_coarse_energy_impl(m, start, end, eBands, oldEBands, budget, + tell, e_prob_model[LM][intra], error, enc, C, LM, 0, max_decay, lfe); + + if (two_pass && (badness1 < badness2 || (badness1 == badness2 && ((opus_int32)ec_tell_frac(enc))+intra_bias > tell_intra))) + { + *enc = enc_intra_state; + /* Copy intra bits to bit-stream */ + OPUS_COPY(intra_buf, intra_bits, nintra_bytes - nstart_bytes); + OPUS_COPY(oldEBands, oldEBands_intra, C*m->nbEBands); + OPUS_COPY(error, error_intra, C*m->nbEBands); + intra = 1; + } + } else { + OPUS_COPY(oldEBands, oldEBands_intra, C*m->nbEBands); + OPUS_COPY(error, error_intra, C*m->nbEBands); + } + + if (intra) + *delayedIntra = new_distortion; + else + *delayedIntra = ADD32(MULT16_32_Q15(MULT16_16_Q15(pred_coef[LM], pred_coef[LM]),*delayedIntra), + new_distortion); + + RESTORE_STACK; +} + +void quant_fine_energy(const CELTMode *m, int start, int end, opus_val16 *oldEBands, opus_val16 *error, int *fine_quant, ec_enc *enc, int C) +{ + int i, c; + + /* Encode finer resolution */ + for (i=start;i<end;i++) + { + opus_int16 frac = 1<<fine_quant[i]; + if (fine_quant[i] <= 0) + continue; + c=0; + do { + int q2; + opus_val16 offset; +#ifdef FIXED_POINT + /* Has to be without rounding */ + q2 = (error[i+c*m->nbEBands]+QCONST16(.5f,DB_SHIFT))>>(DB_SHIFT-fine_quant[i]); +#else + q2 = (int)floor((error[i+c*m->nbEBands]+.5f)*frac); +#endif + if (q2 > frac-1) + q2 = frac-1; + if (q2<0) + q2 = 0; + ec_enc_bits(enc, q2, fine_quant[i]); +#ifdef FIXED_POINT + offset = SUB16(SHR32(SHL32(EXTEND32(q2),DB_SHIFT)+QCONST16(.5f,DB_SHIFT),fine_quant[i]),QCONST16(.5f,DB_SHIFT)); +#else + offset = (q2+.5f)*(1<<(14-fine_quant[i]))*(1.f/16384) - .5f; +#endif + oldEBands[i+c*m->nbEBands] += offset; + error[i+c*m->nbEBands] -= offset; + /*printf ("%f ", error[i] - offset);*/ + } while (++c < C); + } +} + +void quant_energy_finalise(const CELTMode *m, int start, int end, opus_val16 *oldEBands, opus_val16 *error, int *fine_quant, int *fine_priority, int bits_left, ec_enc *enc, int C) +{ + int i, prio, c; + + /* Use up the remaining bits */ + for (prio=0;prio<2;prio++) + { + for (i=start;i<end && bits_left>=C ;i++) + { + if (fine_quant[i] >= MAX_FINE_BITS || fine_priority[i]!=prio) + continue; + c=0; + do { + int q2; + opus_val16 offset; + q2 = error[i+c*m->nbEBands]<0 ? 0 : 1; + ec_enc_bits(enc, q2, 1); +#ifdef FIXED_POINT + offset = SHR16(SHL16(q2,DB_SHIFT)-QCONST16(.5f,DB_SHIFT),fine_quant[i]+1); +#else + offset = (q2-.5f)*(1<<(14-fine_quant[i]-1))*(1.f/16384); +#endif + oldEBands[i+c*m->nbEBands] += offset; + error[i+c*m->nbEBands] -= offset; + bits_left--; + } while (++c < C); + } + } +} + +void unquant_coarse_energy(const CELTMode *m, int start, int end, opus_val16 *oldEBands, int intra, ec_dec *dec, int C, int LM) +{ + const unsigned char *prob_model = e_prob_model[LM][intra]; + int i, c; + opus_val32 prev[2] = {0, 0}; + opus_val16 coef; + opus_val16 beta; + opus_int32 budget; + opus_int32 tell; + + if (intra) + { + coef = 0; + beta = beta_intra; + } else { + beta = beta_coef[LM]; + coef = pred_coef[LM]; + } + + budget = dec->storage*8; + + /* Decode at a fixed coarse resolution */ + for (i=start;i<end;i++) + { + c=0; + do { + int qi; + opus_val32 q; + opus_val32 tmp; + /* It would be better to express this invariant as a + test on C at function entry, but that isn't enough + to make the static analyzer happy. */ + celt_sig_assert(c<2); + tell = ec_tell(dec); + if(budget-tell>=15) + { + int pi; + pi = 2*IMIN(i,20); + qi = ec_laplace_decode(dec, + prob_model[pi]<<7, prob_model[pi+1]<<6); + } + else if(budget-tell>=2) + { + qi = ec_dec_icdf(dec, small_energy_icdf, 2); + qi = (qi>>1)^-(qi&1); + } + else if(budget-tell>=1) + { + qi = -ec_dec_bit_logp(dec, 1); + } + else + qi = -1; + q = (opus_val32)SHL32(EXTEND32(qi),DB_SHIFT); + + oldEBands[i+c*m->nbEBands] = MAX16(-QCONST16(9.f,DB_SHIFT), oldEBands[i+c*m->nbEBands]); + tmp = PSHR32(MULT16_16(coef,oldEBands[i+c*m->nbEBands]),8) + prev[c] + SHL32(q,7); +#ifdef FIXED_POINT + tmp = MAX32(-QCONST32(28.f, DB_SHIFT+7), tmp); +#endif + oldEBands[i+c*m->nbEBands] = PSHR32(tmp, 7); + prev[c] = prev[c] + SHL32(q,7) - MULT16_16(beta,PSHR32(q,8)); + } while (++c < C); + } +} + +void unquant_fine_energy(const CELTMode *m, int start, int end, opus_val16 *oldEBands, int *fine_quant, ec_dec *dec, int C) +{ + int i, c; + /* Decode finer resolution */ + for (i=start;i<end;i++) + { + if (fine_quant[i] <= 0) + continue; + c=0; + do { + int q2; + opus_val16 offset; + q2 = ec_dec_bits(dec, fine_quant[i]); +#ifdef FIXED_POINT + offset = SUB16(SHR32(SHL32(EXTEND32(q2),DB_SHIFT)+QCONST16(.5f,DB_SHIFT),fine_quant[i]),QCONST16(.5f,DB_SHIFT)); +#else + offset = (q2+.5f)*(1<<(14-fine_quant[i]))*(1.f/16384) - .5f; +#endif + oldEBands[i+c*m->nbEBands] += offset; + } while (++c < C); + } +} + +void unquant_energy_finalise(const CELTMode *m, int start, int end, opus_val16 *oldEBands, int *fine_quant, int *fine_priority, int bits_left, ec_dec *dec, int C) +{ + int i, prio, c; + + /* Use up the remaining bits */ + for (prio=0;prio<2;prio++) + { + for (i=start;i<end && bits_left>=C ;i++) + { + if (fine_quant[i] >= MAX_FINE_BITS || fine_priority[i]!=prio) + continue; + c=0; + do { + int q2; + opus_val16 offset; + q2 = ec_dec_bits(dec, 1); +#ifdef FIXED_POINT + offset = SHR16(SHL16(q2,DB_SHIFT)-QCONST16(.5f,DB_SHIFT),fine_quant[i]+1); +#else + offset = (q2-.5f)*(1<<(14-fine_quant[i]-1))*(1.f/16384); +#endif + oldEBands[i+c*m->nbEBands] += offset; + bits_left--; + } while (++c < C); + } + } +} + +void amp2Log2(const CELTMode *m, int effEnd, int end, + celt_ener *bandE, opus_val16 *bandLogE, int C) +{ + int c, i; + c=0; + do { + for (i=0;i<effEnd;i++) + { + bandLogE[i+c*m->nbEBands] = + celt_log2(bandE[i+c*m->nbEBands]) + - SHL16((opus_val16)eMeans[i],6); +#ifdef FIXED_POINT + /* Compensate for bandE[] being Q12 but celt_log2() taking a Q14 input. */ + bandLogE[i+c*m->nbEBands] += QCONST16(2.f, DB_SHIFT); +#endif + } + for (i=effEnd;i<end;i++) + bandLogE[c*m->nbEBands+i] = -QCONST16(14.f,DB_SHIFT); + } while (++c < C); +} diff --git a/media/libopus/celt/quant_bands.h b/media/libopus/celt/quant_bands.h new file mode 100644 index 0000000000..0490bca4b4 --- /dev/null +++ b/media/libopus/celt/quant_bands.h @@ -0,0 +1,66 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef QUANT_BANDS +#define QUANT_BANDS + +#include "arch.h" +#include "modes.h" +#include "entenc.h" +#include "entdec.h" +#include "mathops.h" + +#ifdef FIXED_POINT +extern const signed char eMeans[25]; +#else +extern const opus_val16 eMeans[25]; +#endif + +void amp2Log2(const CELTMode *m, int effEnd, int end, + celt_ener *bandE, opus_val16 *bandLogE, int C); + +void log2Amp(const CELTMode *m, int start, int end, + celt_ener *eBands, const opus_val16 *oldEBands, int C); + +void quant_coarse_energy(const CELTMode *m, int start, int end, int effEnd, + const opus_val16 *eBands, opus_val16 *oldEBands, opus_uint32 budget, + opus_val16 *error, ec_enc *enc, int C, int LM, + int nbAvailableBytes, int force_intra, opus_val32 *delayedIntra, + int two_pass, int loss_rate, int lfe); + +void quant_fine_energy(const CELTMode *m, int start, int end, opus_val16 *oldEBands, opus_val16 *error, int *fine_quant, ec_enc *enc, int C); + +void quant_energy_finalise(const CELTMode *m, int start, int end, opus_val16 *oldEBands, opus_val16 *error, int *fine_quant, int *fine_priority, int bits_left, ec_enc *enc, int C); + +void unquant_coarse_energy(const CELTMode *m, int start, int end, opus_val16 *oldEBands, int intra, ec_dec *dec, int C, int LM); + +void unquant_fine_energy(const CELTMode *m, int start, int end, opus_val16 *oldEBands, int *fine_quant, ec_dec *dec, int C); + +void unquant_energy_finalise(const CELTMode *m, int start, int end, opus_val16 *oldEBands, int *fine_quant, int *fine_priority, int bits_left, ec_dec *dec, int C); + +#endif /* QUANT_BANDS */ diff --git a/media/libopus/celt/rate.c b/media/libopus/celt/rate.c new file mode 100644 index 0000000000..7f7ad3fa0f --- /dev/null +++ b/media/libopus/celt/rate.c @@ -0,0 +1,646 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include <math.h> +#include "modes.h" +#include "cwrs.h" +#include "arch.h" +#include "os_support.h" + +#include "entcode.h" +#include "rate.h" + +static const unsigned char LOG2_FRAC_TABLE[24]={ + 0, + 8,13, + 16,19,21,23, + 24,26,27,28,29,30,31,32, + 32,33,34,34,35,36,36,37,37 +}; + +#ifdef CUSTOM_MODES + +/*Determines if V(N,K) fits in a 32-bit unsigned integer. + N and K are themselves limited to 15 bits.*/ +static int fits_in32(int _n, int _k) +{ + static const opus_int16 maxN[15] = { + 32767, 32767, 32767, 1476, 283, 109, 60, 40, + 29, 24, 20, 18, 16, 14, 13}; + static const opus_int16 maxK[15] = { + 32767, 32767, 32767, 32767, 1172, 238, 95, 53, + 36, 27, 22, 18, 16, 15, 13}; + if (_n>=14) + { + if (_k>=14) + return 0; + else + return _n <= maxN[_k]; + } else { + return _k <= maxK[_n]; + } +} + +void compute_pulse_cache(CELTMode *m, int LM) +{ + int C; + int i; + int j; + int curr=0; + int nbEntries=0; + int entryN[100], entryK[100], entryI[100]; + const opus_int16 *eBands = m->eBands; + PulseCache *cache = &m->cache; + opus_int16 *cindex; + unsigned char *bits; + unsigned char *cap; + + cindex = (opus_int16 *)opus_alloc(sizeof(cache->index[0])*m->nbEBands*(LM+2)); + cache->index = cindex; + + /* Scan for all unique band sizes */ + for (i=0;i<=LM+1;i++) + { + for (j=0;j<m->nbEBands;j++) + { + int k; + int N = (eBands[j+1]-eBands[j])<<i>>1; + cindex[i*m->nbEBands+j] = -1; + /* Find other bands that have the same size */ + for (k=0;k<=i;k++) + { + int n; + for (n=0;n<m->nbEBands && (k!=i || n<j);n++) + { + if (N == (eBands[n+1]-eBands[n])<<k>>1) + { + cindex[i*m->nbEBands+j] = cindex[k*m->nbEBands+n]; + break; + } + } + } + if (cache->index[i*m->nbEBands+j] == -1 && N!=0) + { + int K; + entryN[nbEntries] = N; + K = 0; + while (fits_in32(N,get_pulses(K+1)) && K<MAX_PSEUDO) + K++; + entryK[nbEntries] = K; + cindex[i*m->nbEBands+j] = curr; + entryI[nbEntries] = curr; + + curr += K+1; + nbEntries++; + } + } + } + bits = (unsigned char *)opus_alloc(sizeof(unsigned char)*curr); + cache->bits = bits; + cache->size = curr; + /* Compute the cache for all unique sizes */ + for (i=0;i<nbEntries;i++) + { + unsigned char *ptr = bits+entryI[i]; + opus_int16 tmp[CELT_MAX_PULSES+1]; + get_required_bits(tmp, entryN[i], get_pulses(entryK[i]), BITRES); + for (j=1;j<=entryK[i];j++) + ptr[j] = tmp[get_pulses(j)]-1; + ptr[0] = entryK[i]; + } + + /* Compute the maximum rate for each band at which we'll reliably use as + many bits as we ask for. */ + cache->caps = cap = (unsigned char *)opus_alloc(sizeof(cache->caps[0])*(LM+1)*2*m->nbEBands); + for (i=0;i<=LM;i++) + { + for (C=1;C<=2;C++) + { + for (j=0;j<m->nbEBands;j++) + { + int N0; + int max_bits; + N0 = m->eBands[j+1]-m->eBands[j]; + /* N=1 bands only have a sign bit and fine bits. */ + if (N0<<i == 1) + max_bits = C*(1+MAX_FINE_BITS)<<BITRES; + else + { + const unsigned char *pcache; + opus_int32 num; + opus_int32 den; + int LM0; + int N; + int offset; + int ndof; + int qb; + int k; + LM0 = 0; + /* Even-sized bands bigger than N=2 can be split one more time. + As of commit 44203907 all bands >1 are even, including custom modes.*/ + if (N0 > 2) + { + N0>>=1; + LM0--; + } + /* N0=1 bands can't be split down to N<2. */ + else if (N0 <= 1) + { + LM0=IMIN(i,1); + N0<<=LM0; + } + /* Compute the cost for the lowest-level PVQ of a fully split + band. */ + pcache = bits + cindex[(LM0+1)*m->nbEBands+j]; + max_bits = pcache[pcache[0]]+1; + /* Add in the cost of coding regular splits. */ + N = N0; + for(k=0;k<i-LM0;k++){ + max_bits <<= 1; + /* Offset the number of qtheta bits by log2(N)/2 + + QTHETA_OFFSET compared to their "fair share" of + total/N */ + offset = ((m->logN[j]+((LM0+k)<<BITRES))>>1)-QTHETA_OFFSET; + /* The number of qtheta bits we'll allocate if the remainder + is to be max_bits. + The average measured cost for theta is 0.89701 times qb, + approximated here as 459/512. */ + num=459*(opus_int32)((2*N-1)*offset+max_bits); + den=((opus_int32)(2*N-1)<<9)-459; + qb = IMIN((num+(den>>1))/den, 57); + celt_assert(qb >= 0); + max_bits += qb; + N <<= 1; + } + /* Add in the cost of a stereo split, if necessary. */ + if (C==2) + { + max_bits <<= 1; + offset = ((m->logN[j]+(i<<BITRES))>>1)-(N==2?QTHETA_OFFSET_TWOPHASE:QTHETA_OFFSET); + ndof = 2*N-1-(N==2); + /* The average measured cost for theta with the step PDF is + 0.95164 times qb, approximated here as 487/512. */ + num = (N==2?512:487)*(opus_int32)(max_bits+ndof*offset); + den = ((opus_int32)ndof<<9)-(N==2?512:487); + qb = IMIN((num+(den>>1))/den, (N==2?64:61)); + celt_assert(qb >= 0); + max_bits += qb; + } + /* Add the fine bits we'll use. */ + /* Compensate for the extra DoF in stereo */ + ndof = C*N + ((C==2 && N>2) ? 1 : 0); + /* Offset the number of fine bits by log2(N)/2 + FINE_OFFSET + compared to their "fair share" of total/N */ + offset = ((m->logN[j] + (i<<BITRES))>>1)-FINE_OFFSET; + /* N=2 is the only point that doesn't match the curve */ + if (N==2) + offset += 1<<BITRES>>2; + /* The number of fine bits we'll allocate if the remainder is + to be max_bits. */ + num = max_bits+ndof*offset; + den = (ndof-1)<<BITRES; + qb = IMIN((num+(den>>1))/den, MAX_FINE_BITS); + celt_assert(qb >= 0); + max_bits += C*qb<<BITRES; + } + max_bits = (4*max_bits/(C*((m->eBands[j+1]-m->eBands[j])<<i)))-64; + celt_assert(max_bits >= 0); + celt_assert(max_bits < 256); + *cap++ = (unsigned char)max_bits; + } + } + } +} + +#endif /* CUSTOM_MODES */ + +#define ALLOC_STEPS 6 + +static OPUS_INLINE int interp_bits2pulses(const CELTMode *m, int start, int end, int skip_start, + const int *bits1, const int *bits2, const int *thresh, const int *cap, opus_int32 total, opus_int32 *_balance, + int skip_rsv, int *intensity, int intensity_rsv, int *dual_stereo, int dual_stereo_rsv, int *bits, + int *ebits, int *fine_priority, int C, int LM, ec_ctx *ec, int encode, int prev, int signalBandwidth) +{ + opus_int32 psum; + int lo, hi; + int i, j; + int logM; + int stereo; + int codedBands=-1; + int alloc_floor; + opus_int32 left, percoeff; + int done; + opus_int32 balance; + SAVE_STACK; + + alloc_floor = C<<BITRES; + stereo = C>1; + + logM = LM<<BITRES; + lo = 0; + hi = 1<<ALLOC_STEPS; + for (i=0;i<ALLOC_STEPS;i++) + { + int mid = (lo+hi)>>1; + psum = 0; + done = 0; + for (j=end;j-->start;) + { + int tmp = bits1[j] + (mid*(opus_int32)bits2[j]>>ALLOC_STEPS); + if (tmp >= thresh[j] || done) + { + done = 1; + /* Don't allocate more than we can actually use */ + psum += IMIN(tmp, cap[j]); + } else { + if (tmp >= alloc_floor) + psum += alloc_floor; + } + } + if (psum > total) + hi = mid; + else + lo = mid; + } + psum = 0; + /*printf ("interp bisection gave %d\n", lo);*/ + done = 0; + for (j=end;j-->start;) + { + int tmp = bits1[j] + ((opus_int32)lo*bits2[j]>>ALLOC_STEPS); + if (tmp < thresh[j] && !done) + { + if (tmp >= alloc_floor) + tmp = alloc_floor; + else + tmp = 0; + } else + done = 1; + /* Don't allocate more than we can actually use */ + tmp = IMIN(tmp, cap[j]); + bits[j] = tmp; + psum += tmp; + } + + /* Decide which bands to skip, working backwards from the end. */ + for (codedBands=end;;codedBands--) + { + int band_width; + int band_bits; + int rem; + j = codedBands-1; + /* Never skip the first band, nor a band that has been boosted by + dynalloc. + In the first case, we'd be coding a bit to signal we're going to waste + all the other bits. + In the second case, we'd be coding a bit to redistribute all the bits + we just signaled should be cocentrated in this band. */ + if (j<=skip_start) + { + /* Give the bit we reserved to end skipping back. */ + total += skip_rsv; + break; + } + /*Figure out how many left-over bits we would be adding to this band. + This can include bits we've stolen back from higher, skipped bands.*/ + left = total-psum; + percoeff = celt_udiv(left, m->eBands[codedBands]-m->eBands[start]); + left -= (m->eBands[codedBands]-m->eBands[start])*percoeff; + rem = IMAX(left-(m->eBands[j]-m->eBands[start]),0); + band_width = m->eBands[codedBands]-m->eBands[j]; + band_bits = (int)(bits[j] + percoeff*band_width + rem); + /*Only code a skip decision if we're above the threshold for this band. + Otherwise it is force-skipped. + This ensures that we have enough bits to code the skip flag.*/ + if (band_bits >= IMAX(thresh[j], alloc_floor+(1<<BITRES))) + { + if (encode) + { + /*This if() block is the only part of the allocation function that + is not a mandatory part of the bitstream: any bands we choose to + skip here must be explicitly signaled.*/ + int depth_threshold; + /*We choose a threshold with some hysteresis to keep bands from + fluctuating in and out, but we try not to fold below a certain point. */ + if (codedBands > 17) + depth_threshold = j<prev ? 7 : 9; + else + depth_threshold = 0; +#ifdef FUZZING + (void)signalBandwidth; + (void)depth_threshold; + if ((rand()&0x1) == 0) +#else + if (codedBands<=start+2 || (band_bits > (depth_threshold*band_width<<LM<<BITRES)>>4 && j<=signalBandwidth)) +#endif + { + ec_enc_bit_logp(ec, 1, 1); + break; + } + ec_enc_bit_logp(ec, 0, 1); + } else if (ec_dec_bit_logp(ec, 1)) { + break; + } + /*We used a bit to skip this band.*/ + psum += 1<<BITRES; + band_bits -= 1<<BITRES; + } + /*Reclaim the bits originally allocated to this band.*/ + psum -= bits[j]+intensity_rsv; + if (intensity_rsv > 0) + intensity_rsv = LOG2_FRAC_TABLE[j-start]; + psum += intensity_rsv; + if (band_bits >= alloc_floor) + { + /*If we have enough for a fine energy bit per channel, use it.*/ + psum += alloc_floor; + bits[j] = alloc_floor; + } else { + /*Otherwise this band gets nothing at all.*/ + bits[j] = 0; + } + } + + celt_assert(codedBands > start); + /* Code the intensity and dual stereo parameters. */ + if (intensity_rsv > 0) + { + if (encode) + { + *intensity = IMIN(*intensity, codedBands); + ec_enc_uint(ec, *intensity-start, codedBands+1-start); + } + else + *intensity = start+ec_dec_uint(ec, codedBands+1-start); + } + else + *intensity = 0; + if (*intensity <= start) + { + total += dual_stereo_rsv; + dual_stereo_rsv = 0; + } + if (dual_stereo_rsv > 0) + { + if (encode) + ec_enc_bit_logp(ec, *dual_stereo, 1); + else + *dual_stereo = ec_dec_bit_logp(ec, 1); + } + else + *dual_stereo = 0; + + /* Allocate the remaining bits */ + left = total-psum; + percoeff = celt_udiv(left, m->eBands[codedBands]-m->eBands[start]); + left -= (m->eBands[codedBands]-m->eBands[start])*percoeff; + for (j=start;j<codedBands;j++) + bits[j] += ((int)percoeff*(m->eBands[j+1]-m->eBands[j])); + for (j=start;j<codedBands;j++) + { + int tmp = (int)IMIN(left, m->eBands[j+1]-m->eBands[j]); + bits[j] += tmp; + left -= tmp; + } + /*for (j=0;j<end;j++)printf("%d ", bits[j]);printf("\n");*/ + + balance = 0; + for (j=start;j<codedBands;j++) + { + int N0, N, den; + int offset; + int NClogN; + opus_int32 excess, bit; + + celt_assert(bits[j] >= 0); + N0 = m->eBands[j+1]-m->eBands[j]; + N=N0<<LM; + bit = (opus_int32)bits[j]+balance; + + if (N>1) + { + excess = MAX32(bit-cap[j],0); + bits[j] = bit-excess; + + /* Compensate for the extra DoF in stereo */ + den=(C*N+ ((C==2 && N>2 && !*dual_stereo && j<*intensity) ? 1 : 0)); + + NClogN = den*(m->logN[j] + logM); + + /* Offset for the number of fine bits by log2(N)/2 + FINE_OFFSET + compared to their "fair share" of total/N */ + offset = (NClogN>>1)-den*FINE_OFFSET; + + /* N=2 is the only point that doesn't match the curve */ + if (N==2) + offset += den<<BITRES>>2; + + /* Changing the offset for allocating the second and third + fine energy bit */ + if (bits[j] + offset < den*2<<BITRES) + offset += NClogN>>2; + else if (bits[j] + offset < den*3<<BITRES) + offset += NClogN>>3; + + /* Divide with rounding */ + ebits[j] = IMAX(0, (bits[j] + offset + (den<<(BITRES-1)))); + ebits[j] = celt_udiv(ebits[j], den)>>BITRES; + + /* Make sure not to bust */ + if (C*ebits[j] > (bits[j]>>BITRES)) + ebits[j] = bits[j] >> stereo >> BITRES; + + /* More than that is useless because that's about as far as PVQ can go */ + ebits[j] = IMIN(ebits[j], MAX_FINE_BITS); + + /* If we rounded down or capped this band, make it a candidate for the + final fine energy pass */ + fine_priority[j] = ebits[j]*(den<<BITRES) >= bits[j]+offset; + + /* Remove the allocated fine bits; the rest are assigned to PVQ */ + bits[j] -= C*ebits[j]<<BITRES; + + } else { + /* For N=1, all bits go to fine energy except for a single sign bit */ + excess = MAX32(0,bit-(C<<BITRES)); + bits[j] = bit-excess; + ebits[j] = 0; + fine_priority[j] = 1; + } + + /* Fine energy can't take advantage of the re-balancing in + quant_all_bands(). + Instead, do the re-balancing here.*/ + if(excess > 0) + { + int extra_fine; + int extra_bits; + extra_fine = IMIN(excess>>(stereo+BITRES),MAX_FINE_BITS-ebits[j]); + ebits[j] += extra_fine; + extra_bits = extra_fine*C<<BITRES; + fine_priority[j] = extra_bits >= excess-balance; + excess -= extra_bits; + } + balance = excess; + + celt_assert(bits[j] >= 0); + celt_assert(ebits[j] >= 0); + } + /* Save any remaining bits over the cap for the rebalancing in + quant_all_bands(). */ + *_balance = balance; + + /* The skipped bands use all their bits for fine energy. */ + for (;j<end;j++) + { + ebits[j] = bits[j] >> stereo >> BITRES; + celt_assert(C*ebits[j]<<BITRES == bits[j]); + bits[j] = 0; + fine_priority[j] = ebits[j]<1; + } + RESTORE_STACK; + return codedBands; +} + +int clt_compute_allocation(const CELTMode *m, int start, int end, const int *offsets, const int *cap, int alloc_trim, int *intensity, int *dual_stereo, + opus_int32 total, opus_int32 *balance, int *pulses, int *ebits, int *fine_priority, int C, int LM, ec_ctx *ec, int encode, int prev, int signalBandwidth) +{ + int lo, hi, len, j; + int codedBands; + int skip_start; + int skip_rsv; + int intensity_rsv; + int dual_stereo_rsv; + VARDECL(int, bits1); + VARDECL(int, bits2); + VARDECL(int, thresh); + VARDECL(int, trim_offset); + SAVE_STACK; + + total = IMAX(total, 0); + len = m->nbEBands; + skip_start = start; + /* Reserve a bit to signal the end of manually skipped bands. */ + skip_rsv = total >= 1<<BITRES ? 1<<BITRES : 0; + total -= skip_rsv; + /* Reserve bits for the intensity and dual stereo parameters. */ + intensity_rsv = dual_stereo_rsv = 0; + if (C==2) + { + intensity_rsv = LOG2_FRAC_TABLE[end-start]; + if (intensity_rsv>total) + intensity_rsv = 0; + else + { + total -= intensity_rsv; + dual_stereo_rsv = total>=1<<BITRES ? 1<<BITRES : 0; + total -= dual_stereo_rsv; + } + } + ALLOC(bits1, len, int); + ALLOC(bits2, len, int); + ALLOC(thresh, len, int); + ALLOC(trim_offset, len, int); + + for (j=start;j<end;j++) + { + /* Below this threshold, we're sure not to allocate any PVQ bits */ + thresh[j] = IMAX((C)<<BITRES, (3*(m->eBands[j+1]-m->eBands[j])<<LM<<BITRES)>>4); + /* Tilt of the allocation curve */ + trim_offset[j] = C*(m->eBands[j+1]-m->eBands[j])*(alloc_trim-5-LM)*(end-j-1) + *(1<<(LM+BITRES))>>6; + /* Giving less resolution to single-coefficient bands because they get + more benefit from having one coarse value per coefficient*/ + if ((m->eBands[j+1]-m->eBands[j])<<LM==1) + trim_offset[j] -= C<<BITRES; + } + lo = 1; + hi = m->nbAllocVectors - 1; + do + { + int done = 0; + int psum = 0; + int mid = (lo+hi) >> 1; + for (j=end;j-->start;) + { + int bitsj; + int N = m->eBands[j+1]-m->eBands[j]; + bitsj = C*N*m->allocVectors[mid*len+j]<<LM>>2; + if (bitsj > 0) + bitsj = IMAX(0, bitsj + trim_offset[j]); + bitsj += offsets[j]; + if (bitsj >= thresh[j] || done) + { + done = 1; + /* Don't allocate more than we can actually use */ + psum += IMIN(bitsj, cap[j]); + } else { + if (bitsj >= C<<BITRES) + psum += C<<BITRES; + } + } + if (psum > total) + hi = mid - 1; + else + lo = mid + 1; + /*printf ("lo = %d, hi = %d\n", lo, hi);*/ + } + while (lo <= hi); + hi = lo--; + /*printf ("interp between %d and %d\n", lo, hi);*/ + for (j=start;j<end;j++) + { + int bits1j, bits2j; + int N = m->eBands[j+1]-m->eBands[j]; + bits1j = C*N*m->allocVectors[lo*len+j]<<LM>>2; + bits2j = hi>=m->nbAllocVectors ? + cap[j] : C*N*m->allocVectors[hi*len+j]<<LM>>2; + if (bits1j > 0) + bits1j = IMAX(0, bits1j + trim_offset[j]); + if (bits2j > 0) + bits2j = IMAX(0, bits2j + trim_offset[j]); + if (lo > 0) + bits1j += offsets[j]; + bits2j += offsets[j]; + if (offsets[j]>0) + skip_start = j; + bits2j = IMAX(0,bits2j-bits1j); + bits1[j] = bits1j; + bits2[j] = bits2j; + } + codedBands = interp_bits2pulses(m, start, end, skip_start, bits1, bits2, thresh, cap, + total, balance, skip_rsv, intensity, intensity_rsv, dual_stereo, dual_stereo_rsv, + pulses, ebits, fine_priority, C, LM, ec, encode, prev, signalBandwidth); + RESTORE_STACK; + return codedBands; +} + diff --git a/media/libopus/celt/rate.h b/media/libopus/celt/rate.h new file mode 100644 index 0000000000..fad5e412da --- /dev/null +++ b/media/libopus/celt/rate.h @@ -0,0 +1,101 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef RATE_H +#define RATE_H + +#define MAX_PSEUDO 40 +#define LOG_MAX_PSEUDO 6 + +#define CELT_MAX_PULSES 128 + +#define MAX_FINE_BITS 8 + +#define FINE_OFFSET 21 +#define QTHETA_OFFSET 4 +#define QTHETA_OFFSET_TWOPHASE 16 + +#include "cwrs.h" +#include "modes.h" + +void compute_pulse_cache(CELTMode *m, int LM); + +static OPUS_INLINE int get_pulses(int i) +{ + return i<8 ? i : (8 + (i&7)) << ((i>>3)-1); +} + +static OPUS_INLINE int bits2pulses(const CELTMode *m, int band, int LM, int bits) +{ + int i; + int lo, hi; + const unsigned char *cache; + + LM++; + cache = m->cache.bits + m->cache.index[LM*m->nbEBands+band]; + + lo = 0; + hi = cache[0]; + bits--; + for (i=0;i<LOG_MAX_PSEUDO;i++) + { + int mid = (lo+hi+1)>>1; + /* OPT: Make sure this is implemented with a conditional move */ + if ((int)cache[mid] >= bits) + hi = mid; + else + lo = mid; + } + if (bits- (lo == 0 ? -1 : (int)cache[lo]) <= (int)cache[hi]-bits) + return lo; + else + return hi; +} + +static OPUS_INLINE int pulses2bits(const CELTMode *m, int band, int LM, int pulses) +{ + const unsigned char *cache; + + LM++; + cache = m->cache.bits + m->cache.index[LM*m->nbEBands+band]; + return pulses == 0 ? 0 : cache[pulses]+1; +} + +/** Compute the pulse allocation, i.e. how many pulses will go in each + * band. + @param m mode + @param offsets Requested increase or decrease in the number of bits for + each band + @param total Number of bands + @param pulses Number of pulses per band (returned) + @return Total number of bits allocated +*/ +int clt_compute_allocation(const CELTMode *m, int start, int end, const int *offsets, const int *cap, int alloc_trim, int *intensity, int *dual_stereo, + opus_int32 total, opus_int32 *balance, int *pulses, int *ebits, int *fine_priority, int C, int LM, ec_ctx *ec, int encode, int prev, int signalBandwidth); + +#endif diff --git a/media/libopus/celt/stack_alloc.h b/media/libopus/celt/stack_alloc.h new file mode 100644 index 0000000000..ae40e2a165 --- /dev/null +++ b/media/libopus/celt/stack_alloc.h @@ -0,0 +1,184 @@ +/* Copyright (C) 2002-2003 Jean-Marc Valin + Copyright (C) 2007-2009 Xiph.Org Foundation */ +/** + @file stack_alloc.h + @brief Temporary memory allocation on stack +*/ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef STACK_ALLOC_H +#define STACK_ALLOC_H + +#include "opus_types.h" +#include "opus_defines.h" + +#if (!defined (VAR_ARRAYS) && !defined (USE_ALLOCA) && !defined (NONTHREADSAFE_PSEUDOSTACK)) +#error "Opus requires one of VAR_ARRAYS, USE_ALLOCA, or NONTHREADSAFE_PSEUDOSTACK be defined to select the temporary allocation mode." +#endif + +#ifdef USE_ALLOCA +# ifdef _WIN32 +# include <malloc.h> +# else +# ifdef HAVE_ALLOCA_H +# include <alloca.h> +# else +# include <stdlib.h> +# endif +# endif +#endif + +/** + * @def ALIGN(stack, size) + * + * Aligns the stack to a 'size' boundary + * + * @param stack Stack + * @param size New size boundary + */ + +/** + * @def PUSH(stack, size, type) + * + * Allocates 'size' elements of type 'type' on the stack + * + * @param stack Stack + * @param size Number of elements + * @param type Type of element + */ + +/** + * @def VARDECL(var) + * + * Declare variable on stack + * + * @param var Variable to declare + */ + +/** + * @def ALLOC(var, size, type) + * + * Allocate 'size' elements of 'type' on stack + * + * @param var Name of variable to allocate + * @param size Number of elements + * @param type Type of element + */ + +#if defined(VAR_ARRAYS) + +#define VARDECL(type, var) +#define ALLOC(var, size, type) type var[size] +#define SAVE_STACK +#define RESTORE_STACK +#define ALLOC_STACK +/* C99 does not allow VLAs of size zero */ +#define ALLOC_NONE 1 + +#elif defined(USE_ALLOCA) + +#define VARDECL(type, var) type *var + +# ifdef _WIN32 +# define ALLOC(var, size, type) var = ((type*)_alloca(sizeof(type)*(size))) +# else +# define ALLOC(var, size, type) var = ((type*)alloca(sizeof(type)*(size))) +# endif + +#define SAVE_STACK +#define RESTORE_STACK +#define ALLOC_STACK +#define ALLOC_NONE 0 + +#else + +#ifdef CELT_C +char *scratch_ptr=0; +char *global_stack=0; +#else +extern char *global_stack; +extern char *scratch_ptr; +#endif /* CELT_C */ + +#ifdef ENABLE_VALGRIND + +#include <valgrind/memcheck.h> + +#ifdef CELT_C +char *global_stack_top=0; +#else +extern char *global_stack_top; +#endif /* CELT_C */ + +#define ALIGN(stack, size) ((stack) += ((size) - (long)(stack)) & ((size) - 1)) +#define PUSH(stack, size, type) (VALGRIND_MAKE_MEM_NOACCESS(stack, global_stack_top-stack),ALIGN((stack),sizeof(type)/sizeof(char)),VALGRIND_MAKE_MEM_UNDEFINED(stack, ((size)*sizeof(type)/sizeof(char))),(stack)+=(2*(size)*sizeof(type)/sizeof(char)),(type*)((stack)-(2*(size)*sizeof(type)/sizeof(char)))) +#define RESTORE_STACK ((global_stack = _saved_stack),VALGRIND_MAKE_MEM_NOACCESS(global_stack, global_stack_top-global_stack)) +#define ALLOC_STACK char *_saved_stack; ((global_stack = (global_stack==0) ? ((global_stack_top=opus_alloc_scratch(GLOBAL_STACK_SIZE*2)+(GLOBAL_STACK_SIZE*2))-(GLOBAL_STACK_SIZE*2)) : global_stack),VALGRIND_MAKE_MEM_NOACCESS(global_stack, global_stack_top-global_stack)); _saved_stack = global_stack; + +#else + +#define ALIGN(stack, size) ((stack) += ((size) - (long)(stack)) & ((size) - 1)) +#define PUSH(stack, size, type) (ALIGN((stack),sizeof(type)/sizeof(char)),(stack)+=(size)*(sizeof(type)/sizeof(char)),(type*)((stack)-(size)*(sizeof(type)/sizeof(char)))) +#if 0 /* Set this to 1 to instrument pseudostack usage */ +#define RESTORE_STACK (printf("%ld %s:%d\n", global_stack-scratch_ptr, __FILE__, __LINE__),global_stack = _saved_stack) +#else +#define RESTORE_STACK (global_stack = _saved_stack) +#endif +#define ALLOC_STACK char *_saved_stack; (global_stack = (global_stack==0) ? (scratch_ptr=opus_alloc_scratch(GLOBAL_STACK_SIZE)) : global_stack); _saved_stack = global_stack; + +#endif /* ENABLE_VALGRIND */ + +#include "os_support.h" +#define VARDECL(type, var) type *var +#define ALLOC(var, size, type) var = PUSH(global_stack, size, type) +#define SAVE_STACK char *_saved_stack = global_stack; +#define ALLOC_NONE 0 + +#endif /* VAR_ARRAYS */ + + +#ifdef ENABLE_VALGRIND + +#include <valgrind/memcheck.h> +#define OPUS_CHECK_ARRAY(ptr, len) VALGRIND_CHECK_MEM_IS_DEFINED(ptr, len*sizeof(*ptr)) +#define OPUS_CHECK_VALUE(value) VALGRIND_CHECK_VALUE_IS_DEFINED(value) +#define OPUS_CHECK_ARRAY_COND(ptr, len) VALGRIND_CHECK_MEM_IS_DEFINED(ptr, len*sizeof(*ptr)) +#define OPUS_CHECK_VALUE_COND(value) VALGRIND_CHECK_VALUE_IS_DEFINED(value) +#define OPUS_PRINT_INT(value) do {fprintf(stderr, #value " = %d at %s:%d\n", value, __FILE__, __LINE__);}while(0) +#define OPUS_FPRINTF fprintf + +#else + +static OPUS_INLINE int _opus_false(void) {return 0;} +#define OPUS_CHECK_ARRAY(ptr, len) _opus_false() +#define OPUS_CHECK_VALUE(value) _opus_false() +#define OPUS_PRINT_INT(value) do{}while(0) +#define OPUS_FPRINTF (void) + +#endif + + +#endif /* STACK_ALLOC_H */ diff --git a/media/libopus/celt/static_modes_fixed.h b/media/libopus/celt/static_modes_fixed.h new file mode 100644 index 0000000000..8717d626cb --- /dev/null +++ b/media/libopus/celt/static_modes_fixed.h @@ -0,0 +1,892 @@ +/* The contents of this file was automatically generated by dump_modes.c + with arguments: 48000 960 + It contains static definitions for some pre-defined modes. */ +#include "modes.h" +#include "rate.h" + +#ifdef HAVE_ARM_NE10 +#define OVERRIDE_FFT 1 +#include "static_modes_fixed_arm_ne10.h" +#endif + +#ifndef DEF_WINDOW120 +#define DEF_WINDOW120 +static const opus_val16 window120[120] = { +2, 20, 55, 108, 178, +266, 372, 494, 635, 792, +966, 1157, 1365, 1590, 1831, +2089, 2362, 2651, 2956, 3276, +3611, 3961, 4325, 4703, 5094, +5499, 5916, 6346, 6788, 7241, +7705, 8179, 8663, 9156, 9657, +10167, 10684, 11207, 11736, 12271, +12810, 13353, 13899, 14447, 14997, +15547, 16098, 16648, 17197, 17744, +18287, 18827, 19363, 19893, 20418, +20936, 21447, 21950, 22445, 22931, +23407, 23874, 24330, 24774, 25208, +25629, 26039, 26435, 26819, 27190, +27548, 27893, 28224, 28541, 28845, +29135, 29411, 29674, 29924, 30160, +30384, 30594, 30792, 30977, 31151, +31313, 31463, 31602, 31731, 31849, +31958, 32057, 32148, 32229, 32303, +32370, 32429, 32481, 32528, 32568, +32604, 32634, 32661, 32683, 32701, +32717, 32729, 32740, 32748, 32754, +32758, 32762, 32764, 32766, 32767, +32767, 32767, 32767, 32767, 32767, +}; +#endif + +#ifndef DEF_LOGN400 +#define DEF_LOGN400 +static const opus_int16 logN400[21] = { +0, 0, 0, 0, 0, 0, 0, 0, 8, 8, 8, 8, 16, 16, 16, 21, 21, 24, 29, 34, 36, }; +#endif + +#ifndef DEF_PULSE_CACHE50 +#define DEF_PULSE_CACHE50 +static const opus_int16 cache_index50[105] = { +-1, -1, -1, -1, -1, -1, -1, -1, 0, 0, 0, 0, 41, 41, 41, +82, 82, 123, 164, 200, 222, 0, 0, 0, 0, 0, 0, 0, 0, 41, +41, 41, 41, 123, 123, 123, 164, 164, 240, 266, 283, 295, 41, 41, 41, +41, 41, 41, 41, 41, 123, 123, 123, 123, 240, 240, 240, 266, 266, 305, +318, 328, 336, 123, 123, 123, 123, 123, 123, 123, 123, 240, 240, 240, 240, +305, 305, 305, 318, 318, 343, 351, 358, 364, 240, 240, 240, 240, 240, 240, +240, 240, 305, 305, 305, 305, 343, 343, 343, 351, 351, 370, 376, 382, 387, +}; +static const unsigned char cache_bits50[392] = { +40, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, +7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, +7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 40, 15, 23, 28, +31, 34, 36, 38, 39, 41, 42, 43, 44, 45, 46, 47, 47, 49, 50, +51, 52, 53, 54, 55, 55, 57, 58, 59, 60, 61, 62, 63, 63, 65, +66, 67, 68, 69, 70, 71, 71, 40, 20, 33, 41, 48, 53, 57, 61, +64, 66, 69, 71, 73, 75, 76, 78, 80, 82, 85, 87, 89, 91, 92, +94, 96, 98, 101, 103, 105, 107, 108, 110, 112, 114, 117, 119, 121, 123, +124, 126, 128, 40, 23, 39, 51, 60, 67, 73, 79, 83, 87, 91, 94, +97, 100, 102, 105, 107, 111, 115, 118, 121, 124, 126, 129, 131, 135, 139, +142, 145, 148, 150, 153, 155, 159, 163, 166, 169, 172, 174, 177, 179, 35, +28, 49, 65, 78, 89, 99, 107, 114, 120, 126, 132, 136, 141, 145, 149, +153, 159, 165, 171, 176, 180, 185, 189, 192, 199, 205, 211, 216, 220, 225, +229, 232, 239, 245, 251, 21, 33, 58, 79, 97, 112, 125, 137, 148, 157, +166, 174, 182, 189, 195, 201, 207, 217, 227, 235, 243, 251, 17, 35, 63, +86, 106, 123, 139, 152, 165, 177, 187, 197, 206, 214, 222, 230, 237, 250, +25, 31, 55, 75, 91, 105, 117, 128, 138, 146, 154, 161, 168, 174, 180, +185, 190, 200, 208, 215, 222, 229, 235, 240, 245, 255, 16, 36, 65, 89, +110, 128, 144, 159, 173, 185, 196, 207, 217, 226, 234, 242, 250, 11, 41, +74, 103, 128, 151, 172, 191, 209, 225, 241, 255, 9, 43, 79, 110, 138, +163, 186, 207, 227, 246, 12, 39, 71, 99, 123, 144, 164, 182, 198, 214, +228, 241, 253, 9, 44, 81, 113, 142, 168, 192, 214, 235, 255, 7, 49, +90, 127, 160, 191, 220, 247, 6, 51, 95, 134, 170, 203, 234, 7, 47, +87, 123, 155, 184, 212, 237, 6, 52, 97, 137, 174, 208, 240, 5, 57, +106, 151, 192, 231, 5, 59, 111, 158, 202, 243, 5, 55, 103, 147, 187, +224, 5, 60, 113, 161, 206, 248, 4, 65, 122, 175, 224, 4, 67, 127, +182, 234, }; +static const unsigned char cache_caps50[168] = { +224, 224, 224, 224, 224, 224, 224, 224, 160, 160, 160, 160, 185, 185, 185, +178, 178, 168, 134, 61, 37, 224, 224, 224, 224, 224, 224, 224, 224, 240, +240, 240, 240, 207, 207, 207, 198, 198, 183, 144, 66, 40, 160, 160, 160, +160, 160, 160, 160, 160, 185, 185, 185, 185, 193, 193, 193, 183, 183, 172, +138, 64, 38, 240, 240, 240, 240, 240, 240, 240, 240, 207, 207, 207, 207, +204, 204, 204, 193, 193, 180, 143, 66, 40, 185, 185, 185, 185, 185, 185, +185, 185, 193, 193, 193, 193, 193, 193, 193, 183, 183, 172, 138, 65, 39, +207, 207, 207, 207, 207, 207, 207, 207, 204, 204, 204, 204, 201, 201, 201, +188, 188, 176, 141, 66, 40, 193, 193, 193, 193, 193, 193, 193, 193, 193, +193, 193, 193, 194, 194, 194, 184, 184, 173, 139, 65, 39, 204, 204, 204, +204, 204, 204, 204, 204, 201, 201, 201, 201, 198, 198, 198, 187, 187, 175, +140, 66, 40, }; +#endif + +#ifndef FFT_TWIDDLES48000_960 +#define FFT_TWIDDLES48000_960 +static const kiss_twiddle_cpx fft_twiddles48000_960[480] = { +{32767, 0}, {32766, -429}, +{32757, -858}, {32743, -1287}, +{32724, -1715}, {32698, -2143}, +{32667, -2570}, {32631, -2998}, +{32588, -3425}, {32541, -3851}, +{32488, -4277}, {32429, -4701}, +{32364, -5125}, {32295, -5548}, +{32219, -5971}, {32138, -6393}, +{32051, -6813}, {31960, -7231}, +{31863, -7650}, {31760, -8067}, +{31652, -8481}, {31539, -8895}, +{31419, -9306}, {31294, -9716}, +{31165, -10126}, {31030, -10532}, +{30889, -10937}, {30743, -11340}, +{30592, -11741}, {30436, -12141}, +{30274, -12540}, {30107, -12935}, +{29936, -13328}, {29758, -13718}, +{29577, -14107}, {29390, -14493}, +{29197, -14875}, {29000, -15257}, +{28797, -15635}, {28590, -16010}, +{28379, -16384}, {28162, -16753}, +{27940, -17119}, {27714, -17484}, +{27482, -17845}, {27246, -18205}, +{27006, -18560}, {26760, -18911}, +{26510, -19260}, {26257, -19606}, +{25997, -19947}, {25734, -20286}, +{25466, -20621}, {25194, -20952}, +{24918, -21281}, {24637, -21605}, +{24353, -21926}, {24063, -22242}, +{23770, -22555}, {23473, -22865}, +{23171, -23171}, {22866, -23472}, +{22557, -23769}, {22244, -24063}, +{21927, -24352}, {21606, -24636}, +{21282, -24917}, {20954, -25194}, +{20622, -25465}, {20288, -25733}, +{19949, -25997}, {19607, -26255}, +{19261, -26509}, {18914, -26760}, +{18561, -27004}, {18205, -27246}, +{17846, -27481}, {17485, -27713}, +{17122, -27940}, {16755, -28162}, +{16385, -28378}, {16012, -28590}, +{15636, -28797}, {15258, -28999}, +{14878, -29197}, {14494, -29389}, +{14108, -29576}, {13720, -29757}, +{13329, -29934}, {12937, -30107}, +{12540, -30274}, {12142, -30435}, +{11744, -30592}, {11342, -30743}, +{10939, -30889}, {10534, -31030}, +{10127, -31164}, {9718, -31294}, +{9307, -31418}, {8895, -31537}, +{8482, -31652}, {8067, -31759}, +{7650, -31862}, {7233, -31960}, +{6815, -32051}, {6393, -32138}, +{5973, -32219}, {5549, -32294}, +{5127, -32364}, {4703, -32429}, +{4278, -32487}, {3852, -32541}, +{3426, -32588}, {2999, -32630}, +{2572, -32667}, {2144, -32698}, +{1716, -32724}, {1287, -32742}, +{860, -32757}, {430, -32766}, +{0, -32767}, {-429, -32766}, +{-858, -32757}, {-1287, -32743}, +{-1715, -32724}, {-2143, -32698}, +{-2570, -32667}, {-2998, -32631}, +{-3425, -32588}, {-3851, -32541}, +{-4277, -32488}, {-4701, -32429}, +{-5125, -32364}, {-5548, -32295}, +{-5971, -32219}, {-6393, -32138}, +{-6813, -32051}, {-7231, -31960}, +{-7650, -31863}, {-8067, -31760}, +{-8481, -31652}, {-8895, -31539}, +{-9306, -31419}, {-9716, -31294}, +{-10126, -31165}, {-10532, -31030}, +{-10937, -30889}, {-11340, -30743}, +{-11741, -30592}, {-12141, -30436}, +{-12540, -30274}, {-12935, -30107}, +{-13328, -29936}, {-13718, -29758}, +{-14107, -29577}, {-14493, -29390}, +{-14875, -29197}, {-15257, -29000}, +{-15635, -28797}, {-16010, -28590}, +{-16384, -28379}, {-16753, -28162}, +{-17119, -27940}, {-17484, -27714}, +{-17845, -27482}, {-18205, -27246}, +{-18560, -27006}, {-18911, -26760}, +{-19260, -26510}, {-19606, -26257}, +{-19947, -25997}, {-20286, -25734}, +{-20621, -25466}, {-20952, -25194}, +{-21281, -24918}, {-21605, -24637}, +{-21926, -24353}, {-22242, -24063}, +{-22555, -23770}, {-22865, -23473}, +{-23171, -23171}, {-23472, -22866}, +{-23769, -22557}, {-24063, -22244}, +{-24352, -21927}, {-24636, -21606}, +{-24917, -21282}, {-25194, -20954}, +{-25465, -20622}, {-25733, -20288}, +{-25997, -19949}, {-26255, -19607}, +{-26509, -19261}, {-26760, -18914}, +{-27004, -18561}, {-27246, -18205}, +{-27481, -17846}, {-27713, -17485}, +{-27940, -17122}, {-28162, -16755}, +{-28378, -16385}, {-28590, -16012}, +{-28797, -15636}, {-28999, -15258}, +{-29197, -14878}, {-29389, -14494}, +{-29576, -14108}, {-29757, -13720}, +{-29934, -13329}, {-30107, -12937}, +{-30274, -12540}, {-30435, -12142}, +{-30592, -11744}, {-30743, -11342}, +{-30889, -10939}, {-31030, -10534}, +{-31164, -10127}, {-31294, -9718}, +{-31418, -9307}, {-31537, -8895}, +{-31652, -8482}, {-31759, -8067}, +{-31862, -7650}, {-31960, -7233}, +{-32051, -6815}, {-32138, -6393}, +{-32219, -5973}, {-32294, -5549}, +{-32364, -5127}, {-32429, -4703}, +{-32487, -4278}, {-32541, -3852}, +{-32588, -3426}, {-32630, -2999}, +{-32667, -2572}, {-32698, -2144}, +{-32724, -1716}, {-32742, -1287}, +{-32757, -860}, {-32766, -430}, +{-32767, 0}, {-32766, 429}, +{-32757, 858}, {-32743, 1287}, +{-32724, 1715}, {-32698, 2143}, +{-32667, 2570}, {-32631, 2998}, +{-32588, 3425}, {-32541, 3851}, +{-32488, 4277}, {-32429, 4701}, +{-32364, 5125}, {-32295, 5548}, +{-32219, 5971}, {-32138, 6393}, +{-32051, 6813}, {-31960, 7231}, +{-31863, 7650}, {-31760, 8067}, +{-31652, 8481}, {-31539, 8895}, +{-31419, 9306}, {-31294, 9716}, +{-31165, 10126}, {-31030, 10532}, +{-30889, 10937}, {-30743, 11340}, +{-30592, 11741}, {-30436, 12141}, +{-30274, 12540}, {-30107, 12935}, +{-29936, 13328}, {-29758, 13718}, +{-29577, 14107}, {-29390, 14493}, +{-29197, 14875}, {-29000, 15257}, +{-28797, 15635}, {-28590, 16010}, +{-28379, 16384}, {-28162, 16753}, +{-27940, 17119}, {-27714, 17484}, +{-27482, 17845}, {-27246, 18205}, +{-27006, 18560}, {-26760, 18911}, +{-26510, 19260}, {-26257, 19606}, +{-25997, 19947}, {-25734, 20286}, +{-25466, 20621}, {-25194, 20952}, +{-24918, 21281}, {-24637, 21605}, +{-24353, 21926}, {-24063, 22242}, +{-23770, 22555}, {-23473, 22865}, +{-23171, 23171}, {-22866, 23472}, +{-22557, 23769}, {-22244, 24063}, +{-21927, 24352}, {-21606, 24636}, +{-21282, 24917}, {-20954, 25194}, +{-20622, 25465}, {-20288, 25733}, +{-19949, 25997}, {-19607, 26255}, +{-19261, 26509}, {-18914, 26760}, +{-18561, 27004}, {-18205, 27246}, +{-17846, 27481}, {-17485, 27713}, +{-17122, 27940}, {-16755, 28162}, +{-16385, 28378}, {-16012, 28590}, +{-15636, 28797}, {-15258, 28999}, +{-14878, 29197}, {-14494, 29389}, +{-14108, 29576}, {-13720, 29757}, +{-13329, 29934}, {-12937, 30107}, +{-12540, 30274}, {-12142, 30435}, +{-11744, 30592}, {-11342, 30743}, +{-10939, 30889}, {-10534, 31030}, +{-10127, 31164}, {-9718, 31294}, +{-9307, 31418}, {-8895, 31537}, +{-8482, 31652}, {-8067, 31759}, +{-7650, 31862}, {-7233, 31960}, +{-6815, 32051}, {-6393, 32138}, +{-5973, 32219}, {-5549, 32294}, +{-5127, 32364}, {-4703, 32429}, +{-4278, 32487}, {-3852, 32541}, +{-3426, 32588}, {-2999, 32630}, +{-2572, 32667}, {-2144, 32698}, +{-1716, 32724}, {-1287, 32742}, +{-860, 32757}, {-430, 32766}, +{0, 32767}, {429, 32766}, +{858, 32757}, {1287, 32743}, +{1715, 32724}, {2143, 32698}, +{2570, 32667}, {2998, 32631}, +{3425, 32588}, {3851, 32541}, +{4277, 32488}, {4701, 32429}, +{5125, 32364}, {5548, 32295}, +{5971, 32219}, {6393, 32138}, +{6813, 32051}, {7231, 31960}, +{7650, 31863}, {8067, 31760}, +{8481, 31652}, {8895, 31539}, +{9306, 31419}, {9716, 31294}, +{10126, 31165}, {10532, 31030}, +{10937, 30889}, {11340, 30743}, +{11741, 30592}, {12141, 30436}, +{12540, 30274}, {12935, 30107}, +{13328, 29936}, {13718, 29758}, +{14107, 29577}, {14493, 29390}, +{14875, 29197}, {15257, 29000}, +{15635, 28797}, {16010, 28590}, +{16384, 28379}, {16753, 28162}, +{17119, 27940}, {17484, 27714}, +{17845, 27482}, {18205, 27246}, +{18560, 27006}, {18911, 26760}, +{19260, 26510}, {19606, 26257}, +{19947, 25997}, {20286, 25734}, +{20621, 25466}, {20952, 25194}, +{21281, 24918}, {21605, 24637}, +{21926, 24353}, {22242, 24063}, +{22555, 23770}, {22865, 23473}, +{23171, 23171}, {23472, 22866}, +{23769, 22557}, {24063, 22244}, +{24352, 21927}, {24636, 21606}, +{24917, 21282}, {25194, 20954}, +{25465, 20622}, {25733, 20288}, +{25997, 19949}, {26255, 19607}, +{26509, 19261}, {26760, 18914}, +{27004, 18561}, {27246, 18205}, +{27481, 17846}, {27713, 17485}, +{27940, 17122}, {28162, 16755}, +{28378, 16385}, {28590, 16012}, +{28797, 15636}, {28999, 15258}, +{29197, 14878}, {29389, 14494}, +{29576, 14108}, {29757, 13720}, +{29934, 13329}, {30107, 12937}, +{30274, 12540}, {30435, 12142}, +{30592, 11744}, {30743, 11342}, +{30889, 10939}, {31030, 10534}, +{31164, 10127}, {31294, 9718}, +{31418, 9307}, {31537, 8895}, +{31652, 8482}, {31759, 8067}, +{31862, 7650}, {31960, 7233}, +{32051, 6815}, {32138, 6393}, +{32219, 5973}, {32294, 5549}, +{32364, 5127}, {32429, 4703}, +{32487, 4278}, {32541, 3852}, +{32588, 3426}, {32630, 2999}, +{32667, 2572}, {32698, 2144}, +{32724, 1716}, {32742, 1287}, +{32757, 860}, {32766, 430}, +}; +#ifndef FFT_BITREV480 +#define FFT_BITREV480 +static const opus_int16 fft_bitrev480[480] = { +0, 96, 192, 288, 384, 32, 128, 224, 320, 416, 64, 160, 256, 352, 448, +8, 104, 200, 296, 392, 40, 136, 232, 328, 424, 72, 168, 264, 360, 456, +16, 112, 208, 304, 400, 48, 144, 240, 336, 432, 80, 176, 272, 368, 464, +24, 120, 216, 312, 408, 56, 152, 248, 344, 440, 88, 184, 280, 376, 472, +4, 100, 196, 292, 388, 36, 132, 228, 324, 420, 68, 164, 260, 356, 452, +12, 108, 204, 300, 396, 44, 140, 236, 332, 428, 76, 172, 268, 364, 460, +20, 116, 212, 308, 404, 52, 148, 244, 340, 436, 84, 180, 276, 372, 468, +28, 124, 220, 316, 412, 60, 156, 252, 348, 444, 92, 188, 284, 380, 476, +1, 97, 193, 289, 385, 33, 129, 225, 321, 417, 65, 161, 257, 353, 449, +9, 105, 201, 297, 393, 41, 137, 233, 329, 425, 73, 169, 265, 361, 457, +17, 113, 209, 305, 401, 49, 145, 241, 337, 433, 81, 177, 273, 369, 465, +25, 121, 217, 313, 409, 57, 153, 249, 345, 441, 89, 185, 281, 377, 473, +5, 101, 197, 293, 389, 37, 133, 229, 325, 421, 69, 165, 261, 357, 453, +13, 109, 205, 301, 397, 45, 141, 237, 333, 429, 77, 173, 269, 365, 461, +21, 117, 213, 309, 405, 53, 149, 245, 341, 437, 85, 181, 277, 373, 469, +29, 125, 221, 317, 413, 61, 157, 253, 349, 445, 93, 189, 285, 381, 477, +2, 98, 194, 290, 386, 34, 130, 226, 322, 418, 66, 162, 258, 354, 450, +10, 106, 202, 298, 394, 42, 138, 234, 330, 426, 74, 170, 266, 362, 458, +18, 114, 210, 306, 402, 50, 146, 242, 338, 434, 82, 178, 274, 370, 466, +26, 122, 218, 314, 410, 58, 154, 250, 346, 442, 90, 186, 282, 378, 474, +6, 102, 198, 294, 390, 38, 134, 230, 326, 422, 70, 166, 262, 358, 454, +14, 110, 206, 302, 398, 46, 142, 238, 334, 430, 78, 174, 270, 366, 462, +22, 118, 214, 310, 406, 54, 150, 246, 342, 438, 86, 182, 278, 374, 470, +30, 126, 222, 318, 414, 62, 158, 254, 350, 446, 94, 190, 286, 382, 478, +3, 99, 195, 291, 387, 35, 131, 227, 323, 419, 67, 163, 259, 355, 451, +11, 107, 203, 299, 395, 43, 139, 235, 331, 427, 75, 171, 267, 363, 459, +19, 115, 211, 307, 403, 51, 147, 243, 339, 435, 83, 179, 275, 371, 467, +27, 123, 219, 315, 411, 59, 155, 251, 347, 443, 91, 187, 283, 379, 475, +7, 103, 199, 295, 391, 39, 135, 231, 327, 423, 71, 167, 263, 359, 455, +15, 111, 207, 303, 399, 47, 143, 239, 335, 431, 79, 175, 271, 367, 463, +23, 119, 215, 311, 407, 55, 151, 247, 343, 439, 87, 183, 279, 375, 471, +31, 127, 223, 319, 415, 63, 159, 255, 351, 447, 95, 191, 287, 383, 479, +}; +#endif + +#ifndef FFT_BITREV240 +#define FFT_BITREV240 +static const opus_int16 fft_bitrev240[240] = { +0, 48, 96, 144, 192, 16, 64, 112, 160, 208, 32, 80, 128, 176, 224, +4, 52, 100, 148, 196, 20, 68, 116, 164, 212, 36, 84, 132, 180, 228, +8, 56, 104, 152, 200, 24, 72, 120, 168, 216, 40, 88, 136, 184, 232, +12, 60, 108, 156, 204, 28, 76, 124, 172, 220, 44, 92, 140, 188, 236, +1, 49, 97, 145, 193, 17, 65, 113, 161, 209, 33, 81, 129, 177, 225, +5, 53, 101, 149, 197, 21, 69, 117, 165, 213, 37, 85, 133, 181, 229, +9, 57, 105, 153, 201, 25, 73, 121, 169, 217, 41, 89, 137, 185, 233, +13, 61, 109, 157, 205, 29, 77, 125, 173, 221, 45, 93, 141, 189, 237, +2, 50, 98, 146, 194, 18, 66, 114, 162, 210, 34, 82, 130, 178, 226, +6, 54, 102, 150, 198, 22, 70, 118, 166, 214, 38, 86, 134, 182, 230, +10, 58, 106, 154, 202, 26, 74, 122, 170, 218, 42, 90, 138, 186, 234, +14, 62, 110, 158, 206, 30, 78, 126, 174, 222, 46, 94, 142, 190, 238, +3, 51, 99, 147, 195, 19, 67, 115, 163, 211, 35, 83, 131, 179, 227, +7, 55, 103, 151, 199, 23, 71, 119, 167, 215, 39, 87, 135, 183, 231, +11, 59, 107, 155, 203, 27, 75, 123, 171, 219, 43, 91, 139, 187, 235, +15, 63, 111, 159, 207, 31, 79, 127, 175, 223, 47, 95, 143, 191, 239, +}; +#endif + +#ifndef FFT_BITREV120 +#define FFT_BITREV120 +static const opus_int16 fft_bitrev120[120] = { +0, 24, 48, 72, 96, 8, 32, 56, 80, 104, 16, 40, 64, 88, 112, +4, 28, 52, 76, 100, 12, 36, 60, 84, 108, 20, 44, 68, 92, 116, +1, 25, 49, 73, 97, 9, 33, 57, 81, 105, 17, 41, 65, 89, 113, +5, 29, 53, 77, 101, 13, 37, 61, 85, 109, 21, 45, 69, 93, 117, +2, 26, 50, 74, 98, 10, 34, 58, 82, 106, 18, 42, 66, 90, 114, +6, 30, 54, 78, 102, 14, 38, 62, 86, 110, 22, 46, 70, 94, 118, +3, 27, 51, 75, 99, 11, 35, 59, 83, 107, 19, 43, 67, 91, 115, +7, 31, 55, 79, 103, 15, 39, 63, 87, 111, 23, 47, 71, 95, 119, +}; +#endif + +#ifndef FFT_BITREV60 +#define FFT_BITREV60 +static const opus_int16 fft_bitrev60[60] = { +0, 12, 24, 36, 48, 4, 16, 28, 40, 52, 8, 20, 32, 44, 56, +1, 13, 25, 37, 49, 5, 17, 29, 41, 53, 9, 21, 33, 45, 57, +2, 14, 26, 38, 50, 6, 18, 30, 42, 54, 10, 22, 34, 46, 58, +3, 15, 27, 39, 51, 7, 19, 31, 43, 55, 11, 23, 35, 47, 59, +}; +#endif + +#ifndef FFT_STATE48000_960_0 +#define FFT_STATE48000_960_0 +static const kiss_fft_state fft_state48000_960_0 = { +480, /* nfft */ +17476, /* scale */ +8, /* scale_shift */ +-1, /* shift */ +{5, 96, 3, 32, 4, 8, 2, 4, 4, 1, 0, 0, 0, 0, 0, 0, }, /* factors */ +fft_bitrev480, /* bitrev */ +fft_twiddles48000_960, /* bitrev */ +#ifdef OVERRIDE_FFT +(arch_fft_state *)&cfg_arch_480, +#else +NULL, +#endif +}; +#endif + +#ifndef FFT_STATE48000_960_1 +#define FFT_STATE48000_960_1 +static const kiss_fft_state fft_state48000_960_1 = { +240, /* nfft */ +17476, /* scale */ +7, /* scale_shift */ +1, /* shift */ +{5, 48, 3, 16, 4, 4, 4, 1, 0, 0, 0, 0, 0, 0, 0, 0, }, /* factors */ +fft_bitrev240, /* bitrev */ +fft_twiddles48000_960, /* bitrev */ +#ifdef OVERRIDE_FFT +(arch_fft_state *)&cfg_arch_240, +#else +NULL, +#endif +}; +#endif + +#ifndef FFT_STATE48000_960_2 +#define FFT_STATE48000_960_2 +static const kiss_fft_state fft_state48000_960_2 = { +120, /* nfft */ +17476, /* scale */ +6, /* scale_shift */ +2, /* shift */ +{5, 24, 3, 8, 2, 4, 4, 1, 0, 0, 0, 0, 0, 0, 0, 0, }, /* factors */ +fft_bitrev120, /* bitrev */ +fft_twiddles48000_960, /* bitrev */ +#ifdef OVERRIDE_FFT +(arch_fft_state *)&cfg_arch_120, +#else +NULL, +#endif +}; +#endif + +#ifndef FFT_STATE48000_960_3 +#define FFT_STATE48000_960_3 +static const kiss_fft_state fft_state48000_960_3 = { +60, /* nfft */ +17476, /* scale */ +5, /* scale_shift */ +3, /* shift */ +{5, 12, 3, 4, 4, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, }, /* factors */ +fft_bitrev60, /* bitrev */ +fft_twiddles48000_960, /* bitrev */ +#ifdef OVERRIDE_FFT +(arch_fft_state *)&cfg_arch_60, +#else +NULL, +#endif +}; +#endif + +#endif + +#ifndef MDCT_TWIDDLES960 +#define MDCT_TWIDDLES960 +static const opus_val16 mdct_twiddles960[1800] = { +32767, 32767, 32767, 32766, 32765, +32763, 32761, 32759, 32756, 32753, +32750, 32746, 32742, 32738, 32733, +32728, 32722, 32717, 32710, 32704, +32697, 32690, 32682, 32674, 32666, +32657, 32648, 32639, 32629, 32619, +32609, 32598, 32587, 32576, 32564, +32552, 32539, 32526, 32513, 32500, +32486, 32472, 32457, 32442, 32427, +32411, 32395, 32379, 32362, 32345, +32328, 32310, 32292, 32274, 32255, +32236, 32217, 32197, 32177, 32157, +32136, 32115, 32093, 32071, 32049, +32027, 32004, 31981, 31957, 31933, +31909, 31884, 31859, 31834, 31809, +31783, 31756, 31730, 31703, 31676, +31648, 31620, 31592, 31563, 31534, +31505, 31475, 31445, 31415, 31384, +31353, 31322, 31290, 31258, 31226, +31193, 31160, 31127, 31093, 31059, +31025, 30990, 30955, 30920, 30884, +30848, 30812, 30775, 30738, 30701, +30663, 30625, 30587, 30548, 30509, +30470, 30430, 30390, 30350, 30309, +30269, 30227, 30186, 30144, 30102, +30059, 30016, 29973, 29930, 29886, +29842, 29797, 29752, 29707, 29662, +29616, 29570, 29524, 29477, 29430, +29383, 29335, 29287, 29239, 29190, +29142, 29092, 29043, 28993, 28943, +28892, 28842, 28791, 28739, 28688, +28636, 28583, 28531, 28478, 28425, +28371, 28317, 28263, 28209, 28154, +28099, 28044, 27988, 27932, 27876, +27820, 27763, 27706, 27648, 27591, +27533, 27474, 27416, 27357, 27298, +27238, 27178, 27118, 27058, 26997, +26936, 26875, 26814, 26752, 26690, +26628, 26565, 26502, 26439, 26375, +26312, 26247, 26183, 26119, 26054, +25988, 25923, 25857, 25791, 25725, +25658, 25592, 25524, 25457, 25389, +25322, 25253, 25185, 25116, 25047, +24978, 24908, 24838, 24768, 24698, +24627, 24557, 24485, 24414, 24342, +24270, 24198, 24126, 24053, 23980, +23907, 23834, 23760, 23686, 23612, +23537, 23462, 23387, 23312, 23237, +23161, 23085, 23009, 22932, 22856, +22779, 22701, 22624, 22546, 22468, +22390, 22312, 22233, 22154, 22075, +21996, 21916, 21836, 21756, 21676, +21595, 21515, 21434, 21352, 21271, +21189, 21107, 21025, 20943, 20860, +20777, 20694, 20611, 20528, 20444, +20360, 20276, 20192, 20107, 20022, +19937, 19852, 19767, 19681, 19595, +19509, 19423, 19336, 19250, 19163, +19076, 18988, 18901, 18813, 18725, +18637, 18549, 18460, 18372, 18283, +18194, 18104, 18015, 17925, 17835, +17745, 17655, 17565, 17474, 17383, +17292, 17201, 17110, 17018, 16927, +16835, 16743, 16650, 16558, 16465, +16372, 16279, 16186, 16093, 15999, +15906, 15812, 15718, 15624, 15529, +15435, 15340, 15245, 15150, 15055, +14960, 14864, 14769, 14673, 14577, +14481, 14385, 14288, 14192, 14095, +13998, 13901, 13804, 13706, 13609, +13511, 13414, 13316, 13218, 13119, +13021, 12923, 12824, 12725, 12626, +12527, 12428, 12329, 12230, 12130, +12030, 11930, 11831, 11730, 11630, +11530, 11430, 11329, 11228, 11128, +11027, 10926, 10824, 10723, 10622, +10520, 10419, 10317, 10215, 10113, +10011, 9909, 9807, 9704, 9602, +9499, 9397, 9294, 9191, 9088, +8985, 8882, 8778, 8675, 8572, +8468, 8364, 8261, 8157, 8053, +7949, 7845, 7741, 7637, 7532, +7428, 7323, 7219, 7114, 7009, +6905, 6800, 6695, 6590, 6485, +6380, 6274, 6169, 6064, 5958, +5853, 5747, 5642, 5536, 5430, +5325, 5219, 5113, 5007, 4901, +4795, 4689, 4583, 4476, 4370, +4264, 4157, 4051, 3945, 3838, +3732, 3625, 3518, 3412, 3305, +3198, 3092, 2985, 2878, 2771, +2664, 2558, 2451, 2344, 2237, +2130, 2023, 1916, 1809, 1702, +1594, 1487, 1380, 1273, 1166, +1059, 952, 844, 737, 630, +523, 416, 308, 201, 94, +-13, -121, -228, -335, -442, +-550, -657, -764, -871, -978, +-1086, -1193, -1300, -1407, -1514, +-1621, -1728, -1835, -1942, -2049, +-2157, -2263, -2370, -2477, -2584, +-2691, -2798, -2905, -3012, -3118, +-3225, -3332, -3439, -3545, -3652, +-3758, -3865, -3971, -4078, -4184, +-4290, -4397, -4503, -4609, -4715, +-4821, -4927, -5033, -5139, -5245, +-5351, -5457, -5562, -5668, -5774, +-5879, -5985, -6090, -6195, -6301, +-6406, -6511, -6616, -6721, -6826, +-6931, -7036, -7140, -7245, -7349, +-7454, -7558, -7663, -7767, -7871, +-7975, -8079, -8183, -8287, -8390, +-8494, -8597, -8701, -8804, -8907, +-9011, -9114, -9217, -9319, -9422, +-9525, -9627, -9730, -9832, -9934, +-10037, -10139, -10241, -10342, -10444, +-10546, -10647, -10748, -10850, -10951, +-11052, -11153, -11253, -11354, -11455, +-11555, -11655, -11756, -11856, -11955, +-12055, -12155, -12254, -12354, -12453, +-12552, -12651, -12750, -12849, -12947, +-13046, -13144, -13242, -13340, -13438, +-13536, -13633, -13731, -13828, -13925, +-14022, -14119, -14216, -14312, -14409, +-14505, -14601, -14697, -14793, -14888, +-14984, -15079, -15174, -15269, -15364, +-15459, -15553, -15647, -15741, -15835, +-15929, -16023, -16116, -16210, -16303, +-16396, -16488, -16581, -16673, -16766, +-16858, -16949, -17041, -17133, -17224, +-17315, -17406, -17497, -17587, -17678, +-17768, -17858, -17948, -18037, -18127, +-18216, -18305, -18394, -18483, -18571, +-18659, -18747, -18835, -18923, -19010, +-19098, -19185, -19271, -19358, -19444, +-19531, -19617, -19702, -19788, -19873, +-19959, -20043, -20128, -20213, -20297, +-20381, -20465, -20549, -20632, -20715, +-20798, -20881, -20963, -21046, -21128, +-21210, -21291, -21373, -21454, -21535, +-21616, -21696, -21776, -21856, -21936, +-22016, -22095, -22174, -22253, -22331, +-22410, -22488, -22566, -22643, -22721, +-22798, -22875, -22951, -23028, -23104, +-23180, -23256, -23331, -23406, -23481, +-23556, -23630, -23704, -23778, -23852, +-23925, -23998, -24071, -24144, -24216, +-24288, -24360, -24432, -24503, -24574, +-24645, -24716, -24786, -24856, -24926, +-24995, -25064, -25133, -25202, -25270, +-25339, -25406, -25474, -25541, -25608, +-25675, -25742, -25808, -25874, -25939, +-26005, -26070, -26135, -26199, -26264, +-26327, -26391, -26455, -26518, -26581, +-26643, -26705, -26767, -26829, -26891, +-26952, -27013, -27073, -27133, -27193, +-27253, -27312, -27372, -27430, -27489, +-27547, -27605, -27663, -27720, -27777, +-27834, -27890, -27946, -28002, -28058, +-28113, -28168, -28223, -28277, -28331, +-28385, -28438, -28491, -28544, -28596, +-28649, -28701, -28752, -28803, -28854, +-28905, -28955, -29006, -29055, -29105, +-29154, -29203, -29251, -29299, -29347, +-29395, -29442, -29489, -29535, -29582, +-29628, -29673, -29719, -29764, -29808, +-29853, -29897, -29941, -29984, -30027, +-30070, -30112, -30154, -30196, -30238, +-30279, -30320, -30360, -30400, -30440, +-30480, -30519, -30558, -30596, -30635, +-30672, -30710, -30747, -30784, -30821, +-30857, -30893, -30929, -30964, -30999, +-31033, -31068, -31102, -31135, -31168, +-31201, -31234, -31266, -31298, -31330, +-31361, -31392, -31422, -31453, -31483, +-31512, -31541, -31570, -31599, -31627, +-31655, -31682, -31710, -31737, -31763, +-31789, -31815, -31841, -31866, -31891, +-31915, -31939, -31963, -31986, -32010, +-32032, -32055, -32077, -32099, -32120, +-32141, -32162, -32182, -32202, -32222, +-32241, -32260, -32279, -32297, -32315, +-32333, -32350, -32367, -32383, -32399, +-32415, -32431, -32446, -32461, -32475, +-32489, -32503, -32517, -32530, -32542, +-32555, -32567, -32579, -32590, -32601, +-32612, -32622, -32632, -32641, -32651, +-32659, -32668, -32676, -32684, -32692, +-32699, -32706, -32712, -32718, -32724, +-32729, -32734, -32739, -32743, -32747, +-32751, -32754, -32757, -32760, -32762, +-32764, -32765, -32767, -32767, -32767, +32767, 32767, 32765, 32761, 32756, +32750, 32742, 32732, 32722, 32710, +32696, 32681, 32665, 32647, 32628, +32608, 32586, 32562, 32538, 32512, +32484, 32455, 32425, 32393, 32360, +32326, 32290, 32253, 32214, 32174, +32133, 32090, 32046, 32001, 31954, +31906, 31856, 31805, 31753, 31700, +31645, 31588, 31530, 31471, 31411, +31349, 31286, 31222, 31156, 31089, +31020, 30951, 30880, 30807, 30733, +30658, 30582, 30504, 30425, 30345, +30263, 30181, 30096, 30011, 29924, +29836, 29747, 29656, 29564, 29471, +29377, 29281, 29184, 29086, 28987, +28886, 28784, 28681, 28577, 28471, +28365, 28257, 28147, 28037, 27925, +27812, 27698, 27583, 27467, 27349, +27231, 27111, 26990, 26868, 26744, +26620, 26494, 26367, 26239, 26110, +25980, 25849, 25717, 25583, 25449, +25313, 25176, 25038, 24900, 24760, +24619, 24477, 24333, 24189, 24044, +23898, 23751, 23602, 23453, 23303, +23152, 22999, 22846, 22692, 22537, +22380, 22223, 22065, 21906, 21746, +21585, 21423, 21261, 21097, 20933, +20767, 20601, 20434, 20265, 20096, +19927, 19756, 19584, 19412, 19239, +19065, 18890, 18714, 18538, 18361, +18183, 18004, 17824, 17644, 17463, +17281, 17098, 16915, 16731, 16546, +16361, 16175, 15988, 15800, 15612, +15423, 15234, 15043, 14852, 14661, +14469, 14276, 14083, 13889, 13694, +13499, 13303, 13107, 12910, 12713, +12515, 12317, 12118, 11918, 11718, +11517, 11316, 11115, 10913, 10710, +10508, 10304, 10100, 9896, 9691, +9486, 9281, 9075, 8869, 8662, +8455, 8248, 8040, 7832, 7623, +7415, 7206, 6996, 6787, 6577, +6366, 6156, 5945, 5734, 5523, +5311, 5100, 4888, 4675, 4463, +4251, 4038, 3825, 3612, 3399, +3185, 2972, 2758, 2544, 2330, +2116, 1902, 1688, 1474, 1260, +1045, 831, 617, 402, 188, +-27, -241, -456, -670, -885, +-1099, -1313, -1528, -1742, -1956, +-2170, -2384, -2598, -2811, -3025, +-3239, -3452, -3665, -3878, -4091, +-4304, -4516, -4728, -4941, -5153, +-5364, -5576, -5787, -5998, -6209, +-6419, -6629, -6839, -7049, -7258, +-7467, -7676, -7884, -8092, -8300, +-8507, -8714, -8920, -9127, -9332, +-9538, -9743, -9947, -10151, -10355, +-10558, -10761, -10963, -11165, -11367, +-11568, -11768, -11968, -12167, -12366, +-12565, -12762, -12960, -13156, -13352, +-13548, -13743, -13937, -14131, -14324, +-14517, -14709, -14900, -15091, -15281, +-15470, -15659, -15847, -16035, -16221, +-16407, -16593, -16777, -16961, -17144, +-17326, -17508, -17689, -17869, -18049, +-18227, -18405, -18582, -18758, -18934, +-19108, -19282, -19455, -19627, -19799, +-19969, -20139, -20308, -20475, -20642, +-20809, -20974, -21138, -21301, -21464, +-21626, -21786, -21946, -22105, -22263, +-22420, -22575, -22730, -22884, -23037, +-23189, -23340, -23490, -23640, -23788, +-23935, -24080, -24225, -24369, -24512, +-24654, -24795, -24934, -25073, -25211, +-25347, -25482, -25617, -25750, -25882, +-26013, -26143, -26272, -26399, -26526, +-26651, -26775, -26898, -27020, -27141, +-27260, -27379, -27496, -27612, -27727, +-27841, -27953, -28065, -28175, -28284, +-28391, -28498, -28603, -28707, -28810, +-28911, -29012, -29111, -29209, -29305, +-29401, -29495, -29587, -29679, -29769, +-29858, -29946, -30032, -30118, -30201, +-30284, -30365, -30445, -30524, -30601, +-30677, -30752, -30825, -30897, -30968, +-31038, -31106, -31172, -31238, -31302, +-31365, -31426, -31486, -31545, -31602, +-31658, -31713, -31766, -31818, -31869, +-31918, -31966, -32012, -32058, -32101, +-32144, -32185, -32224, -32262, -32299, +-32335, -32369, -32401, -32433, -32463, +-32491, -32518, -32544, -32568, -32591, +-32613, -32633, -32652, -32669, -32685, +-32700, -32713, -32724, -32735, -32744, +-32751, -32757, -32762, -32766, -32767, +32767, 32764, 32755, 32741, 32720, +32694, 32663, 32626, 32583, 32535, +32481, 32421, 32356, 32286, 32209, +32128, 32041, 31948, 31850, 31747, +31638, 31523, 31403, 31278, 31148, +31012, 30871, 30724, 30572, 30415, +30253, 30086, 29913, 29736, 29553, +29365, 29172, 28974, 28771, 28564, +28351, 28134, 27911, 27684, 27452, +27216, 26975, 26729, 26478, 26223, +25964, 25700, 25432, 25159, 24882, +24601, 24315, 24026, 23732, 23434, +23133, 22827, 22517, 22204, 21886, +21565, 21240, 20912, 20580, 20244, +19905, 19563, 19217, 18868, 18516, +18160, 17802, 17440, 17075, 16708, +16338, 15964, 15588, 15210, 14829, +14445, 14059, 13670, 13279, 12886, +12490, 12093, 11693, 11291, 10888, +10482, 10075, 9666, 9255, 8843, +8429, 8014, 7597, 7180, 6760, +6340, 5919, 5496, 5073, 4649, +4224, 3798, 3372, 2945, 2517, +2090, 1661, 1233, 804, 375, +-54, -483, -911, -1340, -1768, +-2197, -2624, -3052, -3479, -3905, +-4330, -4755, -5179, -5602, -6024, +-6445, -6865, -7284, -7702, -8118, +-8533, -8946, -9358, -9768, -10177, +-10584, -10989, -11392, -11793, -12192, +-12589, -12984, -13377, -13767, -14155, +-14541, -14924, -15305, -15683, -16058, +-16430, -16800, -17167, -17531, -17892, +-18249, -18604, -18956, -19304, -19649, +-19990, -20329, -20663, -20994, -21322, +-21646, -21966, -22282, -22595, -22904, +-23208, -23509, -23806, -24099, -24387, +-24672, -24952, -25228, -25499, -25766, +-26029, -26288, -26541, -26791, -27035, +-27275, -27511, -27741, -27967, -28188, +-28405, -28616, -28823, -29024, -29221, +-29412, -29599, -29780, -29957, -30128, +-30294, -30455, -30611, -30761, -30906, +-31046, -31181, -31310, -31434, -31552, +-31665, -31773, -31875, -31972, -32063, +-32149, -32229, -32304, -32373, -32437, +-32495, -32547, -32594, -32635, -32671, +-32701, -32726, -32745, -32758, -32766, +32767, 32754, 32717, 32658, 32577, +32473, 32348, 32200, 32029, 31837, +31624, 31388, 31131, 30853, 30553, +30232, 29891, 29530, 29148, 28746, +28324, 27883, 27423, 26944, 26447, +25931, 25398, 24847, 24279, 23695, +23095, 22478, 21846, 21199, 20538, +19863, 19174, 18472, 17757, 17030, +16291, 15541, 14781, 14010, 13230, +12441, 11643, 10837, 10024, 9204, +8377, 7545, 6708, 5866, 5020, +4171, 3319, 2464, 1608, 751, +-107, -965, -1822, -2678, -3532, +-4383, -5232, -6077, -6918, -7754, +-8585, -9409, -10228, -11039, -11843, +-12639, -13426, -14204, -14972, -15730, +-16477, -17213, -17937, -18648, -19347, +-20033, -20705, -21363, -22006, -22634, +-23246, -23843, -24423, -24986, -25533, +-26062, -26573, -27066, -27540, -27995, +-28431, -28848, -29245, -29622, -29979, +-30315, -30630, -30924, -31197, -31449, +-31679, -31887, -32074, -32239, -32381, +-32501, -32600, -32675, -32729, -32759, +}; +#endif + +static const CELTMode mode48000_960_120 = { +48000, /* Fs */ +120, /* overlap */ +21, /* nbEBands */ +21, /* effEBands */ +{27853, 0, 4096, 8192, }, /* preemph */ +eband5ms, /* eBands */ +3, /* maxLM */ +8, /* nbShortMdcts */ +120, /* shortMdctSize */ +11, /* nbAllocVectors */ +band_allocation, /* allocVectors */ +logN400, /* logN */ +window120, /* window */ +{1920, 3, {&fft_state48000_960_0, &fft_state48000_960_1, &fft_state48000_960_2, &fft_state48000_960_3, }, mdct_twiddles960}, /* mdct */ +{392, cache_index50, cache_bits50, cache_caps50}, /* cache */ +}; + +/* List of all the available modes */ +#define TOTAL_MODES 1 +static const CELTMode * const static_mode_list[TOTAL_MODES] = { +&mode48000_960_120, +}; diff --git a/media/libopus/celt/static_modes_fixed_arm_ne10.h b/media/libopus/celt/static_modes_fixed_arm_ne10.h new file mode 100644 index 0000000000..7623092192 --- /dev/null +++ b/media/libopus/celt/static_modes_fixed_arm_ne10.h @@ -0,0 +1,388 @@ +/* The contents of this file was automatically generated by + * dump_mode_arm_ne10.c with arguments: 48000 960 + * It contains static definitions for some pre-defined modes. */ +#include <NE10_types.h> + +#ifndef NE10_FFT_PARAMS48000_960 +#define NE10_FFT_PARAMS48000_960 +static const ne10_int32_t ne10_factors_480[64] = { +4, 40, 4, 30, 2, 15, 5, 3, 3, 1, 1, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, }; +static const ne10_int32_t ne10_factors_240[64] = { +3, 20, 4, 15, 5, 3, 3, 1, 1, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, }; +static const ne10_int32_t ne10_factors_120[64] = { +3, 10, 2, 15, 5, 3, 3, 1, 1, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, }; +static const ne10_int32_t ne10_factors_60[64] = { +2, 5, 5, 3, 3, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, }; +static const ne10_fft_cpx_int32_t ne10_twiddles_480[480] = { +{0,0}, {2147483647,0}, {2147483647,0}, +{2147483647,0}, {1961823921,-873460313}, {1436946998,-1595891394}, +{2147483647,0}, {1436946998,-1595891394}, {-224473265,-2135719496}, +{2147483647,0}, {663608871,-2042378339}, {-1737350854,-1262259096}, +{2147483647,0}, {-224473265,-2135719496}, {-2100555935,446487152}, +{2147483647,0}, {2100555974,-446486968}, {1961823921,-873460313}, +{1737350743,-1262259248}, {1436946998,-1595891394}, {1073741769,-1859775424}, +{663608871,-2042378339}, {224473078,-2135719516}, {-224473265,-2135719496}, +{-663609049,-2042378281}, {-1073741932,-1859775330}, {-1436947137,-1595891268}, +{-1737350854,-1262259096}, {-1961823997,-873460141}, {-2100556013,-446486785}, +{2147483647,0}, {2144540595,-112390613}, {2135719506,-224473172}, +{2121044558,-335940465}, {2100555974,-446486968}, {2074309912,-555809682}, +{2042378310,-663608960}, {2004848691,-769589332}, {1961823921,-873460313}, +{1913421927,-974937199}, {1859775377,-1073741851}, {1801031311,-1169603450}, +{1737350743,-1262259248}, {1668908218,-1351455280}, {1595891331,-1436947067}, +{1518500216,-1518500282}, {1436946998,-1595891394}, {1351455207,-1668908277}, +{1262259172,-1737350799}, {1169603371,-1801031362}, {1073741769,-1859775424}, +{974937230,-1913421912}, {873460227,-1961823959}, {769589125,-2004848771}, +{663608871,-2042378339}, {555809715,-2074309903}, {446486876,-2100555994}, +{335940246,-2121044593}, {224473078,-2135719516}, {112390647,-2144540593}, +{2147483647,0}, {2135719506,-224473172}, {2100555974,-446486968}, +{2042378310,-663608960}, {1961823921,-873460313}, {1859775377,-1073741851}, +{1737350743,-1262259248}, {1595891331,-1436947067}, {1436946998,-1595891394}, +{1262259172,-1737350799}, {1073741769,-1859775424}, {873460227,-1961823959}, +{663608871,-2042378339}, {446486876,-2100555994}, {224473078,-2135719516}, +{-94,-2147483647}, {-224473265,-2135719496}, {-446487060,-2100555955}, +{-663609049,-2042378281}, {-873460398,-1961823883}, {-1073741932,-1859775330}, +{-1262259116,-1737350839}, {-1436947137,-1595891268}, {-1595891628,-1436946738}, +{-1737350854,-1262259096}, {-1859775343,-1073741910}, {-1961823997,-873460141}, +{-2042378447,-663608538}, {-2100556013,-446486785}, {-2135719499,-224473240}, +{2147483647,0}, {2121044558,-335940465}, {2042378310,-663608960}, +{1913421927,-974937199}, {1737350743,-1262259248}, {1518500216,-1518500282}, +{1262259172,-1737350799}, {974937230,-1913421912}, {663608871,-2042378339}, +{335940246,-2121044593}, {-94,-2147483647}, {-335940431,-2121044564}, +{-663609049,-2042378281}, {-974937397,-1913421827}, {-1262259116,-1737350839}, +{-1518500258,-1518500240}, {-1737350854,-1262259096}, {-1913422071,-974936918}, +{-2042378447,-663608538}, {-2121044568,-335940406}, {-2147483647,188}, +{-2121044509,335940777}, {-2042378331,663608895}, {-1913421900,974937252}, +{-1737350633,1262259400}, {-1518499993,1518500506}, {-1262258813,1737351059}, +{-974936606,1913422229}, {-663609179,2042378239}, {-335940566,2121044542}, +{2147483647,0}, {2147299667,-28109693}, {2146747758,-56214570}, +{2145828015,-84309815}, {2144540595,-112390613}, {2142885719,-140452154}, +{2140863671,-168489630}, {2138474797,-196498235}, {2135719506,-224473172}, +{2132598271,-252409646}, {2129111626,-280302871}, {2125260168,-308148068}, +{2121044558,-335940465}, {2116465518,-363675300}, {2111523833,-391347822}, +{2106220349,-418953288}, {2100555974,-446486968}, {2094531681,-473944146}, +{2088148500,-501320115}, {2081407525,-528610186}, {2074309912,-555809682}, +{2066856885,-582913912}, {2059049696,-609918325}, {2050889698,-636818231}, +{2042378310,-663608960}, {2033516972,-690285983}, {2024307180,-716844791}, +{2014750533,-743280770}, {2004848691,-769589332}, {1994603329,-795766029}, +{1984016179,-821806435}, {1973089077,-847706028}, {1961823921,-873460313}, +{1950222618,-899064934}, {1938287127,-924515564}, {1926019520,-949807783}, +{1913421927,-974937199}, {1900496481,-999899565}, {1887245364,-1024690661}, +{1873670877,-1049306180}, {1859775377,-1073741851}, {1845561215,-1097993541}, +{1831030826,-1122057097}, {1816186632,-1145928502}, {1801031311,-1169603450}, +{1785567394,-1193077993}, {1769797456,-1216348214}, {1753724345,-1239409914}, +{1737350743,-1262259248}, {1720679456,-1284892300}, {1703713340,-1307305194}, +{1686455222,-1329494189}, {1668908218,-1351455280}, {1651075255,-1373184807}, +{1632959307,-1394679144}, {1614563642,-1415934412}, {1595891331,-1436947067}, +{1576945572,-1457713510}, {1557729613,-1478230181}, {1538246655,-1498493658}, +{1518500216,-1518500282}, {1498493590,-1538246721}, {1478230113,-1557729677}, +{1457713441,-1576945636}, {1436946998,-1595891394}, {1415934341,-1614563704}, +{1394679073,-1632959368}, {1373184735,-1651075315}, {1351455207,-1668908277}, +{1329494115,-1686455280}, {1307305120,-1703713397}, {1284892225,-1720679512}, +{1262259172,-1737350799}, {1239409837,-1753724400}, {1216348136,-1769797510}, +{1193077915,-1785567446}, {1169603371,-1801031362}, {1145928423,-1816186682}, +{1122057017,-1831030875}, {1097993571,-1845561197}, {1073741769,-1859775424}, +{1049305987,-1873670985}, {1024690635,-1887245378}, {999899482,-1900496524}, +{974937230,-1913421912}, {949807699,-1926019561}, {924515422,-1938287195}, +{899064965,-1950222603}, {873460227,-1961823959}, {847705824,-1973089164}, +{821806407,-1984016190}, {795765941,-1994603364}, {769589125,-2004848771}, +{743280682,-2014750566}, {716844642,-2024307233}, {690286016,-2033516961}, +{663608871,-2042378339}, {636818019,-2050889764}, {609918296,-2059049705}, +{582913822,-2066856911}, {555809715,-2074309903}, {528610126,-2081407540}, +{501319962,-2088148536}, {473944148,-2094531680}, {446486876,-2100555994}, +{418953102,-2106220386}, {391347792,-2111523838}, {363675176,-2116465540}, +{335940246,-2121044593}, {308148006,-2125260177}, {280302715,-2129111646}, +{252409648,-2132598271}, {224473078,-2135719516}, {196498046,-2138474814}, +{168489600,-2140863674}, {140452029,-2142885728}, {112390647,-2144540593}, +{84309753,-2145828017}, {56214412,-2146747762}, {28109695,-2147299667}, +{2147483647,0}, {2146747758,-56214570}, {2144540595,-112390613}, +{2140863671,-168489630}, {2135719506,-224473172}, {2129111626,-280302871}, +{2121044558,-335940465}, {2111523833,-391347822}, {2100555974,-446486968}, +{2088148500,-501320115}, {2074309912,-555809682}, {2059049696,-609918325}, +{2042378310,-663608960}, {2024307180,-716844791}, {2004848691,-769589332}, +{1984016179,-821806435}, {1961823921,-873460313}, {1938287127,-924515564}, +{1913421927,-974937199}, {1887245364,-1024690661}, {1859775377,-1073741851}, +{1831030826,-1122057097}, {1801031311,-1169603450}, {1769797456,-1216348214}, +{1737350743,-1262259248}, {1703713340,-1307305194}, {1668908218,-1351455280}, +{1632959307,-1394679144}, {1595891331,-1436947067}, {1557729613,-1478230181}, +{1518500216,-1518500282}, {1478230113,-1557729677}, {1436946998,-1595891394}, +{1394679073,-1632959368}, {1351455207,-1668908277}, {1307305120,-1703713397}, +{1262259172,-1737350799}, {1216348136,-1769797510}, {1169603371,-1801031362}, +{1122057017,-1831030875}, {1073741769,-1859775424}, {1024690635,-1887245378}, +{974937230,-1913421912}, {924515422,-1938287195}, {873460227,-1961823959}, +{821806407,-1984016190}, {769589125,-2004848771}, {716844642,-2024307233}, +{663608871,-2042378339}, {609918296,-2059049705}, {555809715,-2074309903}, +{501319962,-2088148536}, {446486876,-2100555994}, {391347792,-2111523838}, +{335940246,-2121044593}, {280302715,-2129111646}, {224473078,-2135719516}, +{168489600,-2140863674}, {112390647,-2144540593}, {56214412,-2146747762}, +{-94,-2147483647}, {-56214600,-2146747757}, {-112390835,-2144540584}, +{-168489787,-2140863659}, {-224473265,-2135719496}, {-280302901,-2129111622}, +{-335940431,-2121044564}, {-391347977,-2111523804}, {-446487060,-2100555955}, +{-501320144,-2088148493}, {-555809896,-2074309855}, {-609918476,-2059049651}, +{-663609049,-2042378281}, {-716844819,-2024307170}, {-769589300,-2004848703}, +{-821806581,-1984016118}, {-873460398,-1961823883}, {-924515591,-1938287114}, +{-974937397,-1913421827}, {-1024690575,-1887245411}, {-1073741932,-1859775330}, +{-1122057395,-1831030643}, {-1169603421,-1801031330}, {-1216348291,-1769797403}, +{-1262259116,-1737350839}, {-1307305268,-1703713283}, {-1351455453,-1668908078}, +{-1394679021,-1632959413}, {-1436947137,-1595891268}, {-1478230435,-1557729372}, +{-1518500258,-1518500240}, {-1557729742,-1478230045}, {-1595891628,-1436946738}, +{-1632959429,-1394679001}, {-1668908417,-1351455035}, {-1703713298,-1307305248}, +{-1737350854,-1262259096}, {-1769797708,-1216347848}, {-1801031344,-1169603400}, +{-1831030924,-1122056937}, {-1859775343,-1073741910}, {-1887245423,-1024690552}, +{-1913422071,-974936918}, {-1938287125,-924515568}, {-1961823997,-873460141}, +{-1984016324,-821806084}, {-2004848713,-769589276}, {-2024307264,-716844553}, +{-2042378447,-663608538}, {-2059049731,-609918206}, {-2074309994,-555809377}, +{-2088148499,-501320119}, {-2100556013,-446486785}, {-2111523902,-391347448}, +{-2121044568,-335940406}, {-2129111659,-280302621}, {-2135719499,-224473240}, +{-2140863681,-168489506}, {-2144540612,-112390298}, {-2146747758,-56214574}, +{2147483647,0}, {2145828015,-84309815}, {2140863671,-168489630}, +{2132598271,-252409646}, {2121044558,-335940465}, {2106220349,-418953288}, +{2088148500,-501320115}, {2066856885,-582913912}, {2042378310,-663608960}, +{2014750533,-743280770}, {1984016179,-821806435}, {1950222618,-899064934}, +{1913421927,-974937199}, {1873670877,-1049306180}, {1831030826,-1122057097}, +{1785567394,-1193077993}, {1737350743,-1262259248}, {1686455222,-1329494189}, +{1632959307,-1394679144}, {1576945572,-1457713510}, {1518500216,-1518500282}, +{1457713441,-1576945636}, {1394679073,-1632959368}, {1329494115,-1686455280}, +{1262259172,-1737350799}, {1193077915,-1785567446}, {1122057017,-1831030875}, +{1049305987,-1873670985}, {974937230,-1913421912}, {899064965,-1950222603}, +{821806407,-1984016190}, {743280682,-2014750566}, {663608871,-2042378339}, +{582913822,-2066856911}, {501319962,-2088148536}, {418953102,-2106220386}, +{335940246,-2121044593}, {252409648,-2132598271}, {168489600,-2140863674}, +{84309753,-2145828017}, {-94,-2147483647}, {-84309940,-2145828010}, +{-168489787,-2140863659}, {-252409834,-2132598249}, {-335940431,-2121044564}, +{-418953286,-2106220349}, {-501320144,-2088148493}, {-582914003,-2066856860}, +{-663609049,-2042378281}, {-743280858,-2014750501}, {-821806581,-1984016118}, +{-899065136,-1950222525}, {-974937397,-1913421827}, {-1049306374,-1873670768}, +{-1122057395,-1831030643}, {-1193078284,-1785567199}, {-1262259116,-1737350839}, +{-1329494061,-1686455323}, {-1394679021,-1632959413}, {-1457713485,-1576945595}, +{-1518500258,-1518500240}, {-1576945613,-1457713466}, {-1632959429,-1394679001}, +{-1686455338,-1329494041}, {-1737350854,-1262259096}, {-1785567498,-1193077837}, +{-1831030924,-1122056937}, {-1873671031,-1049305905}, {-1913422071,-974936918}, +{-1950222750,-899064648}, {-1984016324,-821806084}, {-2014750687,-743280354}, +{-2042378447,-663608538}, {-2066856867,-582913978}, {-2088148499,-501320119}, +{-2106220354,-418953261}, {-2121044568,-335940406}, {-2132598282,-252409555}, +{-2140863681,-168489506}, {-2145828021,-84309659}, {-2147483647,188}, +{-2145828006,84310034}, {-2140863651,168489881}, {-2132598237,252409928}, +{-2121044509,335940777}, {-2106220281,418953629}, {-2088148411,501320484}, +{-2066856765,582914339}, {-2042378331,663608895}, {-2014750557,743280706}, +{-1984016181,821806431}, {-1950222593,899064989}, {-1913421900,974937252}, +{-1873670848,1049306232}, {-1831030728,1122057257}, {-1785567289,1193078149}, +{-1737350633,1262259400}, {-1686455106,1329494336}, {-1632959185,1394679287}, +{-1576945358,1457713742}, {-1518499993,1518500506}, {-1457713209,1576945850}, +{-1394678735,1632959656}, {-1329493766,1686455555}, {-1262258813,1737351059}, +{-1193077546,1785567692}, {-1122056638,1831031107}, {-1049305599,1873671202}, +{-974936606,1913422229}, {-899064330,1950222896}, {-821805761,1984016458}, +{-743280025,2014750808}, {-663609179,2042378239}, {-582914134,2066856823}, +{-501320277,2088148461}, {-418953420,2106220322}, {-335940566,2121044542}, +{-252409716,2132598263}, {-168489668,2140863668}, {-84309821,2145828015}, +}; +static const ne10_fft_cpx_int32_t ne10_twiddles_240[240] = { +{0,0}, {2147483647,0}, {2147483647,0}, +{2147483647,0}, {1961823921,-873460313}, {1436946998,-1595891394}, +{2147483647,0}, {1436946998,-1595891394}, {-224473265,-2135719496}, +{2147483647,0}, {663608871,-2042378339}, {-1737350854,-1262259096}, +{2147483647,0}, {-224473265,-2135719496}, {-2100555935,446487152}, +{2147483647,0}, {2135719506,-224473172}, {2100555974,-446486968}, +{2042378310,-663608960}, {1961823921,-873460313}, {1859775377,-1073741851}, +{1737350743,-1262259248}, {1595891331,-1436947067}, {1436946998,-1595891394}, +{1262259172,-1737350799}, {1073741769,-1859775424}, {873460227,-1961823959}, +{663608871,-2042378339}, {446486876,-2100555994}, {224473078,-2135719516}, +{2147483647,0}, {2100555974,-446486968}, {1961823921,-873460313}, +{1737350743,-1262259248}, {1436946998,-1595891394}, {1073741769,-1859775424}, +{663608871,-2042378339}, {224473078,-2135719516}, {-224473265,-2135719496}, +{-663609049,-2042378281}, {-1073741932,-1859775330}, {-1436947137,-1595891268}, +{-1737350854,-1262259096}, {-1961823997,-873460141}, {-2100556013,-446486785}, +{2147483647,0}, {2042378310,-663608960}, {1737350743,-1262259248}, +{1262259172,-1737350799}, {663608871,-2042378339}, {-94,-2147483647}, +{-663609049,-2042378281}, {-1262259116,-1737350839}, {-1737350854,-1262259096}, +{-2042378447,-663608538}, {-2147483647,188}, {-2042378331,663608895}, +{-1737350633,1262259400}, {-1262258813,1737351059}, {-663609179,2042378239}, +{2147483647,0}, {2146747758,-56214570}, {2144540595,-112390613}, +{2140863671,-168489630}, {2135719506,-224473172}, {2129111626,-280302871}, +{2121044558,-335940465}, {2111523833,-391347822}, {2100555974,-446486968}, +{2088148500,-501320115}, {2074309912,-555809682}, {2059049696,-609918325}, +{2042378310,-663608960}, {2024307180,-716844791}, {2004848691,-769589332}, +{1984016179,-821806435}, {1961823921,-873460313}, {1938287127,-924515564}, +{1913421927,-974937199}, {1887245364,-1024690661}, {1859775377,-1073741851}, +{1831030826,-1122057097}, {1801031311,-1169603450}, {1769797456,-1216348214}, +{1737350743,-1262259248}, {1703713340,-1307305194}, {1668908218,-1351455280}, +{1632959307,-1394679144}, {1595891331,-1436947067}, {1557729613,-1478230181}, +{1518500216,-1518500282}, {1478230113,-1557729677}, {1436946998,-1595891394}, +{1394679073,-1632959368}, {1351455207,-1668908277}, {1307305120,-1703713397}, +{1262259172,-1737350799}, {1216348136,-1769797510}, {1169603371,-1801031362}, +{1122057017,-1831030875}, {1073741769,-1859775424}, {1024690635,-1887245378}, +{974937230,-1913421912}, {924515422,-1938287195}, {873460227,-1961823959}, +{821806407,-1984016190}, {769589125,-2004848771}, {716844642,-2024307233}, +{663608871,-2042378339}, {609918296,-2059049705}, {555809715,-2074309903}, +{501319962,-2088148536}, {446486876,-2100555994}, {391347792,-2111523838}, +{335940246,-2121044593}, {280302715,-2129111646}, {224473078,-2135719516}, +{168489600,-2140863674}, {112390647,-2144540593}, {56214412,-2146747762}, +{2147483647,0}, {2144540595,-112390613}, {2135719506,-224473172}, +{2121044558,-335940465}, {2100555974,-446486968}, {2074309912,-555809682}, +{2042378310,-663608960}, {2004848691,-769589332}, {1961823921,-873460313}, +{1913421927,-974937199}, {1859775377,-1073741851}, {1801031311,-1169603450}, +{1737350743,-1262259248}, {1668908218,-1351455280}, {1595891331,-1436947067}, +{1518500216,-1518500282}, {1436946998,-1595891394}, {1351455207,-1668908277}, +{1262259172,-1737350799}, {1169603371,-1801031362}, {1073741769,-1859775424}, +{974937230,-1913421912}, {873460227,-1961823959}, {769589125,-2004848771}, +{663608871,-2042378339}, {555809715,-2074309903}, {446486876,-2100555994}, +{335940246,-2121044593}, {224473078,-2135719516}, {112390647,-2144540593}, +{-94,-2147483647}, {-112390835,-2144540584}, {-224473265,-2135719496}, +{-335940431,-2121044564}, {-446487060,-2100555955}, {-555809896,-2074309855}, +{-663609049,-2042378281}, {-769589300,-2004848703}, {-873460398,-1961823883}, +{-974937397,-1913421827}, {-1073741932,-1859775330}, {-1169603421,-1801031330}, +{-1262259116,-1737350839}, {-1351455453,-1668908078}, {-1436947137,-1595891268}, +{-1518500258,-1518500240}, {-1595891628,-1436946738}, {-1668908417,-1351455035}, +{-1737350854,-1262259096}, {-1801031344,-1169603400}, {-1859775343,-1073741910}, +{-1913422071,-974936918}, {-1961823997,-873460141}, {-2004848713,-769589276}, +{-2042378447,-663608538}, {-2074309994,-555809377}, {-2100556013,-446486785}, +{-2121044568,-335940406}, {-2135719499,-224473240}, {-2144540612,-112390298}, +{2147483647,0}, {2140863671,-168489630}, {2121044558,-335940465}, +{2088148500,-501320115}, {2042378310,-663608960}, {1984016179,-821806435}, +{1913421927,-974937199}, {1831030826,-1122057097}, {1737350743,-1262259248}, +{1632959307,-1394679144}, {1518500216,-1518500282}, {1394679073,-1632959368}, +{1262259172,-1737350799}, {1122057017,-1831030875}, {974937230,-1913421912}, +{821806407,-1984016190}, {663608871,-2042378339}, {501319962,-2088148536}, +{335940246,-2121044593}, {168489600,-2140863674}, {-94,-2147483647}, +{-168489787,-2140863659}, {-335940431,-2121044564}, {-501320144,-2088148493}, +{-663609049,-2042378281}, {-821806581,-1984016118}, {-974937397,-1913421827}, +{-1122057395,-1831030643}, {-1262259116,-1737350839}, {-1394679021,-1632959413}, +{-1518500258,-1518500240}, {-1632959429,-1394679001}, {-1737350854,-1262259096}, +{-1831030924,-1122056937}, {-1913422071,-974936918}, {-1984016324,-821806084}, +{-2042378447,-663608538}, {-2088148499,-501320119}, {-2121044568,-335940406}, +{-2140863681,-168489506}, {-2147483647,188}, {-2140863651,168489881}, +{-2121044509,335940777}, {-2088148411,501320484}, {-2042378331,663608895}, +{-1984016181,821806431}, {-1913421900,974937252}, {-1831030728,1122057257}, +{-1737350633,1262259400}, {-1632959185,1394679287}, {-1518499993,1518500506}, +{-1394678735,1632959656}, {-1262258813,1737351059}, {-1122056638,1831031107}, +{-974936606,1913422229}, {-821805761,1984016458}, {-663609179,2042378239}, +{-501320277,2088148461}, {-335940566,2121044542}, {-168489668,2140863668}, +}; +static const ne10_fft_cpx_int32_t ne10_twiddles_120[120] = { +{0,0}, {2147483647,0}, {2147483647,0}, +{2147483647,0}, {1961823921,-873460313}, {1436946998,-1595891394}, +{2147483647,0}, {1436946998,-1595891394}, {-224473265,-2135719496}, +{2147483647,0}, {663608871,-2042378339}, {-1737350854,-1262259096}, +{2147483647,0}, {-224473265,-2135719496}, {-2100555935,446487152}, +{2147483647,0}, {2100555974,-446486968}, {1961823921,-873460313}, +{1737350743,-1262259248}, {1436946998,-1595891394}, {1073741769,-1859775424}, +{663608871,-2042378339}, {224473078,-2135719516}, {-224473265,-2135719496}, +{-663609049,-2042378281}, {-1073741932,-1859775330}, {-1436947137,-1595891268}, +{-1737350854,-1262259096}, {-1961823997,-873460141}, {-2100556013,-446486785}, +{2147483647,0}, {2144540595,-112390613}, {2135719506,-224473172}, +{2121044558,-335940465}, {2100555974,-446486968}, {2074309912,-555809682}, +{2042378310,-663608960}, {2004848691,-769589332}, {1961823921,-873460313}, +{1913421927,-974937199}, {1859775377,-1073741851}, {1801031311,-1169603450}, +{1737350743,-1262259248}, {1668908218,-1351455280}, {1595891331,-1436947067}, +{1518500216,-1518500282}, {1436946998,-1595891394}, {1351455207,-1668908277}, +{1262259172,-1737350799}, {1169603371,-1801031362}, {1073741769,-1859775424}, +{974937230,-1913421912}, {873460227,-1961823959}, {769589125,-2004848771}, +{663608871,-2042378339}, {555809715,-2074309903}, {446486876,-2100555994}, +{335940246,-2121044593}, {224473078,-2135719516}, {112390647,-2144540593}, +{2147483647,0}, {2135719506,-224473172}, {2100555974,-446486968}, +{2042378310,-663608960}, {1961823921,-873460313}, {1859775377,-1073741851}, +{1737350743,-1262259248}, {1595891331,-1436947067}, {1436946998,-1595891394}, +{1262259172,-1737350799}, {1073741769,-1859775424}, {873460227,-1961823959}, +{663608871,-2042378339}, {446486876,-2100555994}, {224473078,-2135719516}, +{-94,-2147483647}, {-224473265,-2135719496}, {-446487060,-2100555955}, +{-663609049,-2042378281}, {-873460398,-1961823883}, {-1073741932,-1859775330}, +{-1262259116,-1737350839}, {-1436947137,-1595891268}, {-1595891628,-1436946738}, +{-1737350854,-1262259096}, {-1859775343,-1073741910}, {-1961823997,-873460141}, +{-2042378447,-663608538}, {-2100556013,-446486785}, {-2135719499,-224473240}, +{2147483647,0}, {2121044558,-335940465}, {2042378310,-663608960}, +{1913421927,-974937199}, {1737350743,-1262259248}, {1518500216,-1518500282}, +{1262259172,-1737350799}, {974937230,-1913421912}, {663608871,-2042378339}, +{335940246,-2121044593}, {-94,-2147483647}, {-335940431,-2121044564}, +{-663609049,-2042378281}, {-974937397,-1913421827}, {-1262259116,-1737350839}, +{-1518500258,-1518500240}, {-1737350854,-1262259096}, {-1913422071,-974936918}, +{-2042378447,-663608538}, {-2121044568,-335940406}, {-2147483647,188}, +{-2121044509,335940777}, {-2042378331,663608895}, {-1913421900,974937252}, +{-1737350633,1262259400}, {-1518499993,1518500506}, {-1262258813,1737351059}, +{-974936606,1913422229}, {-663609179,2042378239}, {-335940566,2121044542}, +}; +static const ne10_fft_cpx_int32_t ne10_twiddles_60[60] = { +{0,0}, {2147483647,0}, {2147483647,0}, +{2147483647,0}, {1961823921,-873460313}, {1436946998,-1595891394}, +{2147483647,0}, {1436946998,-1595891394}, {-224473265,-2135719496}, +{2147483647,0}, {663608871,-2042378339}, {-1737350854,-1262259096}, +{2147483647,0}, {-224473265,-2135719496}, {-2100555935,446487152}, +{2147483647,0}, {2135719506,-224473172}, {2100555974,-446486968}, +{2042378310,-663608960}, {1961823921,-873460313}, {1859775377,-1073741851}, +{1737350743,-1262259248}, {1595891331,-1436947067}, {1436946998,-1595891394}, +{1262259172,-1737350799}, {1073741769,-1859775424}, {873460227,-1961823959}, +{663608871,-2042378339}, {446486876,-2100555994}, {224473078,-2135719516}, +{2147483647,0}, {2100555974,-446486968}, {1961823921,-873460313}, +{1737350743,-1262259248}, {1436946998,-1595891394}, {1073741769,-1859775424}, +{663608871,-2042378339}, {224473078,-2135719516}, {-224473265,-2135719496}, +{-663609049,-2042378281}, {-1073741932,-1859775330}, {-1436947137,-1595891268}, +{-1737350854,-1262259096}, {-1961823997,-873460141}, {-2100556013,-446486785}, +{2147483647,0}, {2042378310,-663608960}, {1737350743,-1262259248}, +{1262259172,-1737350799}, {663608871,-2042378339}, {-94,-2147483647}, +{-663609049,-2042378281}, {-1262259116,-1737350839}, {-1737350854,-1262259096}, +{-2042378447,-663608538}, {-2147483647,188}, {-2042378331,663608895}, +{-1737350633,1262259400}, {-1262258813,1737351059}, {-663609179,2042378239}, +}; +static const ne10_fft_state_int32_t ne10_fft_state_int32_t_480 = { +120, +(ne10_int32_t *)ne10_factors_480, +(ne10_fft_cpx_int32_t *)ne10_twiddles_480, +NULL, +(ne10_fft_cpx_int32_t *)&ne10_twiddles_480[120], +}; +static const arch_fft_state cfg_arch_480 = { +1, +(void *)&ne10_fft_state_int32_t_480, +}; + +static const ne10_fft_state_int32_t ne10_fft_state_int32_t_240 = { +60, +(ne10_int32_t *)ne10_factors_240, +(ne10_fft_cpx_int32_t *)ne10_twiddles_240, +NULL, +(ne10_fft_cpx_int32_t *)&ne10_twiddles_240[60], +}; +static const arch_fft_state cfg_arch_240 = { +1, +(void *)&ne10_fft_state_int32_t_240, +}; + +static const ne10_fft_state_int32_t ne10_fft_state_int32_t_120 = { +30, +(ne10_int32_t *)ne10_factors_120, +(ne10_fft_cpx_int32_t *)ne10_twiddles_120, +NULL, +(ne10_fft_cpx_int32_t *)&ne10_twiddles_120[30], +}; +static const arch_fft_state cfg_arch_120 = { +1, +(void *)&ne10_fft_state_int32_t_120, +}; + +static const ne10_fft_state_int32_t ne10_fft_state_int32_t_60 = { +15, +(ne10_int32_t *)ne10_factors_60, +(ne10_fft_cpx_int32_t *)ne10_twiddles_60, +NULL, +(ne10_fft_cpx_int32_t *)&ne10_twiddles_60[15], +}; +static const arch_fft_state cfg_arch_60 = { +1, +(void *)&ne10_fft_state_int32_t_60, +}; + +#endif /* end NE10_FFT_PARAMS48000_960 */ diff --git a/media/libopus/celt/static_modes_float.h b/media/libopus/celt/static_modes_float.h new file mode 100644 index 0000000000..e102a38391 --- /dev/null +++ b/media/libopus/celt/static_modes_float.h @@ -0,0 +1,888 @@ +/* The contents of this file was automatically generated by dump_modes.c + with arguments: 48000 960 + It contains static definitions for some pre-defined modes. */ +#include "modes.h" +#include "rate.h" + +#ifdef HAVE_ARM_NE10 +#define OVERRIDE_FFT 1 +#include "static_modes_float_arm_ne10.h" +#endif + +#ifndef DEF_WINDOW120 +#define DEF_WINDOW120 +static const opus_val16 window120[120] = { +6.7286966e-05f, 0.00060551348f, 0.0016815970f, 0.0032947962f, 0.0054439943f, +0.0081276923f, 0.011344001f, 0.015090633f, 0.019364886f, 0.024163635f, +0.029483315f, 0.035319905f, 0.041668911f, 0.048525347f, 0.055883718f, +0.063737999f, 0.072081616f, 0.080907428f, 0.090207705f, 0.099974111f, +0.11019769f, 0.12086883f, 0.13197729f, 0.14351214f, 0.15546177f, +0.16781389f, 0.18055550f, 0.19367290f, 0.20715171f, 0.22097682f, +0.23513243f, 0.24960208f, 0.26436860f, 0.27941419f, 0.29472040f, +0.31026818f, 0.32603788f, 0.34200931f, 0.35816177f, 0.37447407f, +0.39092462f, 0.40749142f, 0.42415215f, 0.44088423f, 0.45766484f, +0.47447104f, 0.49127978f, 0.50806798f, 0.52481261f, 0.54149077f, +0.55807973f, 0.57455701f, 0.59090049f, 0.60708841f, 0.62309951f, +0.63891306f, 0.65450896f, 0.66986776f, 0.68497077f, 0.69980010f, +0.71433873f, 0.72857055f, 0.74248043f, 0.75605424f, 0.76927895f, +0.78214257f, 0.79463430f, 0.80674445f, 0.81846456f, 0.82978733f, +0.84070669f, 0.85121779f, 0.86131698f, 0.87100183f, 0.88027111f, +0.88912479f, 0.89756398f, 0.90559094f, 0.91320904f, 0.92042270f, +0.92723738f, 0.93365955f, 0.93969656f, 0.94535671f, 0.95064907f, +0.95558353f, 0.96017067f, 0.96442171f, 0.96834849f, 0.97196334f, +0.97527906f, 0.97830883f, 0.98106616f, 0.98356480f, 0.98581869f, +0.98784191f, 0.98964856f, 0.99125274f, 0.99266849f, 0.99390969f, +0.99499004f, 0.99592297f, 0.99672162f, 0.99739874f, 0.99796667f, +0.99843728f, 0.99882195f, 0.99913147f, 0.99937606f, 0.99956527f, +0.99970802f, 0.99981248f, 0.99988613f, 0.99993565f, 0.99996697f, +0.99998518f, 0.99999457f, 0.99999859f, 0.99999982f, 1.0000000f, +}; +#endif + +#ifndef DEF_LOGN400 +#define DEF_LOGN400 +static const opus_int16 logN400[21] = { +0, 0, 0, 0, 0, 0, 0, 0, 8, 8, 8, 8, 16, 16, 16, 21, 21, 24, 29, 34, 36, }; +#endif + +#ifndef DEF_PULSE_CACHE50 +#define DEF_PULSE_CACHE50 +static const opus_int16 cache_index50[105] = { +-1, -1, -1, -1, -1, -1, -1, -1, 0, 0, 0, 0, 41, 41, 41, +82, 82, 123, 164, 200, 222, 0, 0, 0, 0, 0, 0, 0, 0, 41, +41, 41, 41, 123, 123, 123, 164, 164, 240, 266, 283, 295, 41, 41, 41, +41, 41, 41, 41, 41, 123, 123, 123, 123, 240, 240, 240, 266, 266, 305, +318, 328, 336, 123, 123, 123, 123, 123, 123, 123, 123, 240, 240, 240, 240, +305, 305, 305, 318, 318, 343, 351, 358, 364, 240, 240, 240, 240, 240, 240, +240, 240, 305, 305, 305, 305, 343, 343, 343, 351, 351, 370, 376, 382, 387, +}; +static const unsigned char cache_bits50[392] = { +40, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, +7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, +7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 40, 15, 23, 28, +31, 34, 36, 38, 39, 41, 42, 43, 44, 45, 46, 47, 47, 49, 50, +51, 52, 53, 54, 55, 55, 57, 58, 59, 60, 61, 62, 63, 63, 65, +66, 67, 68, 69, 70, 71, 71, 40, 20, 33, 41, 48, 53, 57, 61, +64, 66, 69, 71, 73, 75, 76, 78, 80, 82, 85, 87, 89, 91, 92, +94, 96, 98, 101, 103, 105, 107, 108, 110, 112, 114, 117, 119, 121, 123, +124, 126, 128, 40, 23, 39, 51, 60, 67, 73, 79, 83, 87, 91, 94, +97, 100, 102, 105, 107, 111, 115, 118, 121, 124, 126, 129, 131, 135, 139, +142, 145, 148, 150, 153, 155, 159, 163, 166, 169, 172, 174, 177, 179, 35, +28, 49, 65, 78, 89, 99, 107, 114, 120, 126, 132, 136, 141, 145, 149, +153, 159, 165, 171, 176, 180, 185, 189, 192, 199, 205, 211, 216, 220, 225, +229, 232, 239, 245, 251, 21, 33, 58, 79, 97, 112, 125, 137, 148, 157, +166, 174, 182, 189, 195, 201, 207, 217, 227, 235, 243, 251, 17, 35, 63, +86, 106, 123, 139, 152, 165, 177, 187, 197, 206, 214, 222, 230, 237, 250, +25, 31, 55, 75, 91, 105, 117, 128, 138, 146, 154, 161, 168, 174, 180, +185, 190, 200, 208, 215, 222, 229, 235, 240, 245, 255, 16, 36, 65, 89, +110, 128, 144, 159, 173, 185, 196, 207, 217, 226, 234, 242, 250, 11, 41, +74, 103, 128, 151, 172, 191, 209, 225, 241, 255, 9, 43, 79, 110, 138, +163, 186, 207, 227, 246, 12, 39, 71, 99, 123, 144, 164, 182, 198, 214, +228, 241, 253, 9, 44, 81, 113, 142, 168, 192, 214, 235, 255, 7, 49, +90, 127, 160, 191, 220, 247, 6, 51, 95, 134, 170, 203, 234, 7, 47, +87, 123, 155, 184, 212, 237, 6, 52, 97, 137, 174, 208, 240, 5, 57, +106, 151, 192, 231, 5, 59, 111, 158, 202, 243, 5, 55, 103, 147, 187, +224, 5, 60, 113, 161, 206, 248, 4, 65, 122, 175, 224, 4, 67, 127, +182, 234, }; +static const unsigned char cache_caps50[168] = { +224, 224, 224, 224, 224, 224, 224, 224, 160, 160, 160, 160, 185, 185, 185, +178, 178, 168, 134, 61, 37, 224, 224, 224, 224, 224, 224, 224, 224, 240, +240, 240, 240, 207, 207, 207, 198, 198, 183, 144, 66, 40, 160, 160, 160, +160, 160, 160, 160, 160, 185, 185, 185, 185, 193, 193, 193, 183, 183, 172, +138, 64, 38, 240, 240, 240, 240, 240, 240, 240, 240, 207, 207, 207, 207, +204, 204, 204, 193, 193, 180, 143, 66, 40, 185, 185, 185, 185, 185, 185, +185, 185, 193, 193, 193, 193, 193, 193, 193, 183, 183, 172, 138, 65, 39, +207, 207, 207, 207, 207, 207, 207, 207, 204, 204, 204, 204, 201, 201, 201, +188, 188, 176, 141, 66, 40, 193, 193, 193, 193, 193, 193, 193, 193, 193, +193, 193, 193, 194, 194, 194, 184, 184, 173, 139, 65, 39, 204, 204, 204, +204, 204, 204, 204, 204, 201, 201, 201, 201, 198, 198, 198, 187, 187, 175, +140, 66, 40, }; +#endif + +#ifndef FFT_TWIDDLES48000_960 +#define FFT_TWIDDLES48000_960 +static const kiss_twiddle_cpx fft_twiddles48000_960[480] = { +{1.0000000f, -0.0000000f}, {0.99991433f, -0.013089596f}, +{0.99965732f, -0.026176948f}, {0.99922904f, -0.039259816f}, +{0.99862953f, -0.052335956f}, {0.99785892f, -0.065403129f}, +{0.99691733f, -0.078459096f}, {0.99580493f, -0.091501619f}, +{0.99452190f, -0.10452846f}, {0.99306846f, -0.11753740f}, +{0.99144486f, -0.13052619f}, {0.98965139f, -0.14349262f}, +{0.98768834f, -0.15643447f}, {0.98555606f, -0.16934950f}, +{0.98325491f, -0.18223553f}, {0.98078528f, -0.19509032f}, +{0.97814760f, -0.20791169f}, {0.97534232f, -0.22069744f}, +{0.97236992f, -0.23344536f}, {0.96923091f, -0.24615329f}, +{0.96592583f, -0.25881905f}, {0.96245524f, -0.27144045f}, +{0.95881973f, -0.28401534f}, {0.95501994f, -0.29654157f}, +{0.95105652f, -0.30901699f}, {0.94693013f, -0.32143947f}, +{0.94264149f, -0.33380686f}, {0.93819134f, -0.34611706f}, +{0.93358043f, -0.35836795f}, {0.92880955f, -0.37055744f}, +{0.92387953f, -0.38268343f}, {0.91879121f, -0.39474386f}, +{0.91354546f, -0.40673664f}, {0.90814317f, -0.41865974f}, +{0.90258528f, -0.43051110f}, {0.89687274f, -0.44228869f}, +{0.89100652f, -0.45399050f}, {0.88498764f, -0.46561452f}, +{0.87881711f, -0.47715876f}, {0.87249601f, -0.48862124f}, +{0.86602540f, -0.50000000f}, {0.85940641f, -0.51129309f}, +{0.85264016f, -0.52249856f}, {0.84572782f, -0.53361452f}, +{0.83867057f, -0.54463904f}, {0.83146961f, -0.55557023f}, +{0.82412619f, -0.56640624f}, {0.81664156f, -0.57714519f}, +{0.80901699f, -0.58778525f}, {0.80125381f, -0.59832460f}, +{0.79335334f, -0.60876143f}, {0.78531693f, -0.61909395f}, +{0.77714596f, -0.62932039f}, {0.76884183f, -0.63943900f}, +{0.76040597f, -0.64944805f}, {0.75183981f, -0.65934582f}, +{0.74314483f, -0.66913061f}, {0.73432251f, -0.67880075f}, +{0.72537437f, -0.68835458f}, {0.71630194f, -0.69779046f}, +{0.70710678f, -0.70710678f}, {0.69779046f, -0.71630194f}, +{0.68835458f, -0.72537437f}, {0.67880075f, -0.73432251f}, +{0.66913061f, -0.74314483f}, {0.65934582f, -0.75183981f}, +{0.64944805f, -0.76040597f}, {0.63943900f, -0.76884183f}, +{0.62932039f, -0.77714596f}, {0.61909395f, -0.78531693f}, +{0.60876143f, -0.79335334f}, {0.59832460f, -0.80125381f}, +{0.58778525f, -0.80901699f}, {0.57714519f, -0.81664156f}, +{0.56640624f, -0.82412619f}, {0.55557023f, -0.83146961f}, +{0.54463904f, -0.83867057f}, {0.53361452f, -0.84572782f}, +{0.52249856f, -0.85264016f}, {0.51129309f, -0.85940641f}, +{0.50000000f, -0.86602540f}, {0.48862124f, -0.87249601f}, +{0.47715876f, -0.87881711f}, {0.46561452f, -0.88498764f}, +{0.45399050f, -0.89100652f}, {0.44228869f, -0.89687274f}, +{0.43051110f, -0.90258528f}, {0.41865974f, -0.90814317f}, +{0.40673664f, -0.91354546f}, {0.39474386f, -0.91879121f}, +{0.38268343f, -0.92387953f}, {0.37055744f, -0.92880955f}, +{0.35836795f, -0.93358043f}, {0.34611706f, -0.93819134f}, +{0.33380686f, -0.94264149f}, {0.32143947f, -0.94693013f}, +{0.30901699f, -0.95105652f}, {0.29654157f, -0.95501994f}, +{0.28401534f, -0.95881973f}, {0.27144045f, -0.96245524f}, +{0.25881905f, -0.96592583f}, {0.24615329f, -0.96923091f}, +{0.23344536f, -0.97236992f}, {0.22069744f, -0.97534232f}, +{0.20791169f, -0.97814760f}, {0.19509032f, -0.98078528f}, +{0.18223553f, -0.98325491f}, {0.16934950f, -0.98555606f}, +{0.15643447f, -0.98768834f}, {0.14349262f, -0.98965139f}, +{0.13052619f, -0.99144486f}, {0.11753740f, -0.99306846f}, +{0.10452846f, -0.99452190f}, {0.091501619f, -0.99580493f}, +{0.078459096f, -0.99691733f}, {0.065403129f, -0.99785892f}, +{0.052335956f, -0.99862953f}, {0.039259816f, -0.99922904f}, +{0.026176948f, -0.99965732f}, {0.013089596f, -0.99991433f}, +{6.1230318e-17f, -1.0000000f}, {-0.013089596f, -0.99991433f}, +{-0.026176948f, -0.99965732f}, {-0.039259816f, -0.99922904f}, +{-0.052335956f, -0.99862953f}, {-0.065403129f, -0.99785892f}, +{-0.078459096f, -0.99691733f}, {-0.091501619f, -0.99580493f}, +{-0.10452846f, -0.99452190f}, {-0.11753740f, -0.99306846f}, +{-0.13052619f, -0.99144486f}, {-0.14349262f, -0.98965139f}, +{-0.15643447f, -0.98768834f}, {-0.16934950f, -0.98555606f}, +{-0.18223553f, -0.98325491f}, {-0.19509032f, -0.98078528f}, +{-0.20791169f, -0.97814760f}, {-0.22069744f, -0.97534232f}, +{-0.23344536f, -0.97236992f}, {-0.24615329f, -0.96923091f}, +{-0.25881905f, -0.96592583f}, {-0.27144045f, -0.96245524f}, +{-0.28401534f, -0.95881973f}, {-0.29654157f, -0.95501994f}, +{-0.30901699f, -0.95105652f}, {-0.32143947f, -0.94693013f}, +{-0.33380686f, -0.94264149f}, {-0.34611706f, -0.93819134f}, +{-0.35836795f, -0.93358043f}, {-0.37055744f, -0.92880955f}, +{-0.38268343f, -0.92387953f}, {-0.39474386f, -0.91879121f}, +{-0.40673664f, -0.91354546f}, {-0.41865974f, -0.90814317f}, +{-0.43051110f, -0.90258528f}, {-0.44228869f, -0.89687274f}, +{-0.45399050f, -0.89100652f}, {-0.46561452f, -0.88498764f}, +{-0.47715876f, -0.87881711f}, {-0.48862124f, -0.87249601f}, +{-0.50000000f, -0.86602540f}, {-0.51129309f, -0.85940641f}, +{-0.52249856f, -0.85264016f}, {-0.53361452f, -0.84572782f}, +{-0.54463904f, -0.83867057f}, {-0.55557023f, -0.83146961f}, +{-0.56640624f, -0.82412619f}, {-0.57714519f, -0.81664156f}, +{-0.58778525f, -0.80901699f}, {-0.59832460f, -0.80125381f}, +{-0.60876143f, -0.79335334f}, {-0.61909395f, -0.78531693f}, +{-0.62932039f, -0.77714596f}, {-0.63943900f, -0.76884183f}, +{-0.64944805f, -0.76040597f}, {-0.65934582f, -0.75183981f}, +{-0.66913061f, -0.74314483f}, {-0.67880075f, -0.73432251f}, +{-0.68835458f, -0.72537437f}, {-0.69779046f, -0.71630194f}, +{-0.70710678f, -0.70710678f}, {-0.71630194f, -0.69779046f}, +{-0.72537437f, -0.68835458f}, {-0.73432251f, -0.67880075f}, +{-0.74314483f, -0.66913061f}, {-0.75183981f, -0.65934582f}, +{-0.76040597f, -0.64944805f}, {-0.76884183f, -0.63943900f}, +{-0.77714596f, -0.62932039f}, {-0.78531693f, -0.61909395f}, +{-0.79335334f, -0.60876143f}, {-0.80125381f, -0.59832460f}, +{-0.80901699f, -0.58778525f}, {-0.81664156f, -0.57714519f}, +{-0.82412619f, -0.56640624f}, {-0.83146961f, -0.55557023f}, +{-0.83867057f, -0.54463904f}, {-0.84572782f, -0.53361452f}, +{-0.85264016f, -0.52249856f}, {-0.85940641f, -0.51129309f}, +{-0.86602540f, -0.50000000f}, {-0.87249601f, -0.48862124f}, +{-0.87881711f, -0.47715876f}, {-0.88498764f, -0.46561452f}, +{-0.89100652f, -0.45399050f}, {-0.89687274f, -0.44228869f}, +{-0.90258528f, -0.43051110f}, {-0.90814317f, -0.41865974f}, +{-0.91354546f, -0.40673664f}, {-0.91879121f, -0.39474386f}, +{-0.92387953f, -0.38268343f}, {-0.92880955f, -0.37055744f}, +{-0.93358043f, -0.35836795f}, {-0.93819134f, -0.34611706f}, +{-0.94264149f, -0.33380686f}, {-0.94693013f, -0.32143947f}, +{-0.95105652f, -0.30901699f}, {-0.95501994f, -0.29654157f}, +{-0.95881973f, -0.28401534f}, {-0.96245524f, -0.27144045f}, +{-0.96592583f, -0.25881905f}, {-0.96923091f, -0.24615329f}, +{-0.97236992f, -0.23344536f}, {-0.97534232f, -0.22069744f}, +{-0.97814760f, -0.20791169f}, {-0.98078528f, -0.19509032f}, +{-0.98325491f, -0.18223553f}, {-0.98555606f, -0.16934950f}, +{-0.98768834f, -0.15643447f}, {-0.98965139f, -0.14349262f}, +{-0.99144486f, -0.13052619f}, {-0.99306846f, -0.11753740f}, +{-0.99452190f, -0.10452846f}, {-0.99580493f, -0.091501619f}, +{-0.99691733f, -0.078459096f}, {-0.99785892f, -0.065403129f}, +{-0.99862953f, -0.052335956f}, {-0.99922904f, -0.039259816f}, +{-0.99965732f, -0.026176948f}, {-0.99991433f, -0.013089596f}, +{-1.0000000f, -1.2246064e-16f}, {-0.99991433f, 0.013089596f}, +{-0.99965732f, 0.026176948f}, {-0.99922904f, 0.039259816f}, +{-0.99862953f, 0.052335956f}, {-0.99785892f, 0.065403129f}, +{-0.99691733f, 0.078459096f}, {-0.99580493f, 0.091501619f}, +{-0.99452190f, 0.10452846f}, {-0.99306846f, 0.11753740f}, +{-0.99144486f, 0.13052619f}, {-0.98965139f, 0.14349262f}, +{-0.98768834f, 0.15643447f}, {-0.98555606f, 0.16934950f}, +{-0.98325491f, 0.18223553f}, {-0.98078528f, 0.19509032f}, +{-0.97814760f, 0.20791169f}, {-0.97534232f, 0.22069744f}, +{-0.97236992f, 0.23344536f}, {-0.96923091f, 0.24615329f}, +{-0.96592583f, 0.25881905f}, {-0.96245524f, 0.27144045f}, +{-0.95881973f, 0.28401534f}, {-0.95501994f, 0.29654157f}, +{-0.95105652f, 0.30901699f}, {-0.94693013f, 0.32143947f}, +{-0.94264149f, 0.33380686f}, {-0.93819134f, 0.34611706f}, +{-0.93358043f, 0.35836795f}, {-0.92880955f, 0.37055744f}, +{-0.92387953f, 0.38268343f}, {-0.91879121f, 0.39474386f}, +{-0.91354546f, 0.40673664f}, {-0.90814317f, 0.41865974f}, +{-0.90258528f, 0.43051110f}, {-0.89687274f, 0.44228869f}, +{-0.89100652f, 0.45399050f}, {-0.88498764f, 0.46561452f}, +{-0.87881711f, 0.47715876f}, {-0.87249601f, 0.48862124f}, +{-0.86602540f, 0.50000000f}, {-0.85940641f, 0.51129309f}, +{-0.85264016f, 0.52249856f}, {-0.84572782f, 0.53361452f}, +{-0.83867057f, 0.54463904f}, {-0.83146961f, 0.55557023f}, +{-0.82412619f, 0.56640624f}, {-0.81664156f, 0.57714519f}, +{-0.80901699f, 0.58778525f}, {-0.80125381f, 0.59832460f}, +{-0.79335334f, 0.60876143f}, {-0.78531693f, 0.61909395f}, +{-0.77714596f, 0.62932039f}, {-0.76884183f, 0.63943900f}, +{-0.76040597f, 0.64944805f}, {-0.75183981f, 0.65934582f}, +{-0.74314483f, 0.66913061f}, {-0.73432251f, 0.67880075f}, +{-0.72537437f, 0.68835458f}, {-0.71630194f, 0.69779046f}, +{-0.70710678f, 0.70710678f}, {-0.69779046f, 0.71630194f}, +{-0.68835458f, 0.72537437f}, {-0.67880075f, 0.73432251f}, +{-0.66913061f, 0.74314483f}, {-0.65934582f, 0.75183981f}, +{-0.64944805f, 0.76040597f}, {-0.63943900f, 0.76884183f}, +{-0.62932039f, 0.77714596f}, {-0.61909395f, 0.78531693f}, +{-0.60876143f, 0.79335334f}, {-0.59832460f, 0.80125381f}, +{-0.58778525f, 0.80901699f}, {-0.57714519f, 0.81664156f}, +{-0.56640624f, 0.82412619f}, {-0.55557023f, 0.83146961f}, +{-0.54463904f, 0.83867057f}, {-0.53361452f, 0.84572782f}, +{-0.52249856f, 0.85264016f}, {-0.51129309f, 0.85940641f}, +{-0.50000000f, 0.86602540f}, {-0.48862124f, 0.87249601f}, +{-0.47715876f, 0.87881711f}, {-0.46561452f, 0.88498764f}, +{-0.45399050f, 0.89100652f}, {-0.44228869f, 0.89687274f}, +{-0.43051110f, 0.90258528f}, {-0.41865974f, 0.90814317f}, +{-0.40673664f, 0.91354546f}, {-0.39474386f, 0.91879121f}, +{-0.38268343f, 0.92387953f}, {-0.37055744f, 0.92880955f}, +{-0.35836795f, 0.93358043f}, {-0.34611706f, 0.93819134f}, +{-0.33380686f, 0.94264149f}, {-0.32143947f, 0.94693013f}, +{-0.30901699f, 0.95105652f}, {-0.29654157f, 0.95501994f}, +{-0.28401534f, 0.95881973f}, {-0.27144045f, 0.96245524f}, +{-0.25881905f, 0.96592583f}, {-0.24615329f, 0.96923091f}, +{-0.23344536f, 0.97236992f}, {-0.22069744f, 0.97534232f}, +{-0.20791169f, 0.97814760f}, {-0.19509032f, 0.98078528f}, +{-0.18223553f, 0.98325491f}, {-0.16934950f, 0.98555606f}, +{-0.15643447f, 0.98768834f}, {-0.14349262f, 0.98965139f}, +{-0.13052619f, 0.99144486f}, {-0.11753740f, 0.99306846f}, +{-0.10452846f, 0.99452190f}, {-0.091501619f, 0.99580493f}, +{-0.078459096f, 0.99691733f}, {-0.065403129f, 0.99785892f}, +{-0.052335956f, 0.99862953f}, {-0.039259816f, 0.99922904f}, +{-0.026176948f, 0.99965732f}, {-0.013089596f, 0.99991433f}, +{-1.8369095e-16f, 1.0000000f}, {0.013089596f, 0.99991433f}, +{0.026176948f, 0.99965732f}, {0.039259816f, 0.99922904f}, +{0.052335956f, 0.99862953f}, {0.065403129f, 0.99785892f}, +{0.078459096f, 0.99691733f}, {0.091501619f, 0.99580493f}, +{0.10452846f, 0.99452190f}, {0.11753740f, 0.99306846f}, +{0.13052619f, 0.99144486f}, {0.14349262f, 0.98965139f}, +{0.15643447f, 0.98768834f}, {0.16934950f, 0.98555606f}, +{0.18223553f, 0.98325491f}, {0.19509032f, 0.98078528f}, +{0.20791169f, 0.97814760f}, {0.22069744f, 0.97534232f}, +{0.23344536f, 0.97236992f}, {0.24615329f, 0.96923091f}, +{0.25881905f, 0.96592583f}, {0.27144045f, 0.96245524f}, +{0.28401534f, 0.95881973f}, {0.29654157f, 0.95501994f}, +{0.30901699f, 0.95105652f}, {0.32143947f, 0.94693013f}, +{0.33380686f, 0.94264149f}, {0.34611706f, 0.93819134f}, +{0.35836795f, 0.93358043f}, {0.37055744f, 0.92880955f}, +{0.38268343f, 0.92387953f}, {0.39474386f, 0.91879121f}, +{0.40673664f, 0.91354546f}, {0.41865974f, 0.90814317f}, +{0.43051110f, 0.90258528f}, {0.44228869f, 0.89687274f}, +{0.45399050f, 0.89100652f}, {0.46561452f, 0.88498764f}, +{0.47715876f, 0.87881711f}, {0.48862124f, 0.87249601f}, +{0.50000000f, 0.86602540f}, {0.51129309f, 0.85940641f}, +{0.52249856f, 0.85264016f}, {0.53361452f, 0.84572782f}, +{0.54463904f, 0.83867057f}, {0.55557023f, 0.83146961f}, +{0.56640624f, 0.82412619f}, {0.57714519f, 0.81664156f}, +{0.58778525f, 0.80901699f}, {0.59832460f, 0.80125381f}, +{0.60876143f, 0.79335334f}, {0.61909395f, 0.78531693f}, +{0.62932039f, 0.77714596f}, {0.63943900f, 0.76884183f}, +{0.64944805f, 0.76040597f}, {0.65934582f, 0.75183981f}, +{0.66913061f, 0.74314483f}, {0.67880075f, 0.73432251f}, +{0.68835458f, 0.72537437f}, {0.69779046f, 0.71630194f}, +{0.70710678f, 0.70710678f}, {0.71630194f, 0.69779046f}, +{0.72537437f, 0.68835458f}, {0.73432251f, 0.67880075f}, +{0.74314483f, 0.66913061f}, {0.75183981f, 0.65934582f}, +{0.76040597f, 0.64944805f}, {0.76884183f, 0.63943900f}, +{0.77714596f, 0.62932039f}, {0.78531693f, 0.61909395f}, +{0.79335334f, 0.60876143f}, {0.80125381f, 0.59832460f}, +{0.80901699f, 0.58778525f}, {0.81664156f, 0.57714519f}, +{0.82412619f, 0.56640624f}, {0.83146961f, 0.55557023f}, +{0.83867057f, 0.54463904f}, {0.84572782f, 0.53361452f}, +{0.85264016f, 0.52249856f}, {0.85940641f, 0.51129309f}, +{0.86602540f, 0.50000000f}, {0.87249601f, 0.48862124f}, +{0.87881711f, 0.47715876f}, {0.88498764f, 0.46561452f}, +{0.89100652f, 0.45399050f}, {0.89687274f, 0.44228869f}, +{0.90258528f, 0.43051110f}, {0.90814317f, 0.41865974f}, +{0.91354546f, 0.40673664f}, {0.91879121f, 0.39474386f}, +{0.92387953f, 0.38268343f}, {0.92880955f, 0.37055744f}, +{0.93358043f, 0.35836795f}, {0.93819134f, 0.34611706f}, +{0.94264149f, 0.33380686f}, {0.94693013f, 0.32143947f}, +{0.95105652f, 0.30901699f}, {0.95501994f, 0.29654157f}, +{0.95881973f, 0.28401534f}, {0.96245524f, 0.27144045f}, +{0.96592583f, 0.25881905f}, {0.96923091f, 0.24615329f}, +{0.97236992f, 0.23344536f}, {0.97534232f, 0.22069744f}, +{0.97814760f, 0.20791169f}, {0.98078528f, 0.19509032f}, +{0.98325491f, 0.18223553f}, {0.98555606f, 0.16934950f}, +{0.98768834f, 0.15643447f}, {0.98965139f, 0.14349262f}, +{0.99144486f, 0.13052619f}, {0.99306846f, 0.11753740f}, +{0.99452190f, 0.10452846f}, {0.99580493f, 0.091501619f}, +{0.99691733f, 0.078459096f}, {0.99785892f, 0.065403129f}, +{0.99862953f, 0.052335956f}, {0.99922904f, 0.039259816f}, +{0.99965732f, 0.026176948f}, {0.99991433f, 0.013089596f}, +}; +#ifndef FFT_BITREV480 +#define FFT_BITREV480 +static const opus_int16 fft_bitrev480[480] = { +0, 96, 192, 288, 384, 32, 128, 224, 320, 416, 64, 160, 256, 352, 448, +8, 104, 200, 296, 392, 40, 136, 232, 328, 424, 72, 168, 264, 360, 456, +16, 112, 208, 304, 400, 48, 144, 240, 336, 432, 80, 176, 272, 368, 464, +24, 120, 216, 312, 408, 56, 152, 248, 344, 440, 88, 184, 280, 376, 472, +4, 100, 196, 292, 388, 36, 132, 228, 324, 420, 68, 164, 260, 356, 452, +12, 108, 204, 300, 396, 44, 140, 236, 332, 428, 76, 172, 268, 364, 460, +20, 116, 212, 308, 404, 52, 148, 244, 340, 436, 84, 180, 276, 372, 468, +28, 124, 220, 316, 412, 60, 156, 252, 348, 444, 92, 188, 284, 380, 476, +1, 97, 193, 289, 385, 33, 129, 225, 321, 417, 65, 161, 257, 353, 449, +9, 105, 201, 297, 393, 41, 137, 233, 329, 425, 73, 169, 265, 361, 457, +17, 113, 209, 305, 401, 49, 145, 241, 337, 433, 81, 177, 273, 369, 465, +25, 121, 217, 313, 409, 57, 153, 249, 345, 441, 89, 185, 281, 377, 473, +5, 101, 197, 293, 389, 37, 133, 229, 325, 421, 69, 165, 261, 357, 453, +13, 109, 205, 301, 397, 45, 141, 237, 333, 429, 77, 173, 269, 365, 461, +21, 117, 213, 309, 405, 53, 149, 245, 341, 437, 85, 181, 277, 373, 469, +29, 125, 221, 317, 413, 61, 157, 253, 349, 445, 93, 189, 285, 381, 477, +2, 98, 194, 290, 386, 34, 130, 226, 322, 418, 66, 162, 258, 354, 450, +10, 106, 202, 298, 394, 42, 138, 234, 330, 426, 74, 170, 266, 362, 458, +18, 114, 210, 306, 402, 50, 146, 242, 338, 434, 82, 178, 274, 370, 466, +26, 122, 218, 314, 410, 58, 154, 250, 346, 442, 90, 186, 282, 378, 474, +6, 102, 198, 294, 390, 38, 134, 230, 326, 422, 70, 166, 262, 358, 454, +14, 110, 206, 302, 398, 46, 142, 238, 334, 430, 78, 174, 270, 366, 462, +22, 118, 214, 310, 406, 54, 150, 246, 342, 438, 86, 182, 278, 374, 470, +30, 126, 222, 318, 414, 62, 158, 254, 350, 446, 94, 190, 286, 382, 478, +3, 99, 195, 291, 387, 35, 131, 227, 323, 419, 67, 163, 259, 355, 451, +11, 107, 203, 299, 395, 43, 139, 235, 331, 427, 75, 171, 267, 363, 459, +19, 115, 211, 307, 403, 51, 147, 243, 339, 435, 83, 179, 275, 371, 467, +27, 123, 219, 315, 411, 59, 155, 251, 347, 443, 91, 187, 283, 379, 475, +7, 103, 199, 295, 391, 39, 135, 231, 327, 423, 71, 167, 263, 359, 455, +15, 111, 207, 303, 399, 47, 143, 239, 335, 431, 79, 175, 271, 367, 463, +23, 119, 215, 311, 407, 55, 151, 247, 343, 439, 87, 183, 279, 375, 471, +31, 127, 223, 319, 415, 63, 159, 255, 351, 447, 95, 191, 287, 383, 479, +}; +#endif + +#ifndef FFT_BITREV240 +#define FFT_BITREV240 +static const opus_int16 fft_bitrev240[240] = { +0, 48, 96, 144, 192, 16, 64, 112, 160, 208, 32, 80, 128, 176, 224, +4, 52, 100, 148, 196, 20, 68, 116, 164, 212, 36, 84, 132, 180, 228, +8, 56, 104, 152, 200, 24, 72, 120, 168, 216, 40, 88, 136, 184, 232, +12, 60, 108, 156, 204, 28, 76, 124, 172, 220, 44, 92, 140, 188, 236, +1, 49, 97, 145, 193, 17, 65, 113, 161, 209, 33, 81, 129, 177, 225, +5, 53, 101, 149, 197, 21, 69, 117, 165, 213, 37, 85, 133, 181, 229, +9, 57, 105, 153, 201, 25, 73, 121, 169, 217, 41, 89, 137, 185, 233, +13, 61, 109, 157, 205, 29, 77, 125, 173, 221, 45, 93, 141, 189, 237, +2, 50, 98, 146, 194, 18, 66, 114, 162, 210, 34, 82, 130, 178, 226, +6, 54, 102, 150, 198, 22, 70, 118, 166, 214, 38, 86, 134, 182, 230, +10, 58, 106, 154, 202, 26, 74, 122, 170, 218, 42, 90, 138, 186, 234, +14, 62, 110, 158, 206, 30, 78, 126, 174, 222, 46, 94, 142, 190, 238, +3, 51, 99, 147, 195, 19, 67, 115, 163, 211, 35, 83, 131, 179, 227, +7, 55, 103, 151, 199, 23, 71, 119, 167, 215, 39, 87, 135, 183, 231, +11, 59, 107, 155, 203, 27, 75, 123, 171, 219, 43, 91, 139, 187, 235, +15, 63, 111, 159, 207, 31, 79, 127, 175, 223, 47, 95, 143, 191, 239, +}; +#endif + +#ifndef FFT_BITREV120 +#define FFT_BITREV120 +static const opus_int16 fft_bitrev120[120] = { +0, 24, 48, 72, 96, 8, 32, 56, 80, 104, 16, 40, 64, 88, 112, +4, 28, 52, 76, 100, 12, 36, 60, 84, 108, 20, 44, 68, 92, 116, +1, 25, 49, 73, 97, 9, 33, 57, 81, 105, 17, 41, 65, 89, 113, +5, 29, 53, 77, 101, 13, 37, 61, 85, 109, 21, 45, 69, 93, 117, +2, 26, 50, 74, 98, 10, 34, 58, 82, 106, 18, 42, 66, 90, 114, +6, 30, 54, 78, 102, 14, 38, 62, 86, 110, 22, 46, 70, 94, 118, +3, 27, 51, 75, 99, 11, 35, 59, 83, 107, 19, 43, 67, 91, 115, +7, 31, 55, 79, 103, 15, 39, 63, 87, 111, 23, 47, 71, 95, 119, +}; +#endif + +#ifndef FFT_BITREV60 +#define FFT_BITREV60 +static const opus_int16 fft_bitrev60[60] = { +0, 12, 24, 36, 48, 4, 16, 28, 40, 52, 8, 20, 32, 44, 56, +1, 13, 25, 37, 49, 5, 17, 29, 41, 53, 9, 21, 33, 45, 57, +2, 14, 26, 38, 50, 6, 18, 30, 42, 54, 10, 22, 34, 46, 58, +3, 15, 27, 39, 51, 7, 19, 31, 43, 55, 11, 23, 35, 47, 59, +}; +#endif + +#ifndef FFT_STATE48000_960_0 +#define FFT_STATE48000_960_0 +static const kiss_fft_state fft_state48000_960_0 = { +480, /* nfft */ +0.002083333f, /* scale */ +-1, /* shift */ +{5, 96, 3, 32, 4, 8, 2, 4, 4, 1, 0, 0, 0, 0, 0, 0, }, /* factors */ +fft_bitrev480, /* bitrev */ +fft_twiddles48000_960, /* bitrev */ +#ifdef OVERRIDE_FFT +(arch_fft_state *)&cfg_arch_480, +#else +NULL, +#endif +}; +#endif + +#ifndef FFT_STATE48000_960_1 +#define FFT_STATE48000_960_1 +static const kiss_fft_state fft_state48000_960_1 = { +240, /* nfft */ +0.004166667f, /* scale */ +1, /* shift */ +{5, 48, 3, 16, 4, 4, 4, 1, 0, 0, 0, 0, 0, 0, 0, 0, }, /* factors */ +fft_bitrev240, /* bitrev */ +fft_twiddles48000_960, /* bitrev */ +#ifdef OVERRIDE_FFT +(arch_fft_state *)&cfg_arch_240, +#else +NULL, +#endif +}; +#endif + +#ifndef FFT_STATE48000_960_2 +#define FFT_STATE48000_960_2 +static const kiss_fft_state fft_state48000_960_2 = { +120, /* nfft */ +0.008333333f, /* scale */ +2, /* shift */ +{5, 24, 3, 8, 2, 4, 4, 1, 0, 0, 0, 0, 0, 0, 0, 0, }, /* factors */ +fft_bitrev120, /* bitrev */ +fft_twiddles48000_960, /* bitrev */ +#ifdef OVERRIDE_FFT +(arch_fft_state *)&cfg_arch_120, +#else +NULL, +#endif +}; +#endif + +#ifndef FFT_STATE48000_960_3 +#define FFT_STATE48000_960_3 +static const kiss_fft_state fft_state48000_960_3 = { +60, /* nfft */ +0.016666667f, /* scale */ +3, /* shift */ +{5, 12, 3, 4, 4, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, }, /* factors */ +fft_bitrev60, /* bitrev */ +fft_twiddles48000_960, /* bitrev */ +#ifdef OVERRIDE_FFT +(arch_fft_state *)&cfg_arch_60, +#else +NULL, +#endif +}; +#endif + +#endif + +#ifndef MDCT_TWIDDLES960 +#define MDCT_TWIDDLES960 +static const opus_val16 mdct_twiddles960[1800] = { +0.99999994f, 0.99999321f, 0.99997580f, 0.99994773f, 0.99990886f, +0.99985933f, 0.99979913f, 0.99972820f, 0.99964654f, 0.99955416f, +0.99945110f, 0.99933738f, 0.99921292f, 0.99907774f, 0.99893188f, +0.99877530f, 0.99860805f, 0.99843007f, 0.99824142f, 0.99804211f, +0.99783206f, 0.99761140f, 0.99737996f, 0.99713790f, 0.99688518f, +0.99662173f, 0.99634761f, 0.99606287f, 0.99576741f, 0.99546129f, +0.99514455f, 0.99481714f, 0.99447906f, 0.99413031f, 0.99377096f, +0.99340093f, 0.99302030f, 0.99262899f, 0.99222708f, 0.99181455f, +0.99139136f, 0.99095762f, 0.99051321f, 0.99005818f, 0.98959261f, +0.98911643f, 0.98862964f, 0.98813224f, 0.98762429f, 0.98710573f, +0.98657662f, 0.98603696f, 0.98548669f, 0.98492593f, 0.98435456f, +0.98377270f, 0.98318028f, 0.98257732f, 0.98196387f, 0.98133987f, +0.98070538f, 0.98006040f, 0.97940493f, 0.97873890f, 0.97806245f, +0.97737551f, 0.97667813f, 0.97597027f, 0.97525197f, 0.97452319f, +0.97378403f, 0.97303438f, 0.97227436f, 0.97150391f, 0.97072303f, +0.96993178f, 0.96913016f, 0.96831810f, 0.96749574f, 0.96666300f, +0.96581990f, 0.96496642f, 0.96410263f, 0.96322852f, 0.96234411f, +0.96144938f, 0.96054435f, 0.95962906f, 0.95870346f, 0.95776761f, +0.95682150f, 0.95586514f, 0.95489854f, 0.95392174f, 0.95293468f, +0.95193744f, 0.95093000f, 0.94991243f, 0.94888461f, 0.94784665f, +0.94679856f, 0.94574034f, 0.94467193f, 0.94359344f, 0.94250488f, +0.94140619f, 0.94029742f, 0.93917859f, 0.93804967f, 0.93691075f, +0.93576175f, 0.93460274f, 0.93343377f, 0.93225473f, 0.93106574f, +0.92986679f, 0.92865789f, 0.92743903f, 0.92621022f, 0.92497152f, +0.92372292f, 0.92246443f, 0.92119598f, 0.91991776f, 0.91862965f, +0.91733170f, 0.91602397f, 0.91470635f, 0.91337901f, 0.91204184f, +0.91069490f, 0.90933824f, 0.90797186f, 0.90659571f, 0.90520984f, +0.90381432f, 0.90240908f, 0.90099424f, 0.89956969f, 0.89813554f, +0.89669174f, 0.89523834f, 0.89377540f, 0.89230281f, 0.89082074f, +0.88932908f, 0.88782793f, 0.88631725f, 0.88479710f, 0.88326746f, +0.88172835f, 0.88017982f, 0.87862182f, 0.87705445f, 0.87547767f, +0.87389153f, 0.87229604f, 0.87069118f, 0.86907703f, 0.86745358f, +0.86582077f, 0.86417878f, 0.86252749f, 0.86086690f, 0.85919720f, +0.85751826f, 0.85583007f, 0.85413277f, 0.85242635f, 0.85071075f, +0.84898609f, 0.84725231f, 0.84550947f, 0.84375757f, 0.84199661f, +0.84022665f, 0.83844769f, 0.83665979f, 0.83486289f, 0.83305705f, +0.83124226f, 0.82941860f, 0.82758605f, 0.82574469f, 0.82389444f, +0.82203537f, 0.82016748f, 0.81829083f, 0.81640542f, 0.81451124f, +0.81260836f, 0.81069672f, 0.80877650f, 0.80684757f, 0.80490994f, +0.80296379f, 0.80100900f, 0.79904562f, 0.79707366f, 0.79509324f, +0.79310423f, 0.79110676f, 0.78910083f, 0.78708643f, 0.78506362f, +0.78303236f, 0.78099275f, 0.77894479f, 0.77688843f, 0.77482378f, +0.77275085f, 0.77066964f, 0.76858020f, 0.76648247f, 0.76437658f, +0.76226246f, 0.76014024f, 0.75800985f, 0.75587130f, 0.75372469f, +0.75157005f, 0.74940729f, 0.74723655f, 0.74505776f, 0.74287105f, +0.74067634f, 0.73847371f, 0.73626316f, 0.73404479f, 0.73181850f, +0.72958434f, 0.72734243f, 0.72509271f, 0.72283524f, 0.72057003f, +0.71829706f, 0.71601641f, 0.71372813f, 0.71143216f, 0.70912862f, +0.70681745f, 0.70449871f, 0.70217246f, 0.69983864f, 0.69749737f, +0.69514859f, 0.69279242f, 0.69042879f, 0.68805778f, 0.68567938f, +0.68329364f, 0.68090063f, 0.67850029f, 0.67609268f, 0.67367786f, +0.67125577f, 0.66882652f, 0.66639012f, 0.66394657f, 0.66149592f, +0.65903819f, 0.65657341f, 0.65410155f, 0.65162271f, 0.64913690f, +0.64664418f, 0.64414448f, 0.64163786f, 0.63912445f, 0.63660413f, +0.63407701f, 0.63154310f, 0.62900239f, 0.62645501f, 0.62390089f, +0.62134010f, 0.61877263f, 0.61619854f, 0.61361790f, 0.61103064f, +0.60843682f, 0.60583651f, 0.60322970f, 0.60061646f, 0.59799677f, +0.59537065f, 0.59273821f, 0.59009939f, 0.58745426f, 0.58480281f, +0.58214509f, 0.57948118f, 0.57681108f, 0.57413477f, 0.57145232f, +0.56876373f, 0.56606907f, 0.56336832f, 0.56066155f, 0.55794877f, +0.55523002f, 0.55250537f, 0.54977477f, 0.54703826f, 0.54429591f, +0.54154772f, 0.53879374f, 0.53603399f, 0.53326851f, 0.53049731f, +0.52772039f, 0.52493787f, 0.52214974f, 0.51935595f, 0.51655668f, +0.51375180f, 0.51094145f, 0.50812566f, 0.50530440f, 0.50247771f, +0.49964568f, 0.49680826f, 0.49396557f, 0.49111754f, 0.48826426f, +0.48540577f, 0.48254207f, 0.47967321f, 0.47679919f, 0.47392011f, +0.47103590f, 0.46814668f, 0.46525243f, 0.46235323f, 0.45944905f, +0.45653993f, 0.45362595f, 0.45070711f, 0.44778344f, 0.44485497f, +0.44192174f, 0.43898380f, 0.43604112f, 0.43309379f, 0.43014181f, +0.42718524f, 0.42422408f, 0.42125839f, 0.41828820f, 0.41531351f, +0.41233435f, 0.40935081f, 0.40636289f, 0.40337059f, 0.40037400f, +0.39737311f, 0.39436796f, 0.39135858f, 0.38834500f, 0.38532731f, +0.38230544f, 0.37927949f, 0.37624949f, 0.37321547f, 0.37017745f, +0.36713544f, 0.36408952f, 0.36103970f, 0.35798600f, 0.35492846f, +0.35186714f, 0.34880206f, 0.34573323f, 0.34266070f, 0.33958447f, +0.33650464f, 0.33342120f, 0.33033419f, 0.32724363f, 0.32414958f, +0.32105204f, 0.31795108f, 0.31484672f, 0.31173897f, 0.30862790f, +0.30551350f, 0.30239585f, 0.29927495f, 0.29615086f, 0.29302359f, +0.28989318f, 0.28675964f, 0.28362307f, 0.28048345f, 0.27734083f, +0.27419522f, 0.27104670f, 0.26789525f, 0.26474094f, 0.26158381f, +0.25842386f, 0.25526115f, 0.25209570f, 0.24892756f, 0.24575676f, +0.24258332f, 0.23940729f, 0.23622867f, 0.23304754f, 0.22986393f, +0.22667783f, 0.22348931f, 0.22029841f, 0.21710514f, 0.21390954f, +0.21071166f, 0.20751151f, 0.20430915f, 0.20110460f, 0.19789790f, +0.19468907f, 0.19147816f, 0.18826519f, 0.18505022f, 0.18183327f, +0.17861435f, 0.17539354f, 0.17217083f, 0.16894630f, 0.16571994f, +0.16249183f, 0.15926196f, 0.15603039f, 0.15279715f, 0.14956227f, +0.14632578f, 0.14308774f, 0.13984816f, 0.13660708f, 0.13336454f, +0.13012058f, 0.12687522f, 0.12362850f, 0.12038045f, 0.11713112f, +0.11388054f, 0.11062872f, 0.10737573f, 0.10412160f, 0.10086634f, +0.097609997f, 0.094352618f, 0.091094226f, 0.087834857f, 0.084574550f, +0.081313334f, 0.078051247f, 0.074788325f, 0.071524605f, 0.068260118f, +0.064994894f, 0.061728980f, 0.058462404f, 0.055195201f, 0.051927410f, +0.048659060f, 0.045390189f, 0.042120833f, 0.038851023f, 0.035580799f, +0.032310195f, 0.029039243f, 0.025767982f, 0.022496443f, 0.019224664f, +0.015952680f, 0.012680525f, 0.0094082337f, 0.0061358409f, 0.0028633832f, +-0.00040910527f, -0.0036815894f, -0.0069540343f, -0.010226404f, -0.013498665f, +-0.016770782f, -0.020042717f, -0.023314439f, -0.026585912f, -0.029857099f, +-0.033127967f, -0.036398482f, -0.039668605f, -0.042938303f, -0.046207540f, +-0.049476285f, -0.052744497f, -0.056012146f, -0.059279196f, -0.062545612f, +-0.065811358f, -0.069076397f, -0.072340697f, -0.075604223f, -0.078866936f, +-0.082128808f, -0.085389800f, -0.088649876f, -0.091909006f, -0.095167145f, +-0.098424271f, -0.10168034f, -0.10493532f, -0.10818918f, -0.11144188f, +-0.11469338f, -0.11794366f, -0.12119267f, -0.12444039f, -0.12768677f, +-0.13093179f, -0.13417540f, -0.13741758f, -0.14065829f, -0.14389749f, +-0.14713514f, -0.15037122f, -0.15360570f, -0.15683852f, -0.16006967f, +-0.16329910f, -0.16652679f, -0.16975269f, -0.17297678f, -0.17619900f, +-0.17941935f, -0.18263777f, -0.18585424f, -0.18906870f, -0.19228116f, +-0.19549155f, -0.19869985f, -0.20190603f, -0.20511003f, -0.20831184f, +-0.21151142f, -0.21470875f, -0.21790376f, -0.22109644f, -0.22428675f, +-0.22747467f, -0.23066014f, -0.23384315f, -0.23702365f, -0.24020162f, +-0.24337701f, -0.24654980f, -0.24971995f, -0.25288740f, -0.25605217f, +-0.25921419f, -0.26237345f, -0.26552987f, -0.26868346f, -0.27183419f, +-0.27498198f, -0.27812684f, -0.28126872f, -0.28440759f, -0.28754342f, +-0.29067615f, -0.29380578f, -0.29693225f, -0.30005556f, -0.30317566f, +-0.30629250f, -0.30940607f, -0.31251630f, -0.31562322f, -0.31872672f, +-0.32182685f, -0.32492352f, -0.32801670f, -0.33110636f, -0.33419248f, +-0.33727503f, -0.34035397f, -0.34342924f, -0.34650084f, -0.34956875f, +-0.35263291f, -0.35569328f, -0.35874987f, -0.36180258f, -0.36485144f, +-0.36789638f, -0.37093741f, -0.37397444f, -0.37700745f, -0.38003644f, +-0.38306138f, -0.38608220f, -0.38909888f, -0.39211139f, -0.39511973f, +-0.39812380f, -0.40112361f, -0.40411916f, -0.40711036f, -0.41009718f, +-0.41307965f, -0.41605768f, -0.41903123f, -0.42200032f, -0.42496487f, +-0.42792490f, -0.43088034f, -0.43383113f, -0.43677729f, -0.43971881f, +-0.44265559f, -0.44558764f, -0.44851488f, -0.45143735f, -0.45435500f, +-0.45726776f, -0.46017563f, -0.46307856f, -0.46597654f, -0.46886954f, +-0.47175750f, -0.47464043f, -0.47751826f, -0.48039100f, -0.48325855f, +-0.48612097f, -0.48897815f, -0.49183011f, -0.49467680f, -0.49751821f, +-0.50035429f, -0.50318497f, -0.50601029f, -0.50883019f, -0.51164466f, +-0.51445359f, -0.51725709f, -0.52005500f, -0.52284735f, -0.52563411f, +-0.52841520f, -0.53119069f, -0.53396046f, -0.53672451f, -0.53948283f, +-0.54223537f, -0.54498214f, -0.54772300f, -0.55045801f, -0.55318713f, +-0.55591035f, -0.55862761f, -0.56133890f, -0.56404412f, -0.56674337f, +-0.56943649f, -0.57212353f, -0.57480448f, -0.57747924f, -0.58014780f, +-0.58281022f, -0.58546633f, -0.58811617f, -0.59075975f, -0.59339696f, +-0.59602785f, -0.59865236f, -0.60127044f, -0.60388207f, -0.60648727f, +-0.60908598f, -0.61167812f, -0.61426371f, -0.61684275f, -0.61941516f, +-0.62198097f, -0.62454009f, -0.62709254f, -0.62963831f, -0.63217729f, +-0.63470948f, -0.63723493f, -0.63975352f, -0.64226526f, -0.64477009f, +-0.64726806f, -0.64975911f, -0.65224314f, -0.65472025f, -0.65719032f, +-0.65965337f, -0.66210932f, -0.66455823f, -0.66700000f, -0.66943461f, +-0.67186207f, -0.67428231f, -0.67669535f, -0.67910111f, -0.68149966f, +-0.68389088f, -0.68627477f, -0.68865126f, -0.69102043f, -0.69338220f, +-0.69573659f, -0.69808346f, -0.70042288f, -0.70275480f, -0.70507920f, +-0.70739603f, -0.70970529f, -0.71200693f, -0.71430099f, -0.71658736f, +-0.71886611f, -0.72113711f, -0.72340041f, -0.72565591f, -0.72790372f, +-0.73014367f, -0.73237586f, -0.73460019f, -0.73681659f, -0.73902518f, +-0.74122584f, -0.74341851f, -0.74560326f, -0.74778003f, -0.74994880f, +-0.75210953f, -0.75426215f, -0.75640678f, -0.75854325f, -0.76067162f, +-0.76279181f, -0.76490390f, -0.76700771f, -0.76910341f, -0.77119076f, +-0.77326995f, -0.77534080f, -0.77740335f, -0.77945763f, -0.78150350f, +-0.78354102f, -0.78557014f, -0.78759086f, -0.78960317f, -0.79160696f, +-0.79360235f, -0.79558921f, -0.79756755f, -0.79953730f, -0.80149853f, +-0.80345118f, -0.80539525f, -0.80733067f, -0.80925739f, -0.81117553f, +-0.81308490f, -0.81498563f, -0.81687760f, -0.81876087f, -0.82063532f, +-0.82250100f, -0.82435787f, -0.82620591f, -0.82804507f, -0.82987541f, +-0.83169687f, -0.83350939f, -0.83531296f, -0.83710766f, -0.83889335f, +-0.84067005f, -0.84243774f, -0.84419644f, -0.84594607f, -0.84768665f, +-0.84941816f, -0.85114056f, -0.85285389f, -0.85455805f, -0.85625303f, +-0.85793889f, -0.85961550f, -0.86128294f, -0.86294121f, -0.86459017f, +-0.86622989f, -0.86786032f, -0.86948150f, -0.87109333f, -0.87269586f, +-0.87428904f, -0.87587279f, -0.87744725f, -0.87901229f, -0.88056785f, +-0.88211405f, -0.88365078f, -0.88517809f, -0.88669586f, -0.88820416f, +-0.88970292f, -0.89119220f, -0.89267188f, -0.89414203f, -0.89560264f, +-0.89705360f, -0.89849502f, -0.89992678f, -0.90134889f, -0.90276134f, +-0.90416414f, -0.90555727f, -0.90694070f, -0.90831441f, -0.90967834f, +-0.91103262f, -0.91237706f, -0.91371179f, -0.91503674f, -0.91635185f, +-0.91765714f, -0.91895264f, -0.92023826f, -0.92151409f, -0.92277998f, +-0.92403603f, -0.92528218f, -0.92651838f, -0.92774469f, -0.92896110f, +-0.93016750f, -0.93136400f, -0.93255049f, -0.93372697f, -0.93489349f, +-0.93604994f, -0.93719643f, -0.93833286f, -0.93945926f, -0.94057560f, +-0.94168180f, -0.94277799f, -0.94386405f, -0.94494003f, -0.94600588f, +-0.94706154f, -0.94810712f, -0.94914252f, -0.95016778f, -0.95118284f, +-0.95218778f, -0.95318246f, -0.95416695f, -0.95514119f, -0.95610523f, +-0.95705903f, -0.95800257f, -0.95893586f, -0.95985889f, -0.96077162f, +-0.96167403f, -0.96256620f, -0.96344805f, -0.96431959f, -0.96518075f, +-0.96603161f, -0.96687216f, -0.96770233f, -0.96852213f, -0.96933156f, +-0.97013056f, -0.97091925f, -0.97169751f, -0.97246534f, -0.97322279f, +-0.97396982f, -0.97470641f, -0.97543252f, -0.97614825f, -0.97685349f, +-0.97754824f, -0.97823256f, -0.97890645f, -0.97956979f, -0.98022264f, +-0.98086500f, -0.98149687f, -0.98211825f, -0.98272908f, -0.98332942f, +-0.98391914f, -0.98449844f, -0.98506713f, -0.98562527f, -0.98617285f, +-0.98670989f, -0.98723638f, -0.98775226f, -0.98825759f, -0.98875231f, +-0.98923647f, -0.98971003f, -0.99017298f, -0.99062532f, -0.99106705f, +-0.99149817f, -0.99191868f, -0.99232858f, -0.99272782f, -0.99311644f, +-0.99349445f, -0.99386179f, -0.99421853f, -0.99456459f, -0.99489999f, +-0.99522477f, -0.99553883f, -0.99584228f, -0.99613506f, -0.99641716f, +-0.99668860f, -0.99694937f, -0.99719942f, -0.99743885f, -0.99766755f, +-0.99788558f, -0.99809295f, -0.99828959f, -0.99847561f, -0.99865085f, +-0.99881548f, -0.99896932f, -0.99911255f, -0.99924499f, -0.99936682f, +-0.99947786f, -0.99957830f, -0.99966794f, -0.99974692f, -0.99981517f, +-0.99987274f, -0.99991959f, -0.99995571f, -0.99998116f, -0.99999589f, +0.99999964f, 0.99997288f, 0.99990326f, 0.99979085f, 0.99963558f, +0.99943751f, 0.99919659f, 0.99891287f, 0.99858636f, 0.99821711f, +0.99780506f, 0.99735034f, 0.99685282f, 0.99631262f, 0.99572974f, +0.99510419f, 0.99443603f, 0.99372530f, 0.99297196f, 0.99217612f, +0.99133772f, 0.99045694f, 0.98953366f, 0.98856801f, 0.98756003f, +0.98650974f, 0.98541719f, 0.98428243f, 0.98310548f, 0.98188645f, +0.98062533f, 0.97932225f, 0.97797716f, 0.97659022f, 0.97516143f, +0.97369087f, 0.97217858f, 0.97062469f, 0.96902919f, 0.96739221f, +0.96571374f, 0.96399397f, 0.96223283f, 0.96043050f, 0.95858705f, +0.95670253f, 0.95477700f, 0.95281059f, 0.95080340f, 0.94875544f, +0.94666684f, 0.94453770f, 0.94236809f, 0.94015813f, 0.93790787f, +0.93561745f, 0.93328691f, 0.93091643f, 0.92850608f, 0.92605597f, +0.92356616f, 0.92103678f, 0.91846794f, 0.91585976f, 0.91321236f, +0.91052586f, 0.90780038f, 0.90503591f, 0.90223277f, 0.89939094f, +0.89651060f, 0.89359182f, 0.89063478f, 0.88763964f, 0.88460642f, +0.88153529f, 0.87842643f, 0.87527996f, 0.87209594f, 0.86887461f, +0.86561602f, 0.86232042f, 0.85898781f, 0.85561842f, 0.85221243f, +0.84876984f, 0.84529096f, 0.84177583f, 0.83822471f, 0.83463764f, +0.83101481f, 0.82735640f, 0.82366252f, 0.81993335f, 0.81616908f, +0.81236988f, 0.80853581f, 0.80466717f, 0.80076402f, 0.79682660f, +0.79285502f, 0.78884947f, 0.78481019f, 0.78073722f, 0.77663082f, +0.77249116f, 0.76831841f, 0.76411277f, 0.75987434f, 0.75560343f, +0.75130010f, 0.74696463f, 0.74259710f, 0.73819780f, 0.73376691f, +0.72930455f, 0.72481096f, 0.72028631f, 0.71573079f, 0.71114463f, +0.70652801f, 0.70188117f, 0.69720417f, 0.69249737f, 0.68776089f, +0.68299496f, 0.67819971f, 0.67337549f, 0.66852236f, 0.66364062f, +0.65873051f, 0.65379208f, 0.64882571f, 0.64383155f, 0.63880974f, +0.63376063f, 0.62868434f, 0.62358117f, 0.61845124f, 0.61329484f, +0.60811216f, 0.60290343f, 0.59766883f, 0.59240872f, 0.58712316f, +0.58181250f, 0.57647687f, 0.57111657f, 0.56573176f, 0.56032276f, +0.55488980f, 0.54943299f, 0.54395270f, 0.53844911f, 0.53292239f, +0.52737290f, 0.52180082f, 0.51620632f, 0.51058978f, 0.50495136f, +0.49929130f, 0.49360985f, 0.48790723f, 0.48218375f, 0.47643960f, +0.47067502f, 0.46489030f, 0.45908567f, 0.45326138f, 0.44741765f, +0.44155475f, 0.43567297f, 0.42977250f, 0.42385364f, 0.41791660f, +0.41196167f, 0.40598908f, 0.39999911f, 0.39399201f, 0.38796803f, +0.38192743f, 0.37587047f, 0.36979741f, 0.36370850f, 0.35760403f, +0.35148421f, 0.34534934f, 0.33919969f, 0.33303553f, 0.32685706f, +0.32066461f, 0.31445843f, 0.30823877f, 0.30200592f, 0.29576012f, +0.28950164f, 0.28323078f, 0.27694780f, 0.27065292f, 0.26434645f, +0.25802869f, 0.25169984f, 0.24536023f, 0.23901010f, 0.23264973f, +0.22627939f, 0.21989937f, 0.21350993f, 0.20711134f, 0.20070387f, +0.19428782f, 0.18786344f, 0.18143101f, 0.17499080f, 0.16854310f, +0.16208819f, 0.15562633f, 0.14915779f, 0.14268288f, 0.13620184f, +0.12971498f, 0.12322257f, 0.11672486f, 0.11022217f, 0.10371475f, +0.097202882f, 0.090686858f, 0.084166944f, 0.077643424f, 0.071116582f, +0.064586692f, 0.058054037f, 0.051518895f, 0.044981543f, 0.038442269f, +0.031901345f, 0.025359053f, 0.018815678f, 0.012271495f, 0.0057267868f, +-0.00081816671f, -0.0073630852f, -0.013907688f, -0.020451695f, -0.026994826f, +-0.033536803f, -0.040077340f, -0.046616159f, -0.053152986f, -0.059687532f, +-0.066219524f, -0.072748676f, -0.079274714f, -0.085797355f, -0.092316322f, +-0.098831341f, -0.10534211f, -0.11184838f, -0.11834986f, -0.12484626f, +-0.13133731f, -0.13782275f, -0.14430228f, -0.15077563f, -0.15724251f, +-0.16370267f, -0.17015581f, -0.17660165f, -0.18303993f, -0.18947038f, +-0.19589271f, -0.20230664f, -0.20871192f, -0.21510825f, -0.22149536f, +-0.22787298f, -0.23424086f, -0.24059868f, -0.24694622f, -0.25328314f, +-0.25960925f, -0.26592422f, -0.27222782f, -0.27851975f, -0.28479972f, +-0.29106751f, -0.29732284f, -0.30356544f, -0.30979502f, -0.31601134f, +-0.32221413f, -0.32840309f, -0.33457801f, -0.34073856f, -0.34688455f, +-0.35301566f, -0.35913166f, -0.36523229f, -0.37131724f, -0.37738630f, +-0.38343921f, -0.38947567f, -0.39549544f, -0.40149832f, -0.40748394f, +-0.41345215f, -0.41940263f, -0.42533514f, -0.43124944f, -0.43714526f, +-0.44302234f, -0.44888046f, -0.45471936f, -0.46053877f, -0.46633846f, +-0.47211814f, -0.47787762f, -0.48361665f, -0.48933494f, -0.49503228f, +-0.50070840f, -0.50636309f, -0.51199609f, -0.51760709f, -0.52319598f, +-0.52876246f, -0.53430629f, -0.53982723f, -0.54532504f, -0.55079949f, +-0.55625033f, -0.56167740f, -0.56708032f, -0.57245898f, -0.57781315f, +-0.58314258f, -0.58844697f, -0.59372622f, -0.59897995f, -0.60420811f, +-0.60941035f, -0.61458647f, -0.61973625f, -0.62485951f, -0.62995601f, +-0.63502556f, -0.64006782f, -0.64508271f, -0.65007001f, -0.65502942f, +-0.65996075f, -0.66486382f, -0.66973841f, -0.67458433f, -0.67940134f, +-0.68418926f, -0.68894786f, -0.69367695f, -0.69837630f, -0.70304573f, +-0.70768511f, -0.71229410f, -0.71687263f, -0.72142041f, -0.72593731f, +-0.73042315f, -0.73487765f, -0.73930067f, -0.74369204f, -0.74805158f, +-0.75237900f, -0.75667429f, -0.76093709f, -0.76516730f, -0.76936477f, +-0.77352923f, -0.77766061f, -0.78175867f, -0.78582323f, -0.78985411f, +-0.79385114f, -0.79781419f, -0.80174309f, -0.80563760f, -0.80949765f, +-0.81332302f, -0.81711352f, -0.82086903f, -0.82458937f, -0.82827437f, +-0.83192390f, -0.83553779f, -0.83911592f, -0.84265804f, -0.84616417f, +-0.84963393f, -0.85306740f, -0.85646427f, -0.85982448f, -0.86314780f, +-0.86643422f, -0.86968350f, -0.87289548f, -0.87607014f, -0.87920725f, +-0.88230664f, -0.88536829f, -0.88839203f, -0.89137769f, -0.89432514f, +-0.89723432f, -0.90010506f, -0.90293723f, -0.90573072f, -0.90848541f, +-0.91120118f, -0.91387796f, -0.91651553f, -0.91911387f, -0.92167282f, +-0.92419231f, -0.92667222f, -0.92911243f, -0.93151283f, -0.93387336f, +-0.93619382f, -0.93847424f, -0.94071442f, -0.94291431f, -0.94507378f, +-0.94719279f, -0.94927126f, -0.95130903f, -0.95330608f, -0.95526224f, +-0.95717752f, -0.95905179f, -0.96088499f, -0.96267700f, -0.96442777f, +-0.96613729f, -0.96780539f, -0.96943200f, -0.97101706f, -0.97256058f, +-0.97406244f, -0.97552258f, -0.97694093f, -0.97831738f, -0.97965199f, +-0.98094457f, -0.98219514f, -0.98340368f, -0.98457009f, -0.98569429f, +-0.98677629f, -0.98781598f, -0.98881340f, -0.98976845f, -0.99068111f, +-0.99155134f, -0.99237907f, -0.99316430f, -0.99390697f, -0.99460709f, +-0.99526459f, -0.99587947f, -0.99645168f, -0.99698120f, -0.99746799f, +-0.99791211f, -0.99831343f, -0.99867201f, -0.99898779f, -0.99926084f, +-0.99949104f, -0.99967843f, -0.99982297f, -0.99992472f, -0.99998361f, +0.99999869f, 0.99989158f, 0.99961317f, 0.99916345f, 0.99854255f, +0.99775058f, 0.99678761f, 0.99565387f, 0.99434954f, 0.99287480f, +0.99122995f, 0.98941529f, 0.98743105f, 0.98527765f, 0.98295540f, +0.98046476f, 0.97780609f, 0.97497988f, 0.97198665f, 0.96882683f, +0.96550101f, 0.96200979f, 0.95835376f, 0.95453346f, 0.95054960f, +0.94640291f, 0.94209403f, 0.93762374f, 0.93299282f, 0.92820197f, +0.92325211f, 0.91814411f, 0.91287869f, 0.90745693f, 0.90187967f, +0.89614785f, 0.89026248f, 0.88422459f, 0.87803519f, 0.87169534f, +0.86520612f, 0.85856867f, 0.85178405f, 0.84485358f, 0.83777827f, +0.83055943f, 0.82319832f, 0.81569612f, 0.80805415f, 0.80027372f, +0.79235619f, 0.78430289f, 0.77611518f, 0.76779449f, 0.75934225f, +0.75075996f, 0.74204898f, 0.73321080f, 0.72424710f, 0.71515924f, +0.70594883f, 0.69661748f, 0.68716675f, 0.67759830f, 0.66791373f, +0.65811473f, 0.64820296f, 0.63818014f, 0.62804794f, 0.61780810f, +0.60746247f, 0.59701276f, 0.58646071f, 0.57580817f, 0.56505698f, +0.55420899f, 0.54326600f, 0.53222996f, 0.52110273f, 0.50988621f, +0.49858227f, 0.48719296f, 0.47572014f, 0.46416581f, 0.45253196f, +0.44082057f, 0.42903364f, 0.41717321f, 0.40524128f, 0.39323992f, +0.38117120f, 0.36903715f, 0.35683987f, 0.34458145f, 0.33226398f, +0.31988961f, 0.30746040f, 0.29497850f, 0.28244606f, 0.26986524f, +0.25723818f, 0.24456702f, 0.23185398f, 0.21910121f, 0.20631088f, +0.19348522f, 0.18062639f, 0.16773662f, 0.15481812f, 0.14187308f, +0.12890373f, 0.11591230f, 0.10290100f, 0.089872077f, 0.076827750f, +0.063770257f, 0.050701842f, 0.037624735f, 0.024541186f, 0.011453429f, +-0.0016362892f, -0.014725727f, -0.027812643f, -0.040894791f, -0.053969935f, +-0.067035832f, -0.080090240f, -0.093130924f, -0.10615565f, -0.11916219f, +-0.13214831f, -0.14511178f, -0.15805040f, -0.17096193f, -0.18384418f, +-0.19669491f, -0.20951195f, -0.22229309f, -0.23503613f, -0.24773891f, +-0.26039925f, -0.27301496f, -0.28558388f, -0.29810387f, -0.31057280f, +-0.32298848f, -0.33534884f, -0.34765175f, -0.35989508f, -0.37207675f, +-0.38419467f, -0.39624676f, -0.40823093f, -0.42014518f, -0.43198743f, +-0.44375566f, -0.45544785f, -0.46706200f, -0.47859612f, -0.49004826f, +-0.50141639f, -0.51269865f, -0.52389306f, -0.53499764f, -0.54601061f, +-0.55693001f, -0.56775403f, -0.57848072f, -0.58910829f, -0.59963489f, +-0.61005878f, -0.62037814f, -0.63059121f, -0.64069623f, -0.65069145f, +-0.66057515f, -0.67034572f, -0.68000144f, -0.68954057f, -0.69896162f, +-0.70826286f, -0.71744281f, -0.72649974f, -0.73543227f, -0.74423873f, +-0.75291771f, -0.76146764f, -0.76988715f, -0.77817470f, -0.78632891f, +-0.79434842f, -0.80223179f, -0.80997771f, -0.81758487f, -0.82505190f, +-0.83237761f, -0.83956063f, -0.84659988f, -0.85349399f, -0.86024189f, +-0.86684239f, -0.87329435f, -0.87959671f, -0.88574833f, -0.89174819f, +-0.89759529f, -0.90328854f, -0.90882701f, -0.91420978f, -0.91943592f, +-0.92450452f, -0.92941469f, -0.93416560f, -0.93875647f, -0.94318646f, +-0.94745487f, -0.95156091f, -0.95550388f, -0.95928317f, -0.96289814f, +-0.96634805f, -0.96963239f, -0.97275060f, -0.97570217f, -0.97848648f, +-0.98110318f, -0.98355180f, -0.98583186f, -0.98794299f, -0.98988485f, +-0.99165714f, -0.99325943f, -0.99469161f, -0.99595332f, -0.99704438f, +-0.99796462f, -0.99871385f, -0.99929196f, -0.99969882f, -0.99993443f, +0.99999464f, 0.99956632f, 0.99845290f, 0.99665523f, 0.99417448f, +0.99101239f, 0.98717111f, 0.98265326f, 0.97746199f, 0.97160077f, +0.96507365f, 0.95788515f, 0.95004016f, 0.94154406f, 0.93240267f, +0.92262226f, 0.91220951f, 0.90117162f, 0.88951606f, 0.87725091f, +0.86438453f, 0.85092574f, 0.83688372f, 0.82226819f, 0.80708915f, +0.79135692f, 0.77508235f, 0.75827658f, 0.74095112f, 0.72311783f, +0.70478898f, 0.68597710f, 0.66669506f, 0.64695615f, 0.62677377f, +0.60616189f, 0.58513457f, 0.56370622f, 0.54189157f, 0.51970547f, +0.49716324f, 0.47428027f, 0.45107225f, 0.42755505f, 0.40374488f, +0.37965798f, 0.35531086f, 0.33072025f, 0.30590299f, 0.28087607f, +0.25565663f, 0.23026201f, 0.20470956f, 0.17901683f, 0.15320139f, +0.12728097f, 0.10127331f, 0.075196236f, 0.049067631f, 0.022905400f, +-0.0032725304f, -0.029448219f, -0.055603724f, -0.081721120f, -0.10778251f, +-0.13377003f, -0.15966587f, -0.18545228f, -0.21111161f, -0.23662624f, +-0.26197869f, -0.28715160f, -0.31212771f, -0.33688989f, -0.36142120f, +-0.38570482f, -0.40972409f, -0.43346253f, -0.45690393f, -0.48003218f, +-0.50283146f, -0.52528608f, -0.54738069f, -0.56910020f, -0.59042966f, +-0.61135447f, -0.63186026f, -0.65193301f, -0.67155898f, -0.69072473f, +-0.70941705f, -0.72762316f, -0.74533063f, -0.76252723f, -0.77920127f, +-0.79534131f, -0.81093621f, -0.82597536f, -0.84044844f, -0.85434550f, +-0.86765707f, -0.88037395f, -0.89248747f, -0.90398932f, -0.91487163f, +-0.92512697f, -0.93474823f, -0.94372886f, -0.95206273f, -0.95974404f, +-0.96676767f, -0.97312868f, -0.97882277f, -0.98384601f, -0.98819500f, +-0.99186671f, -0.99485862f, -0.99716878f, -0.99879545f, -0.99973762f, +}; +#endif + +static const CELTMode mode48000_960_120 = { +48000, /* Fs */ +120, /* overlap */ +21, /* nbEBands */ +21, /* effEBands */ +{0.85000610f, 0.0000000f, 1.0000000f, 1.0000000f, }, /* preemph */ +eband5ms, /* eBands */ +3, /* maxLM */ +8, /* nbShortMdcts */ +120, /* shortMdctSize */ +11, /* nbAllocVectors */ +band_allocation, /* allocVectors */ +logN400, /* logN */ +window120, /* window */ +{1920, 3, {&fft_state48000_960_0, &fft_state48000_960_1, &fft_state48000_960_2, &fft_state48000_960_3, }, mdct_twiddles960}, /* mdct */ +{392, cache_index50, cache_bits50, cache_caps50}, /* cache */ +}; + +/* List of all the available modes */ +#define TOTAL_MODES 1 +static const CELTMode * const static_mode_list[TOTAL_MODES] = { +&mode48000_960_120, +}; diff --git a/media/libopus/celt/static_modes_float_arm_ne10.h b/media/libopus/celt/static_modes_float_arm_ne10.h new file mode 100644 index 0000000000..66e1abb101 --- /dev/null +++ b/media/libopus/celt/static_modes_float_arm_ne10.h @@ -0,0 +1,404 @@ +/* The contents of this file was automatically generated by + * dump_mode_arm_ne10.c with arguments: 48000 960 + * It contains static definitions for some pre-defined modes. */ +#include <NE10_types.h> + +#ifndef NE10_FFT_PARAMS48000_960 +#define NE10_FFT_PARAMS48000_960 +static const ne10_int32_t ne10_factors_480[64] = { +4, 40, 4, 30, 2, 15, 5, 3, 3, 1, 1, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, }; +static const ne10_int32_t ne10_factors_240[64] = { +3, 20, 4, 15, 5, 3, 3, 1, 1, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, }; +static const ne10_int32_t ne10_factors_120[64] = { +3, 10, 2, 15, 5, 3, 3, 1, 1, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, }; +static const ne10_int32_t ne10_factors_60[64] = { +2, 5, 5, 3, 3, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, +0, 0, 0, 0, }; +static const ne10_fft_cpx_float32_t ne10_twiddles_480[480] = { +{1.0000000f,0.0000000f}, {1.0000000f,-0.0000000f}, {1.0000000f,-0.0000000f}, +{1.0000000f,-0.0000000f}, {0.91354543f,-0.40673664f}, {0.66913056f,-0.74314487f}, +{1.0000000f,-0.0000000f}, {0.66913056f,-0.74314487f}, {-0.10452851f,-0.99452192f}, +{1.0000000f,-0.0000000f}, {0.30901697f,-0.95105654f}, {-0.80901700f,-0.58778518f}, +{1.0000000f,-0.0000000f}, {-0.10452851f,-0.99452192f}, {-0.97814757f,0.20791179f}, +{1.0000000f,-0.0000000f}, {0.97814763f,-0.20791170f}, {0.91354543f,-0.40673664f}, +{0.80901700f,-0.58778524f}, {0.66913056f,-0.74314487f}, {0.49999997f,-0.86602545f}, +{0.30901697f,-0.95105654f}, {0.10452842f,-0.99452192f}, {-0.10452851f,-0.99452192f}, +{-0.30901703f,-0.95105648f}, {-0.50000006f,-0.86602533f}, {-0.66913068f,-0.74314475f}, +{-0.80901700f,-0.58778518f}, {-0.91354549f,-0.40673658f}, {-0.97814763f,-0.20791161f}, +{1.0000000f,-0.0000000f}, {0.99862951f,-0.052335959f}, {0.99452192f,-0.10452846f}, +{0.98768836f,-0.15643448f}, {0.97814763f,-0.20791170f}, {0.96592581f,-0.25881904f}, +{0.95105648f,-0.30901700f}, {0.93358040f,-0.35836795f}, {0.91354543f,-0.40673664f}, +{0.89100653f,-0.45399052f}, {0.86602545f,-0.50000000f}, {0.83867055f,-0.54463905f}, +{0.80901700f,-0.58778524f}, {0.77714598f,-0.62932038f}, {0.74314475f,-0.66913062f}, +{0.70710677f,-0.70710683f}, {0.66913056f,-0.74314487f}, {0.62932038f,-0.77714598f}, +{0.58778524f,-0.80901700f}, {0.54463899f,-0.83867055f}, {0.49999997f,-0.86602545f}, +{0.45399052f,-0.89100653f}, {0.40673661f,-0.91354549f}, {0.35836786f,-0.93358046f}, +{0.30901697f,-0.95105654f}, {0.25881907f,-0.96592581f}, {0.20791166f,-0.97814763f}, +{0.15643437f,-0.98768836f}, {0.10452842f,-0.99452192f}, {0.052335974f,-0.99862951f}, +{1.0000000f,-0.0000000f}, {0.99452192f,-0.10452846f}, {0.97814763f,-0.20791170f}, +{0.95105648f,-0.30901700f}, {0.91354543f,-0.40673664f}, {0.86602545f,-0.50000000f}, +{0.80901700f,-0.58778524f}, {0.74314475f,-0.66913062f}, {0.66913056f,-0.74314487f}, +{0.58778524f,-0.80901700f}, {0.49999997f,-0.86602545f}, {0.40673661f,-0.91354549f}, +{0.30901697f,-0.95105654f}, {0.20791166f,-0.97814763f}, {0.10452842f,-0.99452192f}, +{-4.3711388e-08f,-1.0000000f}, {-0.10452851f,-0.99452192f}, {-0.20791174f,-0.97814757f}, +{-0.30901703f,-0.95105648f}, {-0.40673670f,-0.91354543f}, {-0.50000006f,-0.86602533f}, +{-0.58778518f,-0.80901700f}, {-0.66913068f,-0.74314475f}, {-0.74314493f,-0.66913044f}, +{-0.80901700f,-0.58778518f}, {-0.86602539f,-0.50000006f}, {-0.91354549f,-0.40673658f}, +{-0.95105654f,-0.30901679f}, {-0.97814763f,-0.20791161f}, {-0.99452192f,-0.10452849f}, +{1.0000000f,-0.0000000f}, {0.98768836f,-0.15643448f}, {0.95105648f,-0.30901700f}, +{0.89100653f,-0.45399052f}, {0.80901700f,-0.58778524f}, {0.70710677f,-0.70710683f}, +{0.58778524f,-0.80901700f}, {0.45399052f,-0.89100653f}, {0.30901697f,-0.95105654f}, +{0.15643437f,-0.98768836f}, {-4.3711388e-08f,-1.0000000f}, {-0.15643445f,-0.98768836f}, +{-0.30901703f,-0.95105648f}, {-0.45399061f,-0.89100647f}, {-0.58778518f,-0.80901700f}, +{-0.70710677f,-0.70710677f}, {-0.80901700f,-0.58778518f}, {-0.89100659f,-0.45399037f}, +{-0.95105654f,-0.30901679f}, {-0.98768836f,-0.15643445f}, {-1.0000000f,8.7422777e-08f}, +{-0.98768830f,0.15643461f}, {-0.95105654f,0.30901697f}, {-0.89100653f,0.45399055f}, +{-0.80901694f,0.58778536f}, {-0.70710665f,0.70710689f}, {-0.58778507f,0.80901712f}, +{-0.45399022f,0.89100665f}, {-0.30901709f,0.95105648f}, {-0.15643452f,0.98768830f}, +{1.0000000f,-0.0000000f}, {0.99991435f,-0.013089596f}, {0.99965733f,-0.026176950f}, +{0.99922901f,-0.039259817f}, {0.99862951f,-0.052335959f}, {0.99785894f,-0.065403134f}, +{0.99691731f,-0.078459099f}, {0.99580491f,-0.091501623f}, {0.99452192f,-0.10452846f}, +{0.99306846f,-0.11753740f}, {0.99144489f,-0.13052620f}, {0.98965138f,-0.14349262f}, +{0.98768836f,-0.15643448f}, {0.98555607f,-0.16934951f}, {0.98325491f,-0.18223552f}, +{0.98078525f,-0.19509032f}, {0.97814763f,-0.20791170f}, {0.97534233f,-0.22069745f}, +{0.97236991f,-0.23344538f}, {0.96923089f,-0.24615330f}, {0.96592581f,-0.25881904f}, +{0.96245521f,-0.27144045f}, {0.95881975f,-0.28401536f}, {0.95501995f,-0.29654160f}, +{0.95105648f,-0.30901700f}, {0.94693011f,-0.32143945f}, {0.94264150f,-0.33380687f}, +{0.93819129f,-0.34611708f}, {0.93358040f,-0.35836795f}, {0.92880952f,-0.37055743f}, +{0.92387956f,-0.38268346f}, {0.91879117f,-0.39474389f}, {0.91354543f,-0.40673664f}, +{0.90814316f,-0.41865975f}, {0.90258527f,-0.43051112f}, {0.89687270f,-0.44228873f}, +{0.89100653f,-0.45399052f}, {0.88498765f,-0.46561453f}, {0.87881708f,-0.47715878f}, +{0.87249601f,-0.48862126f}, {0.86602545f,-0.50000000f}, {0.85940641f,-0.51129311f}, +{0.85264015f,-0.52249855f}, {0.84572786f,-0.53361452f}, {0.83867055f,-0.54463905f}, +{0.83146960f,-0.55557024f}, {0.82412618f,-0.56640625f}, {0.81664151f,-0.57714522f}, +{0.80901700f,-0.58778524f}, {0.80125380f,-0.59832460f}, {0.79335332f,-0.60876143f}, +{0.78531694f,-0.61909395f}, {0.77714598f,-0.62932038f}, {0.76884180f,-0.63943899f}, +{0.76040596f,-0.64944810f}, {0.75183982f,-0.65934587f}, {0.74314475f,-0.66913062f}, +{0.73432249f,-0.67880076f}, {0.72537434f,-0.68835455f}, {0.71630192f,-0.69779050f}, +{0.70710677f,-0.70710683f}, {0.69779044f,-0.71630198f}, {0.68835455f,-0.72537440f}, +{0.67880070f,-0.73432255f}, {0.66913056f,-0.74314487f}, {0.65934581f,-0.75183982f}, +{0.64944804f,-0.76040596f}, {0.63943899f,-0.76884186f}, {0.62932038f,-0.77714598f}, +{0.61909395f,-0.78531694f}, {0.60876137f,-0.79335338f}, {0.59832460f,-0.80125386f}, +{0.58778524f,-0.80901700f}, {0.57714516f,-0.81664151f}, {0.56640625f,-0.82412618f}, +{0.55557019f,-0.83146960f}, {0.54463899f,-0.83867055f}, {0.53361452f,-0.84572786f}, +{0.52249849f,-0.85264015f}, {0.51129311f,-0.85940641f}, {0.49999997f,-0.86602545f}, +{0.48862118f,-0.87249601f}, {0.47715876f,-0.87881708f}, {0.46561447f,-0.88498765f}, +{0.45399052f,-0.89100653f}, {0.44228867f,-0.89687276f}, {0.43051103f,-0.90258533f}, +{0.41865975f,-0.90814316f}, {0.40673661f,-0.91354549f}, {0.39474380f,-0.91879129f}, +{0.38268343f,-0.92387956f}, {0.37055740f,-0.92880958f}, {0.35836786f,-0.93358046f}, +{0.34611705f,-0.93819135f}, {0.33380681f,-0.94264150f}, {0.32143947f,-0.94693011f}, +{0.30901697f,-0.95105654f}, {0.29654151f,-0.95501995f}, {0.28401533f,-0.95881975f}, +{0.27144039f,-0.96245527f}, {0.25881907f,-0.96592581f}, {0.24615327f,-0.96923089f}, +{0.23344530f,-0.97236991f}, {0.22069745f,-0.97534233f}, {0.20791166f,-0.97814763f}, +{0.19509023f,-0.98078531f}, {0.18223552f,-0.98325491f}, {0.16934945f,-0.98555607f}, +{0.15643437f,-0.98768836f}, {0.14349259f,-0.98965138f}, {0.13052613f,-0.99144489f}, +{0.11753740f,-0.99306846f}, {0.10452842f,-0.99452192f}, {0.091501534f,-0.99580491f}, +{0.078459084f,-0.99691731f}, {0.065403074f,-0.99785894f}, {0.052335974f,-0.99862951f}, +{0.039259788f,-0.99922901f}, {0.026176875f,-0.99965733f}, {0.013089597f,-0.99991435f}, +{1.0000000f,-0.0000000f}, {0.99965733f,-0.026176950f}, {0.99862951f,-0.052335959f}, +{0.99691731f,-0.078459099f}, {0.99452192f,-0.10452846f}, {0.99144489f,-0.13052620f}, +{0.98768836f,-0.15643448f}, {0.98325491f,-0.18223552f}, {0.97814763f,-0.20791170f}, +{0.97236991f,-0.23344538f}, {0.96592581f,-0.25881904f}, {0.95881975f,-0.28401536f}, +{0.95105648f,-0.30901700f}, {0.94264150f,-0.33380687f}, {0.93358040f,-0.35836795f}, +{0.92387956f,-0.38268346f}, {0.91354543f,-0.40673664f}, {0.90258527f,-0.43051112f}, +{0.89100653f,-0.45399052f}, {0.87881708f,-0.47715878f}, {0.86602545f,-0.50000000f}, +{0.85264015f,-0.52249855f}, {0.83867055f,-0.54463905f}, {0.82412618f,-0.56640625f}, +{0.80901700f,-0.58778524f}, {0.79335332f,-0.60876143f}, {0.77714598f,-0.62932038f}, +{0.76040596f,-0.64944810f}, {0.74314475f,-0.66913062f}, {0.72537434f,-0.68835455f}, +{0.70710677f,-0.70710683f}, {0.68835455f,-0.72537440f}, {0.66913056f,-0.74314487f}, +{0.64944804f,-0.76040596f}, {0.62932038f,-0.77714598f}, {0.60876137f,-0.79335338f}, +{0.58778524f,-0.80901700f}, {0.56640625f,-0.82412618f}, {0.54463899f,-0.83867055f}, +{0.52249849f,-0.85264015f}, {0.49999997f,-0.86602545f}, {0.47715876f,-0.87881708f}, +{0.45399052f,-0.89100653f}, {0.43051103f,-0.90258533f}, {0.40673661f,-0.91354549f}, +{0.38268343f,-0.92387956f}, {0.35836786f,-0.93358046f}, {0.33380681f,-0.94264150f}, +{0.30901697f,-0.95105654f}, {0.28401533f,-0.95881975f}, {0.25881907f,-0.96592581f}, +{0.23344530f,-0.97236991f}, {0.20791166f,-0.97814763f}, {0.18223552f,-0.98325491f}, +{0.15643437f,-0.98768836f}, {0.13052613f,-0.99144489f}, {0.10452842f,-0.99452192f}, +{0.078459084f,-0.99691731f}, {0.052335974f,-0.99862951f}, {0.026176875f,-0.99965733f}, +{-4.3711388e-08f,-1.0000000f}, {-0.026176963f,-0.99965733f}, {-0.052336060f,-0.99862951f}, +{-0.078459173f,-0.99691731f}, {-0.10452851f,-0.99452192f}, {-0.13052621f,-0.99144489f}, +{-0.15643445f,-0.98768836f}, {-0.18223560f,-0.98325491f}, {-0.20791174f,-0.97814757f}, +{-0.23344538f,-0.97236991f}, {-0.25881916f,-0.96592581f}, {-0.28401542f,-0.95881969f}, +{-0.30901703f,-0.95105648f}, {-0.33380687f,-0.94264150f}, {-0.35836795f,-0.93358040f}, +{-0.38268352f,-0.92387950f}, {-0.40673670f,-0.91354543f}, {-0.43051112f,-0.90258527f}, +{-0.45399061f,-0.89100647f}, {-0.47715873f,-0.87881708f}, {-0.50000006f,-0.86602533f}, +{-0.52249867f,-0.85264009f}, {-0.54463905f,-0.83867055f}, {-0.56640631f,-0.82412612f}, +{-0.58778518f,-0.80901700f}, {-0.60876143f,-0.79335332f}, {-0.62932050f,-0.77714586f}, +{-0.64944804f,-0.76040596f}, {-0.66913068f,-0.74314475f}, {-0.68835467f,-0.72537428f}, +{-0.70710677f,-0.70710677f}, {-0.72537446f,-0.68835449f}, {-0.74314493f,-0.66913044f}, +{-0.76040596f,-0.64944804f}, {-0.77714604f,-0.62932026f}, {-0.79335332f,-0.60876143f}, +{-0.80901700f,-0.58778518f}, {-0.82412624f,-0.56640613f}, {-0.83867055f,-0.54463899f}, +{-0.85264021f,-0.52249849f}, {-0.86602539f,-0.50000006f}, {-0.87881714f,-0.47715873f}, +{-0.89100659f,-0.45399037f}, {-0.90258527f,-0.43051112f}, {-0.91354549f,-0.40673658f}, +{-0.92387956f,-0.38268328f}, {-0.93358040f,-0.35836792f}, {-0.94264150f,-0.33380675f}, +{-0.95105654f,-0.30901679f}, {-0.95881975f,-0.28401530f}, {-0.96592587f,-0.25881892f}, +{-0.97236991f,-0.23344538f}, {-0.97814763f,-0.20791161f}, {-0.98325491f,-0.18223536f}, +{-0.98768836f,-0.15643445f}, {-0.99144489f,-0.13052608f}, {-0.99452192f,-0.10452849f}, +{-0.99691737f,-0.078459039f}, {-0.99862957f,-0.052335810f}, {-0.99965733f,-0.026176952f}, +{1.0000000f,-0.0000000f}, {0.99922901f,-0.039259817f}, {0.99691731f,-0.078459099f}, +{0.99306846f,-0.11753740f}, {0.98768836f,-0.15643448f}, {0.98078525f,-0.19509032f}, +{0.97236991f,-0.23344538f}, {0.96245521f,-0.27144045f}, {0.95105648f,-0.30901700f}, +{0.93819129f,-0.34611708f}, {0.92387956f,-0.38268346f}, {0.90814316f,-0.41865975f}, +{0.89100653f,-0.45399052f}, {0.87249601f,-0.48862126f}, {0.85264015f,-0.52249855f}, +{0.83146960f,-0.55557024f}, {0.80901700f,-0.58778524f}, {0.78531694f,-0.61909395f}, +{0.76040596f,-0.64944810f}, {0.73432249f,-0.67880076f}, {0.70710677f,-0.70710683f}, +{0.67880070f,-0.73432255f}, {0.64944804f,-0.76040596f}, {0.61909395f,-0.78531694f}, +{0.58778524f,-0.80901700f}, {0.55557019f,-0.83146960f}, {0.52249849f,-0.85264015f}, +{0.48862118f,-0.87249601f}, {0.45399052f,-0.89100653f}, {0.41865975f,-0.90814316f}, +{0.38268343f,-0.92387956f}, {0.34611705f,-0.93819135f}, {0.30901697f,-0.95105654f}, +{0.27144039f,-0.96245527f}, {0.23344530f,-0.97236991f}, {0.19509023f,-0.98078531f}, +{0.15643437f,-0.98768836f}, {0.11753740f,-0.99306846f}, {0.078459084f,-0.99691731f}, +{0.039259788f,-0.99922901f}, {-4.3711388e-08f,-1.0000000f}, {-0.039259877f,-0.99922901f}, +{-0.078459173f,-0.99691731f}, {-0.11753749f,-0.99306846f}, {-0.15643445f,-0.98768836f}, +{-0.19509032f,-0.98078525f}, {-0.23344538f,-0.97236991f}, {-0.27144048f,-0.96245521f}, +{-0.30901703f,-0.95105648f}, {-0.34611711f,-0.93819129f}, {-0.38268352f,-0.92387950f}, +{-0.41865984f,-0.90814310f}, {-0.45399061f,-0.89100647f}, {-0.48862135f,-0.87249595f}, +{-0.52249867f,-0.85264009f}, {-0.55557036f,-0.83146954f}, {-0.58778518f,-0.80901700f}, +{-0.61909389f,-0.78531694f}, {-0.64944804f,-0.76040596f}, {-0.67880076f,-0.73432249f}, +{-0.70710677f,-0.70710677f}, {-0.73432249f,-0.67880070f}, {-0.76040596f,-0.64944804f}, +{-0.78531694f,-0.61909389f}, {-0.80901700f,-0.58778518f}, {-0.83146966f,-0.55557019f}, +{-0.85264021f,-0.52249849f}, {-0.87249607f,-0.48862115f}, {-0.89100659f,-0.45399037f}, +{-0.90814322f,-0.41865960f}, {-0.92387956f,-0.38268328f}, {-0.93819135f,-0.34611690f}, +{-0.95105654f,-0.30901679f}, {-0.96245521f,-0.27144048f}, {-0.97236991f,-0.23344538f}, +{-0.98078531f,-0.19509031f}, {-0.98768836f,-0.15643445f}, {-0.99306846f,-0.11753736f}, +{-0.99691737f,-0.078459039f}, {-0.99922901f,-0.039259743f}, {-1.0000000f,8.7422777e-08f}, +{-0.99922901f,0.039259918f}, {-0.99691731f,0.078459218f}, {-0.99306846f,0.11753753f}, +{-0.98768830f,0.15643461f}, {-0.98078525f,0.19509049f}, {-0.97236985f,0.23344554f}, +{-0.96245515f,0.27144065f}, {-0.95105654f,0.30901697f}, {-0.93819135f,0.34611705f}, +{-0.92387956f,0.38268346f}, {-0.90814316f,0.41865975f}, {-0.89100653f,0.45399055f}, +{-0.87249601f,0.48862129f}, {-0.85264015f,0.52249861f}, {-0.83146960f,0.55557030f}, +{-0.80901694f,0.58778536f}, {-0.78531688f,0.61909401f}, {-0.76040590f,0.64944816f}, +{-0.73432243f,0.67880082f}, {-0.70710665f,0.70710689f}, {-0.67880058f,0.73432261f}, +{-0.64944792f,0.76040608f}, {-0.61909378f,0.78531706f}, {-0.58778507f,0.80901712f}, +{-0.55557001f,0.83146977f}, {-0.52249837f,0.85264033f}, {-0.48862100f,0.87249613f}, +{-0.45399022f,0.89100665f}, {-0.41865945f,0.90814328f}, {-0.38268313f,0.92387968f}, +{-0.34611672f,0.93819147f}, {-0.30901709f,0.95105648f}, {-0.27144054f,0.96245521f}, +{-0.23344545f,0.97236991f}, {-0.19509038f,0.98078525f}, {-0.15643452f,0.98768830f}, +{-0.11753743f,0.99306846f}, {-0.078459114f,0.99691731f}, {-0.039259821f,0.99922901f}, +}; +static const ne10_fft_cpx_float32_t ne10_twiddles_240[240] = { +{1.0000000f,0.0000000f}, {1.0000000f,-0.0000000f}, {1.0000000f,-0.0000000f}, +{1.0000000f,-0.0000000f}, {0.91354543f,-0.40673664f}, {0.66913056f,-0.74314487f}, +{1.0000000f,-0.0000000f}, {0.66913056f,-0.74314487f}, {-0.10452851f,-0.99452192f}, +{1.0000000f,-0.0000000f}, {0.30901697f,-0.95105654f}, {-0.80901700f,-0.58778518f}, +{1.0000000f,-0.0000000f}, {-0.10452851f,-0.99452192f}, {-0.97814757f,0.20791179f}, +{1.0000000f,-0.0000000f}, {0.99452192f,-0.10452846f}, {0.97814763f,-0.20791170f}, +{0.95105648f,-0.30901700f}, {0.91354543f,-0.40673664f}, {0.86602545f,-0.50000000f}, +{0.80901700f,-0.58778524f}, {0.74314475f,-0.66913062f}, {0.66913056f,-0.74314487f}, +{0.58778524f,-0.80901700f}, {0.49999997f,-0.86602545f}, {0.40673661f,-0.91354549f}, +{0.30901697f,-0.95105654f}, {0.20791166f,-0.97814763f}, {0.10452842f,-0.99452192f}, +{1.0000000f,-0.0000000f}, {0.97814763f,-0.20791170f}, {0.91354543f,-0.40673664f}, +{0.80901700f,-0.58778524f}, {0.66913056f,-0.74314487f}, {0.49999997f,-0.86602545f}, +{0.30901697f,-0.95105654f}, {0.10452842f,-0.99452192f}, {-0.10452851f,-0.99452192f}, +{-0.30901703f,-0.95105648f}, {-0.50000006f,-0.86602533f}, {-0.66913068f,-0.74314475f}, +{-0.80901700f,-0.58778518f}, {-0.91354549f,-0.40673658f}, {-0.97814763f,-0.20791161f}, +{1.0000000f,-0.0000000f}, {0.95105648f,-0.30901700f}, {0.80901700f,-0.58778524f}, +{0.58778524f,-0.80901700f}, {0.30901697f,-0.95105654f}, {-4.3711388e-08f,-1.0000000f}, +{-0.30901703f,-0.95105648f}, {-0.58778518f,-0.80901700f}, {-0.80901700f,-0.58778518f}, +{-0.95105654f,-0.30901679f}, {-1.0000000f,8.7422777e-08f}, {-0.95105654f,0.30901697f}, +{-0.80901694f,0.58778536f}, {-0.58778507f,0.80901712f}, {-0.30901709f,0.95105648f}, +{1.0000000f,-0.0000000f}, {0.99965733f,-0.026176950f}, {0.99862951f,-0.052335959f}, +{0.99691731f,-0.078459099f}, {0.99452192f,-0.10452846f}, {0.99144489f,-0.13052620f}, +{0.98768836f,-0.15643448f}, {0.98325491f,-0.18223552f}, {0.97814763f,-0.20791170f}, +{0.97236991f,-0.23344538f}, {0.96592581f,-0.25881904f}, {0.95881975f,-0.28401536f}, +{0.95105648f,-0.30901700f}, {0.94264150f,-0.33380687f}, {0.93358040f,-0.35836795f}, +{0.92387956f,-0.38268346f}, {0.91354543f,-0.40673664f}, {0.90258527f,-0.43051112f}, +{0.89100653f,-0.45399052f}, {0.87881708f,-0.47715878f}, {0.86602545f,-0.50000000f}, +{0.85264015f,-0.52249855f}, {0.83867055f,-0.54463905f}, {0.82412618f,-0.56640625f}, +{0.80901700f,-0.58778524f}, {0.79335332f,-0.60876143f}, {0.77714598f,-0.62932038f}, +{0.76040596f,-0.64944810f}, {0.74314475f,-0.66913062f}, {0.72537434f,-0.68835455f}, +{0.70710677f,-0.70710683f}, {0.68835455f,-0.72537440f}, {0.66913056f,-0.74314487f}, +{0.64944804f,-0.76040596f}, {0.62932038f,-0.77714598f}, {0.60876137f,-0.79335338f}, +{0.58778524f,-0.80901700f}, {0.56640625f,-0.82412618f}, {0.54463899f,-0.83867055f}, +{0.52249849f,-0.85264015f}, {0.49999997f,-0.86602545f}, {0.47715876f,-0.87881708f}, +{0.45399052f,-0.89100653f}, {0.43051103f,-0.90258533f}, {0.40673661f,-0.91354549f}, +{0.38268343f,-0.92387956f}, {0.35836786f,-0.93358046f}, {0.33380681f,-0.94264150f}, +{0.30901697f,-0.95105654f}, {0.28401533f,-0.95881975f}, {0.25881907f,-0.96592581f}, +{0.23344530f,-0.97236991f}, {0.20791166f,-0.97814763f}, {0.18223552f,-0.98325491f}, +{0.15643437f,-0.98768836f}, {0.13052613f,-0.99144489f}, {0.10452842f,-0.99452192f}, +{0.078459084f,-0.99691731f}, {0.052335974f,-0.99862951f}, {0.026176875f,-0.99965733f}, +{1.0000000f,-0.0000000f}, {0.99862951f,-0.052335959f}, {0.99452192f,-0.10452846f}, +{0.98768836f,-0.15643448f}, {0.97814763f,-0.20791170f}, {0.96592581f,-0.25881904f}, +{0.95105648f,-0.30901700f}, {0.93358040f,-0.35836795f}, {0.91354543f,-0.40673664f}, +{0.89100653f,-0.45399052f}, {0.86602545f,-0.50000000f}, {0.83867055f,-0.54463905f}, +{0.80901700f,-0.58778524f}, {0.77714598f,-0.62932038f}, {0.74314475f,-0.66913062f}, +{0.70710677f,-0.70710683f}, {0.66913056f,-0.74314487f}, {0.62932038f,-0.77714598f}, +{0.58778524f,-0.80901700f}, {0.54463899f,-0.83867055f}, {0.49999997f,-0.86602545f}, +{0.45399052f,-0.89100653f}, {0.40673661f,-0.91354549f}, {0.35836786f,-0.93358046f}, +{0.30901697f,-0.95105654f}, {0.25881907f,-0.96592581f}, {0.20791166f,-0.97814763f}, +{0.15643437f,-0.98768836f}, {0.10452842f,-0.99452192f}, {0.052335974f,-0.99862951f}, +{-4.3711388e-08f,-1.0000000f}, {-0.052336060f,-0.99862951f}, {-0.10452851f,-0.99452192f}, +{-0.15643445f,-0.98768836f}, {-0.20791174f,-0.97814757f}, {-0.25881916f,-0.96592581f}, +{-0.30901703f,-0.95105648f}, {-0.35836795f,-0.93358040f}, {-0.40673670f,-0.91354543f}, +{-0.45399061f,-0.89100647f}, {-0.50000006f,-0.86602533f}, {-0.54463905f,-0.83867055f}, +{-0.58778518f,-0.80901700f}, {-0.62932050f,-0.77714586f}, {-0.66913068f,-0.74314475f}, +{-0.70710677f,-0.70710677f}, {-0.74314493f,-0.66913044f}, {-0.77714604f,-0.62932026f}, +{-0.80901700f,-0.58778518f}, {-0.83867055f,-0.54463899f}, {-0.86602539f,-0.50000006f}, +{-0.89100659f,-0.45399037f}, {-0.91354549f,-0.40673658f}, {-0.93358040f,-0.35836792f}, +{-0.95105654f,-0.30901679f}, {-0.96592587f,-0.25881892f}, {-0.97814763f,-0.20791161f}, +{-0.98768836f,-0.15643445f}, {-0.99452192f,-0.10452849f}, {-0.99862957f,-0.052335810f}, +{1.0000000f,-0.0000000f}, {0.99691731f,-0.078459099f}, {0.98768836f,-0.15643448f}, +{0.97236991f,-0.23344538f}, {0.95105648f,-0.30901700f}, {0.92387956f,-0.38268346f}, +{0.89100653f,-0.45399052f}, {0.85264015f,-0.52249855f}, {0.80901700f,-0.58778524f}, +{0.76040596f,-0.64944810f}, {0.70710677f,-0.70710683f}, {0.64944804f,-0.76040596f}, +{0.58778524f,-0.80901700f}, {0.52249849f,-0.85264015f}, {0.45399052f,-0.89100653f}, +{0.38268343f,-0.92387956f}, {0.30901697f,-0.95105654f}, {0.23344530f,-0.97236991f}, +{0.15643437f,-0.98768836f}, {0.078459084f,-0.99691731f}, {-4.3711388e-08f,-1.0000000f}, +{-0.078459173f,-0.99691731f}, {-0.15643445f,-0.98768836f}, {-0.23344538f,-0.97236991f}, +{-0.30901703f,-0.95105648f}, {-0.38268352f,-0.92387950f}, {-0.45399061f,-0.89100647f}, +{-0.52249867f,-0.85264009f}, {-0.58778518f,-0.80901700f}, {-0.64944804f,-0.76040596f}, +{-0.70710677f,-0.70710677f}, {-0.76040596f,-0.64944804f}, {-0.80901700f,-0.58778518f}, +{-0.85264021f,-0.52249849f}, {-0.89100659f,-0.45399037f}, {-0.92387956f,-0.38268328f}, +{-0.95105654f,-0.30901679f}, {-0.97236991f,-0.23344538f}, {-0.98768836f,-0.15643445f}, +{-0.99691737f,-0.078459039f}, {-1.0000000f,8.7422777e-08f}, {-0.99691731f,0.078459218f}, +{-0.98768830f,0.15643461f}, {-0.97236985f,0.23344554f}, {-0.95105654f,0.30901697f}, +{-0.92387956f,0.38268346f}, {-0.89100653f,0.45399055f}, {-0.85264015f,0.52249861f}, +{-0.80901694f,0.58778536f}, {-0.76040590f,0.64944816f}, {-0.70710665f,0.70710689f}, +{-0.64944792f,0.76040608f}, {-0.58778507f,0.80901712f}, {-0.52249837f,0.85264033f}, +{-0.45399022f,0.89100665f}, {-0.38268313f,0.92387968f}, {-0.30901709f,0.95105648f}, +{-0.23344545f,0.97236991f}, {-0.15643452f,0.98768830f}, {-0.078459114f,0.99691731f}, +}; +static const ne10_fft_cpx_float32_t ne10_twiddles_120[120] = { +{1.0000000f,0.0000000f}, {1.0000000f,-0.0000000f}, {1.0000000f,-0.0000000f}, +{1.0000000f,-0.0000000f}, {0.91354543f,-0.40673664f}, {0.66913056f,-0.74314487f}, +{1.0000000f,-0.0000000f}, {0.66913056f,-0.74314487f}, {-0.10452851f,-0.99452192f}, +{1.0000000f,-0.0000000f}, {0.30901697f,-0.95105654f}, {-0.80901700f,-0.58778518f}, +{1.0000000f,-0.0000000f}, {-0.10452851f,-0.99452192f}, {-0.97814757f,0.20791179f}, +{1.0000000f,-0.0000000f}, {0.97814763f,-0.20791170f}, {0.91354543f,-0.40673664f}, +{0.80901700f,-0.58778524f}, {0.66913056f,-0.74314487f}, {0.49999997f,-0.86602545f}, +{0.30901697f,-0.95105654f}, {0.10452842f,-0.99452192f}, {-0.10452851f,-0.99452192f}, +{-0.30901703f,-0.95105648f}, {-0.50000006f,-0.86602533f}, {-0.66913068f,-0.74314475f}, +{-0.80901700f,-0.58778518f}, {-0.91354549f,-0.40673658f}, {-0.97814763f,-0.20791161f}, +{1.0000000f,-0.0000000f}, {0.99862951f,-0.052335959f}, {0.99452192f,-0.10452846f}, +{0.98768836f,-0.15643448f}, {0.97814763f,-0.20791170f}, {0.96592581f,-0.25881904f}, +{0.95105648f,-0.30901700f}, {0.93358040f,-0.35836795f}, {0.91354543f,-0.40673664f}, +{0.89100653f,-0.45399052f}, {0.86602545f,-0.50000000f}, {0.83867055f,-0.54463905f}, +{0.80901700f,-0.58778524f}, {0.77714598f,-0.62932038f}, {0.74314475f,-0.66913062f}, +{0.70710677f,-0.70710683f}, {0.66913056f,-0.74314487f}, {0.62932038f,-0.77714598f}, +{0.58778524f,-0.80901700f}, {0.54463899f,-0.83867055f}, {0.49999997f,-0.86602545f}, +{0.45399052f,-0.89100653f}, {0.40673661f,-0.91354549f}, {0.35836786f,-0.93358046f}, +{0.30901697f,-0.95105654f}, {0.25881907f,-0.96592581f}, {0.20791166f,-0.97814763f}, +{0.15643437f,-0.98768836f}, {0.10452842f,-0.99452192f}, {0.052335974f,-0.99862951f}, +{1.0000000f,-0.0000000f}, {0.99452192f,-0.10452846f}, {0.97814763f,-0.20791170f}, +{0.95105648f,-0.30901700f}, {0.91354543f,-0.40673664f}, {0.86602545f,-0.50000000f}, +{0.80901700f,-0.58778524f}, {0.74314475f,-0.66913062f}, {0.66913056f,-0.74314487f}, +{0.58778524f,-0.80901700f}, {0.49999997f,-0.86602545f}, {0.40673661f,-0.91354549f}, +{0.30901697f,-0.95105654f}, {0.20791166f,-0.97814763f}, {0.10452842f,-0.99452192f}, +{-4.3711388e-08f,-1.0000000f}, {-0.10452851f,-0.99452192f}, {-0.20791174f,-0.97814757f}, +{-0.30901703f,-0.95105648f}, {-0.40673670f,-0.91354543f}, {-0.50000006f,-0.86602533f}, +{-0.58778518f,-0.80901700f}, {-0.66913068f,-0.74314475f}, {-0.74314493f,-0.66913044f}, +{-0.80901700f,-0.58778518f}, {-0.86602539f,-0.50000006f}, {-0.91354549f,-0.40673658f}, +{-0.95105654f,-0.30901679f}, {-0.97814763f,-0.20791161f}, {-0.99452192f,-0.10452849f}, +{1.0000000f,-0.0000000f}, {0.98768836f,-0.15643448f}, {0.95105648f,-0.30901700f}, +{0.89100653f,-0.45399052f}, {0.80901700f,-0.58778524f}, {0.70710677f,-0.70710683f}, +{0.58778524f,-0.80901700f}, {0.45399052f,-0.89100653f}, {0.30901697f,-0.95105654f}, +{0.15643437f,-0.98768836f}, {-4.3711388e-08f,-1.0000000f}, {-0.15643445f,-0.98768836f}, +{-0.30901703f,-0.95105648f}, {-0.45399061f,-0.89100647f}, {-0.58778518f,-0.80901700f}, +{-0.70710677f,-0.70710677f}, {-0.80901700f,-0.58778518f}, {-0.89100659f,-0.45399037f}, +{-0.95105654f,-0.30901679f}, {-0.98768836f,-0.15643445f}, {-1.0000000f,8.7422777e-08f}, +{-0.98768830f,0.15643461f}, {-0.95105654f,0.30901697f}, {-0.89100653f,0.45399055f}, +{-0.80901694f,0.58778536f}, {-0.70710665f,0.70710689f}, {-0.58778507f,0.80901712f}, +{-0.45399022f,0.89100665f}, {-0.30901709f,0.95105648f}, {-0.15643452f,0.98768830f}, +}; +static const ne10_fft_cpx_float32_t ne10_twiddles_60[60] = { +{1.0000000f,0.0000000f}, {1.0000000f,-0.0000000f}, {1.0000000f,-0.0000000f}, +{1.0000000f,-0.0000000f}, {0.91354543f,-0.40673664f}, {0.66913056f,-0.74314487f}, +{1.0000000f,-0.0000000f}, {0.66913056f,-0.74314487f}, {-0.10452851f,-0.99452192f}, +{1.0000000f,-0.0000000f}, {0.30901697f,-0.95105654f}, {-0.80901700f,-0.58778518f}, +{1.0000000f,-0.0000000f}, {-0.10452851f,-0.99452192f}, {-0.97814757f,0.20791179f}, +{1.0000000f,-0.0000000f}, {0.99452192f,-0.10452846f}, {0.97814763f,-0.20791170f}, +{0.95105648f,-0.30901700f}, {0.91354543f,-0.40673664f}, {0.86602545f,-0.50000000f}, +{0.80901700f,-0.58778524f}, {0.74314475f,-0.66913062f}, {0.66913056f,-0.74314487f}, +{0.58778524f,-0.80901700f}, {0.49999997f,-0.86602545f}, {0.40673661f,-0.91354549f}, +{0.30901697f,-0.95105654f}, {0.20791166f,-0.97814763f}, {0.10452842f,-0.99452192f}, +{1.0000000f,-0.0000000f}, {0.97814763f,-0.20791170f}, {0.91354543f,-0.40673664f}, +{0.80901700f,-0.58778524f}, {0.66913056f,-0.74314487f}, {0.49999997f,-0.86602545f}, +{0.30901697f,-0.95105654f}, {0.10452842f,-0.99452192f}, {-0.10452851f,-0.99452192f}, +{-0.30901703f,-0.95105648f}, {-0.50000006f,-0.86602533f}, {-0.66913068f,-0.74314475f}, +{-0.80901700f,-0.58778518f}, {-0.91354549f,-0.40673658f}, {-0.97814763f,-0.20791161f}, +{1.0000000f,-0.0000000f}, {0.95105648f,-0.30901700f}, {0.80901700f,-0.58778524f}, +{0.58778524f,-0.80901700f}, {0.30901697f,-0.95105654f}, {-4.3711388e-08f,-1.0000000f}, +{-0.30901703f,-0.95105648f}, {-0.58778518f,-0.80901700f}, {-0.80901700f,-0.58778518f}, +{-0.95105654f,-0.30901679f}, {-1.0000000f,8.7422777e-08f}, {-0.95105654f,0.30901697f}, +{-0.80901694f,0.58778536f}, {-0.58778507f,0.80901712f}, {-0.30901709f,0.95105648f}, +}; +static const ne10_fft_state_float32_t ne10_fft_state_float32_t_480 = { +120, +(ne10_int32_t *)ne10_factors_480, +(ne10_fft_cpx_float32_t *)ne10_twiddles_480, +NULL, +(ne10_fft_cpx_float32_t *)&ne10_twiddles_480[120], +/* is_forward_scaled = true */ +(ne10_int32_t) 1, +/* is_backward_scaled = false */ +(ne10_int32_t) 0, +}; +static const arch_fft_state cfg_arch_480 = { +1, +(void *)&ne10_fft_state_float32_t_480, +}; + +static const ne10_fft_state_float32_t ne10_fft_state_float32_t_240 = { +60, +(ne10_int32_t *)ne10_factors_240, +(ne10_fft_cpx_float32_t *)ne10_twiddles_240, +NULL, +(ne10_fft_cpx_float32_t *)&ne10_twiddles_240[60], +/* is_forward_scaled = true */ +(ne10_int32_t) 1, +/* is_backward_scaled = false */ +(ne10_int32_t) 0, +}; +static const arch_fft_state cfg_arch_240 = { +1, +(void *)&ne10_fft_state_float32_t_240, +}; + +static const ne10_fft_state_float32_t ne10_fft_state_float32_t_120 = { +30, +(ne10_int32_t *)ne10_factors_120, +(ne10_fft_cpx_float32_t *)ne10_twiddles_120, +NULL, +(ne10_fft_cpx_float32_t *)&ne10_twiddles_120[30], +/* is_forward_scaled = true */ +(ne10_int32_t) 1, +/* is_backward_scaled = false */ +(ne10_int32_t) 0, +}; +static const arch_fft_state cfg_arch_120 = { +1, +(void *)&ne10_fft_state_float32_t_120, +}; + +static const ne10_fft_state_float32_t ne10_fft_state_float32_t_60 = { +15, +(ne10_int32_t *)ne10_factors_60, +(ne10_fft_cpx_float32_t *)ne10_twiddles_60, +NULL, +(ne10_fft_cpx_float32_t *)&ne10_twiddles_60[15], +/* is_forward_scaled = true */ +(ne10_int32_t) 1, +/* is_backward_scaled = false */ +(ne10_int32_t) 0, +}; +static const arch_fft_state cfg_arch_60 = { +1, +(void *)&ne10_fft_state_float32_t_60, +}; + +#endif /* end NE10_FFT_PARAMS48000_960 */ diff --git a/media/libopus/celt/vq.c b/media/libopus/celt/vq.c new file mode 100644 index 0000000000..8011e22548 --- /dev/null +++ b/media/libopus/celt/vq.c @@ -0,0 +1,442 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "mathops.h" +#include "cwrs.h" +#include "vq.h" +#include "arch.h" +#include "os_support.h" +#include "bands.h" +#include "rate.h" +#include "pitch.h" + +#if defined(MIPSr1_ASM) +#include "mips/vq_mipsr1.h" +#endif + +#ifndef OVERRIDE_vq_exp_rotation1 +static void exp_rotation1(celt_norm *X, int len, int stride, opus_val16 c, opus_val16 s) +{ + int i; + opus_val16 ms; + celt_norm *Xptr; + Xptr = X; + ms = NEG16(s); + for (i=0;i<len-stride;i++) + { + celt_norm x1, x2; + x1 = Xptr[0]; + x2 = Xptr[stride]; + Xptr[stride] = EXTRACT16(PSHR32(MAC16_16(MULT16_16(c, x2), s, x1), 15)); + *Xptr++ = EXTRACT16(PSHR32(MAC16_16(MULT16_16(c, x1), ms, x2), 15)); + } + Xptr = &X[len-2*stride-1]; + for (i=len-2*stride-1;i>=0;i--) + { + celt_norm x1, x2; + x1 = Xptr[0]; + x2 = Xptr[stride]; + Xptr[stride] = EXTRACT16(PSHR32(MAC16_16(MULT16_16(c, x2), s, x1), 15)); + *Xptr-- = EXTRACT16(PSHR32(MAC16_16(MULT16_16(c, x1), ms, x2), 15)); + } +} +#endif /* OVERRIDE_vq_exp_rotation1 */ + +void exp_rotation(celt_norm *X, int len, int dir, int stride, int K, int spread) +{ + static const int SPREAD_FACTOR[3]={15,10,5}; + int i; + opus_val16 c, s; + opus_val16 gain, theta; + int stride2=0; + int factor; + + if (2*K>=len || spread==SPREAD_NONE) + return; + factor = SPREAD_FACTOR[spread-1]; + + gain = celt_div((opus_val32)MULT16_16(Q15_ONE,len),(opus_val32)(len+factor*K)); + theta = HALF16(MULT16_16_Q15(gain,gain)); + + c = celt_cos_norm(EXTEND32(theta)); + s = celt_cos_norm(EXTEND32(SUB16(Q15ONE,theta))); /* sin(theta) */ + + if (len>=8*stride) + { + stride2 = 1; + /* This is just a simple (equivalent) way of computing sqrt(len/stride) with rounding. + It's basically incrementing long as (stride2+0.5)^2 < len/stride. */ + while ((stride2*stride2+stride2)*stride + (stride>>2) < len) + stride2++; + } + /*NOTE: As a minor optimization, we could be passing around log2(B), not B, for both this and for + extract_collapse_mask().*/ + len = celt_udiv(len, stride); + for (i=0;i<stride;i++) + { + if (dir < 0) + { + if (stride2) + exp_rotation1(X+i*len, len, stride2, s, c); + exp_rotation1(X+i*len, len, 1, c, s); + } else { + exp_rotation1(X+i*len, len, 1, c, -s); + if (stride2) + exp_rotation1(X+i*len, len, stride2, s, -c); + } + } +} + +/** Takes the pitch vector and the decoded residual vector, computes the gain + that will give ||p+g*y||=1 and mixes the residual with the pitch. */ +static void normalise_residual(int * OPUS_RESTRICT iy, celt_norm * OPUS_RESTRICT X, + int N, opus_val32 Ryy, opus_val16 gain) +{ + int i; +#ifdef FIXED_POINT + int k; +#endif + opus_val32 t; + opus_val16 g; + +#ifdef FIXED_POINT + k = celt_ilog2(Ryy)>>1; +#endif + t = VSHR32(Ryy, 2*(k-7)); + g = MULT16_16_P15(celt_rsqrt_norm(t),gain); + + i=0; + do + X[i] = EXTRACT16(PSHR32(MULT16_16(g, iy[i]), k+1)); + while (++i < N); +} + +static unsigned extract_collapse_mask(int *iy, int N, int B) +{ + unsigned collapse_mask; + int N0; + int i; + if (B<=1) + return 1; + /*NOTE: As a minor optimization, we could be passing around log2(B), not B, for both this and for + exp_rotation().*/ + N0 = celt_udiv(N, B); + collapse_mask = 0; + i=0; do { + int j; + unsigned tmp=0; + j=0; do { + tmp |= iy[i*N0+j]; + } while (++j<N0); + collapse_mask |= (tmp!=0)<<i; + } while (++i<B); + return collapse_mask; +} + +opus_val16 op_pvq_search_c(celt_norm *X, int *iy, int K, int N, int arch) +{ + VARDECL(celt_norm, y); + VARDECL(int, signx); + int i, j; + int pulsesLeft; + opus_val32 sum; + opus_val32 xy; + opus_val16 yy; + SAVE_STACK; + + (void)arch; + ALLOC(y, N, celt_norm); + ALLOC(signx, N, int); + + /* Get rid of the sign */ + sum = 0; + j=0; do { + signx[j] = X[j]<0; + /* OPT: Make sure the compiler doesn't use a branch on ABS16(). */ + X[j] = ABS16(X[j]); + iy[j] = 0; + y[j] = 0; + } while (++j<N); + + xy = yy = 0; + + pulsesLeft = K; + + /* Do a pre-search by projecting on the pyramid */ + if (K > (N>>1)) + { + opus_val16 rcp; + j=0; do { + sum += X[j]; + } while (++j<N); + + /* If X is too small, just replace it with a pulse at 0 */ +#ifdef FIXED_POINT + if (sum <= K) +#else + /* Prevents infinities and NaNs from causing too many pulses + to be allocated. 64 is an approximation of infinity here. */ + if (!(sum > EPSILON && sum < 64)) +#endif + { + X[0] = QCONST16(1.f,14); + j=1; do + X[j]=0; + while (++j<N); + sum = QCONST16(1.f,14); + } +#ifdef FIXED_POINT + rcp = EXTRACT16(MULT16_32_Q16(K, celt_rcp(sum))); +#else + /* Using K+e with e < 1 guarantees we cannot get more than K pulses. */ + rcp = EXTRACT16(MULT16_32_Q16(K+0.8f, celt_rcp(sum))); +#endif + j=0; do { +#ifdef FIXED_POINT + /* It's really important to round *towards zero* here */ + iy[j] = MULT16_16_Q15(X[j],rcp); +#else + iy[j] = (int)floor(rcp*X[j]); +#endif + y[j] = (celt_norm)iy[j]; + yy = MAC16_16(yy, y[j],y[j]); + xy = MAC16_16(xy, X[j],y[j]); + y[j] *= 2; + pulsesLeft -= iy[j]; + } while (++j<N); + } + celt_sig_assert(pulsesLeft>=0); + + /* This should never happen, but just in case it does (e.g. on silence) + we fill the first bin with pulses. */ +#ifdef FIXED_POINT_DEBUG + celt_sig_assert(pulsesLeft<=N+3); +#endif + if (pulsesLeft > N+3) + { + opus_val16 tmp = (opus_val16)pulsesLeft; + yy = MAC16_16(yy, tmp, tmp); + yy = MAC16_16(yy, tmp, y[0]); + iy[0] += pulsesLeft; + pulsesLeft=0; + } + + for (i=0;i<pulsesLeft;i++) + { + opus_val16 Rxy, Ryy; + int best_id; + opus_val32 best_num; + opus_val16 best_den; +#ifdef FIXED_POINT + int rshift; +#endif +#ifdef FIXED_POINT + rshift = 1+celt_ilog2(K-pulsesLeft+i+1); +#endif + best_id = 0; + /* The squared magnitude term gets added anyway, so we might as well + add it outside the loop */ + yy = ADD16(yy, 1); + + /* Calculations for position 0 are out of the loop, in part to reduce + mispredicted branches (since the if condition is usually false) + in the loop. */ + /* Temporary sums of the new pulse(s) */ + Rxy = EXTRACT16(SHR32(ADD32(xy, EXTEND32(X[0])),rshift)); + /* We're multiplying y[j] by two so we don't have to do it here */ + Ryy = ADD16(yy, y[0]); + + /* Approximate score: we maximise Rxy/sqrt(Ryy) (we're guaranteed that + Rxy is positive because the sign is pre-computed) */ + Rxy = MULT16_16_Q15(Rxy,Rxy); + best_den = Ryy; + best_num = Rxy; + j=1; + do { + /* Temporary sums of the new pulse(s) */ + Rxy = EXTRACT16(SHR32(ADD32(xy, EXTEND32(X[j])),rshift)); + /* We're multiplying y[j] by two so we don't have to do it here */ + Ryy = ADD16(yy, y[j]); + + /* Approximate score: we maximise Rxy/sqrt(Ryy) (we're guaranteed that + Rxy is positive because the sign is pre-computed) */ + Rxy = MULT16_16_Q15(Rxy,Rxy); + /* The idea is to check for num/den >= best_num/best_den, but that way + we can do it without any division */ + /* OPT: It's not clear whether a cmov is faster than a branch here + since the condition is more often false than true and using + a cmov introduces data dependencies across iterations. The optimal + choice may be architecture-dependent. */ + if (opus_unlikely(MULT16_16(best_den, Rxy) > MULT16_16(Ryy, best_num))) + { + best_den = Ryy; + best_num = Rxy; + best_id = j; + } + } while (++j<N); + + /* Updating the sums of the new pulse(s) */ + xy = ADD32(xy, EXTEND32(X[best_id])); + /* We're multiplying y[j] by two so we don't have to do it here */ + yy = ADD16(yy, y[best_id]); + + /* Only now that we've made the final choice, update y/iy */ + /* Multiplying y[j] by 2 so we don't have to do it everywhere else */ + y[best_id] += 2; + iy[best_id]++; + } + + /* Put the original sign back */ + j=0; + do { + /*iy[j] = signx[j] ? -iy[j] : iy[j];*/ + /* OPT: The is more likely to be compiled without a branch than the code above + but has the same performance otherwise. */ + iy[j] = (iy[j]^-signx[j]) + signx[j]; + } while (++j<N); + RESTORE_STACK; + return yy; +} + +unsigned alg_quant(celt_norm *X, int N, int K, int spread, int B, ec_enc *enc, + opus_val16 gain, int resynth, int arch) +{ + VARDECL(int, iy); + opus_val16 yy; + unsigned collapse_mask; + SAVE_STACK; + + celt_assert2(K>0, "alg_quant() needs at least one pulse"); + celt_assert2(N>1, "alg_quant() needs at least two dimensions"); + + /* Covers vectorization by up to 4. */ + ALLOC(iy, N+3, int); + + exp_rotation(X, N, 1, B, K, spread); + + yy = op_pvq_search(X, iy, K, N, arch); + + encode_pulses(iy, N, K, enc); + + if (resynth) + { + normalise_residual(iy, X, N, yy, gain); + exp_rotation(X, N, -1, B, K, spread); + } + + collapse_mask = extract_collapse_mask(iy, N, B); + RESTORE_STACK; + return collapse_mask; +} + +/** Decode pulse vector and combine the result with the pitch vector to produce + the final normalised signal in the current band. */ +unsigned alg_unquant(celt_norm *X, int N, int K, int spread, int B, + ec_dec *dec, opus_val16 gain) +{ + opus_val32 Ryy; + unsigned collapse_mask; + VARDECL(int, iy); + SAVE_STACK; + + celt_assert2(K>0, "alg_unquant() needs at least one pulse"); + celt_assert2(N>1, "alg_unquant() needs at least two dimensions"); + ALLOC(iy, N, int); + Ryy = decode_pulses(iy, N, K, dec); + normalise_residual(iy, X, N, Ryy, gain); + exp_rotation(X, N, -1, B, K, spread); + collapse_mask = extract_collapse_mask(iy, N, B); + RESTORE_STACK; + return collapse_mask; +} + +#ifndef OVERRIDE_renormalise_vector +void renormalise_vector(celt_norm *X, int N, opus_val16 gain, int arch) +{ + int i; +#ifdef FIXED_POINT + int k; +#endif + opus_val32 E; + opus_val16 g; + opus_val32 t; + celt_norm *xptr; + E = EPSILON + celt_inner_prod(X, X, N, arch); +#ifdef FIXED_POINT + k = celt_ilog2(E)>>1; +#endif + t = VSHR32(E, 2*(k-7)); + g = MULT16_16_P15(celt_rsqrt_norm(t),gain); + + xptr = X; + for (i=0;i<N;i++) + { + *xptr = EXTRACT16(PSHR32(MULT16_16(g, *xptr), k+1)); + xptr++; + } + /*return celt_sqrt(E);*/ +} +#endif /* OVERRIDE_renormalise_vector */ + +int stereo_itheta(const celt_norm *X, const celt_norm *Y, int stereo, int N, int arch) +{ + int i; + int itheta; + opus_val16 mid, side; + opus_val32 Emid, Eside; + + Emid = Eside = EPSILON; + if (stereo) + { + for (i=0;i<N;i++) + { + celt_norm m, s; + m = ADD16(SHR16(X[i],1),SHR16(Y[i],1)); + s = SUB16(SHR16(X[i],1),SHR16(Y[i],1)); + Emid = MAC16_16(Emid, m, m); + Eside = MAC16_16(Eside, s, s); + } + } else { + Emid += celt_inner_prod(X, X, N, arch); + Eside += celt_inner_prod(Y, Y, N, arch); + } + mid = celt_sqrt(Emid); + side = celt_sqrt(Eside); +#ifdef FIXED_POINT + /* 0.63662 = 2/pi */ + itheta = MULT16_16_Q15(QCONST16(0.63662f,15),celt_atan2p(side, mid)); +#else + itheta = (int)floor(.5f+16384*0.63662f*fast_atan2f(side,mid)); +#endif + + return itheta; +} diff --git a/media/libopus/celt/vq.h b/media/libopus/celt/vq.h new file mode 100644 index 0000000000..45ec55918e --- /dev/null +++ b/media/libopus/celt/vq.h @@ -0,0 +1,79 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/** + @file vq.h + @brief Vector quantisation of the residual + */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef VQ_H +#define VQ_H + +#include "entenc.h" +#include "entdec.h" +#include "modes.h" + +#if (defined(OPUS_X86_MAY_HAVE_SSE2) && !defined(FIXED_POINT)) +#include "x86/vq_sse.h" +#endif + +void exp_rotation(celt_norm *X, int len, int dir, int stride, int K, int spread); + +opus_val16 op_pvq_search_c(celt_norm *X, int *iy, int K, int N, int arch); + +#if !defined(OVERRIDE_OP_PVQ_SEARCH) +#define op_pvq_search(x, iy, K, N, arch) \ + (op_pvq_search_c(x, iy, K, N, arch)) +#endif + +/** Algebraic pulse-vector quantiser. The signal x is replaced by the sum of + * the pitch and a combination of pulses such that its norm is still equal + * to 1. This is the function that will typically require the most CPU. + * @param X Residual signal to quantise/encode (returns quantised version) + * @param N Number of samples to encode + * @param K Number of pulses to use + * @param enc Entropy encoder state + * @ret A mask indicating which blocks in the band received pulses +*/ +unsigned alg_quant(celt_norm *X, int N, int K, int spread, int B, ec_enc *enc, + opus_val16 gain, int resynth, int arch); + +/** Algebraic pulse decoder + * @param X Decoded normalised spectrum (returned) + * @param N Number of samples to decode + * @param K Number of pulses to use + * @param dec Entropy decoder state + * @ret A mask indicating which blocks in the band received pulses + */ +unsigned alg_unquant(celt_norm *X, int N, int K, int spread, int B, + ec_dec *dec, opus_val16 gain); + +void renormalise_vector(celt_norm *X, int N, opus_val16 gain, int arch); + +int stereo_itheta(const celt_norm *X, const celt_norm *Y, int stereo, int N, int arch); + +#endif /* VQ_H */ diff --git a/media/libopus/celt/x86/celt_lpc_sse.h b/media/libopus/celt/x86/celt_lpc_sse.h new file mode 100644 index 0000000000..90e69ecf63 --- /dev/null +++ b/media/libopus/celt/x86/celt_lpc_sse.h @@ -0,0 +1,67 @@ +/* Copyright (c) 2014, Cisco Systems, INC + Written by XiangMingZhu WeiZhou MinPeng YanWang + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef CELT_LPC_SSE_H +#define CELT_LPC_SSE_H + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#if defined(OPUS_X86_MAY_HAVE_SSE4_1) && defined(FIXED_POINT) + +void celt_fir_sse4_1( + const opus_val16 *x, + const opus_val16 *num, + opus_val16 *y, + int N, + int ord, + int arch); + +#if defined(OPUS_X86_PRESUME_SSE4_1) +#define OVERRIDE_CELT_FIR +#define celt_fir(x, num, y, N, ord, arch) \ + ((void)arch, celt_fir_sse4_1(x, num, y, N, ord, arch)) + +#elif defined(OPUS_HAVE_RTCD) + +extern void (*const CELT_FIR_IMPL[OPUS_ARCHMASK + 1])( + const opus_val16 *x, + const opus_val16 *num, + opus_val16 *y, + int N, + int ord, + int arch); + +#define OVERRIDE_CELT_FIR +# define celt_fir(x, num, y, N, ord, arch) \ + ((*CELT_FIR_IMPL[(arch) & OPUS_ARCHMASK])(x, num, y, N, ord, arch)) + +#endif +#endif + +#endif diff --git a/media/libopus/celt/x86/celt_lpc_sse4_1.c b/media/libopus/celt/x86/celt_lpc_sse4_1.c new file mode 100644 index 0000000000..5478568849 --- /dev/null +++ b/media/libopus/celt/x86/celt_lpc_sse4_1.c @@ -0,0 +1,89 @@ +/* Copyright (c) 2014, Cisco Systems, INC + Written by XiangMingZhu WeiZhou MinPeng YanWang + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include <xmmintrin.h> +#include <emmintrin.h> +#include <smmintrin.h> +#include "celt_lpc.h" +#include "stack_alloc.h" +#include "mathops.h" +#include "pitch.h" +#include "x86cpu.h" + +#if defined(FIXED_POINT) + +void celt_fir_sse4_1(const opus_val16 *x, + const opus_val16 *num, + opus_val16 *y, + int N, + int ord, + int arch) +{ + int i,j; + VARDECL(opus_val16, rnum); + + __m128i vecNoA; + opus_int32 noA ; + SAVE_STACK; + + ALLOC(rnum, ord, opus_val16); + for(i=0;i<ord;i++) + rnum[i] = num[ord-i-1]; + noA = EXTEND32(1) << SIG_SHIFT >> 1; + vecNoA = _mm_set_epi32(noA, noA, noA, noA); + + for (i=0;i<N-3;i+=4) + { + opus_val32 sums[4] = {0}; + __m128i vecSum, vecX; + + xcorr_kernel(rnum, x+i-ord, sums, ord, arch); + + vecSum = _mm_loadu_si128((__m128i *)sums); + vecSum = _mm_add_epi32(vecSum, vecNoA); + vecSum = _mm_srai_epi32(vecSum, SIG_SHIFT); + vecX = OP_CVTEPI16_EPI32_M64(x + i); + vecSum = _mm_add_epi32(vecSum, vecX); + vecSum = _mm_packs_epi32(vecSum, vecSum); + _mm_storel_epi64((__m128i *)(y + i), vecSum); + } + for (;i<N;i++) + { + opus_val32 sum = 0; + for (j=0;j<ord;j++) + sum = MAC16_16(sum, rnum[j], x[i+j-ord]); + y[i] = SATURATE16(ADD32(EXTEND32(x[i]), PSHR32(sum, SIG_SHIFT))); + } + + RESTORE_STACK; +} + +#endif diff --git a/media/libopus/celt/x86/pitch_sse.c b/media/libopus/celt/x86/pitch_sse.c new file mode 100644 index 0000000000..20e73126b6 --- /dev/null +++ b/media/libopus/celt/x86/pitch_sse.c @@ -0,0 +1,185 @@ +/* Copyright (c) 2014, Cisco Systems, INC + Written by XiangMingZhu WeiZhou MinPeng YanWang + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "macros.h" +#include "celt_lpc.h" +#include "stack_alloc.h" +#include "mathops.h" +#include "pitch.h" + +#if defined(OPUS_X86_MAY_HAVE_SSE) && !defined(FIXED_POINT) + +#include <xmmintrin.h> +#include "arch.h" + +void xcorr_kernel_sse(const opus_val16 *x, const opus_val16 *y, opus_val32 sum[4], int len) +{ + int j; + __m128 xsum1, xsum2; + xsum1 = _mm_loadu_ps(sum); + xsum2 = _mm_setzero_ps(); + + for (j = 0; j < len-3; j += 4) + { + __m128 x0 = _mm_loadu_ps(x+j); + __m128 yj = _mm_loadu_ps(y+j); + __m128 y3 = _mm_loadu_ps(y+j+3); + + xsum1 = _mm_add_ps(xsum1,_mm_mul_ps(_mm_shuffle_ps(x0,x0,0x00),yj)); + xsum2 = _mm_add_ps(xsum2,_mm_mul_ps(_mm_shuffle_ps(x0,x0,0x55), + _mm_shuffle_ps(yj,y3,0x49))); + xsum1 = _mm_add_ps(xsum1,_mm_mul_ps(_mm_shuffle_ps(x0,x0,0xaa), + _mm_shuffle_ps(yj,y3,0x9e))); + xsum2 = _mm_add_ps(xsum2,_mm_mul_ps(_mm_shuffle_ps(x0,x0,0xff),y3)); + } + if (j < len) + { + xsum1 = _mm_add_ps(xsum1,_mm_mul_ps(_mm_load1_ps(x+j),_mm_loadu_ps(y+j))); + if (++j < len) + { + xsum2 = _mm_add_ps(xsum2,_mm_mul_ps(_mm_load1_ps(x+j),_mm_loadu_ps(y+j))); + if (++j < len) + { + xsum1 = _mm_add_ps(xsum1,_mm_mul_ps(_mm_load1_ps(x+j),_mm_loadu_ps(y+j))); + } + } + } + _mm_storeu_ps(sum,_mm_add_ps(xsum1,xsum2)); +} + + +void dual_inner_prod_sse(const opus_val16 *x, const opus_val16 *y01, const opus_val16 *y02, + int N, opus_val32 *xy1, opus_val32 *xy2) +{ + int i; + __m128 xsum1, xsum2; + xsum1 = _mm_setzero_ps(); + xsum2 = _mm_setzero_ps(); + for (i=0;i<N-3;i+=4) + { + __m128 xi = _mm_loadu_ps(x+i); + __m128 y1i = _mm_loadu_ps(y01+i); + __m128 y2i = _mm_loadu_ps(y02+i); + xsum1 = _mm_add_ps(xsum1,_mm_mul_ps(xi, y1i)); + xsum2 = _mm_add_ps(xsum2,_mm_mul_ps(xi, y2i)); + } + /* Horizontal sum */ + xsum1 = _mm_add_ps(xsum1, _mm_movehl_ps(xsum1, xsum1)); + xsum1 = _mm_add_ss(xsum1, _mm_shuffle_ps(xsum1, xsum1, 0x55)); + _mm_store_ss(xy1, xsum1); + xsum2 = _mm_add_ps(xsum2, _mm_movehl_ps(xsum2, xsum2)); + xsum2 = _mm_add_ss(xsum2, _mm_shuffle_ps(xsum2, xsum2, 0x55)); + _mm_store_ss(xy2, xsum2); + for (;i<N;i++) + { + *xy1 = MAC16_16(*xy1, x[i], y01[i]); + *xy2 = MAC16_16(*xy2, x[i], y02[i]); + } +} + +opus_val32 celt_inner_prod_sse(const opus_val16 *x, const opus_val16 *y, + int N) +{ + int i; + float xy; + __m128 sum; + sum = _mm_setzero_ps(); + /* FIXME: We should probably go 8-way and use 2 sums. */ + for (i=0;i<N-3;i+=4) + { + __m128 xi = _mm_loadu_ps(x+i); + __m128 yi = _mm_loadu_ps(y+i); + sum = _mm_add_ps(sum,_mm_mul_ps(xi, yi)); + } + /* Horizontal sum */ + sum = _mm_add_ps(sum, _mm_movehl_ps(sum, sum)); + sum = _mm_add_ss(sum, _mm_shuffle_ps(sum, sum, 0x55)); + _mm_store_ss(&xy, sum); + for (;i<N;i++) + { + xy = MAC16_16(xy, x[i], y[i]); + } + return xy; +} + +void comb_filter_const_sse(opus_val32 *y, opus_val32 *x, int T, int N, + opus_val16 g10, opus_val16 g11, opus_val16 g12) +{ + int i; + __m128 x0v; + __m128 g10v, g11v, g12v; + g10v = _mm_load1_ps(&g10); + g11v = _mm_load1_ps(&g11); + g12v = _mm_load1_ps(&g12); + x0v = _mm_loadu_ps(&x[-T-2]); + for (i=0;i<N-3;i+=4) + { + __m128 yi, yi2, x1v, x2v, x3v, x4v; + const opus_val32 *xp = &x[i-T-2]; + yi = _mm_loadu_ps(x+i); + x4v = _mm_loadu_ps(xp+4); +#if 0 + /* Slower version with all loads */ + x1v = _mm_loadu_ps(xp+1); + x2v = _mm_loadu_ps(xp+2); + x3v = _mm_loadu_ps(xp+3); +#else + x2v = _mm_shuffle_ps(x0v, x4v, 0x4e); + x1v = _mm_shuffle_ps(x0v, x2v, 0x99); + x3v = _mm_shuffle_ps(x2v, x4v, 0x99); +#endif + + yi = _mm_add_ps(yi, _mm_mul_ps(g10v,x2v)); +#if 0 /* Set to 1 to make it bit-exact with the non-SSE version */ + yi = _mm_add_ps(yi, _mm_mul_ps(g11v,_mm_add_ps(x3v,x1v))); + yi = _mm_add_ps(yi, _mm_mul_ps(g12v,_mm_add_ps(x4v,x0v))); +#else + /* Use partial sums */ + yi2 = _mm_add_ps(_mm_mul_ps(g11v,_mm_add_ps(x3v,x1v)), + _mm_mul_ps(g12v,_mm_add_ps(x4v,x0v))); + yi = _mm_add_ps(yi, yi2); +#endif + x0v=x4v; + _mm_storeu_ps(y+i, yi); + } +#ifdef CUSTOM_MODES + for (;i<N;i++) + { + y[i] = x[i] + + MULT16_32_Q15(g10,x[i-T]) + + MULT16_32_Q15(g11,ADD32(x[i-T+1],x[i-T-1])) + + MULT16_32_Q15(g12,ADD32(x[i-T+2],x[i-T-2])); + } +#endif +} + + +#endif diff --git a/media/libopus/celt/x86/pitch_sse.h b/media/libopus/celt/x86/pitch_sse.h new file mode 100644 index 0000000000..964aef50db --- /dev/null +++ b/media/libopus/celt/x86/pitch_sse.h @@ -0,0 +1,192 @@ +/* Copyright (c) 2013 Jean-Marc Valin and John Ridges + Copyright (c) 2014, Cisco Systems, INC MingXiang WeiZhou MinPeng YanWang*/ +/** + @file pitch_sse.h + @brief Pitch analysis + */ + +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef PITCH_SSE_H +#define PITCH_SSE_H + +#if defined(HAVE_CONFIG_H) +#include "config.h" +#endif + +#if defined(OPUS_X86_MAY_HAVE_SSE4_1) && defined(FIXED_POINT) +void xcorr_kernel_sse4_1( + const opus_int16 *x, + const opus_int16 *y, + opus_val32 sum[4], + int len); +#endif + +#if defined(OPUS_X86_MAY_HAVE_SSE) && !defined(FIXED_POINT) +void xcorr_kernel_sse( + const opus_val16 *x, + const opus_val16 *y, + opus_val32 sum[4], + int len); +#endif + +#if defined(OPUS_X86_PRESUME_SSE4_1) && defined(FIXED_POINT) +#define OVERRIDE_XCORR_KERNEL +#define xcorr_kernel(x, y, sum, len, arch) \ + ((void)arch, xcorr_kernel_sse4_1(x, y, sum, len)) + +#elif defined(OPUS_X86_PRESUME_SSE) && !defined(FIXED_POINT) +#define OVERRIDE_XCORR_KERNEL +#define xcorr_kernel(x, y, sum, len, arch) \ + ((void)arch, xcorr_kernel_sse(x, y, sum, len)) + +#elif defined(OPUS_HAVE_RTCD) && ((defined(OPUS_X86_MAY_HAVE_SSE4_1) && defined(FIXED_POINT)) || (defined(OPUS_X86_MAY_HAVE_SSE) && !defined(FIXED_POINT))) + +extern void (*const XCORR_KERNEL_IMPL[OPUS_ARCHMASK + 1])( + const opus_val16 *x, + const opus_val16 *y, + opus_val32 sum[4], + int len); + +#define OVERRIDE_XCORR_KERNEL +#define xcorr_kernel(x, y, sum, len, arch) \ + ((*XCORR_KERNEL_IMPL[(arch) & OPUS_ARCHMASK])(x, y, sum, len)) + +#endif + +#if defined(OPUS_X86_MAY_HAVE_SSE4_1) && defined(FIXED_POINT) +opus_val32 celt_inner_prod_sse4_1( + const opus_int16 *x, + const opus_int16 *y, + int N); +#endif + +#if defined(OPUS_X86_MAY_HAVE_SSE2) && defined(FIXED_POINT) +opus_val32 celt_inner_prod_sse2( + const opus_int16 *x, + const opus_int16 *y, + int N); +#endif + +#if defined(OPUS_X86_MAY_HAVE_SSE) && !defined(FIXED_POINT) +opus_val32 celt_inner_prod_sse( + const opus_val16 *x, + const opus_val16 *y, + int N); +#endif + + +#if defined(OPUS_X86_PRESUME_SSE4_1) && defined(FIXED_POINT) +#define OVERRIDE_CELT_INNER_PROD +#define celt_inner_prod(x, y, N, arch) \ + ((void)arch, celt_inner_prod_sse4_1(x, y, N)) + +#elif defined(OPUS_X86_PRESUME_SSE2) && defined(FIXED_POINT) && !defined(OPUS_X86_MAY_HAVE_SSE4_1) +#define OVERRIDE_CELT_INNER_PROD +#define celt_inner_prod(x, y, N, arch) \ + ((void)arch, celt_inner_prod_sse2(x, y, N)) + +#elif defined(OPUS_X86_PRESUME_SSE) && !defined(FIXED_POINT) +#define OVERRIDE_CELT_INNER_PROD +#define celt_inner_prod(x, y, N, arch) \ + ((void)arch, celt_inner_prod_sse(x, y, N)) + + +#elif defined(OPUS_HAVE_RTCD) && (((defined(OPUS_X86_MAY_HAVE_SSE4_1) || defined(OPUS_X86_MAY_HAVE_SSE2)) && defined(FIXED_POINT)) || \ + (defined(OPUS_X86_MAY_HAVE_SSE) && !defined(FIXED_POINT))) + +extern opus_val32 (*const CELT_INNER_PROD_IMPL[OPUS_ARCHMASK + 1])( + const opus_val16 *x, + const opus_val16 *y, + int N); + +#define OVERRIDE_CELT_INNER_PROD +#define celt_inner_prod(x, y, N, arch) \ + ((*CELT_INNER_PROD_IMPL[(arch) & OPUS_ARCHMASK])(x, y, N)) + +#endif + +#if defined(OPUS_X86_MAY_HAVE_SSE) && !defined(FIXED_POINT) + +#define OVERRIDE_DUAL_INNER_PROD +#define OVERRIDE_COMB_FILTER_CONST + +#undef dual_inner_prod +#undef comb_filter_const + +void dual_inner_prod_sse(const opus_val16 *x, + const opus_val16 *y01, + const opus_val16 *y02, + int N, + opus_val32 *xy1, + opus_val32 *xy2); + +void comb_filter_const_sse(opus_val32 *y, + opus_val32 *x, + int T, + int N, + opus_val16 g10, + opus_val16 g11, + opus_val16 g12); + + +#if defined(OPUS_X86_PRESUME_SSE) +# define dual_inner_prod(x, y01, y02, N, xy1, xy2, arch) \ + ((void)(arch),dual_inner_prod_sse(x, y01, y02, N, xy1, xy2)) + +# define comb_filter_const(y, x, T, N, g10, g11, g12, arch) \ + ((void)(arch),comb_filter_const_sse(y, x, T, N, g10, g11, g12)) +#else + +extern void (*const DUAL_INNER_PROD_IMPL[OPUS_ARCHMASK + 1])( + const opus_val16 *x, + const opus_val16 *y01, + const opus_val16 *y02, + int N, + opus_val32 *xy1, + opus_val32 *xy2); + +#define dual_inner_prod(x, y01, y02, N, xy1, xy2, arch) \ + ((*DUAL_INNER_PROD_IMPL[(arch) & OPUS_ARCHMASK])(x, y01, y02, N, xy1, xy2)) + +extern void (*const COMB_FILTER_CONST_IMPL[OPUS_ARCHMASK + 1])( + opus_val32 *y, + opus_val32 *x, + int T, + int N, + opus_val16 g10, + opus_val16 g11, + opus_val16 g12); + +#define comb_filter_const(y, x, T, N, g10, g11, g12, arch) \ + ((*COMB_FILTER_CONST_IMPL[(arch) & OPUS_ARCHMASK])(y, x, T, N, g10, g11, g12)) + +#define NON_STATIC_COMB_FILTER_CONST_C + +#endif +#endif + +#endif diff --git a/media/libopus/celt/x86/pitch_sse2.c b/media/libopus/celt/x86/pitch_sse2.c new file mode 100644 index 0000000000..a0e7d1beaf --- /dev/null +++ b/media/libopus/celt/x86/pitch_sse2.c @@ -0,0 +1,95 @@ +/* Copyright (c) 2014, Cisco Systems, INC + Written by XiangMingZhu WeiZhou MinPeng YanWang + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include <xmmintrin.h> +#include <emmintrin.h> + +#include "macros.h" +#include "celt_lpc.h" +#include "stack_alloc.h" +#include "mathops.h" +#include "pitch.h" + +#if defined(OPUS_X86_MAY_HAVE_SSE2) && defined(FIXED_POINT) +opus_val32 celt_inner_prod_sse2(const opus_val16 *x, const opus_val16 *y, + int N) +{ + opus_int i, dataSize16; + opus_int32 sum; + + __m128i inVec1_76543210, inVec1_FEDCBA98, acc1; + __m128i inVec2_76543210, inVec2_FEDCBA98, acc2; + + sum = 0; + dataSize16 = N & ~15; + + acc1 = _mm_setzero_si128(); + acc2 = _mm_setzero_si128(); + + for (i=0;i<dataSize16;i+=16) + { + inVec1_76543210 = _mm_loadu_si128((__m128i *)(&x[i + 0])); + inVec2_76543210 = _mm_loadu_si128((__m128i *)(&y[i + 0])); + + inVec1_FEDCBA98 = _mm_loadu_si128((__m128i *)(&x[i + 8])); + inVec2_FEDCBA98 = _mm_loadu_si128((__m128i *)(&y[i + 8])); + + inVec1_76543210 = _mm_madd_epi16(inVec1_76543210, inVec2_76543210); + inVec1_FEDCBA98 = _mm_madd_epi16(inVec1_FEDCBA98, inVec2_FEDCBA98); + + acc1 = _mm_add_epi32(acc1, inVec1_76543210); + acc2 = _mm_add_epi32(acc2, inVec1_FEDCBA98); + } + + acc1 = _mm_add_epi32( acc1, acc2 ); + + if (N - i >= 8) + { + inVec1_76543210 = _mm_loadu_si128((__m128i *)(&x[i + 0])); + inVec2_76543210 = _mm_loadu_si128((__m128i *)(&y[i + 0])); + + inVec1_76543210 = _mm_madd_epi16(inVec1_76543210, inVec2_76543210); + + acc1 = _mm_add_epi32(acc1, inVec1_76543210); + i += 8; + } + + acc1 = _mm_add_epi32(acc1, _mm_unpackhi_epi64( acc1, acc1)); + acc1 = _mm_add_epi32(acc1, _mm_shufflelo_epi16( acc1, 0x0E)); + sum += _mm_cvtsi128_si32(acc1); + + for (;i<N;i++) { + sum = silk_SMLABB(sum, x[i], y[i]); + } + + return sum; +} +#endif diff --git a/media/libopus/celt/x86/pitch_sse4_1.c b/media/libopus/celt/x86/pitch_sse4_1.c new file mode 100644 index 0000000000..2bc578304f --- /dev/null +++ b/media/libopus/celt/x86/pitch_sse4_1.c @@ -0,0 +1,240 @@ +/* Copyright (c) 2014, Cisco Systems, INC + Written by XiangMingZhu WeiZhou MinPeng YanWang + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include <xmmintrin.h> +#include <emmintrin.h> + +#include "macros.h" +#include "celt_lpc.h" +#include "stack_alloc.h" +#include "mathops.h" +#include "pitch.h" + +#if defined(OPUS_X86_MAY_HAVE_SSE4_1) && defined(FIXED_POINT) +#include <smmintrin.h> +#include "x86cpu.h" + +opus_val32 celt_inner_prod_sse4_1(const opus_val16 *x, const opus_val16 *y, + int N) +{ + opus_int i, dataSize16; + opus_int32 sum; + __m128i inVec1_76543210, inVec1_FEDCBA98, acc1; + __m128i inVec2_76543210, inVec2_FEDCBA98, acc2; + __m128i inVec1_3210, inVec2_3210; + + sum = 0; + dataSize16 = N & ~15; + + acc1 = _mm_setzero_si128(); + acc2 = _mm_setzero_si128(); + + for (i=0;i<dataSize16;i+=16) { + inVec1_76543210 = _mm_loadu_si128((__m128i *)(&x[i + 0])); + inVec2_76543210 = _mm_loadu_si128((__m128i *)(&y[i + 0])); + + inVec1_FEDCBA98 = _mm_loadu_si128((__m128i *)(&x[i + 8])); + inVec2_FEDCBA98 = _mm_loadu_si128((__m128i *)(&y[i + 8])); + + inVec1_76543210 = _mm_madd_epi16(inVec1_76543210, inVec2_76543210); + inVec1_FEDCBA98 = _mm_madd_epi16(inVec1_FEDCBA98, inVec2_FEDCBA98); + + acc1 = _mm_add_epi32(acc1, inVec1_76543210); + acc2 = _mm_add_epi32(acc2, inVec1_FEDCBA98); + } + + acc1 = _mm_add_epi32(acc1, acc2); + + if (N - i >= 8) + { + inVec1_76543210 = _mm_loadu_si128((__m128i *)(&x[i + 0])); + inVec2_76543210 = _mm_loadu_si128((__m128i *)(&y[i + 0])); + + inVec1_76543210 = _mm_madd_epi16(inVec1_76543210, inVec2_76543210); + + acc1 = _mm_add_epi32(acc1, inVec1_76543210); + i += 8; + } + + if (N - i >= 4) + { + inVec1_3210 = OP_CVTEPI16_EPI32_M64(&x[i + 0]); + inVec2_3210 = OP_CVTEPI16_EPI32_M64(&y[i + 0]); + + inVec1_3210 = _mm_mullo_epi32(inVec1_3210, inVec2_3210); + + acc1 = _mm_add_epi32(acc1, inVec1_3210); + i += 4; + } + + acc1 = _mm_add_epi32(acc1, _mm_unpackhi_epi64(acc1, acc1)); + acc1 = _mm_add_epi32(acc1, _mm_shufflelo_epi16(acc1, 0x0E)); + + sum += _mm_cvtsi128_si32(acc1); + + for (;i<N;i++) + { + sum = silk_SMLABB(sum, x[i], y[i]); + } + + return sum; +} + +void xcorr_kernel_sse4_1(const opus_val16 * x, const opus_val16 * y, opus_val32 sum[ 4 ], int len) +{ + int j; + + __m128i vecX, vecX0, vecX1, vecX2, vecX3; + __m128i vecY0, vecY1, vecY2, vecY3; + __m128i sum0, sum1, sum2, sum3, vecSum; + __m128i initSum; + +#ifdef OPUS_CHECK_ASM + opus_val32 sum_c[4]; + for (j=0;j<4;j++) { + sum_c[j] = sum[j]; + } + xcorr_kernel_c(x, y, sum_c, len); +#endif + + celt_assert(len >= 3); + + sum0 = _mm_setzero_si128(); + sum1 = _mm_setzero_si128(); + sum2 = _mm_setzero_si128(); + sum3 = _mm_setzero_si128(); + + for (j=0;j<(len-7);j+=8) + { + vecX = _mm_loadu_si128((__m128i *)(&x[j + 0])); + vecY0 = _mm_loadu_si128((__m128i *)(&y[j + 0])); + vecY1 = _mm_loadu_si128((__m128i *)(&y[j + 1])); + vecY2 = _mm_loadu_si128((__m128i *)(&y[j + 2])); + vecY3 = _mm_loadu_si128((__m128i *)(&y[j + 3])); + + sum0 = _mm_add_epi32(sum0, _mm_madd_epi16(vecX, vecY0)); + sum1 = _mm_add_epi32(sum1, _mm_madd_epi16(vecX, vecY1)); + sum2 = _mm_add_epi32(sum2, _mm_madd_epi16(vecX, vecY2)); + sum3 = _mm_add_epi32(sum3, _mm_madd_epi16(vecX, vecY3)); + } + + sum0 = _mm_add_epi32(sum0, _mm_unpackhi_epi64( sum0, sum0)); + sum0 = _mm_add_epi32(sum0, _mm_shufflelo_epi16( sum0, 0x0E)); + + sum1 = _mm_add_epi32(sum1, _mm_unpackhi_epi64( sum1, sum1)); + sum1 = _mm_add_epi32(sum1, _mm_shufflelo_epi16( sum1, 0x0E)); + + sum2 = _mm_add_epi32(sum2, _mm_unpackhi_epi64( sum2, sum2)); + sum2 = _mm_add_epi32(sum2, _mm_shufflelo_epi16( sum2, 0x0E)); + + sum3 = _mm_add_epi32(sum3, _mm_unpackhi_epi64( sum3, sum3)); + sum3 = _mm_add_epi32(sum3, _mm_shufflelo_epi16( sum3, 0x0E)); + + vecSum = _mm_unpacklo_epi64(_mm_unpacklo_epi32(sum0, sum1), + _mm_unpacklo_epi32(sum2, sum3)); + + for (;j<(len-3);j+=4) + { + vecX = OP_CVTEPI16_EPI32_M64(&x[j + 0]); + vecX0 = _mm_shuffle_epi32(vecX, 0x00); + vecX1 = _mm_shuffle_epi32(vecX, 0x55); + vecX2 = _mm_shuffle_epi32(vecX, 0xaa); + vecX3 = _mm_shuffle_epi32(vecX, 0xff); + + vecY0 = OP_CVTEPI16_EPI32_M64(&y[j + 0]); + vecY1 = OP_CVTEPI16_EPI32_M64(&y[j + 1]); + vecY2 = OP_CVTEPI16_EPI32_M64(&y[j + 2]); + vecY3 = OP_CVTEPI16_EPI32_M64(&y[j + 3]); + + sum0 = _mm_mullo_epi32(vecX0, vecY0); + sum1 = _mm_mullo_epi32(vecX1, vecY1); + sum2 = _mm_mullo_epi32(vecX2, vecY2); + sum3 = _mm_mullo_epi32(vecX3, vecY3); + + sum0 = _mm_add_epi32(sum0, sum1); + sum2 = _mm_add_epi32(sum2, sum3); + vecSum = _mm_add_epi32(vecSum, sum0); + vecSum = _mm_add_epi32(vecSum, sum2); + } + + vecX = OP_CVTEPI16_EPI32_M64(&x[len - 4]); + if (len - j == 3) + { + vecX0 = _mm_shuffle_epi32(vecX, 0x55); + vecX1 = _mm_shuffle_epi32(vecX, 0xaa); + vecX2 = _mm_shuffle_epi32(vecX, 0xff); + + vecY0 = OP_CVTEPI16_EPI32_M64(&y[j + 0]); + vecY1 = OP_CVTEPI16_EPI32_M64(&y[j + 1]); + vecY2 = OP_CVTEPI16_EPI32_M64(&y[j + 2]); + + sum0 = _mm_mullo_epi32(vecX0, vecY0); + sum1 = _mm_mullo_epi32(vecX1, vecY1); + sum2 = _mm_mullo_epi32(vecX2, vecY2); + + vecSum = _mm_add_epi32(vecSum, sum0); + vecSum = _mm_add_epi32(vecSum, sum1); + vecSum = _mm_add_epi32(vecSum, sum2); + } + else if (len - j == 2) + { + vecX0 = _mm_shuffle_epi32(vecX, 0xaa); + vecX1 = _mm_shuffle_epi32(vecX, 0xff); + + vecY0 = OP_CVTEPI16_EPI32_M64(&y[j + 0]); + vecY1 = OP_CVTEPI16_EPI32_M64(&y[j + 1]); + + sum0 = _mm_mullo_epi32(vecX0, vecY0); + sum1 = _mm_mullo_epi32(vecX1, vecY1); + + vecSum = _mm_add_epi32(vecSum, sum0); + vecSum = _mm_add_epi32(vecSum, sum1); + } + else if (len - j == 1) + { + vecX0 = _mm_shuffle_epi32(vecX, 0xff); + + vecY0 = OP_CVTEPI16_EPI32_M64(&y[j + 0]); + + sum0 = _mm_mullo_epi32(vecX0, vecY0); + + vecSum = _mm_add_epi32(vecSum, sum0); + } + + initSum = _mm_loadu_si128((__m128i *)(&sum[0])); + initSum = _mm_add_epi32(initSum, vecSum); + _mm_storeu_si128((__m128i *)sum, initSum); + +#ifdef OPUS_CHECK_ASM + celt_assert(!memcmp(sum_c, sum, sizeof(sum_c))); +#endif +} +#endif diff --git a/media/libopus/celt/x86/vq_sse.h b/media/libopus/celt/x86/vq_sse.h new file mode 100644 index 0000000000..b4efe8f249 --- /dev/null +++ b/media/libopus/celt/x86/vq_sse.h @@ -0,0 +1,50 @@ +/* Copyright (c) 2016 Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef VQ_SSE_H +#define VQ_SSE_H + +#if defined(OPUS_X86_MAY_HAVE_SSE2) && !defined(FIXED_POINT) +#define OVERRIDE_OP_PVQ_SEARCH + +opus_val16 op_pvq_search_sse2(celt_norm *_X, int *iy, int K, int N, int arch); + +#if defined(OPUS_X86_PRESUME_SSE2) +#define op_pvq_search(x, iy, K, N, arch) \ + (op_pvq_search_sse2(x, iy, K, N, arch)) + +#else + +extern opus_val16 (*const OP_PVQ_SEARCH_IMPL[OPUS_ARCHMASK + 1])( + celt_norm *_X, int *iy, int K, int N, int arch); + +# define op_pvq_search(X, iy, K, N, arch) \ + ((*OP_PVQ_SEARCH_IMPL[(arch) & OPUS_ARCHMASK])(X, iy, K, N, arch)) + +#endif +#endif + +#endif diff --git a/media/libopus/celt/x86/vq_sse2.c b/media/libopus/celt/x86/vq_sse2.c new file mode 100644 index 0000000000..775042860d --- /dev/null +++ b/media/libopus/celt/x86/vq_sse2.c @@ -0,0 +1,217 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Copyright (c) 2007-2016 Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include <xmmintrin.h> +#include <emmintrin.h> +#include "celt_lpc.h" +#include "stack_alloc.h" +#include "mathops.h" +#include "vq.h" +#include "x86cpu.h" + + +#ifndef FIXED_POINT + +opus_val16 op_pvq_search_sse2(celt_norm *_X, int *iy, int K, int N, int arch) +{ + int i, j; + int pulsesLeft; + float xy, yy; + VARDECL(celt_norm, y); + VARDECL(celt_norm, X); + VARDECL(float, signy); + __m128 signmask; + __m128 sums; + __m128i fours; + SAVE_STACK; + + (void)arch; + /* All bits set to zero, except for the sign bit. */ + signmask = _mm_set_ps1(-0.f); + fours = _mm_set_epi32(4, 4, 4, 4); + ALLOC(y, N+3, celt_norm); + ALLOC(X, N+3, celt_norm); + ALLOC(signy, N+3, float); + + OPUS_COPY(X, _X, N); + X[N] = X[N+1] = X[N+2] = 0; + sums = _mm_setzero_ps(); + for (j=0;j<N;j+=4) + { + __m128 x4, s4; + x4 = _mm_loadu_ps(&X[j]); + s4 = _mm_cmplt_ps(x4, _mm_setzero_ps()); + /* Get rid of the sign */ + x4 = _mm_andnot_ps(signmask, x4); + sums = _mm_add_ps(sums, x4); + /* Clear y and iy in case we don't do the projection. */ + _mm_storeu_ps(&y[j], _mm_setzero_ps()); + _mm_storeu_si128((__m128i*)&iy[j], _mm_setzero_si128()); + _mm_storeu_ps(&X[j], x4); + _mm_storeu_ps(&signy[j], s4); + } + sums = _mm_add_ps(sums, _mm_shuffle_ps(sums, sums, _MM_SHUFFLE(1, 0, 3, 2))); + sums = _mm_add_ps(sums, _mm_shuffle_ps(sums, sums, _MM_SHUFFLE(2, 3, 0, 1))); + + xy = yy = 0; + + pulsesLeft = K; + + /* Do a pre-search by projecting on the pyramid */ + if (K > (N>>1)) + { + __m128i pulses_sum; + __m128 yy4, xy4; + __m128 rcp4; + opus_val32 sum = _mm_cvtss_f32(sums); + /* If X is too small, just replace it with a pulse at 0 */ + /* Prevents infinities and NaNs from causing too many pulses + to be allocated. 64 is an approximation of infinity here. */ + if (!(sum > EPSILON && sum < 64)) + { + X[0] = QCONST16(1.f,14); + j=1; do + X[j]=0; + while (++j<N); + sums = _mm_set_ps1(1.f); + } + /* Using K+e with e < 1 guarantees we cannot get more than K pulses. */ + rcp4 = _mm_mul_ps(_mm_set_ps1((float)(K+.8)), _mm_rcp_ps(sums)); + xy4 = yy4 = _mm_setzero_ps(); + pulses_sum = _mm_setzero_si128(); + for (j=0;j<N;j+=4) + { + __m128 rx4, x4, y4; + __m128i iy4; + x4 = _mm_loadu_ps(&X[j]); + rx4 = _mm_mul_ps(x4, rcp4); + iy4 = _mm_cvttps_epi32(rx4); + pulses_sum = _mm_add_epi32(pulses_sum, iy4); + _mm_storeu_si128((__m128i*)&iy[j], iy4); + y4 = _mm_cvtepi32_ps(iy4); + xy4 = _mm_add_ps(xy4, _mm_mul_ps(x4, y4)); + yy4 = _mm_add_ps(yy4, _mm_mul_ps(y4, y4)); + /* double the y[] vector so we don't have to do it in the search loop. */ + _mm_storeu_ps(&y[j], _mm_add_ps(y4, y4)); + } + pulses_sum = _mm_add_epi32(pulses_sum, _mm_shuffle_epi32(pulses_sum, _MM_SHUFFLE(1, 0, 3, 2))); + pulses_sum = _mm_add_epi32(pulses_sum, _mm_shuffle_epi32(pulses_sum, _MM_SHUFFLE(2, 3, 0, 1))); + pulsesLeft -= _mm_cvtsi128_si32(pulses_sum); + xy4 = _mm_add_ps(xy4, _mm_shuffle_ps(xy4, xy4, _MM_SHUFFLE(1, 0, 3, 2))); + xy4 = _mm_add_ps(xy4, _mm_shuffle_ps(xy4, xy4, _MM_SHUFFLE(2, 3, 0, 1))); + xy = _mm_cvtss_f32(xy4); + yy4 = _mm_add_ps(yy4, _mm_shuffle_ps(yy4, yy4, _MM_SHUFFLE(1, 0, 3, 2))); + yy4 = _mm_add_ps(yy4, _mm_shuffle_ps(yy4, yy4, _MM_SHUFFLE(2, 3, 0, 1))); + yy = _mm_cvtss_f32(yy4); + } + X[N] = X[N+1] = X[N+2] = -100; + y[N] = y[N+1] = y[N+2] = 100; + celt_sig_assert(pulsesLeft>=0); + + /* This should never happen, but just in case it does (e.g. on silence) + we fill the first bin with pulses. */ + if (pulsesLeft > N+3) + { + opus_val16 tmp = (opus_val16)pulsesLeft; + yy = MAC16_16(yy, tmp, tmp); + yy = MAC16_16(yy, tmp, y[0]); + iy[0] += pulsesLeft; + pulsesLeft=0; + } + + for (i=0;i<pulsesLeft;i++) + { + int best_id; + __m128 xy4, yy4; + __m128 max, max2; + __m128i count; + __m128i pos; + /* The squared magnitude term gets added anyway, so we might as well + add it outside the loop */ + yy = ADD16(yy, 1); + xy4 = _mm_load1_ps(&xy); + yy4 = _mm_load1_ps(&yy); + max = _mm_setzero_ps(); + pos = _mm_setzero_si128(); + count = _mm_set_epi32(3, 2, 1, 0); + for (j=0;j<N;j+=4) + { + __m128 x4, y4, r4; + x4 = _mm_loadu_ps(&X[j]); + y4 = _mm_loadu_ps(&y[j]); + x4 = _mm_add_ps(x4, xy4); + y4 = _mm_add_ps(y4, yy4); + y4 = _mm_rsqrt_ps(y4); + r4 = _mm_mul_ps(x4, y4); + /* Update the index of the max. */ + pos = _mm_max_epi16(pos, _mm_and_si128(count, _mm_castps_si128(_mm_cmpgt_ps(r4, max)))); + /* Update the max. */ + max = _mm_max_ps(max, r4); + /* Update the indices (+4) */ + count = _mm_add_epi32(count, fours); + } + /* Horizontal max */ + max2 = _mm_max_ps(max, _mm_shuffle_ps(max, max, _MM_SHUFFLE(1, 0, 3, 2))); + max2 = _mm_max_ps(max2, _mm_shuffle_ps(max2, max2, _MM_SHUFFLE(2, 3, 0, 1))); + /* Now that max2 contains the max at all positions, look at which value(s) of the + partial max is equal to the global max. */ + pos = _mm_and_si128(pos, _mm_castps_si128(_mm_cmpeq_ps(max, max2))); + pos = _mm_max_epi16(pos, _mm_unpackhi_epi64(pos, pos)); + pos = _mm_max_epi16(pos, _mm_shufflelo_epi16(pos, _MM_SHUFFLE(1, 0, 3, 2))); + best_id = _mm_cvtsi128_si32(pos); + + /* Updating the sums of the new pulse(s) */ + xy = ADD32(xy, EXTEND32(X[best_id])); + /* We're multiplying y[j] by two so we don't have to do it here */ + yy = ADD16(yy, y[best_id]); + + /* Only now that we've made the final choice, update y/iy */ + /* Multiplying y[j] by 2 so we don't have to do it everywhere else */ + y[best_id] += 2; + iy[best_id]++; + } + + /* Put the original sign back */ + for (j=0;j<N;j+=4) + { + __m128i y4; + __m128i s4; + y4 = _mm_loadu_si128((__m128i*)&iy[j]); + s4 = _mm_castps_si128(_mm_loadu_ps(&signy[j])); + y4 = _mm_xor_si128(_mm_add_epi32(y4, s4), s4); + _mm_storeu_si128((__m128i*)&iy[j], y4); + } + RESTORE_STACK; + return yy; +} + +#endif diff --git a/media/libopus/celt/x86/x86_celt_map.c b/media/libopus/celt/x86/x86_celt_map.c new file mode 100644 index 0000000000..d39d88edec --- /dev/null +++ b/media/libopus/celt/x86/x86_celt_map.c @@ -0,0 +1,167 @@ +/* Copyright (c) 2014, Cisco Systems, INC + Written by XiangMingZhu WeiZhou MinPeng YanWang + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#if defined(HAVE_CONFIG_H) +#include "config.h" +#endif + +#include "x86/x86cpu.h" +#include "celt_lpc.h" +#include "pitch.h" +#include "pitch_sse.h" +#include "vq.h" + +#if defined(OPUS_HAVE_RTCD) + +# if defined(FIXED_POINT) + +#if defined(OPUS_X86_MAY_HAVE_SSE4_1) && !defined(OPUS_X86_PRESUME_SSE4_1) + +void (*const CELT_FIR_IMPL[OPUS_ARCHMASK + 1])( + const opus_val16 *x, + const opus_val16 *num, + opus_val16 *y, + int N, + int ord, + int arch +) = { + celt_fir_c, /* non-sse */ + celt_fir_c, + celt_fir_c, + MAY_HAVE_SSE4_1(celt_fir), /* sse4.1 */ + MAY_HAVE_SSE4_1(celt_fir) /* avx */ +}; + +void (*const XCORR_KERNEL_IMPL[OPUS_ARCHMASK + 1])( + const opus_val16 *x, + const opus_val16 *y, + opus_val32 sum[4], + int len +) = { + xcorr_kernel_c, /* non-sse */ + xcorr_kernel_c, + xcorr_kernel_c, + MAY_HAVE_SSE4_1(xcorr_kernel), /* sse4.1 */ + MAY_HAVE_SSE4_1(xcorr_kernel) /* avx */ +}; + +#endif + +#if (defined(OPUS_X86_MAY_HAVE_SSE4_1) && !defined(OPUS_X86_PRESUME_SSE4_1)) || \ + (!defined(OPUS_X86_MAY_HAVE_SSE_4_1) && defined(OPUS_X86_MAY_HAVE_SSE2) && !defined(OPUS_X86_PRESUME_SSE2)) + +opus_val32 (*const CELT_INNER_PROD_IMPL[OPUS_ARCHMASK + 1])( + const opus_val16 *x, + const opus_val16 *y, + int N +) = { + celt_inner_prod_c, /* non-sse */ + celt_inner_prod_c, + MAY_HAVE_SSE2(celt_inner_prod), + MAY_HAVE_SSE4_1(celt_inner_prod), /* sse4.1 */ + MAY_HAVE_SSE4_1(celt_inner_prod) /* avx */ +}; + +#endif + +# else + +#if defined(OPUS_X86_MAY_HAVE_SSE) && !defined(OPUS_X86_PRESUME_SSE) + +void (*const XCORR_KERNEL_IMPL[OPUS_ARCHMASK + 1])( + const opus_val16 *x, + const opus_val16 *y, + opus_val32 sum[4], + int len +) = { + xcorr_kernel_c, /* non-sse */ + MAY_HAVE_SSE(xcorr_kernel), + MAY_HAVE_SSE(xcorr_kernel), + MAY_HAVE_SSE(xcorr_kernel), + MAY_HAVE_SSE(xcorr_kernel) +}; + +opus_val32 (*const CELT_INNER_PROD_IMPL[OPUS_ARCHMASK + 1])( + const opus_val16 *x, + const opus_val16 *y, + int N +) = { + celt_inner_prod_c, /* non-sse */ + MAY_HAVE_SSE(celt_inner_prod), + MAY_HAVE_SSE(celt_inner_prod), + MAY_HAVE_SSE(celt_inner_prod), + MAY_HAVE_SSE(celt_inner_prod) +}; + +void (*const DUAL_INNER_PROD_IMPL[OPUS_ARCHMASK + 1])( + const opus_val16 *x, + const opus_val16 *y01, + const opus_val16 *y02, + int N, + opus_val32 *xy1, + opus_val32 *xy2 +) = { + dual_inner_prod_c, /* non-sse */ + MAY_HAVE_SSE(dual_inner_prod), + MAY_HAVE_SSE(dual_inner_prod), + MAY_HAVE_SSE(dual_inner_prod), + MAY_HAVE_SSE(dual_inner_prod) +}; + +void (*const COMB_FILTER_CONST_IMPL[OPUS_ARCHMASK + 1])( + opus_val32 *y, + opus_val32 *x, + int T, + int N, + opus_val16 g10, + opus_val16 g11, + opus_val16 g12 +) = { + comb_filter_const_c, /* non-sse */ + MAY_HAVE_SSE(comb_filter_const), + MAY_HAVE_SSE(comb_filter_const), + MAY_HAVE_SSE(comb_filter_const), + MAY_HAVE_SSE(comb_filter_const) +}; + + +#endif + +#if defined(OPUS_X86_MAY_HAVE_SSE2) && !defined(OPUS_X86_PRESUME_SSE2) +opus_val16 (*const OP_PVQ_SEARCH_IMPL[OPUS_ARCHMASK + 1])( + celt_norm *_X, int *iy, int K, int N, int arch +) = { + op_pvq_search_c, /* non-sse */ + op_pvq_search_c, + MAY_HAVE_SSE2(op_pvq_search), + MAY_HAVE_SSE2(op_pvq_search), + MAY_HAVE_SSE2(op_pvq_search) +}; +#endif + +#endif +#endif diff --git a/media/libopus/celt/x86/x86cpu.c b/media/libopus/celt/x86/x86cpu.c new file mode 100644 index 0000000000..6a1914dee7 --- /dev/null +++ b/media/libopus/celt/x86/x86cpu.c @@ -0,0 +1,178 @@ +/* Copyright (c) 2014, Cisco Systems, INC + Written by XiangMingZhu WeiZhou MinPeng YanWang + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "cpu_support.h" +#include "macros.h" +#include "main.h" +#include "pitch.h" +#include "x86cpu.h" + +#if defined(OPUS_HAVE_RTCD) && \ + ((defined(OPUS_X86_MAY_HAVE_SSE) && !defined(OPUS_X86_PRESUME_SSE)) || \ + (defined(OPUS_X86_MAY_HAVE_SSE2) && !defined(OPUS_X86_PRESUME_SSE2)) || \ + (defined(OPUS_X86_MAY_HAVE_SSE4_1) && !defined(OPUS_X86_PRESUME_SSE4_1)) || \ + (defined(OPUS_X86_MAY_HAVE_AVX) && !defined(OPUS_X86_PRESUME_AVX))) + +#if defined(_MSC_VER) + +#include <intrin.h> +static _inline void cpuid(unsigned int CPUInfo[4], unsigned int InfoType) +{ + __cpuid((int*)CPUInfo, InfoType); +} + +#else + +#if defined(CPU_INFO_BY_C) +#include <cpuid.h> +#endif + +static void cpuid(unsigned int CPUInfo[4], unsigned int InfoType) +{ +#if defined(CPU_INFO_BY_ASM) +#if defined(__i386__) && defined(__PIC__) +/* %ebx is PIC register in 32-bit, so mustn't clobber it. */ + __asm__ __volatile__ ( + "xchg %%ebx, %1\n" + "cpuid\n" + "xchg %%ebx, %1\n": + "=a" (CPUInfo[0]), + "=r" (CPUInfo[1]), + "=c" (CPUInfo[2]), + "=d" (CPUInfo[3]) : + /* We clear ECX to avoid a valgrind false-positive prior to v3.17.0. */ + "0" (InfoType), "2" (0) + ); +#else + __asm__ __volatile__ ( + "cpuid": + "=a" (CPUInfo[0]), + "=b" (CPUInfo[1]), + "=c" (CPUInfo[2]), + "=d" (CPUInfo[3]) : + /* We clear ECX to avoid a valgrind false-positive prior to v3.17.0. */ + "0" (InfoType), "2" (0) + ); +#endif +#elif defined(CPU_INFO_BY_C) + /* We use __get_cpuid_count to clear ECX to avoid a valgrind false-positive + prior to v3.17.0.*/ + if (!__get_cpuid_count(InfoType, 0, &(CPUInfo[0]), &(CPUInfo[1]), &(CPUInfo[2]), &(CPUInfo[3]))) { + /* Our function cannot fail, but __get_cpuid{_count} can. + Returning all zeroes will effectively disable all SIMD, which is + what we want on CPUs that don't support CPUID. */ + CPUInfo[3] = CPUInfo[2] = CPUInfo[1] = CPUInfo[0] = 0; + } +#else +# error "Configured to use x86 RTCD, but no CPU detection method available. " \ + "Reconfigure with --disable-rtcd (or send patches)." +#endif +} + +#endif + +typedef struct CPU_Feature{ + /* SIMD: 128-bit */ + int HW_SSE; + int HW_SSE2; + int HW_SSE41; + /* SIMD: 256-bit */ + int HW_AVX; +} CPU_Feature; + +static void opus_cpu_feature_check(CPU_Feature *cpu_feature) +{ + unsigned int info[4]; + unsigned int nIds = 0; + + cpuid(info, 0); + nIds = info[0]; + + if (nIds >= 1){ + cpuid(info, 1); + cpu_feature->HW_SSE = (info[3] & (1 << 25)) != 0; + cpu_feature->HW_SSE2 = (info[3] & (1 << 26)) != 0; + cpu_feature->HW_SSE41 = (info[2] & (1 << 19)) != 0; + cpu_feature->HW_AVX = (info[2] & (1 << 28)) != 0; + } + else { + cpu_feature->HW_SSE = 0; + cpu_feature->HW_SSE2 = 0; + cpu_feature->HW_SSE41 = 0; + cpu_feature->HW_AVX = 0; + } +} + +static int opus_select_arch_impl(void) +{ + CPU_Feature cpu_feature; + int arch; + + opus_cpu_feature_check(&cpu_feature); + + arch = 0; + if (!cpu_feature.HW_SSE) + { + return arch; + } + arch++; + + if (!cpu_feature.HW_SSE2) + { + return arch; + } + arch++; + + if (!cpu_feature.HW_SSE41) + { + return arch; + } + arch++; + + if (!cpu_feature.HW_AVX) + { + return arch; + } + arch++; + + return arch; +} + +int opus_select_arch(void) { + int arch = opus_select_arch_impl(); +#ifdef FUZZING + /* Randomly downgrade the architecture. */ + arch = rand()%(arch+1); +#endif + return arch; +} + +#endif diff --git a/media/libopus/celt/x86/x86cpu.h b/media/libopus/celt/x86/x86cpu.h new file mode 100644 index 0000000000..04e80489b1 --- /dev/null +++ b/media/libopus/celt/x86/x86cpu.h @@ -0,0 +1,73 @@ +/* Copyright (c) 2014, Cisco Systems, INC + Written by XiangMingZhu WeiZhou MinPeng YanWang + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#if !defined(X86CPU_H) +# define X86CPU_H + +# if defined(OPUS_X86_MAY_HAVE_SSE) +# define MAY_HAVE_SSE(name) name ## _sse +# else +# define MAY_HAVE_SSE(name) name ## _c +# endif + +# if defined(OPUS_X86_MAY_HAVE_SSE2) +# define MAY_HAVE_SSE2(name) name ## _sse2 +# else +# define MAY_HAVE_SSE2(name) name ## _c +# endif + +# if defined(OPUS_X86_MAY_HAVE_SSE4_1) +# define MAY_HAVE_SSE4_1(name) name ## _sse4_1 +# else +# define MAY_HAVE_SSE4_1(name) name ## _c +# endif + +# if defined(OPUS_X86_MAY_HAVE_AVX) +# define MAY_HAVE_AVX(name) name ## _avx +# else +# define MAY_HAVE_AVX(name) name ## _c +# endif + +# if defined(OPUS_HAVE_RTCD) +int opus_select_arch(void); +# endif + +/*MOVD should not impose any alignment restrictions, but the C standard does, + and UBSan will report errors if we actually make unaligned accesses. + Use this to work around those restrictions (which should hopefully all get + optimized to a single MOVD instruction).*/ +#define OP_LOADU_EPI32(x) \ + (int)((*(unsigned char *)(x) | *((unsigned char *)(x) + 1) << 8U |\ + *((unsigned char *)(x) + 2) << 16U | (opus_uint32)*((unsigned char *)(x) + 3) << 24U)) + +#define OP_CVTEPI8_EPI32_M32(x) \ + (_mm_cvtepi8_epi32(_mm_cvtsi32_si128(OP_LOADU_EPI32(x)))) + +#define OP_CVTEPI16_EPI32_M64(x) \ + (_mm_cvtepi16_epi32(_mm_loadl_epi64((__m128i *)(x)))) + +#endif diff --git a/media/libopus/gen-sources.py b/media/libopus/gen-sources.py new file mode 100644 index 0000000000..a06f7366fd --- /dev/null +++ b/media/libopus/gen-sources.py @@ -0,0 +1,83 @@ +# This Source Code Form is subject to the terms of the Mozilla Public +# License, v. 2.0. If a copy of the MPL was not distributed with this +# file, You can obtain one at http://mozilla.org/MPL/2.0/. + +# This file takes the .mk files from an upstream opus repo and generates the +# sources.mozbuild file. It is invoked as part of update.sh + +import sys +import re + +# These files are not used, ignore them. +ignore_list = [ + 'silk/float/regularize_correlations_FLP.c', + 'silk/float/LPC_inv_pred_gain_FLP.c', + 'src/opus_projection_encoder.c', + 'silk/debug.c', + 'src/mapping_matrix.c', + 'src/opus_projection_decoder.c', +] + +def should_ignore(value): + return any(item in value for item in ignore_list) + +def add_value(values, text): + text = text.replace('\\', '') + text = text.strip() + if text and not should_ignore(text): + values.append(text) + +def write_values(output, values): + for value in sorted(values, key=lambda x: x.lower()): + output.write(" '%s',\n" % value) + output.write(']\n\n') + +def generate_sources_mozbuild(path): + makefiles = [ + 'celt_sources.mk', + 'opus_sources.mk', + 'silk_sources.mk', + ] + + var_definition = re.compile('([A-Z0-9_]*) = (.*)') + with open('sources.mozbuild', 'w') as output: + + output.write('# THIS FILE WAS AUTOMATICALLY GENERATED BY %s. DO NOT EDIT.\n' % sys.argv[0]) + for makefile in makefiles: + values = [] + definition_started = False + + with open('%s/%s' % (path, makefile), 'r') as mk: + for line in mk: + line = line.rstrip() + result = var_definition.match(line) + if result: + if definition_started: + write_values(output, values) + values = [] + definition_started = True + + # Some variable definitions have the first entry on the + # first line. Eg: + # + # CELT_SOURCES = celt/bands.c + # + # So we treat the first group as the variable name and + # the second group as a potential value. + # + # Note that we write the variable name in lower case (so + # "CELT_SOURCES" in the .mk file becomes "celt_sources" + # in the .mozbuild file) because moz.build reserves + # upper-case variable names for build system outputs. + output.write('%s = [\n' % result.group(1).lower()) + add_value(values, result.group(2)) + else: + add_value(values, line) + write_values(output, values) + +if __name__ == '__main__': + if len(sys.argv) != 2: + print("Usage: %s /path/to/opus" % (sys.argv[0])) + sys.exit(1) + + generate_sources_mozbuild(sys.argv[1]) diff --git a/media/libopus/include/opus.h b/media/libopus/include/opus.h new file mode 100644 index 0000000000..0c69c627d9 --- /dev/null +++ b/media/libopus/include/opus.h @@ -0,0 +1,981 @@ +/* Copyright (c) 2010-2011 Xiph.Org Foundation, Skype Limited + Written by Jean-Marc Valin and Koen Vos */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +/** + * @file opus.h + * @brief Opus reference implementation API + */ + +#ifndef OPUS_H +#define OPUS_H + +#include "opus_types.h" +#include "opus_defines.h" + +#ifdef __cplusplus +extern "C" { +#endif + +/** + * @mainpage Opus + * + * The Opus codec is designed for interactive speech and audio transmission over the Internet. + * It is designed by the IETF Codec Working Group and incorporates technology from + * Skype's SILK codec and Xiph.Org's CELT codec. + * + * The Opus codec is designed to handle a wide range of interactive audio applications, + * including Voice over IP, videoconferencing, in-game chat, and even remote live music + * performances. It can scale from low bit-rate narrowband speech to very high quality + * stereo music. Its main features are: + + * @li Sampling rates from 8 to 48 kHz + * @li Bit-rates from 6 kb/s to 510 kb/s + * @li Support for both constant bit-rate (CBR) and variable bit-rate (VBR) + * @li Audio bandwidth from narrowband to full-band + * @li Support for speech and music + * @li Support for mono and stereo + * @li Support for multichannel (up to 255 channels) + * @li Frame sizes from 2.5 ms to 60 ms + * @li Good loss robustness and packet loss concealment (PLC) + * @li Floating point and fixed-point implementation + * + * Documentation sections: + * @li @ref opus_encoder + * @li @ref opus_decoder + * @li @ref opus_repacketizer + * @li @ref opus_multistream + * @li @ref opus_libinfo + * @li @ref opus_custom + */ + +/** @defgroup opus_encoder Opus Encoder + * @{ + * + * @brief This page describes the process and functions used to encode Opus. + * + * Since Opus is a stateful codec, the encoding process starts with creating an encoder + * state. This can be done with: + * + * @code + * int error; + * OpusEncoder *enc; + * enc = opus_encoder_create(Fs, channels, application, &error); + * @endcode + * + * From this point, @c enc can be used for encoding an audio stream. An encoder state + * @b must @b not be used for more than one stream at the same time. Similarly, the encoder + * state @b must @b not be re-initialized for each frame. + * + * While opus_encoder_create() allocates memory for the state, it's also possible + * to initialize pre-allocated memory: + * + * @code + * int size; + * int error; + * OpusEncoder *enc; + * size = opus_encoder_get_size(channels); + * enc = malloc(size); + * error = opus_encoder_init(enc, Fs, channels, application); + * @endcode + * + * where opus_encoder_get_size() returns the required size for the encoder state. Note that + * future versions of this code may change the size, so no assuptions should be made about it. + * + * The encoder state is always continuous in memory and only a shallow copy is sufficient + * to copy it (e.g. memcpy()) + * + * It is possible to change some of the encoder's settings using the opus_encoder_ctl() + * interface. All these settings already default to the recommended value, so they should + * only be changed when necessary. The most common settings one may want to change are: + * + * @code + * opus_encoder_ctl(enc, OPUS_SET_BITRATE(bitrate)); + * opus_encoder_ctl(enc, OPUS_SET_COMPLEXITY(complexity)); + * opus_encoder_ctl(enc, OPUS_SET_SIGNAL(signal_type)); + * @endcode + * + * where + * + * @arg bitrate is in bits per second (b/s) + * @arg complexity is a value from 1 to 10, where 1 is the lowest complexity and 10 is the highest + * @arg signal_type is either OPUS_AUTO (default), OPUS_SIGNAL_VOICE, or OPUS_SIGNAL_MUSIC + * + * See @ref opus_encoderctls and @ref opus_genericctls for a complete list of parameters that can be set or queried. Most parameters can be set or changed at any time during a stream. + * + * To encode a frame, opus_encode() or opus_encode_float() must be called with exactly one frame (2.5, 5, 10, 20, 40 or 60 ms) of audio data: + * @code + * len = opus_encode(enc, audio_frame, frame_size, packet, max_packet); + * @endcode + * + * where + * <ul> + * <li>audio_frame is the audio data in opus_int16 (or float for opus_encode_float())</li> + * <li>frame_size is the duration of the frame in samples (per channel)</li> + * <li>packet is the byte array to which the compressed data is written</li> + * <li>max_packet is the maximum number of bytes that can be written in the packet (4000 bytes is recommended). + * Do not use max_packet to control VBR target bitrate, instead use the #OPUS_SET_BITRATE CTL.</li> + * </ul> + * + * opus_encode() and opus_encode_float() return the number of bytes actually written to the packet. + * The return value <b>can be negative</b>, which indicates that an error has occurred. If the return value + * is 2 bytes or less, then the packet does not need to be transmitted (DTX). + * + * Once the encoder state if no longer needed, it can be destroyed with + * + * @code + * opus_encoder_destroy(enc); + * @endcode + * + * If the encoder was created with opus_encoder_init() rather than opus_encoder_create(), + * then no action is required aside from potentially freeing the memory that was manually + * allocated for it (calling free(enc) for the example above) + * + */ + +/** Opus encoder state. + * This contains the complete state of an Opus encoder. + * It is position independent and can be freely copied. + * @see opus_encoder_create,opus_encoder_init + */ +typedef struct OpusEncoder OpusEncoder; + +/** Gets the size of an <code>OpusEncoder</code> structure. + * @param[in] channels <tt>int</tt>: Number of channels. + * This must be 1 or 2. + * @returns The size in bytes. + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT int opus_encoder_get_size(int channels); + +/** + */ + +/** Allocates and initializes an encoder state. + * There are three coding modes: + * + * @ref OPUS_APPLICATION_VOIP gives best quality at a given bitrate for voice + * signals. It enhances the input signal by high-pass filtering and + * emphasizing formants and harmonics. Optionally it includes in-band + * forward error correction to protect against packet loss. Use this + * mode for typical VoIP applications. Because of the enhancement, + * even at high bitrates the output may sound different from the input. + * + * @ref OPUS_APPLICATION_AUDIO gives best quality at a given bitrate for most + * non-voice signals like music. Use this mode for music and mixed + * (music/voice) content, broadcast, and applications requiring less + * than 15 ms of coding delay. + * + * @ref OPUS_APPLICATION_RESTRICTED_LOWDELAY configures low-delay mode that + * disables the speech-optimized mode in exchange for slightly reduced delay. + * This mode can only be set on an newly initialized or freshly reset encoder + * because it changes the codec delay. + * + * This is useful when the caller knows that the speech-optimized modes will not be needed (use with caution). + * @param [in] Fs <tt>opus_int32</tt>: Sampling rate of input signal (Hz) + * This must be one of 8000, 12000, 16000, + * 24000, or 48000. + * @param [in] channels <tt>int</tt>: Number of channels (1 or 2) in input signal + * @param [in] application <tt>int</tt>: Coding mode (one of @ref OPUS_APPLICATION_VOIP, @ref OPUS_APPLICATION_AUDIO, or @ref OPUS_APPLICATION_RESTRICTED_LOWDELAY) + * @param [out] error <tt>int*</tt>: @ref opus_errorcodes + * @note Regardless of the sampling rate and number channels selected, the Opus encoder + * can switch to a lower audio bandwidth or number of channels if the bitrate + * selected is too low. This also means that it is safe to always use 48 kHz stereo input + * and let the encoder optimize the encoding. + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT OpusEncoder *opus_encoder_create( + opus_int32 Fs, + int channels, + int application, + int *error +); + +/** Initializes a previously allocated encoder state + * The memory pointed to by st must be at least the size returned by opus_encoder_get_size(). + * This is intended for applications which use their own allocator instead of malloc. + * @see opus_encoder_create(),opus_encoder_get_size() + * To reset a previously initialized state, use the #OPUS_RESET_STATE CTL. + * @param [in] st <tt>OpusEncoder*</tt>: Encoder state + * @param [in] Fs <tt>opus_int32</tt>: Sampling rate of input signal (Hz) + * This must be one of 8000, 12000, 16000, + * 24000, or 48000. + * @param [in] channels <tt>int</tt>: Number of channels (1 or 2) in input signal + * @param [in] application <tt>int</tt>: Coding mode (one of OPUS_APPLICATION_VOIP, OPUS_APPLICATION_AUDIO, or OPUS_APPLICATION_RESTRICTED_LOWDELAY) + * @retval #OPUS_OK Success or @ref opus_errorcodes + */ +OPUS_EXPORT int opus_encoder_init( + OpusEncoder *st, + opus_int32 Fs, + int channels, + int application +) OPUS_ARG_NONNULL(1); + +/** Encodes an Opus frame. + * @param [in] st <tt>OpusEncoder*</tt>: Encoder state + * @param [in] pcm <tt>opus_int16*</tt>: Input signal (interleaved if 2 channels). length is frame_size*channels*sizeof(opus_int16) + * @param [in] frame_size <tt>int</tt>: Number of samples per channel in the + * input signal. + * This must be an Opus frame size for + * the encoder's sampling rate. + * For example, at 48 kHz the permitted + * values are 120, 240, 480, 960, 1920, + * and 2880. + * Passing in a duration of less than + * 10 ms (480 samples at 48 kHz) will + * prevent the encoder from using the LPC + * or hybrid modes. + * @param [out] data <tt>unsigned char*</tt>: Output payload. + * This must contain storage for at + * least \a max_data_bytes. + * @param [in] max_data_bytes <tt>opus_int32</tt>: Size of the allocated + * memory for the output + * payload. This may be + * used to impose an upper limit on + * the instant bitrate, but should + * not be used as the only bitrate + * control. Use #OPUS_SET_BITRATE to + * control the bitrate. + * @returns The length of the encoded packet (in bytes) on success or a + * negative error code (see @ref opus_errorcodes) on failure. + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT opus_int32 opus_encode( + OpusEncoder *st, + const opus_int16 *pcm, + int frame_size, + unsigned char *data, + opus_int32 max_data_bytes +) OPUS_ARG_NONNULL(1) OPUS_ARG_NONNULL(2) OPUS_ARG_NONNULL(4); + +/** Encodes an Opus frame from floating point input. + * @param [in] st <tt>OpusEncoder*</tt>: Encoder state + * @param [in] pcm <tt>float*</tt>: Input in float format (interleaved if 2 channels), with a normal range of +/-1.0. + * Samples with a range beyond +/-1.0 are supported but will + * be clipped by decoders using the integer API and should + * only be used if it is known that the far end supports + * extended dynamic range. + * length is frame_size*channels*sizeof(float) + * @param [in] frame_size <tt>int</tt>: Number of samples per channel in the + * input signal. + * This must be an Opus frame size for + * the encoder's sampling rate. + * For example, at 48 kHz the permitted + * values are 120, 240, 480, 960, 1920, + * and 2880. + * Passing in a duration of less than + * 10 ms (480 samples at 48 kHz) will + * prevent the encoder from using the LPC + * or hybrid modes. + * @param [out] data <tt>unsigned char*</tt>: Output payload. + * This must contain storage for at + * least \a max_data_bytes. + * @param [in] max_data_bytes <tt>opus_int32</tt>: Size of the allocated + * memory for the output + * payload. This may be + * used to impose an upper limit on + * the instant bitrate, but should + * not be used as the only bitrate + * control. Use #OPUS_SET_BITRATE to + * control the bitrate. + * @returns The length of the encoded packet (in bytes) on success or a + * negative error code (see @ref opus_errorcodes) on failure. + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT opus_int32 opus_encode_float( + OpusEncoder *st, + const float *pcm, + int frame_size, + unsigned char *data, + opus_int32 max_data_bytes +) OPUS_ARG_NONNULL(1) OPUS_ARG_NONNULL(2) OPUS_ARG_NONNULL(4); + +/** Frees an <code>OpusEncoder</code> allocated by opus_encoder_create(). + * @param[in] st <tt>OpusEncoder*</tt>: State to be freed. + */ +OPUS_EXPORT void opus_encoder_destroy(OpusEncoder *st); + +/** Perform a CTL function on an Opus encoder. + * + * Generally the request and subsequent arguments are generated + * by a convenience macro. + * @param st <tt>OpusEncoder*</tt>: Encoder state. + * @param request This and all remaining parameters should be replaced by one + * of the convenience macros in @ref opus_genericctls or + * @ref opus_encoderctls. + * @see opus_genericctls + * @see opus_encoderctls + */ +OPUS_EXPORT int opus_encoder_ctl(OpusEncoder *st, int request, ...) OPUS_ARG_NONNULL(1); +/**@}*/ + +/** @defgroup opus_decoder Opus Decoder + * @{ + * + * @brief This page describes the process and functions used to decode Opus. + * + * The decoding process also starts with creating a decoder + * state. This can be done with: + * @code + * int error; + * OpusDecoder *dec; + * dec = opus_decoder_create(Fs, channels, &error); + * @endcode + * where + * @li Fs is the sampling rate and must be 8000, 12000, 16000, 24000, or 48000 + * @li channels is the number of channels (1 or 2) + * @li error will hold the error code in case of failure (or #OPUS_OK on success) + * @li the return value is a newly created decoder state to be used for decoding + * + * While opus_decoder_create() allocates memory for the state, it's also possible + * to initialize pre-allocated memory: + * @code + * int size; + * int error; + * OpusDecoder *dec; + * size = opus_decoder_get_size(channels); + * dec = malloc(size); + * error = opus_decoder_init(dec, Fs, channels); + * @endcode + * where opus_decoder_get_size() returns the required size for the decoder state. Note that + * future versions of this code may change the size, so no assuptions should be made about it. + * + * The decoder state is always continuous in memory and only a shallow copy is sufficient + * to copy it (e.g. memcpy()) + * + * To decode a frame, opus_decode() or opus_decode_float() must be called with a packet of compressed audio data: + * @code + * frame_size = opus_decode(dec, packet, len, decoded, max_size, 0); + * @endcode + * where + * + * @li packet is the byte array containing the compressed data + * @li len is the exact number of bytes contained in the packet + * @li decoded is the decoded audio data in opus_int16 (or float for opus_decode_float()) + * @li max_size is the max duration of the frame in samples (per channel) that can fit into the decoded_frame array + * + * opus_decode() and opus_decode_float() return the number of samples (per channel) decoded from the packet. + * If that value is negative, then an error has occurred. This can occur if the packet is corrupted or if the audio + * buffer is too small to hold the decoded audio. + * + * Opus is a stateful codec with overlapping blocks and as a result Opus + * packets are not coded independently of each other. Packets must be + * passed into the decoder serially and in the correct order for a correct + * decode. Lost packets can be replaced with loss concealment by calling + * the decoder with a null pointer and zero length for the missing packet. + * + * A single codec state may only be accessed from a single thread at + * a time and any required locking must be performed by the caller. Separate + * streams must be decoded with separate decoder states and can be decoded + * in parallel unless the library was compiled with NONTHREADSAFE_PSEUDOSTACK + * defined. + * + */ + +/** Opus decoder state. + * This contains the complete state of an Opus decoder. + * It is position independent and can be freely copied. + * @see opus_decoder_create,opus_decoder_init + */ +typedef struct OpusDecoder OpusDecoder; + +/** Gets the size of an <code>OpusDecoder</code> structure. + * @param [in] channels <tt>int</tt>: Number of channels. + * This must be 1 or 2. + * @returns The size in bytes. + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT int opus_decoder_get_size(int channels); + +/** Allocates and initializes a decoder state. + * @param [in] Fs <tt>opus_int32</tt>: Sample rate to decode at (Hz). + * This must be one of 8000, 12000, 16000, + * 24000, or 48000. + * @param [in] channels <tt>int</tt>: Number of channels (1 or 2) to decode + * @param [out] error <tt>int*</tt>: #OPUS_OK Success or @ref opus_errorcodes + * + * Internally Opus stores data at 48000 Hz, so that should be the default + * value for Fs. However, the decoder can efficiently decode to buffers + * at 8, 12, 16, and 24 kHz so if for some reason the caller cannot use + * data at the full sample rate, or knows the compressed data doesn't + * use the full frequency range, it can request decoding at a reduced + * rate. Likewise, the decoder is capable of filling in either mono or + * interleaved stereo pcm buffers, at the caller's request. + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT OpusDecoder *opus_decoder_create( + opus_int32 Fs, + int channels, + int *error +); + +/** Initializes a previously allocated decoder state. + * The state must be at least the size returned by opus_decoder_get_size(). + * This is intended for applications which use their own allocator instead of malloc. @see opus_decoder_create,opus_decoder_get_size + * To reset a previously initialized state, use the #OPUS_RESET_STATE CTL. + * @param [in] st <tt>OpusDecoder*</tt>: Decoder state. + * @param [in] Fs <tt>opus_int32</tt>: Sampling rate to decode to (Hz). + * This must be one of 8000, 12000, 16000, + * 24000, or 48000. + * @param [in] channels <tt>int</tt>: Number of channels (1 or 2) to decode + * @retval #OPUS_OK Success or @ref opus_errorcodes + */ +OPUS_EXPORT int opus_decoder_init( + OpusDecoder *st, + opus_int32 Fs, + int channels +) OPUS_ARG_NONNULL(1); + +/** Decode an Opus packet. + * @param [in] st <tt>OpusDecoder*</tt>: Decoder state + * @param [in] data <tt>char*</tt>: Input payload. Use a NULL pointer to indicate packet loss + * @param [in] len <tt>opus_int32</tt>: Number of bytes in payload* + * @param [out] pcm <tt>opus_int16*</tt>: Output signal (interleaved if 2 channels). length + * is frame_size*channels*sizeof(opus_int16) + * @param [in] frame_size Number of samples per channel of available space in \a pcm. + * If this is less than the maximum packet duration (120ms; 5760 for 48kHz), this function will + * not be capable of decoding some packets. In the case of PLC (data==NULL) or FEC (decode_fec=1), + * then frame_size needs to be exactly the duration of audio that is missing, otherwise the + * decoder will not be in the optimal state to decode the next incoming packet. For the PLC and + * FEC cases, frame_size <b>must</b> be a multiple of 2.5 ms. + * @param [in] decode_fec <tt>int</tt>: Flag (0 or 1) to request that any in-band forward error correction data be + * decoded. If no such data is available, the frame is decoded as if it were lost. + * @returns Number of decoded samples or @ref opus_errorcodes + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT int opus_decode( + OpusDecoder *st, + const unsigned char *data, + opus_int32 len, + opus_int16 *pcm, + int frame_size, + int decode_fec +) OPUS_ARG_NONNULL(1) OPUS_ARG_NONNULL(4); + +/** Decode an Opus packet with floating point output. + * @param [in] st <tt>OpusDecoder*</tt>: Decoder state + * @param [in] data <tt>char*</tt>: Input payload. Use a NULL pointer to indicate packet loss + * @param [in] len <tt>opus_int32</tt>: Number of bytes in payload + * @param [out] pcm <tt>float*</tt>: Output signal (interleaved if 2 channels). length + * is frame_size*channels*sizeof(float) + * @param [in] frame_size Number of samples per channel of available space in \a pcm. + * If this is less than the maximum packet duration (120ms; 5760 for 48kHz), this function will + * not be capable of decoding some packets. In the case of PLC (data==NULL) or FEC (decode_fec=1), + * then frame_size needs to be exactly the duration of audio that is missing, otherwise the + * decoder will not be in the optimal state to decode the next incoming packet. For the PLC and + * FEC cases, frame_size <b>must</b> be a multiple of 2.5 ms. + * @param [in] decode_fec <tt>int</tt>: Flag (0 or 1) to request that any in-band forward error correction data be + * decoded. If no such data is available the frame is decoded as if it were lost. + * @returns Number of decoded samples or @ref opus_errorcodes + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT int opus_decode_float( + OpusDecoder *st, + const unsigned char *data, + opus_int32 len, + float *pcm, + int frame_size, + int decode_fec +) OPUS_ARG_NONNULL(1) OPUS_ARG_NONNULL(4); + +/** Perform a CTL function on an Opus decoder. + * + * Generally the request and subsequent arguments are generated + * by a convenience macro. + * @param st <tt>OpusDecoder*</tt>: Decoder state. + * @param request This and all remaining parameters should be replaced by one + * of the convenience macros in @ref opus_genericctls or + * @ref opus_decoderctls. + * @see opus_genericctls + * @see opus_decoderctls + */ +OPUS_EXPORT int opus_decoder_ctl(OpusDecoder *st, int request, ...) OPUS_ARG_NONNULL(1); + +/** Frees an <code>OpusDecoder</code> allocated by opus_decoder_create(). + * @param[in] st <tt>OpusDecoder*</tt>: State to be freed. + */ +OPUS_EXPORT void opus_decoder_destroy(OpusDecoder *st); + +/** Parse an opus packet into one or more frames. + * Opus_decode will perform this operation internally so most applications do + * not need to use this function. + * This function does not copy the frames, the returned pointers are pointers into + * the input packet. + * @param [in] data <tt>char*</tt>: Opus packet to be parsed + * @param [in] len <tt>opus_int32</tt>: size of data + * @param [out] out_toc <tt>char*</tt>: TOC pointer + * @param [out] frames <tt>char*[48]</tt> encapsulated frames + * @param [out] size <tt>opus_int16[48]</tt> sizes of the encapsulated frames + * @param [out] payload_offset <tt>int*</tt>: returns the position of the payload within the packet (in bytes) + * @returns number of frames + */ +OPUS_EXPORT int opus_packet_parse( + const unsigned char *data, + opus_int32 len, + unsigned char *out_toc, + const unsigned char *frames[48], + opus_int16 size[48], + int *payload_offset +) OPUS_ARG_NONNULL(1) OPUS_ARG_NONNULL(5); + +/** Gets the bandwidth of an Opus packet. + * @param [in] data <tt>char*</tt>: Opus packet + * @retval OPUS_BANDWIDTH_NARROWBAND Narrowband (4kHz bandpass) + * @retval OPUS_BANDWIDTH_MEDIUMBAND Mediumband (6kHz bandpass) + * @retval OPUS_BANDWIDTH_WIDEBAND Wideband (8kHz bandpass) + * @retval OPUS_BANDWIDTH_SUPERWIDEBAND Superwideband (12kHz bandpass) + * @retval OPUS_BANDWIDTH_FULLBAND Fullband (20kHz bandpass) + * @retval OPUS_INVALID_PACKET The compressed data passed is corrupted or of an unsupported type + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT int opus_packet_get_bandwidth(const unsigned char *data) OPUS_ARG_NONNULL(1); + +/** Gets the number of samples per frame from an Opus packet. + * @param [in] data <tt>char*</tt>: Opus packet. + * This must contain at least one byte of + * data. + * @param [in] Fs <tt>opus_int32</tt>: Sampling rate in Hz. + * This must be a multiple of 400, or + * inaccurate results will be returned. + * @returns Number of samples per frame. + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT int opus_packet_get_samples_per_frame(const unsigned char *data, opus_int32 Fs) OPUS_ARG_NONNULL(1); + +/** Gets the number of channels from an Opus packet. + * @param [in] data <tt>char*</tt>: Opus packet + * @returns Number of channels + * @retval OPUS_INVALID_PACKET The compressed data passed is corrupted or of an unsupported type + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT int opus_packet_get_nb_channels(const unsigned char *data) OPUS_ARG_NONNULL(1); + +/** Gets the number of frames in an Opus packet. + * @param [in] packet <tt>char*</tt>: Opus packet + * @param [in] len <tt>opus_int32</tt>: Length of packet + * @returns Number of frames + * @retval OPUS_BAD_ARG Insufficient data was passed to the function + * @retval OPUS_INVALID_PACKET The compressed data passed is corrupted or of an unsupported type + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT int opus_packet_get_nb_frames(const unsigned char packet[], opus_int32 len) OPUS_ARG_NONNULL(1); + +/** Gets the number of samples of an Opus packet. + * @param [in] packet <tt>char*</tt>: Opus packet + * @param [in] len <tt>opus_int32</tt>: Length of packet + * @param [in] Fs <tt>opus_int32</tt>: Sampling rate in Hz. + * This must be a multiple of 400, or + * inaccurate results will be returned. + * @returns Number of samples + * @retval OPUS_BAD_ARG Insufficient data was passed to the function + * @retval OPUS_INVALID_PACKET The compressed data passed is corrupted or of an unsupported type + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT int opus_packet_get_nb_samples(const unsigned char packet[], opus_int32 len, opus_int32 Fs) OPUS_ARG_NONNULL(1); + +/** Gets the number of samples of an Opus packet. + * @param [in] dec <tt>OpusDecoder*</tt>: Decoder state + * @param [in] packet <tt>char*</tt>: Opus packet + * @param [in] len <tt>opus_int32</tt>: Length of packet + * @returns Number of samples + * @retval OPUS_BAD_ARG Insufficient data was passed to the function + * @retval OPUS_INVALID_PACKET The compressed data passed is corrupted or of an unsupported type + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT int opus_decoder_get_nb_samples(const OpusDecoder *dec, const unsigned char packet[], opus_int32 len) OPUS_ARG_NONNULL(1) OPUS_ARG_NONNULL(2); + +/** Applies soft-clipping to bring a float signal within the [-1,1] range. If + * the signal is already in that range, nothing is done. If there are values + * outside of [-1,1], then the signal is clipped as smoothly as possible to + * both fit in the range and avoid creating excessive distortion in the + * process. + * @param [in,out] pcm <tt>float*</tt>: Input PCM and modified PCM + * @param [in] frame_size <tt>int</tt> Number of samples per channel to process + * @param [in] channels <tt>int</tt>: Number of channels + * @param [in,out] softclip_mem <tt>float*</tt>: State memory for the soft clipping process (one float per channel, initialized to zero) + */ +OPUS_EXPORT void opus_pcm_soft_clip(float *pcm, int frame_size, int channels, float *softclip_mem); + + +/**@}*/ + +/** @defgroup opus_repacketizer Repacketizer + * @{ + * + * The repacketizer can be used to merge multiple Opus packets into a single + * packet or alternatively to split Opus packets that have previously been + * merged. Splitting valid Opus packets is always guaranteed to succeed, + * whereas merging valid packets only succeeds if all frames have the same + * mode, bandwidth, and frame size, and when the total duration of the merged + * packet is no more than 120 ms. The 120 ms limit comes from the + * specification and limits decoder memory requirements at a point where + * framing overhead becomes negligible. + * + * The repacketizer currently only operates on elementary Opus + * streams. It will not manipualte multistream packets successfully, except in + * the degenerate case where they consist of data from a single stream. + * + * The repacketizing process starts with creating a repacketizer state, either + * by calling opus_repacketizer_create() or by allocating the memory yourself, + * e.g., + * @code + * OpusRepacketizer *rp; + * rp = (OpusRepacketizer*)malloc(opus_repacketizer_get_size()); + * if (rp != NULL) + * opus_repacketizer_init(rp); + * @endcode + * + * Then the application should submit packets with opus_repacketizer_cat(), + * extract new packets with opus_repacketizer_out() or + * opus_repacketizer_out_range(), and then reset the state for the next set of + * input packets via opus_repacketizer_init(). + * + * For example, to split a sequence of packets into individual frames: + * @code + * unsigned char *data; + * int len; + * while (get_next_packet(&data, &len)) + * { + * unsigned char out[1276]; + * opus_int32 out_len; + * int nb_frames; + * int err; + * int i; + * err = opus_repacketizer_cat(rp, data, len); + * if (err != OPUS_OK) + * { + * release_packet(data); + * return err; + * } + * nb_frames = opus_repacketizer_get_nb_frames(rp); + * for (i = 0; i < nb_frames; i++) + * { + * out_len = opus_repacketizer_out_range(rp, i, i+1, out, sizeof(out)); + * if (out_len < 0) + * { + * release_packet(data); + * return (int)out_len; + * } + * output_next_packet(out, out_len); + * } + * opus_repacketizer_init(rp); + * release_packet(data); + * } + * @endcode + * + * Alternatively, to combine a sequence of frames into packets that each + * contain up to <code>TARGET_DURATION_MS</code> milliseconds of data: + * @code + * // The maximum number of packets with duration TARGET_DURATION_MS occurs + * // when the frame size is 2.5 ms, for a total of (TARGET_DURATION_MS*2/5) + * // packets. + * unsigned char *data[(TARGET_DURATION_MS*2/5)+1]; + * opus_int32 len[(TARGET_DURATION_MS*2/5)+1]; + * int nb_packets; + * unsigned char out[1277*(TARGET_DURATION_MS*2/2)]; + * opus_int32 out_len; + * int prev_toc; + * nb_packets = 0; + * while (get_next_packet(data+nb_packets, len+nb_packets)) + * { + * int nb_frames; + * int err; + * nb_frames = opus_packet_get_nb_frames(data[nb_packets], len[nb_packets]); + * if (nb_frames < 1) + * { + * release_packets(data, nb_packets+1); + * return nb_frames; + * } + * nb_frames += opus_repacketizer_get_nb_frames(rp); + * // If adding the next packet would exceed our target, or it has an + * // incompatible TOC sequence, output the packets we already have before + * // submitting it. + * // N.B., The nb_packets > 0 check ensures we've submitted at least one + * // packet since the last call to opus_repacketizer_init(). Otherwise a + * // single packet longer than TARGET_DURATION_MS would cause us to try to + * // output an (invalid) empty packet. It also ensures that prev_toc has + * // been set to a valid value. Additionally, len[nb_packets] > 0 is + * // guaranteed by the call to opus_packet_get_nb_frames() above, so the + * // reference to data[nb_packets][0] should be valid. + * if (nb_packets > 0 && ( + * ((prev_toc & 0xFC) != (data[nb_packets][0] & 0xFC)) || + * opus_packet_get_samples_per_frame(data[nb_packets], 48000)*nb_frames > + * TARGET_DURATION_MS*48)) + * { + * out_len = opus_repacketizer_out(rp, out, sizeof(out)); + * if (out_len < 0) + * { + * release_packets(data, nb_packets+1); + * return (int)out_len; + * } + * output_next_packet(out, out_len); + * opus_repacketizer_init(rp); + * release_packets(data, nb_packets); + * data[0] = data[nb_packets]; + * len[0] = len[nb_packets]; + * nb_packets = 0; + * } + * err = opus_repacketizer_cat(rp, data[nb_packets], len[nb_packets]); + * if (err != OPUS_OK) + * { + * release_packets(data, nb_packets+1); + * return err; + * } + * prev_toc = data[nb_packets][0]; + * nb_packets++; + * } + * // Output the final, partial packet. + * if (nb_packets > 0) + * { + * out_len = opus_repacketizer_out(rp, out, sizeof(out)); + * release_packets(data, nb_packets); + * if (out_len < 0) + * return (int)out_len; + * output_next_packet(out, out_len); + * } + * @endcode + * + * An alternate way of merging packets is to simply call opus_repacketizer_cat() + * unconditionally until it fails. At that point, the merged packet can be + * obtained with opus_repacketizer_out() and the input packet for which + * opus_repacketizer_cat() needs to be re-added to a newly reinitialized + * repacketizer state. + */ + +typedef struct OpusRepacketizer OpusRepacketizer; + +/** Gets the size of an <code>OpusRepacketizer</code> structure. + * @returns The size in bytes. + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT int opus_repacketizer_get_size(void); + +/** (Re)initializes a previously allocated repacketizer state. + * The state must be at least the size returned by opus_repacketizer_get_size(). + * This can be used for applications which use their own allocator instead of + * malloc(). + * It must also be called to reset the queue of packets waiting to be + * repacketized, which is necessary if the maximum packet duration of 120 ms + * is reached or if you wish to submit packets with a different Opus + * configuration (coding mode, audio bandwidth, frame size, or channel count). + * Failure to do so will prevent a new packet from being added with + * opus_repacketizer_cat(). + * @see opus_repacketizer_create + * @see opus_repacketizer_get_size + * @see opus_repacketizer_cat + * @param rp <tt>OpusRepacketizer*</tt>: The repacketizer state to + * (re)initialize. + * @returns A pointer to the same repacketizer state that was passed in. + */ +OPUS_EXPORT OpusRepacketizer *opus_repacketizer_init(OpusRepacketizer *rp) OPUS_ARG_NONNULL(1); + +/** Allocates memory and initializes the new repacketizer with + * opus_repacketizer_init(). + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT OpusRepacketizer *opus_repacketizer_create(void); + +/** Frees an <code>OpusRepacketizer</code> allocated by + * opus_repacketizer_create(). + * @param[in] rp <tt>OpusRepacketizer*</tt>: State to be freed. + */ +OPUS_EXPORT void opus_repacketizer_destroy(OpusRepacketizer *rp); + +/** Add a packet to the current repacketizer state. + * This packet must match the configuration of any packets already submitted + * for repacketization since the last call to opus_repacketizer_init(). + * This means that it must have the same coding mode, audio bandwidth, frame + * size, and channel count. + * This can be checked in advance by examining the top 6 bits of the first + * byte of the packet, and ensuring they match the top 6 bits of the first + * byte of any previously submitted packet. + * The total duration of audio in the repacketizer state also must not exceed + * 120 ms, the maximum duration of a single packet, after adding this packet. + * + * The contents of the current repacketizer state can be extracted into new + * packets using opus_repacketizer_out() or opus_repacketizer_out_range(). + * + * In order to add a packet with a different configuration or to add more + * audio beyond 120 ms, you must clear the repacketizer state by calling + * opus_repacketizer_init(). + * If a packet is too large to add to the current repacketizer state, no part + * of it is added, even if it contains multiple frames, some of which might + * fit. + * If you wish to be able to add parts of such packets, you should first use + * another repacketizer to split the packet into pieces and add them + * individually. + * @see opus_repacketizer_out_range + * @see opus_repacketizer_out + * @see opus_repacketizer_init + * @param rp <tt>OpusRepacketizer*</tt>: The repacketizer state to which to + * add the packet. + * @param[in] data <tt>const unsigned char*</tt>: The packet data. + * The application must ensure + * this pointer remains valid + * until the next call to + * opus_repacketizer_init() or + * opus_repacketizer_destroy(). + * @param len <tt>opus_int32</tt>: The number of bytes in the packet data. + * @returns An error code indicating whether or not the operation succeeded. + * @retval #OPUS_OK The packet's contents have been added to the repacketizer + * state. + * @retval #OPUS_INVALID_PACKET The packet did not have a valid TOC sequence, + * the packet's TOC sequence was not compatible + * with previously submitted packets (because + * the coding mode, audio bandwidth, frame size, + * or channel count did not match), or adding + * this packet would increase the total amount of + * audio stored in the repacketizer state to more + * than 120 ms. + */ +OPUS_EXPORT int opus_repacketizer_cat(OpusRepacketizer *rp, const unsigned char *data, opus_int32 len) OPUS_ARG_NONNULL(1) OPUS_ARG_NONNULL(2); + + +/** Construct a new packet from data previously submitted to the repacketizer + * state via opus_repacketizer_cat(). + * @param rp <tt>OpusRepacketizer*</tt>: The repacketizer state from which to + * construct the new packet. + * @param begin <tt>int</tt>: The index of the first frame in the current + * repacketizer state to include in the output. + * @param end <tt>int</tt>: One past the index of the last frame in the + * current repacketizer state to include in the + * output. + * @param[out] data <tt>const unsigned char*</tt>: The buffer in which to + * store the output packet. + * @param maxlen <tt>opus_int32</tt>: The maximum number of bytes to store in + * the output buffer. In order to guarantee + * success, this should be at least + * <code>1276</code> for a single frame, + * or for multiple frames, + * <code>1277*(end-begin)</code>. + * However, <code>1*(end-begin)</code> plus + * the size of all packet data submitted to + * the repacketizer since the last call to + * opus_repacketizer_init() or + * opus_repacketizer_create() is also + * sufficient, and possibly much smaller. + * @returns The total size of the output packet on success, or an error code + * on failure. + * @retval #OPUS_BAD_ARG <code>[begin,end)</code> was an invalid range of + * frames (begin < 0, begin >= end, or end > + * opus_repacketizer_get_nb_frames()). + * @retval #OPUS_BUFFER_TOO_SMALL \a maxlen was insufficient to contain the + * complete output packet. + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT opus_int32 opus_repacketizer_out_range(OpusRepacketizer *rp, int begin, int end, unsigned char *data, opus_int32 maxlen) OPUS_ARG_NONNULL(1) OPUS_ARG_NONNULL(4); + +/** Return the total number of frames contained in packet data submitted to + * the repacketizer state so far via opus_repacketizer_cat() since the last + * call to opus_repacketizer_init() or opus_repacketizer_create(). + * This defines the valid range of packets that can be extracted with + * opus_repacketizer_out_range() or opus_repacketizer_out(). + * @param rp <tt>OpusRepacketizer*</tt>: The repacketizer state containing the + * frames. + * @returns The total number of frames contained in the packet data submitted + * to the repacketizer state. + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT int opus_repacketizer_get_nb_frames(OpusRepacketizer *rp) OPUS_ARG_NONNULL(1); + +/** Construct a new packet from data previously submitted to the repacketizer + * state via opus_repacketizer_cat(). + * This is a convenience routine that returns all the data submitted so far + * in a single packet. + * It is equivalent to calling + * @code + * opus_repacketizer_out_range(rp, 0, opus_repacketizer_get_nb_frames(rp), + * data, maxlen) + * @endcode + * @param rp <tt>OpusRepacketizer*</tt>: The repacketizer state from which to + * construct the new packet. + * @param[out] data <tt>const unsigned char*</tt>: The buffer in which to + * store the output packet. + * @param maxlen <tt>opus_int32</tt>: The maximum number of bytes to store in + * the output buffer. In order to guarantee + * success, this should be at least + * <code>1277*opus_repacketizer_get_nb_frames(rp)</code>. + * However, + * <code>1*opus_repacketizer_get_nb_frames(rp)</code> + * plus the size of all packet data + * submitted to the repacketizer since the + * last call to opus_repacketizer_init() or + * opus_repacketizer_create() is also + * sufficient, and possibly much smaller. + * @returns The total size of the output packet on success, or an error code + * on failure. + * @retval #OPUS_BUFFER_TOO_SMALL \a maxlen was insufficient to contain the + * complete output packet. + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT opus_int32 opus_repacketizer_out(OpusRepacketizer *rp, unsigned char *data, opus_int32 maxlen) OPUS_ARG_NONNULL(1); + +/** Pads a given Opus packet to a larger size (possibly changing the TOC sequence). + * @param[in,out] data <tt>const unsigned char*</tt>: The buffer containing the + * packet to pad. + * @param len <tt>opus_int32</tt>: The size of the packet. + * This must be at least 1. + * @param new_len <tt>opus_int32</tt>: The desired size of the packet after padding. + * This must be at least as large as len. + * @returns an error code + * @retval #OPUS_OK \a on success. + * @retval #OPUS_BAD_ARG \a len was less than 1 or new_len was less than len. + * @retval #OPUS_INVALID_PACKET \a data did not contain a valid Opus packet. + */ +OPUS_EXPORT int opus_packet_pad(unsigned char *data, opus_int32 len, opus_int32 new_len); + +/** Remove all padding from a given Opus packet and rewrite the TOC sequence to + * minimize space usage. + * @param[in,out] data <tt>const unsigned char*</tt>: The buffer containing the + * packet to strip. + * @param len <tt>opus_int32</tt>: The size of the packet. + * This must be at least 1. + * @returns The new size of the output packet on success, or an error code + * on failure. + * @retval #OPUS_BAD_ARG \a len was less than 1. + * @retval #OPUS_INVALID_PACKET \a data did not contain a valid Opus packet. + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT opus_int32 opus_packet_unpad(unsigned char *data, opus_int32 len); + +/** Pads a given Opus multi-stream packet to a larger size (possibly changing the TOC sequence). + * @param[in,out] data <tt>const unsigned char*</tt>: The buffer containing the + * packet to pad. + * @param len <tt>opus_int32</tt>: The size of the packet. + * This must be at least 1. + * @param new_len <tt>opus_int32</tt>: The desired size of the packet after padding. + * This must be at least 1. + * @param nb_streams <tt>opus_int32</tt>: The number of streams (not channels) in the packet. + * This must be at least as large as len. + * @returns an error code + * @retval #OPUS_OK \a on success. + * @retval #OPUS_BAD_ARG \a len was less than 1. + * @retval #OPUS_INVALID_PACKET \a data did not contain a valid Opus packet. + */ +OPUS_EXPORT int opus_multistream_packet_pad(unsigned char *data, opus_int32 len, opus_int32 new_len, int nb_streams); + +/** Remove all padding from a given Opus multi-stream packet and rewrite the TOC sequence to + * minimize space usage. + * @param[in,out] data <tt>const unsigned char*</tt>: The buffer containing the + * packet to strip. + * @param len <tt>opus_int32</tt>: The size of the packet. + * This must be at least 1. + * @param nb_streams <tt>opus_int32</tt>: The number of streams (not channels) in the packet. + * This must be at least 1. + * @returns The new size of the output packet on success, or an error code + * on failure. + * @retval #OPUS_BAD_ARG \a len was less than 1 or new_len was less than len. + * @retval #OPUS_INVALID_PACKET \a data did not contain a valid Opus packet. + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT opus_int32 opus_multistream_packet_unpad(unsigned char *data, opus_int32 len, int nb_streams); + +/**@}*/ + +#ifdef __cplusplus +} +#endif + +#endif /* OPUS_H */ diff --git a/media/libopus/include/opus_custom.h b/media/libopus/include/opus_custom.h new file mode 100644 index 0000000000..2f22d4b35a --- /dev/null +++ b/media/libopus/include/opus_custom.h @@ -0,0 +1,343 @@ +/* Copyright (c) 2007-2008 CSIRO + Copyright (c) 2007-2009 Xiph.Org Foundation + Copyright (c) 2008-2012 Gregory Maxwell + Written by Jean-Marc Valin and Gregory Maxwell */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +/** + @file opus_custom.h + @brief Opus-Custom reference implementation API + */ + +#ifndef OPUS_CUSTOM_H +#define OPUS_CUSTOM_H + +#include "opus_defines.h" + +#ifdef __cplusplus +extern "C" { +#endif + +#ifdef CUSTOM_MODES +# define OPUS_CUSTOM_EXPORT OPUS_EXPORT +# define OPUS_CUSTOM_EXPORT_STATIC OPUS_EXPORT +#else +# define OPUS_CUSTOM_EXPORT +# ifdef OPUS_BUILD +# define OPUS_CUSTOM_EXPORT_STATIC static OPUS_INLINE +# else +# define OPUS_CUSTOM_EXPORT_STATIC +# endif +#endif + +/** @defgroup opus_custom Opus Custom + * @{ + * Opus Custom is an optional part of the Opus specification and + * reference implementation which uses a distinct API from the regular + * API and supports frame sizes that are not normally supported.\ Use + * of Opus Custom is discouraged for all but very special applications + * for which a frame size different from 2.5, 5, 10, or 20 ms is needed + * (for either complexity or latency reasons) and where interoperability + * is less important. + * + * In addition to the interoperability limitations the use of Opus custom + * disables a substantial chunk of the codec and generally lowers the + * quality available at a given bitrate. Normally when an application needs + * a different frame size from the codec it should buffer to match the + * sizes but this adds a small amount of delay which may be important + * in some very low latency applications. Some transports (especially + * constant rate RF transports) may also work best with frames of + * particular durations. + * + * Libopus only supports custom modes if they are enabled at compile time. + * + * The Opus Custom API is similar to the regular API but the + * @ref opus_encoder_create and @ref opus_decoder_create calls take + * an additional mode parameter which is a structure produced by + * a call to @ref opus_custom_mode_create. Both the encoder and decoder + * must create a mode using the same sample rate (fs) and frame size + * (frame size) so these parameters must either be signaled out of band + * or fixed in a particular implementation. + * + * Similar to regular Opus the custom modes support on the fly frame size + * switching, but the sizes available depend on the particular frame size in + * use. For some initial frame sizes on a single on the fly size is available. + */ + +/** Contains the state of an encoder. One encoder state is needed + for each stream. It is initialized once at the beginning of the + stream. Do *not* re-initialize the state for every frame. + @brief Encoder state + */ +typedef struct OpusCustomEncoder OpusCustomEncoder; + +/** State of the decoder. One decoder state is needed for each stream. + It is initialized once at the beginning of the stream. Do *not* + re-initialize the state for every frame. + @brief Decoder state + */ +typedef struct OpusCustomDecoder OpusCustomDecoder; + +/** The mode contains all the information necessary to create an + encoder. Both the encoder and decoder need to be initialized + with exactly the same mode, otherwise the output will be + corrupted. The mode MUST NOT BE DESTROYED until the encoders and + decoders that use it are destroyed as well. + @brief Mode configuration + */ +typedef struct OpusCustomMode OpusCustomMode; + +/** Creates a new mode struct. This will be passed to an encoder or + * decoder. The mode MUST NOT BE DESTROYED until the encoders and + * decoders that use it are destroyed as well. + * @param [in] Fs <tt>int</tt>: Sampling rate (8000 to 96000 Hz) + * @param [in] frame_size <tt>int</tt>: Number of samples (per channel) to encode in each + * packet (64 - 1024, prime factorization must contain zero or more 2s, 3s, or 5s and no other primes) + * @param [out] error <tt>int*</tt>: Returned error code (if NULL, no error will be returned) + * @return A newly created mode + */ +OPUS_CUSTOM_EXPORT OPUS_WARN_UNUSED_RESULT OpusCustomMode *opus_custom_mode_create(opus_int32 Fs, int frame_size, int *error); + +/** Destroys a mode struct. Only call this after all encoders and + * decoders using this mode are destroyed as well. + * @param [in] mode <tt>OpusCustomMode*</tt>: Mode to be freed. + */ +OPUS_CUSTOM_EXPORT void opus_custom_mode_destroy(OpusCustomMode *mode); + + +#if !defined(OPUS_BUILD) || defined(CELT_ENCODER_C) + +/* Encoder */ +/** Gets the size of an OpusCustomEncoder structure. + * @param [in] mode <tt>OpusCustomMode *</tt>: Mode configuration + * @param [in] channels <tt>int</tt>: Number of channels + * @returns size + */ +OPUS_CUSTOM_EXPORT_STATIC OPUS_WARN_UNUSED_RESULT int opus_custom_encoder_get_size( + const OpusCustomMode *mode, + int channels +) OPUS_ARG_NONNULL(1); + +# ifdef CUSTOM_MODES +/** Initializes a previously allocated encoder state + * The memory pointed to by st must be the size returned by opus_custom_encoder_get_size. + * This is intended for applications which use their own allocator instead of malloc. + * @see opus_custom_encoder_create(),opus_custom_encoder_get_size() + * To reset a previously initialized state use the OPUS_RESET_STATE CTL. + * @param [in] st <tt>OpusCustomEncoder*</tt>: Encoder state + * @param [in] mode <tt>OpusCustomMode *</tt>: Contains all the information about the characteristics of + * the stream (must be the same characteristics as used for the + * decoder) + * @param [in] channels <tt>int</tt>: Number of channels + * @return OPUS_OK Success or @ref opus_errorcodes + */ +OPUS_CUSTOM_EXPORT int opus_custom_encoder_init( + OpusCustomEncoder *st, + const OpusCustomMode *mode, + int channels +) OPUS_ARG_NONNULL(1) OPUS_ARG_NONNULL(2); +# endif +#endif + + +/** Creates a new encoder state. Each stream needs its own encoder + * state (can't be shared across simultaneous streams). + * @param [in] mode <tt>OpusCustomMode*</tt>: Contains all the information about the characteristics of + * the stream (must be the same characteristics as used for the + * decoder) + * @param [in] channels <tt>int</tt>: Number of channels + * @param [out] error <tt>int*</tt>: Returns an error code + * @return Newly created encoder state. +*/ +OPUS_CUSTOM_EXPORT OPUS_WARN_UNUSED_RESULT OpusCustomEncoder *opus_custom_encoder_create( + const OpusCustomMode *mode, + int channels, + int *error +) OPUS_ARG_NONNULL(1); + + +/** Destroys an encoder state. + * @param[in] st <tt>OpusCustomEncoder*</tt>: State to be freed. + */ +OPUS_CUSTOM_EXPORT void opus_custom_encoder_destroy(OpusCustomEncoder *st); + +/** Encodes a frame of audio. + * @param [in] st <tt>OpusCustomEncoder*</tt>: Encoder state + * @param [in] pcm <tt>float*</tt>: PCM audio in float format, with a normal range of +/-1.0. + * Samples with a range beyond +/-1.0 are supported but will + * be clipped by decoders using the integer API and should + * only be used if it is known that the far end supports + * extended dynamic range. There must be exactly + * frame_size samples per channel. + * @param [in] frame_size <tt>int</tt>: Number of samples per frame of input signal + * @param [out] compressed <tt>char *</tt>: The compressed data is written here. This may not alias pcm and must be at least maxCompressedBytes long. + * @param [in] maxCompressedBytes <tt>int</tt>: Maximum number of bytes to use for compressing the frame + * (can change from one frame to another) + * @return Number of bytes written to "compressed". + * If negative, an error has occurred (see error codes). It is IMPORTANT that + * the length returned be somehow transmitted to the decoder. Otherwise, no + * decoding is possible. + */ +OPUS_CUSTOM_EXPORT OPUS_WARN_UNUSED_RESULT int opus_custom_encode_float( + OpusCustomEncoder *st, + const float *pcm, + int frame_size, + unsigned char *compressed, + int maxCompressedBytes +) OPUS_ARG_NONNULL(1) OPUS_ARG_NONNULL(2) OPUS_ARG_NONNULL(4); + +/** Encodes a frame of audio. + * @param [in] st <tt>OpusCustomEncoder*</tt>: Encoder state + * @param [in] pcm <tt>opus_int16*</tt>: PCM audio in signed 16-bit format (native endian). + * There must be exactly frame_size samples per channel. + * @param [in] frame_size <tt>int</tt>: Number of samples per frame of input signal + * @param [out] compressed <tt>char *</tt>: The compressed data is written here. This may not alias pcm and must be at least maxCompressedBytes long. + * @param [in] maxCompressedBytes <tt>int</tt>: Maximum number of bytes to use for compressing the frame + * (can change from one frame to another) + * @return Number of bytes written to "compressed". + * If negative, an error has occurred (see error codes). It is IMPORTANT that + * the length returned be somehow transmitted to the decoder. Otherwise, no + * decoding is possible. + */ +OPUS_CUSTOM_EXPORT OPUS_WARN_UNUSED_RESULT int opus_custom_encode( + OpusCustomEncoder *st, + const opus_int16 *pcm, + int frame_size, + unsigned char *compressed, + int maxCompressedBytes +) OPUS_ARG_NONNULL(1) OPUS_ARG_NONNULL(2) OPUS_ARG_NONNULL(4); + +/** Perform a CTL function on an Opus custom encoder. + * + * Generally the request and subsequent arguments are generated + * by a convenience macro. + * @see opus_encoderctls + */ +OPUS_CUSTOM_EXPORT int opus_custom_encoder_ctl(OpusCustomEncoder * OPUS_RESTRICT st, int request, ...) OPUS_ARG_NONNULL(1); + + +#if !defined(OPUS_BUILD) || defined(CELT_DECODER_C) +/* Decoder */ + +/** Gets the size of an OpusCustomDecoder structure. + * @param [in] mode <tt>OpusCustomMode *</tt>: Mode configuration + * @param [in] channels <tt>int</tt>: Number of channels + * @returns size + */ +OPUS_CUSTOM_EXPORT_STATIC OPUS_WARN_UNUSED_RESULT int opus_custom_decoder_get_size( + const OpusCustomMode *mode, + int channels +) OPUS_ARG_NONNULL(1); + +/** Initializes a previously allocated decoder state + * The memory pointed to by st must be the size returned by opus_custom_decoder_get_size. + * This is intended for applications which use their own allocator instead of malloc. + * @see opus_custom_decoder_create(),opus_custom_decoder_get_size() + * To reset a previously initialized state use the OPUS_RESET_STATE CTL. + * @param [in] st <tt>OpusCustomDecoder*</tt>: Decoder state + * @param [in] mode <tt>OpusCustomMode *</tt>: Contains all the information about the characteristics of + * the stream (must be the same characteristics as used for the + * encoder) + * @param [in] channels <tt>int</tt>: Number of channels + * @return OPUS_OK Success or @ref opus_errorcodes + */ +OPUS_CUSTOM_EXPORT_STATIC int opus_custom_decoder_init( + OpusCustomDecoder *st, + const OpusCustomMode *mode, + int channels +) OPUS_ARG_NONNULL(1) OPUS_ARG_NONNULL(2); + +#endif + + +/** Creates a new decoder state. Each stream needs its own decoder state (can't + * be shared across simultaneous streams). + * @param [in] mode <tt>OpusCustomMode</tt>: Contains all the information about the characteristics of the + * stream (must be the same characteristics as used for the encoder) + * @param [in] channels <tt>int</tt>: Number of channels + * @param [out] error <tt>int*</tt>: Returns an error code + * @return Newly created decoder state. + */ +OPUS_CUSTOM_EXPORT OPUS_WARN_UNUSED_RESULT OpusCustomDecoder *opus_custom_decoder_create( + const OpusCustomMode *mode, + int channels, + int *error +) OPUS_ARG_NONNULL(1); + +/** Destroys a decoder state. + * @param[in] st <tt>OpusCustomDecoder*</tt>: State to be freed. + */ +OPUS_CUSTOM_EXPORT void opus_custom_decoder_destroy(OpusCustomDecoder *st); + +/** Decode an opus custom frame with floating point output + * @param [in] st <tt>OpusCustomDecoder*</tt>: Decoder state + * @param [in] data <tt>char*</tt>: Input payload. Use a NULL pointer to indicate packet loss + * @param [in] len <tt>int</tt>: Number of bytes in payload + * @param [out] pcm <tt>float*</tt>: Output signal (interleaved if 2 channels). length + * is frame_size*channels*sizeof(float) + * @param [in] frame_size Number of samples per channel of available space in *pcm. + * @returns Number of decoded samples or @ref opus_errorcodes + */ +OPUS_CUSTOM_EXPORT OPUS_WARN_UNUSED_RESULT int opus_custom_decode_float( + OpusCustomDecoder *st, + const unsigned char *data, + int len, + float *pcm, + int frame_size +) OPUS_ARG_NONNULL(1) OPUS_ARG_NONNULL(4); + +/** Decode an opus custom frame + * @param [in] st <tt>OpusCustomDecoder*</tt>: Decoder state + * @param [in] data <tt>char*</tt>: Input payload. Use a NULL pointer to indicate packet loss + * @param [in] len <tt>int</tt>: Number of bytes in payload + * @param [out] pcm <tt>opus_int16*</tt>: Output signal (interleaved if 2 channels). length + * is frame_size*channels*sizeof(opus_int16) + * @param [in] frame_size Number of samples per channel of available space in *pcm. + * @returns Number of decoded samples or @ref opus_errorcodes + */ +OPUS_CUSTOM_EXPORT OPUS_WARN_UNUSED_RESULT int opus_custom_decode( + OpusCustomDecoder *st, + const unsigned char *data, + int len, + opus_int16 *pcm, + int frame_size +) OPUS_ARG_NONNULL(1) OPUS_ARG_NONNULL(4); + +/** Perform a CTL function on an Opus custom decoder. + * + * Generally the request and subsequent arguments are generated + * by a convenience macro. + * @see opus_genericctls + */ +OPUS_CUSTOM_EXPORT int opus_custom_decoder_ctl(OpusCustomDecoder * OPUS_RESTRICT st, int request, ...) OPUS_ARG_NONNULL(1); + +/**@}*/ + +#ifdef __cplusplus +} +#endif + +#endif /* OPUS_CUSTOM_H */ diff --git a/media/libopus/include/opus_defines.h b/media/libopus/include/opus_defines.h new file mode 100644 index 0000000000..94b9e0d9fc --- /dev/null +++ b/media/libopus/include/opus_defines.h @@ -0,0 +1,801 @@ +/* Copyright (c) 2010-2011 Xiph.Org Foundation, Skype Limited + Written by Jean-Marc Valin and Koen Vos */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +/** + * @file opus_defines.h + * @brief Opus reference implementation constants + */ + +#ifndef OPUS_DEFINES_H +#define OPUS_DEFINES_H + +#include "opus_types.h" + +#ifdef __cplusplus +extern "C" { +#endif + +/** @defgroup opus_errorcodes Error codes + * @{ + */ +/** No error @hideinitializer*/ +#define OPUS_OK 0 +/** One or more invalid/out of range arguments @hideinitializer*/ +#define OPUS_BAD_ARG -1 +/** Not enough bytes allocated in the buffer @hideinitializer*/ +#define OPUS_BUFFER_TOO_SMALL -2 +/** An internal error was detected @hideinitializer*/ +#define OPUS_INTERNAL_ERROR -3 +/** The compressed data passed is corrupted @hideinitializer*/ +#define OPUS_INVALID_PACKET -4 +/** Invalid/unsupported request number @hideinitializer*/ +#define OPUS_UNIMPLEMENTED -5 +/** An encoder or decoder structure is invalid or already freed @hideinitializer*/ +#define OPUS_INVALID_STATE -6 +/** Memory allocation has failed @hideinitializer*/ +#define OPUS_ALLOC_FAIL -7 +/**@}*/ + +/** @cond OPUS_INTERNAL_DOC */ +/**Export control for opus functions */ + +#ifndef OPUS_EXPORT +# if defined(_WIN32) +# if defined(OPUS_BUILD) && defined(DLL_EXPORT) +# define OPUS_EXPORT __declspec(dllexport) +# else +# define OPUS_EXPORT +# endif +# elif defined(__GNUC__) && defined(OPUS_BUILD) +# define OPUS_EXPORT __attribute__ ((visibility ("default"))) +# else +# define OPUS_EXPORT +# endif +#endif + +# if !defined(OPUS_GNUC_PREREQ) +# if defined(__GNUC__)&&defined(__GNUC_MINOR__) +# define OPUS_GNUC_PREREQ(_maj,_min) \ + ((__GNUC__<<16)+__GNUC_MINOR__>=((_maj)<<16)+(_min)) +# else +# define OPUS_GNUC_PREREQ(_maj,_min) 0 +# endif +# endif + +#if (!defined(__STDC_VERSION__) || (__STDC_VERSION__ < 199901L) ) +# if OPUS_GNUC_PREREQ(3,0) +# define OPUS_RESTRICT __restrict__ +# elif (defined(_MSC_VER) && _MSC_VER >= 1400) +# define OPUS_RESTRICT __restrict +# else +# define OPUS_RESTRICT +# endif +#else +# define OPUS_RESTRICT restrict +#endif + +#if (!defined(__STDC_VERSION__) || (__STDC_VERSION__ < 199901L) ) +# if OPUS_GNUC_PREREQ(2,7) +# define OPUS_INLINE __inline__ +# elif (defined(_MSC_VER)) +# define OPUS_INLINE __inline +# else +# define OPUS_INLINE +# endif +#else +# define OPUS_INLINE inline +#endif + +/**Warning attributes for opus functions + * NONNULL is not used in OPUS_BUILD to avoid the compiler optimizing out + * some paranoid null checks. */ +#if defined(__GNUC__) && OPUS_GNUC_PREREQ(3, 4) +# define OPUS_WARN_UNUSED_RESULT __attribute__ ((__warn_unused_result__)) +#else +# define OPUS_WARN_UNUSED_RESULT +#endif +#if !defined(OPUS_BUILD) && defined(__GNUC__) && OPUS_GNUC_PREREQ(3, 4) +# define OPUS_ARG_NONNULL(_x) __attribute__ ((__nonnull__(_x))) +#else +# define OPUS_ARG_NONNULL(_x) +#endif + +/** These are the actual Encoder CTL ID numbers. + * They should not be used directly by applications. + * In general, SETs should be even and GETs should be odd.*/ +#define OPUS_SET_APPLICATION_REQUEST 4000 +#define OPUS_GET_APPLICATION_REQUEST 4001 +#define OPUS_SET_BITRATE_REQUEST 4002 +#define OPUS_GET_BITRATE_REQUEST 4003 +#define OPUS_SET_MAX_BANDWIDTH_REQUEST 4004 +#define OPUS_GET_MAX_BANDWIDTH_REQUEST 4005 +#define OPUS_SET_VBR_REQUEST 4006 +#define OPUS_GET_VBR_REQUEST 4007 +#define OPUS_SET_BANDWIDTH_REQUEST 4008 +#define OPUS_GET_BANDWIDTH_REQUEST 4009 +#define OPUS_SET_COMPLEXITY_REQUEST 4010 +#define OPUS_GET_COMPLEXITY_REQUEST 4011 +#define OPUS_SET_INBAND_FEC_REQUEST 4012 +#define OPUS_GET_INBAND_FEC_REQUEST 4013 +#define OPUS_SET_PACKET_LOSS_PERC_REQUEST 4014 +#define OPUS_GET_PACKET_LOSS_PERC_REQUEST 4015 +#define OPUS_SET_DTX_REQUEST 4016 +#define OPUS_GET_DTX_REQUEST 4017 +#define OPUS_SET_VBR_CONSTRAINT_REQUEST 4020 +#define OPUS_GET_VBR_CONSTRAINT_REQUEST 4021 +#define OPUS_SET_FORCE_CHANNELS_REQUEST 4022 +#define OPUS_GET_FORCE_CHANNELS_REQUEST 4023 +#define OPUS_SET_SIGNAL_REQUEST 4024 +#define OPUS_GET_SIGNAL_REQUEST 4025 +#define OPUS_GET_LOOKAHEAD_REQUEST 4027 +/* #define OPUS_RESET_STATE 4028 */ +#define OPUS_GET_SAMPLE_RATE_REQUEST 4029 +#define OPUS_GET_FINAL_RANGE_REQUEST 4031 +#define OPUS_GET_PITCH_REQUEST 4033 +#define OPUS_SET_GAIN_REQUEST 4034 +#define OPUS_GET_GAIN_REQUEST 4045 /* Should have been 4035 */ +#define OPUS_SET_LSB_DEPTH_REQUEST 4036 +#define OPUS_GET_LSB_DEPTH_REQUEST 4037 +#define OPUS_GET_LAST_PACKET_DURATION_REQUEST 4039 +#define OPUS_SET_EXPERT_FRAME_DURATION_REQUEST 4040 +#define OPUS_GET_EXPERT_FRAME_DURATION_REQUEST 4041 +#define OPUS_SET_PREDICTION_DISABLED_REQUEST 4042 +#define OPUS_GET_PREDICTION_DISABLED_REQUEST 4043 +/* Don't use 4045, it's already taken by OPUS_GET_GAIN_REQUEST */ +#define OPUS_SET_PHASE_INVERSION_DISABLED_REQUEST 4046 +#define OPUS_GET_PHASE_INVERSION_DISABLED_REQUEST 4047 +#define OPUS_GET_IN_DTX_REQUEST 4049 + +/** Defines for the presence of extended APIs. */ +#define OPUS_HAVE_OPUS_PROJECTION_H + +/* Macros to trigger compilation errors when the wrong types are provided to a CTL */ +#define __opus_check_int(x) (((void)((x) == (opus_int32)0)), (opus_int32)(x)) +#define __opus_check_int_ptr(ptr) ((ptr) + ((ptr) - (opus_int32*)(ptr))) +#define __opus_check_uint_ptr(ptr) ((ptr) + ((ptr) - (opus_uint32*)(ptr))) +#define __opus_check_val16_ptr(ptr) ((ptr) + ((ptr) - (opus_val16*)(ptr))) +/** @endcond */ + +/** @defgroup opus_ctlvalues Pre-defined values for CTL interface + * @see opus_genericctls, opus_encoderctls + * @{ + */ +/* Values for the various encoder CTLs */ +#define OPUS_AUTO -1000 /**<Auto/default setting @hideinitializer*/ +#define OPUS_BITRATE_MAX -1 /**<Maximum bitrate @hideinitializer*/ + +/** Best for most VoIP/videoconference applications where listening quality and intelligibility matter most + * @hideinitializer */ +#define OPUS_APPLICATION_VOIP 2048 +/** Best for broadcast/high-fidelity application where the decoded audio should be as close as possible to the input + * @hideinitializer */ +#define OPUS_APPLICATION_AUDIO 2049 +/** Only use when lowest-achievable latency is what matters most. Voice-optimized modes cannot be used. + * @hideinitializer */ +#define OPUS_APPLICATION_RESTRICTED_LOWDELAY 2051 + +#define OPUS_SIGNAL_VOICE 3001 /**< Signal being encoded is voice */ +#define OPUS_SIGNAL_MUSIC 3002 /**< Signal being encoded is music */ +#define OPUS_BANDWIDTH_NARROWBAND 1101 /**< 4 kHz bandpass @hideinitializer*/ +#define OPUS_BANDWIDTH_MEDIUMBAND 1102 /**< 6 kHz bandpass @hideinitializer*/ +#define OPUS_BANDWIDTH_WIDEBAND 1103 /**< 8 kHz bandpass @hideinitializer*/ +#define OPUS_BANDWIDTH_SUPERWIDEBAND 1104 /**<12 kHz bandpass @hideinitializer*/ +#define OPUS_BANDWIDTH_FULLBAND 1105 /**<20 kHz bandpass @hideinitializer*/ + +#define OPUS_FRAMESIZE_ARG 5000 /**< Select frame size from the argument (default) */ +#define OPUS_FRAMESIZE_2_5_MS 5001 /**< Use 2.5 ms frames */ +#define OPUS_FRAMESIZE_5_MS 5002 /**< Use 5 ms frames */ +#define OPUS_FRAMESIZE_10_MS 5003 /**< Use 10 ms frames */ +#define OPUS_FRAMESIZE_20_MS 5004 /**< Use 20 ms frames */ +#define OPUS_FRAMESIZE_40_MS 5005 /**< Use 40 ms frames */ +#define OPUS_FRAMESIZE_60_MS 5006 /**< Use 60 ms frames */ +#define OPUS_FRAMESIZE_80_MS 5007 /**< Use 80 ms frames */ +#define OPUS_FRAMESIZE_100_MS 5008 /**< Use 100 ms frames */ +#define OPUS_FRAMESIZE_120_MS 5009 /**< Use 120 ms frames */ + +/**@}*/ + + +/** @defgroup opus_encoderctls Encoder related CTLs + * + * These are convenience macros for use with the \c opus_encode_ctl + * interface. They are used to generate the appropriate series of + * arguments for that call, passing the correct type, size and so + * on as expected for each particular request. + * + * Some usage examples: + * + * @code + * int ret; + * ret = opus_encoder_ctl(enc_ctx, OPUS_SET_BANDWIDTH(OPUS_AUTO)); + * if (ret != OPUS_OK) return ret; + * + * opus_int32 rate; + * opus_encoder_ctl(enc_ctx, OPUS_GET_BANDWIDTH(&rate)); + * + * opus_encoder_ctl(enc_ctx, OPUS_RESET_STATE); + * @endcode + * + * @see opus_genericctls, opus_encoder + * @{ + */ + +/** Configures the encoder's computational complexity. + * The supported range is 0-10 inclusive with 10 representing the highest complexity. + * @see OPUS_GET_COMPLEXITY + * @param[in] x <tt>opus_int32</tt>: Allowed values: 0-10, inclusive. + * + * @hideinitializer */ +#define OPUS_SET_COMPLEXITY(x) OPUS_SET_COMPLEXITY_REQUEST, __opus_check_int(x) +/** Gets the encoder's complexity configuration. + * @see OPUS_SET_COMPLEXITY + * @param[out] x <tt>opus_int32 *</tt>: Returns a value in the range 0-10, + * inclusive. + * @hideinitializer */ +#define OPUS_GET_COMPLEXITY(x) OPUS_GET_COMPLEXITY_REQUEST, __opus_check_int_ptr(x) + +/** Configures the bitrate in the encoder. + * Rates from 500 to 512000 bits per second are meaningful, as well as the + * special values #OPUS_AUTO and #OPUS_BITRATE_MAX. + * The value #OPUS_BITRATE_MAX can be used to cause the codec to use as much + * rate as it can, which is useful for controlling the rate by adjusting the + * output buffer size. + * @see OPUS_GET_BITRATE + * @param[in] x <tt>opus_int32</tt>: Bitrate in bits per second. The default + * is determined based on the number of + * channels and the input sampling rate. + * @hideinitializer */ +#define OPUS_SET_BITRATE(x) OPUS_SET_BITRATE_REQUEST, __opus_check_int(x) +/** Gets the encoder's bitrate configuration. + * @see OPUS_SET_BITRATE + * @param[out] x <tt>opus_int32 *</tt>: Returns the bitrate in bits per second. + * The default is determined based on the + * number of channels and the input + * sampling rate. + * @hideinitializer */ +#define OPUS_GET_BITRATE(x) OPUS_GET_BITRATE_REQUEST, __opus_check_int_ptr(x) + +/** Enables or disables variable bitrate (VBR) in the encoder. + * The configured bitrate may not be met exactly because frames must + * be an integer number of bytes in length. + * @see OPUS_GET_VBR + * @see OPUS_SET_VBR_CONSTRAINT + * @param[in] x <tt>opus_int32</tt>: Allowed values: + * <dl> + * <dt>0</dt><dd>Hard CBR. For LPC/hybrid modes at very low bit-rate, this can + * cause noticeable quality degradation.</dd> + * <dt>1</dt><dd>VBR (default). The exact type of VBR is controlled by + * #OPUS_SET_VBR_CONSTRAINT.</dd> + * </dl> + * @hideinitializer */ +#define OPUS_SET_VBR(x) OPUS_SET_VBR_REQUEST, __opus_check_int(x) +/** Determine if variable bitrate (VBR) is enabled in the encoder. + * @see OPUS_SET_VBR + * @see OPUS_GET_VBR_CONSTRAINT + * @param[out] x <tt>opus_int32 *</tt>: Returns one of the following values: + * <dl> + * <dt>0</dt><dd>Hard CBR.</dd> + * <dt>1</dt><dd>VBR (default). The exact type of VBR may be retrieved via + * #OPUS_GET_VBR_CONSTRAINT.</dd> + * </dl> + * @hideinitializer */ +#define OPUS_GET_VBR(x) OPUS_GET_VBR_REQUEST, __opus_check_int_ptr(x) + +/** Enables or disables constrained VBR in the encoder. + * This setting is ignored when the encoder is in CBR mode. + * @warning Only the MDCT mode of Opus currently heeds the constraint. + * Speech mode ignores it completely, hybrid mode may fail to obey it + * if the LPC layer uses more bitrate than the constraint would have + * permitted. + * @see OPUS_GET_VBR_CONSTRAINT + * @see OPUS_SET_VBR + * @param[in] x <tt>opus_int32</tt>: Allowed values: + * <dl> + * <dt>0</dt><dd>Unconstrained VBR.</dd> + * <dt>1</dt><dd>Constrained VBR (default). This creates a maximum of one + * frame of buffering delay assuming a transport with a + * serialization speed of the nominal bitrate.</dd> + * </dl> + * @hideinitializer */ +#define OPUS_SET_VBR_CONSTRAINT(x) OPUS_SET_VBR_CONSTRAINT_REQUEST, __opus_check_int(x) +/** Determine if constrained VBR is enabled in the encoder. + * @see OPUS_SET_VBR_CONSTRAINT + * @see OPUS_GET_VBR + * @param[out] x <tt>opus_int32 *</tt>: Returns one of the following values: + * <dl> + * <dt>0</dt><dd>Unconstrained VBR.</dd> + * <dt>1</dt><dd>Constrained VBR (default).</dd> + * </dl> + * @hideinitializer */ +#define OPUS_GET_VBR_CONSTRAINT(x) OPUS_GET_VBR_CONSTRAINT_REQUEST, __opus_check_int_ptr(x) + +/** Configures mono/stereo forcing in the encoder. + * This can force the encoder to produce packets encoded as either mono or + * stereo, regardless of the format of the input audio. This is useful when + * the caller knows that the input signal is currently a mono source embedded + * in a stereo stream. + * @see OPUS_GET_FORCE_CHANNELS + * @param[in] x <tt>opus_int32</tt>: Allowed values: + * <dl> + * <dt>#OPUS_AUTO</dt><dd>Not forced (default)</dd> + * <dt>1</dt> <dd>Forced mono</dd> + * <dt>2</dt> <dd>Forced stereo</dd> + * </dl> + * @hideinitializer */ +#define OPUS_SET_FORCE_CHANNELS(x) OPUS_SET_FORCE_CHANNELS_REQUEST, __opus_check_int(x) +/** Gets the encoder's forced channel configuration. + * @see OPUS_SET_FORCE_CHANNELS + * @param[out] x <tt>opus_int32 *</tt>: + * <dl> + * <dt>#OPUS_AUTO</dt><dd>Not forced (default)</dd> + * <dt>1</dt> <dd>Forced mono</dd> + * <dt>2</dt> <dd>Forced stereo</dd> + * </dl> + * @hideinitializer */ +#define OPUS_GET_FORCE_CHANNELS(x) OPUS_GET_FORCE_CHANNELS_REQUEST, __opus_check_int_ptr(x) + +/** Configures the maximum bandpass that the encoder will select automatically. + * Applications should normally use this instead of #OPUS_SET_BANDWIDTH + * (leaving that set to the default, #OPUS_AUTO). This allows the + * application to set an upper bound based on the type of input it is + * providing, but still gives the encoder the freedom to reduce the bandpass + * when the bitrate becomes too low, for better overall quality. + * @see OPUS_GET_MAX_BANDWIDTH + * @param[in] x <tt>opus_int32</tt>: Allowed values: + * <dl> + * <dt>OPUS_BANDWIDTH_NARROWBAND</dt> <dd>4 kHz passband</dd> + * <dt>OPUS_BANDWIDTH_MEDIUMBAND</dt> <dd>6 kHz passband</dd> + * <dt>OPUS_BANDWIDTH_WIDEBAND</dt> <dd>8 kHz passband</dd> + * <dt>OPUS_BANDWIDTH_SUPERWIDEBAND</dt><dd>12 kHz passband</dd> + * <dt>OPUS_BANDWIDTH_FULLBAND</dt> <dd>20 kHz passband (default)</dd> + * </dl> + * @hideinitializer */ +#define OPUS_SET_MAX_BANDWIDTH(x) OPUS_SET_MAX_BANDWIDTH_REQUEST, __opus_check_int(x) + +/** Gets the encoder's configured maximum allowed bandpass. + * @see OPUS_SET_MAX_BANDWIDTH + * @param[out] x <tt>opus_int32 *</tt>: Allowed values: + * <dl> + * <dt>#OPUS_BANDWIDTH_NARROWBAND</dt> <dd>4 kHz passband</dd> + * <dt>#OPUS_BANDWIDTH_MEDIUMBAND</dt> <dd>6 kHz passband</dd> + * <dt>#OPUS_BANDWIDTH_WIDEBAND</dt> <dd>8 kHz passband</dd> + * <dt>#OPUS_BANDWIDTH_SUPERWIDEBAND</dt><dd>12 kHz passband</dd> + * <dt>#OPUS_BANDWIDTH_FULLBAND</dt> <dd>20 kHz passband (default)</dd> + * </dl> + * @hideinitializer */ +#define OPUS_GET_MAX_BANDWIDTH(x) OPUS_GET_MAX_BANDWIDTH_REQUEST, __opus_check_int_ptr(x) + +/** Sets the encoder's bandpass to a specific value. + * This prevents the encoder from automatically selecting the bandpass based + * on the available bitrate. If an application knows the bandpass of the input + * audio it is providing, it should normally use #OPUS_SET_MAX_BANDWIDTH + * instead, which still gives the encoder the freedom to reduce the bandpass + * when the bitrate becomes too low, for better overall quality. + * @see OPUS_GET_BANDWIDTH + * @param[in] x <tt>opus_int32</tt>: Allowed values: + * <dl> + * <dt>#OPUS_AUTO</dt> <dd>(default)</dd> + * <dt>#OPUS_BANDWIDTH_NARROWBAND</dt> <dd>4 kHz passband</dd> + * <dt>#OPUS_BANDWIDTH_MEDIUMBAND</dt> <dd>6 kHz passband</dd> + * <dt>#OPUS_BANDWIDTH_WIDEBAND</dt> <dd>8 kHz passband</dd> + * <dt>#OPUS_BANDWIDTH_SUPERWIDEBAND</dt><dd>12 kHz passband</dd> + * <dt>#OPUS_BANDWIDTH_FULLBAND</dt> <dd>20 kHz passband</dd> + * </dl> + * @hideinitializer */ +#define OPUS_SET_BANDWIDTH(x) OPUS_SET_BANDWIDTH_REQUEST, __opus_check_int(x) + +/** Configures the type of signal being encoded. + * This is a hint which helps the encoder's mode selection. + * @see OPUS_GET_SIGNAL + * @param[in] x <tt>opus_int32</tt>: Allowed values: + * <dl> + * <dt>#OPUS_AUTO</dt> <dd>(default)</dd> + * <dt>#OPUS_SIGNAL_VOICE</dt><dd>Bias thresholds towards choosing LPC or Hybrid modes.</dd> + * <dt>#OPUS_SIGNAL_MUSIC</dt><dd>Bias thresholds towards choosing MDCT modes.</dd> + * </dl> + * @hideinitializer */ +#define OPUS_SET_SIGNAL(x) OPUS_SET_SIGNAL_REQUEST, __opus_check_int(x) +/** Gets the encoder's configured signal type. + * @see OPUS_SET_SIGNAL + * @param[out] x <tt>opus_int32 *</tt>: Returns one of the following values: + * <dl> + * <dt>#OPUS_AUTO</dt> <dd>(default)</dd> + * <dt>#OPUS_SIGNAL_VOICE</dt><dd>Bias thresholds towards choosing LPC or Hybrid modes.</dd> + * <dt>#OPUS_SIGNAL_MUSIC</dt><dd>Bias thresholds towards choosing MDCT modes.</dd> + * </dl> + * @hideinitializer */ +#define OPUS_GET_SIGNAL(x) OPUS_GET_SIGNAL_REQUEST, __opus_check_int_ptr(x) + + +/** Configures the encoder's intended application. + * The initial value is a mandatory argument to the encoder_create function. + * @see OPUS_GET_APPLICATION + * @param[in] x <tt>opus_int32</tt>: Returns one of the following values: + * <dl> + * <dt>#OPUS_APPLICATION_VOIP</dt> + * <dd>Process signal for improved speech intelligibility.</dd> + * <dt>#OPUS_APPLICATION_AUDIO</dt> + * <dd>Favor faithfulness to the original input.</dd> + * <dt>#OPUS_APPLICATION_RESTRICTED_LOWDELAY</dt> + * <dd>Configure the minimum possible coding delay by disabling certain modes + * of operation.</dd> + * </dl> + * @hideinitializer */ +#define OPUS_SET_APPLICATION(x) OPUS_SET_APPLICATION_REQUEST, __opus_check_int(x) +/** Gets the encoder's configured application. + * @see OPUS_SET_APPLICATION + * @param[out] x <tt>opus_int32 *</tt>: Returns one of the following values: + * <dl> + * <dt>#OPUS_APPLICATION_VOIP</dt> + * <dd>Process signal for improved speech intelligibility.</dd> + * <dt>#OPUS_APPLICATION_AUDIO</dt> + * <dd>Favor faithfulness to the original input.</dd> + * <dt>#OPUS_APPLICATION_RESTRICTED_LOWDELAY</dt> + * <dd>Configure the minimum possible coding delay by disabling certain modes + * of operation.</dd> + * </dl> + * @hideinitializer */ +#define OPUS_GET_APPLICATION(x) OPUS_GET_APPLICATION_REQUEST, __opus_check_int_ptr(x) + +/** Gets the total samples of delay added by the entire codec. + * This can be queried by the encoder and then the provided number of samples can be + * skipped on from the start of the decoder's output to provide time aligned input + * and output. From the perspective of a decoding application the real data begins this many + * samples late. + * + * The decoder contribution to this delay is identical for all decoders, but the + * encoder portion of the delay may vary from implementation to implementation, + * version to version, or even depend on the encoder's initial configuration. + * Applications needing delay compensation should call this CTL rather than + * hard-coding a value. + * @param[out] x <tt>opus_int32 *</tt>: Number of lookahead samples + * @hideinitializer */ +#define OPUS_GET_LOOKAHEAD(x) OPUS_GET_LOOKAHEAD_REQUEST, __opus_check_int_ptr(x) + +/** Configures the encoder's use of inband forward error correction (FEC). + * @note This is only applicable to the LPC layer + * @see OPUS_GET_INBAND_FEC + * @param[in] x <tt>opus_int32</tt>: Allowed values: + * <dl> + * <dt>0</dt><dd>Disable inband FEC (default).</dd> + * <dt>1</dt><dd>Inband FEC enabled. If the packet loss rate is sufficiently high, Opus will automatically switch to SILK even at high rates to enable use of that FEC.</dd> + * <dt>2</dt><dd>Inband FEC enabled, but does not necessarily switch to SILK if we have music.</dd> + * </dl> + * @hideinitializer */ +#define OPUS_SET_INBAND_FEC(x) OPUS_SET_INBAND_FEC_REQUEST, __opus_check_int(x) +/** Gets encoder's configured use of inband forward error correction. + * @see OPUS_SET_INBAND_FEC + * @param[out] x <tt>opus_int32 *</tt>: Returns one of the following values: + * <dl> + * <dt>0</dt><dd>Inband FEC disabled (default).</dd> + * <dt>1</dt><dd>Inband FEC enabled. If the packet loss rate is sufficiently high, Opus will automatically switch to SILK even at high rates to enable use of that FEC.</dd> + * <dt>2</dt><dd>Inband FEC enabled, but does not necessarily switch to SILK if we have music.</dd> + * </dl> + * @hideinitializer */ +#define OPUS_GET_INBAND_FEC(x) OPUS_GET_INBAND_FEC_REQUEST, __opus_check_int_ptr(x) + +/** Configures the encoder's expected packet loss percentage. + * Higher values trigger progressively more loss resistant behavior in the encoder + * at the expense of quality at a given bitrate in the absence of packet loss, but + * greater quality under loss. + * @see OPUS_GET_PACKET_LOSS_PERC + * @param[in] x <tt>opus_int32</tt>: Loss percentage in the range 0-100, inclusive (default: 0). + * @hideinitializer */ +#define OPUS_SET_PACKET_LOSS_PERC(x) OPUS_SET_PACKET_LOSS_PERC_REQUEST, __opus_check_int(x) +/** Gets the encoder's configured packet loss percentage. + * @see OPUS_SET_PACKET_LOSS_PERC + * @param[out] x <tt>opus_int32 *</tt>: Returns the configured loss percentage + * in the range 0-100, inclusive (default: 0). + * @hideinitializer */ +#define OPUS_GET_PACKET_LOSS_PERC(x) OPUS_GET_PACKET_LOSS_PERC_REQUEST, __opus_check_int_ptr(x) + +/** Configures the encoder's use of discontinuous transmission (DTX). + * @note This is only applicable to the LPC layer + * @see OPUS_GET_DTX + * @param[in] x <tt>opus_int32</tt>: Allowed values: + * <dl> + * <dt>0</dt><dd>Disable DTX (default).</dd> + * <dt>1</dt><dd>Enabled DTX.</dd> + * </dl> + * @hideinitializer */ +#define OPUS_SET_DTX(x) OPUS_SET_DTX_REQUEST, __opus_check_int(x) +/** Gets encoder's configured use of discontinuous transmission. + * @see OPUS_SET_DTX + * @param[out] x <tt>opus_int32 *</tt>: Returns one of the following values: + * <dl> + * <dt>0</dt><dd>DTX disabled (default).</dd> + * <dt>1</dt><dd>DTX enabled.</dd> + * </dl> + * @hideinitializer */ +#define OPUS_GET_DTX(x) OPUS_GET_DTX_REQUEST, __opus_check_int_ptr(x) +/** Configures the depth of signal being encoded. + * + * This is a hint which helps the encoder identify silence and near-silence. + * It represents the number of significant bits of linear intensity below + * which the signal contains ignorable quantization or other noise. + * + * For example, OPUS_SET_LSB_DEPTH(14) would be an appropriate setting + * for G.711 u-law input. OPUS_SET_LSB_DEPTH(16) would be appropriate + * for 16-bit linear pcm input with opus_encode_float(). + * + * When using opus_encode() instead of opus_encode_float(), or when libopus + * is compiled for fixed-point, the encoder uses the minimum of the value + * set here and the value 16. + * + * @see OPUS_GET_LSB_DEPTH + * @param[in] x <tt>opus_int32</tt>: Input precision in bits, between 8 and 24 + * (default: 24). + * @hideinitializer */ +#define OPUS_SET_LSB_DEPTH(x) OPUS_SET_LSB_DEPTH_REQUEST, __opus_check_int(x) +/** Gets the encoder's configured signal depth. + * @see OPUS_SET_LSB_DEPTH + * @param[out] x <tt>opus_int32 *</tt>: Input precision in bits, between 8 and + * 24 (default: 24). + * @hideinitializer */ +#define OPUS_GET_LSB_DEPTH(x) OPUS_GET_LSB_DEPTH_REQUEST, __opus_check_int_ptr(x) + +/** Configures the encoder's use of variable duration frames. + * When variable duration is enabled, the encoder is free to use a shorter frame + * size than the one requested in the opus_encode*() call. + * It is then the user's responsibility + * to verify how much audio was encoded by checking the ToC byte of the encoded + * packet. The part of the audio that was not encoded needs to be resent to the + * encoder for the next call. Do not use this option unless you <b>really</b> + * know what you are doing. + * @see OPUS_GET_EXPERT_FRAME_DURATION + * @param[in] x <tt>opus_int32</tt>: Allowed values: + * <dl> + * <dt>OPUS_FRAMESIZE_ARG</dt><dd>Select frame size from the argument (default).</dd> + * <dt>OPUS_FRAMESIZE_2_5_MS</dt><dd>Use 2.5 ms frames.</dd> + * <dt>OPUS_FRAMESIZE_5_MS</dt><dd>Use 5 ms frames.</dd> + * <dt>OPUS_FRAMESIZE_10_MS</dt><dd>Use 10 ms frames.</dd> + * <dt>OPUS_FRAMESIZE_20_MS</dt><dd>Use 20 ms frames.</dd> + * <dt>OPUS_FRAMESIZE_40_MS</dt><dd>Use 40 ms frames.</dd> + * <dt>OPUS_FRAMESIZE_60_MS</dt><dd>Use 60 ms frames.</dd> + * <dt>OPUS_FRAMESIZE_80_MS</dt><dd>Use 80 ms frames.</dd> + * <dt>OPUS_FRAMESIZE_100_MS</dt><dd>Use 100 ms frames.</dd> + * <dt>OPUS_FRAMESIZE_120_MS</dt><dd>Use 120 ms frames.</dd> + * </dl> + * @hideinitializer */ +#define OPUS_SET_EXPERT_FRAME_DURATION(x) OPUS_SET_EXPERT_FRAME_DURATION_REQUEST, __opus_check_int(x) +/** Gets the encoder's configured use of variable duration frames. + * @see OPUS_SET_EXPERT_FRAME_DURATION + * @param[out] x <tt>opus_int32 *</tt>: Returns one of the following values: + * <dl> + * <dt>OPUS_FRAMESIZE_ARG</dt><dd>Select frame size from the argument (default).</dd> + * <dt>OPUS_FRAMESIZE_2_5_MS</dt><dd>Use 2.5 ms frames.</dd> + * <dt>OPUS_FRAMESIZE_5_MS</dt><dd>Use 5 ms frames.</dd> + * <dt>OPUS_FRAMESIZE_10_MS</dt><dd>Use 10 ms frames.</dd> + * <dt>OPUS_FRAMESIZE_20_MS</dt><dd>Use 20 ms frames.</dd> + * <dt>OPUS_FRAMESIZE_40_MS</dt><dd>Use 40 ms frames.</dd> + * <dt>OPUS_FRAMESIZE_60_MS</dt><dd>Use 60 ms frames.</dd> + * <dt>OPUS_FRAMESIZE_80_MS</dt><dd>Use 80 ms frames.</dd> + * <dt>OPUS_FRAMESIZE_100_MS</dt><dd>Use 100 ms frames.</dd> + * <dt>OPUS_FRAMESIZE_120_MS</dt><dd>Use 120 ms frames.</dd> + * </dl> + * @hideinitializer */ +#define OPUS_GET_EXPERT_FRAME_DURATION(x) OPUS_GET_EXPERT_FRAME_DURATION_REQUEST, __opus_check_int_ptr(x) + +/** If set to 1, disables almost all use of prediction, making frames almost + * completely independent. This reduces quality. + * @see OPUS_GET_PREDICTION_DISABLED + * @param[in] x <tt>opus_int32</tt>: Allowed values: + * <dl> + * <dt>0</dt><dd>Enable prediction (default).</dd> + * <dt>1</dt><dd>Disable prediction.</dd> + * </dl> + * @hideinitializer */ +#define OPUS_SET_PREDICTION_DISABLED(x) OPUS_SET_PREDICTION_DISABLED_REQUEST, __opus_check_int(x) +/** Gets the encoder's configured prediction status. + * @see OPUS_SET_PREDICTION_DISABLED + * @param[out] x <tt>opus_int32 *</tt>: Returns one of the following values: + * <dl> + * <dt>0</dt><dd>Prediction enabled (default).</dd> + * <dt>1</dt><dd>Prediction disabled.</dd> + * </dl> + * @hideinitializer */ +#define OPUS_GET_PREDICTION_DISABLED(x) OPUS_GET_PREDICTION_DISABLED_REQUEST, __opus_check_int_ptr(x) + +/**@}*/ + +/** @defgroup opus_genericctls Generic CTLs + * + * These macros are used with the \c opus_decoder_ctl and + * \c opus_encoder_ctl calls to generate a particular + * request. + * + * When called on an \c OpusDecoder they apply to that + * particular decoder instance. When called on an + * \c OpusEncoder they apply to the corresponding setting + * on that encoder instance, if present. + * + * Some usage examples: + * + * @code + * int ret; + * opus_int32 pitch; + * ret = opus_decoder_ctl(dec_ctx, OPUS_GET_PITCH(&pitch)); + * if (ret == OPUS_OK) return ret; + * + * opus_encoder_ctl(enc_ctx, OPUS_RESET_STATE); + * opus_decoder_ctl(dec_ctx, OPUS_RESET_STATE); + * + * opus_int32 enc_bw, dec_bw; + * opus_encoder_ctl(enc_ctx, OPUS_GET_BANDWIDTH(&enc_bw)); + * opus_decoder_ctl(dec_ctx, OPUS_GET_BANDWIDTH(&dec_bw)); + * if (enc_bw != dec_bw) { + * printf("packet bandwidth mismatch!\n"); + * } + * @endcode + * + * @see opus_encoder, opus_decoder_ctl, opus_encoder_ctl, opus_decoderctls, opus_encoderctls + * @{ + */ + +/** Resets the codec state to be equivalent to a freshly initialized state. + * This should be called when switching streams in order to prevent + * the back to back decoding from giving different results from + * one at a time decoding. + * @hideinitializer */ +#define OPUS_RESET_STATE 4028 + +/** Gets the final state of the codec's entropy coder. + * This is used for testing purposes, + * The encoder and decoder state should be identical after coding a payload + * (assuming no data corruption or software bugs) + * + * @param[out] x <tt>opus_uint32 *</tt>: Entropy coder state + * + * @hideinitializer */ +#define OPUS_GET_FINAL_RANGE(x) OPUS_GET_FINAL_RANGE_REQUEST, __opus_check_uint_ptr(x) + +/** Gets the encoder's configured bandpass or the decoder's last bandpass. + * @see OPUS_SET_BANDWIDTH + * @param[out] x <tt>opus_int32 *</tt>: Returns one of the following values: + * <dl> + * <dt>#OPUS_AUTO</dt> <dd>(default)</dd> + * <dt>#OPUS_BANDWIDTH_NARROWBAND</dt> <dd>4 kHz passband</dd> + * <dt>#OPUS_BANDWIDTH_MEDIUMBAND</dt> <dd>6 kHz passband</dd> + * <dt>#OPUS_BANDWIDTH_WIDEBAND</dt> <dd>8 kHz passband</dd> + * <dt>#OPUS_BANDWIDTH_SUPERWIDEBAND</dt><dd>12 kHz passband</dd> + * <dt>#OPUS_BANDWIDTH_FULLBAND</dt> <dd>20 kHz passband</dd> + * </dl> + * @hideinitializer */ +#define OPUS_GET_BANDWIDTH(x) OPUS_GET_BANDWIDTH_REQUEST, __opus_check_int_ptr(x) + +/** Gets the sampling rate the encoder or decoder was initialized with. + * This simply returns the <code>Fs</code> value passed to opus_encoder_init() + * or opus_decoder_init(). + * @param[out] x <tt>opus_int32 *</tt>: Sampling rate of encoder or decoder. + * @hideinitializer + */ +#define OPUS_GET_SAMPLE_RATE(x) OPUS_GET_SAMPLE_RATE_REQUEST, __opus_check_int_ptr(x) + +/** If set to 1, disables the use of phase inversion for intensity stereo, + * improving the quality of mono downmixes, but slightly reducing normal + * stereo quality. Disabling phase inversion in the decoder does not comply + * with RFC 6716, although it does not cause any interoperability issue and + * is expected to become part of the Opus standard once RFC 6716 is updated + * by draft-ietf-codec-opus-update. + * @see OPUS_GET_PHASE_INVERSION_DISABLED + * @param[in] x <tt>opus_int32</tt>: Allowed values: + * <dl> + * <dt>0</dt><dd>Enable phase inversion (default).</dd> + * <dt>1</dt><dd>Disable phase inversion.</dd> + * </dl> + * @hideinitializer */ +#define OPUS_SET_PHASE_INVERSION_DISABLED(x) OPUS_SET_PHASE_INVERSION_DISABLED_REQUEST, __opus_check_int(x) +/** Gets the encoder's configured phase inversion status. + * @see OPUS_SET_PHASE_INVERSION_DISABLED + * @param[out] x <tt>opus_int32 *</tt>: Returns one of the following values: + * <dl> + * <dt>0</dt><dd>Stereo phase inversion enabled (default).</dd> + * <dt>1</dt><dd>Stereo phase inversion disabled.</dd> + * </dl> + * @hideinitializer */ +#define OPUS_GET_PHASE_INVERSION_DISABLED(x) OPUS_GET_PHASE_INVERSION_DISABLED_REQUEST, __opus_check_int_ptr(x) +/** Gets the DTX state of the encoder. + * Returns whether the last encoded frame was either a comfort noise update + * during DTX or not encoded because of DTX. + * @param[out] x <tt>opus_int32 *</tt>: Returns one of the following values: + * <dl> + * <dt>0</dt><dd>The encoder is not in DTX.</dd> + * <dt>1</dt><dd>The encoder is in DTX.</dd> + * </dl> + * @hideinitializer */ +#define OPUS_GET_IN_DTX(x) OPUS_GET_IN_DTX_REQUEST, __opus_check_int_ptr(x) + +/**@}*/ + +/** @defgroup opus_decoderctls Decoder related CTLs + * @see opus_genericctls, opus_encoderctls, opus_decoder + * @{ + */ + +/** Configures decoder gain adjustment. + * Scales the decoded output by a factor specified in Q8 dB units. + * This has a maximum range of -32768 to 32767 inclusive, and returns + * OPUS_BAD_ARG otherwise. The default is zero indicating no adjustment. + * This setting survives decoder reset. + * + * gain = pow(10, x/(20.0*256)) + * + * @param[in] x <tt>opus_int32</tt>: Amount to scale PCM signal by in Q8 dB units. + * @hideinitializer */ +#define OPUS_SET_GAIN(x) OPUS_SET_GAIN_REQUEST, __opus_check_int(x) +/** Gets the decoder's configured gain adjustment. @see OPUS_SET_GAIN + * + * @param[out] x <tt>opus_int32 *</tt>: Amount to scale PCM signal by in Q8 dB units. + * @hideinitializer */ +#define OPUS_GET_GAIN(x) OPUS_GET_GAIN_REQUEST, __opus_check_int_ptr(x) + +/** Gets the duration (in samples) of the last packet successfully decoded or concealed. + * @param[out] x <tt>opus_int32 *</tt>: Number of samples (at current sampling rate). + * @hideinitializer */ +#define OPUS_GET_LAST_PACKET_DURATION(x) OPUS_GET_LAST_PACKET_DURATION_REQUEST, __opus_check_int_ptr(x) + +/** Gets the pitch of the last decoded frame, if available. + * This can be used for any post-processing algorithm requiring the use of pitch, + * e.g. time stretching/shortening. If the last frame was not voiced, or if the + * pitch was not coded in the frame, then zero is returned. + * + * This CTL is only implemented for decoder instances. + * + * @param[out] x <tt>opus_int32 *</tt>: pitch period at 48 kHz (or 0 if not available) + * + * @hideinitializer */ +#define OPUS_GET_PITCH(x) OPUS_GET_PITCH_REQUEST, __opus_check_int_ptr(x) + +/**@}*/ + +/** @defgroup opus_libinfo Opus library information functions + * @{ + */ + +/** Converts an opus error code into a human readable string. + * + * @param[in] error <tt>int</tt>: Error number + * @returns Error string + */ +OPUS_EXPORT const char *opus_strerror(int error); + +/** Gets the libopus version string. + * + * Applications may look for the substring "-fixed" in the version string to + * determine whether they have a fixed-point or floating-point build at + * runtime. + * + * @returns Version string + */ +OPUS_EXPORT const char *opus_get_version_string(void); +/**@}*/ + +#ifdef __cplusplus +} +#endif + +#endif /* OPUS_DEFINES_H */ diff --git a/media/libopus/include/opus_multistream.h b/media/libopus/include/opus_multistream.h new file mode 100644 index 0000000000..babcee6905 --- /dev/null +++ b/media/libopus/include/opus_multistream.h @@ -0,0 +1,660 @@ +/* Copyright (c) 2011 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +/** + * @file opus_multistream.h + * @brief Opus reference implementation multistream API + */ + +#ifndef OPUS_MULTISTREAM_H +#define OPUS_MULTISTREAM_H + +#include "opus.h" + +#ifdef __cplusplus +extern "C" { +#endif + +/** @cond OPUS_INTERNAL_DOC */ + +/** Macros to trigger compilation errors when the wrong types are provided to a + * CTL. */ +/**@{*/ +#define __opus_check_encstate_ptr(ptr) ((ptr) + ((ptr) - (OpusEncoder**)(ptr))) +#define __opus_check_decstate_ptr(ptr) ((ptr) + ((ptr) - (OpusDecoder**)(ptr))) +/**@}*/ + +/** These are the actual encoder and decoder CTL ID numbers. + * They should not be used directly by applications. + * In general, SETs should be even and GETs should be odd.*/ +/**@{*/ +#define OPUS_MULTISTREAM_GET_ENCODER_STATE_REQUEST 5120 +#define OPUS_MULTISTREAM_GET_DECODER_STATE_REQUEST 5122 +/**@}*/ + +/** @endcond */ + +/** @defgroup opus_multistream_ctls Multistream specific encoder and decoder CTLs + * + * These are convenience macros that are specific to the + * opus_multistream_encoder_ctl() and opus_multistream_decoder_ctl() + * interface. + * The CTLs from @ref opus_genericctls, @ref opus_encoderctls, and + * @ref opus_decoderctls may be applied to a multistream encoder or decoder as + * well. + * In addition, you may retrieve the encoder or decoder state for an specific + * stream via #OPUS_MULTISTREAM_GET_ENCODER_STATE or + * #OPUS_MULTISTREAM_GET_DECODER_STATE and apply CTLs to it individually. + */ +/**@{*/ + +/** Gets the encoder state for an individual stream of a multistream encoder. + * @param[in] x <tt>opus_int32</tt>: The index of the stream whose encoder you + * wish to retrieve. + * This must be non-negative and less than + * the <code>streams</code> parameter used + * to initialize the encoder. + * @param[out] y <tt>OpusEncoder**</tt>: Returns a pointer to the given + * encoder state. + * @retval OPUS_BAD_ARG The index of the requested stream was out of range. + * @hideinitializer + */ +#define OPUS_MULTISTREAM_GET_ENCODER_STATE(x,y) OPUS_MULTISTREAM_GET_ENCODER_STATE_REQUEST, __opus_check_int(x), __opus_check_encstate_ptr(y) + +/** Gets the decoder state for an individual stream of a multistream decoder. + * @param[in] x <tt>opus_int32</tt>: The index of the stream whose decoder you + * wish to retrieve. + * This must be non-negative and less than + * the <code>streams</code> parameter used + * to initialize the decoder. + * @param[out] y <tt>OpusDecoder**</tt>: Returns a pointer to the given + * decoder state. + * @retval OPUS_BAD_ARG The index of the requested stream was out of range. + * @hideinitializer + */ +#define OPUS_MULTISTREAM_GET_DECODER_STATE(x,y) OPUS_MULTISTREAM_GET_DECODER_STATE_REQUEST, __opus_check_int(x), __opus_check_decstate_ptr(y) + +/**@}*/ + +/** @defgroup opus_multistream Opus Multistream API + * @{ + * + * The multistream API allows individual Opus streams to be combined into a + * single packet, enabling support for up to 255 channels. Unlike an + * elementary Opus stream, the encoder and decoder must negotiate the channel + * configuration before the decoder can successfully interpret the data in the + * packets produced by the encoder. Some basic information, such as packet + * duration, can be computed without any special negotiation. + * + * The format for multistream Opus packets is defined in + * <a href="https://tools.ietf.org/html/rfc7845">RFC 7845</a> + * and is based on the self-delimited Opus framing described in Appendix B of + * <a href="https://tools.ietf.org/html/rfc6716">RFC 6716</a>. + * Normal Opus packets are just a degenerate case of multistream Opus packets, + * and can be encoded or decoded with the multistream API by setting + * <code>streams</code> to <code>1</code> when initializing the encoder or + * decoder. + * + * Multistream Opus streams can contain up to 255 elementary Opus streams. + * These may be either "uncoupled" or "coupled", indicating that the decoder + * is configured to decode them to either 1 or 2 channels, respectively. + * The streams are ordered so that all coupled streams appear at the + * beginning. + * + * A <code>mapping</code> table defines which decoded channel <code>i</code> + * should be used for each input/output (I/O) channel <code>j</code>. This table is + * typically provided as an unsigned char array. + * Let <code>i = mapping[j]</code> be the index for I/O channel <code>j</code>. + * If <code>i < 2*coupled_streams</code>, then I/O channel <code>j</code> is + * encoded as the left channel of stream <code>(i/2)</code> if <code>i</code> + * is even, or as the right channel of stream <code>(i/2)</code> if + * <code>i</code> is odd. Otherwise, I/O channel <code>j</code> is encoded as + * mono in stream <code>(i - coupled_streams)</code>, unless it has the special + * value 255, in which case it is omitted from the encoding entirely (the + * decoder will reproduce it as silence). Each value <code>i</code> must either + * be the special value 255 or be less than <code>streams + coupled_streams</code>. + * + * The output channels specified by the encoder + * should use the + * <a href="https://www.xiph.org/vorbis/doc/Vorbis_I_spec.html#x1-810004.3.9">Vorbis + * channel ordering</a>. A decoder may wish to apply an additional permutation + * to the mapping the encoder used to achieve a different output channel + * order (e.g. for outputing in WAV order). + * + * Each multistream packet contains an Opus packet for each stream, and all of + * the Opus packets in a single multistream packet must have the same + * duration. Therefore the duration of a multistream packet can be extracted + * from the TOC sequence of the first stream, which is located at the + * beginning of the packet, just like an elementary Opus stream: + * + * @code + * int nb_samples; + * int nb_frames; + * nb_frames = opus_packet_get_nb_frames(data, len); + * if (nb_frames < 1) + * return nb_frames; + * nb_samples = opus_packet_get_samples_per_frame(data, 48000) * nb_frames; + * @endcode + * + * The general encoding and decoding process proceeds exactly the same as in + * the normal @ref opus_encoder and @ref opus_decoder APIs. + * See their documentation for an overview of how to use the corresponding + * multistream functions. + */ + +/** Opus multistream encoder state. + * This contains the complete state of a multistream Opus encoder. + * It is position independent and can be freely copied. + * @see opus_multistream_encoder_create + * @see opus_multistream_encoder_init + */ +typedef struct OpusMSEncoder OpusMSEncoder; + +/** Opus multistream decoder state. + * This contains the complete state of a multistream Opus decoder. + * It is position independent and can be freely copied. + * @see opus_multistream_decoder_create + * @see opus_multistream_decoder_init + */ +typedef struct OpusMSDecoder OpusMSDecoder; + +/**\name Multistream encoder functions */ +/**@{*/ + +/** Gets the size of an OpusMSEncoder structure. + * @param streams <tt>int</tt>: The total number of streams to encode from the + * input. + * This must be no more than 255. + * @param coupled_streams <tt>int</tt>: Number of coupled (2 channel) streams + * to encode. + * This must be no larger than the total + * number of streams. + * Additionally, The total number of + * encoded channels (<code>streams + + * coupled_streams</code>) must be no + * more than 255. + * @returns The size in bytes on success, or a negative error code + * (see @ref opus_errorcodes) on error. + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT opus_int32 opus_multistream_encoder_get_size( + int streams, + int coupled_streams +); + +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT opus_int32 opus_multistream_surround_encoder_get_size( + int channels, + int mapping_family +); + + +/** Allocates and initializes a multistream encoder state. + * Call opus_multistream_encoder_destroy() to release + * this object when finished. + * @param Fs <tt>opus_int32</tt>: Sampling rate of the input signal (in Hz). + * This must be one of 8000, 12000, 16000, + * 24000, or 48000. + * @param channels <tt>int</tt>: Number of channels in the input signal. + * This must be at most 255. + * It may be greater than the number of + * coded channels (<code>streams + + * coupled_streams</code>). + * @param streams <tt>int</tt>: The total number of streams to encode from the + * input. + * This must be no more than the number of channels. + * @param coupled_streams <tt>int</tt>: Number of coupled (2 channel) streams + * to encode. + * This must be no larger than the total + * number of streams. + * Additionally, The total number of + * encoded channels (<code>streams + + * coupled_streams</code>) must be no + * more than the number of input channels. + * @param[in] mapping <code>const unsigned char[channels]</code>: Mapping from + * encoded channels to input channels, as described in + * @ref opus_multistream. As an extra constraint, the + * multistream encoder does not allow encoding coupled + * streams for which one channel is unused since this + * is never a good idea. + * @param application <tt>int</tt>: The target encoder application. + * This must be one of the following: + * <dl> + * <dt>#OPUS_APPLICATION_VOIP</dt> + * <dd>Process signal for improved speech intelligibility.</dd> + * <dt>#OPUS_APPLICATION_AUDIO</dt> + * <dd>Favor faithfulness to the original input.</dd> + * <dt>#OPUS_APPLICATION_RESTRICTED_LOWDELAY</dt> + * <dd>Configure the minimum possible coding delay by disabling certain modes + * of operation.</dd> + * </dl> + * @param[out] error <tt>int *</tt>: Returns #OPUS_OK on success, or an error + * code (see @ref opus_errorcodes) on + * failure. + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT OpusMSEncoder *opus_multistream_encoder_create( + opus_int32 Fs, + int channels, + int streams, + int coupled_streams, + const unsigned char *mapping, + int application, + int *error +) OPUS_ARG_NONNULL(5); + +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT OpusMSEncoder *opus_multistream_surround_encoder_create( + opus_int32 Fs, + int channels, + int mapping_family, + int *streams, + int *coupled_streams, + unsigned char *mapping, + int application, + int *error +) OPUS_ARG_NONNULL(4) OPUS_ARG_NONNULL(5) OPUS_ARG_NONNULL(6); + +/** Initialize a previously allocated multistream encoder state. + * The memory pointed to by \a st must be at least the size returned by + * opus_multistream_encoder_get_size(). + * This is intended for applications which use their own allocator instead of + * malloc. + * To reset a previously initialized state, use the #OPUS_RESET_STATE CTL. + * @see opus_multistream_encoder_create + * @see opus_multistream_encoder_get_size + * @param st <tt>OpusMSEncoder*</tt>: Multistream encoder state to initialize. + * @param Fs <tt>opus_int32</tt>: Sampling rate of the input signal (in Hz). + * This must be one of 8000, 12000, 16000, + * 24000, or 48000. + * @param channels <tt>int</tt>: Number of channels in the input signal. + * This must be at most 255. + * It may be greater than the number of + * coded channels (<code>streams + + * coupled_streams</code>). + * @param streams <tt>int</tt>: The total number of streams to encode from the + * input. + * This must be no more than the number of channels. + * @param coupled_streams <tt>int</tt>: Number of coupled (2 channel) streams + * to encode. + * This must be no larger than the total + * number of streams. + * Additionally, The total number of + * encoded channels (<code>streams + + * coupled_streams</code>) must be no + * more than the number of input channels. + * @param[in] mapping <code>const unsigned char[channels]</code>: Mapping from + * encoded channels to input channels, as described in + * @ref opus_multistream. As an extra constraint, the + * multistream encoder does not allow encoding coupled + * streams for which one channel is unused since this + * is never a good idea. + * @param application <tt>int</tt>: The target encoder application. + * This must be one of the following: + * <dl> + * <dt>#OPUS_APPLICATION_VOIP</dt> + * <dd>Process signal for improved speech intelligibility.</dd> + * <dt>#OPUS_APPLICATION_AUDIO</dt> + * <dd>Favor faithfulness to the original input.</dd> + * <dt>#OPUS_APPLICATION_RESTRICTED_LOWDELAY</dt> + * <dd>Configure the minimum possible coding delay by disabling certain modes + * of operation.</dd> + * </dl> + * @returns #OPUS_OK on success, or an error code (see @ref opus_errorcodes) + * on failure. + */ +OPUS_EXPORT int opus_multistream_encoder_init( + OpusMSEncoder *st, + opus_int32 Fs, + int channels, + int streams, + int coupled_streams, + const unsigned char *mapping, + int application +) OPUS_ARG_NONNULL(1) OPUS_ARG_NONNULL(6); + +OPUS_EXPORT int opus_multistream_surround_encoder_init( + OpusMSEncoder *st, + opus_int32 Fs, + int channels, + int mapping_family, + int *streams, + int *coupled_streams, + unsigned char *mapping, + int application +) OPUS_ARG_NONNULL(1) OPUS_ARG_NONNULL(5) OPUS_ARG_NONNULL(6) OPUS_ARG_NONNULL(7); + +/** Encodes a multistream Opus frame. + * @param st <tt>OpusMSEncoder*</tt>: Multistream encoder state. + * @param[in] pcm <tt>const opus_int16*</tt>: The input signal as interleaved + * samples. + * This must contain + * <code>frame_size*channels</code> + * samples. + * @param frame_size <tt>int</tt>: Number of samples per channel in the input + * signal. + * This must be an Opus frame size for the + * encoder's sampling rate. + * For example, at 48 kHz the permitted values + * are 120, 240, 480, 960, 1920, and 2880. + * Passing in a duration of less than 10 ms + * (480 samples at 48 kHz) will prevent the + * encoder from using the LPC or hybrid modes. + * @param[out] data <tt>unsigned char*</tt>: Output payload. + * This must contain storage for at + * least \a max_data_bytes. + * @param [in] max_data_bytes <tt>opus_int32</tt>: Size of the allocated + * memory for the output + * payload. This may be + * used to impose an upper limit on + * the instant bitrate, but should + * not be used as the only bitrate + * control. Use #OPUS_SET_BITRATE to + * control the bitrate. + * @returns The length of the encoded packet (in bytes) on success or a + * negative error code (see @ref opus_errorcodes) on failure. + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT int opus_multistream_encode( + OpusMSEncoder *st, + const opus_int16 *pcm, + int frame_size, + unsigned char *data, + opus_int32 max_data_bytes +) OPUS_ARG_NONNULL(1) OPUS_ARG_NONNULL(2) OPUS_ARG_NONNULL(4); + +/** Encodes a multistream Opus frame from floating point input. + * @param st <tt>OpusMSEncoder*</tt>: Multistream encoder state. + * @param[in] pcm <tt>const float*</tt>: The input signal as interleaved + * samples with a normal range of + * +/-1.0. + * Samples with a range beyond +/-1.0 + * are supported but will be clipped by + * decoders using the integer API and + * should only be used if it is known + * that the far end supports extended + * dynamic range. + * This must contain + * <code>frame_size*channels</code> + * samples. + * @param frame_size <tt>int</tt>: Number of samples per channel in the input + * signal. + * This must be an Opus frame size for the + * encoder's sampling rate. + * For example, at 48 kHz the permitted values + * are 120, 240, 480, 960, 1920, and 2880. + * Passing in a duration of less than 10 ms + * (480 samples at 48 kHz) will prevent the + * encoder from using the LPC or hybrid modes. + * @param[out] data <tt>unsigned char*</tt>: Output payload. + * This must contain storage for at + * least \a max_data_bytes. + * @param [in] max_data_bytes <tt>opus_int32</tt>: Size of the allocated + * memory for the output + * payload. This may be + * used to impose an upper limit on + * the instant bitrate, but should + * not be used as the only bitrate + * control. Use #OPUS_SET_BITRATE to + * control the bitrate. + * @returns The length of the encoded packet (in bytes) on success or a + * negative error code (see @ref opus_errorcodes) on failure. + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT int opus_multistream_encode_float( + OpusMSEncoder *st, + const float *pcm, + int frame_size, + unsigned char *data, + opus_int32 max_data_bytes +) OPUS_ARG_NONNULL(1) OPUS_ARG_NONNULL(2) OPUS_ARG_NONNULL(4); + +/** Frees an <code>OpusMSEncoder</code> allocated by + * opus_multistream_encoder_create(). + * @param st <tt>OpusMSEncoder*</tt>: Multistream encoder state to be freed. + */ +OPUS_EXPORT void opus_multistream_encoder_destroy(OpusMSEncoder *st); + +/** Perform a CTL function on a multistream Opus encoder. + * + * Generally the request and subsequent arguments are generated by a + * convenience macro. + * @param st <tt>OpusMSEncoder*</tt>: Multistream encoder state. + * @param request This and all remaining parameters should be replaced by one + * of the convenience macros in @ref opus_genericctls, + * @ref opus_encoderctls, or @ref opus_multistream_ctls. + * @see opus_genericctls + * @see opus_encoderctls + * @see opus_multistream_ctls + */ +OPUS_EXPORT int opus_multistream_encoder_ctl(OpusMSEncoder *st, int request, ...) OPUS_ARG_NONNULL(1); + +/**@}*/ + +/**\name Multistream decoder functions */ +/**@{*/ + +/** Gets the size of an <code>OpusMSDecoder</code> structure. + * @param streams <tt>int</tt>: The total number of streams coded in the + * input. + * This must be no more than 255. + * @param coupled_streams <tt>int</tt>: Number streams to decode as coupled + * (2 channel) streams. + * This must be no larger than the total + * number of streams. + * Additionally, The total number of + * coded channels (<code>streams + + * coupled_streams</code>) must be no + * more than 255. + * @returns The size in bytes on success, or a negative error code + * (see @ref opus_errorcodes) on error. + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT opus_int32 opus_multistream_decoder_get_size( + int streams, + int coupled_streams +); + +/** Allocates and initializes a multistream decoder state. + * Call opus_multistream_decoder_destroy() to release + * this object when finished. + * @param Fs <tt>opus_int32</tt>: Sampling rate to decode at (in Hz). + * This must be one of 8000, 12000, 16000, + * 24000, or 48000. + * @param channels <tt>int</tt>: Number of channels to output. + * This must be at most 255. + * It may be different from the number of coded + * channels (<code>streams + + * coupled_streams</code>). + * @param streams <tt>int</tt>: The total number of streams coded in the + * input. + * This must be no more than 255. + * @param coupled_streams <tt>int</tt>: Number of streams to decode as coupled + * (2 channel) streams. + * This must be no larger than the total + * number of streams. + * Additionally, The total number of + * coded channels (<code>streams + + * coupled_streams</code>) must be no + * more than 255. + * @param[in] mapping <code>const unsigned char[channels]</code>: Mapping from + * coded channels to output channels, as described in + * @ref opus_multistream. + * @param[out] error <tt>int *</tt>: Returns #OPUS_OK on success, or an error + * code (see @ref opus_errorcodes) on + * failure. + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT OpusMSDecoder *opus_multistream_decoder_create( + opus_int32 Fs, + int channels, + int streams, + int coupled_streams, + const unsigned char *mapping, + int *error +) OPUS_ARG_NONNULL(5); + +/** Intialize a previously allocated decoder state object. + * The memory pointed to by \a st must be at least the size returned by + * opus_multistream_encoder_get_size(). + * This is intended for applications which use their own allocator instead of + * malloc. + * To reset a previously initialized state, use the #OPUS_RESET_STATE CTL. + * @see opus_multistream_decoder_create + * @see opus_multistream_deocder_get_size + * @param st <tt>OpusMSEncoder*</tt>: Multistream encoder state to initialize. + * @param Fs <tt>opus_int32</tt>: Sampling rate to decode at (in Hz). + * This must be one of 8000, 12000, 16000, + * 24000, or 48000. + * @param channels <tt>int</tt>: Number of channels to output. + * This must be at most 255. + * It may be different from the number of coded + * channels (<code>streams + + * coupled_streams</code>). + * @param streams <tt>int</tt>: The total number of streams coded in the + * input. + * This must be no more than 255. + * @param coupled_streams <tt>int</tt>: Number of streams to decode as coupled + * (2 channel) streams. + * This must be no larger than the total + * number of streams. + * Additionally, The total number of + * coded channels (<code>streams + + * coupled_streams</code>) must be no + * more than 255. + * @param[in] mapping <code>const unsigned char[channels]</code>: Mapping from + * coded channels to output channels, as described in + * @ref opus_multistream. + * @returns #OPUS_OK on success, or an error code (see @ref opus_errorcodes) + * on failure. + */ +OPUS_EXPORT int opus_multistream_decoder_init( + OpusMSDecoder *st, + opus_int32 Fs, + int channels, + int streams, + int coupled_streams, + const unsigned char *mapping +) OPUS_ARG_NONNULL(1) OPUS_ARG_NONNULL(6); + +/** Decode a multistream Opus packet. + * @param st <tt>OpusMSDecoder*</tt>: Multistream decoder state. + * @param[in] data <tt>const unsigned char*</tt>: Input payload. + * Use a <code>NULL</code> + * pointer to indicate packet + * loss. + * @param len <tt>opus_int32</tt>: Number of bytes in payload. + * @param[out] pcm <tt>opus_int16*</tt>: Output signal, with interleaved + * samples. + * This must contain room for + * <code>frame_size*channels</code> + * samples. + * @param frame_size <tt>int</tt>: The number of samples per channel of + * available space in \a pcm. + * If this is less than the maximum packet duration + * (120 ms; 5760 for 48kHz), this function will not be capable + * of decoding some packets. In the case of PLC (data==NULL) + * or FEC (decode_fec=1), then frame_size needs to be exactly + * the duration of audio that is missing, otherwise the + * decoder will not be in the optimal state to decode the + * next incoming packet. For the PLC and FEC cases, frame_size + * <b>must</b> be a multiple of 2.5 ms. + * @param decode_fec <tt>int</tt>: Flag (0 or 1) to request that any in-band + * forward error correction data be decoded. + * If no such data is available, the frame is + * decoded as if it were lost. + * @returns Number of samples decoded on success or a negative error code + * (see @ref opus_errorcodes) on failure. + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT int opus_multistream_decode( + OpusMSDecoder *st, + const unsigned char *data, + opus_int32 len, + opus_int16 *pcm, + int frame_size, + int decode_fec +) OPUS_ARG_NONNULL(1) OPUS_ARG_NONNULL(4); + +/** Decode a multistream Opus packet with floating point output. + * @param st <tt>OpusMSDecoder*</tt>: Multistream decoder state. + * @param[in] data <tt>const unsigned char*</tt>: Input payload. + * Use a <code>NULL</code> + * pointer to indicate packet + * loss. + * @param len <tt>opus_int32</tt>: Number of bytes in payload. + * @param[out] pcm <tt>opus_int16*</tt>: Output signal, with interleaved + * samples. + * This must contain room for + * <code>frame_size*channels</code> + * samples. + * @param frame_size <tt>int</tt>: The number of samples per channel of + * available space in \a pcm. + * If this is less than the maximum packet duration + * (120 ms; 5760 for 48kHz), this function will not be capable + * of decoding some packets. In the case of PLC (data==NULL) + * or FEC (decode_fec=1), then frame_size needs to be exactly + * the duration of audio that is missing, otherwise the + * decoder will not be in the optimal state to decode the + * next incoming packet. For the PLC and FEC cases, frame_size + * <b>must</b> be a multiple of 2.5 ms. + * @param decode_fec <tt>int</tt>: Flag (0 or 1) to request that any in-band + * forward error correction data be decoded. + * If no such data is available, the frame is + * decoded as if it were lost. + * @returns Number of samples decoded on success or a negative error code + * (see @ref opus_errorcodes) on failure. + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT int opus_multistream_decode_float( + OpusMSDecoder *st, + const unsigned char *data, + opus_int32 len, + float *pcm, + int frame_size, + int decode_fec +) OPUS_ARG_NONNULL(1) OPUS_ARG_NONNULL(4); + +/** Perform a CTL function on a multistream Opus decoder. + * + * Generally the request and subsequent arguments are generated by a + * convenience macro. + * @param st <tt>OpusMSDecoder*</tt>: Multistream decoder state. + * @param request This and all remaining parameters should be replaced by one + * of the convenience macros in @ref opus_genericctls, + * @ref opus_decoderctls, or @ref opus_multistream_ctls. + * @see opus_genericctls + * @see opus_decoderctls + * @see opus_multistream_ctls + */ +OPUS_EXPORT int opus_multistream_decoder_ctl(OpusMSDecoder *st, int request, ...) OPUS_ARG_NONNULL(1); + +/** Frees an <code>OpusMSDecoder</code> allocated by + * opus_multistream_decoder_create(). + * @param st <tt>OpusMSDecoder</tt>: Multistream decoder state to be freed. + */ +OPUS_EXPORT void opus_multistream_decoder_destroy(OpusMSDecoder *st); + +/**@}*/ + +/**@}*/ + +#ifdef __cplusplus +} +#endif + +#endif /* OPUS_MULTISTREAM_H */ diff --git a/media/libopus/include/opus_projection.h b/media/libopus/include/opus_projection.h new file mode 100644 index 0000000000..9dabf4e85c --- /dev/null +++ b/media/libopus/include/opus_projection.h @@ -0,0 +1,568 @@ +/* Copyright (c) 2017 Google Inc. + Written by Andrew Allen */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +/** + * @file opus_projection.h + * @brief Opus projection reference API + */ + +#ifndef OPUS_PROJECTION_H +#define OPUS_PROJECTION_H + +#include "opus_multistream.h" + +#ifdef __cplusplus +extern "C" { +#endif + +/** @cond OPUS_INTERNAL_DOC */ + +/** These are the actual encoder and decoder CTL ID numbers. + * They should not be used directly by applications.c + * In general, SETs should be even and GETs should be odd.*/ +/**@{*/ +#define OPUS_PROJECTION_GET_DEMIXING_MATRIX_GAIN_REQUEST 6001 +#define OPUS_PROJECTION_GET_DEMIXING_MATRIX_SIZE_REQUEST 6003 +#define OPUS_PROJECTION_GET_DEMIXING_MATRIX_REQUEST 6005 +/**@}*/ + + +/** @endcond */ + +/** @defgroup opus_projection_ctls Projection specific encoder and decoder CTLs + * + * These are convenience macros that are specific to the + * opus_projection_encoder_ctl() and opus_projection_decoder_ctl() + * interface. + * The CTLs from @ref opus_genericctls, @ref opus_encoderctls, + * @ref opus_decoderctls, and @ref opus_multistream_ctls may be applied to a + * projection encoder or decoder as well. + */ +/**@{*/ + +/** Gets the gain (in dB. S7.8-format) of the demixing matrix from the encoder. + * @param[out] x <tt>opus_int32 *</tt>: Returns the gain (in dB. S7.8-format) + * of the demixing matrix. + * @hideinitializer + */ +#define OPUS_PROJECTION_GET_DEMIXING_MATRIX_GAIN(x) OPUS_PROJECTION_GET_DEMIXING_MATRIX_GAIN_REQUEST, __opus_check_int_ptr(x) + + +/** Gets the size in bytes of the demixing matrix from the encoder. + * @param[out] x <tt>opus_int32 *</tt>: Returns the size in bytes of the + * demixing matrix. + * @hideinitializer + */ +#define OPUS_PROJECTION_GET_DEMIXING_MATRIX_SIZE(x) OPUS_PROJECTION_GET_DEMIXING_MATRIX_SIZE_REQUEST, __opus_check_int_ptr(x) + + +/** Copies the demixing matrix to the supplied pointer location. + * @param[out] x <tt>unsigned char *</tt>: Returns the demixing matrix to the + * supplied pointer location. + * @param y <tt>opus_int32</tt>: The size in bytes of the reserved memory at the + * pointer location. + * @hideinitializer + */ +#define OPUS_PROJECTION_GET_DEMIXING_MATRIX(x,y) OPUS_PROJECTION_GET_DEMIXING_MATRIX_REQUEST, x, __opus_check_int(y) + + +/**@}*/ + +/** Opus projection encoder state. + * This contains the complete state of a projection Opus encoder. + * It is position independent and can be freely copied. + * @see opus_projection_ambisonics_encoder_create + */ +typedef struct OpusProjectionEncoder OpusProjectionEncoder; + + +/** Opus projection decoder state. + * This contains the complete state of a projection Opus decoder. + * It is position independent and can be freely copied. + * @see opus_projection_decoder_create + * @see opus_projection_decoder_init + */ +typedef struct OpusProjectionDecoder OpusProjectionDecoder; + + +/**\name Projection encoder functions */ +/**@{*/ + +/** Gets the size of an OpusProjectionEncoder structure. + * @param channels <tt>int</tt>: The total number of input channels to encode. + * This must be no more than 255. + * @param mapping_family <tt>int</tt>: The mapping family to use for selecting + * the appropriate projection. + * @returns The size in bytes on success, or a negative error code + * (see @ref opus_errorcodes) on error. + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT opus_int32 opus_projection_ambisonics_encoder_get_size( + int channels, + int mapping_family +); + + +/** Allocates and initializes a projection encoder state. + * Call opus_projection_encoder_destroy() to release + * this object when finished. + * @param Fs <tt>opus_int32</tt>: Sampling rate of the input signal (in Hz). + * This must be one of 8000, 12000, 16000, + * 24000, or 48000. + * @param channels <tt>int</tt>: Number of channels in the input signal. + * This must be at most 255. + * It may be greater than the number of + * coded channels (<code>streams + + * coupled_streams</code>). + * @param mapping_family <tt>int</tt>: The mapping family to use for selecting + * the appropriate projection. + * @param[out] streams <tt>int *</tt>: The total number of streams that will + * be encoded from the input. + * @param[out] coupled_streams <tt>int *</tt>: Number of coupled (2 channel) + * streams that will be encoded from the input. + * @param application <tt>int</tt>: The target encoder application. + * This must be one of the following: + * <dl> + * <dt>#OPUS_APPLICATION_VOIP</dt> + * <dd>Process signal for improved speech intelligibility.</dd> + * <dt>#OPUS_APPLICATION_AUDIO</dt> + * <dd>Favor faithfulness to the original input.</dd> + * <dt>#OPUS_APPLICATION_RESTRICTED_LOWDELAY</dt> + * <dd>Configure the minimum possible coding delay by disabling certain modes + * of operation.</dd> + * </dl> + * @param[out] error <tt>int *</tt>: Returns #OPUS_OK on success, or an error + * code (see @ref opus_errorcodes) on + * failure. + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT OpusProjectionEncoder *opus_projection_ambisonics_encoder_create( + opus_int32 Fs, + int channels, + int mapping_family, + int *streams, + int *coupled_streams, + int application, + int *error +) OPUS_ARG_NONNULL(4) OPUS_ARG_NONNULL(5); + + +/** Initialize a previously allocated projection encoder state. + * The memory pointed to by \a st must be at least the size returned by + * opus_projection_ambisonics_encoder_get_size(). + * This is intended for applications which use their own allocator instead of + * malloc. + * To reset a previously initialized state, use the #OPUS_RESET_STATE CTL. + * @see opus_projection_ambisonics_encoder_create + * @see opus_projection_ambisonics_encoder_get_size + * @param st <tt>OpusProjectionEncoder*</tt>: Projection encoder state to initialize. + * @param Fs <tt>opus_int32</tt>: Sampling rate of the input signal (in Hz). + * This must be one of 8000, 12000, 16000, + * 24000, or 48000. + * @param channels <tt>int</tt>: Number of channels in the input signal. + * This must be at most 255. + * It may be greater than the number of + * coded channels (<code>streams + + * coupled_streams</code>). + * @param streams <tt>int</tt>: The total number of streams to encode from the + * input. + * This must be no more than the number of channels. + * @param coupled_streams <tt>int</tt>: Number of coupled (2 channel) streams + * to encode. + * This must be no larger than the total + * number of streams. + * Additionally, The total number of + * encoded channels (<code>streams + + * coupled_streams</code>) must be no + * more than the number of input channels. + * @param application <tt>int</tt>: The target encoder application. + * This must be one of the following: + * <dl> + * <dt>#OPUS_APPLICATION_VOIP</dt> + * <dd>Process signal for improved speech intelligibility.</dd> + * <dt>#OPUS_APPLICATION_AUDIO</dt> + * <dd>Favor faithfulness to the original input.</dd> + * <dt>#OPUS_APPLICATION_RESTRICTED_LOWDELAY</dt> + * <dd>Configure the minimum possible coding delay by disabling certain modes + * of operation.</dd> + * </dl> + * @returns #OPUS_OK on success, or an error code (see @ref opus_errorcodes) + * on failure. + */ +OPUS_EXPORT int opus_projection_ambisonics_encoder_init( + OpusProjectionEncoder *st, + opus_int32 Fs, + int channels, + int mapping_family, + int *streams, + int *coupled_streams, + int application +) OPUS_ARG_NONNULL(1) OPUS_ARG_NONNULL(5) OPUS_ARG_NONNULL(6); + + +/** Encodes a projection Opus frame. + * @param st <tt>OpusProjectionEncoder*</tt>: Projection encoder state. + * @param[in] pcm <tt>const opus_int16*</tt>: The input signal as interleaved + * samples. + * This must contain + * <code>frame_size*channels</code> + * samples. + * @param frame_size <tt>int</tt>: Number of samples per channel in the input + * signal. + * This must be an Opus frame size for the + * encoder's sampling rate. + * For example, at 48 kHz the permitted values + * are 120, 240, 480, 960, 1920, and 2880. + * Passing in a duration of less than 10 ms + * (480 samples at 48 kHz) will prevent the + * encoder from using the LPC or hybrid modes. + * @param[out] data <tt>unsigned char*</tt>: Output payload. + * This must contain storage for at + * least \a max_data_bytes. + * @param [in] max_data_bytes <tt>opus_int32</tt>: Size of the allocated + * memory for the output + * payload. This may be + * used to impose an upper limit on + * the instant bitrate, but should + * not be used as the only bitrate + * control. Use #OPUS_SET_BITRATE to + * control the bitrate. + * @returns The length of the encoded packet (in bytes) on success or a + * negative error code (see @ref opus_errorcodes) on failure. + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT int opus_projection_encode( + OpusProjectionEncoder *st, + const opus_int16 *pcm, + int frame_size, + unsigned char *data, + opus_int32 max_data_bytes +) OPUS_ARG_NONNULL(1) OPUS_ARG_NONNULL(2) OPUS_ARG_NONNULL(4); + + +/** Encodes a projection Opus frame from floating point input. + * @param st <tt>OpusProjectionEncoder*</tt>: Projection encoder state. + * @param[in] pcm <tt>const float*</tt>: The input signal as interleaved + * samples with a normal range of + * +/-1.0. + * Samples with a range beyond +/-1.0 + * are supported but will be clipped by + * decoders using the integer API and + * should only be used if it is known + * that the far end supports extended + * dynamic range. + * This must contain + * <code>frame_size*channels</code> + * samples. + * @param frame_size <tt>int</tt>: Number of samples per channel in the input + * signal. + * This must be an Opus frame size for the + * encoder's sampling rate. + * For example, at 48 kHz the permitted values + * are 120, 240, 480, 960, 1920, and 2880. + * Passing in a duration of less than 10 ms + * (480 samples at 48 kHz) will prevent the + * encoder from using the LPC or hybrid modes. + * @param[out] data <tt>unsigned char*</tt>: Output payload. + * This must contain storage for at + * least \a max_data_bytes. + * @param [in] max_data_bytes <tt>opus_int32</tt>: Size of the allocated + * memory for the output + * payload. This may be + * used to impose an upper limit on + * the instant bitrate, but should + * not be used as the only bitrate + * control. Use #OPUS_SET_BITRATE to + * control the bitrate. + * @returns The length of the encoded packet (in bytes) on success or a + * negative error code (see @ref opus_errorcodes) on failure. + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT int opus_projection_encode_float( + OpusProjectionEncoder *st, + const float *pcm, + int frame_size, + unsigned char *data, + opus_int32 max_data_bytes +) OPUS_ARG_NONNULL(1) OPUS_ARG_NONNULL(2) OPUS_ARG_NONNULL(4); + + +/** Frees an <code>OpusProjectionEncoder</code> allocated by + * opus_projection_ambisonics_encoder_create(). + * @param st <tt>OpusProjectionEncoder*</tt>: Projection encoder state to be freed. + */ +OPUS_EXPORT void opus_projection_encoder_destroy(OpusProjectionEncoder *st); + + +/** Perform a CTL function on a projection Opus encoder. + * + * Generally the request and subsequent arguments are generated by a + * convenience macro. + * @param st <tt>OpusProjectionEncoder*</tt>: Projection encoder state. + * @param request This and all remaining parameters should be replaced by one + * of the convenience macros in @ref opus_genericctls, + * @ref opus_encoderctls, @ref opus_multistream_ctls, or + * @ref opus_projection_ctls + * @see opus_genericctls + * @see opus_encoderctls + * @see opus_multistream_ctls + * @see opus_projection_ctls + */ +OPUS_EXPORT int opus_projection_encoder_ctl(OpusProjectionEncoder *st, int request, ...) OPUS_ARG_NONNULL(1); + + +/**@}*/ + +/**\name Projection decoder functions */ +/**@{*/ + +/** Gets the size of an <code>OpusProjectionDecoder</code> structure. + * @param channels <tt>int</tt>: The total number of output channels. + * This must be no more than 255. + * @param streams <tt>int</tt>: The total number of streams coded in the + * input. + * This must be no more than 255. + * @param coupled_streams <tt>int</tt>: Number streams to decode as coupled + * (2 channel) streams. + * This must be no larger than the total + * number of streams. + * Additionally, The total number of + * coded channels (<code>streams + + * coupled_streams</code>) must be no + * more than 255. + * @returns The size in bytes on success, or a negative error code + * (see @ref opus_errorcodes) on error. + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT opus_int32 opus_projection_decoder_get_size( + int channels, + int streams, + int coupled_streams +); + + +/** Allocates and initializes a projection decoder state. + * Call opus_projection_decoder_destroy() to release + * this object when finished. + * @param Fs <tt>opus_int32</tt>: Sampling rate to decode at (in Hz). + * This must be one of 8000, 12000, 16000, + * 24000, or 48000. + * @param channels <tt>int</tt>: Number of channels to output. + * This must be at most 255. + * It may be different from the number of coded + * channels (<code>streams + + * coupled_streams</code>). + * @param streams <tt>int</tt>: The total number of streams coded in the + * input. + * This must be no more than 255. + * @param coupled_streams <tt>int</tt>: Number of streams to decode as coupled + * (2 channel) streams. + * This must be no larger than the total + * number of streams. + * Additionally, The total number of + * coded channels (<code>streams + + * coupled_streams</code>) must be no + * more than 255. + * @param[in] demixing_matrix <tt>const unsigned char[demixing_matrix_size]</tt>: Demixing matrix + * that mapping from coded channels to output channels, + * as described in @ref opus_projection and + * @ref opus_projection_ctls. + * @param demixing_matrix_size <tt>opus_int32</tt>: The size in bytes of the + * demixing matrix, as + * described in @ref + * opus_projection_ctls. + * @param[out] error <tt>int *</tt>: Returns #OPUS_OK on success, or an error + * code (see @ref opus_errorcodes) on + * failure. + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT OpusProjectionDecoder *opus_projection_decoder_create( + opus_int32 Fs, + int channels, + int streams, + int coupled_streams, + unsigned char *demixing_matrix, + opus_int32 demixing_matrix_size, + int *error +) OPUS_ARG_NONNULL(5); + + +/** Intialize a previously allocated projection decoder state object. + * The memory pointed to by \a st must be at least the size returned by + * opus_projection_decoder_get_size(). + * This is intended for applications which use their own allocator instead of + * malloc. + * To reset a previously initialized state, use the #OPUS_RESET_STATE CTL. + * @see opus_projection_decoder_create + * @see opus_projection_deocder_get_size + * @param st <tt>OpusProjectionDecoder*</tt>: Projection encoder state to initialize. + * @param Fs <tt>opus_int32</tt>: Sampling rate to decode at (in Hz). + * This must be one of 8000, 12000, 16000, + * 24000, or 48000. + * @param channels <tt>int</tt>: Number of channels to output. + * This must be at most 255. + * It may be different from the number of coded + * channels (<code>streams + + * coupled_streams</code>). + * @param streams <tt>int</tt>: The total number of streams coded in the + * input. + * This must be no more than 255. + * @param coupled_streams <tt>int</tt>: Number of streams to decode as coupled + * (2 channel) streams. + * This must be no larger than the total + * number of streams. + * Additionally, The total number of + * coded channels (<code>streams + + * coupled_streams</code>) must be no + * more than 255. + * @param[in] demixing_matrix <tt>const unsigned char[demixing_matrix_size]</tt>: Demixing matrix + * that mapping from coded channels to output channels, + * as described in @ref opus_projection and + * @ref opus_projection_ctls. + * @param demixing_matrix_size <tt>opus_int32</tt>: The size in bytes of the + * demixing matrix, as + * described in @ref + * opus_projection_ctls. + * @returns #OPUS_OK on success, or an error code (see @ref opus_errorcodes) + * on failure. + */ +OPUS_EXPORT int opus_projection_decoder_init( + OpusProjectionDecoder *st, + opus_int32 Fs, + int channels, + int streams, + int coupled_streams, + unsigned char *demixing_matrix, + opus_int32 demixing_matrix_size +) OPUS_ARG_NONNULL(1) OPUS_ARG_NONNULL(6); + + +/** Decode a projection Opus packet. + * @param st <tt>OpusProjectionDecoder*</tt>: Projection decoder state. + * @param[in] data <tt>const unsigned char*</tt>: Input payload. + * Use a <code>NULL</code> + * pointer to indicate packet + * loss. + * @param len <tt>opus_int32</tt>: Number of bytes in payload. + * @param[out] pcm <tt>opus_int16*</tt>: Output signal, with interleaved + * samples. + * This must contain room for + * <code>frame_size*channels</code> + * samples. + * @param frame_size <tt>int</tt>: The number of samples per channel of + * available space in \a pcm. + * If this is less than the maximum packet duration + * (120 ms; 5760 for 48kHz), this function will not be capable + * of decoding some packets. In the case of PLC (data==NULL) + * or FEC (decode_fec=1), then frame_size needs to be exactly + * the duration of audio that is missing, otherwise the + * decoder will not be in the optimal state to decode the + * next incoming packet. For the PLC and FEC cases, frame_size + * <b>must</b> be a multiple of 2.5 ms. + * @param decode_fec <tt>int</tt>: Flag (0 or 1) to request that any in-band + * forward error correction data be decoded. + * If no such data is available, the frame is + * decoded as if it were lost. + * @returns Number of samples decoded on success or a negative error code + * (see @ref opus_errorcodes) on failure. + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT int opus_projection_decode( + OpusProjectionDecoder *st, + const unsigned char *data, + opus_int32 len, + opus_int16 *pcm, + int frame_size, + int decode_fec +) OPUS_ARG_NONNULL(1) OPUS_ARG_NONNULL(4); + + +/** Decode a projection Opus packet with floating point output. + * @param st <tt>OpusProjectionDecoder*</tt>: Projection decoder state. + * @param[in] data <tt>const unsigned char*</tt>: Input payload. + * Use a <code>NULL</code> + * pointer to indicate packet + * loss. + * @param len <tt>opus_int32</tt>: Number of bytes in payload. + * @param[out] pcm <tt>opus_int16*</tt>: Output signal, with interleaved + * samples. + * This must contain room for + * <code>frame_size*channels</code> + * samples. + * @param frame_size <tt>int</tt>: The number of samples per channel of + * available space in \a pcm. + * If this is less than the maximum packet duration + * (120 ms; 5760 for 48kHz), this function will not be capable + * of decoding some packets. In the case of PLC (data==NULL) + * or FEC (decode_fec=1), then frame_size needs to be exactly + * the duration of audio that is missing, otherwise the + * decoder will not be in the optimal state to decode the + * next incoming packet. For the PLC and FEC cases, frame_size + * <b>must</b> be a multiple of 2.5 ms. + * @param decode_fec <tt>int</tt>: Flag (0 or 1) to request that any in-band + * forward error correction data be decoded. + * If no such data is available, the frame is + * decoded as if it were lost. + * @returns Number of samples decoded on success or a negative error code + * (see @ref opus_errorcodes) on failure. + */ +OPUS_EXPORT OPUS_WARN_UNUSED_RESULT int opus_projection_decode_float( + OpusProjectionDecoder *st, + const unsigned char *data, + opus_int32 len, + float *pcm, + int frame_size, + int decode_fec +) OPUS_ARG_NONNULL(1) OPUS_ARG_NONNULL(4); + + +/** Perform a CTL function on a projection Opus decoder. + * + * Generally the request and subsequent arguments are generated by a + * convenience macro. + * @param st <tt>OpusProjectionDecoder*</tt>: Projection decoder state. + * @param request This and all remaining parameters should be replaced by one + * of the convenience macros in @ref opus_genericctls, + * @ref opus_decoderctls, @ref opus_multistream_ctls, or + * @ref opus_projection_ctls. + * @see opus_genericctls + * @see opus_decoderctls + * @see opus_multistream_ctls + * @see opus_projection_ctls + */ +OPUS_EXPORT int opus_projection_decoder_ctl(OpusProjectionDecoder *st, int request, ...) OPUS_ARG_NONNULL(1); + + +/** Frees an <code>OpusProjectionDecoder</code> allocated by + * opus_projection_decoder_create(). + * @param st <tt>OpusProjectionDecoder</tt>: Projection decoder state to be freed. + */ +OPUS_EXPORT void opus_projection_decoder_destroy(OpusProjectionDecoder *st); + + +/**@}*/ + +/**@}*/ + +#ifdef __cplusplus +} +#endif + +#endif /* OPUS_PROJECTION_H */ diff --git a/media/libopus/include/opus_types.h b/media/libopus/include/opus_types.h new file mode 100644 index 0000000000..7cf675580f --- /dev/null +++ b/media/libopus/include/opus_types.h @@ -0,0 +1,166 @@ +/* (C) COPYRIGHT 1994-2002 Xiph.Org Foundation */ +/* Modified by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ +/* opus_types.h based on ogg_types.h from libogg */ + +/** + @file opus_types.h + @brief Opus reference implementation types +*/ +#ifndef OPUS_TYPES_H +#define OPUS_TYPES_H + +#define opus_int int /* used for counters etc; at least 16 bits */ +#define opus_int64 long long +#define opus_int8 signed char + +#define opus_uint unsigned int /* used for counters etc; at least 16 bits */ +#define opus_uint64 unsigned long long +#define opus_uint8 unsigned char + +/* Use the real stdint.h if it's there (taken from Paul Hsieh's pstdint.h) */ +#if (defined(__STDC__) && __STDC__ && defined(__STDC_VERSION__) && __STDC_VERSION__ >= 199901L) || (defined(__GNUC__) && (defined(_STDINT_H) || defined(_STDINT_H_)) || defined (HAVE_STDINT_H)) +#include <stdint.h> +# undef opus_int64 +# undef opus_int8 +# undef opus_uint64 +# undef opus_uint8 + typedef int8_t opus_int8; + typedef uint8_t opus_uint8; + typedef int16_t opus_int16; + typedef uint16_t opus_uint16; + typedef int32_t opus_int32; + typedef uint32_t opus_uint32; + typedef int64_t opus_int64; + typedef uint64_t opus_uint64; +#elif defined(_WIN32) + +# if defined(__CYGWIN__) +# include <_G_config.h> + typedef _G_int32_t opus_int32; + typedef _G_uint32_t opus_uint32; + typedef _G_int16 opus_int16; + typedef _G_uint16 opus_uint16; +# elif defined(__MINGW32__) + typedef short opus_int16; + typedef unsigned short opus_uint16; + typedef int opus_int32; + typedef unsigned int opus_uint32; +# elif defined(__MWERKS__) + typedef int opus_int32; + typedef unsigned int opus_uint32; + typedef short opus_int16; + typedef unsigned short opus_uint16; +# else + /* MSVC/Borland */ + typedef __int32 opus_int32; + typedef unsigned __int32 opus_uint32; + typedef __int16 opus_int16; + typedef unsigned __int16 opus_uint16; +# endif + +#elif defined(__MACOS__) + +# include <sys/types.h> + typedef SInt16 opus_int16; + typedef UInt16 opus_uint16; + typedef SInt32 opus_int32; + typedef UInt32 opus_uint32; + +#elif (defined(__APPLE__) && defined(__MACH__)) /* MacOS X Framework build */ + +# include <sys/types.h> + typedef int16_t opus_int16; + typedef u_int16_t opus_uint16; + typedef int32_t opus_int32; + typedef u_int32_t opus_uint32; + +#elif defined(__BEOS__) + + /* Be */ +# include <inttypes.h> + typedef int16 opus_int16; + typedef u_int16 opus_uint16; + typedef int32_t opus_int32; + typedef u_int32_t opus_uint32; + +#elif defined (__EMX__) + + /* OS/2 GCC */ + typedef short opus_int16; + typedef unsigned short opus_uint16; + typedef int opus_int32; + typedef unsigned int opus_uint32; + +#elif defined (DJGPP) + + /* DJGPP */ + typedef short opus_int16; + typedef unsigned short opus_uint16; + typedef int opus_int32; + typedef unsigned int opus_uint32; + +#elif defined(R5900) + + /* PS2 EE */ + typedef int opus_int32; + typedef unsigned opus_uint32; + typedef short opus_int16; + typedef unsigned short opus_uint16; + +#elif defined(__SYMBIAN32__) + + /* Symbian GCC */ + typedef signed short opus_int16; + typedef unsigned short opus_uint16; + typedef signed int opus_int32; + typedef unsigned int opus_uint32; + +#elif defined(CONFIG_TI_C54X) || defined (CONFIG_TI_C55X) + + typedef short opus_int16; + typedef unsigned short opus_uint16; + typedef long opus_int32; + typedef unsigned long opus_uint32; + +#elif defined(CONFIG_TI_C6X) + + typedef short opus_int16; + typedef unsigned short opus_uint16; + typedef int opus_int32; + typedef unsigned int opus_uint32; + +#else + + /* Give up, take a reasonable guess */ + typedef short opus_int16; + typedef unsigned short opus_uint16; + typedef int opus_int32; + typedef unsigned int opus_uint32; + +#endif + +#endif /* OPUS_TYPES_H */ diff --git a/media/libopus/moz.build b/media/libopus/moz.build new file mode 100644 index 0000000000..442cc0b8ea --- /dev/null +++ b/media/libopus/moz.build @@ -0,0 +1,149 @@ +# -*- Mode: python; indent-tabs-mode: nil; tab-width: 40 -*- +# vim: set filetype=python: +# This Source Code Form is subject to the terms of the Mozilla Public +# License, v. 2.0. If a copy of the MPL was not distributed with this +# file, You can obtain one at http://mozilla.org/MPL/2.0/. + +with Files("*"): + BUG_COMPONENT = ("Core", "Audio/Video") + +EXPORTS.opus += [ + "include/opus.h", + "include/opus_defines.h", + "include/opus_multistream.h", + "include/opus_types.h", +] + +# We allow warnings for third-party code that can be updated from upstream. +AllowCompilerWarnings() + +FINAL_LIBRARY = "gkmedias" + +DEFINES["OPUS_BUILD"] = True +DEFINES["OPUS_VERSION"] = "f3de7ca74358eff2aec766b076ded84f463ac606" +DEFINES["USE_ALLOCA"] = True +DEFINES["ENABLE_HARDENING"] = True + +# Don"t export symbols +DEFINES["OPUS_EXPORT"] = "" + +if CONFIG["CPU_ARCH"] == "arm" and CONFIG["GNU_AS"]: + DEFINES["OPUS_ARM_ASM"] = True + DEFINES["OPUS_ARM_EXTERNAL_ASM"] = True + DEFINES["OPUS_ARM_INLINE_ASM"] = True + if int(CONFIG["ARM_ARCH"]) >= 6: + DEFINES["OPUS_ARM_INLINE_EDSP"] = True + DEFINES["OPUS_ARM_MAY_HAVE_EDSP"] = True + DEFINES["OPUS_ARM_MAY_HAVE_MEDIA"] = True + DEFINES["OPUS_ARM_MAY_HAVE_NEON"] = True + +if CONFIG["MOZ_DEBUG"]: + DEFINES["ENABLE_ASSERTIONS"] = True + +if CONFIG["OS_ARCH"] in ("Linux", "Darwin", "DragonFly", "FreeBSD", + "NetBSD", "OpenBSD"): + DEFINES["HAVE_LRINTF"] = True + +if CONFIG["OS_ARCH"] == "WINNT": + DEFINES["inline"] = "__inline" + if CONFIG["CC_TYPE"] in ("clang", "gcc"): + DEFINES["HAVE_LRINTF"] = True + +if CONFIG["OS_ARCH"] == "AIX": + DEFINES["alloca"] = "__alloca" + +if CONFIG["OS_ARCH"] == "SunOS": + DEFINES["HAVE_ALLOCA_H"] = True + +if not CONFIG["MOZ_SAMPLE_TYPE_FLOAT32"]: + DEFINES["FIXED_POINT"] = 1 + DEFINES["DISABLE_FLOAT_API"] = True + +LOCAL_INCLUDES += [ + "celt", + "include", + "silk", + "silk/fixed", + "silk/float", + "src", +] + +# sources.mozbuild is generated from gen-sources.py when a new libopus is +# imported. +include("sources.mozbuild") + +UNIFIED_SOURCES += celt_sources +UNIFIED_SOURCES += silk_sources +UNIFIED_SOURCES += opus_sources +SOURCES += opus_nonunified_sources + +if CONFIG["MOZ_SAMPLE_TYPE_FLOAT32"]: + UNIFIED_SOURCES += silk_sources_float + UNIFIED_SOURCES += opus_sources_float +else: + UNIFIED_SOURCES += silk_sources_fixed + +if CONFIG["CPU_ARCH"] in ("x86", "x86_64"): + DEFINES["OPUS_HAVE_RTCD"] = True + DEFINES["CPU_INFO_BY_ASM"] = True + DEFINES["OPUS_X86_MAY_HAVE_SSE"] = True + DEFINES["OPUS_X86_MAY_HAVE_SSE2"] = True + DEFINES["OPUS_X86_MAY_HAVE_SSE4_1"] = True + DEFINES["OPUS_X86_MAY_HAVE_AVX"] = True + SOURCES += celt_sources_x86_rtcd + SOURCES += celt_sources_sse + SOURCES += celt_sources_sse2 + SOURCES += celt_sources_sse4_1 + SOURCES += silk_sources_x86_rtcd + SOURCES += silk_sources_sse4_1 + if not CONFIG["MOZ_SAMPLE_TYPE_FLOAT32"]: + SOURCES += silk_sources_fixed_sse4_1 + for f in SOURCES: + if f in celt_sources_sse: + SOURCES[f].flags += CONFIG["SSE_FLAGS"] + if f in celt_sources_sse2: + SOURCES[f].flags += CONFIG["SSE2_FLAGS"] + if f in celt_sources_sse4_1 or \ + f in silk_sources_sse4_1 or \ + f in silk_sources_fixed_sse4_1: + SOURCES[f].flags += ["-msse4.1"] + +if CONFIG["CPU_ARCH"] == "arm" and CONFIG["GNU_AS"]: + SOURCES += celt_sources_arm_rtcd + SOURCES += [ + "!celt_pitch_xcorr_arm-gnu.s" + ] + # -Os is significantly slower, enable -O3 unless optimization is disabled + if CONFIG["MOZ_OPTIMIZE"]: + CFLAGS += [ + "-O3", + ] + CXXFLAGS += [ + "-O3", + ] + # These flags are a lie; they"re just used to enable the requisite + # opcodes; actual arch detection is done at runtime. + ASFLAGS += [ + "-march=armv7-a", + ] + ASFLAGS += CONFIG["NEON_FLAGS"] + +if CONFIG["CPU_ARCH"] == "aarch64" and CONFIG["CC_TYPE"] in ("clang", "gcc"): + DEFINES["OPUS_ARM_PRESUME_AARCH64_NEON_INTR"] = True + DEFINES["OPUS_ARM_PRESUME_NEON"] = True + DEFINES["OPUS_ARM_PRESUME_NEON_INTR"] = True + SOURCES += celt_sources_arm_neon_intr + SOURCES += silk_sources_arm_neon_intr + if CONFIG["MOZ_SAMPLE_TYPE_FLOAT32"]: + DEFINES["OPUS_ARM_MAY_HAVE_NEON"] = True + DEFINES["OPUS_ARM_MAY_HAVE_NEON_INTR"] = True + else: + SOURCES += silk_sources_fixed_arm_neon_intr + +# Suppress warnings in third-party code. +if CONFIG["CC_TYPE"] in ("clang", "gcc"): + if CONFIG["CC_TYPE"] == "clang": + CFLAGS += ["-Wno-#pragma-messages"] + +# Add libFuzzer configuration directives +include("/tools/fuzzing/libfuzzer-config.mozbuild") diff --git a/media/libopus/moz.yaml b/media/libopus/moz.yaml new file mode 100644 index 0000000000..72849d9dd2 --- /dev/null +++ b/media/libopus/moz.yaml @@ -0,0 +1,69 @@ +# Version of this schema +schema: 1 + +bugzilla: + # Bugzilla product and component for this directory and subdirectories + product: Core + component: "Audio/Video: Playback" + +# Document the source of externally hosted code +origin: + + # Short name of the package/library + name: opus + + description: opus, an open, royalty-free, highly versatile audio codec + + # Full URL for the package's homepage/etc + # Usually different from repository url + url: https://opus-codec.org/ + + # Human-readable identifier for this version/release + # Generally "version NNN", "tag SSS", "bookmark SSS" + release: f3de7ca74358eff2aec766b076ded84f463ac606 (2023-04-20T15:13:19.000-07:00). + + # Revision to pull in + # Must be a long or short commit SHA (long preferred) + revision: f3de7ca74358eff2aec766b076ded84f463ac606 + + # The package's license, where possible using the mnemonic from + # https://spdx.org/licenses/ + # Multiple licenses can be specified (as a YAML list) + # A "LICENSE" file must exist containing the full license text + license: BSD-2-Clause + + license-file: COPYING + +updatebot: + maintainer-phab: kinetik + maintainer-bz: kinetik@flim.org + tasks: + - type: vendoring + enabled: true + frequency: every + +vendoring: + url: https://gitlab.xiph.org/xiph/opus + source-hosting: gitlab + vendor-directory: media/libopus + skip-vendoring-steps: ['move-contents', 'update-moz-build'] + + keep: + - COPYING + - gen-sources.py + - "*.patch" + - Makefile.in + - README_MOZILLA + - sources.mozbuild + - update.sh + + update-actions: + - action: run-script + script: 'update.sh' + cwd: '{yaml_dir}' + args: ['{tmpextractdir}'] + - action: replace-in-file-regex + file: moz.build + pattern: 'DEFINES\["OPUS_VERSION"\] = "(.+)"' + with: 'DEFINES["OPUS_VERSION"] = "{revision}"' + diff --git a/media/libopus/nonunified.patch b/media/libopus/nonunified.patch new file mode 100644 index 0000000000..693881b3cf --- /dev/null +++ b/media/libopus/nonunified.patch @@ -0,0 +1,40 @@ +diff --git a/media/libopus/sources.mozbuild b/media/libopus/sources.mozbuild +index 8a39b9f..dfc2c62 100644 +--- a/media/libopus/sources.mozbuild ++++ b/media/libopus/sources.mozbuild +@@ -2,8 +2,5 @@ + celt_sources = [ + 'celt/bands.c', +- 'celt/celt.c', +- 'celt/celt_decoder.c', +- 'celt/celt_encoder.c', + 'celt/celt_lpc.c', + 'celt/cwrs.c', + 'celt/entcode.c', +@@ -20,6 +18,17 @@ celt_sources = [ + 'celt/vq.c', + ] + ++opus_nonunified_sources = [ ++ # Disabled because of undefined reference to celt_fatal at link time ++ 'celt/celt.c', ++ # Disabled because of name clash of opus_custom_encoder_get_size. ++ 'celt/celt_decoder.c', ++ 'celt/celt_encoder.c', ++ # Disabled for (safe) warning about QA redefinition. ++ 'silk/LPC_inv_pred_gain.c', ++ 'silk/NLSF2A.c', ++] ++ + celt_sources_sse = [ + 'celt/x86/pitch_sse.c', + 'celt/x86/x86_celt_map.c', +@@ -114,8 +114,6 @@ silk_sources = [ + 'silk/LP_variable_cutoff.c', + 'silk/LPC_analysis_filter.c', + 'silk/LPC_fit.c', +- 'silk/LPC_inv_pred_gain.c', +- 'silk/NLSF2A.c', + 'silk/NLSF_decode.c', + 'silk/NLSF_del_dec_quant.c', + 'silk/NLSF_encode.c', diff --git a/media/libopus/silk/A2NLSF.c b/media/libopus/silk/A2NLSF.c new file mode 100644 index 0000000000..b487686ff9 --- /dev/null +++ b/media/libopus/silk/A2NLSF.c @@ -0,0 +1,267 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +/* Conversion between prediction filter coefficients and NLSFs */ +/* Requires the order to be an even number */ +/* A piecewise linear approximation maps LSF <-> cos(LSF) */ +/* Therefore the result is not accurate NLSFs, but the two */ +/* functions are accurate inverses of each other */ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "SigProc_FIX.h" +#include "tables.h" + +/* Number of binary divisions, when not in low complexity mode */ +#define BIN_DIV_STEPS_A2NLSF_FIX 3 /* must be no higher than 16 - log2( LSF_COS_TAB_SZ_FIX ) */ +#define MAX_ITERATIONS_A2NLSF_FIX 16 + +/* Helper function for A2NLSF(..) */ +/* Transforms polynomials from cos(n*f) to cos(f)^n */ +static OPUS_INLINE void silk_A2NLSF_trans_poly( + opus_int32 *p, /* I/O Polynomial */ + const opus_int dd /* I Polynomial order (= filter order / 2 ) */ +) +{ + opus_int k, n; + + for( k = 2; k <= dd; k++ ) { + for( n = dd; n > k; n-- ) { + p[ n - 2 ] -= p[ n ]; + } + p[ k - 2 ] -= silk_LSHIFT( p[ k ], 1 ); + } +} +/* Helper function for A2NLSF(..) */ +/* Polynomial evaluation */ +static OPUS_INLINE opus_int32 silk_A2NLSF_eval_poly( /* return the polynomial evaluation, in Q16 */ + opus_int32 *p, /* I Polynomial, Q16 */ + const opus_int32 x, /* I Evaluation point, Q12 */ + const opus_int dd /* I Order */ +) +{ + opus_int n; + opus_int32 x_Q16, y32; + + y32 = p[ dd ]; /* Q16 */ + x_Q16 = silk_LSHIFT( x, 4 ); + + if ( opus_likely( 8 == dd ) ) + { + y32 = silk_SMLAWW( p[ 7 ], y32, x_Q16 ); + y32 = silk_SMLAWW( p[ 6 ], y32, x_Q16 ); + y32 = silk_SMLAWW( p[ 5 ], y32, x_Q16 ); + y32 = silk_SMLAWW( p[ 4 ], y32, x_Q16 ); + y32 = silk_SMLAWW( p[ 3 ], y32, x_Q16 ); + y32 = silk_SMLAWW( p[ 2 ], y32, x_Q16 ); + y32 = silk_SMLAWW( p[ 1 ], y32, x_Q16 ); + y32 = silk_SMLAWW( p[ 0 ], y32, x_Q16 ); + } + else + { + for( n = dd - 1; n >= 0; n-- ) { + y32 = silk_SMLAWW( p[ n ], y32, x_Q16 ); /* Q16 */ + } + } + return y32; +} + +static OPUS_INLINE void silk_A2NLSF_init( + const opus_int32 *a_Q16, + opus_int32 *P, + opus_int32 *Q, + const opus_int dd +) +{ + opus_int k; + + /* Convert filter coefs to even and odd polynomials */ + P[dd] = silk_LSHIFT( 1, 16 ); + Q[dd] = silk_LSHIFT( 1, 16 ); + for( k = 0; k < dd; k++ ) { + P[ k ] = -a_Q16[ dd - k - 1 ] - a_Q16[ dd + k ]; /* Q16 */ + Q[ k ] = -a_Q16[ dd - k - 1 ] + a_Q16[ dd + k ]; /* Q16 */ + } + + /* Divide out zeros as we have that for even filter orders, */ + /* z = 1 is always a root in Q, and */ + /* z = -1 is always a root in P */ + for( k = dd; k > 0; k-- ) { + P[ k - 1 ] -= P[ k ]; + Q[ k - 1 ] += Q[ k ]; + } + + /* Transform polynomials from cos(n*f) to cos(f)^n */ + silk_A2NLSF_trans_poly( P, dd ); + silk_A2NLSF_trans_poly( Q, dd ); +} + +/* Compute Normalized Line Spectral Frequencies (NLSFs) from whitening filter coefficients */ +/* If not all roots are found, the a_Q16 coefficients are bandwidth expanded until convergence. */ +void silk_A2NLSF( + opus_int16 *NLSF, /* O Normalized Line Spectral Frequencies in Q15 (0..2^15-1) [d] */ + opus_int32 *a_Q16, /* I/O Monic whitening filter coefficients in Q16 [d] */ + const opus_int d /* I Filter order (must be even) */ +) +{ + opus_int i, k, m, dd, root_ix, ffrac; + opus_int32 xlo, xhi, xmid; + opus_int32 ylo, yhi, ymid, thr; + opus_int32 nom, den; + opus_int32 P[ SILK_MAX_ORDER_LPC / 2 + 1 ]; + opus_int32 Q[ SILK_MAX_ORDER_LPC / 2 + 1 ]; + opus_int32 *PQ[ 2 ]; + opus_int32 *p; + + /* Store pointers to array */ + PQ[ 0 ] = P; + PQ[ 1 ] = Q; + + dd = silk_RSHIFT( d, 1 ); + + silk_A2NLSF_init( a_Q16, P, Q, dd ); + + /* Find roots, alternating between P and Q */ + p = P; /* Pointer to polynomial */ + + xlo = silk_LSFCosTab_FIX_Q12[ 0 ]; /* Q12*/ + ylo = silk_A2NLSF_eval_poly( p, xlo, dd ); + + if( ylo < 0 ) { + /* Set the first NLSF to zero and move on to the next */ + NLSF[ 0 ] = 0; + p = Q; /* Pointer to polynomial */ + ylo = silk_A2NLSF_eval_poly( p, xlo, dd ); + root_ix = 1; /* Index of current root */ + } else { + root_ix = 0; /* Index of current root */ + } + k = 1; /* Loop counter */ + i = 0; /* Counter for bandwidth expansions applied */ + thr = 0; + while( 1 ) { + /* Evaluate polynomial */ + xhi = silk_LSFCosTab_FIX_Q12[ k ]; /* Q12 */ + yhi = silk_A2NLSF_eval_poly( p, xhi, dd ); + + /* Detect zero crossing */ + if( ( ylo <= 0 && yhi >= thr ) || ( ylo >= 0 && yhi <= -thr ) ) { + if( yhi == 0 ) { + /* If the root lies exactly at the end of the current */ + /* interval, look for the next root in the next interval */ + thr = 1; + } else { + thr = 0; + } + /* Binary division */ + ffrac = -256; + for( m = 0; m < BIN_DIV_STEPS_A2NLSF_FIX; m++ ) { + /* Evaluate polynomial */ + xmid = silk_RSHIFT_ROUND( xlo + xhi, 1 ); + ymid = silk_A2NLSF_eval_poly( p, xmid, dd ); + + /* Detect zero crossing */ + if( ( ylo <= 0 && ymid >= 0 ) || ( ylo >= 0 && ymid <= 0 ) ) { + /* Reduce frequency */ + xhi = xmid; + yhi = ymid; + } else { + /* Increase frequency */ + xlo = xmid; + ylo = ymid; + ffrac = silk_ADD_RSHIFT( ffrac, 128, m ); + } + } + + /* Interpolate */ + if( silk_abs( ylo ) < 65536 ) { + /* Avoid dividing by zero */ + den = ylo - yhi; + nom = silk_LSHIFT( ylo, 8 - BIN_DIV_STEPS_A2NLSF_FIX ) + silk_RSHIFT( den, 1 ); + if( den != 0 ) { + ffrac += silk_DIV32( nom, den ); + } + } else { + /* No risk of dividing by zero because abs(ylo - yhi) >= abs(ylo) >= 65536 */ + ffrac += silk_DIV32( ylo, silk_RSHIFT( ylo - yhi, 8 - BIN_DIV_STEPS_A2NLSF_FIX ) ); + } + NLSF[ root_ix ] = (opus_int16)silk_min_32( silk_LSHIFT( (opus_int32)k, 8 ) + ffrac, silk_int16_MAX ); + + silk_assert( NLSF[ root_ix ] >= 0 ); + + root_ix++; /* Next root */ + if( root_ix >= d ) { + /* Found all roots */ + break; + } + /* Alternate pointer to polynomial */ + p = PQ[ root_ix & 1 ]; + + /* Evaluate polynomial */ + xlo = silk_LSFCosTab_FIX_Q12[ k - 1 ]; /* Q12*/ + ylo = silk_LSHIFT( 1 - ( root_ix & 2 ), 12 ); + } else { + /* Increment loop counter */ + k++; + xlo = xhi; + ylo = yhi; + thr = 0; + + if( k > LSF_COS_TAB_SZ_FIX ) { + i++; + if( i > MAX_ITERATIONS_A2NLSF_FIX ) { + /* Set NLSFs to white spectrum and exit */ + NLSF[ 0 ] = (opus_int16)silk_DIV32_16( 1 << 15, d + 1 ); + for( k = 1; k < d; k++ ) { + NLSF[ k ] = (opus_int16)silk_ADD16( NLSF[ k-1 ], NLSF[ 0 ] ); + } + return; + } + + /* Error: Apply progressively more bandwidth expansion and run again */ + silk_bwexpander_32( a_Q16, d, 65536 - silk_LSHIFT( 1, i ) ); + + silk_A2NLSF_init( a_Q16, P, Q, dd ); + p = P; /* Pointer to polynomial */ + xlo = silk_LSFCosTab_FIX_Q12[ 0 ]; /* Q12*/ + ylo = silk_A2NLSF_eval_poly( p, xlo, dd ); + if( ylo < 0 ) { + /* Set the first NLSF to zero and move on to the next */ + NLSF[ 0 ] = 0; + p = Q; /* Pointer to polynomial */ + ylo = silk_A2NLSF_eval_poly( p, xlo, dd ); + root_ix = 1; /* Index of current root */ + } else { + root_ix = 0; /* Index of current root */ + } + k = 1; /* Reset loop counter */ + } + } + } +} diff --git a/media/libopus/silk/API.h b/media/libopus/silk/API.h new file mode 100644 index 0000000000..4d90ff9aa3 --- /dev/null +++ b/media/libopus/silk/API.h @@ -0,0 +1,135 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef SILK_API_H +#define SILK_API_H + +#include "control.h" +#include "typedef.h" +#include "errors.h" +#include "entenc.h" +#include "entdec.h" + +#ifdef __cplusplus +extern "C" +{ +#endif + +#define SILK_MAX_FRAMES_PER_PACKET 3 + +/* Struct for TOC (Table of Contents) */ +typedef struct { + opus_int VADFlag; /* Voice activity for packet */ + opus_int VADFlags[ SILK_MAX_FRAMES_PER_PACKET ]; /* Voice activity for each frame in packet */ + opus_int inbandFECFlag; /* Flag indicating if packet contains in-band FEC */ +} silk_TOC_struct; + +/****************************************/ +/* Encoder functions */ +/****************************************/ + +/***********************************************/ +/* Get size in bytes of the Silk encoder state */ +/***********************************************/ +opus_int silk_Get_Encoder_Size( /* O Returns error code */ + opus_int *encSizeBytes /* O Number of bytes in SILK encoder state */ +); + +/*************************/ +/* Init or reset encoder */ +/*************************/ +opus_int silk_InitEncoder( /* O Returns error code */ + void *encState, /* I/O State */ + int arch, /* I Run-time architecture */ + silk_EncControlStruct *encStatus /* O Encoder Status */ +); + +/**************************/ +/* Encode frame with Silk */ +/**************************/ +/* Note: if prefillFlag is set, the input must contain 10 ms of audio, irrespective of what */ +/* encControl->payloadSize_ms is set to */ +opus_int silk_Encode( /* O Returns error code */ + void *encState, /* I/O State */ + silk_EncControlStruct *encControl, /* I Control status */ + const opus_int16 *samplesIn, /* I Speech sample input vector */ + opus_int nSamplesIn, /* I Number of samples in input vector */ + ec_enc *psRangeEnc, /* I/O Compressor data structure */ + opus_int32 *nBytesOut, /* I/O Number of bytes in payload (input: Max bytes) */ + const opus_int prefillFlag, /* I Flag to indicate prefilling buffers no coding */ + int activity /* I Decision of Opus voice activity detector */ +); + +/****************************************/ +/* Decoder functions */ +/****************************************/ + +/***********************************************/ +/* Get size in bytes of the Silk decoder state */ +/***********************************************/ +opus_int silk_Get_Decoder_Size( /* O Returns error code */ + opus_int *decSizeBytes /* O Number of bytes in SILK decoder state */ +); + +/*************************/ +/* Init or Reset decoder */ +/*************************/ +opus_int silk_InitDecoder( /* O Returns error code */ + void *decState /* I/O State */ +); + +/******************/ +/* Decode a frame */ +/******************/ +opus_int silk_Decode( /* O Returns error code */ + void* decState, /* I/O State */ + silk_DecControlStruct* decControl, /* I/O Control Structure */ + opus_int lostFlag, /* I 0: no loss, 1 loss, 2 decode fec */ + opus_int newPacketFlag, /* I Indicates first decoder call for this packet */ + ec_dec *psRangeDec, /* I/O Compressor data structure */ + opus_int16 *samplesOut, /* O Decoded output speech vector */ + opus_int32 *nSamplesOut, /* O Number of samples decoded */ + int arch /* I Run-time architecture */ +); + +#if 0 +/**************************************/ +/* Get table of contents for a packet */ +/**************************************/ +opus_int silk_get_TOC( + const opus_uint8 *payload, /* I Payload data */ + const opus_int nBytesIn, /* I Number of input bytes */ + const opus_int nFramesPerPayload, /* I Number of SILK frames per payload */ + silk_TOC_struct *Silk_TOC /* O Type of content */ +); +#endif + +#ifdef __cplusplus +} +#endif + +#endif diff --git a/media/libopus/silk/CNG.c b/media/libopus/silk/CNG.c new file mode 100644 index 0000000000..2a910099ed --- /dev/null +++ b/media/libopus/silk/CNG.c @@ -0,0 +1,188 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" +#include "stack_alloc.h" + +/* Generates excitation for CNG LPC synthesis */ +static OPUS_INLINE void silk_CNG_exc( + opus_int32 exc_Q14[], /* O CNG excitation signal Q10 */ + opus_int32 exc_buf_Q14[], /* I Random samples buffer Q10 */ + opus_int length, /* I Length */ + opus_int32 *rand_seed /* I/O Seed to random index generator */ +) +{ + opus_int32 seed; + opus_int i, idx, exc_mask; + + exc_mask = CNG_BUF_MASK_MAX; + while( exc_mask > length ) { + exc_mask = silk_RSHIFT( exc_mask, 1 ); + } + + seed = *rand_seed; + for( i = 0; i < length; i++ ) { + seed = silk_RAND( seed ); + idx = (opus_int)( silk_RSHIFT( seed, 24 ) & exc_mask ); + silk_assert( idx >= 0 ); + silk_assert( idx <= CNG_BUF_MASK_MAX ); + exc_Q14[ i ] = exc_buf_Q14[ idx ]; + } + *rand_seed = seed; +} + +void silk_CNG_Reset( + silk_decoder_state *psDec /* I/O Decoder state */ +) +{ + opus_int i, NLSF_step_Q15, NLSF_acc_Q15; + + NLSF_step_Q15 = silk_DIV32_16( silk_int16_MAX, psDec->LPC_order + 1 ); + NLSF_acc_Q15 = 0; + for( i = 0; i < psDec->LPC_order; i++ ) { + NLSF_acc_Q15 += NLSF_step_Q15; + psDec->sCNG.CNG_smth_NLSF_Q15[ i ] = NLSF_acc_Q15; + } + psDec->sCNG.CNG_smth_Gain_Q16 = 0; + psDec->sCNG.rand_seed = 3176576; +} + +/* Updates CNG estimate, and applies the CNG when packet was lost */ +void silk_CNG( + silk_decoder_state *psDec, /* I/O Decoder state */ + silk_decoder_control *psDecCtrl, /* I/O Decoder control */ + opus_int16 frame[], /* I/O Signal */ + opus_int length /* I Length of residual */ +) +{ + opus_int i, subfr; + opus_int32 LPC_pred_Q10, max_Gain_Q16, gain_Q16, gain_Q10; + opus_int16 A_Q12[ MAX_LPC_ORDER ]; + silk_CNG_struct *psCNG = &psDec->sCNG; + SAVE_STACK; + + if( psDec->fs_kHz != psCNG->fs_kHz ) { + /* Reset state */ + silk_CNG_Reset( psDec ); + + psCNG->fs_kHz = psDec->fs_kHz; + } + if( psDec->lossCnt == 0 && psDec->prevSignalType == TYPE_NO_VOICE_ACTIVITY ) { + /* Update CNG parameters */ + + /* Smoothing of LSF's */ + for( i = 0; i < psDec->LPC_order; i++ ) { + psCNG->CNG_smth_NLSF_Q15[ i ] += silk_SMULWB( (opus_int32)psDec->prevNLSF_Q15[ i ] - (opus_int32)psCNG->CNG_smth_NLSF_Q15[ i ], CNG_NLSF_SMTH_Q16 ); + } + /* Find the subframe with the highest gain */ + max_Gain_Q16 = 0; + subfr = 0; + for( i = 0; i < psDec->nb_subfr; i++ ) { + if( psDecCtrl->Gains_Q16[ i ] > max_Gain_Q16 ) { + max_Gain_Q16 = psDecCtrl->Gains_Q16[ i ]; + subfr = i; + } + } + /* Update CNG excitation buffer with excitation from this subframe */ + silk_memmove( &psCNG->CNG_exc_buf_Q14[ psDec->subfr_length ], psCNG->CNG_exc_buf_Q14, ( psDec->nb_subfr - 1 ) * psDec->subfr_length * sizeof( opus_int32 ) ); + silk_memcpy( psCNG->CNG_exc_buf_Q14, &psDec->exc_Q14[ subfr * psDec->subfr_length ], psDec->subfr_length * sizeof( opus_int32 ) ); + + /* Smooth gains */ + for( i = 0; i < psDec->nb_subfr; i++ ) { + psCNG->CNG_smth_Gain_Q16 += silk_SMULWB( psDecCtrl->Gains_Q16[ i ] - psCNG->CNG_smth_Gain_Q16, CNG_GAIN_SMTH_Q16 ); + /* If the smoothed gain is 3 dB greater than this subframe's gain, use this subframe's gain to adapt faster. */ + if( silk_SMULWW( psCNG->CNG_smth_Gain_Q16, CNG_GAIN_SMTH_THRESHOLD_Q16 ) > psDecCtrl->Gains_Q16[ i ] ) { + psCNG->CNG_smth_Gain_Q16 = psDecCtrl->Gains_Q16[ i ]; + } + } + } + + /* Add CNG when packet is lost or during DTX */ + if( psDec->lossCnt ) { + VARDECL( opus_int32, CNG_sig_Q14 ); + ALLOC( CNG_sig_Q14, length + MAX_LPC_ORDER, opus_int32 ); + + /* Generate CNG excitation */ + gain_Q16 = silk_SMULWW( psDec->sPLC.randScale_Q14, psDec->sPLC.prevGain_Q16[1] ); + if( gain_Q16 >= (1 << 21) || psCNG->CNG_smth_Gain_Q16 > (1 << 23) ) { + gain_Q16 = silk_SMULTT( gain_Q16, gain_Q16 ); + gain_Q16 = silk_SUB_LSHIFT32(silk_SMULTT( psCNG->CNG_smth_Gain_Q16, psCNG->CNG_smth_Gain_Q16 ), gain_Q16, 5 ); + gain_Q16 = silk_LSHIFT32( silk_SQRT_APPROX( gain_Q16 ), 16 ); + } else { + gain_Q16 = silk_SMULWW( gain_Q16, gain_Q16 ); + gain_Q16 = silk_SUB_LSHIFT32(silk_SMULWW( psCNG->CNG_smth_Gain_Q16, psCNG->CNG_smth_Gain_Q16 ), gain_Q16, 5 ); + gain_Q16 = silk_LSHIFT32( silk_SQRT_APPROX( gain_Q16 ), 8 ); + } + gain_Q10 = silk_RSHIFT( gain_Q16, 6 ); + + silk_CNG_exc( CNG_sig_Q14 + MAX_LPC_ORDER, psCNG->CNG_exc_buf_Q14, length, &psCNG->rand_seed ); + + /* Convert CNG NLSF to filter representation */ + silk_NLSF2A( A_Q12, psCNG->CNG_smth_NLSF_Q15, psDec->LPC_order, psDec->arch ); + + /* Generate CNG signal, by synthesis filtering */ + silk_memcpy( CNG_sig_Q14, psCNG->CNG_synth_state, MAX_LPC_ORDER * sizeof( opus_int32 ) ); + celt_assert( psDec->LPC_order == 10 || psDec->LPC_order == 16 ); + for( i = 0; i < length; i++ ) { + /* Avoids introducing a bias because silk_SMLAWB() always rounds to -inf */ + LPC_pred_Q10 = silk_RSHIFT( psDec->LPC_order, 1 ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, CNG_sig_Q14[ MAX_LPC_ORDER + i - 1 ], A_Q12[ 0 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, CNG_sig_Q14[ MAX_LPC_ORDER + i - 2 ], A_Q12[ 1 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, CNG_sig_Q14[ MAX_LPC_ORDER + i - 3 ], A_Q12[ 2 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, CNG_sig_Q14[ MAX_LPC_ORDER + i - 4 ], A_Q12[ 3 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, CNG_sig_Q14[ MAX_LPC_ORDER + i - 5 ], A_Q12[ 4 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, CNG_sig_Q14[ MAX_LPC_ORDER + i - 6 ], A_Q12[ 5 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, CNG_sig_Q14[ MAX_LPC_ORDER + i - 7 ], A_Q12[ 6 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, CNG_sig_Q14[ MAX_LPC_ORDER + i - 8 ], A_Q12[ 7 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, CNG_sig_Q14[ MAX_LPC_ORDER + i - 9 ], A_Q12[ 8 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, CNG_sig_Q14[ MAX_LPC_ORDER + i - 10 ], A_Q12[ 9 ] ); + if( psDec->LPC_order == 16 ) { + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, CNG_sig_Q14[ MAX_LPC_ORDER + i - 11 ], A_Q12[ 10 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, CNG_sig_Q14[ MAX_LPC_ORDER + i - 12 ], A_Q12[ 11 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, CNG_sig_Q14[ MAX_LPC_ORDER + i - 13 ], A_Q12[ 12 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, CNG_sig_Q14[ MAX_LPC_ORDER + i - 14 ], A_Q12[ 13 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, CNG_sig_Q14[ MAX_LPC_ORDER + i - 15 ], A_Q12[ 14 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, CNG_sig_Q14[ MAX_LPC_ORDER + i - 16 ], A_Q12[ 15 ] ); + } + + /* Update states */ + CNG_sig_Q14[ MAX_LPC_ORDER + i ] = silk_ADD_SAT32( CNG_sig_Q14[ MAX_LPC_ORDER + i ], silk_LSHIFT_SAT32( LPC_pred_Q10, 4 ) ); + + /* Scale with Gain and add to input signal */ + frame[ i ] = (opus_int16)silk_ADD_SAT16( frame[ i ], silk_SAT16( silk_RSHIFT_ROUND( silk_SMULWW( CNG_sig_Q14[ MAX_LPC_ORDER + i ], gain_Q10 ), 8 ) ) ); + + } + silk_memcpy( psCNG->CNG_synth_state, &CNG_sig_Q14[ length ], MAX_LPC_ORDER * sizeof( opus_int32 ) ); + } else { + silk_memset( psCNG->CNG_synth_state, 0, psDec->LPC_order * sizeof( opus_int32 ) ); + } + RESTORE_STACK; +} diff --git a/media/libopus/silk/HP_variable_cutoff.c b/media/libopus/silk/HP_variable_cutoff.c new file mode 100644 index 0000000000..bbe10f04ce --- /dev/null +++ b/media/libopus/silk/HP_variable_cutoff.c @@ -0,0 +1,77 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif +#ifdef FIXED_POINT +#include "main_FIX.h" +#else +#include "main_FLP.h" +#endif +#include "tuning_parameters.h" + +/* High-pass filter with cutoff frequency adaptation based on pitch lag statistics */ +void silk_HP_variable_cutoff( + silk_encoder_state_Fxx state_Fxx[] /* I/O Encoder states */ +) +{ + opus_int quality_Q15; + opus_int32 pitch_freq_Hz_Q16, pitch_freq_log_Q7, delta_freq_Q7; + silk_encoder_state *psEncC1 = &state_Fxx[ 0 ].sCmn; + + /* Adaptive cutoff frequency: estimate low end of pitch frequency range */ + if( psEncC1->prevSignalType == TYPE_VOICED ) { + /* difference, in log domain */ + pitch_freq_Hz_Q16 = silk_DIV32_16( silk_LSHIFT( silk_MUL( psEncC1->fs_kHz, 1000 ), 16 ), psEncC1->prevLag ); + pitch_freq_log_Q7 = silk_lin2log( pitch_freq_Hz_Q16 ) - ( 16 << 7 ); + + /* adjustment based on quality */ + quality_Q15 = psEncC1->input_quality_bands_Q15[ 0 ]; + pitch_freq_log_Q7 = silk_SMLAWB( pitch_freq_log_Q7, silk_SMULWB( silk_LSHIFT( -quality_Q15, 2 ), quality_Q15 ), + pitch_freq_log_Q7 - ( silk_lin2log( SILK_FIX_CONST( VARIABLE_HP_MIN_CUTOFF_HZ, 16 ) ) - ( 16 << 7 ) ) ); + + /* delta_freq = pitch_freq_log - psEnc->variable_HP_smth1; */ + delta_freq_Q7 = pitch_freq_log_Q7 - silk_RSHIFT( psEncC1->variable_HP_smth1_Q15, 8 ); + if( delta_freq_Q7 < 0 ) { + /* less smoothing for decreasing pitch frequency, to track something close to the minimum */ + delta_freq_Q7 = silk_MUL( delta_freq_Q7, 3 ); + } + + /* limit delta, to reduce impact of outliers in pitch estimation */ + delta_freq_Q7 = silk_LIMIT_32( delta_freq_Q7, -SILK_FIX_CONST( VARIABLE_HP_MAX_DELTA_FREQ, 7 ), SILK_FIX_CONST( VARIABLE_HP_MAX_DELTA_FREQ, 7 ) ); + + /* update smoother */ + psEncC1->variable_HP_smth1_Q15 = silk_SMLAWB( psEncC1->variable_HP_smth1_Q15, + silk_SMULBB( psEncC1->speech_activity_Q8, delta_freq_Q7 ), SILK_FIX_CONST( VARIABLE_HP_SMTH_COEF1, 16 ) ); + + /* limit frequency range */ + psEncC1->variable_HP_smth1_Q15 = silk_LIMIT_32( psEncC1->variable_HP_smth1_Q15, + silk_LSHIFT( silk_lin2log( VARIABLE_HP_MIN_CUTOFF_HZ ), 8 ), + silk_LSHIFT( silk_lin2log( VARIABLE_HP_MAX_CUTOFF_HZ ), 8 ) ); + } +} diff --git a/media/libopus/silk/Inlines.h b/media/libopus/silk/Inlines.h new file mode 100644 index 0000000000..ec986cdfdd --- /dev/null +++ b/media/libopus/silk/Inlines.h @@ -0,0 +1,188 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +/*! \file silk_Inlines.h + * \brief silk_Inlines.h defines OPUS_INLINE signal processing functions. + */ + +#ifndef SILK_FIX_INLINES_H +#define SILK_FIX_INLINES_H + +#ifdef __cplusplus +extern "C" +{ +#endif + +/* count leading zeros of opus_int64 */ +static OPUS_INLINE opus_int32 silk_CLZ64( opus_int64 in ) +{ + opus_int32 in_upper; + + in_upper = (opus_int32)silk_RSHIFT64(in, 32); + if (in_upper == 0) { + /* Search in the lower 32 bits */ + return 32 + silk_CLZ32( (opus_int32) in ); + } else { + /* Search in the upper 32 bits */ + return silk_CLZ32( in_upper ); + } +} + +/* get number of leading zeros and fractional part (the bits right after the leading one */ +static OPUS_INLINE void silk_CLZ_FRAC( + opus_int32 in, /* I input */ + opus_int32 *lz, /* O number of leading zeros */ + opus_int32 *frac_Q7 /* O the 7 bits right after the leading one */ +) +{ + opus_int32 lzeros = silk_CLZ32(in); + + * lz = lzeros; + * frac_Q7 = silk_ROR32(in, 24 - lzeros) & 0x7f; +} + +/* Approximation of square root */ +/* Accuracy: < +/- 10% for output values > 15 */ +/* < +/- 2.5% for output values > 120 */ +static OPUS_INLINE opus_int32 silk_SQRT_APPROX( opus_int32 x ) +{ + opus_int32 y, lz, frac_Q7; + + if( x <= 0 ) { + return 0; + } + + silk_CLZ_FRAC(x, &lz, &frac_Q7); + + if( lz & 1 ) { + y = 32768; + } else { + y = 46214; /* 46214 = sqrt(2) * 32768 */ + } + + /* get scaling right */ + y >>= silk_RSHIFT(lz, 1); + + /* increment using fractional part of input */ + y = silk_SMLAWB(y, y, silk_SMULBB(213, frac_Q7)); + + return y; +} + +/* Divide two int32 values and return result as int32 in a given Q-domain */ +static OPUS_INLINE opus_int32 silk_DIV32_varQ( /* O returns a good approximation of "(a32 << Qres) / b32" */ + const opus_int32 a32, /* I numerator (Q0) */ + const opus_int32 b32, /* I denominator (Q0) */ + const opus_int Qres /* I Q-domain of result (>= 0) */ +) +{ + opus_int a_headrm, b_headrm, lshift; + opus_int32 b32_inv, a32_nrm, b32_nrm, result; + + silk_assert( b32 != 0 ); + silk_assert( Qres >= 0 ); + + /* Compute number of bits head room and normalize inputs */ + a_headrm = silk_CLZ32( silk_abs(a32) ) - 1; + a32_nrm = silk_LSHIFT(a32, a_headrm); /* Q: a_headrm */ + b_headrm = silk_CLZ32( silk_abs(b32) ) - 1; + b32_nrm = silk_LSHIFT(b32, b_headrm); /* Q: b_headrm */ + + /* Inverse of b32, with 14 bits of precision */ + b32_inv = silk_DIV32_16( silk_int32_MAX >> 2, silk_RSHIFT(b32_nrm, 16) ); /* Q: 29 + 16 - b_headrm */ + + /* First approximation */ + result = silk_SMULWB(a32_nrm, b32_inv); /* Q: 29 + a_headrm - b_headrm */ + + /* Compute residual by subtracting product of denominator and first approximation */ + /* It's OK to overflow because the final value of a32_nrm should always be small */ + a32_nrm = silk_SUB32_ovflw(a32_nrm, silk_LSHIFT_ovflw( silk_SMMUL(b32_nrm, result), 3 )); /* Q: a_headrm */ + + /* Refinement */ + result = silk_SMLAWB(result, a32_nrm, b32_inv); /* Q: 29 + a_headrm - b_headrm */ + + /* Convert to Qres domain */ + lshift = 29 + a_headrm - b_headrm - Qres; + if( lshift < 0 ) { + return silk_LSHIFT_SAT32(result, -lshift); + } else { + if( lshift < 32){ + return silk_RSHIFT(result, lshift); + } else { + /* Avoid undefined result */ + return 0; + } + } +} + +/* Invert int32 value and return result as int32 in a given Q-domain */ +static OPUS_INLINE opus_int32 silk_INVERSE32_varQ( /* O returns a good approximation of "(1 << Qres) / b32" */ + const opus_int32 b32, /* I denominator (Q0) */ + const opus_int Qres /* I Q-domain of result (> 0) */ +) +{ + opus_int b_headrm, lshift; + opus_int32 b32_inv, b32_nrm, err_Q32, result; + + silk_assert( b32 != 0 ); + silk_assert( Qres > 0 ); + + /* Compute number of bits head room and normalize input */ + b_headrm = silk_CLZ32( silk_abs(b32) ) - 1; + b32_nrm = silk_LSHIFT(b32, b_headrm); /* Q: b_headrm */ + + /* Inverse of b32, with 14 bits of precision */ + b32_inv = silk_DIV32_16( silk_int32_MAX >> 2, silk_RSHIFT(b32_nrm, 16) ); /* Q: 29 + 16 - b_headrm */ + + /* First approximation */ + result = silk_LSHIFT(b32_inv, 16); /* Q: 61 - b_headrm */ + + /* Compute residual by subtracting product of denominator and first approximation from one */ + err_Q32 = silk_LSHIFT( ((opus_int32)1<<29) - silk_SMULWB(b32_nrm, b32_inv), 3 ); /* Q32 */ + + /* Refinement */ + result = silk_SMLAWW(result, err_Q32, b32_inv); /* Q: 61 - b_headrm */ + + /* Convert to Qres domain */ + lshift = 61 - b_headrm - Qres; + if( lshift <= 0 ) { + return silk_LSHIFT_SAT32(result, -lshift); + } else { + if( lshift < 32){ + return silk_RSHIFT(result, lshift); + }else{ + /* Avoid undefined result */ + return 0; + } + } +} + +#ifdef __cplusplus +} +#endif + +#endif /* SILK_FIX_INLINES_H */ diff --git a/media/libopus/silk/LPC_analysis_filter.c b/media/libopus/silk/LPC_analysis_filter.c new file mode 100644 index 0000000000..d34b5eb709 --- /dev/null +++ b/media/libopus/silk/LPC_analysis_filter.c @@ -0,0 +1,111 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "SigProc_FIX.h" +#include "celt_lpc.h" + +/*******************************************/ +/* LPC analysis filter */ +/* NB! State is kept internally and the */ +/* filter always starts with zero state */ +/* first d output samples are set to zero */ +/*******************************************/ + +/* OPT: Using celt_fir() for this function should be faster, but it may cause + integer overflows in intermediate values (not final results), which the + current implementation silences by casting to unsigned. Enabling + this should be safe in pretty much all cases, even though it is not technically + C89-compliant. */ +#define USE_CELT_FIR 0 + +void silk_LPC_analysis_filter( + opus_int16 *out, /* O Output signal */ + const opus_int16 *in, /* I Input signal */ + const opus_int16 *B, /* I MA prediction coefficients, Q12 [order] */ + const opus_int32 len, /* I Signal length */ + const opus_int32 d, /* I Filter order */ + int arch /* I Run-time architecture */ +) +{ + opus_int j; +#if defined(FIXED_POINT) && USE_CELT_FIR + opus_int16 num[SILK_MAX_ORDER_LPC]; +#else + int ix; + opus_int32 out32_Q12, out32; + const opus_int16 *in_ptr; +#endif + + celt_assert( d >= 6 ); + celt_assert( (d & 1) == 0 ); + celt_assert( d <= len ); + +#if defined(FIXED_POINT) && USE_CELT_FIR + celt_assert( d <= SILK_MAX_ORDER_LPC ); + for ( j = 0; j < d; j++ ) { + num[ j ] = -B[ j ]; + } + celt_fir( in + d, num, out + d, len - d, d, arch ); + for ( j = 0; j < d; j++ ) { + out[ j ] = 0; + } +#else + (void)arch; + for( ix = d; ix < len; ix++ ) { + in_ptr = &in[ ix - 1 ]; + + out32_Q12 = silk_SMULBB( in_ptr[ 0 ], B[ 0 ] ); + /* Allowing wrap around so that two wraps can cancel each other. The rare + cases where the result wraps around can only be triggered by invalid streams*/ + out32_Q12 = silk_SMLABB_ovflw( out32_Q12, in_ptr[ -1 ], B[ 1 ] ); + out32_Q12 = silk_SMLABB_ovflw( out32_Q12, in_ptr[ -2 ], B[ 2 ] ); + out32_Q12 = silk_SMLABB_ovflw( out32_Q12, in_ptr[ -3 ], B[ 3 ] ); + out32_Q12 = silk_SMLABB_ovflw( out32_Q12, in_ptr[ -4 ], B[ 4 ] ); + out32_Q12 = silk_SMLABB_ovflw( out32_Q12, in_ptr[ -5 ], B[ 5 ] ); + for( j = 6; j < d; j += 2 ) { + out32_Q12 = silk_SMLABB_ovflw( out32_Q12, in_ptr[ -j ], B[ j ] ); + out32_Q12 = silk_SMLABB_ovflw( out32_Q12, in_ptr[ -j - 1 ], B[ j + 1 ] ); + } + + /* Subtract prediction */ + out32_Q12 = silk_SUB32_ovflw( silk_LSHIFT( (opus_int32)in_ptr[ 1 ], 12 ), out32_Q12 ); + + /* Scale to Q0 */ + out32 = silk_RSHIFT_ROUND( out32_Q12, 12 ); + + /* Saturate output */ + out[ ix ] = (opus_int16)silk_SAT16( out32 ); + } + + /* Set first d output samples to zero */ + silk_memset( out, 0, d * sizeof( opus_int16 ) ); +#endif +} diff --git a/media/libopus/silk/LPC_fit.c b/media/libopus/silk/LPC_fit.c new file mode 100644 index 0000000000..c0690a1fc9 --- /dev/null +++ b/media/libopus/silk/LPC_fit.c @@ -0,0 +1,82 @@ +/*********************************************************************** +Copyright (c) 2013, Koen Vos. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "SigProc_FIX.h" + +/* Convert int32 coefficients to int16 coefs and make sure there's no wrap-around. + This logic is reused in _celt_lpc(). Any bug fixes should also be applied there. */ +void silk_LPC_fit( + opus_int16 *a_QOUT, /* O Output signal */ + opus_int32 *a_QIN, /* I/O Input signal */ + const opus_int QOUT, /* I Input Q domain */ + const opus_int QIN, /* I Input Q domain */ + const opus_int d /* I Filter order */ +) +{ + opus_int i, k, idx = 0; + opus_int32 maxabs, absval, chirp_Q16; + + /* Limit the maximum absolute value of the prediction coefficients, so that they'll fit in int16 */ + for( i = 0; i < 10; i++ ) { + /* Find maximum absolute value and its index */ + maxabs = 0; + for( k = 0; k < d; k++ ) { + absval = silk_abs( a_QIN[k] ); + if( absval > maxabs ) { + maxabs = absval; + idx = k; + } + } + maxabs = silk_RSHIFT_ROUND( maxabs, QIN - QOUT ); + + if( maxabs > silk_int16_MAX ) { + /* Reduce magnitude of prediction coefficients */ + maxabs = silk_min( maxabs, 163838 ); /* ( silk_int32_MAX >> 14 ) + silk_int16_MAX = 163838 */ + chirp_Q16 = SILK_FIX_CONST( 0.999, 16 ) - silk_DIV32( silk_LSHIFT( maxabs - silk_int16_MAX, 14 ), + silk_RSHIFT32( silk_MUL( maxabs, idx + 1), 2 ) ); + silk_bwexpander_32( a_QIN, d, chirp_Q16 ); + } else { + break; + } + } + + if( i == 10 ) { + /* Reached the last iteration, clip the coefficients */ + for( k = 0; k < d; k++ ) { + a_QOUT[ k ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( a_QIN[ k ], QIN - QOUT ) ); + a_QIN[ k ] = silk_LSHIFT( (opus_int32)a_QOUT[ k ], QIN - QOUT ); + } + } else { + for( k = 0; k < d; k++ ) { + a_QOUT[ k ] = (opus_int16)silk_RSHIFT_ROUND( a_QIN[ k ], QIN - QOUT ); + } + } +} diff --git a/media/libopus/silk/LPC_inv_pred_gain.c b/media/libopus/silk/LPC_inv_pred_gain.c new file mode 100644 index 0000000000..a3746a6ef9 --- /dev/null +++ b/media/libopus/silk/LPC_inv_pred_gain.c @@ -0,0 +1,141 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "SigProc_FIX.h" +#include "define.h" + +#define QA 24 +#define A_LIMIT SILK_FIX_CONST( 0.99975, QA ) + +#define MUL32_FRAC_Q(a32, b32, Q) ((opus_int32)(silk_RSHIFT_ROUND64(silk_SMULL(a32, b32), Q))) + +/* Compute inverse of LPC prediction gain, and */ +/* test if LPC coefficients are stable (all poles within unit circle) */ +static opus_int32 LPC_inverse_pred_gain_QA_c( /* O Returns inverse prediction gain in energy domain, Q30 */ + opus_int32 A_QA[ SILK_MAX_ORDER_LPC ], /* I Prediction coefficients */ + const opus_int order /* I Prediction order */ +) +{ + opus_int k, n, mult2Q; + opus_int32 invGain_Q30, rc_Q31, rc_mult1_Q30, rc_mult2, tmp1, tmp2; + + invGain_Q30 = SILK_FIX_CONST( 1, 30 ); + for( k = order - 1; k > 0; k-- ) { + /* Check for stability */ + if( ( A_QA[ k ] > A_LIMIT ) || ( A_QA[ k ] < -A_LIMIT ) ) { + return 0; + } + + /* Set RC equal to negated AR coef */ + rc_Q31 = -silk_LSHIFT( A_QA[ k ], 31 - QA ); + + /* rc_mult1_Q30 range: [ 1 : 2^30 ] */ + rc_mult1_Q30 = silk_SUB32( SILK_FIX_CONST( 1, 30 ), silk_SMMUL( rc_Q31, rc_Q31 ) ); + silk_assert( rc_mult1_Q30 > ( 1 << 15 ) ); /* reduce A_LIMIT if fails */ + silk_assert( rc_mult1_Q30 <= ( 1 << 30 ) ); + + /* Update inverse gain */ + /* invGain_Q30 range: [ 0 : 2^30 ] */ + invGain_Q30 = silk_LSHIFT( silk_SMMUL( invGain_Q30, rc_mult1_Q30 ), 2 ); + silk_assert( invGain_Q30 >= 0 ); + silk_assert( invGain_Q30 <= ( 1 << 30 ) ); + if( invGain_Q30 < SILK_FIX_CONST( 1.0f / MAX_PREDICTION_POWER_GAIN, 30 ) ) { + return 0; + } + + /* rc_mult2 range: [ 2^30 : silk_int32_MAX ] */ + mult2Q = 32 - silk_CLZ32( silk_abs( rc_mult1_Q30 ) ); + rc_mult2 = silk_INVERSE32_varQ( rc_mult1_Q30, mult2Q + 30 ); + + /* Update AR coefficient */ + for( n = 0; n < (k + 1) >> 1; n++ ) { + opus_int64 tmp64; + tmp1 = A_QA[ n ]; + tmp2 = A_QA[ k - n - 1 ]; + tmp64 = silk_RSHIFT_ROUND64( silk_SMULL( silk_SUB_SAT32(tmp1, + MUL32_FRAC_Q( tmp2, rc_Q31, 31 ) ), rc_mult2 ), mult2Q); + if( tmp64 > silk_int32_MAX || tmp64 < silk_int32_MIN ) { + return 0; + } + A_QA[ n ] = ( opus_int32 )tmp64; + tmp64 = silk_RSHIFT_ROUND64( silk_SMULL( silk_SUB_SAT32(tmp2, + MUL32_FRAC_Q( tmp1, rc_Q31, 31 ) ), rc_mult2), mult2Q); + if( tmp64 > silk_int32_MAX || tmp64 < silk_int32_MIN ) { + return 0; + } + A_QA[ k - n - 1 ] = ( opus_int32 )tmp64; + } + } + + /* Check for stability */ + if( ( A_QA[ k ] > A_LIMIT ) || ( A_QA[ k ] < -A_LIMIT ) ) { + return 0; + } + + /* Set RC equal to negated AR coef */ + rc_Q31 = -silk_LSHIFT( A_QA[ 0 ], 31 - QA ); + + /* Range: [ 1 : 2^30 ] */ + rc_mult1_Q30 = silk_SUB32( SILK_FIX_CONST( 1, 30 ), silk_SMMUL( rc_Q31, rc_Q31 ) ); + + /* Update inverse gain */ + /* Range: [ 0 : 2^30 ] */ + invGain_Q30 = silk_LSHIFT( silk_SMMUL( invGain_Q30, rc_mult1_Q30 ), 2 ); + silk_assert( invGain_Q30 >= 0 ); + silk_assert( invGain_Q30 <= ( 1 << 30 ) ); + if( invGain_Q30 < SILK_FIX_CONST( 1.0f / MAX_PREDICTION_POWER_GAIN, 30 ) ) { + return 0; + } + + return invGain_Q30; +} + +/* For input in Q12 domain */ +opus_int32 silk_LPC_inverse_pred_gain_c( /* O Returns inverse prediction gain in energy domain, Q30 */ + const opus_int16 *A_Q12, /* I Prediction coefficients, Q12 [order] */ + const opus_int order /* I Prediction order */ +) +{ + opus_int k; + opus_int32 Atmp_QA[ SILK_MAX_ORDER_LPC ]; + opus_int32 DC_resp = 0; + + /* Increase Q domain of the AR coefficients */ + for( k = 0; k < order; k++ ) { + DC_resp += (opus_int32)A_Q12[ k ]; + Atmp_QA[ k ] = silk_LSHIFT32( (opus_int32)A_Q12[ k ], QA - 12 ); + } + /* If the DC is unstable, we don't even need to do the full calculations */ + if( DC_resp >= 4096 ) { + return 0; + } + return LPC_inverse_pred_gain_QA_c( Atmp_QA, order ); +} diff --git a/media/libopus/silk/LP_variable_cutoff.c b/media/libopus/silk/LP_variable_cutoff.c new file mode 100644 index 0000000000..79112ad354 --- /dev/null +++ b/media/libopus/silk/LP_variable_cutoff.c @@ -0,0 +1,135 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +/* + Elliptic/Cauer filters designed with 0.1 dB passband ripple, + 80 dB minimum stopband attenuation, and + [0.95 : 0.15 : 0.35] normalized cut off frequencies. +*/ + +#include "main.h" + +/* Helper function, interpolates the filter taps */ +static OPUS_INLINE void silk_LP_interpolate_filter_taps( + opus_int32 B_Q28[ TRANSITION_NB ], + opus_int32 A_Q28[ TRANSITION_NA ], + const opus_int ind, + const opus_int32 fac_Q16 +) +{ + opus_int nb, na; + + if( ind < TRANSITION_INT_NUM - 1 ) { + if( fac_Q16 > 0 ) { + if( fac_Q16 < 32768 ) { /* fac_Q16 is in range of a 16-bit int */ + /* Piece-wise linear interpolation of B and A */ + for( nb = 0; nb < TRANSITION_NB; nb++ ) { + B_Q28[ nb ] = silk_SMLAWB( + silk_Transition_LP_B_Q28[ ind ][ nb ], + silk_Transition_LP_B_Q28[ ind + 1 ][ nb ] - + silk_Transition_LP_B_Q28[ ind ][ nb ], + fac_Q16 ); + } + for( na = 0; na < TRANSITION_NA; na++ ) { + A_Q28[ na ] = silk_SMLAWB( + silk_Transition_LP_A_Q28[ ind ][ na ], + silk_Transition_LP_A_Q28[ ind + 1 ][ na ] - + silk_Transition_LP_A_Q28[ ind ][ na ], + fac_Q16 ); + } + } else { /* ( fac_Q16 - ( 1 << 16 ) ) is in range of a 16-bit int */ + silk_assert( fac_Q16 - ( 1 << 16 ) == silk_SAT16( fac_Q16 - ( 1 << 16 ) ) ); + /* Piece-wise linear interpolation of B and A */ + for( nb = 0; nb < TRANSITION_NB; nb++ ) { + B_Q28[ nb ] = silk_SMLAWB( + silk_Transition_LP_B_Q28[ ind + 1 ][ nb ], + silk_Transition_LP_B_Q28[ ind + 1 ][ nb ] - + silk_Transition_LP_B_Q28[ ind ][ nb ], + fac_Q16 - ( (opus_int32)1 << 16 ) ); + } + for( na = 0; na < TRANSITION_NA; na++ ) { + A_Q28[ na ] = silk_SMLAWB( + silk_Transition_LP_A_Q28[ ind + 1 ][ na ], + silk_Transition_LP_A_Q28[ ind + 1 ][ na ] - + silk_Transition_LP_A_Q28[ ind ][ na ], + fac_Q16 - ( (opus_int32)1 << 16 ) ); + } + } + } else { + silk_memcpy( B_Q28, silk_Transition_LP_B_Q28[ ind ], TRANSITION_NB * sizeof( opus_int32 ) ); + silk_memcpy( A_Q28, silk_Transition_LP_A_Q28[ ind ], TRANSITION_NA * sizeof( opus_int32 ) ); + } + } else { + silk_memcpy( B_Q28, silk_Transition_LP_B_Q28[ TRANSITION_INT_NUM - 1 ], TRANSITION_NB * sizeof( opus_int32 ) ); + silk_memcpy( A_Q28, silk_Transition_LP_A_Q28[ TRANSITION_INT_NUM - 1 ], TRANSITION_NA * sizeof( opus_int32 ) ); + } +} + +/* Low-pass filter with variable cutoff frequency based on */ +/* piece-wise linear interpolation between elliptic filters */ +/* Start by setting psEncC->mode <> 0; */ +/* Deactivate by setting psEncC->mode = 0; */ +void silk_LP_variable_cutoff( + silk_LP_state *psLP, /* I/O LP filter state */ + opus_int16 *frame, /* I/O Low-pass filtered output signal */ + const opus_int frame_length /* I Frame length */ +) +{ + opus_int32 B_Q28[ TRANSITION_NB ], A_Q28[ TRANSITION_NA ], fac_Q16 = 0; + opus_int ind = 0; + + silk_assert( psLP->transition_frame_no >= 0 && psLP->transition_frame_no <= TRANSITION_FRAMES ); + + /* Run filter if needed */ + if( psLP->mode != 0 ) { + /* Calculate index and interpolation factor for interpolation */ +#if( TRANSITION_INT_STEPS == 64 ) + fac_Q16 = silk_LSHIFT( TRANSITION_FRAMES - psLP->transition_frame_no, 16 - 6 ); +#else + fac_Q16 = silk_DIV32_16( silk_LSHIFT( TRANSITION_FRAMES - psLP->transition_frame_no, 16 ), TRANSITION_FRAMES ); +#endif + ind = silk_RSHIFT( fac_Q16, 16 ); + fac_Q16 -= silk_LSHIFT( ind, 16 ); + + silk_assert( ind >= 0 ); + silk_assert( ind < TRANSITION_INT_NUM ); + + /* Interpolate filter coefficients */ + silk_LP_interpolate_filter_taps( B_Q28, A_Q28, ind, fac_Q16 ); + + /* Update transition frame number for next frame */ + psLP->transition_frame_no = silk_LIMIT( psLP->transition_frame_no + psLP->mode, 0, TRANSITION_FRAMES ); + + /* ARMA low-pass filtering */ + silk_assert( TRANSITION_NB == 3 && TRANSITION_NA == 2 ); + silk_biquad_alt_stride1( frame, B_Q28, A_Q28, psLP->In_LP_State, frame, frame_length); + } +} diff --git a/media/libopus/silk/MacroCount.h b/media/libopus/silk/MacroCount.h new file mode 100644 index 0000000000..dab2f57a68 --- /dev/null +++ b/media/libopus/silk/MacroCount.h @@ -0,0 +1,710 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef SIGPROCFIX_API_MACROCOUNT_H +#define SIGPROCFIX_API_MACROCOUNT_H + +#ifdef silk_MACRO_COUNT +#include <stdio.h> +#define varDefine opus_int64 ops_count = 0; + +extern opus_int64 ops_count; + +static OPUS_INLINE opus_int64 silk_SaveCount(){ + return(ops_count); +} + +static OPUS_INLINE opus_int64 silk_SaveResetCount(){ + opus_int64 ret; + + ret = ops_count; + ops_count = 0; + return(ret); +} + +static OPUS_INLINE silk_PrintCount(){ + printf("ops_count = %d \n ", (opus_int32)ops_count); +} + +#undef silk_MUL +static OPUS_INLINE opus_int32 silk_MUL(opus_int32 a32, opus_int32 b32){ + opus_int32 ret; + ops_count += 4; + ret = a32 * b32; + return ret; +} + +#undef silk_MUL_uint +static OPUS_INLINE opus_uint32 silk_MUL_uint(opus_uint32 a32, opus_uint32 b32){ + opus_uint32 ret; + ops_count += 4; + ret = a32 * b32; + return ret; +} +#undef silk_MLA +static OPUS_INLINE opus_int32 silk_MLA(opus_int32 a32, opus_int32 b32, opus_int32 c32){ + opus_int32 ret; + ops_count += 4; + ret = a32 + b32 * c32; + return ret; +} + +#undef silk_MLA_uint +static OPUS_INLINE opus_int32 silk_MLA_uint(opus_uint32 a32, opus_uint32 b32, opus_uint32 c32){ + opus_uint32 ret; + ops_count += 4; + ret = a32 + b32 * c32; + return ret; +} + +#undef silk_SMULWB +static OPUS_INLINE opus_int32 silk_SMULWB(opus_int32 a32, opus_int32 b32){ + opus_int32 ret; + ops_count += 5; + ret = (a32 >> 16) * (opus_int32)((opus_int16)b32) + (((a32 & 0x0000FFFF) * (opus_int32)((opus_int16)b32)) >> 16); + return ret; +} +#undef silk_SMLAWB +static OPUS_INLINE opus_int32 silk_SMLAWB(opus_int32 a32, opus_int32 b32, opus_int32 c32){ + opus_int32 ret; + ops_count += 5; + ret = ((a32) + ((((b32) >> 16) * (opus_int32)((opus_int16)(c32))) + ((((b32) & 0x0000FFFF) * (opus_int32)((opus_int16)(c32))) >> 16))); + return ret; +} + +#undef silk_SMULWT +static OPUS_INLINE opus_int32 silk_SMULWT(opus_int32 a32, opus_int32 b32){ + opus_int32 ret; + ops_count += 4; + ret = (a32 >> 16) * (b32 >> 16) + (((a32 & 0x0000FFFF) * (b32 >> 16)) >> 16); + return ret; +} +#undef silk_SMLAWT +static OPUS_INLINE opus_int32 silk_SMLAWT(opus_int32 a32, opus_int32 b32, opus_int32 c32){ + opus_int32 ret; + ops_count += 4; + ret = a32 + ((b32 >> 16) * (c32 >> 16)) + (((b32 & 0x0000FFFF) * ((c32 >> 16)) >> 16)); + return ret; +} + +#undef silk_SMULBB +static OPUS_INLINE opus_int32 silk_SMULBB(opus_int32 a32, opus_int32 b32){ + opus_int32 ret; + ops_count += 1; + ret = (opus_int32)((opus_int16)a32) * (opus_int32)((opus_int16)b32); + return ret; +} +#undef silk_SMLABB +static OPUS_INLINE opus_int32 silk_SMLABB(opus_int32 a32, opus_int32 b32, opus_int32 c32){ + opus_int32 ret; + ops_count += 1; + ret = a32 + (opus_int32)((opus_int16)b32) * (opus_int32)((opus_int16)c32); + return ret; +} + +#undef silk_SMULBT +static OPUS_INLINE opus_int32 silk_SMULBT(opus_int32 a32, opus_int32 b32 ){ + opus_int32 ret; + ops_count += 4; + ret = ((opus_int32)((opus_int16)a32)) * (b32 >> 16); + return ret; +} + +#undef silk_SMLABT +static OPUS_INLINE opus_int32 silk_SMLABT(opus_int32 a32, opus_int32 b32, opus_int32 c32){ + opus_int32 ret; + ops_count += 1; + ret = a32 + ((opus_int32)((opus_int16)b32)) * (c32 >> 16); + return ret; +} + +#undef silk_SMULTT +static OPUS_INLINE opus_int32 silk_SMULTT(opus_int32 a32, opus_int32 b32){ + opus_int32 ret; + ops_count += 1; + ret = (a32 >> 16) * (b32 >> 16); + return ret; +} + +#undef silk_SMLATT +static OPUS_INLINE opus_int32 silk_SMLATT(opus_int32 a32, opus_int32 b32, opus_int32 c32){ + opus_int32 ret; + ops_count += 1; + ret = a32 + (b32 >> 16) * (c32 >> 16); + return ret; +} + + +/* multiply-accumulate macros that allow overflow in the addition (ie, no asserts in debug mode)*/ +#undef silk_MLA_ovflw +#define silk_MLA_ovflw silk_MLA + +#undef silk_SMLABB_ovflw +#define silk_SMLABB_ovflw silk_SMLABB + +#undef silk_SMLABT_ovflw +#define silk_SMLABT_ovflw silk_SMLABT + +#undef silk_SMLATT_ovflw +#define silk_SMLATT_ovflw silk_SMLATT + +#undef silk_SMLAWB_ovflw +#define silk_SMLAWB_ovflw silk_SMLAWB + +#undef silk_SMLAWT_ovflw +#define silk_SMLAWT_ovflw silk_SMLAWT + +#undef silk_SMULL +static OPUS_INLINE opus_int64 silk_SMULL(opus_int32 a32, opus_int32 b32){ + opus_int64 ret; + ops_count += 8; + ret = ((opus_int64)(a32) * /*(opus_int64)*/(b32)); + return ret; +} + +#undef silk_SMLAL +static OPUS_INLINE opus_int64 silk_SMLAL(opus_int64 a64, opus_int32 b32, opus_int32 c32){ + opus_int64 ret; + ops_count += 8; + ret = a64 + ((opus_int64)(b32) * /*(opus_int64)*/(c32)); + return ret; +} +#undef silk_SMLALBB +static OPUS_INLINE opus_int64 silk_SMLALBB(opus_int64 a64, opus_int16 b16, opus_int16 c16){ + opus_int64 ret; + ops_count += 4; + ret = a64 + ((opus_int64)(b16) * /*(opus_int64)*/(c16)); + return ret; +} + +#undef SigProcFIX_CLZ16 +static OPUS_INLINE opus_int32 SigProcFIX_CLZ16(opus_int16 in16) +{ + opus_int32 out32 = 0; + ops_count += 10; + if( in16 == 0 ) { + return 16; + } + /* test nibbles */ + if( in16 & 0xFF00 ) { + if( in16 & 0xF000 ) { + in16 >>= 12; + } else { + out32 += 4; + in16 >>= 8; + } + } else { + if( in16 & 0xFFF0 ) { + out32 += 8; + in16 >>= 4; + } else { + out32 += 12; + } + } + /* test bits and return */ + if( in16 & 0xC ) { + if( in16 & 0x8 ) + return out32 + 0; + else + return out32 + 1; + } else { + if( in16 & 0xE ) + return out32 + 2; + else + return out32 + 3; + } +} + +#undef SigProcFIX_CLZ32 +static OPUS_INLINE opus_int32 SigProcFIX_CLZ32(opus_int32 in32) +{ + /* test highest 16 bits and convert to opus_int16 */ + ops_count += 2; + if( in32 & 0xFFFF0000 ) { + return SigProcFIX_CLZ16((opus_int16)(in32 >> 16)); + } else { + return SigProcFIX_CLZ16((opus_int16)in32) + 16; + } +} + +#undef silk_DIV32 +static OPUS_INLINE opus_int32 silk_DIV32(opus_int32 a32, opus_int32 b32){ + ops_count += 64; + return a32 / b32; +} + +#undef silk_DIV32_16 +static OPUS_INLINE opus_int32 silk_DIV32_16(opus_int32 a32, opus_int32 b32){ + ops_count += 32; + return a32 / b32; +} + +#undef silk_SAT8 +static OPUS_INLINE opus_int8 silk_SAT8(opus_int64 a){ + opus_int8 tmp; + ops_count += 1; + tmp = (opus_int8)((a) > silk_int8_MAX ? silk_int8_MAX : \ + ((a) < silk_int8_MIN ? silk_int8_MIN : (a))); + return(tmp); +} + +#undef silk_SAT16 +static OPUS_INLINE opus_int16 silk_SAT16(opus_int64 a){ + opus_int16 tmp; + ops_count += 1; + tmp = (opus_int16)((a) > silk_int16_MAX ? silk_int16_MAX : \ + ((a) < silk_int16_MIN ? silk_int16_MIN : (a))); + return(tmp); +} +#undef silk_SAT32 +static OPUS_INLINE opus_int32 silk_SAT32(opus_int64 a){ + opus_int32 tmp; + ops_count += 1; + tmp = (opus_int32)((a) > silk_int32_MAX ? silk_int32_MAX : \ + ((a) < silk_int32_MIN ? silk_int32_MIN : (a))); + return(tmp); +} +#undef silk_POS_SAT32 +static OPUS_INLINE opus_int32 silk_POS_SAT32(opus_int64 a){ + opus_int32 tmp; + ops_count += 1; + tmp = (opus_int32)((a) > silk_int32_MAX ? silk_int32_MAX : (a)); + return(tmp); +} + +#undef silk_ADD_POS_SAT8 +static OPUS_INLINE opus_int8 silk_ADD_POS_SAT8(opus_int64 a, opus_int64 b){ + opus_int8 tmp; + ops_count += 1; + tmp = (opus_int8)((((a)+(b)) & 0x80) ? silk_int8_MAX : ((a)+(b))); + return(tmp); +} +#undef silk_ADD_POS_SAT16 +static OPUS_INLINE opus_int16 silk_ADD_POS_SAT16(opus_int64 a, opus_int64 b){ + opus_int16 tmp; + ops_count += 1; + tmp = (opus_int16)((((a)+(b)) & 0x8000) ? silk_int16_MAX : ((a)+(b))); + return(tmp); +} + +#undef silk_ADD_POS_SAT32 +static OPUS_INLINE opus_int32 silk_ADD_POS_SAT32(opus_int64 a, opus_int64 b){ + opus_int32 tmp; + ops_count += 1; + tmp = (opus_int32)((((a)+(b)) & 0x80000000) ? silk_int32_MAX : ((a)+(b))); + return(tmp); +} + +#undef silk_LSHIFT8 +static OPUS_INLINE opus_int8 silk_LSHIFT8(opus_int8 a, opus_int32 shift){ + opus_int8 ret; + ops_count += 1; + ret = a << shift; + return ret; +} +#undef silk_LSHIFT16 +static OPUS_INLINE opus_int16 silk_LSHIFT16(opus_int16 a, opus_int32 shift){ + opus_int16 ret; + ops_count += 1; + ret = a << shift; + return ret; +} +#undef silk_LSHIFT32 +static OPUS_INLINE opus_int32 silk_LSHIFT32(opus_int32 a, opus_int32 shift){ + opus_int32 ret; + ops_count += 1; + ret = a << shift; + return ret; +} +#undef silk_LSHIFT64 +static OPUS_INLINE opus_int64 silk_LSHIFT64(opus_int64 a, opus_int shift){ + ops_count += 1; + return a << shift; +} + +#undef silk_LSHIFT_ovflw +static OPUS_INLINE opus_int32 silk_LSHIFT_ovflw(opus_int32 a, opus_int32 shift){ + ops_count += 1; + return a << shift; +} + +#undef silk_LSHIFT_uint +static OPUS_INLINE opus_uint32 silk_LSHIFT_uint(opus_uint32 a, opus_int32 shift){ + opus_uint32 ret; + ops_count += 1; + ret = a << shift; + return ret; +} + +#undef silk_RSHIFT8 +static OPUS_INLINE opus_int8 silk_RSHIFT8(opus_int8 a, opus_int32 shift){ + ops_count += 1; + return a >> shift; +} +#undef silk_RSHIFT16 +static OPUS_INLINE opus_int16 silk_RSHIFT16(opus_int16 a, opus_int32 shift){ + ops_count += 1; + return a >> shift; +} +#undef silk_RSHIFT32 +static OPUS_INLINE opus_int32 silk_RSHIFT32(opus_int32 a, opus_int32 shift){ + ops_count += 1; + return a >> shift; +} +#undef silk_RSHIFT64 +static OPUS_INLINE opus_int64 silk_RSHIFT64(opus_int64 a, opus_int64 shift){ + ops_count += 1; + return a >> shift; +} + +#undef silk_RSHIFT_uint +static OPUS_INLINE opus_uint32 silk_RSHIFT_uint(opus_uint32 a, opus_int32 shift){ + ops_count += 1; + return a >> shift; +} + +#undef silk_ADD_LSHIFT +static OPUS_INLINE opus_int32 silk_ADD_LSHIFT(opus_int32 a, opus_int32 b, opus_int32 shift){ + opus_int32 ret; + ops_count += 1; + ret = a + (b << shift); + return ret; /* shift >= 0*/ +} +#undef silk_ADD_LSHIFT32 +static OPUS_INLINE opus_int32 silk_ADD_LSHIFT32(opus_int32 a, opus_int32 b, opus_int32 shift){ + opus_int32 ret; + ops_count += 1; + ret = a + (b << shift); + return ret; /* shift >= 0*/ +} +#undef silk_ADD_LSHIFT_uint +static OPUS_INLINE opus_uint32 silk_ADD_LSHIFT_uint(opus_uint32 a, opus_uint32 b, opus_int32 shift){ + opus_uint32 ret; + ops_count += 1; + ret = a + (b << shift); + return ret; /* shift >= 0*/ +} +#undef silk_ADD_RSHIFT +static OPUS_INLINE opus_int32 silk_ADD_RSHIFT(opus_int32 a, opus_int32 b, opus_int32 shift){ + opus_int32 ret; + ops_count += 1; + ret = a + (b >> shift); + return ret; /* shift > 0*/ +} +#undef silk_ADD_RSHIFT32 +static OPUS_INLINE opus_int32 silk_ADD_RSHIFT32(opus_int32 a, opus_int32 b, opus_int32 shift){ + opus_int32 ret; + ops_count += 1; + ret = a + (b >> shift); + return ret; /* shift > 0*/ +} +#undef silk_ADD_RSHIFT_uint +static OPUS_INLINE opus_uint32 silk_ADD_RSHIFT_uint(opus_uint32 a, opus_uint32 b, opus_int32 shift){ + opus_uint32 ret; + ops_count += 1; + ret = a + (b >> shift); + return ret; /* shift > 0*/ +} +#undef silk_SUB_LSHIFT32 +static OPUS_INLINE opus_int32 silk_SUB_LSHIFT32(opus_int32 a, opus_int32 b, opus_int32 shift){ + opus_int32 ret; + ops_count += 1; + ret = a - (b << shift); + return ret; /* shift >= 0*/ +} +#undef silk_SUB_RSHIFT32 +static OPUS_INLINE opus_int32 silk_SUB_RSHIFT32(opus_int32 a, opus_int32 b, opus_int32 shift){ + opus_int32 ret; + ops_count += 1; + ret = a - (b >> shift); + return ret; /* shift > 0*/ +} + +#undef silk_RSHIFT_ROUND +static OPUS_INLINE opus_int32 silk_RSHIFT_ROUND(opus_int32 a, opus_int32 shift){ + opus_int32 ret; + ops_count += 3; + ret = shift == 1 ? (a >> 1) + (a & 1) : ((a >> (shift - 1)) + 1) >> 1; + return ret; +} + +#undef silk_RSHIFT_ROUND64 +static OPUS_INLINE opus_int64 silk_RSHIFT_ROUND64(opus_int64 a, opus_int32 shift){ + opus_int64 ret; + ops_count += 6; + ret = shift == 1 ? (a >> 1) + (a & 1) : ((a >> (shift - 1)) + 1) >> 1; + return ret; +} + +#undef silk_abs_int64 +static OPUS_INLINE opus_int64 silk_abs_int64(opus_int64 a){ + ops_count += 1; + return (((a) > 0) ? (a) : -(a)); /* Be careful, silk_abs returns wrong when input equals to silk_intXX_MIN*/ +} + +#undef silk_abs_int32 +static OPUS_INLINE opus_int32 silk_abs_int32(opus_int32 a){ + ops_count += 1; + return silk_abs(a); +} + + +#undef silk_min +static silk_min(a, b){ + ops_count += 1; + return (((a) < (b)) ? (a) : (b)); +} +#undef silk_max +static silk_max(a, b){ + ops_count += 1; + return (((a) > (b)) ? (a) : (b)); +} +#undef silk_sign +static silk_sign(a){ + ops_count += 1; + return ((a) > 0 ? 1 : ( (a) < 0 ? -1 : 0 )); +} + +#undef silk_ADD16 +static OPUS_INLINE opus_int16 silk_ADD16(opus_int16 a, opus_int16 b){ + opus_int16 ret; + ops_count += 1; + ret = a + b; + return ret; +} + +#undef silk_ADD32 +static OPUS_INLINE opus_int32 silk_ADD32(opus_int32 a, opus_int32 b){ + opus_int32 ret; + ops_count += 1; + ret = a + b; + return ret; +} + +#undef silk_ADD64 +static OPUS_INLINE opus_int64 silk_ADD64(opus_int64 a, opus_int64 b){ + opus_int64 ret; + ops_count += 2; + ret = a + b; + return ret; +} + +#undef silk_SUB16 +static OPUS_INLINE opus_int16 silk_SUB16(opus_int16 a, opus_int16 b){ + opus_int16 ret; + ops_count += 1; + ret = a - b; + return ret; +} + +#undef silk_SUB32 +static OPUS_INLINE opus_int32 silk_SUB32(opus_int32 a, opus_int32 b){ + opus_int32 ret; + ops_count += 1; + ret = a - b; + return ret; +} + +#undef silk_SUB64 +static OPUS_INLINE opus_int64 silk_SUB64(opus_int64 a, opus_int64 b){ + opus_int64 ret; + ops_count += 2; + ret = a - b; + return ret; +} + +#undef silk_ADD_SAT16 +static OPUS_INLINE opus_int16 silk_ADD_SAT16( opus_int16 a16, opus_int16 b16 ) { + opus_int16 res; + /* Nb will be counted in AKP_add32 and silk_SAT16*/ + res = (opus_int16)silk_SAT16( silk_ADD32( (opus_int32)(a16), (b16) ) ); + return res; +} + +#undef silk_ADD_SAT32 +static OPUS_INLINE opus_int32 silk_ADD_SAT32(opus_int32 a32, opus_int32 b32){ + opus_int32 res; + ops_count += 1; + res = ((((a32) + (b32)) & 0x80000000) == 0 ? \ + ((((a32) & (b32)) & 0x80000000) != 0 ? silk_int32_MIN : (a32)+(b32)) : \ + ((((a32) | (b32)) & 0x80000000) == 0 ? silk_int32_MAX : (a32)+(b32)) ); + return res; +} + +#undef silk_ADD_SAT64 +static OPUS_INLINE opus_int64 silk_ADD_SAT64( opus_int64 a64, opus_int64 b64 ) { + opus_int64 res; + ops_count += 1; + res = ((((a64) + (b64)) & 0x8000000000000000LL) == 0 ? \ + ((((a64) & (b64)) & 0x8000000000000000LL) != 0 ? silk_int64_MIN : (a64)+(b64)) : \ + ((((a64) | (b64)) & 0x8000000000000000LL) == 0 ? silk_int64_MAX : (a64)+(b64)) ); + return res; +} + +#undef silk_SUB_SAT16 +static OPUS_INLINE opus_int16 silk_SUB_SAT16( opus_int16 a16, opus_int16 b16 ) { + opus_int16 res; + silk_assert(0); + /* Nb will be counted in sub-macros*/ + res = (opus_int16)silk_SAT16( silk_SUB32( (opus_int32)(a16), (b16) ) ); + return res; +} + +#undef silk_SUB_SAT32 +static OPUS_INLINE opus_int32 silk_SUB_SAT32( opus_int32 a32, opus_int32 b32 ) { + opus_int32 res; + ops_count += 1; + res = ((((a32)-(b32)) & 0x80000000) == 0 ? \ + (( (a32) & ((b32)^0x80000000) & 0x80000000) ? silk_int32_MIN : (a32)-(b32)) : \ + ((((a32)^0x80000000) & (b32) & 0x80000000) ? silk_int32_MAX : (a32)-(b32)) ); + return res; +} + +#undef silk_SUB_SAT64 +static OPUS_INLINE opus_int64 silk_SUB_SAT64( opus_int64 a64, opus_int64 b64 ) { + opus_int64 res; + ops_count += 1; + res = ((((a64)-(b64)) & 0x8000000000000000LL) == 0 ? \ + (( (a64) & ((b64)^0x8000000000000000LL) & 0x8000000000000000LL) ? silk_int64_MIN : (a64)-(b64)) : \ + ((((a64)^0x8000000000000000LL) & (b64) & 0x8000000000000000LL) ? silk_int64_MAX : (a64)-(b64)) ); + + return res; +} + +#undef silk_SMULWW +static OPUS_INLINE opus_int32 silk_SMULWW(opus_int32 a32, opus_int32 b32){ + opus_int32 ret; + /* Nb will be counted in sub-macros*/ + ret = silk_MLA(silk_SMULWB((a32), (b32)), (a32), silk_RSHIFT_ROUND((b32), 16)); + return ret; +} + +#undef silk_SMLAWW +static OPUS_INLINE opus_int32 silk_SMLAWW(opus_int32 a32, opus_int32 b32, opus_int32 c32){ + opus_int32 ret; + /* Nb will be counted in sub-macros*/ + ret = silk_MLA(silk_SMLAWB((a32), (b32), (c32)), (b32), silk_RSHIFT_ROUND((c32), 16)); + return ret; +} + +#undef silk_min_int +static OPUS_INLINE opus_int silk_min_int(opus_int a, opus_int b) +{ + ops_count += 1; + return (((a) < (b)) ? (a) : (b)); +} + +#undef silk_min_16 +static OPUS_INLINE opus_int16 silk_min_16(opus_int16 a, opus_int16 b) +{ + ops_count += 1; + return (((a) < (b)) ? (a) : (b)); +} +#undef silk_min_32 +static OPUS_INLINE opus_int32 silk_min_32(opus_int32 a, opus_int32 b) +{ + ops_count += 1; + return (((a) < (b)) ? (a) : (b)); +} +#undef silk_min_64 +static OPUS_INLINE opus_int64 silk_min_64(opus_int64 a, opus_int64 b) +{ + ops_count += 1; + return (((a) < (b)) ? (a) : (b)); +} + +/* silk_min() versions with typecast in the function call */ +#undef silk_max_int +static OPUS_INLINE opus_int silk_max_int(opus_int a, opus_int b) +{ + ops_count += 1; + return (((a) > (b)) ? (a) : (b)); +} +#undef silk_max_16 +static OPUS_INLINE opus_int16 silk_max_16(opus_int16 a, opus_int16 b) +{ + ops_count += 1; + return (((a) > (b)) ? (a) : (b)); +} +#undef silk_max_32 +static OPUS_INLINE opus_int32 silk_max_32(opus_int32 a, opus_int32 b) +{ + ops_count += 1; + return (((a) > (b)) ? (a) : (b)); +} + +#undef silk_max_64 +static OPUS_INLINE opus_int64 silk_max_64(opus_int64 a, opus_int64 b) +{ + ops_count += 1; + return (((a) > (b)) ? (a) : (b)); +} + + +#undef silk_LIMIT_int +static OPUS_INLINE opus_int silk_LIMIT_int(opus_int a, opus_int limit1, opus_int limit2) +{ + opus_int ret; + ops_count += 6; + + ret = ((limit1) > (limit2) ? ((a) > (limit1) ? (limit1) : ((a) < (limit2) ? (limit2) : (a))) \ + : ((a) > (limit2) ? (limit2) : ((a) < (limit1) ? (limit1) : (a)))); + + return(ret); +} + +#undef silk_LIMIT_16 +static OPUS_INLINE opus_int16 silk_LIMIT_16(opus_int16 a, opus_int16 limit1, opus_int16 limit2) +{ + opus_int16 ret; + ops_count += 6; + + ret = ((limit1) > (limit2) ? ((a) > (limit1) ? (limit1) : ((a) < (limit2) ? (limit2) : (a))) \ + : ((a) > (limit2) ? (limit2) : ((a) < (limit1) ? (limit1) : (a)))); + +return(ret); +} + + +#undef silk_LIMIT_32 +static OPUS_INLINE opus_int32 silk_LIMIT_32(opus_int32 a, opus_int32 limit1, opus_int32 limit2) +{ + opus_int32 ret; + ops_count += 6; + + ret = ((limit1) > (limit2) ? ((a) > (limit1) ? (limit1) : ((a) < (limit2) ? (limit2) : (a))) \ + : ((a) > (limit2) ? (limit2) : ((a) < (limit1) ? (limit1) : (a)))); + return(ret); +} + +#else +#define varDefine +#define silk_SaveCount() + +#endif +#endif + diff --git a/media/libopus/silk/MacroDebug.h b/media/libopus/silk/MacroDebug.h new file mode 100644 index 0000000000..3110da9a75 --- /dev/null +++ b/media/libopus/silk/MacroDebug.h @@ -0,0 +1,945 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Copyright (C) 2012 Xiph.Org Foundation +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef MACRO_DEBUG_H +#define MACRO_DEBUG_H + +/* Redefine macro functions with extensive assertion in DEBUG mode. + As functions can't be undefined, this file can't work with SigProcFIX_MacroCount.h */ + +#if ( defined (FIXED_DEBUG) || ( 0 && defined (_DEBUG) ) ) && !defined (silk_MACRO_COUNT) + +#undef silk_ADD16 +#define silk_ADD16(a,b) silk_ADD16_((a), (b), __FILE__, __LINE__) +static OPUS_INLINE opus_int16 silk_ADD16_(opus_int16 a, opus_int16 b, char *file, int line){ + opus_int16 ret; + + ret = a + b; + if ( ret != silk_ADD_SAT16( a, b ) ) + { + fprintf (stderr, "silk_ADD16(%d, %d) in %s: line %d\n", a, b, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return ret; +} + +#undef silk_ADD32 +#define silk_ADD32(a,b) silk_ADD32_((a), (b), __FILE__, __LINE__) +static OPUS_INLINE opus_int32 silk_ADD32_(opus_int32 a, opus_int32 b, char *file, int line){ + opus_int32 ret; + + ret = (opus_int32)((opus_uint32)a + (opus_uint32)b); + if ( ret != silk_ADD_SAT32( a, b ) ) + { + fprintf (stderr, "silk_ADD32(%d, %d) in %s: line %d\n", a, b, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return ret; +} + +#undef silk_ADD64 +#define silk_ADD64(a,b) silk_ADD64_((a), (b), __FILE__, __LINE__) +static OPUS_INLINE opus_int64 silk_ADD64_(opus_int64 a, opus_int64 b, char *file, int line){ + opus_int64 ret; + + ret = a + b; + if ( ret != silk_ADD_SAT64( a, b ) ) + { + fprintf (stderr, "silk_ADD64(%lld, %lld) in %s: line %d\n", (long long)a, (long long)b, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return ret; +} + +#undef silk_SUB16 +#define silk_SUB16(a,b) silk_SUB16_((a), (b), __FILE__, __LINE__) +static OPUS_INLINE opus_int16 silk_SUB16_(opus_int16 a, opus_int16 b, char *file, int line){ + opus_int16 ret; + + ret = a - b; + if ( ret != silk_SUB_SAT16( a, b ) ) + { + fprintf (stderr, "silk_SUB16(%d, %d) in %s: line %d\n", a, b, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return ret; +} + +#undef silk_SUB32 +#define silk_SUB32(a,b) silk_SUB32_((a), (b), __FILE__, __LINE__) +static OPUS_INLINE opus_int32 silk_SUB32_(opus_int32 a, opus_int32 b, char *file, int line){ + opus_int64 ret; + + ret = a - (opus_int64)b; + if ( ret != silk_SUB_SAT32( a, b ) ) + { + fprintf (stderr, "silk_SUB32(%d, %d) in %s: line %d\n", a, b, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return ret; +} + +#undef silk_SUB64 +#define silk_SUB64(a,b) silk_SUB64_((a), (b), __FILE__, __LINE__) +static OPUS_INLINE opus_int64 silk_SUB64_(opus_int64 a, opus_int64 b, char *file, int line){ + opus_int64 ret; + + ret = a - b; + if ( ret != silk_SUB_SAT64( a, b ) ) + { + fprintf (stderr, "silk_SUB64(%lld, %lld) in %s: line %d\n", (long long)a, (long long)b, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return ret; +} + +#undef silk_ADD_SAT16 +#define silk_ADD_SAT16(a,b) silk_ADD_SAT16_((a), (b), __FILE__, __LINE__) +static OPUS_INLINE opus_int16 silk_ADD_SAT16_( opus_int16 a16, opus_int16 b16, char *file, int line) { + opus_int16 res; + res = (opus_int16)silk_SAT16( silk_ADD32( (opus_int32)(a16), (b16) ) ); + if ( res != silk_SAT16( (opus_int32)a16 + (opus_int32)b16 ) ) + { + fprintf (stderr, "silk_ADD_SAT16(%d, %d) in %s: line %d\n", a16, b16, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return res; +} + +#undef silk_ADD_SAT32 +#define silk_ADD_SAT32(a,b) silk_ADD_SAT32_((a), (b), __FILE__, __LINE__) +static OPUS_INLINE opus_int32 silk_ADD_SAT32_(opus_int32 a32, opus_int32 b32, char *file, int line){ + opus_int32 res; + res = ((((opus_uint32)(a32) + (opus_uint32)(b32)) & 0x80000000) == 0 ? \ + ((((a32) & (b32)) & 0x80000000) != 0 ? silk_int32_MIN : (a32)+(b32)) : \ + ((((a32) | (b32)) & 0x80000000) == 0 ? silk_int32_MAX : (a32)+(b32)) ); + if ( res != silk_SAT32( (opus_int64)a32 + (opus_int64)b32 ) ) + { + fprintf (stderr, "silk_ADD_SAT32(%d, %d) in %s: line %d\n", a32, b32, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return res; +} + +#undef silk_ADD_SAT64 +#define silk_ADD_SAT64(a,b) silk_ADD_SAT64_((a), (b), __FILE__, __LINE__) +static OPUS_INLINE opus_int64 silk_ADD_SAT64_( opus_int64 a64, opus_int64 b64, char *file, int line) { + opus_int64 res; + int fail = 0; + res = ((((a64) + (b64)) & 0x8000000000000000LL) == 0 ? \ + ((((a64) & (b64)) & 0x8000000000000000LL) != 0 ? silk_int64_MIN : (a64)+(b64)) : \ + ((((a64) | (b64)) & 0x8000000000000000LL) == 0 ? silk_int64_MAX : (a64)+(b64)) ); + if( res != a64 + b64 ) { + /* Check that we saturated to the correct extreme value */ + if ( !(( res == silk_int64_MAX && ( ( a64 >> 1 ) + ( b64 >> 1 ) > ( silk_int64_MAX >> 3 ) ) ) || + ( res == silk_int64_MIN && ( ( a64 >> 1 ) + ( b64 >> 1 ) < ( silk_int64_MIN >> 3 ) ) ) ) ) + { + fail = 1; + } + } else { + /* Saturation not necessary */ + fail = res != a64 + b64; + } + if ( fail ) + { + fprintf (stderr, "silk_ADD_SAT64(%lld, %lld) in %s: line %d\n", (long long)a64, (long long)b64, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return res; +} + +#undef silk_SUB_SAT16 +#define silk_SUB_SAT16(a,b) silk_SUB_SAT16_((a), (b), __FILE__, __LINE__) +static OPUS_INLINE opus_int16 silk_SUB_SAT16_( opus_int16 a16, opus_int16 b16, char *file, int line ) { + opus_int16 res; + res = (opus_int16)silk_SAT16( silk_SUB32( (opus_int32)(a16), (b16) ) ); + if ( res != silk_SAT16( (opus_int32)a16 - (opus_int32)b16 ) ) + { + fprintf (stderr, "silk_SUB_SAT16(%d, %d) in %s: line %d\n", a16, b16, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return res; +} + +#undef silk_SUB_SAT32 +#define silk_SUB_SAT32(a,b) silk_SUB_SAT32_((a), (b), __FILE__, __LINE__) +static OPUS_INLINE opus_int32 silk_SUB_SAT32_( opus_int32 a32, opus_int32 b32, char *file, int line ) { + opus_int32 res; + res = ((((opus_uint32)(a32)-(opus_uint32)(b32)) & 0x80000000) == 0 ? \ + (( (a32) & ((b32)^0x80000000) & 0x80000000) ? silk_int32_MIN : (a32)-(b32)) : \ + ((((a32)^0x80000000) & (b32) & 0x80000000) ? silk_int32_MAX : (a32)-(b32)) ); + if ( res != silk_SAT32( (opus_int64)a32 - (opus_int64)b32 ) ) + { + fprintf (stderr, "silk_SUB_SAT32(%d, %d) in %s: line %d\n", a32, b32, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return res; +} + +#undef silk_SUB_SAT64 +#define silk_SUB_SAT64(a,b) silk_SUB_SAT64_((a), (b), __FILE__, __LINE__) +static OPUS_INLINE opus_int64 silk_SUB_SAT64_( opus_int64 a64, opus_int64 b64, char *file, int line ) { + opus_int64 res; + int fail = 0; + res = ((((a64)-(b64)) & 0x8000000000000000LL) == 0 ? \ + (( (a64) & ((b64)^0x8000000000000000LL) & 0x8000000000000000LL) ? silk_int64_MIN : (a64)-(b64)) : \ + ((((a64)^0x8000000000000000LL) & (b64) & 0x8000000000000000LL) ? silk_int64_MAX : (a64)-(b64)) ); + if( res != a64 - b64 ) { + /* Check that we saturated to the correct extreme value */ + if( !(( res == silk_int64_MAX && ( ( a64 >> 1 ) + ( b64 >> 1 ) > ( silk_int64_MAX >> 3 ) ) ) || + ( res == silk_int64_MIN && ( ( a64 >> 1 ) + ( b64 >> 1 ) < ( silk_int64_MIN >> 3 ) ) ) )) + { + fail = 1; + } + } else { + /* Saturation not necessary */ + fail = res != a64 - b64; + } + if ( fail ) + { + fprintf (stderr, "silk_SUB_SAT64(%lld, %lld) in %s: line %d\n", (long long)a64, (long long)b64, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return res; +} + +#undef silk_MUL +#define silk_MUL(a,b) silk_MUL_((a), (b), __FILE__, __LINE__) +static OPUS_INLINE opus_int32 silk_MUL_(opus_int32 a32, opus_int32 b32, char *file, int line){ + opus_int32 ret; + opus_int64 ret64; + ret = (opus_int32)((opus_uint32)a32 * (opus_uint32)b32); + ret64 = (opus_int64)a32 * (opus_int64)b32; + if ( (opus_int64)ret != ret64 ) + { + fprintf (stderr, "silk_MUL(%d, %d) in %s: line %d\n", a32, b32, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return ret; +} + +#undef silk_MUL_uint +#define silk_MUL_uint(a,b) silk_MUL_uint_((a), (b), __FILE__, __LINE__) +static OPUS_INLINE opus_uint32 silk_MUL_uint_(opus_uint32 a32, opus_uint32 b32, char *file, int line){ + opus_uint32 ret; + ret = a32 * b32; + if ( (opus_uint64)ret != (opus_uint64)a32 * (opus_uint64)b32 ) + { + fprintf (stderr, "silk_MUL_uint(%u, %u) in %s: line %d\n", a32, b32, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return ret; +} + +#undef silk_MLA +#define silk_MLA(a,b,c) silk_MLA_((a), (b), (c), __FILE__, __LINE__) +static OPUS_INLINE opus_int32 silk_MLA_(opus_int32 a32, opus_int32 b32, opus_int32 c32, char *file, int line){ + opus_int32 ret; + ret = a32 + b32 * c32; + if ( (opus_int64)ret != (opus_int64)a32 + (opus_int64)b32 * (opus_int64)c32 ) + { + fprintf (stderr, "silk_MLA(%d, %d, %d) in %s: line %d\n", a32, b32, c32, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return ret; +} + +#undef silk_MLA_uint +#define silk_MLA_uint(a,b,c) silk_MLA_uint_((a), (b), (c), __FILE__, __LINE__) +static OPUS_INLINE opus_int32 silk_MLA_uint_(opus_uint32 a32, opus_uint32 b32, opus_uint32 c32, char *file, int line){ + opus_uint32 ret; + ret = a32 + b32 * c32; + if ( (opus_int64)ret != (opus_int64)a32 + (opus_int64)b32 * (opus_int64)c32 ) + { + fprintf (stderr, "silk_MLA_uint(%d, %d, %d) in %s: line %d\n", a32, b32, c32, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return ret; +} + +#undef silk_SMULWB +#define silk_SMULWB(a,b) silk_SMULWB_((a), (b), __FILE__, __LINE__) +static OPUS_INLINE opus_int32 silk_SMULWB_(opus_int32 a32, opus_int32 b32, char *file, int line){ + opus_int32 ret; + ret = (a32 >> 16) * (opus_int32)((opus_int16)b32) + (((a32 & 0x0000FFFF) * (opus_int32)((opus_int16)b32)) >> 16); + if ( (opus_int64)ret != ((opus_int64)a32 * (opus_int16)b32) >> 16 ) + { + fprintf (stderr, "silk_SMULWB(%d, %d) in %s: line %d\n", a32, b32, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return ret; +} + +#undef silk_SMLAWB +#define silk_SMLAWB(a,b,c) silk_SMLAWB_((a), (b), (c), __FILE__, __LINE__) +static OPUS_INLINE opus_int32 silk_SMLAWB_(opus_int32 a32, opus_int32 b32, opus_int32 c32, char *file, int line){ + opus_int32 ret; + ret = silk_ADD32_ovflw( a32, silk_SMULWB( b32, c32 ) ); + if ( ret != silk_ADD_SAT32( a32, silk_SMULWB( b32, c32 ) ) ) + { + fprintf (stderr, "silk_SMLAWB(%d, %d, %d) in %s: line %d\n", a32, b32, c32, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return ret; +} + +#undef silk_SMULWT +#define silk_SMULWT(a,b) silk_SMULWT_((a), (b), __FILE__, __LINE__) +static OPUS_INLINE opus_int32 silk_SMULWT_(opus_int32 a32, opus_int32 b32, char *file, int line){ + opus_int32 ret; + ret = (a32 >> 16) * (b32 >> 16) + (((a32 & 0x0000FFFF) * (b32 >> 16)) >> 16); + if ( (opus_int64)ret != ((opus_int64)a32 * (b32 >> 16)) >> 16 ) + { + fprintf (stderr, "silk_SMULWT(%d, %d) in %s: line %d\n", a32, b32, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return ret; +} + +#undef silk_SMLAWT +#define silk_SMLAWT(a,b,c) silk_SMLAWT_((a), (b), (c), __FILE__, __LINE__) +static OPUS_INLINE opus_int32 silk_SMLAWT_(opus_int32 a32, opus_int32 b32, opus_int32 c32, char *file, int line){ + opus_int32 ret; + ret = a32 + ((b32 >> 16) * (c32 >> 16)) + (((b32 & 0x0000FFFF) * ((c32 >> 16)) >> 16)); + if ( (opus_int64)ret != (opus_int64)a32 + (((opus_int64)b32 * (c32 >> 16)) >> 16) ) + { + fprintf (stderr, "silk_SMLAWT(%d, %d, %d) in %s: line %d\n", a32, b32, c32, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return ret; +} + +#undef silk_SMULL +#define silk_SMULL(a,b) silk_SMULL_((a), (b), __FILE__, __LINE__) +static OPUS_INLINE opus_int64 silk_SMULL_(opus_int64 a64, opus_int64 b64, char *file, int line){ + opus_int64 ret64; + int fail = 0; + ret64 = a64 * b64; + if( b64 != 0 ) { + fail = a64 != (ret64 / b64); + } else if( a64 != 0 ) { + fail = b64 != (ret64 / a64); + } + if ( fail ) + { + fprintf (stderr, "silk_SMULL(%lld, %lld) in %s: line %d\n", (long long)a64, (long long)b64, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return ret64; +} + +/* no checking needed for silk_SMULBB */ +#undef silk_SMLABB +#define silk_SMLABB(a,b,c) silk_SMLABB_((a), (b), (c), __FILE__, __LINE__) +static OPUS_INLINE opus_int32 silk_SMLABB_(opus_int32 a32, opus_int32 b32, opus_int32 c32, char *file, int line){ + opus_int32 ret; + ret = a32 + (opus_int32)((opus_int16)b32) * (opus_int32)((opus_int16)c32); + if ( (opus_int64)ret != (opus_int64)a32 + (opus_int64)b32 * (opus_int16)c32 ) + { + fprintf (stderr, "silk_SMLABB(%d, %d, %d) in %s: line %d\n", a32, b32, c32, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return ret; +} + +/* no checking needed for silk_SMULBT */ +#undef silk_SMLABT +#define silk_SMLABT(a,b,c) silk_SMLABT_((a), (b), (c), __FILE__, __LINE__) +static OPUS_INLINE opus_int32 silk_SMLABT_(opus_int32 a32, opus_int32 b32, opus_int32 c32, char *file, int line){ + opus_int32 ret; + ret = a32 + ((opus_int32)((opus_int16)b32)) * (c32 >> 16); + if ( (opus_int64)ret != (opus_int64)a32 + (opus_int64)b32 * (c32 >> 16) ) + { + fprintf (stderr, "silk_SMLABT(%d, %d, %d) in %s: line %d\n", a32, b32, c32, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return ret; +} + +/* no checking needed for silk_SMULTT */ +#undef silk_SMLATT +#define silk_SMLATT(a,b,c) silk_SMLATT_((a), (b), (c), __FILE__, __LINE__) +static OPUS_INLINE opus_int32 silk_SMLATT_(opus_int32 a32, opus_int32 b32, opus_int32 c32, char *file, int line){ + opus_int32 ret; + ret = a32 + (b32 >> 16) * (c32 >> 16); + if ( (opus_int64)ret != (opus_int64)a32 + (b32 >> 16) * (c32 >> 16) ) + { + fprintf (stderr, "silk_SMLATT(%d, %d, %d) in %s: line %d\n", a32, b32, c32, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return ret; +} + +#undef silk_SMULWW +#define silk_SMULWW(a,b) silk_SMULWW_((a), (b), __FILE__, __LINE__) +static OPUS_INLINE opus_int32 silk_SMULWW_(opus_int32 a32, opus_int32 b32, char *file, int line){ + opus_int32 ret, tmp1, tmp2; + opus_int64 ret64; + int fail = 0; + + ret = silk_SMULWB( a32, b32 ); + tmp1 = silk_RSHIFT_ROUND( b32, 16 ); + tmp2 = silk_MUL( a32, tmp1 ); + + fail |= (opus_int64)tmp2 != (opus_int64) a32 * (opus_int64) tmp1; + + tmp1 = ret; + ret = silk_ADD32( tmp1, tmp2 ); + fail |= silk_ADD32( tmp1, tmp2 ) != silk_ADD_SAT32( tmp1, tmp2 ); + + ret64 = silk_RSHIFT64( silk_SMULL( a32, b32 ), 16 ); + fail |= (opus_int64)ret != ret64; + + if ( fail ) + { + fprintf (stderr, "silk_SMULWW(%d, %d) in %s: line %d\n", a32, b32, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + + return ret; +} + +#undef silk_SMLAWW +#define silk_SMLAWW(a,b,c) silk_SMLAWW_((a), (b), (c), __FILE__, __LINE__) +static OPUS_INLINE opus_int32 silk_SMLAWW_(opus_int32 a32, opus_int32 b32, opus_int32 c32, char *file, int line){ + opus_int32 ret, tmp; + + tmp = silk_SMULWW( b32, c32 ); + ret = silk_ADD32( a32, tmp ); + if ( ret != silk_ADD_SAT32( a32, tmp ) ) + { + fprintf (stderr, "silk_SMLAWW(%d, %d, %d) in %s: line %d\n", a32, b32, c32, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return ret; +} + +/* no checking needed for silk_SMULL + no checking needed for silk_SMLAL + no checking needed for silk_SMLALBB + no checking needed for SigProcFIX_CLZ16 + no checking needed for SigProcFIX_CLZ32*/ + +#undef silk_DIV32 +#define silk_DIV32(a,b) silk_DIV32_((a), (b), __FILE__, __LINE__) +static OPUS_INLINE opus_int32 silk_DIV32_(opus_int32 a32, opus_int32 b32, char *file, int line){ + if ( b32 == 0 ) + { + fprintf (stderr, "silk_DIV32(%d, %d) in %s: line %d\n", a32, b32, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return a32 / b32; +} + +#undef silk_DIV32_16 +#define silk_DIV32_16(a,b) silk_DIV32_16_((a), (b), __FILE__, __LINE__) +static OPUS_INLINE opus_int32 silk_DIV32_16_(opus_int32 a32, opus_int32 b32, char *file, int line){ + int fail = 0; + fail |= b32 == 0; + fail |= b32 > silk_int16_MAX; + fail |= b32 < silk_int16_MIN; + if ( fail ) + { + fprintf (stderr, "silk_DIV32_16(%d, %d) in %s: line %d\n", a32, b32, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return a32 / b32; +} + +/* no checking needed for silk_SAT8 + no checking needed for silk_SAT16 + no checking needed for silk_SAT32 + no checking needed for silk_POS_SAT32 + no checking needed for silk_ADD_POS_SAT8 + no checking needed for silk_ADD_POS_SAT16 + no checking needed for silk_ADD_POS_SAT32 */ + +#undef silk_LSHIFT8 +#define silk_LSHIFT8(a,b) silk_LSHIFT8_((a), (b), __FILE__, __LINE__) +static OPUS_INLINE opus_int8 silk_LSHIFT8_(opus_int8 a, opus_int32 shift, char *file, int line){ + opus_int8 ret; + int fail = 0; + ret = (opus_int8)((opus_uint8)a << shift); + fail |= shift < 0; + fail |= shift >= 8; + fail |= (opus_int64)ret != (opus_int64)(((opus_uint64)a) << shift); + if ( fail ) + { + fprintf (stderr, "silk_LSHIFT8(%d, %d) in %s: line %d\n", a, shift, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return ret; +} + +#undef silk_LSHIFT16 +#define silk_LSHIFT16(a,b) silk_LSHIFT16_((a), (b), __FILE__, __LINE__) +static OPUS_INLINE opus_int16 silk_LSHIFT16_(opus_int16 a, opus_int32 shift, char *file, int line){ + opus_int16 ret; + int fail = 0; + ret = (opus_int16)((opus_uint16)a << shift); + fail |= shift < 0; + fail |= shift >= 16; + fail |= (opus_int64)ret != (opus_int64)(((opus_uint64)a) << shift); + if ( fail ) + { + fprintf (stderr, "silk_LSHIFT16(%d, %d) in %s: line %d\n", a, shift, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return ret; +} + +#undef silk_LSHIFT32 +#define silk_LSHIFT32(a,b) silk_LSHIFT32_((a), (b), __FILE__, __LINE__) +static OPUS_INLINE opus_int32 silk_LSHIFT32_(opus_int32 a, opus_int32 shift, char *file, int line){ + opus_int32 ret; + int fail = 0; + ret = (opus_int32)((opus_uint32)a << shift); + fail |= shift < 0; + fail |= shift >= 32; + fail |= (opus_int64)ret != (opus_int64)(((opus_uint64)a) << shift); + if ( fail ) + { + fprintf (stderr, "silk_LSHIFT32(%d, %d) in %s: line %d\n", a, shift, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return ret; +} + +#undef silk_LSHIFT64 +#define silk_LSHIFT64(a,b) silk_LSHIFT64_((a), (b), __FILE__, __LINE__) +static OPUS_INLINE opus_int64 silk_LSHIFT64_(opus_int64 a, opus_int shift, char *file, int line){ + opus_int64 ret; + int fail = 0; + ret = (opus_int64)((opus_uint64)a << shift); + fail |= shift < 0; + fail |= shift >= 64; + fail |= (ret>>shift) != ((opus_int64)a); + if ( fail ) + { + fprintf (stderr, "silk_LSHIFT64(%lld, %d) in %s: line %d\n", (long long)a, shift, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return ret; +} + +#undef silk_LSHIFT_ovflw +#define silk_LSHIFT_ovflw(a,b) silk_LSHIFT_ovflw_((a), (b), __FILE__, __LINE__) +static OPUS_INLINE opus_int32 silk_LSHIFT_ovflw_(opus_int32 a, opus_int32 shift, char *file, int line){ + if ( (shift < 0) || (shift >= 32) ) /* no check for overflow */ + { + fprintf (stderr, "silk_LSHIFT_ovflw(%d, %d) in %s: line %d\n", a, shift, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return a << shift; +} + +#undef silk_LSHIFT_uint +#define silk_LSHIFT_uint(a,b) silk_LSHIFT_uint_((a), (b), __FILE__, __LINE__) +static OPUS_INLINE opus_uint32 silk_LSHIFT_uint_(opus_uint32 a, opus_int32 shift, char *file, int line){ + opus_uint32 ret; + ret = a << shift; + if ( (shift < 0) || ((opus_int64)ret != ((opus_int64)a) << shift)) + { + fprintf (stderr, "silk_LSHIFT_uint(%u, %d) in %s: line %d\n", a, shift, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return ret; +} + +#undef silk_RSHIFT8 +#define silk_RSHITF8(a,b) silk_RSHIFT8_((a), (b), __FILE__, __LINE__) +static OPUS_INLINE opus_int8 silk_RSHIFT8_(opus_int8 a, opus_int32 shift, char *file, int line){ + if ( (shift < 0) || (shift>=8) ) + { + fprintf (stderr, "silk_RSHITF8(%d, %d) in %s: line %d\n", a, shift, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return a >> shift; +} + +#undef silk_RSHIFT16 +#define silk_RSHITF16(a,b) silk_RSHIFT16_((a), (b), __FILE__, __LINE__) +static OPUS_INLINE opus_int16 silk_RSHIFT16_(opus_int16 a, opus_int32 shift, char *file, int line){ + if ( (shift < 0) || (shift>=16) ) + { + fprintf (stderr, "silk_RSHITF16(%d, %d) in %s: line %d\n", a, shift, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return a >> shift; +} + +#undef silk_RSHIFT32 +#define silk_RSHIFT32(a,b) silk_RSHIFT32_((a), (b), __FILE__, __LINE__) +static OPUS_INLINE opus_int32 silk_RSHIFT32_(opus_int32 a, opus_int32 shift, char *file, int line){ + if ( (shift < 0) || (shift>=32) ) + { + fprintf (stderr, "silk_RSHITF32(%d, %d) in %s: line %d\n", a, shift, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return a >> shift; +} + +#undef silk_RSHIFT64 +#define silk_RSHIFT64(a,b) silk_RSHIFT64_((a), (b), __FILE__, __LINE__) +static OPUS_INLINE opus_int64 silk_RSHIFT64_(opus_int64 a, opus_int64 shift, char *file, int line){ + if ( (shift < 0) || (shift>=64) ) + { + fprintf (stderr, "silk_RSHITF64(%lld, %lld) in %s: line %d\n", (long long)a, (long long)shift, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return a >> shift; +} + +#undef silk_RSHIFT_uint +#define silk_RSHIFT_uint(a,b) silk_RSHIFT_uint_((a), (b), __FILE__, __LINE__) +static OPUS_INLINE opus_uint32 silk_RSHIFT_uint_(opus_uint32 a, opus_int32 shift, char *file, int line){ + if ( (shift < 0) || (shift>32) ) + { + fprintf (stderr, "silk_RSHIFT_uint(%u, %d) in %s: line %d\n", a, shift, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return a >> shift; +} + +#undef silk_ADD_LSHIFT +#define silk_ADD_LSHIFT(a,b,c) silk_ADD_LSHIFT_((a), (b), (c), __FILE__, __LINE__) +static OPUS_INLINE int silk_ADD_LSHIFT_(int a, int b, int shift, char *file, int line){ + opus_int16 ret; + ret = a + (opus_int16)((opus_uint16)b << shift); + if ( (shift < 0) || (shift>15) || ((opus_int64)ret != (opus_int64)a + (opus_int64)(((opus_uint64)b) << shift)) ) + { + fprintf (stderr, "silk_ADD_LSHIFT(%d, %d, %d) in %s: line %d\n", a, b, shift, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return ret; /* shift >= 0 */ +} + +#undef silk_ADD_LSHIFT32 +#define silk_ADD_LSHIFT32(a,b,c) silk_ADD_LSHIFT32_((a), (b), (c), __FILE__, __LINE__) +static OPUS_INLINE opus_int32 silk_ADD_LSHIFT32_(opus_int32 a, opus_int32 b, opus_int32 shift, char *file, int line){ + opus_int32 ret; + ret = silk_ADD32_ovflw(a, (opus_int32)((opus_uint32)b << shift)); + if ( (shift < 0) || (shift>31) || ((opus_int64)ret != (opus_int64)a + (opus_int64)(((opus_uint64)b) << shift)) ) + { + fprintf (stderr, "silk_ADD_LSHIFT32(%d, %d, %d) in %s: line %d\n", a, b, shift, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return ret; /* shift >= 0 */ +} + +#undef silk_ADD_LSHIFT_uint +#define silk_ADD_LSHIFT_uint(a,b,c) silk_ADD_LSHIFT_uint_((a), (b), (c), __FILE__, __LINE__) +static OPUS_INLINE opus_uint32 silk_ADD_LSHIFT_uint_(opus_uint32 a, opus_uint32 b, opus_int32 shift, char *file, int line){ + opus_uint32 ret; + ret = a + (b << shift); + if ( (shift < 0) || (shift>32) || ((opus_int64)ret != (opus_int64)a + (((opus_int64)b) << shift)) ) + { + fprintf (stderr, "silk_ADD_LSHIFT_uint(%u, %u, %d) in %s: line %d\n", a, b, shift, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return ret; /* shift >= 0 */ +} + +#undef silk_ADD_RSHIFT +#define silk_ADD_RSHIFT(a,b,c) silk_ADD_RSHIFT_((a), (b), (c), __FILE__, __LINE__) +static OPUS_INLINE int silk_ADD_RSHIFT_(int a, int b, int shift, char *file, int line){ + opus_int16 ret; + ret = a + (b >> shift); + if ( (shift < 0) || (shift>15) || ((opus_int64)ret != (opus_int64)a + (((opus_int64)b) >> shift)) ) + { + fprintf (stderr, "silk_ADD_RSHIFT(%d, %d, %d) in %s: line %d\n", a, b, shift, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return ret; /* shift > 0 */ +} + +#undef silk_ADD_RSHIFT32 +#define silk_ADD_RSHIFT32(a,b,c) silk_ADD_RSHIFT32_((a), (b), (c), __FILE__, __LINE__) +static OPUS_INLINE opus_int32 silk_ADD_RSHIFT32_(opus_int32 a, opus_int32 b, opus_int32 shift, char *file, int line){ + opus_int32 ret; + ret = silk_ADD32_ovflw(a, (b >> shift)); + if ( (shift < 0) || (shift>31) || ((opus_int64)ret != (opus_int64)a + (((opus_int64)b) >> shift)) ) + { + fprintf (stderr, "silk_ADD_RSHIFT32(%d, %d, %d) in %s: line %d\n", a, b, shift, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return ret; /* shift > 0 */ +} + +#undef silk_ADD_RSHIFT_uint +#define silk_ADD_RSHIFT_uint(a,b,c) silk_ADD_RSHIFT_uint_((a), (b), (c), __FILE__, __LINE__) +static OPUS_INLINE opus_uint32 silk_ADD_RSHIFT_uint_(opus_uint32 a, opus_uint32 b, opus_int32 shift, char *file, int line){ + opus_uint32 ret; + ret = a + (b >> shift); + if ( (shift < 0) || (shift>32) || ((opus_int64)ret != (opus_int64)a + (((opus_int64)b) >> shift)) ) + { + fprintf (stderr, "silk_ADD_RSHIFT_uint(%u, %u, %d) in %s: line %d\n", a, b, shift, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return ret; /* shift > 0 */ +} + +#undef silk_SUB_LSHIFT32 +#define silk_SUB_LSHIFT32(a,b,c) silk_SUB_LSHIFT32_((a), (b), (c), __FILE__, __LINE__) +static OPUS_INLINE opus_int32 silk_SUB_LSHIFT32_(opus_int32 a, opus_int32 b, opus_int32 shift, char *file, int line){ + opus_int32 ret; + ret = silk_SUB32_ovflw(a, (opus_int32)((opus_uint32)b << shift)); + if ( (shift < 0) || (shift>31) || ((opus_int64)ret != (opus_int64)a - (opus_int64)(((opus_uint64)b) << shift)) ) + { + fprintf (stderr, "silk_SUB_LSHIFT32(%d, %d, %d) in %s: line %d\n", a, b, shift, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return ret; /* shift >= 0 */ +} + +#undef silk_SUB_RSHIFT32 +#define silk_SUB_RSHIFT32(a,b,c) silk_SUB_RSHIFT32_((a), (b), (c), __FILE__, __LINE__) +static OPUS_INLINE opus_int32 silk_SUB_RSHIFT32_(opus_int32 a, opus_int32 b, opus_int32 shift, char *file, int line){ + opus_int32 ret; + ret = silk_SUB32_ovflw(a, (b >> shift)); + if ( (shift < 0) || (shift>31) || ((opus_int64)ret != (opus_int64)a - (((opus_int64)b) >> shift)) ) + { + fprintf (stderr, "silk_SUB_RSHIFT32(%d, %d, %d) in %s: line %d\n", a, b, shift, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return ret; /* shift > 0 */ +} + +#undef silk_RSHIFT_ROUND +#define silk_RSHIFT_ROUND(a,b) silk_RSHIFT_ROUND_((a), (b), __FILE__, __LINE__) +static OPUS_INLINE opus_int32 silk_RSHIFT_ROUND_(opus_int32 a, opus_int32 shift, char *file, int line){ + opus_int32 ret; + ret = shift == 1 ? (a >> 1) + (a & 1) : ((a >> (shift - 1)) + 1) >> 1; + /* the macro definition can't handle a shift of zero */ + if ( (shift <= 0) || (shift>31) || ((opus_int64)ret != ((opus_int64)a + ((opus_int64)1 << (shift - 1))) >> shift) ) + { + fprintf (stderr, "silk_RSHIFT_ROUND(%d, %d) in %s: line %d\n", a, shift, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return ret; +} + +#undef silk_RSHIFT_ROUND64 +#define silk_RSHIFT_ROUND64(a,b) silk_RSHIFT_ROUND64_((a), (b), __FILE__, __LINE__) +static OPUS_INLINE opus_int64 silk_RSHIFT_ROUND64_(opus_int64 a, opus_int32 shift, char *file, int line){ + opus_int64 ret; + /* the macro definition can't handle a shift of zero */ + if ( (shift <= 0) || (shift>=64) ) + { + fprintf (stderr, "silk_RSHIFT_ROUND64(%lld, %d) in %s: line %d\n", (long long)a, shift, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + ret = shift == 1 ? (a >> 1) + (a & 1) : ((a >> (shift - 1)) + 1) >> 1; + return ret; +} + +/* silk_abs is used on floats also, so doesn't work... */ +/*#undef silk_abs +static OPUS_INLINE opus_int32 silk_abs(opus_int32 a){ + silk_assert(a != 0x80000000); + return (((a) > 0) ? (a) : -(a)); // Be careful, silk_abs returns wrong when input equals to silk_intXX_MIN +}*/ + +#undef silk_abs_int64 +#define silk_abs_int64(a) silk_abs_int64_((a), __FILE__, __LINE__) +static OPUS_INLINE opus_int64 silk_abs_int64_(opus_int64 a, char *file, int line){ + if ( a == silk_int64_MIN ) + { + fprintf (stderr, "silk_abs_int64(%lld) in %s: line %d\n", (long long)a, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return (((a) > 0) ? (a) : -(a)); /* Be careful, silk_abs returns wrong when input equals to silk_intXX_MIN */ +} + +#undef silk_abs_int32 +#define silk_abs_int32(a) silk_abs_int32_((a), __FILE__, __LINE__) +static OPUS_INLINE opus_int32 silk_abs_int32_(opus_int32 a, char *file, int line){ + if ( a == silk_int32_MIN ) + { + fprintf (stderr, "silk_abs_int32(%d) in %s: line %d\n", a, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return silk_abs(a); +} + +#undef silk_CHECK_FIT8 +#define silk_CHECK_FIT8(a) silk_CHECK_FIT8_((a), __FILE__, __LINE__) +static OPUS_INLINE opus_int8 silk_CHECK_FIT8_( opus_int64 a, char *file, int line ){ + opus_int8 ret; + ret = (opus_int8)a; + if ( (opus_int64)ret != a ) + { + fprintf (stderr, "silk_CHECK_FIT8(%lld) in %s: line %d\n", (long long)a, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return( ret ); +} + +#undef silk_CHECK_FIT16 +#define silk_CHECK_FIT16(a) silk_CHECK_FIT16_((a), __FILE__, __LINE__) +static OPUS_INLINE opus_int16 silk_CHECK_FIT16_( opus_int64 a, char *file, int line ){ + opus_int16 ret; + ret = (opus_int16)a; + if ( (opus_int64)ret != a ) + { + fprintf (stderr, "silk_CHECK_FIT16(%lld) in %s: line %d\n", (long long)a, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return( ret ); +} + +#undef silk_CHECK_FIT32 +#define silk_CHECK_FIT32(a) silk_CHECK_FIT32_((a), __FILE__, __LINE__) +static OPUS_INLINE opus_int32 silk_CHECK_FIT32_( opus_int64 a, char *file, int line ){ + opus_int32 ret; + ret = (opus_int32)a; + if ( (opus_int64)ret != a ) + { + fprintf (stderr, "silk_CHECK_FIT32(%lld) in %s: line %d\n", (long long)a, file, line); +#ifdef FIXED_DEBUG_ASSERT + silk_assert( 0 ); +#endif + } + return( ret ); +} + +/* no checking for silk_NSHIFT_MUL_32_32 + no checking for silk_NSHIFT_MUL_16_16 + no checking needed for silk_min + no checking needed for silk_max + no checking needed for silk_sign +*/ + +#endif +#endif /* MACRO_DEBUG_H */ diff --git a/media/libopus/silk/NLSF2A.c b/media/libopus/silk/NLSF2A.c new file mode 100644 index 0000000000..d5b7730638 --- /dev/null +++ b/media/libopus/silk/NLSF2A.c @@ -0,0 +1,141 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +/* conversion between prediction filter coefficients and LSFs */ +/* order should be even */ +/* a piecewise linear approximation maps LSF <-> cos(LSF) */ +/* therefore the result is not accurate LSFs, but the two */ +/* functions are accurate inverses of each other */ + +#include "SigProc_FIX.h" +#include "tables.h" + +#define QA 16 + +/* helper function for NLSF2A(..) */ +static OPUS_INLINE void silk_NLSF2A_find_poly( + opus_int32 *out, /* O intermediate polynomial, QA [dd+1] */ + const opus_int32 *cLSF, /* I vector of interleaved 2*cos(LSFs), QA [d] */ + opus_int dd /* I polynomial order (= 1/2 * filter order) */ +) +{ + opus_int k, n; + opus_int32 ftmp; + + out[0] = silk_LSHIFT( 1, QA ); + out[1] = -cLSF[0]; + for( k = 1; k < dd; k++ ) { + ftmp = cLSF[2*k]; /* QA*/ + out[k+1] = silk_LSHIFT( out[k-1], 1 ) - (opus_int32)silk_RSHIFT_ROUND64( silk_SMULL( ftmp, out[k] ), QA ); + for( n = k; n > 1; n-- ) { + out[n] += out[n-2] - (opus_int32)silk_RSHIFT_ROUND64( silk_SMULL( ftmp, out[n-1] ), QA ); + } + out[1] -= ftmp; + } +} + +/* compute whitening filter coefficients from normalized line spectral frequencies */ +void silk_NLSF2A( + opus_int16 *a_Q12, /* O monic whitening filter coefficients in Q12, [ d ] */ + const opus_int16 *NLSF, /* I normalized line spectral frequencies in Q15, [ d ] */ + const opus_int d, /* I filter order (should be even) */ + int arch /* I Run-time architecture */ +) +{ + /* This ordering was found to maximize quality. It improves numerical accuracy of + silk_NLSF2A_find_poly() compared to "standard" ordering. */ + static const unsigned char ordering16[16] = { + 0, 15, 8, 7, 4, 11, 12, 3, 2, 13, 10, 5, 6, 9, 14, 1 + }; + static const unsigned char ordering10[10] = { + 0, 9, 6, 3, 4, 5, 8, 1, 2, 7 + }; + const unsigned char *ordering; + opus_int k, i, dd; + opus_int32 cos_LSF_QA[ SILK_MAX_ORDER_LPC ]; + opus_int32 P[ SILK_MAX_ORDER_LPC / 2 + 1 ], Q[ SILK_MAX_ORDER_LPC / 2 + 1 ]; + opus_int32 Ptmp, Qtmp, f_int, f_frac, cos_val, delta; + opus_int32 a32_QA1[ SILK_MAX_ORDER_LPC ]; + + silk_assert( LSF_COS_TAB_SZ_FIX == 128 ); + celt_assert( d==10 || d==16 ); + + /* convert LSFs to 2*cos(LSF), using piecewise linear curve from table */ + ordering = d == 16 ? ordering16 : ordering10; + for( k = 0; k < d; k++ ) { + silk_assert( NLSF[k] >= 0 ); + + /* f_int on a scale 0-127 (rounded down) */ + f_int = silk_RSHIFT( NLSF[k], 15 - 7 ); + + /* f_frac, range: 0..255 */ + f_frac = NLSF[k] - silk_LSHIFT( f_int, 15 - 7 ); + + silk_assert(f_int >= 0); + silk_assert(f_int < LSF_COS_TAB_SZ_FIX ); + + /* Read start and end value from table */ + cos_val = silk_LSFCosTab_FIX_Q12[ f_int ]; /* Q12 */ + delta = silk_LSFCosTab_FIX_Q12[ f_int + 1 ] - cos_val; /* Q12, with a range of 0..200 */ + + /* Linear interpolation */ + cos_LSF_QA[ordering[k]] = silk_RSHIFT_ROUND( silk_LSHIFT( cos_val, 8 ) + silk_MUL( delta, f_frac ), 20 - QA ); /* QA */ + } + + dd = silk_RSHIFT( d, 1 ); + + /* generate even and odd polynomials using convolution */ + silk_NLSF2A_find_poly( P, &cos_LSF_QA[ 0 ], dd ); + silk_NLSF2A_find_poly( Q, &cos_LSF_QA[ 1 ], dd ); + + /* convert even and odd polynomials to opus_int32 Q12 filter coefs */ + for( k = 0; k < dd; k++ ) { + Ptmp = P[ k+1 ] + P[ k ]; + Qtmp = Q[ k+1 ] - Q[ k ]; + + /* the Ptmp and Qtmp values at this stage need to fit in int32 */ + a32_QA1[ k ] = -Qtmp - Ptmp; /* QA+1 */ + a32_QA1[ d-k-1 ] = Qtmp - Ptmp; /* QA+1 */ + } + + /* Convert int32 coefficients to Q12 int16 coefs */ + silk_LPC_fit( a_Q12, a32_QA1, 12, QA + 1, d ); + + for( i = 0; silk_LPC_inverse_pred_gain( a_Q12, d, arch ) == 0 && i < MAX_LPC_STABILIZE_ITERATIONS; i++ ) { + /* Prediction coefficients are (too close to) unstable; apply bandwidth expansion */ + /* on the unscaled coefficients, convert to Q12 and measure again */ + silk_bwexpander_32( a32_QA1, d, 65536 - silk_LSHIFT( 2, i ) ); + for( k = 0; k < d; k++ ) { + a_Q12[ k ] = (opus_int16)silk_RSHIFT_ROUND( a32_QA1[ k ], QA + 1 - 12 ); /* QA+1 -> Q12 */ + } + } +} + diff --git a/media/libopus/silk/NLSF_VQ.c b/media/libopus/silk/NLSF_VQ.c new file mode 100644 index 0000000000..b83182a79c --- /dev/null +++ b/media/libopus/silk/NLSF_VQ.c @@ -0,0 +1,76 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" + +/* Compute quantization errors for an LPC_order element input vector for a VQ codebook */ +void silk_NLSF_VQ( + opus_int32 err_Q24[], /* O Quantization errors [K] */ + const opus_int16 in_Q15[], /* I Input vectors to be quantized [LPC_order] */ + const opus_uint8 pCB_Q8[], /* I Codebook vectors [K*LPC_order] */ + const opus_int16 pWght_Q9[], /* I Codebook weights [K*LPC_order] */ + const opus_int K, /* I Number of codebook vectors */ + const opus_int LPC_order /* I Number of LPCs */ +) +{ + opus_int i, m; + opus_int32 diff_Q15, diffw_Q24, sum_error_Q24, pred_Q24; + const opus_int16 *w_Q9_ptr; + const opus_uint8 *cb_Q8_ptr; + + celt_assert( ( LPC_order & 1 ) == 0 ); + + /* Loop over codebook */ + cb_Q8_ptr = pCB_Q8; + w_Q9_ptr = pWght_Q9; + for( i = 0; i < K; i++ ) { + sum_error_Q24 = 0; + pred_Q24 = 0; + for( m = LPC_order-2; m >= 0; m -= 2 ) { + /* Compute weighted absolute predictive quantization error for index m + 1 */ + diff_Q15 = silk_SUB_LSHIFT32( in_Q15[ m + 1 ], (opus_int32)cb_Q8_ptr[ m + 1 ], 7 ); /* range: [ -32767 : 32767 ]*/ + diffw_Q24 = silk_SMULBB( diff_Q15, w_Q9_ptr[ m + 1 ] ); + sum_error_Q24 = silk_ADD32( sum_error_Q24, silk_abs( silk_SUB_RSHIFT32( diffw_Q24, pred_Q24, 1 ) ) ); + pred_Q24 = diffw_Q24; + + /* Compute weighted absolute predictive quantization error for index m */ + diff_Q15 = silk_SUB_LSHIFT32( in_Q15[ m ], (opus_int32)cb_Q8_ptr[ m ], 7 ); /* range: [ -32767 : 32767 ]*/ + diffw_Q24 = silk_SMULBB( diff_Q15, w_Q9_ptr[ m ] ); + sum_error_Q24 = silk_ADD32( sum_error_Q24, silk_abs( silk_SUB_RSHIFT32( diffw_Q24, pred_Q24, 1 ) ) ); + pred_Q24 = diffw_Q24; + + silk_assert( sum_error_Q24 >= 0 ); + } + err_Q24[ i ] = sum_error_Q24; + cb_Q8_ptr += LPC_order; + w_Q9_ptr += LPC_order; + } +} diff --git a/media/libopus/silk/NLSF_VQ_weights_laroia.c b/media/libopus/silk/NLSF_VQ_weights_laroia.c new file mode 100644 index 0000000000..9873bcde10 --- /dev/null +++ b/media/libopus/silk/NLSF_VQ_weights_laroia.c @@ -0,0 +1,80 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "define.h" +#include "SigProc_FIX.h" + +/* +R. Laroia, N. Phamdo and N. Farvardin, "Robust and Efficient Quantization of Speech LSP +Parameters Using Structured Vector Quantization", Proc. IEEE Int. Conf. Acoust., Speech, +Signal Processing, pp. 641-644, 1991. +*/ + +/* Laroia low complexity NLSF weights */ +void silk_NLSF_VQ_weights_laroia( + opus_int16 *pNLSFW_Q_OUT, /* O Pointer to input vector weights [D] */ + const opus_int16 *pNLSF_Q15, /* I Pointer to input vector [D] */ + const opus_int D /* I Input vector dimension (even) */ +) +{ + opus_int k; + opus_int32 tmp1_int, tmp2_int; + + celt_assert( D > 0 ); + celt_assert( ( D & 1 ) == 0 ); + + /* First value */ + tmp1_int = silk_max_int( pNLSF_Q15[ 0 ], 1 ); + tmp1_int = silk_DIV32_16( (opus_int32)1 << ( 15 + NLSF_W_Q ), tmp1_int ); + tmp2_int = silk_max_int( pNLSF_Q15[ 1 ] - pNLSF_Q15[ 0 ], 1 ); + tmp2_int = silk_DIV32_16( (opus_int32)1 << ( 15 + NLSF_W_Q ), tmp2_int ); + pNLSFW_Q_OUT[ 0 ] = (opus_int16)silk_min_int( tmp1_int + tmp2_int, silk_int16_MAX ); + silk_assert( pNLSFW_Q_OUT[ 0 ] > 0 ); + + /* Main loop */ + for( k = 1; k < D - 1; k += 2 ) { + tmp1_int = silk_max_int( pNLSF_Q15[ k + 1 ] - pNLSF_Q15[ k ], 1 ); + tmp1_int = silk_DIV32_16( (opus_int32)1 << ( 15 + NLSF_W_Q ), tmp1_int ); + pNLSFW_Q_OUT[ k ] = (opus_int16)silk_min_int( tmp1_int + tmp2_int, silk_int16_MAX ); + silk_assert( pNLSFW_Q_OUT[ k ] > 0 ); + + tmp2_int = silk_max_int( pNLSF_Q15[ k + 2 ] - pNLSF_Q15[ k + 1 ], 1 ); + tmp2_int = silk_DIV32_16( (opus_int32)1 << ( 15 + NLSF_W_Q ), tmp2_int ); + pNLSFW_Q_OUT[ k + 1 ] = (opus_int16)silk_min_int( tmp1_int + tmp2_int, silk_int16_MAX ); + silk_assert( pNLSFW_Q_OUT[ k + 1 ] > 0 ); + } + + /* Last value */ + tmp1_int = silk_max_int( ( 1 << 15 ) - pNLSF_Q15[ D - 1 ], 1 ); + tmp1_int = silk_DIV32_16( (opus_int32)1 << ( 15 + NLSF_W_Q ), tmp1_int ); + pNLSFW_Q_OUT[ D - 1 ] = (opus_int16)silk_min_int( tmp1_int + tmp2_int, silk_int16_MAX ); + silk_assert( pNLSFW_Q_OUT[ D - 1 ] > 0 ); +} diff --git a/media/libopus/silk/NLSF_decode.c b/media/libopus/silk/NLSF_decode.c new file mode 100644 index 0000000000..eeb0ba8c92 --- /dev/null +++ b/media/libopus/silk/NLSF_decode.c @@ -0,0 +1,93 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" + +/* Predictive dequantizer for NLSF residuals */ +static OPUS_INLINE void silk_NLSF_residual_dequant( /* O Returns RD value in Q30 */ + opus_int16 x_Q10[], /* O Output [ order ] */ + const opus_int8 indices[], /* I Quantization indices [ order ] */ + const opus_uint8 pred_coef_Q8[], /* I Backward predictor coefs [ order ] */ + const opus_int quant_step_size_Q16, /* I Quantization step size */ + const opus_int16 order /* I Number of input values */ +) +{ + opus_int i, out_Q10, pred_Q10; + + out_Q10 = 0; + for( i = order-1; i >= 0; i-- ) { + pred_Q10 = silk_RSHIFT( silk_SMULBB( out_Q10, (opus_int16)pred_coef_Q8[ i ] ), 8 ); + out_Q10 = silk_LSHIFT( indices[ i ], 10 ); + if( out_Q10 > 0 ) { + out_Q10 = silk_SUB16( out_Q10, SILK_FIX_CONST( NLSF_QUANT_LEVEL_ADJ, 10 ) ); + } else if( out_Q10 < 0 ) { + out_Q10 = silk_ADD16( out_Q10, SILK_FIX_CONST( NLSF_QUANT_LEVEL_ADJ, 10 ) ); + } + out_Q10 = silk_SMLAWB( pred_Q10, (opus_int32)out_Q10, quant_step_size_Q16 ); + x_Q10[ i ] = out_Q10; + } +} + + +/***********************/ +/* NLSF vector decoder */ +/***********************/ +void silk_NLSF_decode( + opus_int16 *pNLSF_Q15, /* O Quantized NLSF vector [ LPC_ORDER ] */ + opus_int8 *NLSFIndices, /* I Codebook path vector [ LPC_ORDER + 1 ] */ + const silk_NLSF_CB_struct *psNLSF_CB /* I Codebook object */ +) +{ + opus_int i; + opus_uint8 pred_Q8[ MAX_LPC_ORDER ]; + opus_int16 ec_ix[ MAX_LPC_ORDER ]; + opus_int16 res_Q10[ MAX_LPC_ORDER ]; + opus_int32 NLSF_Q15_tmp; + const opus_uint8 *pCB_element; + const opus_int16 *pCB_Wght_Q9; + + /* Unpack entropy table indices and predictor for current CB1 index */ + silk_NLSF_unpack( ec_ix, pred_Q8, psNLSF_CB, NLSFIndices[ 0 ] ); + + /* Predictive residual dequantizer */ + silk_NLSF_residual_dequant( res_Q10, &NLSFIndices[ 1 ], pred_Q8, psNLSF_CB->quantStepSize_Q16, psNLSF_CB->order ); + + /* Apply inverse square-rooted weights to first stage and add to output */ + pCB_element = &psNLSF_CB->CB1_NLSF_Q8[ NLSFIndices[ 0 ] * psNLSF_CB->order ]; + pCB_Wght_Q9 = &psNLSF_CB->CB1_Wght_Q9[ NLSFIndices[ 0 ] * psNLSF_CB->order ]; + for( i = 0; i < psNLSF_CB->order; i++ ) { + NLSF_Q15_tmp = silk_ADD_LSHIFT32( silk_DIV32_16( silk_LSHIFT( (opus_int32)res_Q10[ i ], 14 ), pCB_Wght_Q9[ i ] ), (opus_int16)pCB_element[ i ], 7 ); + pNLSF_Q15[ i ] = (opus_int16)silk_LIMIT( NLSF_Q15_tmp, 0, 32767 ); + } + + /* NLSF stabilization */ + silk_NLSF_stabilize( pNLSF_Q15, psNLSF_CB->deltaMin_Q15, psNLSF_CB->order ); +} diff --git a/media/libopus/silk/NLSF_del_dec_quant.c b/media/libopus/silk/NLSF_del_dec_quant.c new file mode 100644 index 0000000000..44a16acd0b --- /dev/null +++ b/media/libopus/silk/NLSF_del_dec_quant.c @@ -0,0 +1,215 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" + +/* Delayed-decision quantizer for NLSF residuals */ +opus_int32 silk_NLSF_del_dec_quant( /* O Returns RD value in Q25 */ + opus_int8 indices[], /* O Quantization indices [ order ] */ + const opus_int16 x_Q10[], /* I Input [ order ] */ + const opus_int16 w_Q5[], /* I Weights [ order ] */ + const opus_uint8 pred_coef_Q8[], /* I Backward predictor coefs [ order ] */ + const opus_int16 ec_ix[], /* I Indices to entropy coding tables [ order ] */ + const opus_uint8 ec_rates_Q5[], /* I Rates [] */ + const opus_int quant_step_size_Q16, /* I Quantization step size */ + const opus_int16 inv_quant_step_size_Q6, /* I Inverse quantization step size */ + const opus_int32 mu_Q20, /* I R/D tradeoff */ + const opus_int16 order /* I Number of input values */ +) +{ + opus_int i, j, nStates, ind_tmp, ind_min_max, ind_max_min, in_Q10, res_Q10; + opus_int pred_Q10, diff_Q10, rate0_Q5, rate1_Q5; + opus_int16 out0_Q10, out1_Q10; + opus_int32 RD_tmp_Q25, min_Q25, min_max_Q25, max_min_Q25; + opus_int ind_sort[ NLSF_QUANT_DEL_DEC_STATES ]; + opus_int8 ind[ NLSF_QUANT_DEL_DEC_STATES ][ MAX_LPC_ORDER ]; + opus_int16 prev_out_Q10[ 2 * NLSF_QUANT_DEL_DEC_STATES ]; + opus_int32 RD_Q25[ 2 * NLSF_QUANT_DEL_DEC_STATES ]; + opus_int32 RD_min_Q25[ NLSF_QUANT_DEL_DEC_STATES ]; + opus_int32 RD_max_Q25[ NLSF_QUANT_DEL_DEC_STATES ]; + const opus_uint8 *rates_Q5; + + opus_int out0_Q10_table[2 * NLSF_QUANT_MAX_AMPLITUDE_EXT]; + opus_int out1_Q10_table[2 * NLSF_QUANT_MAX_AMPLITUDE_EXT]; + + for (i = -NLSF_QUANT_MAX_AMPLITUDE_EXT; i <= NLSF_QUANT_MAX_AMPLITUDE_EXT-1; i++) + { + out0_Q10 = silk_LSHIFT( i, 10 ); + out1_Q10 = silk_ADD16( out0_Q10, 1024 ); + if( i > 0 ) { + out0_Q10 = silk_SUB16( out0_Q10, SILK_FIX_CONST( NLSF_QUANT_LEVEL_ADJ, 10 ) ); + out1_Q10 = silk_SUB16( out1_Q10, SILK_FIX_CONST( NLSF_QUANT_LEVEL_ADJ, 10 ) ); + } else if( i == 0 ) { + out1_Q10 = silk_SUB16( out1_Q10, SILK_FIX_CONST( NLSF_QUANT_LEVEL_ADJ, 10 ) ); + } else if( i == -1 ) { + out0_Q10 = silk_ADD16( out0_Q10, SILK_FIX_CONST( NLSF_QUANT_LEVEL_ADJ, 10 ) ); + } else { + out0_Q10 = silk_ADD16( out0_Q10, SILK_FIX_CONST( NLSF_QUANT_LEVEL_ADJ, 10 ) ); + out1_Q10 = silk_ADD16( out1_Q10, SILK_FIX_CONST( NLSF_QUANT_LEVEL_ADJ, 10 ) ); + } + out0_Q10_table[ i + NLSF_QUANT_MAX_AMPLITUDE_EXT ] = silk_RSHIFT( silk_SMULBB( out0_Q10, quant_step_size_Q16 ), 16 ); + out1_Q10_table[ i + NLSF_QUANT_MAX_AMPLITUDE_EXT ] = silk_RSHIFT( silk_SMULBB( out1_Q10, quant_step_size_Q16 ), 16 ); + } + + silk_assert( (NLSF_QUANT_DEL_DEC_STATES & (NLSF_QUANT_DEL_DEC_STATES-1)) == 0 ); /* must be power of two */ + + nStates = 1; + RD_Q25[ 0 ] = 0; + prev_out_Q10[ 0 ] = 0; + for( i = order - 1; i >= 0; i-- ) { + rates_Q5 = &ec_rates_Q5[ ec_ix[ i ] ]; + in_Q10 = x_Q10[ i ]; + for( j = 0; j < nStates; j++ ) { + pred_Q10 = silk_RSHIFT( silk_SMULBB( (opus_int16)pred_coef_Q8[ i ], prev_out_Q10[ j ] ), 8 ); + res_Q10 = silk_SUB16( in_Q10, pred_Q10 ); + ind_tmp = silk_RSHIFT( silk_SMULBB( inv_quant_step_size_Q6, res_Q10 ), 16 ); + ind_tmp = silk_LIMIT( ind_tmp, -NLSF_QUANT_MAX_AMPLITUDE_EXT, NLSF_QUANT_MAX_AMPLITUDE_EXT-1 ); + ind[ j ][ i ] = (opus_int8)ind_tmp; + + /* compute outputs for ind_tmp and ind_tmp + 1 */ + out0_Q10 = out0_Q10_table[ ind_tmp + NLSF_QUANT_MAX_AMPLITUDE_EXT ]; + out1_Q10 = out1_Q10_table[ ind_tmp + NLSF_QUANT_MAX_AMPLITUDE_EXT ]; + + out0_Q10 = silk_ADD16( out0_Q10, pred_Q10 ); + out1_Q10 = silk_ADD16( out1_Q10, pred_Q10 ); + prev_out_Q10[ j ] = out0_Q10; + prev_out_Q10[ j + nStates ] = out1_Q10; + + /* compute RD for ind_tmp and ind_tmp + 1 */ + if( ind_tmp + 1 >= NLSF_QUANT_MAX_AMPLITUDE ) { + if( ind_tmp + 1 == NLSF_QUANT_MAX_AMPLITUDE ) { + rate0_Q5 = rates_Q5[ ind_tmp + NLSF_QUANT_MAX_AMPLITUDE ]; + rate1_Q5 = 280; + } else { + rate0_Q5 = silk_SMLABB( 280 - 43 * NLSF_QUANT_MAX_AMPLITUDE, 43, ind_tmp ); + rate1_Q5 = silk_ADD16( rate0_Q5, 43 ); + } + } else if( ind_tmp <= -NLSF_QUANT_MAX_AMPLITUDE ) { + if( ind_tmp == -NLSF_QUANT_MAX_AMPLITUDE ) { + rate0_Q5 = 280; + rate1_Q5 = rates_Q5[ ind_tmp + 1 + NLSF_QUANT_MAX_AMPLITUDE ]; + } else { + rate0_Q5 = silk_SMLABB( 280 - 43 * NLSF_QUANT_MAX_AMPLITUDE, -43, ind_tmp ); + rate1_Q5 = silk_SUB16( rate0_Q5, 43 ); + } + } else { + rate0_Q5 = rates_Q5[ ind_tmp + NLSF_QUANT_MAX_AMPLITUDE ]; + rate1_Q5 = rates_Q5[ ind_tmp + 1 + NLSF_QUANT_MAX_AMPLITUDE ]; + } + RD_tmp_Q25 = RD_Q25[ j ]; + diff_Q10 = silk_SUB16( in_Q10, out0_Q10 ); + RD_Q25[ j ] = silk_SMLABB( silk_MLA( RD_tmp_Q25, silk_SMULBB( diff_Q10, diff_Q10 ), w_Q5[ i ] ), mu_Q20, rate0_Q5 ); + diff_Q10 = silk_SUB16( in_Q10, out1_Q10 ); + RD_Q25[ j + nStates ] = silk_SMLABB( silk_MLA( RD_tmp_Q25, silk_SMULBB( diff_Q10, diff_Q10 ), w_Q5[ i ] ), mu_Q20, rate1_Q5 ); + } + + if( nStates <= NLSF_QUANT_DEL_DEC_STATES/2 ) { + /* double number of states and copy */ + for( j = 0; j < nStates; j++ ) { + ind[ j + nStates ][ i ] = ind[ j ][ i ] + 1; + } + nStates = silk_LSHIFT( nStates, 1 ); + for( j = nStates; j < NLSF_QUANT_DEL_DEC_STATES; j++ ) { + ind[ j ][ i ] = ind[ j - nStates ][ i ]; + } + } else { + /* sort lower and upper half of RD_Q25, pairwise */ + for( j = 0; j < NLSF_QUANT_DEL_DEC_STATES; j++ ) { + if( RD_Q25[ j ] > RD_Q25[ j + NLSF_QUANT_DEL_DEC_STATES ] ) { + RD_max_Q25[ j ] = RD_Q25[ j ]; + RD_min_Q25[ j ] = RD_Q25[ j + NLSF_QUANT_DEL_DEC_STATES ]; + RD_Q25[ j ] = RD_min_Q25[ j ]; + RD_Q25[ j + NLSF_QUANT_DEL_DEC_STATES ] = RD_max_Q25[ j ]; + /* swap prev_out values */ + out0_Q10 = prev_out_Q10[ j ]; + prev_out_Q10[ j ] = prev_out_Q10[ j + NLSF_QUANT_DEL_DEC_STATES ]; + prev_out_Q10[ j + NLSF_QUANT_DEL_DEC_STATES ] = out0_Q10; + ind_sort[ j ] = j + NLSF_QUANT_DEL_DEC_STATES; + } else { + RD_min_Q25[ j ] = RD_Q25[ j ]; + RD_max_Q25[ j ] = RD_Q25[ j + NLSF_QUANT_DEL_DEC_STATES ]; + ind_sort[ j ] = j; + } + } + /* compare the highest RD values of the winning half with the lowest one in the losing half, and copy if necessary */ + /* afterwards ind_sort[] will contain the indices of the NLSF_QUANT_DEL_DEC_STATES winning RD values */ + while( 1 ) { + min_max_Q25 = silk_int32_MAX; + max_min_Q25 = 0; + ind_min_max = 0; + ind_max_min = 0; + for( j = 0; j < NLSF_QUANT_DEL_DEC_STATES; j++ ) { + if( min_max_Q25 > RD_max_Q25[ j ] ) { + min_max_Q25 = RD_max_Q25[ j ]; + ind_min_max = j; + } + if( max_min_Q25 < RD_min_Q25[ j ] ) { + max_min_Q25 = RD_min_Q25[ j ]; + ind_max_min = j; + } + } + if( min_max_Q25 >= max_min_Q25 ) { + break; + } + /* copy ind_min_max to ind_max_min */ + ind_sort[ ind_max_min ] = ind_sort[ ind_min_max ] ^ NLSF_QUANT_DEL_DEC_STATES; + RD_Q25[ ind_max_min ] = RD_Q25[ ind_min_max + NLSF_QUANT_DEL_DEC_STATES ]; + prev_out_Q10[ ind_max_min ] = prev_out_Q10[ ind_min_max + NLSF_QUANT_DEL_DEC_STATES ]; + RD_min_Q25[ ind_max_min ] = 0; + RD_max_Q25[ ind_min_max ] = silk_int32_MAX; + silk_memcpy( ind[ ind_max_min ], ind[ ind_min_max ], MAX_LPC_ORDER * sizeof( opus_int8 ) ); + } + /* increment index if it comes from the upper half */ + for( j = 0; j < NLSF_QUANT_DEL_DEC_STATES; j++ ) { + ind[ j ][ i ] += silk_RSHIFT( ind_sort[ j ], NLSF_QUANT_DEL_DEC_STATES_LOG2 ); + } + } + } + + /* last sample: find winner, copy indices and return RD value */ + ind_tmp = 0; + min_Q25 = silk_int32_MAX; + for( j = 0; j < 2 * NLSF_QUANT_DEL_DEC_STATES; j++ ) { + if( min_Q25 > RD_Q25[ j ] ) { + min_Q25 = RD_Q25[ j ]; + ind_tmp = j; + } + } + for( j = 0; j < order; j++ ) { + indices[ j ] = ind[ ind_tmp & ( NLSF_QUANT_DEL_DEC_STATES - 1 ) ][ j ]; + silk_assert( indices[ j ] >= -NLSF_QUANT_MAX_AMPLITUDE_EXT ); + silk_assert( indices[ j ] <= NLSF_QUANT_MAX_AMPLITUDE_EXT ); + } + indices[ 0 ] += silk_RSHIFT( ind_tmp, NLSF_QUANT_DEL_DEC_STATES_LOG2 ); + silk_assert( indices[ 0 ] <= NLSF_QUANT_MAX_AMPLITUDE_EXT ); + silk_assert( min_Q25 >= 0 ); + return min_Q25; +} diff --git a/media/libopus/silk/NLSF_encode.c b/media/libopus/silk/NLSF_encode.c new file mode 100644 index 0000000000..01ac7db78c --- /dev/null +++ b/media/libopus/silk/NLSF_encode.c @@ -0,0 +1,124 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" +#include "stack_alloc.h" + +/***********************/ +/* NLSF vector encoder */ +/***********************/ +opus_int32 silk_NLSF_encode( /* O Returns RD value in Q25 */ + opus_int8 *NLSFIndices, /* I Codebook path vector [ LPC_ORDER + 1 ] */ + opus_int16 *pNLSF_Q15, /* I/O (Un)quantized NLSF vector [ LPC_ORDER ] */ + const silk_NLSF_CB_struct *psNLSF_CB, /* I Codebook object */ + const opus_int16 *pW_Q2, /* I NLSF weight vector [ LPC_ORDER ] */ + const opus_int NLSF_mu_Q20, /* I Rate weight for the RD optimization */ + const opus_int nSurvivors, /* I Max survivors after first stage */ + const opus_int signalType /* I Signal type: 0/1/2 */ +) +{ + opus_int i, s, ind1, bestIndex, prob_Q8, bits_q7; + opus_int32 W_tmp_Q9, ret; + VARDECL( opus_int32, err_Q24 ); + VARDECL( opus_int32, RD_Q25 ); + VARDECL( opus_int, tempIndices1 ); + VARDECL( opus_int8, tempIndices2 ); + opus_int16 res_Q10[ MAX_LPC_ORDER ]; + opus_int16 NLSF_tmp_Q15[ MAX_LPC_ORDER ]; + opus_int16 W_adj_Q5[ MAX_LPC_ORDER ]; + opus_uint8 pred_Q8[ MAX_LPC_ORDER ]; + opus_int16 ec_ix[ MAX_LPC_ORDER ]; + const opus_uint8 *pCB_element, *iCDF_ptr; + const opus_int16 *pCB_Wght_Q9; + SAVE_STACK; + + celt_assert( signalType >= 0 && signalType <= 2 ); + silk_assert( NLSF_mu_Q20 <= 32767 && NLSF_mu_Q20 >= 0 ); + + /* NLSF stabilization */ + silk_NLSF_stabilize( pNLSF_Q15, psNLSF_CB->deltaMin_Q15, psNLSF_CB->order ); + + /* First stage: VQ */ + ALLOC( err_Q24, psNLSF_CB->nVectors, opus_int32 ); + silk_NLSF_VQ( err_Q24, pNLSF_Q15, psNLSF_CB->CB1_NLSF_Q8, psNLSF_CB->CB1_Wght_Q9, psNLSF_CB->nVectors, psNLSF_CB->order ); + + /* Sort the quantization errors */ + ALLOC( tempIndices1, nSurvivors, opus_int ); + silk_insertion_sort_increasing( err_Q24, tempIndices1, psNLSF_CB->nVectors, nSurvivors ); + + ALLOC( RD_Q25, nSurvivors, opus_int32 ); + ALLOC( tempIndices2, nSurvivors * MAX_LPC_ORDER, opus_int8 ); + + /* Loop over survivors */ + for( s = 0; s < nSurvivors; s++ ) { + ind1 = tempIndices1[ s ]; + + /* Residual after first stage */ + pCB_element = &psNLSF_CB->CB1_NLSF_Q8[ ind1 * psNLSF_CB->order ]; + pCB_Wght_Q9 = &psNLSF_CB->CB1_Wght_Q9[ ind1 * psNLSF_CB->order ]; + for( i = 0; i < psNLSF_CB->order; i++ ) { + NLSF_tmp_Q15[ i ] = silk_LSHIFT16( (opus_int16)pCB_element[ i ], 7 ); + W_tmp_Q9 = pCB_Wght_Q9[ i ]; + res_Q10[ i ] = (opus_int16)silk_RSHIFT( silk_SMULBB( pNLSF_Q15[ i ] - NLSF_tmp_Q15[ i ], W_tmp_Q9 ), 14 ); + W_adj_Q5[ i ] = silk_DIV32_varQ( (opus_int32)pW_Q2[ i ], silk_SMULBB( W_tmp_Q9, W_tmp_Q9 ), 21 ); + } + + /* Unpack entropy table indices and predictor for current CB1 index */ + silk_NLSF_unpack( ec_ix, pred_Q8, psNLSF_CB, ind1 ); + + /* Trellis quantizer */ + RD_Q25[ s ] = silk_NLSF_del_dec_quant( &tempIndices2[ s * MAX_LPC_ORDER ], res_Q10, W_adj_Q5, pred_Q8, ec_ix, + psNLSF_CB->ec_Rates_Q5, psNLSF_CB->quantStepSize_Q16, psNLSF_CB->invQuantStepSize_Q6, NLSF_mu_Q20, psNLSF_CB->order ); + + /* Add rate for first stage */ + iCDF_ptr = &psNLSF_CB->CB1_iCDF[ ( signalType >> 1 ) * psNLSF_CB->nVectors ]; + if( ind1 == 0 ) { + prob_Q8 = 256 - iCDF_ptr[ ind1 ]; + } else { + prob_Q8 = iCDF_ptr[ ind1 - 1 ] - iCDF_ptr[ ind1 ]; + } + bits_q7 = ( 8 << 7 ) - silk_lin2log( prob_Q8 ); + RD_Q25[ s ] = silk_SMLABB( RD_Q25[ s ], bits_q7, silk_RSHIFT( NLSF_mu_Q20, 2 ) ); + } + + /* Find the lowest rate-distortion error */ + silk_insertion_sort_increasing( RD_Q25, &bestIndex, nSurvivors, 1 ); + + NLSFIndices[ 0 ] = (opus_int8)tempIndices1[ bestIndex ]; + silk_memcpy( &NLSFIndices[ 1 ], &tempIndices2[ bestIndex * MAX_LPC_ORDER ], psNLSF_CB->order * sizeof( opus_int8 ) ); + + /* Decode */ + silk_NLSF_decode( pNLSF_Q15, NLSFIndices, psNLSF_CB ); + + ret = RD_Q25[ 0 ]; + RESTORE_STACK; + return ret; +} diff --git a/media/libopus/silk/NLSF_stabilize.c b/media/libopus/silk/NLSF_stabilize.c new file mode 100644 index 0000000000..8f3426b91e --- /dev/null +++ b/media/libopus/silk/NLSF_stabilize.c @@ -0,0 +1,142 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +/* NLSF stabilizer: */ +/* */ +/* - Moves NLSFs further apart if they are too close */ +/* - Moves NLSFs away from borders if they are too close */ +/* - High effort to achieve a modification with minimum */ +/* Euclidean distance to input vector */ +/* - Output are sorted NLSF coefficients */ +/* */ + +#include "SigProc_FIX.h" + +/* Constant Definitions */ +#define MAX_LOOPS 20 + +/* NLSF stabilizer, for a single input data vector */ +void silk_NLSF_stabilize( + opus_int16 *NLSF_Q15, /* I/O Unstable/stabilized normalized LSF vector in Q15 [L] */ + const opus_int16 *NDeltaMin_Q15, /* I Min distance vector, NDeltaMin_Q15[L] must be >= 1 [L+1] */ + const opus_int L /* I Number of NLSF parameters in the input vector */ +) +{ + opus_int i, I=0, k, loops; + opus_int16 center_freq_Q15; + opus_int32 diff_Q15, min_diff_Q15, min_center_Q15, max_center_Q15; + + /* This is necessary to ensure an output within range of a opus_int16 */ + silk_assert( NDeltaMin_Q15[L] >= 1 ); + + for( loops = 0; loops < MAX_LOOPS; loops++ ) { + /**************************/ + /* Find smallest distance */ + /**************************/ + /* First element */ + min_diff_Q15 = NLSF_Q15[0] - NDeltaMin_Q15[0]; + I = 0; + /* Middle elements */ + for( i = 1; i <= L-1; i++ ) { + diff_Q15 = NLSF_Q15[i] - ( NLSF_Q15[i-1] + NDeltaMin_Q15[i] ); + if( diff_Q15 < min_diff_Q15 ) { + min_diff_Q15 = diff_Q15; + I = i; + } + } + /* Last element */ + diff_Q15 = ( 1 << 15 ) - ( NLSF_Q15[L-1] + NDeltaMin_Q15[L] ); + if( diff_Q15 < min_diff_Q15 ) { + min_diff_Q15 = diff_Q15; + I = L; + } + + /***************************************************/ + /* Now check if the smallest distance non-negative */ + /***************************************************/ + if( min_diff_Q15 >= 0 ) { + return; + } + + if( I == 0 ) { + /* Move away from lower limit */ + NLSF_Q15[0] = NDeltaMin_Q15[0]; + + } else if( I == L) { + /* Move away from higher limit */ + NLSF_Q15[L-1] = ( 1 << 15 ) - NDeltaMin_Q15[L]; + + } else { + /* Find the lower extreme for the location of the current center frequency */ + min_center_Q15 = 0; + for( k = 0; k < I; k++ ) { + min_center_Q15 += NDeltaMin_Q15[k]; + } + min_center_Q15 += silk_RSHIFT( NDeltaMin_Q15[I], 1 ); + + /* Find the upper extreme for the location of the current center frequency */ + max_center_Q15 = 1 << 15; + for( k = L; k > I; k-- ) { + max_center_Q15 -= NDeltaMin_Q15[k]; + } + max_center_Q15 -= silk_RSHIFT( NDeltaMin_Q15[I], 1 ); + + /* Move apart, sorted by value, keeping the same center frequency */ + center_freq_Q15 = (opus_int16)silk_LIMIT_32( silk_RSHIFT_ROUND( (opus_int32)NLSF_Q15[I-1] + (opus_int32)NLSF_Q15[I], 1 ), + min_center_Q15, max_center_Q15 ); + NLSF_Q15[I-1] = center_freq_Q15 - silk_RSHIFT( NDeltaMin_Q15[I], 1 ); + NLSF_Q15[I] = NLSF_Q15[I-1] + NDeltaMin_Q15[I]; + } + } + + /* Safe and simple fall back method, which is less ideal than the above */ + if( loops == MAX_LOOPS ) + { + /* Insertion sort (fast for already almost sorted arrays): */ + /* Best case: O(n) for an already sorted array */ + /* Worst case: O(n^2) for an inversely sorted array */ + silk_insertion_sort_increasing_all_values_int16( &NLSF_Q15[0], L ); + + /* First NLSF should be no less than NDeltaMin[0] */ + NLSF_Q15[0] = silk_max_int( NLSF_Q15[0], NDeltaMin_Q15[0] ); + + /* Keep delta_min distance between the NLSFs */ + for( i = 1; i < L; i++ ) + NLSF_Q15[i] = silk_max_int( NLSF_Q15[i], silk_ADD_SAT16( NLSF_Q15[i-1], NDeltaMin_Q15[i] ) ); + + /* Last NLSF should be no higher than 1 - NDeltaMin[L] */ + NLSF_Q15[L-1] = silk_min_int( NLSF_Q15[L-1], (1<<15) - NDeltaMin_Q15[L] ); + + /* Keep NDeltaMin distance between the NLSFs */ + for( i = L-2; i >= 0; i-- ) + NLSF_Q15[i] = silk_min_int( NLSF_Q15[i], NLSF_Q15[i+1] - NDeltaMin_Q15[i+1] ); + } +} diff --git a/media/libopus/silk/NLSF_unpack.c b/media/libopus/silk/NLSF_unpack.c new file mode 100644 index 0000000000..17bd23f752 --- /dev/null +++ b/media/libopus/silk/NLSF_unpack.c @@ -0,0 +1,55 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" + +/* Unpack predictor values and indices for entropy coding tables */ +void silk_NLSF_unpack( + opus_int16 ec_ix[], /* O Indices to entropy tables [ LPC_ORDER ] */ + opus_uint8 pred_Q8[], /* O LSF predictor [ LPC_ORDER ] */ + const silk_NLSF_CB_struct *psNLSF_CB, /* I Codebook object */ + const opus_int CB1_index /* I Index of vector in first LSF codebook */ +) +{ + opus_int i; + opus_uint8 entry; + const opus_uint8 *ec_sel_ptr; + + ec_sel_ptr = &psNLSF_CB->ec_sel[ CB1_index * psNLSF_CB->order / 2 ]; + for( i = 0; i < psNLSF_CB->order; i += 2 ) { + entry = *ec_sel_ptr++; + ec_ix [ i ] = silk_SMULBB( silk_RSHIFT( entry, 1 ) & 7, 2 * NLSF_QUANT_MAX_AMPLITUDE + 1 ); + pred_Q8[ i ] = psNLSF_CB->pred_Q8[ i + ( entry & 1 ) * ( psNLSF_CB->order - 1 ) ]; + ec_ix [ i + 1 ] = silk_SMULBB( silk_RSHIFT( entry, 5 ) & 7, 2 * NLSF_QUANT_MAX_AMPLITUDE + 1 ); + pred_Q8[ i + 1 ] = psNLSF_CB->pred_Q8[ i + ( silk_RSHIFT( entry, 4 ) & 1 ) * ( psNLSF_CB->order - 1 ) + 1 ]; + } +} + diff --git a/media/libopus/silk/NSQ.c b/media/libopus/silk/NSQ.c new file mode 100644 index 0000000000..45dd45ce8d --- /dev/null +++ b/media/libopus/silk/NSQ.c @@ -0,0 +1,437 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" +#include "stack_alloc.h" +#include "NSQ.h" + + +static OPUS_INLINE void silk_nsq_scale_states( + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + const opus_int16 x16[], /* I input */ + opus_int32 x_sc_Q10[], /* O input scaled with 1/Gain */ + const opus_int16 sLTP[], /* I re-whitened LTP state in Q0 */ + opus_int32 sLTP_Q15[], /* O LTP state matching scaled input */ + opus_int subfr, /* I subframe number */ + const opus_int LTP_scale_Q14, /* I */ + const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lag */ + const opus_int signal_type /* I Signal type */ +); + +#if !defined(OPUS_X86_MAY_HAVE_SSE4_1) +static OPUS_INLINE void silk_noise_shape_quantizer( + silk_nsq_state *NSQ, /* I/O NSQ state */ + opus_int signalType, /* I Signal type */ + const opus_int32 x_sc_Q10[], /* I */ + opus_int8 pulses[], /* O */ + opus_int16 xq[], /* O */ + opus_int32 sLTP_Q15[], /* I/O LTP state */ + const opus_int16 a_Q12[], /* I Short term prediction coefs */ + const opus_int16 b_Q14[], /* I Long term prediction coefs */ + const opus_int16 AR_shp_Q13[], /* I Noise shaping AR coefs */ + opus_int lag, /* I Pitch lag */ + opus_int32 HarmShapeFIRPacked_Q14, /* I */ + opus_int Tilt_Q14, /* I Spectral tilt */ + opus_int32 LF_shp_Q14, /* I */ + opus_int32 Gain_Q16, /* I */ + opus_int Lambda_Q10, /* I */ + opus_int offset_Q10, /* I */ + opus_int length, /* I Input length */ + opus_int shapingLPCOrder, /* I Noise shaping AR filter order */ + opus_int predictLPCOrder, /* I Prediction filter order */ + int arch /* I Architecture */ +); +#endif + +void silk_NSQ_c +( + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + SideInfoIndices *psIndices, /* I/O Quantization Indices */ + const opus_int16 x16[], /* I Input */ + opus_int8 pulses[], /* O Quantized pulse signal */ + const opus_int16 PredCoef_Q12[ 2 * MAX_LPC_ORDER ], /* I Short term prediction coefs */ + const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ + const opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ + const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ + const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ + const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ + const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ + const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ + const opus_int LTP_scale_Q14 /* I LTP state scaling */ +) +{ + opus_int k, lag, start_idx, LSF_interpolation_flag; + const opus_int16 *A_Q12, *B_Q14, *AR_shp_Q13; + opus_int16 *pxq; + VARDECL( opus_int32, sLTP_Q15 ); + VARDECL( opus_int16, sLTP ); + opus_int32 HarmShapeFIRPacked_Q14; + opus_int offset_Q10; + VARDECL( opus_int32, x_sc_Q10 ); + SAVE_STACK; + + NSQ->rand_seed = psIndices->Seed; + + /* Set unvoiced lag to the previous one, overwrite later for voiced */ + lag = NSQ->lagPrev; + + silk_assert( NSQ->prev_gain_Q16 != 0 ); + + offset_Q10 = silk_Quantization_Offsets_Q10[ psIndices->signalType >> 1 ][ psIndices->quantOffsetType ]; + + if( psIndices->NLSFInterpCoef_Q2 == 4 ) { + LSF_interpolation_flag = 0; + } else { + LSF_interpolation_flag = 1; + } + + ALLOC( sLTP_Q15, psEncC->ltp_mem_length + psEncC->frame_length, opus_int32 ); + ALLOC( sLTP, psEncC->ltp_mem_length + psEncC->frame_length, opus_int16 ); + ALLOC( x_sc_Q10, psEncC->subfr_length, opus_int32 ); + /* Set up pointers to start of sub frame */ + NSQ->sLTP_shp_buf_idx = psEncC->ltp_mem_length; + NSQ->sLTP_buf_idx = psEncC->ltp_mem_length; + pxq = &NSQ->xq[ psEncC->ltp_mem_length ]; + for( k = 0; k < psEncC->nb_subfr; k++ ) { + A_Q12 = &PredCoef_Q12[ (( k >> 1 ) | ( 1 - LSF_interpolation_flag )) * MAX_LPC_ORDER ]; + B_Q14 = <PCoef_Q14[ k * LTP_ORDER ]; + AR_shp_Q13 = &AR_Q13[ k * MAX_SHAPE_LPC_ORDER ]; + + /* Noise shape parameters */ + silk_assert( HarmShapeGain_Q14[ k ] >= 0 ); + HarmShapeFIRPacked_Q14 = silk_RSHIFT( HarmShapeGain_Q14[ k ], 2 ); + HarmShapeFIRPacked_Q14 |= silk_LSHIFT( (opus_int32)silk_RSHIFT( HarmShapeGain_Q14[ k ], 1 ), 16 ); + + NSQ->rewhite_flag = 0; + if( psIndices->signalType == TYPE_VOICED ) { + /* Voiced */ + lag = pitchL[ k ]; + + /* Re-whitening */ + if( ( k & ( 3 - silk_LSHIFT( LSF_interpolation_flag, 1 ) ) ) == 0 ) { + /* Rewhiten with new A coefs */ + start_idx = psEncC->ltp_mem_length - lag - psEncC->predictLPCOrder - LTP_ORDER / 2; + celt_assert( start_idx > 0 ); + + silk_LPC_analysis_filter( &sLTP[ start_idx ], &NSQ->xq[ start_idx + k * psEncC->subfr_length ], + A_Q12, psEncC->ltp_mem_length - start_idx, psEncC->predictLPCOrder, psEncC->arch ); + + NSQ->rewhite_flag = 1; + NSQ->sLTP_buf_idx = psEncC->ltp_mem_length; + } + } + + silk_nsq_scale_states( psEncC, NSQ, x16, x_sc_Q10, sLTP, sLTP_Q15, k, LTP_scale_Q14, Gains_Q16, pitchL, psIndices->signalType ); + + silk_noise_shape_quantizer( NSQ, psIndices->signalType, x_sc_Q10, pulses, pxq, sLTP_Q15, A_Q12, B_Q14, + AR_shp_Q13, lag, HarmShapeFIRPacked_Q14, Tilt_Q14[ k ], LF_shp_Q14[ k ], Gains_Q16[ k ], Lambda_Q10, + offset_Q10, psEncC->subfr_length, psEncC->shapingLPCOrder, psEncC->predictLPCOrder, psEncC->arch ); + + x16 += psEncC->subfr_length; + pulses += psEncC->subfr_length; + pxq += psEncC->subfr_length; + } + + /* Update lagPrev for next frame */ + NSQ->lagPrev = pitchL[ psEncC->nb_subfr - 1 ]; + + /* Save quantized speech and noise shaping signals */ + silk_memmove( NSQ->xq, &NSQ->xq[ psEncC->frame_length ], psEncC->ltp_mem_length * sizeof( opus_int16 ) ); + silk_memmove( NSQ->sLTP_shp_Q14, &NSQ->sLTP_shp_Q14[ psEncC->frame_length ], psEncC->ltp_mem_length * sizeof( opus_int32 ) ); + RESTORE_STACK; +} + +/******************************/ +/* silk_noise_shape_quantizer */ +/******************************/ + +#if !defined(OPUS_X86_MAY_HAVE_SSE4_1) +static OPUS_INLINE +#endif +void silk_noise_shape_quantizer( + silk_nsq_state *NSQ, /* I/O NSQ state */ + opus_int signalType, /* I Signal type */ + const opus_int32 x_sc_Q10[], /* I */ + opus_int8 pulses[], /* O */ + opus_int16 xq[], /* O */ + opus_int32 sLTP_Q15[], /* I/O LTP state */ + const opus_int16 a_Q12[], /* I Short term prediction coefs */ + const opus_int16 b_Q14[], /* I Long term prediction coefs */ + const opus_int16 AR_shp_Q13[], /* I Noise shaping AR coefs */ + opus_int lag, /* I Pitch lag */ + opus_int32 HarmShapeFIRPacked_Q14, /* I */ + opus_int Tilt_Q14, /* I Spectral tilt */ + opus_int32 LF_shp_Q14, /* I */ + opus_int32 Gain_Q16, /* I */ + opus_int Lambda_Q10, /* I */ + opus_int offset_Q10, /* I */ + opus_int length, /* I Input length */ + opus_int shapingLPCOrder, /* I Noise shaping AR filter order */ + opus_int predictLPCOrder, /* I Prediction filter order */ + int arch /* I Architecture */ +) +{ + opus_int i; + opus_int32 LTP_pred_Q13, LPC_pred_Q10, n_AR_Q12, n_LTP_Q13; + opus_int32 n_LF_Q12, r_Q10, rr_Q10, q1_Q0, q1_Q10, q2_Q10, rd1_Q20, rd2_Q20; + opus_int32 exc_Q14, LPC_exc_Q14, xq_Q14, Gain_Q10; + opus_int32 tmp1, tmp2, sLF_AR_shp_Q14; + opus_int32 *psLPC_Q14, *shp_lag_ptr, *pred_lag_ptr; +#ifdef silk_short_prediction_create_arch_coef + opus_int32 a_Q12_arch[MAX_LPC_ORDER]; +#endif + + shp_lag_ptr = &NSQ->sLTP_shp_Q14[ NSQ->sLTP_shp_buf_idx - lag + HARM_SHAPE_FIR_TAPS / 2 ]; + pred_lag_ptr = &sLTP_Q15[ NSQ->sLTP_buf_idx - lag + LTP_ORDER / 2 ]; + Gain_Q10 = silk_RSHIFT( Gain_Q16, 6 ); + + /* Set up short term AR state */ + psLPC_Q14 = &NSQ->sLPC_Q14[ NSQ_LPC_BUF_LENGTH - 1 ]; + +#ifdef silk_short_prediction_create_arch_coef + silk_short_prediction_create_arch_coef(a_Q12_arch, a_Q12, predictLPCOrder); +#endif + + for( i = 0; i < length; i++ ) { + /* Generate dither */ + NSQ->rand_seed = silk_RAND( NSQ->rand_seed ); + + /* Short-term prediction */ + LPC_pred_Q10 = silk_noise_shape_quantizer_short_prediction(psLPC_Q14, a_Q12, a_Q12_arch, predictLPCOrder, arch); + + /* Long-term prediction */ + if( signalType == TYPE_VOICED ) { + /* Unrolled loop */ + /* Avoids introducing a bias because silk_SMLAWB() always rounds to -inf */ + LTP_pred_Q13 = 2; + LTP_pred_Q13 = silk_SMLAWB( LTP_pred_Q13, pred_lag_ptr[ 0 ], b_Q14[ 0 ] ); + LTP_pred_Q13 = silk_SMLAWB( LTP_pred_Q13, pred_lag_ptr[ -1 ], b_Q14[ 1 ] ); + LTP_pred_Q13 = silk_SMLAWB( LTP_pred_Q13, pred_lag_ptr[ -2 ], b_Q14[ 2 ] ); + LTP_pred_Q13 = silk_SMLAWB( LTP_pred_Q13, pred_lag_ptr[ -3 ], b_Q14[ 3 ] ); + LTP_pred_Q13 = silk_SMLAWB( LTP_pred_Q13, pred_lag_ptr[ -4 ], b_Q14[ 4 ] ); + pred_lag_ptr++; + } else { + LTP_pred_Q13 = 0; + } + + /* Noise shape feedback */ + celt_assert( ( shapingLPCOrder & 1 ) == 0 ); /* check that order is even */ + n_AR_Q12 = silk_NSQ_noise_shape_feedback_loop(&NSQ->sDiff_shp_Q14, NSQ->sAR2_Q14, AR_shp_Q13, shapingLPCOrder, arch); + + n_AR_Q12 = silk_SMLAWB( n_AR_Q12, NSQ->sLF_AR_shp_Q14, Tilt_Q14 ); + + n_LF_Q12 = silk_SMULWB( NSQ->sLTP_shp_Q14[ NSQ->sLTP_shp_buf_idx - 1 ], LF_shp_Q14 ); + n_LF_Q12 = silk_SMLAWT( n_LF_Q12, NSQ->sLF_AR_shp_Q14, LF_shp_Q14 ); + + celt_assert( lag > 0 || signalType != TYPE_VOICED ); + + /* Combine prediction and noise shaping signals */ + tmp1 = silk_SUB32( silk_LSHIFT32( LPC_pred_Q10, 2 ), n_AR_Q12 ); /* Q12 */ + tmp1 = silk_SUB32( tmp1, n_LF_Q12 ); /* Q12 */ + if( lag > 0 ) { + /* Symmetric, packed FIR coefficients */ + n_LTP_Q13 = silk_SMULWB( silk_ADD_SAT32( shp_lag_ptr[ 0 ], shp_lag_ptr[ -2 ] ), HarmShapeFIRPacked_Q14 ); + n_LTP_Q13 = silk_SMLAWT( n_LTP_Q13, shp_lag_ptr[ -1 ], HarmShapeFIRPacked_Q14 ); + n_LTP_Q13 = silk_LSHIFT( n_LTP_Q13, 1 ); + shp_lag_ptr++; + + tmp2 = silk_SUB32( LTP_pred_Q13, n_LTP_Q13 ); /* Q13 */ + tmp1 = silk_ADD_LSHIFT32( tmp2, tmp1, 1 ); /* Q13 */ + tmp1 = silk_RSHIFT_ROUND( tmp1, 3 ); /* Q10 */ + } else { + tmp1 = silk_RSHIFT_ROUND( tmp1, 2 ); /* Q10 */ + } + + r_Q10 = silk_SUB32( x_sc_Q10[ i ], tmp1 ); /* residual error Q10 */ + + /* Flip sign depending on dither */ + if( NSQ->rand_seed < 0 ) { + r_Q10 = -r_Q10; + } + r_Q10 = silk_LIMIT_32( r_Q10, -(31 << 10), 30 << 10 ); + + /* Find two quantization level candidates and measure their rate-distortion */ + q1_Q10 = silk_SUB32( r_Q10, offset_Q10 ); + q1_Q0 = silk_RSHIFT( q1_Q10, 10 ); + if (Lambda_Q10 > 2048) { + /* For aggressive RDO, the bias becomes more than one pulse. */ + int rdo_offset = Lambda_Q10/2 - 512; + if (q1_Q10 > rdo_offset) { + q1_Q0 = silk_RSHIFT( q1_Q10 - rdo_offset, 10 ); + } else if (q1_Q10 < -rdo_offset) { + q1_Q0 = silk_RSHIFT( q1_Q10 + rdo_offset, 10 ); + } else if (q1_Q10 < 0) { + q1_Q0 = -1; + } else { + q1_Q0 = 0; + } + } + if( q1_Q0 > 0 ) { + q1_Q10 = silk_SUB32( silk_LSHIFT( q1_Q0, 10 ), QUANT_LEVEL_ADJUST_Q10 ); + q1_Q10 = silk_ADD32( q1_Q10, offset_Q10 ); + q2_Q10 = silk_ADD32( q1_Q10, 1024 ); + rd1_Q20 = silk_SMULBB( q1_Q10, Lambda_Q10 ); + rd2_Q20 = silk_SMULBB( q2_Q10, Lambda_Q10 ); + } else if( q1_Q0 == 0 ) { + q1_Q10 = offset_Q10; + q2_Q10 = silk_ADD32( q1_Q10, 1024 - QUANT_LEVEL_ADJUST_Q10 ); + rd1_Q20 = silk_SMULBB( q1_Q10, Lambda_Q10 ); + rd2_Q20 = silk_SMULBB( q2_Q10, Lambda_Q10 ); + } else if( q1_Q0 == -1 ) { + q2_Q10 = offset_Q10; + q1_Q10 = silk_SUB32( q2_Q10, 1024 - QUANT_LEVEL_ADJUST_Q10 ); + rd1_Q20 = silk_SMULBB( -q1_Q10, Lambda_Q10 ); + rd2_Q20 = silk_SMULBB( q2_Q10, Lambda_Q10 ); + } else { /* Q1_Q0 < -1 */ + q1_Q10 = silk_ADD32( silk_LSHIFT( q1_Q0, 10 ), QUANT_LEVEL_ADJUST_Q10 ); + q1_Q10 = silk_ADD32( q1_Q10, offset_Q10 ); + q2_Q10 = silk_ADD32( q1_Q10, 1024 ); + rd1_Q20 = silk_SMULBB( -q1_Q10, Lambda_Q10 ); + rd2_Q20 = silk_SMULBB( -q2_Q10, Lambda_Q10 ); + } + rr_Q10 = silk_SUB32( r_Q10, q1_Q10 ); + rd1_Q20 = silk_SMLABB( rd1_Q20, rr_Q10, rr_Q10 ); + rr_Q10 = silk_SUB32( r_Q10, q2_Q10 ); + rd2_Q20 = silk_SMLABB( rd2_Q20, rr_Q10, rr_Q10 ); + + if( rd2_Q20 < rd1_Q20 ) { + q1_Q10 = q2_Q10; + } + + pulses[ i ] = (opus_int8)silk_RSHIFT_ROUND( q1_Q10, 10 ); + + /* Excitation */ + exc_Q14 = silk_LSHIFT( q1_Q10, 4 ); + if ( NSQ->rand_seed < 0 ) { + exc_Q14 = -exc_Q14; + } + + /* Add predictions */ + LPC_exc_Q14 = silk_ADD_LSHIFT32( exc_Q14, LTP_pred_Q13, 1 ); + xq_Q14 = silk_ADD_LSHIFT32( LPC_exc_Q14, LPC_pred_Q10, 4 ); + + /* Scale XQ back to normal level before saving */ + xq[ i ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( silk_SMULWW( xq_Q14, Gain_Q10 ), 8 ) ); + + /* Update states */ + psLPC_Q14++; + *psLPC_Q14 = xq_Q14; + NSQ->sDiff_shp_Q14 = silk_SUB_LSHIFT32( xq_Q14, x_sc_Q10[ i ], 4 ); + sLF_AR_shp_Q14 = silk_SUB_LSHIFT32( NSQ->sDiff_shp_Q14, n_AR_Q12, 2 ); + NSQ->sLF_AR_shp_Q14 = sLF_AR_shp_Q14; + + NSQ->sLTP_shp_Q14[ NSQ->sLTP_shp_buf_idx ] = silk_SUB_LSHIFT32( sLF_AR_shp_Q14, n_LF_Q12, 2 ); + sLTP_Q15[ NSQ->sLTP_buf_idx ] = silk_LSHIFT( LPC_exc_Q14, 1 ); + NSQ->sLTP_shp_buf_idx++; + NSQ->sLTP_buf_idx++; + + /* Make dither dependent on quantized signal */ + NSQ->rand_seed = silk_ADD32_ovflw( NSQ->rand_seed, pulses[ i ] ); + } + + /* Update LPC synth buffer */ + silk_memcpy( NSQ->sLPC_Q14, &NSQ->sLPC_Q14[ length ], NSQ_LPC_BUF_LENGTH * sizeof( opus_int32 ) ); +} + +static OPUS_INLINE void silk_nsq_scale_states( + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + const opus_int16 x16[], /* I input */ + opus_int32 x_sc_Q10[], /* O input scaled with 1/Gain */ + const opus_int16 sLTP[], /* I re-whitened LTP state in Q0 */ + opus_int32 sLTP_Q15[], /* O LTP state matching scaled input */ + opus_int subfr, /* I subframe number */ + const opus_int LTP_scale_Q14, /* I */ + const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lag */ + const opus_int signal_type /* I Signal type */ +) +{ + opus_int i, lag; + opus_int32 gain_adj_Q16, inv_gain_Q31, inv_gain_Q26; + + lag = pitchL[ subfr ]; + inv_gain_Q31 = silk_INVERSE32_varQ( silk_max( Gains_Q16[ subfr ], 1 ), 47 ); + silk_assert( inv_gain_Q31 != 0 ); + + /* Scale input */ + inv_gain_Q26 = silk_RSHIFT_ROUND( inv_gain_Q31, 5 ); + for( i = 0; i < psEncC->subfr_length; i++ ) { + x_sc_Q10[ i ] = silk_SMULWW( x16[ i ], inv_gain_Q26 ); + } + + /* After rewhitening the LTP state is un-scaled, so scale with inv_gain_Q16 */ + if( NSQ->rewhite_flag ) { + if( subfr == 0 ) { + /* Do LTP downscaling */ + inv_gain_Q31 = silk_LSHIFT( silk_SMULWB( inv_gain_Q31, LTP_scale_Q14 ), 2 ); + } + for( i = NSQ->sLTP_buf_idx - lag - LTP_ORDER / 2; i < NSQ->sLTP_buf_idx; i++ ) { + silk_assert( i < MAX_FRAME_LENGTH ); + sLTP_Q15[ i ] = silk_SMULWB( inv_gain_Q31, sLTP[ i ] ); + } + } + + /* Adjust for changing gain */ + if( Gains_Q16[ subfr ] != NSQ->prev_gain_Q16 ) { + gain_adj_Q16 = silk_DIV32_varQ( NSQ->prev_gain_Q16, Gains_Q16[ subfr ], 16 ); + + /* Scale long-term shaping state */ + for( i = NSQ->sLTP_shp_buf_idx - psEncC->ltp_mem_length; i < NSQ->sLTP_shp_buf_idx; i++ ) { + NSQ->sLTP_shp_Q14[ i ] = silk_SMULWW( gain_adj_Q16, NSQ->sLTP_shp_Q14[ i ] ); + } + + /* Scale long-term prediction state */ + if( signal_type == TYPE_VOICED && NSQ->rewhite_flag == 0 ) { + for( i = NSQ->sLTP_buf_idx - lag - LTP_ORDER / 2; i < NSQ->sLTP_buf_idx; i++ ) { + sLTP_Q15[ i ] = silk_SMULWW( gain_adj_Q16, sLTP_Q15[ i ] ); + } + } + + NSQ->sLF_AR_shp_Q14 = silk_SMULWW( gain_adj_Q16, NSQ->sLF_AR_shp_Q14 ); + NSQ->sDiff_shp_Q14 = silk_SMULWW( gain_adj_Q16, NSQ->sDiff_shp_Q14 ); + + /* Scale short-term prediction and shaping states */ + for( i = 0; i < NSQ_LPC_BUF_LENGTH; i++ ) { + NSQ->sLPC_Q14[ i ] = silk_SMULWW( gain_adj_Q16, NSQ->sLPC_Q14[ i ] ); + } + for( i = 0; i < MAX_SHAPE_LPC_ORDER; i++ ) { + NSQ->sAR2_Q14[ i ] = silk_SMULWW( gain_adj_Q16, NSQ->sAR2_Q14[ i ] ); + } + + /* Save inverse gain */ + NSQ->prev_gain_Q16 = Gains_Q16[ subfr ]; + } +} diff --git a/media/libopus/silk/NSQ.h b/media/libopus/silk/NSQ.h new file mode 100644 index 0000000000..971832f660 --- /dev/null +++ b/media/libopus/silk/NSQ.h @@ -0,0 +1,101 @@ +/*********************************************************************** +Copyright (c) 2014 Vidyo. +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ +#ifndef SILK_NSQ_H +#define SILK_NSQ_H + +#include "SigProc_FIX.h" + +#undef silk_short_prediction_create_arch_coef + +static OPUS_INLINE opus_int32 silk_noise_shape_quantizer_short_prediction_c(const opus_int32 *buf32, const opus_int16 *coef16, opus_int order) +{ + opus_int32 out; + silk_assert( order == 10 || order == 16 ); + + /* Avoids introducing a bias because silk_SMLAWB() always rounds to -inf */ + out = silk_RSHIFT( order, 1 ); + out = silk_SMLAWB( out, buf32[ 0 ], coef16[ 0 ] ); + out = silk_SMLAWB( out, buf32[ -1 ], coef16[ 1 ] ); + out = silk_SMLAWB( out, buf32[ -2 ], coef16[ 2 ] ); + out = silk_SMLAWB( out, buf32[ -3 ], coef16[ 3 ] ); + out = silk_SMLAWB( out, buf32[ -4 ], coef16[ 4 ] ); + out = silk_SMLAWB( out, buf32[ -5 ], coef16[ 5 ] ); + out = silk_SMLAWB( out, buf32[ -6 ], coef16[ 6 ] ); + out = silk_SMLAWB( out, buf32[ -7 ], coef16[ 7 ] ); + out = silk_SMLAWB( out, buf32[ -8 ], coef16[ 8 ] ); + out = silk_SMLAWB( out, buf32[ -9 ], coef16[ 9 ] ); + + if( order == 16 ) + { + out = silk_SMLAWB( out, buf32[ -10 ], coef16[ 10 ] ); + out = silk_SMLAWB( out, buf32[ -11 ], coef16[ 11 ] ); + out = silk_SMLAWB( out, buf32[ -12 ], coef16[ 12 ] ); + out = silk_SMLAWB( out, buf32[ -13 ], coef16[ 13 ] ); + out = silk_SMLAWB( out, buf32[ -14 ], coef16[ 14 ] ); + out = silk_SMLAWB( out, buf32[ -15 ], coef16[ 15 ] ); + } + return out; +} + +#define silk_noise_shape_quantizer_short_prediction(in, coef, coefRev, order, arch) ((void)arch,silk_noise_shape_quantizer_short_prediction_c(in, coef, order)) + +static OPUS_INLINE opus_int32 silk_NSQ_noise_shape_feedback_loop_c(const opus_int32 *data0, opus_int32 *data1, const opus_int16 *coef, opus_int order) +{ + opus_int32 out; + opus_int32 tmp1, tmp2; + opus_int j; + + tmp2 = data0[0]; + tmp1 = data1[0]; + data1[0] = tmp2; + + out = silk_RSHIFT(order, 1); + out = silk_SMLAWB(out, tmp2, coef[0]); + + for (j = 2; j < order; j += 2) { + tmp2 = data1[j - 1]; + data1[j - 1] = tmp1; + out = silk_SMLAWB(out, tmp1, coef[j - 1]); + tmp1 = data1[j + 0]; + data1[j + 0] = tmp2; + out = silk_SMLAWB(out, tmp2, coef[j]); + } + data1[order - 1] = tmp1; + out = silk_SMLAWB(out, tmp1, coef[order - 1]); + /* Q11 -> Q12 */ + out = silk_LSHIFT32( out, 1 ); + return out; +} + +#define silk_NSQ_noise_shape_feedback_loop(data0, data1, coef, order, arch) ((void)arch,silk_NSQ_noise_shape_feedback_loop_c(data0, data1, coef, order)) + +#if defined(OPUS_ARM_MAY_HAVE_NEON_INTR) +#include "arm/NSQ_neon.h" +#endif + +#endif /* SILK_NSQ_H */ diff --git a/media/libopus/silk/NSQ_del_dec.c b/media/libopus/silk/NSQ_del_dec.c new file mode 100644 index 0000000000..41f3fc93ef --- /dev/null +++ b/media/libopus/silk/NSQ_del_dec.c @@ -0,0 +1,733 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" +#include "stack_alloc.h" +#include "NSQ.h" + + +typedef struct { + opus_int32 sLPC_Q14[ MAX_SUB_FRAME_LENGTH + NSQ_LPC_BUF_LENGTH ]; + opus_int32 RandState[ DECISION_DELAY ]; + opus_int32 Q_Q10[ DECISION_DELAY ]; + opus_int32 Xq_Q14[ DECISION_DELAY ]; + opus_int32 Pred_Q15[ DECISION_DELAY ]; + opus_int32 Shape_Q14[ DECISION_DELAY ]; + opus_int32 sAR2_Q14[ MAX_SHAPE_LPC_ORDER ]; + opus_int32 LF_AR_Q14; + opus_int32 Diff_Q14; + opus_int32 Seed; + opus_int32 SeedInit; + opus_int32 RD_Q10; +} NSQ_del_dec_struct; + +typedef struct { + opus_int32 Q_Q10; + opus_int32 RD_Q10; + opus_int32 xq_Q14; + opus_int32 LF_AR_Q14; + opus_int32 Diff_Q14; + opus_int32 sLTP_shp_Q14; + opus_int32 LPC_exc_Q14; +} NSQ_sample_struct; + +typedef NSQ_sample_struct NSQ_sample_pair[ 2 ]; + +#if defined(MIPSr1_ASM) +#include "mips/NSQ_del_dec_mipsr1.h" +#endif +static OPUS_INLINE void silk_nsq_del_dec_scale_states( + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + NSQ_del_dec_struct psDelDec[], /* I/O Delayed decision states */ + const opus_int16 x16[], /* I Input */ + opus_int32 x_sc_Q10[], /* O Input scaled with 1/Gain in Q10 */ + const opus_int16 sLTP[], /* I Re-whitened LTP state in Q0 */ + opus_int32 sLTP_Q15[], /* O LTP state matching scaled input */ + opus_int subfr, /* I Subframe number */ + opus_int nStatesDelayedDecision, /* I Number of del dec states */ + const opus_int LTP_scale_Q14, /* I LTP state scaling */ + const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lag */ + const opus_int signal_type, /* I Signal type */ + const opus_int decisionDelay /* I Decision delay */ +); + +/******************************************/ +/* Noise shape quantizer for one subframe */ +/******************************************/ +static OPUS_INLINE void silk_noise_shape_quantizer_del_dec( + silk_nsq_state *NSQ, /* I/O NSQ state */ + NSQ_del_dec_struct psDelDec[], /* I/O Delayed decision states */ + opus_int signalType, /* I Signal type */ + const opus_int32 x_Q10[], /* I */ + opus_int8 pulses[], /* O */ + opus_int16 xq[], /* O */ + opus_int32 sLTP_Q15[], /* I/O LTP filter state */ + opus_int32 delayedGain_Q10[], /* I/O Gain delay buffer */ + const opus_int16 a_Q12[], /* I Short term prediction coefs */ + const opus_int16 b_Q14[], /* I Long term prediction coefs */ + const opus_int16 AR_shp_Q13[], /* I Noise shaping coefs */ + opus_int lag, /* I Pitch lag */ + opus_int32 HarmShapeFIRPacked_Q14, /* I */ + opus_int Tilt_Q14, /* I Spectral tilt */ + opus_int32 LF_shp_Q14, /* I */ + opus_int32 Gain_Q16, /* I */ + opus_int Lambda_Q10, /* I */ + opus_int offset_Q10, /* I */ + opus_int length, /* I Input length */ + opus_int subfr, /* I Subframe number */ + opus_int shapingLPCOrder, /* I Shaping LPC filter order */ + opus_int predictLPCOrder, /* I Prediction filter order */ + opus_int warping_Q16, /* I */ + opus_int nStatesDelayedDecision, /* I Number of states in decision tree */ + opus_int *smpl_buf_idx, /* I/O Index to newest samples in buffers */ + opus_int decisionDelay, /* I */ + int arch /* I */ +); + +void silk_NSQ_del_dec_c( + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + SideInfoIndices *psIndices, /* I/O Quantization Indices */ + const opus_int16 x16[], /* I Input */ + opus_int8 pulses[], /* O Quantized pulse signal */ + const opus_int16 PredCoef_Q12[ 2 * MAX_LPC_ORDER ], /* I Short term prediction coefs */ + const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ + const opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ + const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ + const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ + const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ + const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ + const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ + const opus_int LTP_scale_Q14 /* I LTP state scaling */ +) +{ + opus_int i, k, lag, start_idx, LSF_interpolation_flag, Winner_ind, subfr; + opus_int last_smple_idx, smpl_buf_idx, decisionDelay; + const opus_int16 *A_Q12, *B_Q14, *AR_shp_Q13; + opus_int16 *pxq; + VARDECL( opus_int32, sLTP_Q15 ); + VARDECL( opus_int16, sLTP ); + opus_int32 HarmShapeFIRPacked_Q14; + opus_int offset_Q10; + opus_int32 RDmin_Q10, Gain_Q10; + VARDECL( opus_int32, x_sc_Q10 ); + VARDECL( opus_int32, delayedGain_Q10 ); + VARDECL( NSQ_del_dec_struct, psDelDec ); + NSQ_del_dec_struct *psDD; + SAVE_STACK; + + /* Set unvoiced lag to the previous one, overwrite later for voiced */ + lag = NSQ->lagPrev; + + silk_assert( NSQ->prev_gain_Q16 != 0 ); + + /* Initialize delayed decision states */ + ALLOC( psDelDec, psEncC->nStatesDelayedDecision, NSQ_del_dec_struct ); + silk_memset( psDelDec, 0, psEncC->nStatesDelayedDecision * sizeof( NSQ_del_dec_struct ) ); + for( k = 0; k < psEncC->nStatesDelayedDecision; k++ ) { + psDD = &psDelDec[ k ]; + psDD->Seed = ( k + psIndices->Seed ) & 3; + psDD->SeedInit = psDD->Seed; + psDD->RD_Q10 = 0; + psDD->LF_AR_Q14 = NSQ->sLF_AR_shp_Q14; + psDD->Diff_Q14 = NSQ->sDiff_shp_Q14; + psDD->Shape_Q14[ 0 ] = NSQ->sLTP_shp_Q14[ psEncC->ltp_mem_length - 1 ]; + silk_memcpy( psDD->sLPC_Q14, NSQ->sLPC_Q14, NSQ_LPC_BUF_LENGTH * sizeof( opus_int32 ) ); + silk_memcpy( psDD->sAR2_Q14, NSQ->sAR2_Q14, sizeof( NSQ->sAR2_Q14 ) ); + } + + offset_Q10 = silk_Quantization_Offsets_Q10[ psIndices->signalType >> 1 ][ psIndices->quantOffsetType ]; + smpl_buf_idx = 0; /* index of oldest samples */ + + decisionDelay = silk_min_int( DECISION_DELAY, psEncC->subfr_length ); + + /* For voiced frames limit the decision delay to lower than the pitch lag */ + if( psIndices->signalType == TYPE_VOICED ) { + for( k = 0; k < psEncC->nb_subfr; k++ ) { + decisionDelay = silk_min_int( decisionDelay, pitchL[ k ] - LTP_ORDER / 2 - 1 ); + } + } else { + if( lag > 0 ) { + decisionDelay = silk_min_int( decisionDelay, lag - LTP_ORDER / 2 - 1 ); + } + } + + if( psIndices->NLSFInterpCoef_Q2 == 4 ) { + LSF_interpolation_flag = 0; + } else { + LSF_interpolation_flag = 1; + } + + ALLOC( sLTP_Q15, psEncC->ltp_mem_length + psEncC->frame_length, opus_int32 ); + ALLOC( sLTP, psEncC->ltp_mem_length + psEncC->frame_length, opus_int16 ); + ALLOC( x_sc_Q10, psEncC->subfr_length, opus_int32 ); + ALLOC( delayedGain_Q10, DECISION_DELAY, opus_int32 ); + /* Set up pointers to start of sub frame */ + pxq = &NSQ->xq[ psEncC->ltp_mem_length ]; + NSQ->sLTP_shp_buf_idx = psEncC->ltp_mem_length; + NSQ->sLTP_buf_idx = psEncC->ltp_mem_length; + subfr = 0; + for( k = 0; k < psEncC->nb_subfr; k++ ) { + A_Q12 = &PredCoef_Q12[ ( ( k >> 1 ) | ( 1 - LSF_interpolation_flag ) ) * MAX_LPC_ORDER ]; + B_Q14 = <PCoef_Q14[ k * LTP_ORDER ]; + AR_shp_Q13 = &AR_Q13[ k * MAX_SHAPE_LPC_ORDER ]; + + /* Noise shape parameters */ + silk_assert( HarmShapeGain_Q14[ k ] >= 0 ); + HarmShapeFIRPacked_Q14 = silk_RSHIFT( HarmShapeGain_Q14[ k ], 2 ); + HarmShapeFIRPacked_Q14 |= silk_LSHIFT( (opus_int32)silk_RSHIFT( HarmShapeGain_Q14[ k ], 1 ), 16 ); + + NSQ->rewhite_flag = 0; + if( psIndices->signalType == TYPE_VOICED ) { + /* Voiced */ + lag = pitchL[ k ]; + + /* Re-whitening */ + if( ( k & ( 3 - silk_LSHIFT( LSF_interpolation_flag, 1 ) ) ) == 0 ) { + if( k == 2 ) { + /* RESET DELAYED DECISIONS */ + /* Find winner */ + RDmin_Q10 = psDelDec[ 0 ].RD_Q10; + Winner_ind = 0; + for( i = 1; i < psEncC->nStatesDelayedDecision; i++ ) { + if( psDelDec[ i ].RD_Q10 < RDmin_Q10 ) { + RDmin_Q10 = psDelDec[ i ].RD_Q10; + Winner_ind = i; + } + } + for( i = 0; i < psEncC->nStatesDelayedDecision; i++ ) { + if( i != Winner_ind ) { + psDelDec[ i ].RD_Q10 += ( silk_int32_MAX >> 4 ); + silk_assert( psDelDec[ i ].RD_Q10 >= 0 ); + } + } + + /* Copy final part of signals from winner state to output and long-term filter states */ + psDD = &psDelDec[ Winner_ind ]; + last_smple_idx = smpl_buf_idx + decisionDelay; + for( i = 0; i < decisionDelay; i++ ) { + last_smple_idx = ( last_smple_idx - 1 ) % DECISION_DELAY; + if( last_smple_idx < 0 ) last_smple_idx += DECISION_DELAY; + pulses[ i - decisionDelay ] = (opus_int8)silk_RSHIFT_ROUND( psDD->Q_Q10[ last_smple_idx ], 10 ); + pxq[ i - decisionDelay ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( + silk_SMULWW( psDD->Xq_Q14[ last_smple_idx ], Gains_Q16[ 1 ] ), 14 ) ); + NSQ->sLTP_shp_Q14[ NSQ->sLTP_shp_buf_idx - decisionDelay + i ] = psDD->Shape_Q14[ last_smple_idx ]; + } + + subfr = 0; + } + + /* Rewhiten with new A coefs */ + start_idx = psEncC->ltp_mem_length - lag - psEncC->predictLPCOrder - LTP_ORDER / 2; + celt_assert( start_idx > 0 ); + + silk_LPC_analysis_filter( &sLTP[ start_idx ], &NSQ->xq[ start_idx + k * psEncC->subfr_length ], + A_Q12, psEncC->ltp_mem_length - start_idx, psEncC->predictLPCOrder, psEncC->arch ); + + NSQ->sLTP_buf_idx = psEncC->ltp_mem_length; + NSQ->rewhite_flag = 1; + } + } + + silk_nsq_del_dec_scale_states( psEncC, NSQ, psDelDec, x16, x_sc_Q10, sLTP, sLTP_Q15, k, + psEncC->nStatesDelayedDecision, LTP_scale_Q14, Gains_Q16, pitchL, psIndices->signalType, decisionDelay ); + + silk_noise_shape_quantizer_del_dec( NSQ, psDelDec, psIndices->signalType, x_sc_Q10, pulses, pxq, sLTP_Q15, + delayedGain_Q10, A_Q12, B_Q14, AR_shp_Q13, lag, HarmShapeFIRPacked_Q14, Tilt_Q14[ k ], LF_shp_Q14[ k ], + Gains_Q16[ k ], Lambda_Q10, offset_Q10, psEncC->subfr_length, subfr++, psEncC->shapingLPCOrder, + psEncC->predictLPCOrder, psEncC->warping_Q16, psEncC->nStatesDelayedDecision, &smpl_buf_idx, decisionDelay, psEncC->arch ); + + x16 += psEncC->subfr_length; + pulses += psEncC->subfr_length; + pxq += psEncC->subfr_length; + } + + /* Find winner */ + RDmin_Q10 = psDelDec[ 0 ].RD_Q10; + Winner_ind = 0; + for( k = 1; k < psEncC->nStatesDelayedDecision; k++ ) { + if( psDelDec[ k ].RD_Q10 < RDmin_Q10 ) { + RDmin_Q10 = psDelDec[ k ].RD_Q10; + Winner_ind = k; + } + } + + /* Copy final part of signals from winner state to output and long-term filter states */ + psDD = &psDelDec[ Winner_ind ]; + psIndices->Seed = psDD->SeedInit; + last_smple_idx = smpl_buf_idx + decisionDelay; + Gain_Q10 = silk_RSHIFT32( Gains_Q16[ psEncC->nb_subfr - 1 ], 6 ); + for( i = 0; i < decisionDelay; i++ ) { + last_smple_idx = ( last_smple_idx - 1 ) % DECISION_DELAY; + if( last_smple_idx < 0 ) last_smple_idx += DECISION_DELAY; + + pulses[ i - decisionDelay ] = (opus_int8)silk_RSHIFT_ROUND( psDD->Q_Q10[ last_smple_idx ], 10 ); + pxq[ i - decisionDelay ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( + silk_SMULWW( psDD->Xq_Q14[ last_smple_idx ], Gain_Q10 ), 8 ) ); + NSQ->sLTP_shp_Q14[ NSQ->sLTP_shp_buf_idx - decisionDelay + i ] = psDD->Shape_Q14[ last_smple_idx ]; + } + silk_memcpy( NSQ->sLPC_Q14, &psDD->sLPC_Q14[ psEncC->subfr_length ], NSQ_LPC_BUF_LENGTH * sizeof( opus_int32 ) ); + silk_memcpy( NSQ->sAR2_Q14, psDD->sAR2_Q14, sizeof( psDD->sAR2_Q14 ) ); + + /* Update states */ + NSQ->sLF_AR_shp_Q14 = psDD->LF_AR_Q14; + NSQ->sDiff_shp_Q14 = psDD->Diff_Q14; + NSQ->lagPrev = pitchL[ psEncC->nb_subfr - 1 ]; + + /* Save quantized speech signal */ + silk_memmove( NSQ->xq, &NSQ->xq[ psEncC->frame_length ], psEncC->ltp_mem_length * sizeof( opus_int16 ) ); + silk_memmove( NSQ->sLTP_shp_Q14, &NSQ->sLTP_shp_Q14[ psEncC->frame_length ], psEncC->ltp_mem_length * sizeof( opus_int32 ) ); + RESTORE_STACK; +} + +/******************************************/ +/* Noise shape quantizer for one subframe */ +/******************************************/ +#ifndef OVERRIDE_silk_noise_shape_quantizer_del_dec +static OPUS_INLINE void silk_noise_shape_quantizer_del_dec( + silk_nsq_state *NSQ, /* I/O NSQ state */ + NSQ_del_dec_struct psDelDec[], /* I/O Delayed decision states */ + opus_int signalType, /* I Signal type */ + const opus_int32 x_Q10[], /* I */ + opus_int8 pulses[], /* O */ + opus_int16 xq[], /* O */ + opus_int32 sLTP_Q15[], /* I/O LTP filter state */ + opus_int32 delayedGain_Q10[], /* I/O Gain delay buffer */ + const opus_int16 a_Q12[], /* I Short term prediction coefs */ + const opus_int16 b_Q14[], /* I Long term prediction coefs */ + const opus_int16 AR_shp_Q13[], /* I Noise shaping coefs */ + opus_int lag, /* I Pitch lag */ + opus_int32 HarmShapeFIRPacked_Q14, /* I */ + opus_int Tilt_Q14, /* I Spectral tilt */ + opus_int32 LF_shp_Q14, /* I */ + opus_int32 Gain_Q16, /* I */ + opus_int Lambda_Q10, /* I */ + opus_int offset_Q10, /* I */ + opus_int length, /* I Input length */ + opus_int subfr, /* I Subframe number */ + opus_int shapingLPCOrder, /* I Shaping LPC filter order */ + opus_int predictLPCOrder, /* I Prediction filter order */ + opus_int warping_Q16, /* I */ + opus_int nStatesDelayedDecision, /* I Number of states in decision tree */ + opus_int *smpl_buf_idx, /* I/O Index to newest samples in buffers */ + opus_int decisionDelay, /* I */ + int arch /* I */ +) +{ + opus_int i, j, k, Winner_ind, RDmin_ind, RDmax_ind, last_smple_idx; + opus_int32 Winner_rand_state; + opus_int32 LTP_pred_Q14, LPC_pred_Q14, n_AR_Q14, n_LTP_Q14; + opus_int32 n_LF_Q14, r_Q10, rr_Q10, rd1_Q10, rd2_Q10, RDmin_Q10, RDmax_Q10; + opus_int32 q1_Q0, q1_Q10, q2_Q10, exc_Q14, LPC_exc_Q14, xq_Q14, Gain_Q10; + opus_int32 tmp1, tmp2, sLF_AR_shp_Q14; + opus_int32 *pred_lag_ptr, *shp_lag_ptr, *psLPC_Q14; +#ifdef silk_short_prediction_create_arch_coef + opus_int32 a_Q12_arch[MAX_LPC_ORDER]; +#endif + + VARDECL( NSQ_sample_pair, psSampleState ); + NSQ_del_dec_struct *psDD; + NSQ_sample_struct *psSS; + SAVE_STACK; + + celt_assert( nStatesDelayedDecision > 0 ); + ALLOC( psSampleState, nStatesDelayedDecision, NSQ_sample_pair ); + + shp_lag_ptr = &NSQ->sLTP_shp_Q14[ NSQ->sLTP_shp_buf_idx - lag + HARM_SHAPE_FIR_TAPS / 2 ]; + pred_lag_ptr = &sLTP_Q15[ NSQ->sLTP_buf_idx - lag + LTP_ORDER / 2 ]; + Gain_Q10 = silk_RSHIFT( Gain_Q16, 6 ); + +#ifdef silk_short_prediction_create_arch_coef + silk_short_prediction_create_arch_coef(a_Q12_arch, a_Q12, predictLPCOrder); +#endif + + for( i = 0; i < length; i++ ) { + /* Perform common calculations used in all states */ + + /* Long-term prediction */ + if( signalType == TYPE_VOICED ) { + /* Unrolled loop */ + /* Avoids introducing a bias because silk_SMLAWB() always rounds to -inf */ + LTP_pred_Q14 = 2; + LTP_pred_Q14 = silk_SMLAWB( LTP_pred_Q14, pred_lag_ptr[ 0 ], b_Q14[ 0 ] ); + LTP_pred_Q14 = silk_SMLAWB( LTP_pred_Q14, pred_lag_ptr[ -1 ], b_Q14[ 1 ] ); + LTP_pred_Q14 = silk_SMLAWB( LTP_pred_Q14, pred_lag_ptr[ -2 ], b_Q14[ 2 ] ); + LTP_pred_Q14 = silk_SMLAWB( LTP_pred_Q14, pred_lag_ptr[ -3 ], b_Q14[ 3 ] ); + LTP_pred_Q14 = silk_SMLAWB( LTP_pred_Q14, pred_lag_ptr[ -4 ], b_Q14[ 4 ] ); + LTP_pred_Q14 = silk_LSHIFT( LTP_pred_Q14, 1 ); /* Q13 -> Q14 */ + pred_lag_ptr++; + } else { + LTP_pred_Q14 = 0; + } + + /* Long-term shaping */ + if( lag > 0 ) { + /* Symmetric, packed FIR coefficients */ + n_LTP_Q14 = silk_SMULWB( silk_ADD_SAT32( shp_lag_ptr[ 0 ], shp_lag_ptr[ -2 ] ), HarmShapeFIRPacked_Q14 ); + n_LTP_Q14 = silk_SMLAWT( n_LTP_Q14, shp_lag_ptr[ -1 ], HarmShapeFIRPacked_Q14 ); + n_LTP_Q14 = silk_SUB_LSHIFT32( LTP_pred_Q14, n_LTP_Q14, 2 ); /* Q12 -> Q14 */ + shp_lag_ptr++; + } else { + n_LTP_Q14 = 0; + } + + for( k = 0; k < nStatesDelayedDecision; k++ ) { + /* Delayed decision state */ + psDD = &psDelDec[ k ]; + + /* Sample state */ + psSS = psSampleState[ k ]; + + /* Generate dither */ + psDD->Seed = silk_RAND( psDD->Seed ); + + /* Pointer used in short term prediction and shaping */ + psLPC_Q14 = &psDD->sLPC_Q14[ NSQ_LPC_BUF_LENGTH - 1 + i ]; + /* Short-term prediction */ + LPC_pred_Q14 = silk_noise_shape_quantizer_short_prediction(psLPC_Q14, a_Q12, a_Q12_arch, predictLPCOrder, arch); + LPC_pred_Q14 = silk_LSHIFT( LPC_pred_Q14, 4 ); /* Q10 -> Q14 */ + + /* Noise shape feedback */ + celt_assert( ( shapingLPCOrder & 1 ) == 0 ); /* check that order is even */ + /* Output of lowpass section */ + tmp2 = silk_SMLAWB( psDD->Diff_Q14, psDD->sAR2_Q14[ 0 ], warping_Q16 ); + /* Output of allpass section */ + tmp1 = silk_SMLAWB( psDD->sAR2_Q14[ 0 ], psDD->sAR2_Q14[ 1 ] - tmp2, warping_Q16 ); + psDD->sAR2_Q14[ 0 ] = tmp2; + n_AR_Q14 = silk_RSHIFT( shapingLPCOrder, 1 ); + n_AR_Q14 = silk_SMLAWB( n_AR_Q14, tmp2, AR_shp_Q13[ 0 ] ); + /* Loop over allpass sections */ + for( j = 2; j < shapingLPCOrder; j += 2 ) { + /* Output of allpass section */ + tmp2 = silk_SMLAWB( psDD->sAR2_Q14[ j - 1 ], psDD->sAR2_Q14[ j + 0 ] - tmp1, warping_Q16 ); + psDD->sAR2_Q14[ j - 1 ] = tmp1; + n_AR_Q14 = silk_SMLAWB( n_AR_Q14, tmp1, AR_shp_Q13[ j - 1 ] ); + /* Output of allpass section */ + tmp1 = silk_SMLAWB( psDD->sAR2_Q14[ j + 0 ], psDD->sAR2_Q14[ j + 1 ] - tmp2, warping_Q16 ); + psDD->sAR2_Q14[ j + 0 ] = tmp2; + n_AR_Q14 = silk_SMLAWB( n_AR_Q14, tmp2, AR_shp_Q13[ j ] ); + } + psDD->sAR2_Q14[ shapingLPCOrder - 1 ] = tmp1; + n_AR_Q14 = silk_SMLAWB( n_AR_Q14, tmp1, AR_shp_Q13[ shapingLPCOrder - 1 ] ); + + n_AR_Q14 = silk_LSHIFT( n_AR_Q14, 1 ); /* Q11 -> Q12 */ + n_AR_Q14 = silk_SMLAWB( n_AR_Q14, psDD->LF_AR_Q14, Tilt_Q14 ); /* Q12 */ + n_AR_Q14 = silk_LSHIFT( n_AR_Q14, 2 ); /* Q12 -> Q14 */ + + n_LF_Q14 = silk_SMULWB( psDD->Shape_Q14[ *smpl_buf_idx ], LF_shp_Q14 ); /* Q12 */ + n_LF_Q14 = silk_SMLAWT( n_LF_Q14, psDD->LF_AR_Q14, LF_shp_Q14 ); /* Q12 */ + n_LF_Q14 = silk_LSHIFT( n_LF_Q14, 2 ); /* Q12 -> Q14 */ + + /* Input minus prediction plus noise feedback */ + /* r = x[ i ] - LTP_pred - LPC_pred + n_AR + n_Tilt + n_LF + n_LTP */ + tmp1 = silk_ADD_SAT32( n_AR_Q14, n_LF_Q14 ); /* Q14 */ + tmp2 = silk_ADD32( n_LTP_Q14, LPC_pred_Q14 ); /* Q13 */ + tmp1 = silk_SUB_SAT32( tmp2, tmp1 ); /* Q13 */ + tmp1 = silk_RSHIFT_ROUND( tmp1, 4 ); /* Q10 */ + + r_Q10 = silk_SUB32( x_Q10[ i ], tmp1 ); /* residual error Q10 */ + + /* Flip sign depending on dither */ + if ( psDD->Seed < 0 ) { + r_Q10 = -r_Q10; + } + r_Q10 = silk_LIMIT_32( r_Q10, -(31 << 10), 30 << 10 ); + + /* Find two quantization level candidates and measure their rate-distortion */ + q1_Q10 = silk_SUB32( r_Q10, offset_Q10 ); + q1_Q0 = silk_RSHIFT( q1_Q10, 10 ); + if (Lambda_Q10 > 2048) { + /* For aggressive RDO, the bias becomes more than one pulse. */ + int rdo_offset = Lambda_Q10/2 - 512; + if (q1_Q10 > rdo_offset) { + q1_Q0 = silk_RSHIFT( q1_Q10 - rdo_offset, 10 ); + } else if (q1_Q10 < -rdo_offset) { + q1_Q0 = silk_RSHIFT( q1_Q10 + rdo_offset, 10 ); + } else if (q1_Q10 < 0) { + q1_Q0 = -1; + } else { + q1_Q0 = 0; + } + } + if( q1_Q0 > 0 ) { + q1_Q10 = silk_SUB32( silk_LSHIFT( q1_Q0, 10 ), QUANT_LEVEL_ADJUST_Q10 ); + q1_Q10 = silk_ADD32( q1_Q10, offset_Q10 ); + q2_Q10 = silk_ADD32( q1_Q10, 1024 ); + rd1_Q10 = silk_SMULBB( q1_Q10, Lambda_Q10 ); + rd2_Q10 = silk_SMULBB( q2_Q10, Lambda_Q10 ); + } else if( q1_Q0 == 0 ) { + q1_Q10 = offset_Q10; + q2_Q10 = silk_ADD32( q1_Q10, 1024 - QUANT_LEVEL_ADJUST_Q10 ); + rd1_Q10 = silk_SMULBB( q1_Q10, Lambda_Q10 ); + rd2_Q10 = silk_SMULBB( q2_Q10, Lambda_Q10 ); + } else if( q1_Q0 == -1 ) { + q2_Q10 = offset_Q10; + q1_Q10 = silk_SUB32( q2_Q10, 1024 - QUANT_LEVEL_ADJUST_Q10 ); + rd1_Q10 = silk_SMULBB( -q1_Q10, Lambda_Q10 ); + rd2_Q10 = silk_SMULBB( q2_Q10, Lambda_Q10 ); + } else { /* q1_Q0 < -1 */ + q1_Q10 = silk_ADD32( silk_LSHIFT( q1_Q0, 10 ), QUANT_LEVEL_ADJUST_Q10 ); + q1_Q10 = silk_ADD32( q1_Q10, offset_Q10 ); + q2_Q10 = silk_ADD32( q1_Q10, 1024 ); + rd1_Q10 = silk_SMULBB( -q1_Q10, Lambda_Q10 ); + rd2_Q10 = silk_SMULBB( -q2_Q10, Lambda_Q10 ); + } + rr_Q10 = silk_SUB32( r_Q10, q1_Q10 ); + rd1_Q10 = silk_RSHIFT( silk_SMLABB( rd1_Q10, rr_Q10, rr_Q10 ), 10 ); + rr_Q10 = silk_SUB32( r_Q10, q2_Q10 ); + rd2_Q10 = silk_RSHIFT( silk_SMLABB( rd2_Q10, rr_Q10, rr_Q10 ), 10 ); + + if( rd1_Q10 < rd2_Q10 ) { + psSS[ 0 ].RD_Q10 = silk_ADD32( psDD->RD_Q10, rd1_Q10 ); + psSS[ 1 ].RD_Q10 = silk_ADD32( psDD->RD_Q10, rd2_Q10 ); + psSS[ 0 ].Q_Q10 = q1_Q10; + psSS[ 1 ].Q_Q10 = q2_Q10; + } else { + psSS[ 0 ].RD_Q10 = silk_ADD32( psDD->RD_Q10, rd2_Q10 ); + psSS[ 1 ].RD_Q10 = silk_ADD32( psDD->RD_Q10, rd1_Q10 ); + psSS[ 0 ].Q_Q10 = q2_Q10; + psSS[ 1 ].Q_Q10 = q1_Q10; + } + + /* Update states for best quantization */ + + /* Quantized excitation */ + exc_Q14 = silk_LSHIFT32( psSS[ 0 ].Q_Q10, 4 ); + if ( psDD->Seed < 0 ) { + exc_Q14 = -exc_Q14; + } + + /* Add predictions */ + LPC_exc_Q14 = silk_ADD32( exc_Q14, LTP_pred_Q14 ); + xq_Q14 = silk_ADD32( LPC_exc_Q14, LPC_pred_Q14 ); + + /* Update states */ + psSS[ 0 ].Diff_Q14 = silk_SUB_LSHIFT32( xq_Q14, x_Q10[ i ], 4 ); + sLF_AR_shp_Q14 = silk_SUB32( psSS[ 0 ].Diff_Q14, n_AR_Q14 ); + psSS[ 0 ].sLTP_shp_Q14 = silk_SUB_SAT32( sLF_AR_shp_Q14, n_LF_Q14 ); + psSS[ 0 ].LF_AR_Q14 = sLF_AR_shp_Q14; + psSS[ 0 ].LPC_exc_Q14 = LPC_exc_Q14; + psSS[ 0 ].xq_Q14 = xq_Q14; + + /* Update states for second best quantization */ + + /* Quantized excitation */ + exc_Q14 = silk_LSHIFT32( psSS[ 1 ].Q_Q10, 4 ); + if ( psDD->Seed < 0 ) { + exc_Q14 = -exc_Q14; + } + + /* Add predictions */ + LPC_exc_Q14 = silk_ADD32( exc_Q14, LTP_pred_Q14 ); + xq_Q14 = silk_ADD32( LPC_exc_Q14, LPC_pred_Q14 ); + + /* Update states */ + psSS[ 1 ].Diff_Q14 = silk_SUB_LSHIFT32( xq_Q14, x_Q10[ i ], 4 ); + sLF_AR_shp_Q14 = silk_SUB32( psSS[ 1 ].Diff_Q14, n_AR_Q14 ); + psSS[ 1 ].sLTP_shp_Q14 = silk_SUB_SAT32( sLF_AR_shp_Q14, n_LF_Q14 ); + psSS[ 1 ].LF_AR_Q14 = sLF_AR_shp_Q14; + psSS[ 1 ].LPC_exc_Q14 = LPC_exc_Q14; + psSS[ 1 ].xq_Q14 = xq_Q14; + } + + *smpl_buf_idx = ( *smpl_buf_idx - 1 ) % DECISION_DELAY; + if( *smpl_buf_idx < 0 ) *smpl_buf_idx += DECISION_DELAY; + last_smple_idx = ( *smpl_buf_idx + decisionDelay ) % DECISION_DELAY; + + /* Find winner */ + RDmin_Q10 = psSampleState[ 0 ][ 0 ].RD_Q10; + Winner_ind = 0; + for( k = 1; k < nStatesDelayedDecision; k++ ) { + if( psSampleState[ k ][ 0 ].RD_Q10 < RDmin_Q10 ) { + RDmin_Q10 = psSampleState[ k ][ 0 ].RD_Q10; + Winner_ind = k; + } + } + + /* Increase RD values of expired states */ + Winner_rand_state = psDelDec[ Winner_ind ].RandState[ last_smple_idx ]; + for( k = 0; k < nStatesDelayedDecision; k++ ) { + if( psDelDec[ k ].RandState[ last_smple_idx ] != Winner_rand_state ) { + psSampleState[ k ][ 0 ].RD_Q10 = silk_ADD32( psSampleState[ k ][ 0 ].RD_Q10, silk_int32_MAX >> 4 ); + psSampleState[ k ][ 1 ].RD_Q10 = silk_ADD32( psSampleState[ k ][ 1 ].RD_Q10, silk_int32_MAX >> 4 ); + silk_assert( psSampleState[ k ][ 0 ].RD_Q10 >= 0 ); + } + } + + /* Find worst in first set and best in second set */ + RDmax_Q10 = psSampleState[ 0 ][ 0 ].RD_Q10; + RDmin_Q10 = psSampleState[ 0 ][ 1 ].RD_Q10; + RDmax_ind = 0; + RDmin_ind = 0; + for( k = 1; k < nStatesDelayedDecision; k++ ) { + /* find worst in first set */ + if( psSampleState[ k ][ 0 ].RD_Q10 > RDmax_Q10 ) { + RDmax_Q10 = psSampleState[ k ][ 0 ].RD_Q10; + RDmax_ind = k; + } + /* find best in second set */ + if( psSampleState[ k ][ 1 ].RD_Q10 < RDmin_Q10 ) { + RDmin_Q10 = psSampleState[ k ][ 1 ].RD_Q10; + RDmin_ind = k; + } + } + + /* Replace a state if best from second set outperforms worst in first set */ + if( RDmin_Q10 < RDmax_Q10 ) { + silk_memcpy( ( (opus_int32 *)&psDelDec[ RDmax_ind ] ) + i, + ( (opus_int32 *)&psDelDec[ RDmin_ind ] ) + i, sizeof( NSQ_del_dec_struct ) - i * sizeof( opus_int32) ); + silk_memcpy( &psSampleState[ RDmax_ind ][ 0 ], &psSampleState[ RDmin_ind ][ 1 ], sizeof( NSQ_sample_struct ) ); + } + + /* Write samples from winner to output and long-term filter states */ + psDD = &psDelDec[ Winner_ind ]; + if( subfr > 0 || i >= decisionDelay ) { + pulses[ i - decisionDelay ] = (opus_int8)silk_RSHIFT_ROUND( psDD->Q_Q10[ last_smple_idx ], 10 ); + xq[ i - decisionDelay ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( + silk_SMULWW( psDD->Xq_Q14[ last_smple_idx ], delayedGain_Q10[ last_smple_idx ] ), 8 ) ); + NSQ->sLTP_shp_Q14[ NSQ->sLTP_shp_buf_idx - decisionDelay ] = psDD->Shape_Q14[ last_smple_idx ]; + sLTP_Q15[ NSQ->sLTP_buf_idx - decisionDelay ] = psDD->Pred_Q15[ last_smple_idx ]; + } + NSQ->sLTP_shp_buf_idx++; + NSQ->sLTP_buf_idx++; + + /* Update states */ + for( k = 0; k < nStatesDelayedDecision; k++ ) { + psDD = &psDelDec[ k ]; + psSS = &psSampleState[ k ][ 0 ]; + psDD->LF_AR_Q14 = psSS->LF_AR_Q14; + psDD->Diff_Q14 = psSS->Diff_Q14; + psDD->sLPC_Q14[ NSQ_LPC_BUF_LENGTH + i ] = psSS->xq_Q14; + psDD->Xq_Q14[ *smpl_buf_idx ] = psSS->xq_Q14; + psDD->Q_Q10[ *smpl_buf_idx ] = psSS->Q_Q10; + psDD->Pred_Q15[ *smpl_buf_idx ] = silk_LSHIFT32( psSS->LPC_exc_Q14, 1 ); + psDD->Shape_Q14[ *smpl_buf_idx ] = psSS->sLTP_shp_Q14; + psDD->Seed = silk_ADD32_ovflw( psDD->Seed, silk_RSHIFT_ROUND( psSS->Q_Q10, 10 ) ); + psDD->RandState[ *smpl_buf_idx ] = psDD->Seed; + psDD->RD_Q10 = psSS->RD_Q10; + } + delayedGain_Q10[ *smpl_buf_idx ] = Gain_Q10; + } + /* Update LPC states */ + for( k = 0; k < nStatesDelayedDecision; k++ ) { + psDD = &psDelDec[ k ]; + silk_memcpy( psDD->sLPC_Q14, &psDD->sLPC_Q14[ length ], NSQ_LPC_BUF_LENGTH * sizeof( opus_int32 ) ); + } + RESTORE_STACK; +} +#endif /* OVERRIDE_silk_noise_shape_quantizer_del_dec */ + +static OPUS_INLINE void silk_nsq_del_dec_scale_states( + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + NSQ_del_dec_struct psDelDec[], /* I/O Delayed decision states */ + const opus_int16 x16[], /* I Input */ + opus_int32 x_sc_Q10[], /* O Input scaled with 1/Gain in Q10 */ + const opus_int16 sLTP[], /* I Re-whitened LTP state in Q0 */ + opus_int32 sLTP_Q15[], /* O LTP state matching scaled input */ + opus_int subfr, /* I Subframe number */ + opus_int nStatesDelayedDecision, /* I Number of del dec states */ + const opus_int LTP_scale_Q14, /* I LTP state scaling */ + const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lag */ + const opus_int signal_type, /* I Signal type */ + const opus_int decisionDelay /* I Decision delay */ +) +{ + opus_int i, k, lag; + opus_int32 gain_adj_Q16, inv_gain_Q31, inv_gain_Q26; + NSQ_del_dec_struct *psDD; + + lag = pitchL[ subfr ]; + inv_gain_Q31 = silk_INVERSE32_varQ( silk_max( Gains_Q16[ subfr ], 1 ), 47 ); + silk_assert( inv_gain_Q31 != 0 ); + + /* Scale input */ + inv_gain_Q26 = silk_RSHIFT_ROUND( inv_gain_Q31, 5 ); + for( i = 0; i < psEncC->subfr_length; i++ ) { + x_sc_Q10[ i ] = silk_SMULWW( x16[ i ], inv_gain_Q26 ); + } + + /* After rewhitening the LTP state is un-scaled, so scale with inv_gain_Q16 */ + if( NSQ->rewhite_flag ) { + if( subfr == 0 ) { + /* Do LTP downscaling */ + inv_gain_Q31 = silk_LSHIFT( silk_SMULWB( inv_gain_Q31, LTP_scale_Q14 ), 2 ); + } + for( i = NSQ->sLTP_buf_idx - lag - LTP_ORDER / 2; i < NSQ->sLTP_buf_idx; i++ ) { + silk_assert( i < MAX_FRAME_LENGTH ); + sLTP_Q15[ i ] = silk_SMULWB( inv_gain_Q31, sLTP[ i ] ); + } + } + + /* Adjust for changing gain */ + if( Gains_Q16[ subfr ] != NSQ->prev_gain_Q16 ) { + gain_adj_Q16 = silk_DIV32_varQ( NSQ->prev_gain_Q16, Gains_Q16[ subfr ], 16 ); + + /* Scale long-term shaping state */ + for( i = NSQ->sLTP_shp_buf_idx - psEncC->ltp_mem_length; i < NSQ->sLTP_shp_buf_idx; i++ ) { + NSQ->sLTP_shp_Q14[ i ] = silk_SMULWW( gain_adj_Q16, NSQ->sLTP_shp_Q14[ i ] ); + } + + /* Scale long-term prediction state */ + if( signal_type == TYPE_VOICED && NSQ->rewhite_flag == 0 ) { + for( i = NSQ->sLTP_buf_idx - lag - LTP_ORDER / 2; i < NSQ->sLTP_buf_idx - decisionDelay; i++ ) { + sLTP_Q15[ i ] = silk_SMULWW( gain_adj_Q16, sLTP_Q15[ i ] ); + } + } + + for( k = 0; k < nStatesDelayedDecision; k++ ) { + psDD = &psDelDec[ k ]; + + /* Scale scalar states */ + psDD->LF_AR_Q14 = silk_SMULWW( gain_adj_Q16, psDD->LF_AR_Q14 ); + psDD->Diff_Q14 = silk_SMULWW( gain_adj_Q16, psDD->Diff_Q14 ); + + /* Scale short-term prediction and shaping states */ + for( i = 0; i < NSQ_LPC_BUF_LENGTH; i++ ) { + psDD->sLPC_Q14[ i ] = silk_SMULWW( gain_adj_Q16, psDD->sLPC_Q14[ i ] ); + } + for( i = 0; i < MAX_SHAPE_LPC_ORDER; i++ ) { + psDD->sAR2_Q14[ i ] = silk_SMULWW( gain_adj_Q16, psDD->sAR2_Q14[ i ] ); + } + for( i = 0; i < DECISION_DELAY; i++ ) { + psDD->Pred_Q15[ i ] = silk_SMULWW( gain_adj_Q16, psDD->Pred_Q15[ i ] ); + psDD->Shape_Q14[ i ] = silk_SMULWW( gain_adj_Q16, psDD->Shape_Q14[ i ] ); + } + } + + /* Save inverse gain */ + NSQ->prev_gain_Q16 = Gains_Q16[ subfr ]; + } +} diff --git a/media/libopus/silk/PLC.c b/media/libopus/silk/PLC.c new file mode 100644 index 0000000000..4667440db2 --- /dev/null +++ b/media/libopus/silk/PLC.c @@ -0,0 +1,446 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" +#include "stack_alloc.h" +#include "PLC.h" + +#define NB_ATT 2 +static const opus_int16 HARM_ATT_Q15[NB_ATT] = { 32440, 31130 }; /* 0.99, 0.95 */ +static const opus_int16 PLC_RAND_ATTENUATE_V_Q15[NB_ATT] = { 31130, 26214 }; /* 0.95, 0.8 */ +static const opus_int16 PLC_RAND_ATTENUATE_UV_Q15[NB_ATT] = { 32440, 29491 }; /* 0.99, 0.9 */ + +static OPUS_INLINE void silk_PLC_update( + silk_decoder_state *psDec, /* I/O Decoder state */ + silk_decoder_control *psDecCtrl /* I/O Decoder control */ +); + +static OPUS_INLINE void silk_PLC_conceal( + silk_decoder_state *psDec, /* I/O Decoder state */ + silk_decoder_control *psDecCtrl, /* I/O Decoder control */ + opus_int16 frame[], /* O LPC residual signal */ + int arch /* I Run-time architecture */ +); + + +void silk_PLC_Reset( + silk_decoder_state *psDec /* I/O Decoder state */ +) +{ + psDec->sPLC.pitchL_Q8 = silk_LSHIFT( psDec->frame_length, 8 - 1 ); + psDec->sPLC.prevGain_Q16[ 0 ] = SILK_FIX_CONST( 1, 16 ); + psDec->sPLC.prevGain_Q16[ 1 ] = SILK_FIX_CONST( 1, 16 ); + psDec->sPLC.subfr_length = 20; + psDec->sPLC.nb_subfr = 2; +} + +void silk_PLC( + silk_decoder_state *psDec, /* I/O Decoder state */ + silk_decoder_control *psDecCtrl, /* I/O Decoder control */ + opus_int16 frame[], /* I/O signal */ + opus_int lost, /* I Loss flag */ + int arch /* I Run-time architecture */ +) +{ + /* PLC control function */ + if( psDec->fs_kHz != psDec->sPLC.fs_kHz ) { + silk_PLC_Reset( psDec ); + psDec->sPLC.fs_kHz = psDec->fs_kHz; + } + + if( lost ) { + /****************************/ + /* Generate Signal */ + /****************************/ + silk_PLC_conceal( psDec, psDecCtrl, frame, arch ); + + psDec->lossCnt++; + } else { + /****************************/ + /* Update state */ + /****************************/ + silk_PLC_update( psDec, psDecCtrl ); + } +} + +/**************************************************/ +/* Update state of PLC */ +/**************************************************/ +static OPUS_INLINE void silk_PLC_update( + silk_decoder_state *psDec, /* I/O Decoder state */ + silk_decoder_control *psDecCtrl /* I/O Decoder control */ +) +{ + opus_int32 LTP_Gain_Q14, temp_LTP_Gain_Q14; + opus_int i, j; + silk_PLC_struct *psPLC; + + psPLC = &psDec->sPLC; + + /* Update parameters used in case of packet loss */ + psDec->prevSignalType = psDec->indices.signalType; + LTP_Gain_Q14 = 0; + if( psDec->indices.signalType == TYPE_VOICED ) { + /* Find the parameters for the last subframe which contains a pitch pulse */ + for( j = 0; j * psDec->subfr_length < psDecCtrl->pitchL[ psDec->nb_subfr - 1 ]; j++ ) { + if( j == psDec->nb_subfr ) { + break; + } + temp_LTP_Gain_Q14 = 0; + for( i = 0; i < LTP_ORDER; i++ ) { + temp_LTP_Gain_Q14 += psDecCtrl->LTPCoef_Q14[ ( psDec->nb_subfr - 1 - j ) * LTP_ORDER + i ]; + } + if( temp_LTP_Gain_Q14 > LTP_Gain_Q14 ) { + LTP_Gain_Q14 = temp_LTP_Gain_Q14; + silk_memcpy( psPLC->LTPCoef_Q14, + &psDecCtrl->LTPCoef_Q14[ silk_SMULBB( psDec->nb_subfr - 1 - j, LTP_ORDER ) ], + LTP_ORDER * sizeof( opus_int16 ) ); + + psPLC->pitchL_Q8 = silk_LSHIFT( psDecCtrl->pitchL[ psDec->nb_subfr - 1 - j ], 8 ); + } + } + + silk_memset( psPLC->LTPCoef_Q14, 0, LTP_ORDER * sizeof( opus_int16 ) ); + psPLC->LTPCoef_Q14[ LTP_ORDER / 2 ] = LTP_Gain_Q14; + + /* Limit LT coefs */ + if( LTP_Gain_Q14 < V_PITCH_GAIN_START_MIN_Q14 ) { + opus_int scale_Q10; + opus_int32 tmp; + + tmp = silk_LSHIFT( V_PITCH_GAIN_START_MIN_Q14, 10 ); + scale_Q10 = silk_DIV32( tmp, silk_max( LTP_Gain_Q14, 1 ) ); + for( i = 0; i < LTP_ORDER; i++ ) { + psPLC->LTPCoef_Q14[ i ] = silk_RSHIFT( silk_SMULBB( psPLC->LTPCoef_Q14[ i ], scale_Q10 ), 10 ); + } + } else if( LTP_Gain_Q14 > V_PITCH_GAIN_START_MAX_Q14 ) { + opus_int scale_Q14; + opus_int32 tmp; + + tmp = silk_LSHIFT( V_PITCH_GAIN_START_MAX_Q14, 14 ); + scale_Q14 = silk_DIV32( tmp, silk_max( LTP_Gain_Q14, 1 ) ); + for( i = 0; i < LTP_ORDER; i++ ) { + psPLC->LTPCoef_Q14[ i ] = silk_RSHIFT( silk_SMULBB( psPLC->LTPCoef_Q14[ i ], scale_Q14 ), 14 ); + } + } + } else { + psPLC->pitchL_Q8 = silk_LSHIFT( silk_SMULBB( psDec->fs_kHz, 18 ), 8 ); + silk_memset( psPLC->LTPCoef_Q14, 0, LTP_ORDER * sizeof( opus_int16 )); + } + + /* Save LPC coeficients */ + silk_memcpy( psPLC->prevLPC_Q12, psDecCtrl->PredCoef_Q12[ 1 ], psDec->LPC_order * sizeof( opus_int16 ) ); + psPLC->prevLTP_scale_Q14 = psDecCtrl->LTP_scale_Q14; + + /* Save last two gains */ + silk_memcpy( psPLC->prevGain_Q16, &psDecCtrl->Gains_Q16[ psDec->nb_subfr - 2 ], 2 * sizeof( opus_int32 ) ); + + psPLC->subfr_length = psDec->subfr_length; + psPLC->nb_subfr = psDec->nb_subfr; +} + +static OPUS_INLINE void silk_PLC_energy(opus_int32 *energy1, opus_int *shift1, opus_int32 *energy2, opus_int *shift2, + const opus_int32 *exc_Q14, const opus_int32 *prevGain_Q10, int subfr_length, int nb_subfr) +{ + int i, k; + VARDECL( opus_int16, exc_buf ); + opus_int16 *exc_buf_ptr; + SAVE_STACK; + ALLOC( exc_buf, 2*subfr_length, opus_int16 ); + /* Find random noise component */ + /* Scale previous excitation signal */ + exc_buf_ptr = exc_buf; + for( k = 0; k < 2; k++ ) { + for( i = 0; i < subfr_length; i++ ) { + exc_buf_ptr[ i ] = (opus_int16)silk_SAT16( silk_RSHIFT( + silk_SMULWW( exc_Q14[ i + ( k + nb_subfr - 2 ) * subfr_length ], prevGain_Q10[ k ] ), 8 ) ); + } + exc_buf_ptr += subfr_length; + } + /* Find the subframe with lowest energy of the last two and use that as random noise generator */ + silk_sum_sqr_shift( energy1, shift1, exc_buf, subfr_length ); + silk_sum_sqr_shift( energy2, shift2, &exc_buf[ subfr_length ], subfr_length ); + RESTORE_STACK; +} + +static OPUS_INLINE void silk_PLC_conceal( + silk_decoder_state *psDec, /* I/O Decoder state */ + silk_decoder_control *psDecCtrl, /* I/O Decoder control */ + opus_int16 frame[], /* O LPC residual signal */ + int arch /* I Run-time architecture */ +) +{ + opus_int i, j, k; + opus_int lag, idx, sLTP_buf_idx, shift1, shift2; + opus_int32 rand_seed, harm_Gain_Q15, rand_Gain_Q15, inv_gain_Q30; + opus_int32 energy1, energy2, *rand_ptr, *pred_lag_ptr; + opus_int32 LPC_pred_Q10, LTP_pred_Q12; + opus_int16 rand_scale_Q14; + opus_int16 *B_Q14; + opus_int32 *sLPC_Q14_ptr; + opus_int16 A_Q12[ MAX_LPC_ORDER ]; +#ifdef SMALL_FOOTPRINT + opus_int16 *sLTP; +#else + VARDECL( opus_int16, sLTP ); +#endif + VARDECL( opus_int32, sLTP_Q14 ); + silk_PLC_struct *psPLC = &psDec->sPLC; + opus_int32 prevGain_Q10[2]; + SAVE_STACK; + + ALLOC( sLTP_Q14, psDec->ltp_mem_length + psDec->frame_length, opus_int32 ); +#ifdef SMALL_FOOTPRINT + /* Ugly hack that breaks aliasing rules to save stack: put sLTP at the very end of sLTP_Q14. */ + sLTP = ((opus_int16*)&sLTP_Q14[psDec->ltp_mem_length + psDec->frame_length])-psDec->ltp_mem_length; +#else + ALLOC( sLTP, psDec->ltp_mem_length, opus_int16 ); +#endif + + prevGain_Q10[0] = silk_RSHIFT( psPLC->prevGain_Q16[ 0 ], 6); + prevGain_Q10[1] = silk_RSHIFT( psPLC->prevGain_Q16[ 1 ], 6); + + if( psDec->first_frame_after_reset ) { + silk_memset( psPLC->prevLPC_Q12, 0, sizeof( psPLC->prevLPC_Q12 ) ); + } + + silk_PLC_energy(&energy1, &shift1, &energy2, &shift2, psDec->exc_Q14, prevGain_Q10, psDec->subfr_length, psDec->nb_subfr); + + if( silk_RSHIFT( energy1, shift2 ) < silk_RSHIFT( energy2, shift1 ) ) { + /* First sub-frame has lowest energy */ + rand_ptr = &psDec->exc_Q14[ silk_max_int( 0, ( psPLC->nb_subfr - 1 ) * psPLC->subfr_length - RAND_BUF_SIZE ) ]; + } else { + /* Second sub-frame has lowest energy */ + rand_ptr = &psDec->exc_Q14[ silk_max_int( 0, psPLC->nb_subfr * psPLC->subfr_length - RAND_BUF_SIZE ) ]; + } + + /* Set up Gain to random noise component */ + B_Q14 = psPLC->LTPCoef_Q14; + rand_scale_Q14 = psPLC->randScale_Q14; + + /* Set up attenuation gains */ + harm_Gain_Q15 = HARM_ATT_Q15[ silk_min_int( NB_ATT - 1, psDec->lossCnt ) ]; + if( psDec->prevSignalType == TYPE_VOICED ) { + rand_Gain_Q15 = PLC_RAND_ATTENUATE_V_Q15[ silk_min_int( NB_ATT - 1, psDec->lossCnt ) ]; + } else { + rand_Gain_Q15 = PLC_RAND_ATTENUATE_UV_Q15[ silk_min_int( NB_ATT - 1, psDec->lossCnt ) ]; + } + + /* LPC concealment. Apply BWE to previous LPC */ + silk_bwexpander( psPLC->prevLPC_Q12, psDec->LPC_order, SILK_FIX_CONST( BWE_COEF, 16 ) ); + + /* Preload LPC coeficients to array on stack. Gives small performance gain */ + silk_memcpy( A_Q12, psPLC->prevLPC_Q12, psDec->LPC_order * sizeof( opus_int16 ) ); + + /* First Lost frame */ + if( psDec->lossCnt == 0 ) { + rand_scale_Q14 = 1 << 14; + + /* Reduce random noise Gain for voiced frames */ + if( psDec->prevSignalType == TYPE_VOICED ) { + for( i = 0; i < LTP_ORDER; i++ ) { + rand_scale_Q14 -= B_Q14[ i ]; + } + rand_scale_Q14 = silk_max_16( 3277, rand_scale_Q14 ); /* 0.2 */ + rand_scale_Q14 = (opus_int16)silk_RSHIFT( silk_SMULBB( rand_scale_Q14, psPLC->prevLTP_scale_Q14 ), 14 ); + } else { + /* Reduce random noise for unvoiced frames with high LPC gain */ + opus_int32 invGain_Q30, down_scale_Q30; + + invGain_Q30 = silk_LPC_inverse_pred_gain( psPLC->prevLPC_Q12, psDec->LPC_order, arch ); + + down_scale_Q30 = silk_min_32( silk_RSHIFT( (opus_int32)1 << 30, LOG2_INV_LPC_GAIN_HIGH_THRES ), invGain_Q30 ); + down_scale_Q30 = silk_max_32( silk_RSHIFT( (opus_int32)1 << 30, LOG2_INV_LPC_GAIN_LOW_THRES ), down_scale_Q30 ); + down_scale_Q30 = silk_LSHIFT( down_scale_Q30, LOG2_INV_LPC_GAIN_HIGH_THRES ); + + rand_Gain_Q15 = silk_RSHIFT( silk_SMULWB( down_scale_Q30, rand_Gain_Q15 ), 14 ); + } + } + + rand_seed = psPLC->rand_seed; + lag = silk_RSHIFT_ROUND( psPLC->pitchL_Q8, 8 ); + sLTP_buf_idx = psDec->ltp_mem_length; + + /* Rewhiten LTP state */ + idx = psDec->ltp_mem_length - lag - psDec->LPC_order - LTP_ORDER / 2; + celt_assert( idx > 0 ); + silk_LPC_analysis_filter( &sLTP[ idx ], &psDec->outBuf[ idx ], A_Q12, psDec->ltp_mem_length - idx, psDec->LPC_order, arch ); + /* Scale LTP state */ + inv_gain_Q30 = silk_INVERSE32_varQ( psPLC->prevGain_Q16[ 1 ], 46 ); + inv_gain_Q30 = silk_min( inv_gain_Q30, silk_int32_MAX >> 1 ); + for( i = idx + psDec->LPC_order; i < psDec->ltp_mem_length; i++ ) { + sLTP_Q14[ i ] = silk_SMULWB( inv_gain_Q30, sLTP[ i ] ); + } + + /***************************/ + /* LTP synthesis filtering */ + /***************************/ + for( k = 0; k < psDec->nb_subfr; k++ ) { + /* Set up pointer */ + pred_lag_ptr = &sLTP_Q14[ sLTP_buf_idx - lag + LTP_ORDER / 2 ]; + for( i = 0; i < psDec->subfr_length; i++ ) { + /* Unrolled loop */ + /* Avoids introducing a bias because silk_SMLAWB() always rounds to -inf */ + LTP_pred_Q12 = 2; + LTP_pred_Q12 = silk_SMLAWB( LTP_pred_Q12, pred_lag_ptr[ 0 ], B_Q14[ 0 ] ); + LTP_pred_Q12 = silk_SMLAWB( LTP_pred_Q12, pred_lag_ptr[ -1 ], B_Q14[ 1 ] ); + LTP_pred_Q12 = silk_SMLAWB( LTP_pred_Q12, pred_lag_ptr[ -2 ], B_Q14[ 2 ] ); + LTP_pred_Q12 = silk_SMLAWB( LTP_pred_Q12, pred_lag_ptr[ -3 ], B_Q14[ 3 ] ); + LTP_pred_Q12 = silk_SMLAWB( LTP_pred_Q12, pred_lag_ptr[ -4 ], B_Q14[ 4 ] ); + pred_lag_ptr++; + + /* Generate LPC excitation */ + rand_seed = silk_RAND( rand_seed ); + idx = silk_RSHIFT( rand_seed, 25 ) & RAND_BUF_MASK; + sLTP_Q14[ sLTP_buf_idx ] = silk_LSHIFT32( silk_SMLAWB( LTP_pred_Q12, rand_ptr[ idx ], rand_scale_Q14 ), 2 ); + sLTP_buf_idx++; + } + + /* Gradually reduce LTP gain */ + for( j = 0; j < LTP_ORDER; j++ ) { + B_Q14[ j ] = silk_RSHIFT( silk_SMULBB( harm_Gain_Q15, B_Q14[ j ] ), 15 ); + } + /* Gradually reduce excitation gain */ + rand_scale_Q14 = silk_RSHIFT( silk_SMULBB( rand_scale_Q14, rand_Gain_Q15 ), 15 ); + + /* Slowly increase pitch lag */ + psPLC->pitchL_Q8 = silk_SMLAWB( psPLC->pitchL_Q8, psPLC->pitchL_Q8, PITCH_DRIFT_FAC_Q16 ); + psPLC->pitchL_Q8 = silk_min_32( psPLC->pitchL_Q8, silk_LSHIFT( silk_SMULBB( MAX_PITCH_LAG_MS, psDec->fs_kHz ), 8 ) ); + lag = silk_RSHIFT_ROUND( psPLC->pitchL_Q8, 8 ); + } + + /***************************/ + /* LPC synthesis filtering */ + /***************************/ + sLPC_Q14_ptr = &sLTP_Q14[ psDec->ltp_mem_length - MAX_LPC_ORDER ]; + + /* Copy LPC state */ + silk_memcpy( sLPC_Q14_ptr, psDec->sLPC_Q14_buf, MAX_LPC_ORDER * sizeof( opus_int32 ) ); + + celt_assert( psDec->LPC_order >= 10 ); /* check that unrolling works */ + for( i = 0; i < psDec->frame_length; i++ ) { + /* partly unrolled */ + /* Avoids introducing a bias because silk_SMLAWB() always rounds to -inf */ + LPC_pred_Q10 = silk_RSHIFT( psDec->LPC_order, 1 ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, sLPC_Q14_ptr[ MAX_LPC_ORDER + i - 1 ], A_Q12[ 0 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, sLPC_Q14_ptr[ MAX_LPC_ORDER + i - 2 ], A_Q12[ 1 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, sLPC_Q14_ptr[ MAX_LPC_ORDER + i - 3 ], A_Q12[ 2 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, sLPC_Q14_ptr[ MAX_LPC_ORDER + i - 4 ], A_Q12[ 3 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, sLPC_Q14_ptr[ MAX_LPC_ORDER + i - 5 ], A_Q12[ 4 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, sLPC_Q14_ptr[ MAX_LPC_ORDER + i - 6 ], A_Q12[ 5 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, sLPC_Q14_ptr[ MAX_LPC_ORDER + i - 7 ], A_Q12[ 6 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, sLPC_Q14_ptr[ MAX_LPC_ORDER + i - 8 ], A_Q12[ 7 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, sLPC_Q14_ptr[ MAX_LPC_ORDER + i - 9 ], A_Q12[ 8 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, sLPC_Q14_ptr[ MAX_LPC_ORDER + i - 10 ], A_Q12[ 9 ] ); + for( j = 10; j < psDec->LPC_order; j++ ) { + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, sLPC_Q14_ptr[ MAX_LPC_ORDER + i - j - 1 ], A_Q12[ j ] ); + } + + /* Add prediction to LPC excitation */ + sLPC_Q14_ptr[ MAX_LPC_ORDER + i ] = silk_ADD_SAT32( sLPC_Q14_ptr[ MAX_LPC_ORDER + i ], + silk_LSHIFT_SAT32( LPC_pred_Q10, 4 )); + + /* Scale with Gain */ + frame[ i ] = (opus_int16)silk_SAT16( silk_SAT16( silk_RSHIFT_ROUND( silk_SMULWW( sLPC_Q14_ptr[ MAX_LPC_ORDER + i ], prevGain_Q10[ 1 ] ), 8 ) ) ); + } + + /* Save LPC state */ + silk_memcpy( psDec->sLPC_Q14_buf, &sLPC_Q14_ptr[ psDec->frame_length ], MAX_LPC_ORDER * sizeof( opus_int32 ) ); + + /**************************************/ + /* Update states */ + /**************************************/ + psPLC->rand_seed = rand_seed; + psPLC->randScale_Q14 = rand_scale_Q14; + for( i = 0; i < MAX_NB_SUBFR; i++ ) { + psDecCtrl->pitchL[ i ] = lag; + } + RESTORE_STACK; +} + +/* Glues concealed frames with new good received frames */ +void silk_PLC_glue_frames( + silk_decoder_state *psDec, /* I/O decoder state */ + opus_int16 frame[], /* I/O signal */ + opus_int length /* I length of signal */ +) +{ + opus_int i, energy_shift; + opus_int32 energy; + silk_PLC_struct *psPLC; + psPLC = &psDec->sPLC; + + if( psDec->lossCnt ) { + /* Calculate energy in concealed residual */ + silk_sum_sqr_shift( &psPLC->conc_energy, &psPLC->conc_energy_shift, frame, length ); + + psPLC->last_frame_lost = 1; + } else { + if( psDec->sPLC.last_frame_lost ) { + /* Calculate residual in decoded signal if last frame was lost */ + silk_sum_sqr_shift( &energy, &energy_shift, frame, length ); + + /* Normalize energies */ + if( energy_shift > psPLC->conc_energy_shift ) { + psPLC->conc_energy = silk_RSHIFT( psPLC->conc_energy, energy_shift - psPLC->conc_energy_shift ); + } else if( energy_shift < psPLC->conc_energy_shift ) { + energy = silk_RSHIFT( energy, psPLC->conc_energy_shift - energy_shift ); + } + + /* Fade in the energy difference */ + if( energy > psPLC->conc_energy ) { + opus_int32 frac_Q24, LZ; + opus_int32 gain_Q16, slope_Q16; + + LZ = silk_CLZ32( psPLC->conc_energy ); + LZ = LZ - 1; + psPLC->conc_energy = silk_LSHIFT( psPLC->conc_energy, LZ ); + energy = silk_RSHIFT( energy, silk_max_32( 24 - LZ, 0 ) ); + + frac_Q24 = silk_DIV32( psPLC->conc_energy, silk_max( energy, 1 ) ); + + gain_Q16 = silk_LSHIFT( silk_SQRT_APPROX( frac_Q24 ), 4 ); + slope_Q16 = silk_DIV32_16( ( (opus_int32)1 << 16 ) - gain_Q16, length ); + /* Make slope 4x steeper to avoid missing onsets after DTX */ + slope_Q16 = silk_LSHIFT( slope_Q16, 2 ); + + for( i = 0; i < length; i++ ) { + frame[ i ] = silk_SMULWB( gain_Q16, frame[ i ] ); + gain_Q16 += slope_Q16; + if( gain_Q16 > (opus_int32)1 << 16 ) { + break; + } + } + } + } + psPLC->last_frame_lost = 0; + } +} diff --git a/media/libopus/silk/PLC.h b/media/libopus/silk/PLC.h new file mode 100644 index 0000000000..6438f51633 --- /dev/null +++ b/media/libopus/silk/PLC.h @@ -0,0 +1,62 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef SILK_PLC_H +#define SILK_PLC_H + +#include "main.h" + +#define BWE_COEF 0.99 +#define V_PITCH_GAIN_START_MIN_Q14 11469 /* 0.7 in Q14 */ +#define V_PITCH_GAIN_START_MAX_Q14 15565 /* 0.95 in Q14 */ +#define MAX_PITCH_LAG_MS 18 +#define RAND_BUF_SIZE 128 +#define RAND_BUF_MASK ( RAND_BUF_SIZE - 1 ) +#define LOG2_INV_LPC_GAIN_HIGH_THRES 3 /* 2^3 = 8 dB LPC gain */ +#define LOG2_INV_LPC_GAIN_LOW_THRES 8 /* 2^8 = 24 dB LPC gain */ +#define PITCH_DRIFT_FAC_Q16 655 /* 0.01 in Q16 */ + +void silk_PLC_Reset( + silk_decoder_state *psDec /* I/O Decoder state */ +); + +void silk_PLC( + silk_decoder_state *psDec, /* I/O Decoder state */ + silk_decoder_control *psDecCtrl, /* I/O Decoder control */ + opus_int16 frame[], /* I/O signal */ + opus_int lost, /* I Loss flag */ + int arch /* I Run-time architecture */ +); + +void silk_PLC_glue_frames( + silk_decoder_state *psDec, /* I/O decoder state */ + opus_int16 frame[], /* I/O signal */ + opus_int length /* I length of signal */ +); + +#endif + diff --git a/media/libopus/silk/SigProc_FIX.h b/media/libopus/silk/SigProc_FIX.h new file mode 100644 index 0000000000..fbdfa82e26 --- /dev/null +++ b/media/libopus/silk/SigProc_FIX.h @@ -0,0 +1,643 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef SILK_SIGPROC_FIX_H +#define SILK_SIGPROC_FIX_H + +#ifdef __cplusplus +extern "C" +{ +#endif + +/*#define silk_MACRO_COUNT */ /* Used to enable WMOPS counting */ + +#define SILK_MAX_ORDER_LPC 24 /* max order of the LPC analysis in schur() and k2a() */ + +#include <string.h> /* for memset(), memcpy(), memmove() */ +#include "typedef.h" +#include "resampler_structs.h" +#include "macros.h" +#include "cpu_support.h" + +#if defined(OPUS_X86_MAY_HAVE_SSE4_1) +#include "x86/SigProc_FIX_sse.h" +#endif + +#if (defined(OPUS_ARM_ASM) || defined(OPUS_ARM_MAY_HAVE_NEON_INTR)) +#include "arm/biquad_alt_arm.h" +#include "arm/LPC_inv_pred_gain_arm.h" +#endif + +/********************************************************************/ +/* SIGNAL PROCESSING FUNCTIONS */ +/********************************************************************/ + +/*! + * Initialize/reset the resampler state for a given pair of input/output sampling rates +*/ +opus_int silk_resampler_init( + silk_resampler_state_struct *S, /* I/O Resampler state */ + opus_int32 Fs_Hz_in, /* I Input sampling rate (Hz) */ + opus_int32 Fs_Hz_out, /* I Output sampling rate (Hz) */ + opus_int forEnc /* I If 1: encoder; if 0: decoder */ +); + +/*! + * Resampler: convert from one sampling rate to another + */ +opus_int silk_resampler( + silk_resampler_state_struct *S, /* I/O Resampler state */ + opus_int16 out[], /* O Output signal */ + const opus_int16 in[], /* I Input signal */ + opus_int32 inLen /* I Number of input samples */ +); + +/*! +* Downsample 2x, mediocre quality +*/ +void silk_resampler_down2( + opus_int32 *S, /* I/O State vector [ 2 ] */ + opus_int16 *out, /* O Output signal [ len ] */ + const opus_int16 *in, /* I Input signal [ floor(len/2) ] */ + opus_int32 inLen /* I Number of input samples */ +); + +/*! + * Downsample by a factor 2/3, low quality +*/ +void silk_resampler_down2_3( + opus_int32 *S, /* I/O State vector [ 6 ] */ + opus_int16 *out, /* O Output signal [ floor(2*inLen/3) ] */ + const opus_int16 *in, /* I Input signal [ inLen ] */ + opus_int32 inLen /* I Number of input samples */ +); + +/*! + * second order ARMA filter; + * slower than biquad() but uses more precise coefficients + * can handle (slowly) varying coefficients + */ +void silk_biquad_alt_stride1( + const opus_int16 *in, /* I input signal */ + const opus_int32 *B_Q28, /* I MA coefficients [3] */ + const opus_int32 *A_Q28, /* I AR coefficients [2] */ + opus_int32 *S, /* I/O State vector [2] */ + opus_int16 *out, /* O output signal */ + const opus_int32 len /* I signal length (must be even) */ +); + +void silk_biquad_alt_stride2_c( + const opus_int16 *in, /* I input signal */ + const opus_int32 *B_Q28, /* I MA coefficients [3] */ + const opus_int32 *A_Q28, /* I AR coefficients [2] */ + opus_int32 *S, /* I/O State vector [4] */ + opus_int16 *out, /* O output signal */ + const opus_int32 len /* I signal length (must be even) */ +); + +/* Variable order MA prediction error filter. */ +void silk_LPC_analysis_filter( + opus_int16 *out, /* O Output signal */ + const opus_int16 *in, /* I Input signal */ + const opus_int16 *B, /* I MA prediction coefficients, Q12 [order] */ + const opus_int32 len, /* I Signal length */ + const opus_int32 d, /* I Filter order */ + int arch /* I Run-time architecture */ +); + +/* Chirp (bandwidth expand) LP AR filter */ +void silk_bwexpander( + opus_int16 *ar, /* I/O AR filter to be expanded (without leading 1) */ + const opus_int d, /* I Length of ar */ + opus_int32 chirp_Q16 /* I Chirp factor (typically in the range 0 to 1) */ +); + +/* Chirp (bandwidth expand) LP AR filter */ +void silk_bwexpander_32( + opus_int32 *ar, /* I/O AR filter to be expanded (without leading 1) */ + const opus_int d, /* I Length of ar */ + opus_int32 chirp_Q16 /* I Chirp factor in Q16 */ +); + +/* Compute inverse of LPC prediction gain, and */ +/* test if LPC coefficients are stable (all poles within unit circle) */ +opus_int32 silk_LPC_inverse_pred_gain_c( /* O Returns inverse prediction gain in energy domain, Q30 */ + const opus_int16 *A_Q12, /* I Prediction coefficients, Q12 [order] */ + const opus_int order /* I Prediction order */ +); + +/* Split signal in two decimated bands using first-order allpass filters */ +void silk_ana_filt_bank_1( + const opus_int16 *in, /* I Input signal [N] */ + opus_int32 *S, /* I/O State vector [2] */ + opus_int16 *outL, /* O Low band [N/2] */ + opus_int16 *outH, /* O High band [N/2] */ + const opus_int32 N /* I Number of input samples */ +); + +#if !defined(OVERRIDE_silk_biquad_alt_stride2) +#define silk_biquad_alt_stride2(in, B_Q28, A_Q28, S, out, len, arch) ((void)(arch), silk_biquad_alt_stride2_c(in, B_Q28, A_Q28, S, out, len)) +#endif + +#if !defined(OVERRIDE_silk_LPC_inverse_pred_gain) +#define silk_LPC_inverse_pred_gain(A_Q12, order, arch) ((void)(arch), silk_LPC_inverse_pred_gain_c(A_Q12, order)) +#endif + +/********************************************************************/ +/* SCALAR FUNCTIONS */ +/********************************************************************/ + +/* Approximation of 128 * log2() (exact inverse of approx 2^() below) */ +/* Convert input to a log scale */ +opus_int32 silk_lin2log( + const opus_int32 inLin /* I input in linear scale */ +); + +/* Approximation of a sigmoid function */ +opus_int silk_sigm_Q15( + opus_int in_Q5 /* I */ +); + +/* Approximation of 2^() (exact inverse of approx log2() above) */ +/* Convert input to a linear scale */ +opus_int32 silk_log2lin( + const opus_int32 inLog_Q7 /* I input on log scale */ +); + +/* Compute number of bits to right shift the sum of squares of a vector */ +/* of int16s to make it fit in an int32 */ +void silk_sum_sqr_shift( + opus_int32 *energy, /* O Energy of x, after shifting to the right */ + opus_int *shift, /* O Number of bits right shift applied to energy */ + const opus_int16 *x, /* I Input vector */ + opus_int len /* I Length of input vector */ +); + +/* Calculates the reflection coefficients from the correlation sequence */ +/* Faster than schur64(), but much less accurate. */ +/* uses SMLAWB(), requiring armv5E and higher. */ +opus_int32 silk_schur( /* O Returns residual energy */ + opus_int16 *rc_Q15, /* O reflection coefficients [order] Q15 */ + const opus_int32 *c, /* I correlations [order+1] */ + const opus_int32 order /* I prediction order */ +); + +/* Calculates the reflection coefficients from the correlation sequence */ +/* Slower than schur(), but more accurate. */ +/* Uses SMULL(), available on armv4 */ +opus_int32 silk_schur64( /* O returns residual energy */ + opus_int32 rc_Q16[], /* O Reflection coefficients [order] Q16 */ + const opus_int32 c[], /* I Correlations [order+1] */ + opus_int32 order /* I Prediction order */ +); + +/* Step up function, converts reflection coefficients to prediction coefficients */ +void silk_k2a( + opus_int32 *A_Q24, /* O Prediction coefficients [order] Q24 */ + const opus_int16 *rc_Q15, /* I Reflection coefficients [order] Q15 */ + const opus_int32 order /* I Prediction order */ +); + +/* Step up function, converts reflection coefficients to prediction coefficients */ +void silk_k2a_Q16( + opus_int32 *A_Q24, /* O Prediction coefficients [order] Q24 */ + const opus_int32 *rc_Q16, /* I Reflection coefficients [order] Q16 */ + const opus_int32 order /* I Prediction order */ +); + +/* Apply sine window to signal vector. */ +/* Window types: */ +/* 1 -> sine window from 0 to pi/2 */ +/* 2 -> sine window from pi/2 to pi */ +/* every other sample of window is linearly interpolated, for speed */ +void silk_apply_sine_window( + opus_int16 px_win[], /* O Pointer to windowed signal */ + const opus_int16 px[], /* I Pointer to input signal */ + const opus_int win_type, /* I Selects a window type */ + const opus_int length /* I Window length, multiple of 4 */ +); + +/* Compute autocorrelation */ +void silk_autocorr( + opus_int32 *results, /* O Result (length correlationCount) */ + opus_int *scale, /* O Scaling of the correlation vector */ + const opus_int16 *inputData, /* I Input data to correlate */ + const opus_int inputDataSize, /* I Length of input */ + const opus_int correlationCount, /* I Number of correlation taps to compute */ + int arch /* I Run-time architecture */ +); + +void silk_decode_pitch( + opus_int16 lagIndex, /* I */ + opus_int8 contourIndex, /* O */ + opus_int pitch_lags[], /* O 4 pitch values */ + const opus_int Fs_kHz, /* I sampling frequency (kHz) */ + const opus_int nb_subfr /* I number of sub frames */ +); + +opus_int silk_pitch_analysis_core( /* O Voicing estimate: 0 voiced, 1 unvoiced */ + const opus_int16 *frame, /* I Signal of length PE_FRAME_LENGTH_MS*Fs_kHz */ + opus_int *pitch_out, /* O 4 pitch lag values */ + opus_int16 *lagIndex, /* O Lag Index */ + opus_int8 *contourIndex, /* O Pitch contour Index */ + opus_int *LTPCorr_Q15, /* I/O Normalized correlation; input: value from previous frame */ + opus_int prevLag, /* I Last lag of previous frame; set to zero is unvoiced */ + const opus_int32 search_thres1_Q16, /* I First stage threshold for lag candidates 0 - 1 */ + const opus_int search_thres2_Q13, /* I Final threshold for lag candidates 0 - 1 */ + const opus_int Fs_kHz, /* I Sample frequency (kHz) */ + const opus_int complexity, /* I Complexity setting, 0-2, where 2 is highest */ + const opus_int nb_subfr, /* I number of 5 ms subframes */ + int arch /* I Run-time architecture */ +); + +/* Compute Normalized Line Spectral Frequencies (NLSFs) from whitening filter coefficients */ +/* If not all roots are found, the a_Q16 coefficients are bandwidth expanded until convergence. */ +void silk_A2NLSF( + opus_int16 *NLSF, /* O Normalized Line Spectral Frequencies in Q15 (0..2^15-1) [d] */ + opus_int32 *a_Q16, /* I/O Monic whitening filter coefficients in Q16 [d] */ + const opus_int d /* I Filter order (must be even) */ +); + +/* compute whitening filter coefficients from normalized line spectral frequencies */ +void silk_NLSF2A( + opus_int16 *a_Q12, /* O monic whitening filter coefficients in Q12, [ d ] */ + const opus_int16 *NLSF, /* I normalized line spectral frequencies in Q15, [ d ] */ + const opus_int d, /* I filter order (should be even) */ + int arch /* I Run-time architecture */ +); + +/* Convert int32 coefficients to int16 coefs and make sure there's no wrap-around */ +void silk_LPC_fit( + opus_int16 *a_QOUT, /* O Output signal */ + opus_int32 *a_QIN, /* I/O Input signal */ + const opus_int QOUT, /* I Input Q domain */ + const opus_int QIN, /* I Input Q domain */ + const opus_int d /* I Filter order */ +); + +void silk_insertion_sort_increasing( + opus_int32 *a, /* I/O Unsorted / Sorted vector */ + opus_int *idx, /* O Index vector for the sorted elements */ + const opus_int L, /* I Vector length */ + const opus_int K /* I Number of correctly sorted positions */ +); + +void silk_insertion_sort_decreasing_int16( + opus_int16 *a, /* I/O Unsorted / Sorted vector */ + opus_int *idx, /* O Index vector for the sorted elements */ + const opus_int L, /* I Vector length */ + const opus_int K /* I Number of correctly sorted positions */ +); + +void silk_insertion_sort_increasing_all_values_int16( + opus_int16 *a, /* I/O Unsorted / Sorted vector */ + const opus_int L /* I Vector length */ +); + +/* NLSF stabilizer, for a single input data vector */ +void silk_NLSF_stabilize( + opus_int16 *NLSF_Q15, /* I/O Unstable/stabilized normalized LSF vector in Q15 [L] */ + const opus_int16 *NDeltaMin_Q15, /* I Min distance vector, NDeltaMin_Q15[L] must be >= 1 [L+1] */ + const opus_int L /* I Number of NLSF parameters in the input vector */ +); + +/* Laroia low complexity NLSF weights */ +void silk_NLSF_VQ_weights_laroia( + opus_int16 *pNLSFW_Q_OUT, /* O Pointer to input vector weights [D] */ + const opus_int16 *pNLSF_Q15, /* I Pointer to input vector [D] */ + const opus_int D /* I Input vector dimension (even) */ +); + +/* Compute reflection coefficients from input signal */ +void silk_burg_modified_c( + opus_int32 *res_nrg, /* O Residual energy */ + opus_int *res_nrg_Q, /* O Residual energy Q value */ + opus_int32 A_Q16[], /* O Prediction coefficients (length order) */ + const opus_int16 x[], /* I Input signal, length: nb_subfr * ( D + subfr_length ) */ + const opus_int32 minInvGain_Q30, /* I Inverse of max prediction gain */ + const opus_int subfr_length, /* I Input signal subframe length (incl. D preceding samples) */ + const opus_int nb_subfr, /* I Number of subframes stacked in x */ + const opus_int D, /* I Order */ + int arch /* I Run-time architecture */ +); + +/* Copy and multiply a vector by a constant */ +void silk_scale_copy_vector16( + opus_int16 *data_out, + const opus_int16 *data_in, + opus_int32 gain_Q16, /* I Gain in Q16 */ + const opus_int dataSize /* I Length */ +); + +/* Some for the LTP related function requires Q26 to work.*/ +void silk_scale_vector32_Q26_lshift_18( + opus_int32 *data1, /* I/O Q0/Q18 */ + opus_int32 gain_Q26, /* I Q26 */ + opus_int dataSize /* I length */ +); + +/********************************************************************/ +/* INLINE ARM MATH */ +/********************************************************************/ + +/* return sum( inVec1[i] * inVec2[i] ) */ + +opus_int32 silk_inner_prod_aligned( + const opus_int16 *const inVec1, /* I input vector 1 */ + const opus_int16 *const inVec2, /* I input vector 2 */ + const opus_int len, /* I vector lengths */ + int arch /* I Run-time architecture */ +); + + +opus_int32 silk_inner_prod_aligned_scale( + const opus_int16 *const inVec1, /* I input vector 1 */ + const opus_int16 *const inVec2, /* I input vector 2 */ + const opus_int scale, /* I number of bits to shift */ + const opus_int len /* I vector lengths */ +); + +opus_int64 silk_inner_prod16_c( + const opus_int16 *inVec1, /* I input vector 1 */ + const opus_int16 *inVec2, /* I input vector 2 */ + const opus_int len /* I vector lengths */ +); + +/********************************************************************/ +/* MACROS */ +/********************************************************************/ + +/* Rotate a32 right by 'rot' bits. Negative rot values result in rotating + left. Output is 32bit int. + Note: contemporary compilers recognize the C expression below and + compile it into a 'ror' instruction if available. No need for OPUS_INLINE ASM! */ +static OPUS_INLINE opus_int32 silk_ROR32( opus_int32 a32, opus_int rot ) +{ + opus_uint32 x = (opus_uint32) a32; + opus_uint32 r = (opus_uint32) rot; + opus_uint32 m = (opus_uint32) -rot; + if( rot == 0 ) { + return a32; + } else if( rot < 0 ) { + return (opus_int32) ((x << m) | (x >> (32 - m))); + } else { + return (opus_int32) ((x << (32 - r)) | (x >> r)); + } +} + +/* Allocate opus_int16 aligned to 4-byte memory address */ +#if EMBEDDED_ARM +#define silk_DWORD_ALIGN __attribute__((aligned(4))) +#else +#define silk_DWORD_ALIGN +#endif + +/* Useful Macros that can be adjusted to other platforms */ +#define silk_memcpy(dest, src, size) memcpy((dest), (src), (size)) +#define silk_memset(dest, src, size) memset((dest), (src), (size)) +#define silk_memmove(dest, src, size) memmove((dest), (src), (size)) + +/* Fixed point macros */ + +/* (a32 * b32) output have to be 32bit int */ +#define silk_MUL(a32, b32) ((a32) * (b32)) + +/* (a32 * b32) output have to be 32bit uint */ +#define silk_MUL_uint(a32, b32) silk_MUL(a32, b32) + +/* a32 + (b32 * c32) output have to be 32bit int */ +#define silk_MLA(a32, b32, c32) silk_ADD32((a32),((b32) * (c32))) + +/* a32 + (b32 * c32) output have to be 32bit uint */ +#define silk_MLA_uint(a32, b32, c32) silk_MLA(a32, b32, c32) + +/* ((a32 >> 16) * (b32 >> 16)) output have to be 32bit int */ +#define silk_SMULTT(a32, b32) (((a32) >> 16) * ((b32) >> 16)) + +/* a32 + ((a32 >> 16) * (b32 >> 16)) output have to be 32bit int */ +#define silk_SMLATT(a32, b32, c32) silk_ADD32((a32),((b32) >> 16) * ((c32) >> 16)) + +#define silk_SMLALBB(a64, b16, c16) silk_ADD64((a64),(opus_int64)((opus_int32)(b16) * (opus_int32)(c16))) + +/* (a32 * b32) */ +#define silk_SMULL(a32, b32) ((opus_int64)(a32) * /*(opus_int64)*/(b32)) + +/* Adds two signed 32-bit values in a way that can overflow, while not relying on undefined behaviour + (just standard two's complement implementation-specific behaviour) */ +#define silk_ADD32_ovflw(a, b) ((opus_int32)((opus_uint32)(a) + (opus_uint32)(b))) +/* Subtractss two signed 32-bit values in a way that can overflow, while not relying on undefined behaviour + (just standard two's complement implementation-specific behaviour) */ +#define silk_SUB32_ovflw(a, b) ((opus_int32)((opus_uint32)(a) - (opus_uint32)(b))) + +/* Multiply-accumulate macros that allow overflow in the addition (ie, no asserts in debug mode) */ +#define silk_MLA_ovflw(a32, b32, c32) silk_ADD32_ovflw((a32), (opus_uint32)(b32) * (opus_uint32)(c32)) +#define silk_SMLABB_ovflw(a32, b32, c32) (silk_ADD32_ovflw((a32) , ((opus_int32)((opus_int16)(b32))) * (opus_int32)((opus_int16)(c32)))) + +#define silk_DIV32_16(a32, b16) ((opus_int32)((a32) / (b16))) +#define silk_DIV32(a32, b32) ((opus_int32)((a32) / (b32))) + +/* These macros enables checking for overflow in silk_API_Debug.h*/ +#define silk_ADD16(a, b) ((a) + (b)) +#define silk_ADD32(a, b) ((a) + (b)) +#define silk_ADD64(a, b) ((a) + (b)) + +#define silk_SUB16(a, b) ((a) - (b)) +#define silk_SUB32(a, b) ((a) - (b)) +#define silk_SUB64(a, b) ((a) - (b)) + +#define silk_SAT8(a) ((a) > silk_int8_MAX ? silk_int8_MAX : \ + ((a) < silk_int8_MIN ? silk_int8_MIN : (a))) +#define silk_SAT16(a) ((a) > silk_int16_MAX ? silk_int16_MAX : \ + ((a) < silk_int16_MIN ? silk_int16_MIN : (a))) +#define silk_SAT32(a) ((a) > silk_int32_MAX ? silk_int32_MAX : \ + ((a) < silk_int32_MIN ? silk_int32_MIN : (a))) + +#define silk_CHECK_FIT8(a) (a) +#define silk_CHECK_FIT16(a) (a) +#define silk_CHECK_FIT32(a) (a) + +#define silk_ADD_SAT16(a, b) (opus_int16)silk_SAT16( silk_ADD32( (opus_int32)(a), (b) ) ) +#define silk_ADD_SAT64(a, b) ((((a) + (b)) & 0x8000000000000000LL) == 0 ? \ + ((((a) & (b)) & 0x8000000000000000LL) != 0 ? silk_int64_MIN : (a)+(b)) : \ + ((((a) | (b)) & 0x8000000000000000LL) == 0 ? silk_int64_MAX : (a)+(b)) ) + +#define silk_SUB_SAT16(a, b) (opus_int16)silk_SAT16( silk_SUB32( (opus_int32)(a), (b) ) ) +#define silk_SUB_SAT64(a, b) ((((a)-(b)) & 0x8000000000000000LL) == 0 ? \ + (( (a) & ((b)^0x8000000000000000LL) & 0x8000000000000000LL) ? silk_int64_MIN : (a)-(b)) : \ + ((((a)^0x8000000000000000LL) & (b) & 0x8000000000000000LL) ? silk_int64_MAX : (a)-(b)) ) + +/* Saturation for positive input values */ +#define silk_POS_SAT32(a) ((a) > silk_int32_MAX ? silk_int32_MAX : (a)) + +/* Add with saturation for positive input values */ +#define silk_ADD_POS_SAT8(a, b) ((((a)+(b)) & 0x80) ? silk_int8_MAX : ((a)+(b))) +#define silk_ADD_POS_SAT16(a, b) ((((a)+(b)) & 0x8000) ? silk_int16_MAX : ((a)+(b))) +#define silk_ADD_POS_SAT32(a, b) ((((opus_uint32)(a)+(opus_uint32)(b)) & 0x80000000) ? silk_int32_MAX : ((a)+(b))) + +#define silk_LSHIFT8(a, shift) ((opus_int8)((opus_uint8)(a)<<(shift))) /* shift >= 0, shift < 8 */ +#define silk_LSHIFT16(a, shift) ((opus_int16)((opus_uint16)(a)<<(shift))) /* shift >= 0, shift < 16 */ +#define silk_LSHIFT32(a, shift) ((opus_int32)((opus_uint32)(a)<<(shift))) /* shift >= 0, shift < 32 */ +#define silk_LSHIFT64(a, shift) ((opus_int64)((opus_uint64)(a)<<(shift))) /* shift >= 0, shift < 64 */ +#define silk_LSHIFT(a, shift) silk_LSHIFT32(a, shift) /* shift >= 0, shift < 32 */ + +#define silk_RSHIFT8(a, shift) ((a)>>(shift)) /* shift >= 0, shift < 8 */ +#define silk_RSHIFT16(a, shift) ((a)>>(shift)) /* shift >= 0, shift < 16 */ +#define silk_RSHIFT32(a, shift) ((a)>>(shift)) /* shift >= 0, shift < 32 */ +#define silk_RSHIFT64(a, shift) ((a)>>(shift)) /* shift >= 0, shift < 64 */ +#define silk_RSHIFT(a, shift) silk_RSHIFT32(a, shift) /* shift >= 0, shift < 32 */ + +/* saturates before shifting */ +#define silk_LSHIFT_SAT32(a, shift) (silk_LSHIFT32( silk_LIMIT( (a), silk_RSHIFT32( silk_int32_MIN, (shift) ), \ + silk_RSHIFT32( silk_int32_MAX, (shift) ) ), (shift) )) + +#define silk_LSHIFT_ovflw(a, shift) ((opus_int32)((opus_uint32)(a) << (shift))) /* shift >= 0, allowed to overflow */ +#define silk_LSHIFT_uint(a, shift) ((a) << (shift)) /* shift >= 0 */ +#define silk_RSHIFT_uint(a, shift) ((a) >> (shift)) /* shift >= 0 */ + +#define silk_ADD_LSHIFT(a, b, shift) ((a) + silk_LSHIFT((b), (shift))) /* shift >= 0 */ +#define silk_ADD_LSHIFT32(a, b, shift) silk_ADD32((a), silk_LSHIFT32((b), (shift))) /* shift >= 0 */ +#define silk_ADD_LSHIFT_uint(a, b, shift) ((a) + silk_LSHIFT_uint((b), (shift))) /* shift >= 0 */ +#define silk_ADD_RSHIFT(a, b, shift) ((a) + silk_RSHIFT((b), (shift))) /* shift >= 0 */ +#define silk_ADD_RSHIFT32(a, b, shift) silk_ADD32((a), silk_RSHIFT32((b), (shift))) /* shift >= 0 */ +#define silk_ADD_RSHIFT_uint(a, b, shift) ((a) + silk_RSHIFT_uint((b), (shift))) /* shift >= 0 */ +#define silk_SUB_LSHIFT32(a, b, shift) silk_SUB32((a), silk_LSHIFT32((b), (shift))) /* shift >= 0 */ +#define silk_SUB_RSHIFT32(a, b, shift) silk_SUB32((a), silk_RSHIFT32((b), (shift))) /* shift >= 0 */ + +/* Requires that shift > 0 */ +#define silk_RSHIFT_ROUND(a, shift) ((shift) == 1 ? ((a) >> 1) + ((a) & 1) : (((a) >> ((shift) - 1)) + 1) >> 1) +#define silk_RSHIFT_ROUND64(a, shift) ((shift) == 1 ? ((a) >> 1) + ((a) & 1) : (((a) >> ((shift) - 1)) + 1) >> 1) + +/* Number of rightshift required to fit the multiplication */ +#define silk_NSHIFT_MUL_32_32(a, b) ( -(31- (32-silk_CLZ32(silk_abs(a)) + (32-silk_CLZ32(silk_abs(b))))) ) +#define silk_NSHIFT_MUL_16_16(a, b) ( -(15- (16-silk_CLZ16(silk_abs(a)) + (16-silk_CLZ16(silk_abs(b))))) ) + + +#define silk_min(a, b) (((a) < (b)) ? (a) : (b)) +#define silk_max(a, b) (((a) > (b)) ? (a) : (b)) + +/* Macro to convert floating-point constants to fixed-point */ +#define SILK_FIX_CONST( C, Q ) ((opus_int32)((C) * ((opus_int64)1 << (Q)) + 0.5)) + +/* silk_min() versions with typecast in the function call */ +static OPUS_INLINE opus_int silk_min_int(opus_int a, opus_int b) +{ + return (((a) < (b)) ? (a) : (b)); +} +static OPUS_INLINE opus_int16 silk_min_16(opus_int16 a, opus_int16 b) +{ + return (((a) < (b)) ? (a) : (b)); +} +static OPUS_INLINE opus_int32 silk_min_32(opus_int32 a, opus_int32 b) +{ + return (((a) < (b)) ? (a) : (b)); +} +static OPUS_INLINE opus_int64 silk_min_64(opus_int64 a, opus_int64 b) +{ + return (((a) < (b)) ? (a) : (b)); +} + +/* silk_min() versions with typecast in the function call */ +static OPUS_INLINE opus_int silk_max_int(opus_int a, opus_int b) +{ + return (((a) > (b)) ? (a) : (b)); +} +static OPUS_INLINE opus_int16 silk_max_16(opus_int16 a, opus_int16 b) +{ + return (((a) > (b)) ? (a) : (b)); +} +static OPUS_INLINE opus_int32 silk_max_32(opus_int32 a, opus_int32 b) +{ + return (((a) > (b)) ? (a) : (b)); +} +static OPUS_INLINE opus_int64 silk_max_64(opus_int64 a, opus_int64 b) +{ + return (((a) > (b)) ? (a) : (b)); +} + +#define silk_LIMIT( a, limit1, limit2) ((limit1) > (limit2) ? ((a) > (limit1) ? (limit1) : ((a) < (limit2) ? (limit2) : (a))) \ + : ((a) > (limit2) ? (limit2) : ((a) < (limit1) ? (limit1) : (a)))) + +#define silk_LIMIT_int silk_LIMIT +#define silk_LIMIT_16 silk_LIMIT +#define silk_LIMIT_32 silk_LIMIT + +#define silk_abs(a) (((a) > 0) ? (a) : -(a)) /* Be careful, silk_abs returns wrong when input equals to silk_intXX_MIN */ +#define silk_abs_int(a) (((a) ^ ((a) >> (8 * sizeof(a) - 1))) - ((a) >> (8 * sizeof(a) - 1))) +#define silk_abs_int32(a) (((a) ^ ((a) >> 31)) - ((a) >> 31)) +#define silk_abs_int64(a) (((a) > 0) ? (a) : -(a)) + +#define silk_sign(a) ((a) > 0 ? 1 : ( (a) < 0 ? -1 : 0 )) + +/* PSEUDO-RANDOM GENERATOR */ +/* Make sure to store the result as the seed for the next call (also in between */ +/* frames), otherwise result won't be random at all. When only using some of the */ +/* bits, take the most significant bits by right-shifting. */ +#define RAND_MULTIPLIER 196314165 +#define RAND_INCREMENT 907633515 +#define silk_RAND(seed) (silk_MLA_ovflw((RAND_INCREMENT), (seed), (RAND_MULTIPLIER))) + +/* Add some multiplication functions that can be easily mapped to ARM. */ + +/* silk_SMMUL: Signed top word multiply. + ARMv6 2 instruction cycles. + ARMv3M+ 3 instruction cycles. use SMULL and ignore LSB registers.(except xM)*/ +/*#define silk_SMMUL(a32, b32) (opus_int32)silk_RSHIFT(silk_SMLAL(silk_SMULWB((a32), (b32)), (a32), silk_RSHIFT_ROUND((b32), 16)), 16)*/ +/* the following seems faster on x86 */ +#define silk_SMMUL(a32, b32) (opus_int32)silk_RSHIFT64(silk_SMULL((a32), (b32)), 32) + +#if !defined(OVERRIDE_silk_burg_modified) +#define silk_burg_modified(res_nrg, res_nrg_Q, A_Q16, x, minInvGain_Q30, subfr_length, nb_subfr, D, arch) \ + ((void)(arch), silk_burg_modified_c(res_nrg, res_nrg_Q, A_Q16, x, minInvGain_Q30, subfr_length, nb_subfr, D, arch)) +#endif + +#if !defined(OVERRIDE_silk_inner_prod16) +#define silk_inner_prod16(inVec1, inVec2, len, arch) \ + ((void)(arch),silk_inner_prod16_c(inVec1, inVec2, len)) +#endif + +#include "Inlines.h" +#include "MacroCount.h" +#include "MacroDebug.h" + +#ifdef OPUS_ARM_INLINE_ASM +#include "arm/SigProc_FIX_armv4.h" +#endif + +#ifdef OPUS_ARM_INLINE_EDSP +#include "arm/SigProc_FIX_armv5e.h" +#endif + +#if defined(MIPSr1_ASM) +#include "mips/sigproc_fix_mipsr1.h" +#endif + + +#ifdef __cplusplus +} +#endif + +#endif /* SILK_SIGPROC_FIX_H */ diff --git a/media/libopus/silk/VAD.c b/media/libopus/silk/VAD.c new file mode 100644 index 0000000000..d0cda52162 --- /dev/null +++ b/media/libopus/silk/VAD.c @@ -0,0 +1,360 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" +#include "stack_alloc.h" + +/* Silk VAD noise level estimation */ +# if !defined(OPUS_X86_MAY_HAVE_SSE4_1) +static OPUS_INLINE void silk_VAD_GetNoiseLevels( + const opus_int32 pX[ VAD_N_BANDS ], /* I subband energies */ + silk_VAD_state *psSilk_VAD /* I/O Pointer to Silk VAD state */ +); +#endif + +/**********************************/ +/* Initialization of the Silk VAD */ +/**********************************/ +opus_int silk_VAD_Init( /* O Return value, 0 if success */ + silk_VAD_state *psSilk_VAD /* I/O Pointer to Silk VAD state */ +) +{ + opus_int b, ret = 0; + + /* reset state memory */ + silk_memset( psSilk_VAD, 0, sizeof( silk_VAD_state ) ); + + /* init noise levels */ + /* Initialize array with approx pink noise levels (psd proportional to inverse of frequency) */ + for( b = 0; b < VAD_N_BANDS; b++ ) { + psSilk_VAD->NoiseLevelBias[ b ] = silk_max_32( silk_DIV32_16( VAD_NOISE_LEVELS_BIAS, b + 1 ), 1 ); + } + + /* Initialize state */ + for( b = 0; b < VAD_N_BANDS; b++ ) { + psSilk_VAD->NL[ b ] = silk_MUL( 100, psSilk_VAD->NoiseLevelBias[ b ] ); + psSilk_VAD->inv_NL[ b ] = silk_DIV32( silk_int32_MAX, psSilk_VAD->NL[ b ] ); + } + psSilk_VAD->counter = 15; + + /* init smoothed energy-to-noise ratio*/ + for( b = 0; b < VAD_N_BANDS; b++ ) { + psSilk_VAD->NrgRatioSmth_Q8[ b ] = 100 * 256; /* 100 * 256 --> 20 dB SNR */ + } + + return( ret ); +} + +/* Weighting factors for tilt measure */ +static const opus_int32 tiltWeights[ VAD_N_BANDS ] = { 30000, 6000, -12000, -12000 }; + +/***************************************/ +/* Get the speech activity level in Q8 */ +/***************************************/ +opus_int silk_VAD_GetSA_Q8_c( /* O Return value, 0 if success */ + silk_encoder_state *psEncC, /* I/O Encoder state */ + const opus_int16 pIn[] /* I PCM input */ +) +{ + opus_int SA_Q15, pSNR_dB_Q7, input_tilt; + opus_int decimated_framelength1, decimated_framelength2; + opus_int decimated_framelength; + opus_int dec_subframe_length, dec_subframe_offset, SNR_Q7, i, b, s; + opus_int32 sumSquared, smooth_coef_Q16; + opus_int16 HPstateTmp; + VARDECL( opus_int16, X ); + opus_int32 Xnrg[ VAD_N_BANDS ]; + opus_int32 NrgToNoiseRatio_Q8[ VAD_N_BANDS ]; + opus_int32 speech_nrg, x_tmp; + opus_int X_offset[ VAD_N_BANDS ]; + opus_int ret = 0; + silk_VAD_state *psSilk_VAD = &psEncC->sVAD; + SAVE_STACK; + + /* Safety checks */ + silk_assert( VAD_N_BANDS == 4 ); + celt_assert( MAX_FRAME_LENGTH >= psEncC->frame_length ); + celt_assert( psEncC->frame_length <= 512 ); + celt_assert( psEncC->frame_length == 8 * silk_RSHIFT( psEncC->frame_length, 3 ) ); + + /***********************/ + /* Filter and Decimate */ + /***********************/ + decimated_framelength1 = silk_RSHIFT( psEncC->frame_length, 1 ); + decimated_framelength2 = silk_RSHIFT( psEncC->frame_length, 2 ); + decimated_framelength = silk_RSHIFT( psEncC->frame_length, 3 ); + /* Decimate into 4 bands: + 0 L 3L L 3L 5L + - -- - -- -- + 8 8 2 4 4 + + [0-1 kHz| temp. |1-2 kHz| 2-4 kHz | 4-8 kHz | + + They're arranged to allow the minimal ( frame_length / 4 ) extra + scratch space during the downsampling process */ + X_offset[ 0 ] = 0; + X_offset[ 1 ] = decimated_framelength + decimated_framelength2; + X_offset[ 2 ] = X_offset[ 1 ] + decimated_framelength; + X_offset[ 3 ] = X_offset[ 2 ] + decimated_framelength2; + ALLOC( X, X_offset[ 3 ] + decimated_framelength1, opus_int16 ); + + /* 0-8 kHz to 0-4 kHz and 4-8 kHz */ + silk_ana_filt_bank_1( pIn, &psSilk_VAD->AnaState[ 0 ], + X, &X[ X_offset[ 3 ] ], psEncC->frame_length ); + + /* 0-4 kHz to 0-2 kHz and 2-4 kHz */ + silk_ana_filt_bank_1( X, &psSilk_VAD->AnaState1[ 0 ], + X, &X[ X_offset[ 2 ] ], decimated_framelength1 ); + + /* 0-2 kHz to 0-1 kHz and 1-2 kHz */ + silk_ana_filt_bank_1( X, &psSilk_VAD->AnaState2[ 0 ], + X, &X[ X_offset[ 1 ] ], decimated_framelength2 ); + + /*********************************************/ + /* HP filter on lowest band (differentiator) */ + /*********************************************/ + X[ decimated_framelength - 1 ] = silk_RSHIFT( X[ decimated_framelength - 1 ], 1 ); + HPstateTmp = X[ decimated_framelength - 1 ]; + for( i = decimated_framelength - 1; i > 0; i-- ) { + X[ i - 1 ] = silk_RSHIFT( X[ i - 1 ], 1 ); + X[ i ] -= X[ i - 1 ]; + } + X[ 0 ] -= psSilk_VAD->HPstate; + psSilk_VAD->HPstate = HPstateTmp; + + /*************************************/ + /* Calculate the energy in each band */ + /*************************************/ + for( b = 0; b < VAD_N_BANDS; b++ ) { + /* Find the decimated framelength in the non-uniformly divided bands */ + decimated_framelength = silk_RSHIFT( psEncC->frame_length, silk_min_int( VAD_N_BANDS - b, VAD_N_BANDS - 1 ) ); + + /* Split length into subframe lengths */ + dec_subframe_length = silk_RSHIFT( decimated_framelength, VAD_INTERNAL_SUBFRAMES_LOG2 ); + dec_subframe_offset = 0; + + /* Compute energy per sub-frame */ + /* initialize with summed energy of last subframe */ + Xnrg[ b ] = psSilk_VAD->XnrgSubfr[ b ]; + for( s = 0; s < VAD_INTERNAL_SUBFRAMES; s++ ) { + sumSquared = 0; + for( i = 0; i < dec_subframe_length; i++ ) { + /* The energy will be less than dec_subframe_length * ( silk_int16_MIN / 8 ) ^ 2. */ + /* Therefore we can accumulate with no risk of overflow (unless dec_subframe_length > 128) */ + x_tmp = silk_RSHIFT( + X[ X_offset[ b ] + i + dec_subframe_offset ], 3 ); + sumSquared = silk_SMLABB( sumSquared, x_tmp, x_tmp ); + + /* Safety check */ + silk_assert( sumSquared >= 0 ); + } + + /* Add/saturate summed energy of current subframe */ + if( s < VAD_INTERNAL_SUBFRAMES - 1 ) { + Xnrg[ b ] = silk_ADD_POS_SAT32( Xnrg[ b ], sumSquared ); + } else { + /* Look-ahead subframe */ + Xnrg[ b ] = silk_ADD_POS_SAT32( Xnrg[ b ], silk_RSHIFT( sumSquared, 1 ) ); + } + + dec_subframe_offset += dec_subframe_length; + } + psSilk_VAD->XnrgSubfr[ b ] = sumSquared; + } + + /********************/ + /* Noise estimation */ + /********************/ + silk_VAD_GetNoiseLevels( &Xnrg[ 0 ], psSilk_VAD ); + + /***********************************************/ + /* Signal-plus-noise to noise ratio estimation */ + /***********************************************/ + sumSquared = 0; + input_tilt = 0; + for( b = 0; b < VAD_N_BANDS; b++ ) { + speech_nrg = Xnrg[ b ] - psSilk_VAD->NL[ b ]; + if( speech_nrg > 0 ) { + /* Divide, with sufficient resolution */ + if( ( Xnrg[ b ] & 0xFF800000 ) == 0 ) { + NrgToNoiseRatio_Q8[ b ] = silk_DIV32( silk_LSHIFT( Xnrg[ b ], 8 ), psSilk_VAD->NL[ b ] + 1 ); + } else { + NrgToNoiseRatio_Q8[ b ] = silk_DIV32( Xnrg[ b ], silk_RSHIFT( psSilk_VAD->NL[ b ], 8 ) + 1 ); + } + + /* Convert to log domain */ + SNR_Q7 = silk_lin2log( NrgToNoiseRatio_Q8[ b ] ) - 8 * 128; + + /* Sum-of-squares */ + sumSquared = silk_SMLABB( sumSquared, SNR_Q7, SNR_Q7 ); /* Q14 */ + + /* Tilt measure */ + if( speech_nrg < ( (opus_int32)1 << 20 ) ) { + /* Scale down SNR value for small subband speech energies */ + SNR_Q7 = silk_SMULWB( silk_LSHIFT( silk_SQRT_APPROX( speech_nrg ), 6 ), SNR_Q7 ); + } + input_tilt = silk_SMLAWB( input_tilt, tiltWeights[ b ], SNR_Q7 ); + } else { + NrgToNoiseRatio_Q8[ b ] = 256; + } + } + + /* Mean-of-squares */ + sumSquared = silk_DIV32_16( sumSquared, VAD_N_BANDS ); /* Q14 */ + + /* Root-mean-square approximation, scale to dBs, and write to output pointer */ + pSNR_dB_Q7 = (opus_int16)( 3 * silk_SQRT_APPROX( sumSquared ) ); /* Q7 */ + + /*********************************/ + /* Speech Probability Estimation */ + /*********************************/ + SA_Q15 = silk_sigm_Q15( silk_SMULWB( VAD_SNR_FACTOR_Q16, pSNR_dB_Q7 ) - VAD_NEGATIVE_OFFSET_Q5 ); + + /**************************/ + /* Frequency Tilt Measure */ + /**************************/ + psEncC->input_tilt_Q15 = silk_LSHIFT( silk_sigm_Q15( input_tilt ) - 16384, 1 ); + + /**************************************************/ + /* Scale the sigmoid output based on power levels */ + /**************************************************/ + speech_nrg = 0; + for( b = 0; b < VAD_N_BANDS; b++ ) { + /* Accumulate signal-without-noise energies, higher frequency bands have more weight */ + speech_nrg += ( b + 1 ) * silk_RSHIFT( Xnrg[ b ] - psSilk_VAD->NL[ b ], 4 ); + } + + if( psEncC->frame_length == 20 * psEncC->fs_kHz ) { + speech_nrg = silk_RSHIFT32( speech_nrg, 1 ); + } + /* Power scaling */ + if( speech_nrg <= 0 ) { + SA_Q15 = silk_RSHIFT( SA_Q15, 1 ); + } else if( speech_nrg < 16384 ) { + speech_nrg = silk_LSHIFT32( speech_nrg, 16 ); + + /* square-root */ + speech_nrg = silk_SQRT_APPROX( speech_nrg ); + SA_Q15 = silk_SMULWB( 32768 + speech_nrg, SA_Q15 ); + } + + /* Copy the resulting speech activity in Q8 */ + psEncC->speech_activity_Q8 = silk_min_int( silk_RSHIFT( SA_Q15, 7 ), silk_uint8_MAX ); + + /***********************************/ + /* Energy Level and SNR estimation */ + /***********************************/ + /* Smoothing coefficient */ + smooth_coef_Q16 = silk_SMULWB( VAD_SNR_SMOOTH_COEF_Q18, silk_SMULWB( (opus_int32)SA_Q15, SA_Q15 ) ); + + if( psEncC->frame_length == 10 * psEncC->fs_kHz ) { + smooth_coef_Q16 >>= 1; + } + + for( b = 0; b < VAD_N_BANDS; b++ ) { + /* compute smoothed energy-to-noise ratio per band */ + psSilk_VAD->NrgRatioSmth_Q8[ b ] = silk_SMLAWB( psSilk_VAD->NrgRatioSmth_Q8[ b ], + NrgToNoiseRatio_Q8[ b ] - psSilk_VAD->NrgRatioSmth_Q8[ b ], smooth_coef_Q16 ); + + /* signal to noise ratio in dB per band */ + SNR_Q7 = 3 * ( silk_lin2log( psSilk_VAD->NrgRatioSmth_Q8[b] ) - 8 * 128 ); + /* quality = sigmoid( 0.25 * ( SNR_dB - 16 ) ); */ + psEncC->input_quality_bands_Q15[ b ] = silk_sigm_Q15( silk_RSHIFT( SNR_Q7 - 16 * 128, 4 ) ); + } + + RESTORE_STACK; + return( ret ); +} + +/**************************/ +/* Noise level estimation */ +/**************************/ +# if !defined(OPUS_X86_MAY_HAVE_SSE4_1) +static OPUS_INLINE +#endif +void silk_VAD_GetNoiseLevels( + const opus_int32 pX[ VAD_N_BANDS ], /* I subband energies */ + silk_VAD_state *psSilk_VAD /* I/O Pointer to Silk VAD state */ +) +{ + opus_int k; + opus_int32 nl, nrg, inv_nrg; + opus_int coef, min_coef; + + /* Initially faster smoothing */ + if( psSilk_VAD->counter < 1000 ) { /* 1000 = 20 sec */ + min_coef = silk_DIV32_16( silk_int16_MAX, silk_RSHIFT( psSilk_VAD->counter, 4 ) + 1 ); + /* Increment frame counter */ + psSilk_VAD->counter++; + } else { + min_coef = 0; + } + + for( k = 0; k < VAD_N_BANDS; k++ ) { + /* Get old noise level estimate for current band */ + nl = psSilk_VAD->NL[ k ]; + silk_assert( nl >= 0 ); + + /* Add bias */ + nrg = silk_ADD_POS_SAT32( pX[ k ], psSilk_VAD->NoiseLevelBias[ k ] ); + silk_assert( nrg > 0 ); + + /* Invert energies */ + inv_nrg = silk_DIV32( silk_int32_MAX, nrg ); + silk_assert( inv_nrg >= 0 ); + + /* Less update when subband energy is high */ + if( nrg > silk_LSHIFT( nl, 3 ) ) { + coef = VAD_NOISE_LEVEL_SMOOTH_COEF_Q16 >> 3; + } else if( nrg < nl ) { + coef = VAD_NOISE_LEVEL_SMOOTH_COEF_Q16; + } else { + coef = silk_SMULWB( silk_SMULWW( inv_nrg, nl ), VAD_NOISE_LEVEL_SMOOTH_COEF_Q16 << 1 ); + } + + /* Initially faster smoothing */ + coef = silk_max_int( coef, min_coef ); + + /* Smooth inverse energies */ + psSilk_VAD->inv_NL[ k ] = silk_SMLAWB( psSilk_VAD->inv_NL[ k ], inv_nrg - psSilk_VAD->inv_NL[ k ], coef ); + silk_assert( psSilk_VAD->inv_NL[ k ] >= 0 ); + + /* Compute noise level by inverting again */ + nl = silk_DIV32( silk_int32_MAX, psSilk_VAD->inv_NL[ k ] ); + silk_assert( nl >= 0 ); + + /* Limit noise levels (guarantee 7 bits of head room) */ + nl = silk_min( nl, 0x00FFFFFF ); + + /* Store as part of state */ + psSilk_VAD->NL[ k ] = nl; + } +} diff --git a/media/libopus/silk/VQ_WMat_EC.c b/media/libopus/silk/VQ_WMat_EC.c new file mode 100644 index 0000000000..245a7e4b06 --- /dev/null +++ b/media/libopus/silk/VQ_WMat_EC.c @@ -0,0 +1,131 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" + +/* Entropy constrained matrix-weighted VQ, hard-coded to 5-element vectors, for a single input data vector */ +void silk_VQ_WMat_EC_c( + opus_int8 *ind, /* O index of best codebook vector */ + opus_int32 *res_nrg_Q15, /* O best residual energy */ + opus_int32 *rate_dist_Q8, /* O best total bitrate */ + opus_int *gain_Q7, /* O sum of absolute LTP coefficients */ + const opus_int32 *XX_Q17, /* I correlation matrix */ + const opus_int32 *xX_Q17, /* I correlation vector */ + const opus_int8 *cb_Q7, /* I codebook */ + const opus_uint8 *cb_gain_Q7, /* I codebook effective gain */ + const opus_uint8 *cl_Q5, /* I code length for each codebook vector */ + const opus_int subfr_len, /* I number of samples per subframe */ + const opus_int32 max_gain_Q7, /* I maximum sum of absolute LTP coefficients */ + const opus_int L /* I number of vectors in codebook */ +) +{ + opus_int k, gain_tmp_Q7; + const opus_int8 *cb_row_Q7; + opus_int32 neg_xX_Q24[ 5 ]; + opus_int32 sum1_Q15, sum2_Q24; + opus_int32 bits_res_Q8, bits_tot_Q8; + + /* Negate and convert to new Q domain */ + neg_xX_Q24[ 0 ] = -silk_LSHIFT32( xX_Q17[ 0 ], 7 ); + neg_xX_Q24[ 1 ] = -silk_LSHIFT32( xX_Q17[ 1 ], 7 ); + neg_xX_Q24[ 2 ] = -silk_LSHIFT32( xX_Q17[ 2 ], 7 ); + neg_xX_Q24[ 3 ] = -silk_LSHIFT32( xX_Q17[ 3 ], 7 ); + neg_xX_Q24[ 4 ] = -silk_LSHIFT32( xX_Q17[ 4 ], 7 ); + + /* Loop over codebook */ + *rate_dist_Q8 = silk_int32_MAX; + *res_nrg_Q15 = silk_int32_MAX; + cb_row_Q7 = cb_Q7; + /* If things go really bad, at least *ind is set to something safe. */ + *ind = 0; + for( k = 0; k < L; k++ ) { + opus_int32 penalty; + gain_tmp_Q7 = cb_gain_Q7[k]; + /* Weighted rate */ + /* Quantization error: 1 - 2 * xX * cb + cb' * XX * cb */ + sum1_Q15 = SILK_FIX_CONST( 1.001, 15 ); + + /* Penalty for too large gain */ + penalty = silk_LSHIFT32( silk_max( silk_SUB32( gain_tmp_Q7, max_gain_Q7 ), 0 ), 11 ); + + /* first row of XX_Q17 */ + sum2_Q24 = silk_MLA( neg_xX_Q24[ 0 ], XX_Q17[ 1 ], cb_row_Q7[ 1 ] ); + sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 2 ], cb_row_Q7[ 2 ] ); + sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 3 ], cb_row_Q7[ 3 ] ); + sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 4 ], cb_row_Q7[ 4 ] ); + sum2_Q24 = silk_LSHIFT32( sum2_Q24, 1 ); + sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 0 ], cb_row_Q7[ 0 ] ); + sum1_Q15 = silk_SMLAWB( sum1_Q15, sum2_Q24, cb_row_Q7[ 0 ] ); + + /* second row of XX_Q17 */ + sum2_Q24 = silk_MLA( neg_xX_Q24[ 1 ], XX_Q17[ 7 ], cb_row_Q7[ 2 ] ); + sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 8 ], cb_row_Q7[ 3 ] ); + sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 9 ], cb_row_Q7[ 4 ] ); + sum2_Q24 = silk_LSHIFT32( sum2_Q24, 1 ); + sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 6 ], cb_row_Q7[ 1 ] ); + sum1_Q15 = silk_SMLAWB( sum1_Q15, sum2_Q24, cb_row_Q7[ 1 ] ); + + /* third row of XX_Q17 */ + sum2_Q24 = silk_MLA( neg_xX_Q24[ 2 ], XX_Q17[ 13 ], cb_row_Q7[ 3 ] ); + sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 14 ], cb_row_Q7[ 4 ] ); + sum2_Q24 = silk_LSHIFT32( sum2_Q24, 1 ); + sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 12 ], cb_row_Q7[ 2 ] ); + sum1_Q15 = silk_SMLAWB( sum1_Q15, sum2_Q24, cb_row_Q7[ 2 ] ); + + /* fourth row of XX_Q17 */ + sum2_Q24 = silk_MLA( neg_xX_Q24[ 3 ], XX_Q17[ 19 ], cb_row_Q7[ 4 ] ); + sum2_Q24 = silk_LSHIFT32( sum2_Q24, 1 ); + sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 18 ], cb_row_Q7[ 3 ] ); + sum1_Q15 = silk_SMLAWB( sum1_Q15, sum2_Q24, cb_row_Q7[ 3 ] ); + + /* last row of XX_Q17 */ + sum2_Q24 = silk_LSHIFT32( neg_xX_Q24[ 4 ], 1 ); + sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 24 ], cb_row_Q7[ 4 ] ); + sum1_Q15 = silk_SMLAWB( sum1_Q15, sum2_Q24, cb_row_Q7[ 4 ] ); + + /* find best */ + if( sum1_Q15 >= 0 ) { + /* Translate residual energy to bits using high-rate assumption (6 dB ==> 1 bit/sample) */ + bits_res_Q8 = silk_SMULBB( subfr_len, silk_lin2log( sum1_Q15 + penalty) - (15 << 7) ); + /* In the following line we reduce the codelength component by half ("-1"); seems to slightly improve quality */ + bits_tot_Q8 = silk_ADD_LSHIFT32( bits_res_Q8, cl_Q5[ k ], 3-1 ); + if( bits_tot_Q8 <= *rate_dist_Q8 ) { + *rate_dist_Q8 = bits_tot_Q8; + *res_nrg_Q15 = sum1_Q15 + penalty; + *ind = (opus_int8)k; + *gain_Q7 = gain_tmp_Q7; + } + } + + /* Go to next cbk vector */ + cb_row_Q7 += LTP_ORDER; + } +} diff --git a/media/libopus/silk/ana_filt_bank_1.c b/media/libopus/silk/ana_filt_bank_1.c new file mode 100644 index 0000000000..24cfb03fdb --- /dev/null +++ b/media/libopus/silk/ana_filt_bank_1.c @@ -0,0 +1,74 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "SigProc_FIX.h" + +/* Coefficients for 2-band filter bank based on first-order allpass filters */ +static opus_int16 A_fb1_20 = 5394 << 1; +static opus_int16 A_fb1_21 = -24290; /* (opus_int16)(20623 << 1) */ + +/* Split signal into two decimated bands using first-order allpass filters */ +void silk_ana_filt_bank_1( + const opus_int16 *in, /* I Input signal [N] */ + opus_int32 *S, /* I/O State vector [2] */ + opus_int16 *outL, /* O Low band [N/2] */ + opus_int16 *outH, /* O High band [N/2] */ + const opus_int32 N /* I Number of input samples */ +) +{ + opus_int k, N2 = silk_RSHIFT( N, 1 ); + opus_int32 in32, X, Y, out_1, out_2; + + /* Internal variables and state are in Q10 format */ + for( k = 0; k < N2; k++ ) { + /* Convert to Q10 */ + in32 = silk_LSHIFT( (opus_int32)in[ 2 * k ], 10 ); + + /* All-pass section for even input sample */ + Y = silk_SUB32( in32, S[ 0 ] ); + X = silk_SMLAWB( Y, Y, A_fb1_21 ); + out_1 = silk_ADD32( S[ 0 ], X ); + S[ 0 ] = silk_ADD32( in32, X ); + + /* Convert to Q10 */ + in32 = silk_LSHIFT( (opus_int32)in[ 2 * k + 1 ], 10 ); + + /* All-pass section for odd input sample, and add to output of previous section */ + Y = silk_SUB32( in32, S[ 1 ] ); + X = silk_SMULWB( Y, A_fb1_20 ); + out_2 = silk_ADD32( S[ 1 ], X ); + S[ 1 ] = silk_ADD32( in32, X ); + + /* Add/subtract, convert back to int16 and store to output */ + outL[ k ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( silk_ADD32( out_2, out_1 ), 11 ) ); + outH[ k ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( silk_SUB32( out_2, out_1 ), 11 ) ); + } +} diff --git a/media/libopus/silk/arm/LPC_inv_pred_gain_arm.h b/media/libopus/silk/arm/LPC_inv_pred_gain_arm.h new file mode 100644 index 0000000000..9895b555c8 --- /dev/null +++ b/media/libopus/silk/arm/LPC_inv_pred_gain_arm.h @@ -0,0 +1,57 @@ +/*********************************************************************** +Copyright (c) 2017 Google Inc. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef SILK_LPC_INV_PRED_GAIN_ARM_H +# define SILK_LPC_INV_PRED_GAIN_ARM_H + +# include "celt/arm/armcpu.h" + +# if defined(OPUS_ARM_MAY_HAVE_NEON_INTR) +opus_int32 silk_LPC_inverse_pred_gain_neon( /* O Returns inverse prediction gain in energy domain, Q30 */ + const opus_int16 *A_Q12, /* I Prediction coefficients, Q12 [order] */ + const opus_int order /* I Prediction order */ +); + +# if !defined(OPUS_HAVE_RTCD) && defined(OPUS_ARM_PRESUME_NEON) +# define OVERRIDE_silk_LPC_inverse_pred_gain (1) +# define silk_LPC_inverse_pred_gain(A_Q12, order, arch) ((void)(arch), PRESUME_NEON(silk_LPC_inverse_pred_gain)(A_Q12, order)) +# endif +# endif + +# if !defined(OVERRIDE_silk_LPC_inverse_pred_gain) +/*Is run-time CPU detection enabled on this platform?*/ +# if defined(OPUS_HAVE_RTCD) && (defined(OPUS_ARM_MAY_HAVE_NEON_INTR) && !defined(OPUS_ARM_PRESUME_NEON_INTR)) +extern opus_int32 (*const SILK_LPC_INVERSE_PRED_GAIN_IMPL[OPUS_ARCHMASK+1])(const opus_int16 *A_Q12, const opus_int order); +# define OVERRIDE_silk_LPC_inverse_pred_gain (1) +# define silk_LPC_inverse_pred_gain(A_Q12, order, arch) ((*SILK_LPC_INVERSE_PRED_GAIN_IMPL[(arch)&OPUS_ARCHMASK])(A_Q12, order)) +# elif defined(OPUS_ARM_PRESUME_NEON_INTR) +# define OVERRIDE_silk_LPC_inverse_pred_gain (1) +# define silk_LPC_inverse_pred_gain(A_Q12, order, arch) ((void)(arch), silk_LPC_inverse_pred_gain_neon(A_Q12, order)) +# endif +# endif + +#endif /* end SILK_LPC_INV_PRED_GAIN_ARM_H */ diff --git a/media/libopus/silk/arm/LPC_inv_pred_gain_neon_intr.c b/media/libopus/silk/arm/LPC_inv_pred_gain_neon_intr.c new file mode 100644 index 0000000000..726e6667b4 --- /dev/null +++ b/media/libopus/silk/arm/LPC_inv_pred_gain_neon_intr.c @@ -0,0 +1,288 @@ +/*********************************************************************** +Copyright (c) 2017 Google Inc. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include <arm_neon.h> +#include "SigProc_FIX.h" +#include "define.h" + +#define QA 24 +#define A_LIMIT SILK_FIX_CONST( 0.99975, QA ) + +#define MUL32_FRAC_Q(a32, b32, Q) ((opus_int32)(silk_RSHIFT_ROUND64(silk_SMULL(a32, b32), Q))) + +/* The difficulty is how to judge a 64-bit signed integer tmp64 is 32-bit overflowed, + * since NEON has no 64-bit min, max or comparison instructions. + * A failed idea is to compare the results of vmovn(tmp64) and vqmovn(tmp64) whether they are equal or not. + * However, this idea fails when the tmp64 is something like 0xFFFFFFF980000000. + * Here we know that mult2Q >= 1, so the highest bit (bit 63, sign bit) of tmp64 must equal to bit 62. + * tmp64 was shifted left by 1 and we got tmp64'. If high_half(tmp64') != 0 and high_half(tmp64') != -1, + * then we know that bit 31 to bit 63 of tmp64 can not all be the sign bit, and therefore tmp64 is 32-bit overflowed. + * That is, we judge if tmp64' > 0x00000000FFFFFFFF, or tmp64' <= 0xFFFFFFFF00000000. + * We use narrowing shift right 31 bits to tmp32' to save data bandwidth and instructions. + * That is, we judge if tmp32' > 0x00000000, or tmp32' <= 0xFFFFFFFF. + */ + +/* Compute inverse of LPC prediction gain, and */ +/* test if LPC coefficients are stable (all poles within unit circle) */ +static OPUS_INLINE opus_int32 LPC_inverse_pred_gain_QA_neon( /* O Returns inverse prediction gain in energy domain, Q30 */ + opus_int32 A_QA[ SILK_MAX_ORDER_LPC ], /* I Prediction coefficients */ + const opus_int order /* I Prediction order */ +) +{ + opus_int k, n, mult2Q; + opus_int32 invGain_Q30, rc_Q31, rc_mult1_Q30, rc_mult2, tmp1, tmp2; + opus_int32 max, min; + int32x4_t max_s32x4, min_s32x4; + int32x2_t max_s32x2, min_s32x2; + + max_s32x4 = vdupq_n_s32( silk_int32_MIN ); + min_s32x4 = vdupq_n_s32( silk_int32_MAX ); + invGain_Q30 = SILK_FIX_CONST( 1, 30 ); + for( k = order - 1; k > 0; k-- ) { + int32x2_t rc_Q31_s32x2, rc_mult2_s32x2; + int64x2_t mult2Q_s64x2; + + /* Check for stability */ + if( ( A_QA[ k ] > A_LIMIT ) || ( A_QA[ k ] < -A_LIMIT ) ) { + return 0; + } + + /* Set RC equal to negated AR coef */ + rc_Q31 = -silk_LSHIFT( A_QA[ k ], 31 - QA ); + + /* rc_mult1_Q30 range: [ 1 : 2^30 ] */ + rc_mult1_Q30 = silk_SUB32( SILK_FIX_CONST( 1, 30 ), silk_SMMUL( rc_Q31, rc_Q31 ) ); + silk_assert( rc_mult1_Q30 > ( 1 << 15 ) ); /* reduce A_LIMIT if fails */ + silk_assert( rc_mult1_Q30 <= ( 1 << 30 ) ); + + /* Update inverse gain */ + /* invGain_Q30 range: [ 0 : 2^30 ] */ + invGain_Q30 = silk_LSHIFT( silk_SMMUL( invGain_Q30, rc_mult1_Q30 ), 2 ); + silk_assert( invGain_Q30 >= 0 ); + silk_assert( invGain_Q30 <= ( 1 << 30 ) ); + if( invGain_Q30 < SILK_FIX_CONST( 1.0f / MAX_PREDICTION_POWER_GAIN, 30 ) ) { + return 0; + } + + /* rc_mult2 range: [ 2^30 : silk_int32_MAX ] */ + mult2Q = 32 - silk_CLZ32( silk_abs( rc_mult1_Q30 ) ); + rc_mult2 = silk_INVERSE32_varQ( rc_mult1_Q30, mult2Q + 30 ); + + /* Update AR coefficient */ + rc_Q31_s32x2 = vdup_n_s32( rc_Q31 ); + mult2Q_s64x2 = vdupq_n_s64( -mult2Q ); + rc_mult2_s32x2 = vdup_n_s32( rc_mult2 ); + + for( n = 0; n < ( ( k + 1 ) >> 1 ) - 3; n += 4 ) { + /* We always calculate extra elements of A_QA buffer when ( k % 4 ) != 0, to take the advantage of SIMD parallelization. */ + int32x4_t tmp1_s32x4, tmp2_s32x4, t0_s32x4, t1_s32x4, s0_s32x4, s1_s32x4, t_QA0_s32x4, t_QA1_s32x4; + int64x2_t t0_s64x2, t1_s64x2, t2_s64x2, t3_s64x2; + tmp1_s32x4 = vld1q_s32( A_QA + n ); + tmp2_s32x4 = vld1q_s32( A_QA + k - n - 4 ); + tmp2_s32x4 = vrev64q_s32( tmp2_s32x4 ); + tmp2_s32x4 = vcombine_s32( vget_high_s32( tmp2_s32x4 ), vget_low_s32( tmp2_s32x4 ) ); + t0_s32x4 = vqrdmulhq_lane_s32( tmp2_s32x4, rc_Q31_s32x2, 0 ); + t1_s32x4 = vqrdmulhq_lane_s32( tmp1_s32x4, rc_Q31_s32x2, 0 ); + t_QA0_s32x4 = vqsubq_s32( tmp1_s32x4, t0_s32x4 ); + t_QA1_s32x4 = vqsubq_s32( tmp2_s32x4, t1_s32x4 ); + t0_s64x2 = vmull_s32( vget_low_s32 ( t_QA0_s32x4 ), rc_mult2_s32x2 ); + t1_s64x2 = vmull_s32( vget_high_s32( t_QA0_s32x4 ), rc_mult2_s32x2 ); + t2_s64x2 = vmull_s32( vget_low_s32 ( t_QA1_s32x4 ), rc_mult2_s32x2 ); + t3_s64x2 = vmull_s32( vget_high_s32( t_QA1_s32x4 ), rc_mult2_s32x2 ); + t0_s64x2 = vrshlq_s64( t0_s64x2, mult2Q_s64x2 ); + t1_s64x2 = vrshlq_s64( t1_s64x2, mult2Q_s64x2 ); + t2_s64x2 = vrshlq_s64( t2_s64x2, mult2Q_s64x2 ); + t3_s64x2 = vrshlq_s64( t3_s64x2, mult2Q_s64x2 ); + t0_s32x4 = vcombine_s32( vmovn_s64( t0_s64x2 ), vmovn_s64( t1_s64x2 ) ); + t1_s32x4 = vcombine_s32( vmovn_s64( t2_s64x2 ), vmovn_s64( t3_s64x2 ) ); + s0_s32x4 = vcombine_s32( vshrn_n_s64( t0_s64x2, 31 ), vshrn_n_s64( t1_s64x2, 31 ) ); + s1_s32x4 = vcombine_s32( vshrn_n_s64( t2_s64x2, 31 ), vshrn_n_s64( t3_s64x2, 31 ) ); + max_s32x4 = vmaxq_s32( max_s32x4, s0_s32x4 ); + min_s32x4 = vminq_s32( min_s32x4, s0_s32x4 ); + max_s32x4 = vmaxq_s32( max_s32x4, s1_s32x4 ); + min_s32x4 = vminq_s32( min_s32x4, s1_s32x4 ); + t1_s32x4 = vrev64q_s32( t1_s32x4 ); + t1_s32x4 = vcombine_s32( vget_high_s32( t1_s32x4 ), vget_low_s32( t1_s32x4 ) ); + vst1q_s32( A_QA + n, t0_s32x4 ); + vst1q_s32( A_QA + k - n - 4, t1_s32x4 ); + } + for( ; n < (k + 1) >> 1; n++ ) { + opus_int64 tmp64; + tmp1 = A_QA[ n ]; + tmp2 = A_QA[ k - n - 1 ]; + tmp64 = silk_RSHIFT_ROUND64( silk_SMULL( silk_SUB_SAT32(tmp1, + MUL32_FRAC_Q( tmp2, rc_Q31, 31 ) ), rc_mult2 ), mult2Q); + if( tmp64 > silk_int32_MAX || tmp64 < silk_int32_MIN ) { + return 0; + } + A_QA[ n ] = ( opus_int32 )tmp64; + tmp64 = silk_RSHIFT_ROUND64( silk_SMULL( silk_SUB_SAT32(tmp2, + MUL32_FRAC_Q( tmp1, rc_Q31, 31 ) ), rc_mult2), mult2Q); + if( tmp64 > silk_int32_MAX || tmp64 < silk_int32_MIN ) { + return 0; + } + A_QA[ k - n - 1 ] = ( opus_int32 )tmp64; + } + } + + /* Check for stability */ + if( ( A_QA[ k ] > A_LIMIT ) || ( A_QA[ k ] < -A_LIMIT ) ) { + return 0; + } + + max_s32x2 = vmax_s32( vget_low_s32( max_s32x4 ), vget_high_s32( max_s32x4 ) ); + min_s32x2 = vmin_s32( vget_low_s32( min_s32x4 ), vget_high_s32( min_s32x4 ) ); + max_s32x2 = vmax_s32( max_s32x2, vreinterpret_s32_s64( vshr_n_s64( vreinterpret_s64_s32( max_s32x2 ), 32 ) ) ); + min_s32x2 = vmin_s32( min_s32x2, vreinterpret_s32_s64( vshr_n_s64( vreinterpret_s64_s32( min_s32x2 ), 32 ) ) ); + max = vget_lane_s32( max_s32x2, 0 ); + min = vget_lane_s32( min_s32x2, 0 ); + if( ( max > 0 ) || ( min < -1 ) ) { + return 0; + } + + /* Set RC equal to negated AR coef */ + rc_Q31 = -silk_LSHIFT( A_QA[ 0 ], 31 - QA ); + + /* Range: [ 1 : 2^30 ] */ + rc_mult1_Q30 = silk_SUB32( SILK_FIX_CONST( 1, 30 ), silk_SMMUL( rc_Q31, rc_Q31 ) ); + + /* Update inverse gain */ + /* Range: [ 0 : 2^30 ] */ + invGain_Q30 = silk_LSHIFT( silk_SMMUL( invGain_Q30, rc_mult1_Q30 ), 2 ); + silk_assert( invGain_Q30 >= 0 ); + silk_assert( invGain_Q30 <= ( 1 << 30 ) ); + if( invGain_Q30 < SILK_FIX_CONST( 1.0f / MAX_PREDICTION_POWER_GAIN, 30 ) ) { + return 0; + } + + return invGain_Q30; +} + +/* For input in Q12 domain */ +opus_int32 silk_LPC_inverse_pred_gain_neon( /* O Returns inverse prediction gain in energy domain, Q30 */ + const opus_int16 *A_Q12, /* I Prediction coefficients, Q12 [order] */ + const opus_int order /* I Prediction order */ +) +{ +#ifdef OPUS_CHECK_ASM + const opus_int32 invGain_Q30_c = silk_LPC_inverse_pred_gain_c( A_Q12, order ); +#endif + + opus_int32 invGain_Q30; + if( ( SILK_MAX_ORDER_LPC != 24 ) || ( order & 1 )) { + invGain_Q30 = silk_LPC_inverse_pred_gain_c( A_Q12, order ); + } + else { + opus_int32 Atmp_QA[ SILK_MAX_ORDER_LPC ]; + opus_int32 DC_resp; + int16x8_t t0_s16x8, t1_s16x8, t2_s16x8; + int32x4_t t0_s32x4; + const opus_int leftover = order & 7; + + /* Increase Q domain of the AR coefficients */ + t0_s16x8 = vld1q_s16( A_Q12 + 0 ); + t1_s16x8 = vld1q_s16( A_Q12 + 8 ); + if ( order > 16 ) { + t2_s16x8 = vld1q_s16( A_Q12 + 16 ); + } + t0_s32x4 = vpaddlq_s16( t0_s16x8 ); + + switch( order - leftover ) + { + case 24: + t0_s32x4 = vpadalq_s16( t0_s32x4, t2_s16x8 ); + vst1q_s32( Atmp_QA + 16, vshll_n_s16( vget_low_s16 ( t2_s16x8 ), QA - 12 ) ); + vst1q_s32( Atmp_QA + 20, vshll_n_s16( vget_high_s16( t2_s16x8 ), QA - 12 ) ); + /* FALLTHROUGH */ + + case 16: + t0_s32x4 = vpadalq_s16( t0_s32x4, t1_s16x8 ); + vst1q_s32( Atmp_QA + 8, vshll_n_s16( vget_low_s16 ( t1_s16x8 ), QA - 12 ) ); + vst1q_s32( Atmp_QA + 12, vshll_n_s16( vget_high_s16( t1_s16x8 ), QA - 12 ) ); + /* FALLTHROUGH */ + + case 8: + { + const int32x2_t t_s32x2 = vpadd_s32( vget_low_s32( t0_s32x4 ), vget_high_s32( t0_s32x4 ) ); + const int64x1_t t_s64x1 = vpaddl_s32( t_s32x2 ); + DC_resp = vget_lane_s32( vreinterpret_s32_s64( t_s64x1 ), 0 ); + vst1q_s32( Atmp_QA + 0, vshll_n_s16( vget_low_s16 ( t0_s16x8 ), QA - 12 ) ); + vst1q_s32( Atmp_QA + 4, vshll_n_s16( vget_high_s16( t0_s16x8 ), QA - 12 ) ); + } + break; + + default: + DC_resp = 0; + break; + } + A_Q12 += order - leftover; + + switch( leftover ) + { + case 6: + DC_resp += (opus_int32)A_Q12[ 5 ]; + DC_resp += (opus_int32)A_Q12[ 4 ]; + Atmp_QA[ order - leftover + 5 ] = silk_LSHIFT32( (opus_int32)A_Q12[ 5 ], QA - 12 ); + Atmp_QA[ order - leftover + 4 ] = silk_LSHIFT32( (opus_int32)A_Q12[ 4 ], QA - 12 ); + /* FALLTHROUGH */ + + case 4: + DC_resp += (opus_int32)A_Q12[ 3 ]; + DC_resp += (opus_int32)A_Q12[ 2 ]; + Atmp_QA[ order - leftover + 3 ] = silk_LSHIFT32( (opus_int32)A_Q12[ 3 ], QA - 12 ); + Atmp_QA[ order - leftover + 2 ] = silk_LSHIFT32( (opus_int32)A_Q12[ 2 ], QA - 12 ); + /* FALLTHROUGH */ + + case 2: + DC_resp += (opus_int32)A_Q12[ 1 ]; + DC_resp += (opus_int32)A_Q12[ 0 ]; + Atmp_QA[ order - leftover + 1 ] = silk_LSHIFT32( (opus_int32)A_Q12[ 1 ], QA - 12 ); + Atmp_QA[ order - leftover + 0 ] = silk_LSHIFT32( (opus_int32)A_Q12[ 0 ], QA - 12 ); + /* FALLTHROUGH */ + + default: + break; + } + + /* If the DC is unstable, we don't even need to do the full calculations */ + if( DC_resp >= 4096 ) { + invGain_Q30 = 0; + } else { + invGain_Q30 = LPC_inverse_pred_gain_QA_neon( Atmp_QA, order ); + } + } + +#ifdef OPUS_CHECK_ASM + silk_assert( invGain_Q30_c == invGain_Q30 ); +#endif + + return invGain_Q30; +} diff --git a/media/libopus/silk/arm/NSQ_del_dec_arm.h b/media/libopus/silk/arm/NSQ_del_dec_arm.h new file mode 100644 index 0000000000..9e76e16927 --- /dev/null +++ b/media/libopus/silk/arm/NSQ_del_dec_arm.h @@ -0,0 +1,100 @@ +/*********************************************************************** +Copyright (c) 2017 Google Inc. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef SILK_NSQ_DEL_DEC_ARM_H +#define SILK_NSQ_DEL_DEC_ARM_H + +#include "celt/arm/armcpu.h" + +#if defined(OPUS_ARM_MAY_HAVE_NEON_INTR) +void silk_NSQ_del_dec_neon( + const silk_encoder_state *psEncC, silk_nsq_state *NSQ, + SideInfoIndices *psIndices, const opus_int16 x16[], opus_int8 pulses[], + const opus_int16 PredCoef_Q12[2 * MAX_LPC_ORDER], + const opus_int16 LTPCoef_Q14[LTP_ORDER * MAX_NB_SUBFR], + const opus_int16 AR_Q13[MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER], + const opus_int HarmShapeGain_Q14[MAX_NB_SUBFR], + const opus_int Tilt_Q14[MAX_NB_SUBFR], + const opus_int32 LF_shp_Q14[MAX_NB_SUBFR], + const opus_int32 Gains_Q16[MAX_NB_SUBFR], + const opus_int pitchL[MAX_NB_SUBFR], const opus_int Lambda_Q10, + const opus_int LTP_scale_Q14); + +#if !defined(OPUS_HAVE_RTCD) +#define OVERRIDE_silk_NSQ_del_dec (1) +#define silk_NSQ_del_dec(psEncC, NSQ, psIndices, x16, pulses, PredCoef_Q12, \ + LTPCoef_Q14, AR_Q13, HarmShapeGain_Q14, Tilt_Q14, \ + LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, \ + LTP_scale_Q14, arch) \ + ((void)(arch), \ + PRESUME_NEON(silk_NSQ_del_dec)( \ + psEncC, NSQ, psIndices, x16, pulses, PredCoef_Q12, LTPCoef_Q14, \ + AR_Q13, HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, \ + Lambda_Q10, LTP_scale_Q14)) +#endif +#endif + +#if !defined(OVERRIDE_silk_NSQ_del_dec) +/*Is run-time CPU detection enabled on this platform?*/ +#if defined(OPUS_HAVE_RTCD) && (defined(OPUS_ARM_MAY_HAVE_NEON_INTR) && \ + !defined(OPUS_ARM_PRESUME_NEON_INTR)) +extern void (*const SILK_NSQ_DEL_DEC_IMPL[OPUS_ARCHMASK + 1])( + const silk_encoder_state *psEncC, silk_nsq_state *NSQ, + SideInfoIndices *psIndices, const opus_int16 x16[], opus_int8 pulses[], + const opus_int16 PredCoef_Q12[2 * MAX_LPC_ORDER], + const opus_int16 LTPCoef_Q14[LTP_ORDER * MAX_NB_SUBFR], + const opus_int16 AR_Q13[MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER], + const opus_int HarmShapeGain_Q14[MAX_NB_SUBFR], + const opus_int Tilt_Q14[MAX_NB_SUBFR], + const opus_int32 LF_shp_Q14[MAX_NB_SUBFR], + const opus_int32 Gains_Q16[MAX_NB_SUBFR], + const opus_int pitchL[MAX_NB_SUBFR], const opus_int Lambda_Q10, + const opus_int LTP_scale_Q14); +#define OVERRIDE_silk_NSQ_del_dec (1) +#define silk_NSQ_del_dec(psEncC, NSQ, psIndices, x16, pulses, PredCoef_Q12, \ + LTPCoef_Q14, AR_Q13, HarmShapeGain_Q14, Tilt_Q14, \ + LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, \ + LTP_scale_Q14, arch) \ + ((*SILK_NSQ_DEL_DEC_IMPL[(arch)&OPUS_ARCHMASK])( \ + psEncC, NSQ, psIndices, x16, pulses, PredCoef_Q12, LTPCoef_Q14, \ + AR_Q13, HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, \ + Lambda_Q10, LTP_scale_Q14)) +#elif defined(OPUS_ARM_PRESUME_NEON_INTR) +#define OVERRIDE_silk_NSQ_del_dec (1) +#define silk_NSQ_del_dec(psEncC, NSQ, psIndices, x16, pulses, PredCoef_Q12, \ + LTPCoef_Q14, AR_Q13, HarmShapeGain_Q14, Tilt_Q14, \ + LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, \ + LTP_scale_Q14, arch) \ + ((void)(arch), \ + silk_NSQ_del_dec_neon(psEncC, NSQ, psIndices, x16, pulses, PredCoef_Q12, \ + LTPCoef_Q14, AR_Q13, HarmShapeGain_Q14, Tilt_Q14, \ + LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, \ + LTP_scale_Q14)) +#endif +#endif + +#endif /* end SILK_NSQ_DEL_DEC_ARM_H */ diff --git a/media/libopus/silk/arm/NSQ_del_dec_neon_intr.c b/media/libopus/silk/arm/NSQ_del_dec_neon_intr.c new file mode 100644 index 0000000000..212410f362 --- /dev/null +++ b/media/libopus/silk/arm/NSQ_del_dec_neon_intr.c @@ -0,0 +1,1124 @@ +/*********************************************************************** +Copyright (c) 2017 Google Inc. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include <arm_neon.h> +#ifdef OPUS_CHECK_ASM +# include <string.h> +#endif +#include "main.h" +#include "stack_alloc.h" + +/* NEON intrinsics optimization now can only parallelize up to 4 delay decision states. */ +/* If there are more states, C function is called, and this optimization must be expanded. */ +#define NEON_MAX_DEL_DEC_STATES 4 + +typedef struct { + opus_int32 sLPC_Q14[ MAX_SUB_FRAME_LENGTH + NSQ_LPC_BUF_LENGTH ][ NEON_MAX_DEL_DEC_STATES ]; + opus_int32 RandState[ DECISION_DELAY ][ NEON_MAX_DEL_DEC_STATES ]; + opus_int32 Q_Q10[ DECISION_DELAY ][ NEON_MAX_DEL_DEC_STATES ]; + opus_int32 Xq_Q14[ DECISION_DELAY ][ NEON_MAX_DEL_DEC_STATES ]; + opus_int32 Pred_Q15[ DECISION_DELAY ][ NEON_MAX_DEL_DEC_STATES ]; + opus_int32 Shape_Q14[ DECISION_DELAY ][ NEON_MAX_DEL_DEC_STATES ]; + opus_int32 sAR2_Q14[ MAX_SHAPE_LPC_ORDER ][ NEON_MAX_DEL_DEC_STATES ]; + opus_int32 LF_AR_Q14[ NEON_MAX_DEL_DEC_STATES ]; + opus_int32 Diff_Q14[ NEON_MAX_DEL_DEC_STATES ]; + opus_int32 Seed[ NEON_MAX_DEL_DEC_STATES ]; + opus_int32 SeedInit[ NEON_MAX_DEL_DEC_STATES ]; + opus_int32 RD_Q10[ NEON_MAX_DEL_DEC_STATES ]; +} NSQ_del_decs_struct; + +typedef struct { + opus_int32 Q_Q10[ NEON_MAX_DEL_DEC_STATES ]; + opus_int32 RD_Q10[ NEON_MAX_DEL_DEC_STATES ]; + opus_int32 xq_Q14[ NEON_MAX_DEL_DEC_STATES ]; + opus_int32 LF_AR_Q14[ NEON_MAX_DEL_DEC_STATES ]; + opus_int32 Diff_Q14[ NEON_MAX_DEL_DEC_STATES ]; + opus_int32 sLTP_shp_Q14[ NEON_MAX_DEL_DEC_STATES ]; + opus_int32 LPC_exc_Q14[ NEON_MAX_DEL_DEC_STATES ]; +} NSQ_samples_struct; + +static OPUS_INLINE void silk_nsq_del_dec_scale_states_neon( + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + NSQ_del_decs_struct psDelDec[], /* I/O Delayed decision states */ + const opus_int16 x16[], /* I Input */ + opus_int32 x_sc_Q10[], /* O Input scaled with 1/Gain in Q10 */ + const opus_int16 sLTP[], /* I Re-whitened LTP state in Q0 */ + opus_int32 sLTP_Q15[], /* O LTP state matching scaled input */ + opus_int subfr, /* I Subframe number */ + const opus_int LTP_scale_Q14, /* I LTP state scaling */ + const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lag */ + const opus_int signal_type, /* I Signal type */ + const opus_int decisionDelay /* I Decision delay */ +); + +/******************************************/ +/* Noise shape quantizer for one subframe */ +/******************************************/ +static OPUS_INLINE void silk_noise_shape_quantizer_del_dec_neon( + silk_nsq_state *NSQ, /* I/O NSQ state */ + NSQ_del_decs_struct psDelDec[], /* I/O Delayed decision states */ + opus_int signalType, /* I Signal type */ + const opus_int32 x_Q10[], /* I */ + opus_int8 pulses[], /* O */ + opus_int16 xq[], /* O */ + opus_int32 sLTP_Q15[], /* I/O LTP filter state */ + opus_int32 delayedGain_Q10[], /* I/O Gain delay buffer */ + const opus_int16 a_Q12[], /* I Short term prediction coefs */ + const opus_int16 b_Q14[], /* I Long term prediction coefs */ + const opus_int16 AR_shp_Q13[], /* I Noise shaping coefs */ + opus_int lag, /* I Pitch lag */ + opus_int32 HarmShapeFIRPacked_Q14, /* I */ + opus_int Tilt_Q14, /* I Spectral tilt */ + opus_int32 LF_shp_Q14, /* I */ + opus_int32 Gain_Q16, /* I */ + opus_int Lambda_Q10, /* I */ + opus_int offset_Q10, /* I */ + opus_int length, /* I Input length */ + opus_int subfr, /* I Subframe number */ + opus_int shapingLPCOrder, /* I Shaping LPC filter order */ + opus_int predictLPCOrder, /* I Prediction filter order */ + opus_int warping_Q16, /* I */ + opus_int nStatesDelayedDecision, /* I Number of states in decision tree */ + opus_int *smpl_buf_idx, /* I/O Index to newest samples in buffers */ + opus_int decisionDelay /* I */ +); + +static OPUS_INLINE void copy_winner_state_kernel( + const NSQ_del_decs_struct *psDelDec, + const opus_int offset, + const opus_int last_smple_idx, + const opus_int Winner_ind, + const int32x2_t gain_lo_s32x2, + const int32x2_t gain_hi_s32x2, + const int32x4_t shift_s32x4, + int32x4_t t0_s32x4, + int32x4_t t1_s32x4, + opus_int8 *const pulses, + opus_int16 *pxq, + silk_nsq_state *NSQ +) +{ + int16x8_t t_s16x8; + int32x4_t o0_s32x4, o1_s32x4; + + t0_s32x4 = vld1q_lane_s32( &psDelDec->Q_Q10[ last_smple_idx - 0 ][ Winner_ind ], t0_s32x4, 0 ); + t0_s32x4 = vld1q_lane_s32( &psDelDec->Q_Q10[ last_smple_idx - 1 ][ Winner_ind ], t0_s32x4, 1 ); + t0_s32x4 = vld1q_lane_s32( &psDelDec->Q_Q10[ last_smple_idx - 2 ][ Winner_ind ], t0_s32x4, 2 ); + t0_s32x4 = vld1q_lane_s32( &psDelDec->Q_Q10[ last_smple_idx - 3 ][ Winner_ind ], t0_s32x4, 3 ); + t1_s32x4 = vld1q_lane_s32( &psDelDec->Q_Q10[ last_smple_idx - 4 ][ Winner_ind ], t1_s32x4, 0 ); + t1_s32x4 = vld1q_lane_s32( &psDelDec->Q_Q10[ last_smple_idx - 5 ][ Winner_ind ], t1_s32x4, 1 ); + t1_s32x4 = vld1q_lane_s32( &psDelDec->Q_Q10[ last_smple_idx - 6 ][ Winner_ind ], t1_s32x4, 2 ); + t1_s32x4 = vld1q_lane_s32( &psDelDec->Q_Q10[ last_smple_idx - 7 ][ Winner_ind ], t1_s32x4, 3 ); + t_s16x8 = vcombine_s16( vrshrn_n_s32( t0_s32x4, 10 ), vrshrn_n_s32( t1_s32x4, 10 ) ); + vst1_s8( &pulses[ offset ], vmovn_s16( t_s16x8 ) ); + + t0_s32x4 = vld1q_lane_s32( &psDelDec->Xq_Q14[ last_smple_idx - 0 ][ Winner_ind ], t0_s32x4, 0 ); + t0_s32x4 = vld1q_lane_s32( &psDelDec->Xq_Q14[ last_smple_idx - 1 ][ Winner_ind ], t0_s32x4, 1 ); + t0_s32x4 = vld1q_lane_s32( &psDelDec->Xq_Q14[ last_smple_idx - 2 ][ Winner_ind ], t0_s32x4, 2 ); + t0_s32x4 = vld1q_lane_s32( &psDelDec->Xq_Q14[ last_smple_idx - 3 ][ Winner_ind ], t0_s32x4, 3 ); + t1_s32x4 = vld1q_lane_s32( &psDelDec->Xq_Q14[ last_smple_idx - 4 ][ Winner_ind ], t1_s32x4, 0 ); + t1_s32x4 = vld1q_lane_s32( &psDelDec->Xq_Q14[ last_smple_idx - 5 ][ Winner_ind ], t1_s32x4, 1 ); + t1_s32x4 = vld1q_lane_s32( &psDelDec->Xq_Q14[ last_smple_idx - 6 ][ Winner_ind ], t1_s32x4, 2 ); + t1_s32x4 = vld1q_lane_s32( &psDelDec->Xq_Q14[ last_smple_idx - 7 ][ Winner_ind ], t1_s32x4, 3 ); + o0_s32x4 = vqdmulhq_lane_s32( t0_s32x4, gain_lo_s32x2, 0 ); + o1_s32x4 = vqdmulhq_lane_s32( t1_s32x4, gain_lo_s32x2, 0 ); + o0_s32x4 = vmlaq_lane_s32( o0_s32x4, t0_s32x4, gain_hi_s32x2, 0 ); + o1_s32x4 = vmlaq_lane_s32( o1_s32x4, t1_s32x4, gain_hi_s32x2, 0 ); + o0_s32x4 = vrshlq_s32( o0_s32x4, shift_s32x4 ); + o1_s32x4 = vrshlq_s32( o1_s32x4, shift_s32x4 ); + vst1_s16( &pxq[ offset + 0 ], vqmovn_s32( o0_s32x4 ) ); + vst1_s16( &pxq[ offset + 4 ], vqmovn_s32( o1_s32x4 ) ); + + t0_s32x4 = vld1q_lane_s32( &psDelDec->Shape_Q14[ last_smple_idx - 0 ][ Winner_ind ], t0_s32x4, 0 ); + t0_s32x4 = vld1q_lane_s32( &psDelDec->Shape_Q14[ last_smple_idx - 1 ][ Winner_ind ], t0_s32x4, 1 ); + t0_s32x4 = vld1q_lane_s32( &psDelDec->Shape_Q14[ last_smple_idx - 2 ][ Winner_ind ], t0_s32x4, 2 ); + t0_s32x4 = vld1q_lane_s32( &psDelDec->Shape_Q14[ last_smple_idx - 3 ][ Winner_ind ], t0_s32x4, 3 ); + t1_s32x4 = vld1q_lane_s32( &psDelDec->Shape_Q14[ last_smple_idx - 4 ][ Winner_ind ], t1_s32x4, 0 ); + t1_s32x4 = vld1q_lane_s32( &psDelDec->Shape_Q14[ last_smple_idx - 5 ][ Winner_ind ], t1_s32x4, 1 ); + t1_s32x4 = vld1q_lane_s32( &psDelDec->Shape_Q14[ last_smple_idx - 6 ][ Winner_ind ], t1_s32x4, 2 ); + t1_s32x4 = vld1q_lane_s32( &psDelDec->Shape_Q14[ last_smple_idx - 7 ][ Winner_ind ], t1_s32x4, 3 ); + vst1q_s32( &NSQ->sLTP_shp_Q14[ NSQ->sLTP_shp_buf_idx + offset + 0 ], t0_s32x4 ); + vst1q_s32( &NSQ->sLTP_shp_Q14[ NSQ->sLTP_shp_buf_idx + offset + 4 ], t1_s32x4 ); +} + +static OPUS_INLINE void copy_winner_state( + const NSQ_del_decs_struct *psDelDec, + const opus_int decisionDelay, + const opus_int smpl_buf_idx, + const opus_int Winner_ind, + const opus_int32 gain, + const opus_int32 shift, + opus_int8 *const pulses, + opus_int16 *pxq, + silk_nsq_state *NSQ +) +{ + opus_int i, last_smple_idx; + const int32x2_t gain_lo_s32x2 = vdup_n_s32( silk_LSHIFT32( gain & 0x0000FFFF, 15 ) ); + const int32x2_t gain_hi_s32x2 = vdup_n_s32( gain >> 16 ); + const int32x4_t shift_s32x4 = vdupq_n_s32( -shift ); + int32x4_t t0_s32x4, t1_s32x4; + + t0_s32x4 = t1_s32x4 = vdupq_n_s32( 0 ); /* initialization */ + last_smple_idx = smpl_buf_idx + decisionDelay - 1 + DECISION_DELAY; + if( last_smple_idx >= DECISION_DELAY ) last_smple_idx -= DECISION_DELAY; + if( last_smple_idx >= DECISION_DELAY ) last_smple_idx -= DECISION_DELAY; + + for( i = 0; ( i < ( decisionDelay - 7 ) ) && ( last_smple_idx >= 7 ); i += 8, last_smple_idx -= 8 ) { + copy_winner_state_kernel( psDelDec, i - decisionDelay, last_smple_idx, Winner_ind, gain_lo_s32x2, gain_hi_s32x2, shift_s32x4, t0_s32x4, t1_s32x4, pulses, pxq, NSQ ); + } + for( ; ( i < decisionDelay ) && ( last_smple_idx >= 0 ); i++, last_smple_idx-- ) { + pulses[ i - decisionDelay ] = (opus_int8)silk_RSHIFT_ROUND( psDelDec->Q_Q10[ last_smple_idx ][ Winner_ind ], 10 ); + pxq[ i - decisionDelay ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( silk_SMULWW( psDelDec->Xq_Q14[ last_smple_idx ][ Winner_ind ], gain ), shift ) ); + NSQ->sLTP_shp_Q14[ NSQ->sLTP_shp_buf_idx - decisionDelay + i ] = psDelDec->Shape_Q14[ last_smple_idx ][ Winner_ind ]; + } + + last_smple_idx += DECISION_DELAY; + for( ; i < ( decisionDelay - 7 ); i++, last_smple_idx-- ) { + copy_winner_state_kernel( psDelDec, i - decisionDelay, last_smple_idx, Winner_ind, gain_lo_s32x2, gain_hi_s32x2, shift_s32x4, t0_s32x4, t1_s32x4, pulses, pxq, NSQ ); + } + for( ; i < decisionDelay; i++, last_smple_idx-- ) { + pulses[ i - decisionDelay ] = (opus_int8)silk_RSHIFT_ROUND( psDelDec->Q_Q10[ last_smple_idx ][ Winner_ind ], 10 ); + pxq[ i - decisionDelay ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( silk_SMULWW( psDelDec->Xq_Q14[ last_smple_idx ][ Winner_ind ], gain ), shift ) ); + NSQ->sLTP_shp_Q14[ NSQ->sLTP_shp_buf_idx - decisionDelay + i ] = psDelDec->Shape_Q14[ last_smple_idx ][ Winner_ind ]; + } +} + +void silk_NSQ_del_dec_neon( + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + SideInfoIndices *psIndices, /* I/O Quantization Indices */ + const opus_int16 x16[], /* I Input */ + opus_int8 pulses[], /* O Quantized pulse signal */ + const opus_int16 PredCoef_Q12[ 2 * MAX_LPC_ORDER ], /* I Short term prediction coefs */ + const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ + const opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ + const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ + const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ + const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ + const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ + const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ + const opus_int LTP_scale_Q14 /* I LTP state scaling */ +) +{ +#ifdef OPUS_CHECK_ASM + silk_nsq_state NSQ_c; + SideInfoIndices psIndices_c; + opus_int8 pulses_c[ MAX_FRAME_LENGTH ]; + const opus_int8 *const pulses_a = pulses; + + ( void )pulses_a; + silk_memcpy( &NSQ_c, NSQ, sizeof( NSQ_c ) ); + silk_memcpy( &psIndices_c, psIndices, sizeof( psIndices_c ) ); + silk_memcpy( pulses_c, pulses, sizeof( pulses_c ) ); + silk_NSQ_del_dec_c( psEncC, &NSQ_c, &psIndices_c, x16, pulses_c, PredCoef_Q12, LTPCoef_Q14, AR_Q13, HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, + pitchL, Lambda_Q10, LTP_scale_Q14 ); +#endif + + /* The optimization parallelizes the different delay decision states. */ + if(( psEncC->nStatesDelayedDecision > NEON_MAX_DEL_DEC_STATES ) || ( psEncC->nStatesDelayedDecision <= 2 )) { + /* NEON intrinsics optimization now can only parallelize up to 4 delay decision states. */ + /* If there are more states, C function is called, and this optimization must be expanded. */ + /* When the number of delay decision states is less than 3, there are penalties using this */ + /* optimization, and C function is called. */ + /* When the number of delay decision states is 2, it's better to specialize another */ + /* structure NSQ_del_dec2_struct and optimize with shorter NEON registers. (Low priority) */ + silk_NSQ_del_dec_c( psEncC, NSQ, psIndices, x16, pulses, PredCoef_Q12, LTPCoef_Q14, AR_Q13, HarmShapeGain_Q14, + Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14 ); + } else { + opus_int i, k, lag, start_idx, LSF_interpolation_flag, Winner_ind, subfr; + opus_int smpl_buf_idx, decisionDelay; + const opus_int16 *A_Q12, *B_Q14, *AR_shp_Q13; + opus_int16 *pxq; + VARDECL( opus_int32, sLTP_Q15 ); + VARDECL( opus_int16, sLTP ); + opus_int32 HarmShapeFIRPacked_Q14; + opus_int offset_Q10; + opus_int32 RDmin_Q10, Gain_Q10; + VARDECL( opus_int32, x_sc_Q10 ); + VARDECL( opus_int32, delayedGain_Q10 ); + VARDECL( NSQ_del_decs_struct, psDelDec ); + int32x4_t t_s32x4; + SAVE_STACK; + + /* Set unvoiced lag to the previous one, overwrite later for voiced */ + lag = NSQ->lagPrev; + + silk_assert( NSQ->prev_gain_Q16 != 0 ); + + /* Initialize delayed decision states */ + ALLOC( psDelDec, 1, NSQ_del_decs_struct ); + /* Only RandState and RD_Q10 need to be initialized to 0. */ + silk_memset( psDelDec->RandState, 0, sizeof( psDelDec->RandState ) ); + vst1q_s32( psDelDec->RD_Q10, vdupq_n_s32( 0 ) ); + + for( k = 0; k < psEncC->nStatesDelayedDecision; k++ ) { + psDelDec->SeedInit[ k ] = psDelDec->Seed[ k ] = ( k + psIndices->Seed ) & 3; + } + vst1q_s32( psDelDec->LF_AR_Q14, vld1q_dup_s32( &NSQ->sLF_AR_shp_Q14 ) ); + vst1q_s32( psDelDec->Diff_Q14, vld1q_dup_s32( &NSQ->sDiff_shp_Q14 ) ); + vst1q_s32( psDelDec->Shape_Q14[ 0 ], vld1q_dup_s32( &NSQ->sLTP_shp_Q14[ psEncC->ltp_mem_length - 1 ] ) ); + for( i = 0; i < NSQ_LPC_BUF_LENGTH; i++ ) { + vst1q_s32( psDelDec->sLPC_Q14[ i ], vld1q_dup_s32( &NSQ->sLPC_Q14[ i ] ) ); + } + for( i = 0; i < (opus_int)( sizeof( NSQ->sAR2_Q14 ) / sizeof( NSQ->sAR2_Q14[ 0 ] ) ); i++ ) { + vst1q_s32( psDelDec->sAR2_Q14[ i ], vld1q_dup_s32( &NSQ->sAR2_Q14[ i ] ) ); + } + + offset_Q10 = silk_Quantization_Offsets_Q10[ psIndices->signalType >> 1 ][ psIndices->quantOffsetType ]; + smpl_buf_idx = 0; /* index of oldest samples */ + + decisionDelay = silk_min_int( DECISION_DELAY, psEncC->subfr_length ); + + /* For voiced frames limit the decision delay to lower than the pitch lag */ + if( psIndices->signalType == TYPE_VOICED ) { + opus_int pitch_min = pitchL[ 0 ]; + for( k = 1; k < psEncC->nb_subfr; k++ ) { + pitch_min = silk_min_int( pitch_min, pitchL[ k ] ); + } + decisionDelay = silk_min_int( decisionDelay, pitch_min - LTP_ORDER / 2 - 1 ); + } else { + if( lag > 0 ) { + decisionDelay = silk_min_int( decisionDelay, lag - LTP_ORDER / 2 - 1 ); + } + } + + if( psIndices->NLSFInterpCoef_Q2 == 4 ) { + LSF_interpolation_flag = 0; + } else { + LSF_interpolation_flag = 1; + } + + ALLOC( sLTP_Q15, psEncC->ltp_mem_length + psEncC->frame_length, opus_int32 ); + ALLOC( sLTP, psEncC->ltp_mem_length + psEncC->frame_length, opus_int16 ); + ALLOC( x_sc_Q10, psEncC->subfr_length, opus_int32 ); + ALLOC( delayedGain_Q10, DECISION_DELAY, opus_int32 ); + /* Set up pointers to start of sub frame */ + pxq = &NSQ->xq[ psEncC->ltp_mem_length ]; + NSQ->sLTP_shp_buf_idx = psEncC->ltp_mem_length; + NSQ->sLTP_buf_idx = psEncC->ltp_mem_length; + subfr = 0; + for( k = 0; k < psEncC->nb_subfr; k++ ) { + A_Q12 = &PredCoef_Q12[ ( ( k >> 1 ) | ( 1 - LSF_interpolation_flag ) ) * MAX_LPC_ORDER ]; + B_Q14 = <PCoef_Q14[ k * LTP_ORDER ]; + AR_shp_Q13 = &AR_Q13[ k * MAX_SHAPE_LPC_ORDER ]; + + /* Noise shape parameters */ + silk_assert( HarmShapeGain_Q14[ k ] >= 0 ); + HarmShapeFIRPacked_Q14 = silk_RSHIFT( HarmShapeGain_Q14[ k ], 2 ); + HarmShapeFIRPacked_Q14 |= silk_LSHIFT( (opus_int32)silk_RSHIFT( HarmShapeGain_Q14[ k ], 1 ), 16 ); + + NSQ->rewhite_flag = 0; + if( psIndices->signalType == TYPE_VOICED ) { + /* Voiced */ + lag = pitchL[ k ]; + + /* Re-whitening */ + if( ( k & ( 3 - silk_LSHIFT( LSF_interpolation_flag, 1 ) ) ) == 0 ) { + if( k == 2 ) { + /* RESET DELAYED DECISIONS */ + /* Find winner */ + int32x4_t RD_Q10_s32x4; + RDmin_Q10 = psDelDec->RD_Q10[ 0 ]; + Winner_ind = 0; + for( i = 1; i < psEncC->nStatesDelayedDecision; i++ ) { + if( psDelDec->RD_Q10[ i ] < RDmin_Q10 ) { + RDmin_Q10 = psDelDec->RD_Q10[ i ]; + Winner_ind = i; + } + } + psDelDec->RD_Q10[ Winner_ind ] -= ( silk_int32_MAX >> 4 ); + RD_Q10_s32x4 = vld1q_s32( psDelDec->RD_Q10 ); + RD_Q10_s32x4 = vaddq_s32( RD_Q10_s32x4, vdupq_n_s32( silk_int32_MAX >> 4 ) ); + vst1q_s32( psDelDec->RD_Q10, RD_Q10_s32x4 ); + + /* Copy final part of signals from winner state to output and long-term filter states */ + copy_winner_state( psDelDec, decisionDelay, smpl_buf_idx, Winner_ind, Gains_Q16[ 1 ], 14, pulses, pxq, NSQ ); + + subfr = 0; + } + + /* Rewhiten with new A coefs */ + start_idx = psEncC->ltp_mem_length - lag - psEncC->predictLPCOrder - LTP_ORDER / 2; + silk_assert( start_idx > 0 ); + + silk_LPC_analysis_filter( &sLTP[ start_idx ], &NSQ->xq[ start_idx + k * psEncC->subfr_length ], + A_Q12, psEncC->ltp_mem_length - start_idx, psEncC->predictLPCOrder, psEncC->arch ); + + NSQ->sLTP_buf_idx = psEncC->ltp_mem_length; + NSQ->rewhite_flag = 1; + } + } + + silk_nsq_del_dec_scale_states_neon( psEncC, NSQ, psDelDec, x16, x_sc_Q10, sLTP, sLTP_Q15, k, + LTP_scale_Q14, Gains_Q16, pitchL, psIndices->signalType, decisionDelay ); + + silk_noise_shape_quantizer_del_dec_neon( NSQ, psDelDec, psIndices->signalType, x_sc_Q10, pulses, pxq, sLTP_Q15, + delayedGain_Q10, A_Q12, B_Q14, AR_shp_Q13, lag, HarmShapeFIRPacked_Q14, Tilt_Q14[ k ], LF_shp_Q14[ k ], + Gains_Q16[ k ], Lambda_Q10, offset_Q10, psEncC->subfr_length, subfr++, psEncC->shapingLPCOrder, + psEncC->predictLPCOrder, psEncC->warping_Q16, psEncC->nStatesDelayedDecision, &smpl_buf_idx, decisionDelay ); + + x16 += psEncC->subfr_length; + pulses += psEncC->subfr_length; + pxq += psEncC->subfr_length; + } + + /* Find winner */ + RDmin_Q10 = psDelDec->RD_Q10[ 0 ]; + Winner_ind = 0; + for( k = 1; k < psEncC->nStatesDelayedDecision; k++ ) { + if( psDelDec->RD_Q10[ k ] < RDmin_Q10 ) { + RDmin_Q10 = psDelDec->RD_Q10[ k ]; + Winner_ind = k; + } + } + + /* Copy final part of signals from winner state to output and long-term filter states */ + psIndices->Seed = psDelDec->SeedInit[ Winner_ind ]; + Gain_Q10 = silk_RSHIFT32( Gains_Q16[ psEncC->nb_subfr - 1 ], 6 ); + copy_winner_state( psDelDec, decisionDelay, smpl_buf_idx, Winner_ind, Gain_Q10, 8, pulses, pxq, NSQ ); + + t_s32x4 = vdupq_n_s32( 0 ); /* initialization */ + for( i = 0; i < ( NSQ_LPC_BUF_LENGTH - 3 ); i += 4 ) { + t_s32x4 = vld1q_lane_s32( &psDelDec->sLPC_Q14[ i + 0 ][ Winner_ind ], t_s32x4, 0 ); + t_s32x4 = vld1q_lane_s32( &psDelDec->sLPC_Q14[ i + 1 ][ Winner_ind ], t_s32x4, 1 ); + t_s32x4 = vld1q_lane_s32( &psDelDec->sLPC_Q14[ i + 2 ][ Winner_ind ], t_s32x4, 2 ); + t_s32x4 = vld1q_lane_s32( &psDelDec->sLPC_Q14[ i + 3 ][ Winner_ind ], t_s32x4, 3 ); + vst1q_s32( &NSQ->sLPC_Q14[ i ], t_s32x4 ); + } + + for( ; i < NSQ_LPC_BUF_LENGTH; i++ ) { + NSQ->sLPC_Q14[ i ] = psDelDec->sLPC_Q14[ i ][ Winner_ind ]; + } + + for( i = 0; i < (opus_int)( sizeof( NSQ->sAR2_Q14 ) / sizeof( NSQ->sAR2_Q14[ 0 ] ) - 3 ); i += 4 ) { + t_s32x4 = vld1q_lane_s32( &psDelDec->sAR2_Q14[ i + 0 ][ Winner_ind ], t_s32x4, 0 ); + t_s32x4 = vld1q_lane_s32( &psDelDec->sAR2_Q14[ i + 1 ][ Winner_ind ], t_s32x4, 1 ); + t_s32x4 = vld1q_lane_s32( &psDelDec->sAR2_Q14[ i + 2 ][ Winner_ind ], t_s32x4, 2 ); + t_s32x4 = vld1q_lane_s32( &psDelDec->sAR2_Q14[ i + 3 ][ Winner_ind ], t_s32x4, 3 ); + vst1q_s32( &NSQ->sAR2_Q14[ i ], t_s32x4 ); + } + + for( ; i < (opus_int)( sizeof( NSQ->sAR2_Q14 ) / sizeof( NSQ->sAR2_Q14[ 0 ] ) ); i++ ) { + NSQ->sAR2_Q14[ i ] = psDelDec->sAR2_Q14[ i ][ Winner_ind ]; + } + + /* Update states */ + NSQ->sLF_AR_shp_Q14 = psDelDec->LF_AR_Q14[ Winner_ind ]; + NSQ->sDiff_shp_Q14 = psDelDec->Diff_Q14[ Winner_ind ]; + NSQ->lagPrev = pitchL[ psEncC->nb_subfr - 1 ]; + + /* Save quantized speech signal */ + silk_memmove( NSQ->xq, &NSQ->xq[ psEncC->frame_length ], psEncC->ltp_mem_length * sizeof( opus_int16 ) ); + silk_memmove( NSQ->sLTP_shp_Q14, &NSQ->sLTP_shp_Q14[ psEncC->frame_length ], psEncC->ltp_mem_length * sizeof( opus_int32 ) ); + RESTORE_STACK; + } + +#ifdef OPUS_CHECK_ASM + silk_assert( !memcmp( &NSQ_c, NSQ, sizeof( NSQ_c ) ) ); + silk_assert( !memcmp( &psIndices_c, psIndices, sizeof( psIndices_c ) ) ); + silk_assert( !memcmp( pulses_c, pulses_a, sizeof( pulses_c ) ) ); +#endif +} + +/******************************************/ +/* Noise shape quantizer for one subframe */ +/******************************************/ +/* Note: Function silk_short_prediction_create_arch_coef_neon() defined in NSQ_neon.h is actually a hacking C function. */ +/* Therefore here we append "_local" to the NEON function name to avoid confusion. */ +static OPUS_INLINE void silk_short_prediction_create_arch_coef_neon_local(opus_int32 *out, const opus_int16 *in, opus_int order) +{ + int16x8_t t_s16x8; + int32x4_t t0_s32x4, t1_s32x4, t2_s32x4, t3_s32x4; + silk_assert( order == 10 || order == 16 ); + + t_s16x8 = vld1q_s16( in + 0 ); /* 7 6 5 4 3 2 1 0 */ + t_s16x8 = vrev64q_s16( t_s16x8 ); /* 4 5 6 7 0 1 2 3 */ + t2_s32x4 = vshll_n_s16( vget_high_s16( t_s16x8 ), 15 ); /* 4 5 6 7 */ + t3_s32x4 = vshll_n_s16( vget_low_s16( t_s16x8 ), 15 ); /* 0 1 2 3 */ + + if( order == 16 ) { + t_s16x8 = vld1q_s16( in + 8 ); /* F E D C B A 9 8 */ + t_s16x8 = vrev64q_s16( t_s16x8 ); /* C D E F 8 9 A B */ + t0_s32x4 = vshll_n_s16( vget_high_s16( t_s16x8 ), 15 ); /* C D E F */ + t1_s32x4 = vshll_n_s16( vget_low_s16( t_s16x8 ), 15 ); /* 8 9 A B */ + } else { + int16x4_t t_s16x4; + + t0_s32x4 = vdupq_n_s32( 0 ); /* zero zero zero zero */ + t_s16x4 = vld1_s16( in + 6 ); /* 9 8 7 6 */ + t_s16x4 = vrev64_s16( t_s16x4 ); /* 6 7 8 9 */ + t1_s32x4 = vshll_n_s16( t_s16x4, 15 ); + t1_s32x4 = vcombine_s32( vget_low_s32(t0_s32x4), vget_low_s32( t1_s32x4 ) ); /* 8 9 zero zero */ + } + vst1q_s32( out + 0, t0_s32x4 ); + vst1q_s32( out + 4, t1_s32x4 ); + vst1q_s32( out + 8, t2_s32x4 ); + vst1q_s32( out + 12, t3_s32x4 ); +} + +static OPUS_INLINE int32x4_t silk_SMLAWB_lane0_neon( + const int32x4_t out_s32x4, + const int32x4_t in_s32x4, + const int32x2_t coef_s32x2 +) +{ + return vaddq_s32( out_s32x4, vqdmulhq_lane_s32( in_s32x4, coef_s32x2, 0 ) ); +} + +static OPUS_INLINE int32x4_t silk_SMLAWB_lane1_neon( + const int32x4_t out_s32x4, + const int32x4_t in_s32x4, + const int32x2_t coef_s32x2 +) +{ + return vaddq_s32( out_s32x4, vqdmulhq_lane_s32( in_s32x4, coef_s32x2, 1 ) ); +} + +/* Note: This function has different return value than silk_noise_shape_quantizer_short_prediction_neon(). */ +/* Therefore here we append "_local" to the function name to avoid confusion. */ +static OPUS_INLINE int32x4_t silk_noise_shape_quantizer_short_prediction_neon_local(const opus_int32 *buf32, const opus_int32 *a_Q12_arch, opus_int order) +{ + const int32x4_t a_Q12_arch0_s32x4 = vld1q_s32( a_Q12_arch + 0 ); + const int32x4_t a_Q12_arch1_s32x4 = vld1q_s32( a_Q12_arch + 4 ); + const int32x4_t a_Q12_arch2_s32x4 = vld1q_s32( a_Q12_arch + 8 ); + const int32x4_t a_Q12_arch3_s32x4 = vld1q_s32( a_Q12_arch + 12 ); + int32x4_t LPC_pred_Q14_s32x4; + + silk_assert( order == 10 || order == 16 ); + /* Avoids introducing a bias because silk_SMLAWB() always rounds to -inf */ + LPC_pred_Q14_s32x4 = vdupq_n_s32( silk_RSHIFT( order, 1 ) ); + LPC_pred_Q14_s32x4 = silk_SMLAWB_lane0_neon( LPC_pred_Q14_s32x4, vld1q_s32( buf32 + 0 * NEON_MAX_DEL_DEC_STATES ), vget_low_s32( a_Q12_arch0_s32x4 ) ); + LPC_pred_Q14_s32x4 = silk_SMLAWB_lane1_neon( LPC_pred_Q14_s32x4, vld1q_s32( buf32 + 1 * NEON_MAX_DEL_DEC_STATES ), vget_low_s32( a_Q12_arch0_s32x4 ) ); + LPC_pred_Q14_s32x4 = silk_SMLAWB_lane0_neon( LPC_pred_Q14_s32x4, vld1q_s32( buf32 + 2 * NEON_MAX_DEL_DEC_STATES ), vget_high_s32( a_Q12_arch0_s32x4 ) ); + LPC_pred_Q14_s32x4 = silk_SMLAWB_lane1_neon( LPC_pred_Q14_s32x4, vld1q_s32( buf32 + 3 * NEON_MAX_DEL_DEC_STATES ), vget_high_s32( a_Q12_arch0_s32x4 ) ); + LPC_pred_Q14_s32x4 = silk_SMLAWB_lane0_neon( LPC_pred_Q14_s32x4, vld1q_s32( buf32 + 4 * NEON_MAX_DEL_DEC_STATES ), vget_low_s32( a_Q12_arch1_s32x4 ) ); + LPC_pred_Q14_s32x4 = silk_SMLAWB_lane1_neon( LPC_pred_Q14_s32x4, vld1q_s32( buf32 + 5 * NEON_MAX_DEL_DEC_STATES ), vget_low_s32( a_Q12_arch1_s32x4 ) ); + LPC_pred_Q14_s32x4 = silk_SMLAWB_lane0_neon( LPC_pred_Q14_s32x4, vld1q_s32( buf32 + 6 * NEON_MAX_DEL_DEC_STATES ), vget_high_s32( a_Q12_arch1_s32x4 ) ); + LPC_pred_Q14_s32x4 = silk_SMLAWB_lane1_neon( LPC_pred_Q14_s32x4, vld1q_s32( buf32 + 7 * NEON_MAX_DEL_DEC_STATES ), vget_high_s32( a_Q12_arch1_s32x4 ) ); + LPC_pred_Q14_s32x4 = silk_SMLAWB_lane0_neon( LPC_pred_Q14_s32x4, vld1q_s32( buf32 + 8 * NEON_MAX_DEL_DEC_STATES ), vget_low_s32( a_Q12_arch2_s32x4 ) ); + LPC_pred_Q14_s32x4 = silk_SMLAWB_lane1_neon( LPC_pred_Q14_s32x4, vld1q_s32( buf32 + 9 * NEON_MAX_DEL_DEC_STATES ), vget_low_s32( a_Q12_arch2_s32x4 ) ); + LPC_pred_Q14_s32x4 = silk_SMLAWB_lane0_neon( LPC_pred_Q14_s32x4, vld1q_s32( buf32 + 10 * NEON_MAX_DEL_DEC_STATES ), vget_high_s32( a_Q12_arch2_s32x4 ) ); + LPC_pred_Q14_s32x4 = silk_SMLAWB_lane1_neon( LPC_pred_Q14_s32x4, vld1q_s32( buf32 + 11 * NEON_MAX_DEL_DEC_STATES ), vget_high_s32( a_Q12_arch2_s32x4 ) ); + LPC_pred_Q14_s32x4 = silk_SMLAWB_lane0_neon( LPC_pred_Q14_s32x4, vld1q_s32( buf32 + 12 * NEON_MAX_DEL_DEC_STATES ), vget_low_s32( a_Q12_arch3_s32x4 ) ); + LPC_pred_Q14_s32x4 = silk_SMLAWB_lane1_neon( LPC_pred_Q14_s32x4, vld1q_s32( buf32 + 13 * NEON_MAX_DEL_DEC_STATES ), vget_low_s32( a_Q12_arch3_s32x4 ) ); + LPC_pred_Q14_s32x4 = silk_SMLAWB_lane0_neon( LPC_pred_Q14_s32x4, vld1q_s32( buf32 + 14 * NEON_MAX_DEL_DEC_STATES ), vget_high_s32( a_Q12_arch3_s32x4 ) ); + LPC_pred_Q14_s32x4 = silk_SMLAWB_lane1_neon( LPC_pred_Q14_s32x4, vld1q_s32( buf32 + 15 * NEON_MAX_DEL_DEC_STATES ), vget_high_s32( a_Q12_arch3_s32x4 ) ); + + return LPC_pred_Q14_s32x4; +} + +static OPUS_INLINE void silk_noise_shape_quantizer_del_dec_neon( + silk_nsq_state *NSQ, /* I/O NSQ state */ + NSQ_del_decs_struct psDelDec[], /* I/O Delayed decision states */ + opus_int signalType, /* I Signal type */ + const opus_int32 x_Q10[], /* I */ + opus_int8 pulses[], /* O */ + opus_int16 xq[], /* O */ + opus_int32 sLTP_Q15[], /* I/O LTP filter state */ + opus_int32 delayedGain_Q10[], /* I/O Gain delay buffer */ + const opus_int16 a_Q12[], /* I Short term prediction coefs */ + const opus_int16 b_Q14[], /* I Long term prediction coefs */ + const opus_int16 AR_shp_Q13[], /* I Noise shaping coefs */ + opus_int lag, /* I Pitch lag */ + opus_int32 HarmShapeFIRPacked_Q14, /* I */ + opus_int Tilt_Q14, /* I Spectral tilt */ + opus_int32 LF_shp_Q14, /* I */ + opus_int32 Gain_Q16, /* I */ + opus_int Lambda_Q10, /* I */ + opus_int offset_Q10, /* I */ + opus_int length, /* I Input length */ + opus_int subfr, /* I Subframe number */ + opus_int shapingLPCOrder, /* I Shaping LPC filter order */ + opus_int predictLPCOrder, /* I Prediction filter order */ + opus_int warping_Q16, /* I */ + opus_int nStatesDelayedDecision, /* I Number of states in decision tree */ + opus_int *smpl_buf_idx, /* I/O Index to newest samples in buffers */ + opus_int decisionDelay /* I */ +) +{ + opus_int i, j, k, Winner_ind, RDmin_ind, RDmax_ind, last_smple_idx; + opus_int32 Winner_rand_state; + opus_int32 LTP_pred_Q14, n_LTP_Q14; + opus_int32 RDmin_Q10, RDmax_Q10; + opus_int32 Gain_Q10; + opus_int32 *pred_lag_ptr, *shp_lag_ptr; + opus_int32 a_Q12_arch[MAX_LPC_ORDER]; + const int32x2_t warping_Q16_s32x2 = vdup_n_s32( silk_LSHIFT32( warping_Q16, 16 ) >> 1 ); + const opus_int32 LF_shp_Q29 = silk_LSHIFT32( LF_shp_Q14, 16 ) >> 1; + opus_int32 AR_shp_Q28[ MAX_SHAPE_LPC_ORDER ]; + const uint32x4_t rand_multiplier_u32x4 = vdupq_n_u32( RAND_MULTIPLIER ); + const uint32x4_t rand_increment_u32x4 = vdupq_n_u32( RAND_INCREMENT ); + + VARDECL( NSQ_samples_struct, psSampleState ); + SAVE_STACK; + + silk_assert( nStatesDelayedDecision > 0 ); + silk_assert( ( shapingLPCOrder & 1 ) == 0 ); /* check that order is even */ + ALLOC( psSampleState, 2, NSQ_samples_struct ); + + shp_lag_ptr = &NSQ->sLTP_shp_Q14[ NSQ->sLTP_shp_buf_idx - lag + HARM_SHAPE_FIR_TAPS / 2 ]; + pred_lag_ptr = &sLTP_Q15[ NSQ->sLTP_buf_idx - lag + LTP_ORDER / 2 ]; + Gain_Q10 = silk_RSHIFT( Gain_Q16, 6 ); + + for( i = 0; i < ( MAX_SHAPE_LPC_ORDER - 7 ); i += 8 ) { + const int16x8_t t_s16x8 = vld1q_s16( AR_shp_Q13 + i ); + vst1q_s32( AR_shp_Q28 + i + 0, vshll_n_s16( vget_low_s16( t_s16x8 ), 15 ) ); + vst1q_s32( AR_shp_Q28 + i + 4, vshll_n_s16( vget_high_s16( t_s16x8 ), 15 ) ); + } + + for( ; i < MAX_SHAPE_LPC_ORDER; i++ ) { + AR_shp_Q28[i] = silk_LSHIFT32( AR_shp_Q13[i], 15 ); + } + + silk_short_prediction_create_arch_coef_neon_local( a_Q12_arch, a_Q12, predictLPCOrder ); + + for( i = 0; i < length; i++ ) { + int32x4_t Seed_s32x4, LPC_pred_Q14_s32x4; + int32x4_t sign_s32x4, tmp1_s32x4, tmp2_s32x4; + int32x4_t n_AR_Q14_s32x4, n_LF_Q14_s32x4; + int32x2_t AR_shp_Q28_s32x2; + int16x4_t r_Q10_s16x4, rr_Q10_s16x4; + + /* Perform common calculations used in all states */ + + /* Long-term prediction */ + if( signalType == TYPE_VOICED ) { + /* Unrolled loop */ + /* Avoids introducing a bias because silk_SMLAWB() always rounds to -inf */ + LTP_pred_Q14 = 2; + LTP_pred_Q14 = silk_SMLAWB( LTP_pred_Q14, pred_lag_ptr[ 0 ], b_Q14[ 0 ] ); + LTP_pred_Q14 = silk_SMLAWB( LTP_pred_Q14, pred_lag_ptr[ -1 ], b_Q14[ 1 ] ); + LTP_pred_Q14 = silk_SMLAWB( LTP_pred_Q14, pred_lag_ptr[ -2 ], b_Q14[ 2 ] ); + LTP_pred_Q14 = silk_SMLAWB( LTP_pred_Q14, pred_lag_ptr[ -3 ], b_Q14[ 3 ] ); + LTP_pred_Q14 = silk_SMLAWB( LTP_pred_Q14, pred_lag_ptr[ -4 ], b_Q14[ 4 ] ); + LTP_pred_Q14 = silk_LSHIFT( LTP_pred_Q14, 1 ); /* Q13 -> Q14 */ + pred_lag_ptr++; + } else { + LTP_pred_Q14 = 0; + } + + /* Long-term shaping */ + if( lag > 0 ) { + /* Symmetric, packed FIR coefficients */ + n_LTP_Q14 = silk_SMULWB( silk_ADD32( shp_lag_ptr[ 0 ], shp_lag_ptr[ -2 ] ), HarmShapeFIRPacked_Q14 ); + n_LTP_Q14 = silk_SMLAWT( n_LTP_Q14, shp_lag_ptr[ -1 ], HarmShapeFIRPacked_Q14 ); + n_LTP_Q14 = silk_SUB_LSHIFT32( LTP_pred_Q14, n_LTP_Q14, 2 ); /* Q12 -> Q14 */ + shp_lag_ptr++; + } else { + n_LTP_Q14 = 0; + } + + /* Generate dither */ + Seed_s32x4 = vld1q_s32( psDelDec->Seed ); + Seed_s32x4 = vreinterpretq_s32_u32( vmlaq_u32( rand_increment_u32x4, vreinterpretq_u32_s32( Seed_s32x4 ), rand_multiplier_u32x4 ) ); + vst1q_s32( psDelDec->Seed, Seed_s32x4 ); + + /* Short-term prediction */ + LPC_pred_Q14_s32x4 = silk_noise_shape_quantizer_short_prediction_neon_local(psDelDec->sLPC_Q14[ NSQ_LPC_BUF_LENGTH - 16 + i ], a_Q12_arch, predictLPCOrder); + LPC_pred_Q14_s32x4 = vshlq_n_s32( LPC_pred_Q14_s32x4, 4 ); /* Q10 -> Q14 */ + + /* Noise shape feedback */ + /* Output of lowpass section */ + tmp2_s32x4 = silk_SMLAWB_lane0_neon( vld1q_s32( psDelDec->Diff_Q14 ), vld1q_s32( psDelDec->sAR2_Q14[ 0 ] ), warping_Q16_s32x2 ); + /* Output of allpass section */ + tmp1_s32x4 = vsubq_s32( vld1q_s32( psDelDec->sAR2_Q14[ 1 ] ), tmp2_s32x4 ); + tmp1_s32x4 = silk_SMLAWB_lane0_neon( vld1q_s32( psDelDec->sAR2_Q14[ 0 ] ), tmp1_s32x4, warping_Q16_s32x2 ); + vst1q_s32( psDelDec->sAR2_Q14[ 0 ], tmp2_s32x4 ); + AR_shp_Q28_s32x2 = vld1_s32( AR_shp_Q28 ); + n_AR_Q14_s32x4 = vaddq_s32( vdupq_n_s32( silk_RSHIFT( shapingLPCOrder, 1 ) ), vqdmulhq_lane_s32( tmp2_s32x4, AR_shp_Q28_s32x2, 0 ) ); + + /* Loop over allpass sections */ + for( j = 2; j < shapingLPCOrder; j += 2 ) { + /* Output of allpass section */ + tmp2_s32x4 = vsubq_s32( vld1q_s32( psDelDec->sAR2_Q14[ j + 0 ] ), tmp1_s32x4 ); + tmp2_s32x4 = silk_SMLAWB_lane0_neon( vld1q_s32( psDelDec->sAR2_Q14[ j - 1 ] ), tmp2_s32x4, warping_Q16_s32x2 ); + vst1q_s32( psDelDec->sAR2_Q14[ j - 1 ], tmp1_s32x4 ); + n_AR_Q14_s32x4 = vaddq_s32( n_AR_Q14_s32x4, vqdmulhq_lane_s32( tmp1_s32x4, AR_shp_Q28_s32x2, 1 ) ); + /* Output of allpass section */ + tmp1_s32x4 = vsubq_s32( vld1q_s32( psDelDec->sAR2_Q14[ j + 1 ] ), tmp2_s32x4 ); + tmp1_s32x4 = silk_SMLAWB_lane0_neon( vld1q_s32( psDelDec->sAR2_Q14[ j + 0 ] ), tmp1_s32x4, warping_Q16_s32x2 ); + vst1q_s32( psDelDec->sAR2_Q14[ j + 0 ], tmp2_s32x4 ); + AR_shp_Q28_s32x2 = vld1_s32( &AR_shp_Q28[ j ] ); + n_AR_Q14_s32x4 = vaddq_s32( n_AR_Q14_s32x4, vqdmulhq_lane_s32( tmp2_s32x4, AR_shp_Q28_s32x2, 0 ) ); + } + vst1q_s32( psDelDec->sAR2_Q14[ shapingLPCOrder - 1 ], tmp1_s32x4 ); + n_AR_Q14_s32x4 = vaddq_s32( n_AR_Q14_s32x4, vqdmulhq_lane_s32( tmp1_s32x4, AR_shp_Q28_s32x2, 1 ) ); + n_AR_Q14_s32x4 = vshlq_n_s32( n_AR_Q14_s32x4, 1 ); /* Q11 -> Q12 */ + n_AR_Q14_s32x4 = vaddq_s32( n_AR_Q14_s32x4, vqdmulhq_n_s32( vld1q_s32( psDelDec->LF_AR_Q14 ), silk_LSHIFT32( Tilt_Q14, 16 ) >> 1 ) ); /* Q12 */ + n_AR_Q14_s32x4 = vshlq_n_s32( n_AR_Q14_s32x4, 2 ); /* Q12 -> Q14 */ + n_LF_Q14_s32x4 = vqdmulhq_n_s32( vld1q_s32( psDelDec->Shape_Q14[ *smpl_buf_idx ] ), LF_shp_Q29 ); /* Q12 */ + n_LF_Q14_s32x4 = vaddq_s32( n_LF_Q14_s32x4, vqdmulhq_n_s32( vld1q_s32( psDelDec->LF_AR_Q14 ), silk_LSHIFT32( LF_shp_Q14 >> 16 , 15 ) ) ); /* Q12 */ + n_LF_Q14_s32x4 = vshlq_n_s32( n_LF_Q14_s32x4, 2 ); /* Q12 -> Q14 */ + + /* Input minus prediction plus noise feedback */ + /* r = x[ i ] - LTP_pred - LPC_pred + n_AR + n_Tilt + n_LF + n_LTP */ + tmp1_s32x4 = vaddq_s32( n_AR_Q14_s32x4, n_LF_Q14_s32x4 ); /* Q14 */ + tmp2_s32x4 = vaddq_s32( vdupq_n_s32( n_LTP_Q14 ), LPC_pred_Q14_s32x4 ); /* Q13 */ + tmp1_s32x4 = vsubq_s32( tmp2_s32x4, tmp1_s32x4 ); /* Q13 */ + tmp1_s32x4 = vrshrq_n_s32( tmp1_s32x4, 4 ); /* Q10 */ + tmp1_s32x4 = vsubq_s32( vdupq_n_s32( x_Q10[ i ] ), tmp1_s32x4 ); /* residual error Q10 */ + + /* Flip sign depending on dither */ + sign_s32x4 = vreinterpretq_s32_u32( vcltq_s32( Seed_s32x4, vdupq_n_s32( 0 ) ) ); + tmp1_s32x4 = veorq_s32( tmp1_s32x4, sign_s32x4 ); + tmp1_s32x4 = vsubq_s32( tmp1_s32x4, sign_s32x4 ); + tmp1_s32x4 = vmaxq_s32( tmp1_s32x4, vdupq_n_s32( -( 31 << 10 ) ) ); + tmp1_s32x4 = vminq_s32( tmp1_s32x4, vdupq_n_s32( 30 << 10 ) ); + r_Q10_s16x4 = vmovn_s32( tmp1_s32x4 ); + + /* Find two quantization level candidates and measure their rate-distortion */ + { + int16x4_t q1_Q10_s16x4 = vsub_s16( r_Q10_s16x4, vdup_n_s16( offset_Q10 ) ); + int16x4_t q1_Q0_s16x4 = vshr_n_s16( q1_Q10_s16x4, 10 ); + int16x4_t q2_Q10_s16x4; + int32x4_t rd1_Q10_s32x4, rd2_Q10_s32x4; + uint32x4_t t_u32x4; + + if( Lambda_Q10 > 2048 ) { + /* For aggressive RDO, the bias becomes more than one pulse. */ + const int rdo_offset = Lambda_Q10/2 - 512; + const uint16x4_t greaterThanRdo = vcgt_s16( q1_Q10_s16x4, vdup_n_s16( rdo_offset ) ); + const uint16x4_t lessThanMinusRdo = vclt_s16( q1_Q10_s16x4, vdup_n_s16( -rdo_offset ) ); + /* If Lambda_Q10 > 32767, then q1_Q0, q1_Q10 and q2_Q10 must change to 32-bit. */ + silk_assert( Lambda_Q10 <= 32767 ); + + q1_Q0_s16x4 = vreinterpret_s16_u16( vclt_s16( q1_Q10_s16x4, vdup_n_s16( 0 ) ) ); + q1_Q0_s16x4 = vbsl_s16( greaterThanRdo, vsub_s16( q1_Q10_s16x4, vdup_n_s16( rdo_offset ) ), q1_Q0_s16x4 ); + q1_Q0_s16x4 = vbsl_s16( lessThanMinusRdo, vadd_s16( q1_Q10_s16x4, vdup_n_s16( rdo_offset ) ), q1_Q0_s16x4 ); + q1_Q0_s16x4 = vshr_n_s16( q1_Q0_s16x4, 10 ); + } + { + const uint16x4_t equal0_u16x4 = vceq_s16( q1_Q0_s16x4, vdup_n_s16( 0 ) ); + const uint16x4_t equalMinus1_u16x4 = vceq_s16( q1_Q0_s16x4, vdup_n_s16( -1 ) ); + const uint16x4_t lessThanMinus1_u16x4 = vclt_s16( q1_Q0_s16x4, vdup_n_s16( -1 ) ); + int16x4_t tmp1_s16x4, tmp2_s16x4; + + q1_Q10_s16x4 = vshl_n_s16( q1_Q0_s16x4, 10 ); + tmp1_s16x4 = vadd_s16( q1_Q10_s16x4, vdup_n_s16( offset_Q10 - QUANT_LEVEL_ADJUST_Q10 ) ); + q1_Q10_s16x4 = vadd_s16( q1_Q10_s16x4, vdup_n_s16( offset_Q10 + QUANT_LEVEL_ADJUST_Q10 ) ); + q1_Q10_s16x4 = vbsl_s16( lessThanMinus1_u16x4, q1_Q10_s16x4, tmp1_s16x4 ); + q1_Q10_s16x4 = vbsl_s16( equal0_u16x4, vdup_n_s16( offset_Q10 ), q1_Q10_s16x4 ); + q1_Q10_s16x4 = vbsl_s16( equalMinus1_u16x4, vdup_n_s16( offset_Q10 - ( 1024 - QUANT_LEVEL_ADJUST_Q10 ) ), q1_Q10_s16x4 ); + q2_Q10_s16x4 = vadd_s16( q1_Q10_s16x4, vdup_n_s16( 1024 ) ); + q2_Q10_s16x4 = vbsl_s16( equal0_u16x4, vdup_n_s16( offset_Q10 + 1024 - QUANT_LEVEL_ADJUST_Q10 ), q2_Q10_s16x4 ); + q2_Q10_s16x4 = vbsl_s16( equalMinus1_u16x4, vdup_n_s16( offset_Q10 ), q2_Q10_s16x4 ); + tmp1_s16x4 = q1_Q10_s16x4; + tmp2_s16x4 = q2_Q10_s16x4; + tmp1_s16x4 = vbsl_s16( vorr_u16( equalMinus1_u16x4, lessThanMinus1_u16x4 ), vneg_s16( tmp1_s16x4 ), tmp1_s16x4 ); + tmp2_s16x4 = vbsl_s16( lessThanMinus1_u16x4, vneg_s16( tmp2_s16x4 ), tmp2_s16x4 ); + rd1_Q10_s32x4 = vmull_s16( tmp1_s16x4, vdup_n_s16( Lambda_Q10 ) ); + rd2_Q10_s32x4 = vmull_s16( tmp2_s16x4, vdup_n_s16( Lambda_Q10 ) ); + } + + rr_Q10_s16x4 = vsub_s16( r_Q10_s16x4, q1_Q10_s16x4 ); + rd1_Q10_s32x4 = vmlal_s16( rd1_Q10_s32x4, rr_Q10_s16x4, rr_Q10_s16x4 ); + rd1_Q10_s32x4 = vshrq_n_s32( rd1_Q10_s32x4, 10 ); + + rr_Q10_s16x4 = vsub_s16( r_Q10_s16x4, q2_Q10_s16x4 ); + rd2_Q10_s32x4 = vmlal_s16( rd2_Q10_s32x4, rr_Q10_s16x4, rr_Q10_s16x4 ); + rd2_Q10_s32x4 = vshrq_n_s32( rd2_Q10_s32x4, 10 ); + + tmp2_s32x4 = vld1q_s32( psDelDec->RD_Q10 ); + tmp1_s32x4 = vaddq_s32( tmp2_s32x4, vminq_s32( rd1_Q10_s32x4, rd2_Q10_s32x4 ) ); + tmp2_s32x4 = vaddq_s32( tmp2_s32x4, vmaxq_s32( rd1_Q10_s32x4, rd2_Q10_s32x4 ) ); + vst1q_s32( psSampleState[ 0 ].RD_Q10, tmp1_s32x4 ); + vst1q_s32( psSampleState[ 1 ].RD_Q10, tmp2_s32x4 ); + t_u32x4 = vcltq_s32( rd1_Q10_s32x4, rd2_Q10_s32x4 ); + tmp1_s32x4 = vbslq_s32( t_u32x4, vmovl_s16( q1_Q10_s16x4 ), vmovl_s16( q2_Q10_s16x4 ) ); + tmp2_s32x4 = vbslq_s32( t_u32x4, vmovl_s16( q2_Q10_s16x4 ), vmovl_s16( q1_Q10_s16x4 ) ); + vst1q_s32( psSampleState[ 0 ].Q_Q10, tmp1_s32x4 ); + vst1q_s32( psSampleState[ 1 ].Q_Q10, tmp2_s32x4 ); + } + + { + /* Update states for best quantization */ + int32x4_t exc_Q14_s32x4, LPC_exc_Q14_s32x4, xq_Q14_s32x4, sLF_AR_shp_Q14_s32x4; + + /* Quantized excitation */ + exc_Q14_s32x4 = vshlq_n_s32( tmp1_s32x4, 4 ); + exc_Q14_s32x4 = veorq_s32( exc_Q14_s32x4, sign_s32x4 ); + exc_Q14_s32x4 = vsubq_s32( exc_Q14_s32x4, sign_s32x4 ); + + /* Add predictions */ + LPC_exc_Q14_s32x4 = vaddq_s32( exc_Q14_s32x4, vdupq_n_s32( LTP_pred_Q14 ) ); + xq_Q14_s32x4 = vaddq_s32( LPC_exc_Q14_s32x4, LPC_pred_Q14_s32x4 ); + + /* Update states */ + tmp1_s32x4 = vsubq_s32( xq_Q14_s32x4, vshlq_n_s32( vdupq_n_s32( x_Q10[ i ] ), 4 ) ); + vst1q_s32( psSampleState[ 0 ].Diff_Q14, tmp1_s32x4 ); + sLF_AR_shp_Q14_s32x4 = vsubq_s32( tmp1_s32x4, n_AR_Q14_s32x4 ); + vst1q_s32( psSampleState[ 0 ].sLTP_shp_Q14, vsubq_s32( sLF_AR_shp_Q14_s32x4, n_LF_Q14_s32x4 ) ); + vst1q_s32( psSampleState[ 0 ].LF_AR_Q14, sLF_AR_shp_Q14_s32x4 ); + vst1q_s32( psSampleState[ 0 ].LPC_exc_Q14, LPC_exc_Q14_s32x4 ); + vst1q_s32( psSampleState[ 0 ].xq_Q14, xq_Q14_s32x4 ); + + /* Quantized excitation */ + exc_Q14_s32x4 = vshlq_n_s32( tmp2_s32x4, 4 ); + exc_Q14_s32x4 = veorq_s32( exc_Q14_s32x4, sign_s32x4 ); + exc_Q14_s32x4 = vsubq_s32( exc_Q14_s32x4, sign_s32x4 ); + + /* Add predictions */ + LPC_exc_Q14_s32x4 = vaddq_s32( exc_Q14_s32x4, vdupq_n_s32( LTP_pred_Q14 ) ); + xq_Q14_s32x4 = vaddq_s32( LPC_exc_Q14_s32x4, LPC_pred_Q14_s32x4 ); + + /* Update states */ + tmp1_s32x4 = vsubq_s32( xq_Q14_s32x4, vshlq_n_s32( vdupq_n_s32( x_Q10[ i ] ), 4 ) ); + vst1q_s32( psSampleState[ 1 ].Diff_Q14, tmp1_s32x4 ); + sLF_AR_shp_Q14_s32x4 = vsubq_s32( tmp1_s32x4, n_AR_Q14_s32x4 ); + vst1q_s32( psSampleState[ 1 ].sLTP_shp_Q14, vsubq_s32( sLF_AR_shp_Q14_s32x4, n_LF_Q14_s32x4 ) ); + vst1q_s32( psSampleState[ 1 ].LF_AR_Q14, sLF_AR_shp_Q14_s32x4 ); + vst1q_s32( psSampleState[ 1 ].LPC_exc_Q14, LPC_exc_Q14_s32x4 ); + vst1q_s32( psSampleState[ 1 ].xq_Q14, xq_Q14_s32x4 ); + } + + *smpl_buf_idx = *smpl_buf_idx ? ( *smpl_buf_idx - 1 ) : ( DECISION_DELAY - 1); + last_smple_idx = *smpl_buf_idx + decisionDelay + DECISION_DELAY; + if( last_smple_idx >= DECISION_DELAY ) last_smple_idx -= DECISION_DELAY; + if( last_smple_idx >= DECISION_DELAY ) last_smple_idx -= DECISION_DELAY; + + /* Find winner */ + RDmin_Q10 = psSampleState[ 0 ].RD_Q10[ 0 ]; + Winner_ind = 0; + for( k = 1; k < nStatesDelayedDecision; k++ ) { + if( psSampleState[ 0 ].RD_Q10[ k ] < RDmin_Q10 ) { + RDmin_Q10 = psSampleState[ 0 ].RD_Q10[ k ]; + Winner_ind = k; + } + } + + /* Increase RD values of expired states */ + { + uint32x4_t t_u32x4; + Winner_rand_state = psDelDec->RandState[ last_smple_idx ][ Winner_ind ]; + t_u32x4 = vceqq_s32( vld1q_s32( psDelDec->RandState[ last_smple_idx ] ), vdupq_n_s32( Winner_rand_state ) ); + t_u32x4 = vmvnq_u32( t_u32x4 ); + t_u32x4 = vshrq_n_u32( t_u32x4, 5 ); + tmp1_s32x4 = vld1q_s32( psSampleState[ 0 ].RD_Q10 ); + tmp2_s32x4 = vld1q_s32( psSampleState[ 1 ].RD_Q10 ); + tmp1_s32x4 = vaddq_s32( tmp1_s32x4, vreinterpretq_s32_u32( t_u32x4 ) ); + tmp2_s32x4 = vaddq_s32( tmp2_s32x4, vreinterpretq_s32_u32( t_u32x4 ) ); + vst1q_s32( psSampleState[ 0 ].RD_Q10, tmp1_s32x4 ); + vst1q_s32( psSampleState[ 1 ].RD_Q10, tmp2_s32x4 ); + + /* Find worst in first set and best in second set */ + RDmax_Q10 = psSampleState[ 0 ].RD_Q10[ 0 ]; + RDmin_Q10 = psSampleState[ 1 ].RD_Q10[ 0 ]; + RDmax_ind = 0; + RDmin_ind = 0; + for( k = 1; k < nStatesDelayedDecision; k++ ) { + /* find worst in first set */ + if( psSampleState[ 0 ].RD_Q10[ k ] > RDmax_Q10 ) { + RDmax_Q10 = psSampleState[ 0 ].RD_Q10[ k ]; + RDmax_ind = k; + } + /* find best in second set */ + if( psSampleState[ 1 ].RD_Q10[ k ] < RDmin_Q10 ) { + RDmin_Q10 = psSampleState[ 1 ].RD_Q10[ k ]; + RDmin_ind = k; + } + } + } + + /* Replace a state if best from second set outperforms worst in first set */ + if( RDmin_Q10 < RDmax_Q10 ) { + opus_int32 (*ptr)[NEON_MAX_DEL_DEC_STATES] = psDelDec->RandState; + const int numOthers = (int)( ( sizeof( NSQ_del_decs_struct ) - sizeof( ( (NSQ_del_decs_struct *)0 )->sLPC_Q14 ) ) + / ( NEON_MAX_DEL_DEC_STATES * sizeof( opus_int32 ) ) ); + /* Only ( predictLPCOrder - 1 ) of sLPC_Q14 buffer need to be updated, though the first several */ + /* useless sLPC_Q14[] will be different comparing with C when predictLPCOrder < NSQ_LPC_BUF_LENGTH. */ + /* Here just update constant ( NSQ_LPC_BUF_LENGTH - 1 ) for simplicity. */ + for( j = i + 1; j < i + NSQ_LPC_BUF_LENGTH; j++ ) { + psDelDec->sLPC_Q14[ j ][ RDmax_ind ] = psDelDec->sLPC_Q14[ j ][ RDmin_ind ]; + } + for( j = 0; j < numOthers; j++ ) { + ptr[ j ][ RDmax_ind ] = ptr[ j ][ RDmin_ind ]; + } + + psSampleState[ 0 ].Q_Q10[ RDmax_ind ] = psSampleState[ 1 ].Q_Q10[ RDmin_ind ]; + psSampleState[ 0 ].RD_Q10[ RDmax_ind ] = psSampleState[ 1 ].RD_Q10[ RDmin_ind ]; + psSampleState[ 0 ].xq_Q14[ RDmax_ind ] = psSampleState[ 1 ].xq_Q14[ RDmin_ind ]; + psSampleState[ 0 ].LF_AR_Q14[ RDmax_ind ] = psSampleState[ 1 ].LF_AR_Q14[ RDmin_ind ]; + psSampleState[ 0 ].Diff_Q14[ RDmax_ind ] = psSampleState[ 1 ].Diff_Q14[ RDmin_ind ]; + psSampleState[ 0 ].sLTP_shp_Q14[ RDmax_ind ] = psSampleState[ 1 ].sLTP_shp_Q14[ RDmin_ind ]; + psSampleState[ 0 ].LPC_exc_Q14[ RDmax_ind ] = psSampleState[ 1 ].LPC_exc_Q14[ RDmin_ind ]; + } + + /* Write samples from winner to output and long-term filter states */ + if( subfr > 0 || i >= decisionDelay ) { + pulses[ i - decisionDelay ] = (opus_int8)silk_RSHIFT_ROUND( psDelDec->Q_Q10[ last_smple_idx ][ Winner_ind ], 10 ); + xq[ i - decisionDelay ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( + silk_SMULWW( psDelDec->Xq_Q14[ last_smple_idx ][ Winner_ind ], delayedGain_Q10[ last_smple_idx ] ), 8 ) ); + NSQ->sLTP_shp_Q14[ NSQ->sLTP_shp_buf_idx - decisionDelay ] = psDelDec->Shape_Q14[ last_smple_idx ][ Winner_ind ]; + sLTP_Q15[ NSQ->sLTP_buf_idx - decisionDelay ] = psDelDec->Pred_Q15[ last_smple_idx ][ Winner_ind ]; + } + NSQ->sLTP_shp_buf_idx++; + NSQ->sLTP_buf_idx++; + + /* Update states */ + vst1q_s32( psDelDec->LF_AR_Q14, vld1q_s32( psSampleState[ 0 ].LF_AR_Q14 ) ); + vst1q_s32( psDelDec->Diff_Q14, vld1q_s32( psSampleState[ 0 ].Diff_Q14 ) ); + vst1q_s32( psDelDec->sLPC_Q14[ NSQ_LPC_BUF_LENGTH + i ], vld1q_s32( psSampleState[ 0 ].xq_Q14 ) ); + vst1q_s32( psDelDec->Xq_Q14[ *smpl_buf_idx ], vld1q_s32( psSampleState[ 0 ].xq_Q14 ) ); + tmp1_s32x4 = vld1q_s32( psSampleState[ 0 ].Q_Q10 ); + vst1q_s32( psDelDec->Q_Q10[ *smpl_buf_idx ], tmp1_s32x4 ); + vst1q_s32( psDelDec->Pred_Q15[ *smpl_buf_idx ], vshlq_n_s32( vld1q_s32( psSampleState[ 0 ].LPC_exc_Q14 ), 1 ) ); + vst1q_s32( psDelDec->Shape_Q14[ *smpl_buf_idx ], vld1q_s32( psSampleState[ 0 ].sLTP_shp_Q14 ) ); + tmp1_s32x4 = vrshrq_n_s32( tmp1_s32x4, 10 ); + tmp1_s32x4 = vaddq_s32( vld1q_s32( psDelDec->Seed ), tmp1_s32x4 ); + vst1q_s32( psDelDec->Seed, tmp1_s32x4 ); + vst1q_s32( psDelDec->RandState[ *smpl_buf_idx ], tmp1_s32x4 ); + vst1q_s32( psDelDec->RD_Q10, vld1q_s32( psSampleState[ 0 ].RD_Q10 ) ); + delayedGain_Q10[ *smpl_buf_idx ] = Gain_Q10; + } + /* Update LPC states */ + silk_memcpy( psDelDec->sLPC_Q14[ 0 ], psDelDec->sLPC_Q14[ length ], NEON_MAX_DEL_DEC_STATES * NSQ_LPC_BUF_LENGTH * sizeof( opus_int32 ) ); + + RESTORE_STACK; +} + +static OPUS_INLINE void silk_SMULWB_8_neon( + const opus_int16 *a, + const int32x2_t b, + opus_int32 *o +) +{ + const int16x8_t a_s16x8 = vld1q_s16( a ); + int32x4_t o0_s32x4, o1_s32x4; + + o0_s32x4 = vshll_n_s16( vget_low_s16( a_s16x8 ), 15 ); + o1_s32x4 = vshll_n_s16( vget_high_s16( a_s16x8 ), 15 ); + o0_s32x4 = vqdmulhq_lane_s32( o0_s32x4, b, 0 ); + o1_s32x4 = vqdmulhq_lane_s32( o1_s32x4, b, 0 ); + vst1q_s32( o, o0_s32x4 ); + vst1q_s32( o + 4, o1_s32x4 ); +} + +/* Only works when ( b >= -65536 ) && ( b < 65536 ). */ +static OPUS_INLINE void silk_SMULWW_small_b_4_neon( + opus_int32 *a, + const int32x2_t b_s32x2) +{ + int32x4_t o_s32x4; + + o_s32x4 = vld1q_s32( a ); + o_s32x4 = vqdmulhq_lane_s32( o_s32x4, b_s32x2, 0 ); + vst1q_s32( a, o_s32x4 ); +} + +/* Only works when ( b >= -65536 ) && ( b < 65536 ). */ +static OPUS_INLINE void silk_SMULWW_small_b_8_neon( + opus_int32 *a, + const int32x2_t b_s32x2 +) +{ + int32x4_t o0_s32x4, o1_s32x4; + + o0_s32x4 = vld1q_s32( a ); + o1_s32x4 = vld1q_s32( a + 4 ); + o0_s32x4 = vqdmulhq_lane_s32( o0_s32x4, b_s32x2, 0 ); + o1_s32x4 = vqdmulhq_lane_s32( o1_s32x4, b_s32x2, 0 ); + vst1q_s32( a, o0_s32x4 ); + vst1q_s32( a + 4, o1_s32x4 ); +} + +static OPUS_INLINE void silk_SMULWW_4_neon( + opus_int32 *a, + const int32x2_t b_s32x2) +{ + int32x4_t a_s32x4, o_s32x4; + + a_s32x4 = vld1q_s32( a ); + o_s32x4 = vqdmulhq_lane_s32( a_s32x4, b_s32x2, 0 ); + o_s32x4 = vmlaq_lane_s32( o_s32x4, a_s32x4, b_s32x2, 1 ); + vst1q_s32( a, o_s32x4 ); +} + +static OPUS_INLINE void silk_SMULWW_8_neon( + opus_int32 *a, + const int32x2_t b_s32x2 +) +{ + int32x4_t a0_s32x4, a1_s32x4, o0_s32x4, o1_s32x4; + + a0_s32x4 = vld1q_s32( a ); + a1_s32x4 = vld1q_s32( a + 4 ); + o0_s32x4 = vqdmulhq_lane_s32( a0_s32x4, b_s32x2, 0 ); + o1_s32x4 = vqdmulhq_lane_s32( a1_s32x4, b_s32x2, 0 ); + o0_s32x4 = vmlaq_lane_s32( o0_s32x4, a0_s32x4, b_s32x2, 1 ); + o1_s32x4 = vmlaq_lane_s32( o1_s32x4, a1_s32x4, b_s32x2, 1 ); + vst1q_s32( a, o0_s32x4 ); + vst1q_s32( a + 4, o1_s32x4 ); +} + +static OPUS_INLINE void silk_SMULWW_loop_neon( + const opus_int16 *a, + const opus_int32 b, + opus_int32 *o, + const opus_int loop_num +) +{ + opus_int i; + int32x2_t b_s32x2; + + b_s32x2 = vdup_n_s32( b ); + for( i = 0; i < loop_num - 7; i += 8 ) { + silk_SMULWB_8_neon( a + i, b_s32x2, o + i ); + } + for( ; i < loop_num; i++ ) { + o[ i ] = silk_SMULWW( a[ i ], b ); + } +} + +static OPUS_INLINE void silk_nsq_del_dec_scale_states_neon( + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + NSQ_del_decs_struct psDelDec[], /* I/O Delayed decision states */ + const opus_int16 x16[], /* I Input */ + opus_int32 x_sc_Q10[], /* O Input scaled with 1/Gain in Q10 */ + const opus_int16 sLTP[], /* I Re-whitened LTP state in Q0 */ + opus_int32 sLTP_Q15[], /* O LTP state matching scaled input */ + opus_int subfr, /* I Subframe number */ + const opus_int LTP_scale_Q14, /* I LTP state scaling */ + const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lag */ + const opus_int signal_type, /* I Signal type */ + const opus_int decisionDelay /* I Decision delay */ +) +{ + opus_int i, lag; + opus_int32 gain_adj_Q16, inv_gain_Q31, inv_gain_Q26; + + lag = pitchL[ subfr ]; + inv_gain_Q31 = silk_INVERSE32_varQ( silk_max( Gains_Q16[ subfr ], 1 ), 47 ); + silk_assert( inv_gain_Q31 != 0 ); + + /* Scale input */ + inv_gain_Q26 = silk_RSHIFT_ROUND( inv_gain_Q31, 5 ); + silk_SMULWW_loop_neon( x16, inv_gain_Q26, x_sc_Q10, psEncC->subfr_length ); + + /* After rewhitening the LTP state is un-scaled, so scale with inv_gain_Q16 */ + if( NSQ->rewhite_flag ) { + if( subfr == 0 ) { + /* Do LTP downscaling */ + inv_gain_Q31 = silk_LSHIFT( silk_SMULWB( inv_gain_Q31, LTP_scale_Q14 ), 2 ); + } + silk_SMULWW_loop_neon( sLTP + NSQ->sLTP_buf_idx - lag - LTP_ORDER / 2, inv_gain_Q31, sLTP_Q15 + NSQ->sLTP_buf_idx - lag - LTP_ORDER / 2, lag + LTP_ORDER / 2 ); + } + + /* Adjust for changing gain */ + if( Gains_Q16[ subfr ] != NSQ->prev_gain_Q16 ) { + int32x2_t gain_adj_Q16_s32x2; + gain_adj_Q16 = silk_DIV32_varQ( NSQ->prev_gain_Q16, Gains_Q16[ subfr ], 16 ); + + /* Scale long-term shaping state */ + if( ( gain_adj_Q16 >= -65536 ) && ( gain_adj_Q16 < 65536 ) ) { + gain_adj_Q16_s32x2 = vdup_n_s32( silk_LSHIFT32( gain_adj_Q16, 15 ) ); + for( i = NSQ->sLTP_shp_buf_idx - psEncC->ltp_mem_length; i < NSQ->sLTP_shp_buf_idx - 7; i += 8 ) { + silk_SMULWW_small_b_8_neon( NSQ->sLTP_shp_Q14 + i, gain_adj_Q16_s32x2 ); + } + for( ; i < NSQ->sLTP_shp_buf_idx; i++ ) { + NSQ->sLTP_shp_Q14[ i ] = silk_SMULWW( gain_adj_Q16, NSQ->sLTP_shp_Q14[ i ] ); + } + + /* Scale long-term prediction state */ + if( signal_type == TYPE_VOICED && NSQ->rewhite_flag == 0 ) { + for( i = NSQ->sLTP_buf_idx - lag - LTP_ORDER / 2; i < NSQ->sLTP_buf_idx - decisionDelay - 7; i += 8 ) { + silk_SMULWW_small_b_8_neon( sLTP_Q15 + i, gain_adj_Q16_s32x2 ); + } + for( ; i < NSQ->sLTP_buf_idx - decisionDelay; i++ ) { + sLTP_Q15[ i ] = silk_SMULWW( gain_adj_Q16, sLTP_Q15[ i ] ); + } + } + + /* Scale scalar states */ + silk_SMULWW_small_b_4_neon( psDelDec->LF_AR_Q14, gain_adj_Q16_s32x2 ); + silk_SMULWW_small_b_4_neon( psDelDec->Diff_Q14, gain_adj_Q16_s32x2 ); + + /* Scale short-term prediction and shaping states */ + for( i = 0; i < NSQ_LPC_BUF_LENGTH; i++ ) { + silk_SMULWW_small_b_4_neon( psDelDec->sLPC_Q14[ i ], gain_adj_Q16_s32x2 ); + } + + for( i = 0; i < MAX_SHAPE_LPC_ORDER; i++ ) { + silk_SMULWW_small_b_4_neon( psDelDec->sAR2_Q14[ i ], gain_adj_Q16_s32x2 ); + } + + for( i = 0; i < DECISION_DELAY; i++ ) { + silk_SMULWW_small_b_4_neon( psDelDec->Pred_Q15[ i ], gain_adj_Q16_s32x2 ); + silk_SMULWW_small_b_4_neon( psDelDec->Shape_Q14[ i ], gain_adj_Q16_s32x2 ); + } + } else { + gain_adj_Q16_s32x2 = vdup_n_s32( silk_LSHIFT32( gain_adj_Q16 & 0x0000FFFF, 15 ) ); + gain_adj_Q16_s32x2 = vset_lane_s32( gain_adj_Q16 >> 16, gain_adj_Q16_s32x2, 1 ); + for( i = NSQ->sLTP_shp_buf_idx - psEncC->ltp_mem_length; i < NSQ->sLTP_shp_buf_idx - 7; i += 8 ) { + silk_SMULWW_8_neon( NSQ->sLTP_shp_Q14 + i, gain_adj_Q16_s32x2 ); + } + for( ; i < NSQ->sLTP_shp_buf_idx; i++ ) { + NSQ->sLTP_shp_Q14[ i ] = silk_SMULWW( gain_adj_Q16, NSQ->sLTP_shp_Q14[ i ] ); + } + + /* Scale long-term prediction state */ + if( signal_type == TYPE_VOICED && NSQ->rewhite_flag == 0 ) { + for( i = NSQ->sLTP_buf_idx - lag - LTP_ORDER / 2; i < NSQ->sLTP_buf_idx - decisionDelay - 7; i += 8 ) { + silk_SMULWW_8_neon( sLTP_Q15 + i, gain_adj_Q16_s32x2 ); + } + for( ; i < NSQ->sLTP_buf_idx - decisionDelay; i++ ) { + sLTP_Q15[ i ] = silk_SMULWW( gain_adj_Q16, sLTP_Q15[ i ] ); + } + } + + /* Scale scalar states */ + silk_SMULWW_4_neon( psDelDec->LF_AR_Q14, gain_adj_Q16_s32x2 ); + silk_SMULWW_4_neon( psDelDec->Diff_Q14, gain_adj_Q16_s32x2 ); + + /* Scale short-term prediction and shaping states */ + for( i = 0; i < NSQ_LPC_BUF_LENGTH; i++ ) { + silk_SMULWW_4_neon( psDelDec->sLPC_Q14[ i ], gain_adj_Q16_s32x2 ); + } + + for( i = 0; i < MAX_SHAPE_LPC_ORDER; i++ ) { + silk_SMULWW_4_neon( psDelDec->sAR2_Q14[ i ], gain_adj_Q16_s32x2 ); + } + + for( i = 0; i < DECISION_DELAY; i++ ) { + silk_SMULWW_4_neon( psDelDec->Pred_Q15[ i ], gain_adj_Q16_s32x2 ); + silk_SMULWW_4_neon( psDelDec->Shape_Q14[ i ], gain_adj_Q16_s32x2 ); + } + } + + /* Save inverse gain */ + NSQ->prev_gain_Q16 = Gains_Q16[ subfr ]; + } +} diff --git a/media/libopus/silk/arm/NSQ_neon.c b/media/libopus/silk/arm/NSQ_neon.c new file mode 100644 index 0000000000..9642529973 --- /dev/null +++ b/media/libopus/silk/arm/NSQ_neon.c @@ -0,0 +1,112 @@ +/*********************************************************************** +Copyright (C) 2014 Vidyo +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include <arm_neon.h> +#include "main.h" +#include "stack_alloc.h" +#include "NSQ.h" +#include "celt/cpu_support.h" +#include "celt/arm/armcpu.h" + +opus_int32 silk_noise_shape_quantizer_short_prediction_neon(const opus_int32 *buf32, const opus_int32 *coef32, opus_int order) +{ + int32x4_t coef0 = vld1q_s32(coef32); + int32x4_t coef1 = vld1q_s32(coef32 + 4); + int32x4_t coef2 = vld1q_s32(coef32 + 8); + int32x4_t coef3 = vld1q_s32(coef32 + 12); + + int32x4_t a0 = vld1q_s32(buf32 - 15); + int32x4_t a1 = vld1q_s32(buf32 - 11); + int32x4_t a2 = vld1q_s32(buf32 - 7); + int32x4_t a3 = vld1q_s32(buf32 - 3); + + int32x4_t b0 = vqdmulhq_s32(coef0, a0); + int32x4_t b1 = vqdmulhq_s32(coef1, a1); + int32x4_t b2 = vqdmulhq_s32(coef2, a2); + int32x4_t b3 = vqdmulhq_s32(coef3, a3); + + int32x4_t c0 = vaddq_s32(b0, b1); + int32x4_t c1 = vaddq_s32(b2, b3); + + int32x4_t d = vaddq_s32(c0, c1); + + int64x2_t e = vpaddlq_s32(d); + + int64x1_t f = vadd_s64(vget_low_s64(e), vget_high_s64(e)); + + opus_int32 out = vget_lane_s32(vreinterpret_s32_s64(f), 0); + + out += silk_RSHIFT( order, 1 ); + + return out; +} + + +opus_int32 silk_NSQ_noise_shape_feedback_loop_neon(const opus_int32 *data0, opus_int32 *data1, const opus_int16 *coef, opus_int order) +{ + opus_int32 out; + if (order == 8) + { + int32x4_t a00 = vdupq_n_s32(data0[0]); + int32x4_t a01 = vld1q_s32(data1); /* data1[0] ... [3] */ + + int32x4_t a0 = vextq_s32 (a00, a01, 3); /* data0[0] data1[0] ...[2] */ + int32x4_t a1 = vld1q_s32(data1 + 3); /* data1[3] ... [6] */ + + /*TODO: Convert these once in advance instead of once per sample, like + silk_noise_shape_quantizer_short_prediction_neon() does.*/ + int16x8_t coef16 = vld1q_s16(coef); + int32x4_t coef0 = vmovl_s16(vget_low_s16(coef16)); + int32x4_t coef1 = vmovl_s16(vget_high_s16(coef16)); + + /*This is not bit-exact with the C version, since we do not drop the + lower 16 bits of each multiply, but wait until the end to truncate + precision. This is an encoder-specific calculation (and unlike + silk_noise_shape_quantizer_short_prediction_neon(), is not meant to + simulate what the decoder will do). We still could use vqdmulhq_s32() + like silk_noise_shape_quantizer_short_prediction_neon() and save + half the multiplies, but the speed difference is not large, since we + then need two extra adds.*/ + int64x2_t b0 = vmull_s32(vget_low_s32(a0), vget_low_s32(coef0)); + int64x2_t b1 = vmlal_s32(b0, vget_high_s32(a0), vget_high_s32(coef0)); + int64x2_t b2 = vmlal_s32(b1, vget_low_s32(a1), vget_low_s32(coef1)); + int64x2_t b3 = vmlal_s32(b2, vget_high_s32(a1), vget_high_s32(coef1)); + + int64x1_t c = vadd_s64(vget_low_s64(b3), vget_high_s64(b3)); + int64x1_t cS = vrshr_n_s64(c, 15); + int32x2_t d = vreinterpret_s32_s64(cS); + + out = vget_lane_s32(d, 0); + vst1q_s32(data1, a0); + vst1q_s32(data1 + 4, a1); + return out; + } + return silk_NSQ_noise_shape_feedback_loop_c(data0, data1, coef, order); +} diff --git a/media/libopus/silk/arm/NSQ_neon.h b/media/libopus/silk/arm/NSQ_neon.h new file mode 100644 index 0000000000..b31d9442d6 --- /dev/null +++ b/media/libopus/silk/arm/NSQ_neon.h @@ -0,0 +1,114 @@ +/*********************************************************************** +Copyright (C) 2014 Vidyo +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ +#ifndef SILK_NSQ_NEON_H +#define SILK_NSQ_NEON_H + +#include "cpu_support.h" +#include "SigProc_FIX.h" + +#undef silk_short_prediction_create_arch_coef +/* For vectorized calc, reverse a_Q12 coefs, convert to 32-bit, and shift for vqdmulhq_s32. */ +static OPUS_INLINE void silk_short_prediction_create_arch_coef_neon(opus_int32 *out, const opus_int16 *in, opus_int order) +{ + out[15] = silk_LSHIFT32(in[0], 15); + out[14] = silk_LSHIFT32(in[1], 15); + out[13] = silk_LSHIFT32(in[2], 15); + out[12] = silk_LSHIFT32(in[3], 15); + out[11] = silk_LSHIFT32(in[4], 15); + out[10] = silk_LSHIFT32(in[5], 15); + out[9] = silk_LSHIFT32(in[6], 15); + out[8] = silk_LSHIFT32(in[7], 15); + out[7] = silk_LSHIFT32(in[8], 15); + out[6] = silk_LSHIFT32(in[9], 15); + + if (order == 16) + { + out[5] = silk_LSHIFT32(in[10], 15); + out[4] = silk_LSHIFT32(in[11], 15); + out[3] = silk_LSHIFT32(in[12], 15); + out[2] = silk_LSHIFT32(in[13], 15); + out[1] = silk_LSHIFT32(in[14], 15); + out[0] = silk_LSHIFT32(in[15], 15); + } + else + { + out[5] = 0; + out[4] = 0; + out[3] = 0; + out[2] = 0; + out[1] = 0; + out[0] = 0; + } +} + +#if defined(OPUS_ARM_PRESUME_NEON_INTR) + +#define silk_short_prediction_create_arch_coef(out, in, order) \ + (silk_short_prediction_create_arch_coef_neon(out, in, order)) + +#elif defined(OPUS_HAVE_RTCD) && defined(OPUS_ARM_MAY_HAVE_NEON_INTR) + +#define silk_short_prediction_create_arch_coef(out, in, order) \ + do { if (arch == OPUS_ARCH_ARM_NEON) { silk_short_prediction_create_arch_coef_neon(out, in, order); } } while (0) + +#endif + +opus_int32 silk_noise_shape_quantizer_short_prediction_neon(const opus_int32 *buf32, const opus_int32 *coef32, opus_int order); + +opus_int32 silk_NSQ_noise_shape_feedback_loop_neon(const opus_int32 *data0, opus_int32 *data1, const opus_int16 *coef, opus_int order); + +#if defined(OPUS_ARM_PRESUME_NEON_INTR) +#undef silk_noise_shape_quantizer_short_prediction +#define silk_noise_shape_quantizer_short_prediction(in, coef, coefRev, order, arch) \ + ((void)arch,silk_noise_shape_quantizer_short_prediction_neon(in, coefRev, order)) + +#undef silk_NSQ_noise_shape_feedback_loop +#define silk_NSQ_noise_shape_feedback_loop(data0, data1, coef, order, arch) ((void)arch,silk_NSQ_noise_shape_feedback_loop_neon(data0, data1, coef, order)) + +#elif defined(OPUS_HAVE_RTCD) && defined(OPUS_ARM_MAY_HAVE_NEON_INTR) + +/* silk_noise_shape_quantizer_short_prediction implementations take different parameters based on arch + (coef vs. coefRev) so can't use the usual IMPL table implementation */ +#undef silk_noise_shape_quantizer_short_prediction +#define silk_noise_shape_quantizer_short_prediction(in, coef, coefRev, order, arch) \ + (arch == OPUS_ARCH_ARM_NEON ? \ + silk_noise_shape_quantizer_short_prediction_neon(in, coefRev, order) : \ + silk_noise_shape_quantizer_short_prediction_c(in, coef, order)) + +extern opus_int32 + (*const SILK_NSQ_NOISE_SHAPE_FEEDBACK_LOOP_IMPL[OPUS_ARCHMASK+1])( + const opus_int32 *data0, opus_int32 *data1, const opus_int16 *coef, + opus_int order); + +#undef silk_NSQ_noise_shape_feedback_loop +#define silk_NSQ_noise_shape_feedback_loop(data0, data1, coef, order, arch) \ + (SILK_NSQ_NOISE_SHAPE_FEEDBACK_LOOP_IMPL[(arch)&OPUS_ARCHMASK](data0, data1, \ + coef, order)) + +#endif + +#endif /* SILK_NSQ_NEON_H */ diff --git a/media/libopus/silk/arm/SigProc_FIX_armv4.h b/media/libopus/silk/arm/SigProc_FIX_armv4.h new file mode 100644 index 0000000000..ff62b1e5d6 --- /dev/null +++ b/media/libopus/silk/arm/SigProc_FIX_armv4.h @@ -0,0 +1,47 @@ +/*********************************************************************** +Copyright (C) 2013 Xiph.Org Foundation and contributors +Copyright (c) 2013 Parrot +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef SILK_SIGPROC_FIX_ARMv4_H +#define SILK_SIGPROC_FIX_ARMv4_H + +#undef silk_MLA +static OPUS_INLINE opus_int32 silk_MLA_armv4(opus_int32 a, opus_int32 b, + opus_int32 c) +{ + opus_int32 res; + __asm__( + "#silk_MLA\n\t" + "mla %0, %1, %2, %3\n\t" + : "=&r"(res) + : "r"(b), "r"(c), "r"(a) + ); + return res; +} +#define silk_MLA(a, b, c) (silk_MLA_armv4(a, b, c)) + +#endif diff --git a/media/libopus/silk/arm/SigProc_FIX_armv5e.h b/media/libopus/silk/arm/SigProc_FIX_armv5e.h new file mode 100644 index 0000000000..617a09cab1 --- /dev/null +++ b/media/libopus/silk/arm/SigProc_FIX_armv5e.h @@ -0,0 +1,61 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Copyright (c) 2013 Parrot +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef SILK_SIGPROC_FIX_ARMv5E_H +#define SILK_SIGPROC_FIX_ARMv5E_H + +#undef silk_SMULTT +static OPUS_INLINE opus_int32 silk_SMULTT_armv5e(opus_int32 a, opus_int32 b) +{ + opus_int32 res; + __asm__( + "#silk_SMULTT\n\t" + "smultt %0, %1, %2\n\t" + : "=r"(res) + : "%r"(a), "r"(b) + ); + return res; +} +#define silk_SMULTT(a, b) (silk_SMULTT_armv5e(a, b)) + +#undef silk_SMLATT +static OPUS_INLINE opus_int32 silk_SMLATT_armv5e(opus_int32 a, opus_int32 b, + opus_int32 c) +{ + opus_int32 res; + __asm__( + "#silk_SMLATT\n\t" + "smlatt %0, %1, %2, %3\n\t" + : "=r"(res) + : "%r"(b), "r"(c), "r"(a) + ); + return res; +} +#define silk_SMLATT(a, b, c) (silk_SMLATT_armv5e(a, b, c)) + +#endif diff --git a/media/libopus/silk/arm/arm_silk_map.c b/media/libopus/silk/arm/arm_silk_map.c new file mode 100644 index 0000000000..0b9bfec2ca --- /dev/null +++ b/media/libopus/silk/arm/arm_silk_map.c @@ -0,0 +1,123 @@ +/*********************************************************************** +Copyright (C) 2014 Vidyo +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ +#ifdef HAVE_CONFIG_H +# include "config.h" +#endif + +#include "main_FIX.h" +#include "NSQ.h" +#include "SigProc_FIX.h" + +#if defined(OPUS_HAVE_RTCD) + +# if (defined(OPUS_ARM_MAY_HAVE_NEON_INTR) && \ + !defined(OPUS_ARM_PRESUME_NEON_INTR)) + +void (*const SILK_BIQUAD_ALT_STRIDE2_IMPL[OPUS_ARCHMASK + 1])( + const opus_int16 *in, /* I input signal */ + const opus_int32 *B_Q28, /* I MA coefficients [3] */ + const opus_int32 *A_Q28, /* I AR coefficients [2] */ + opus_int32 *S, /* I/O State vector [4] */ + opus_int16 *out, /* O output signal */ + const opus_int32 len /* I signal length (must be even) */ +) = { + silk_biquad_alt_stride2_c, /* ARMv4 */ + silk_biquad_alt_stride2_c, /* EDSP */ + silk_biquad_alt_stride2_c, /* Media */ + silk_biquad_alt_stride2_neon, /* Neon */ +}; + +opus_int32 (*const SILK_LPC_INVERSE_PRED_GAIN_IMPL[OPUS_ARCHMASK + 1])( /* O Returns inverse prediction gain in energy domain, Q30 */ + const opus_int16 *A_Q12, /* I Prediction coefficients, Q12 [order] */ + const opus_int order /* I Prediction order */ +) = { + silk_LPC_inverse_pred_gain_c, /* ARMv4 */ + silk_LPC_inverse_pred_gain_c, /* EDSP */ + silk_LPC_inverse_pred_gain_c, /* Media */ + silk_LPC_inverse_pred_gain_neon, /* Neon */ +}; + +void (*const SILK_NSQ_DEL_DEC_IMPL[OPUS_ARCHMASK + 1])( + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + SideInfoIndices *psIndices, /* I/O Quantization Indices */ + const opus_int16 x16[], /* I Input */ + opus_int8 pulses[], /* O Quantized pulse signal */ + const opus_int16 PredCoef_Q12[ 2 * MAX_LPC_ORDER ], /* I Short term prediction coefs */ + const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ + const opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ + const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ + const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ + const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ + const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ + const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ + const opus_int LTP_scale_Q14 /* I LTP state scaling */ +) = { + silk_NSQ_del_dec_c, /* ARMv4 */ + silk_NSQ_del_dec_c, /* EDSP */ + silk_NSQ_del_dec_c, /* Media */ + silk_NSQ_del_dec_neon, /* Neon */ +}; + +/*There is no table for silk_noise_shape_quantizer_short_prediction because the + NEON version takes different parameters than the C version. + Instead RTCD is done via if statements at the call sites. + See NSQ_neon.h for details.*/ + +opus_int32 + (*const SILK_NSQ_NOISE_SHAPE_FEEDBACK_LOOP_IMPL[OPUS_ARCHMASK+1])( + const opus_int32 *data0, opus_int32 *data1, const opus_int16 *coef, + opus_int order) = { + silk_NSQ_noise_shape_feedback_loop_c, /* ARMv4 */ + silk_NSQ_noise_shape_feedback_loop_c, /* EDSP */ + silk_NSQ_noise_shape_feedback_loop_c, /* Media */ + silk_NSQ_noise_shape_feedback_loop_neon, /* NEON */ +}; + +# endif + +# if defined(FIXED_POINT) && \ + defined(OPUS_ARM_MAY_HAVE_NEON_INTR) && !defined(OPUS_ARM_PRESUME_NEON_INTR) + +void (*const SILK_WARPED_AUTOCORRELATION_FIX_IMPL[OPUS_ARCHMASK + 1])( + opus_int32 *corr, /* O Result [order + 1] */ + opus_int *scale, /* O Scaling of the correlation vector */ + const opus_int16 *input, /* I Input data to correlate */ + const opus_int warping_Q16, /* I Warping coefficient */ + const opus_int length, /* I Length of input */ + const opus_int order /* I Correlation order (even) */ +) = { + silk_warped_autocorrelation_FIX_c, /* ARMv4 */ + silk_warped_autocorrelation_FIX_c, /* EDSP */ + silk_warped_autocorrelation_FIX_c, /* Media */ + silk_warped_autocorrelation_FIX_neon, /* Neon */ +}; + +# endif + +#endif /* OPUS_HAVE_RTCD */ diff --git a/media/libopus/silk/arm/biquad_alt_arm.h b/media/libopus/silk/arm/biquad_alt_arm.h new file mode 100644 index 0000000000..66ea9f43dd --- /dev/null +++ b/media/libopus/silk/arm/biquad_alt_arm.h @@ -0,0 +1,68 @@ +/*********************************************************************** +Copyright (c) 2017 Google Inc. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef SILK_BIQUAD_ALT_ARM_H +# define SILK_BIQUAD_ALT_ARM_H + +# include "celt/arm/armcpu.h" + +# if defined(OPUS_ARM_MAY_HAVE_NEON_INTR) +void silk_biquad_alt_stride2_neon( + const opus_int16 *in, /* I input signal */ + const opus_int32 *B_Q28, /* I MA coefficients [3] */ + const opus_int32 *A_Q28, /* I AR coefficients [2] */ + opus_int32 *S, /* I/O State vector [4] */ + opus_int16 *out, /* O output signal */ + const opus_int32 len /* I signal length (must be even) */ +); + +# if !defined(OPUS_HAVE_RTCD) && defined(OPUS_ARM_PRESUME_NEON) +# define OVERRIDE_silk_biquad_alt_stride2 (1) +# define silk_biquad_alt_stride2(in, B_Q28, A_Q28, S, out, len, arch) ((void)(arch), PRESUME_NEON(silk_biquad_alt_stride2)(in, B_Q28, A_Q28, S, out, len)) +# endif +# endif + +# if !defined(OVERRIDE_silk_biquad_alt_stride2) +/*Is run-time CPU detection enabled on this platform?*/ +# if defined(OPUS_HAVE_RTCD) && (defined(OPUS_ARM_MAY_HAVE_NEON_INTR) && !defined(OPUS_ARM_PRESUME_NEON_INTR)) +extern void (*const SILK_BIQUAD_ALT_STRIDE2_IMPL[OPUS_ARCHMASK+1])( + const opus_int16 *in, /* I input signal */ + const opus_int32 *B_Q28, /* I MA coefficients [3] */ + const opus_int32 *A_Q28, /* I AR coefficients [2] */ + opus_int32 *S, /* I/O State vector [4] */ + opus_int16 *out, /* O output signal */ + const opus_int32 len /* I signal length (must be even) */ + ); +# define OVERRIDE_silk_biquad_alt_stride2 (1) +# define silk_biquad_alt_stride2(in, B_Q28, A_Q28, S, out, len, arch) ((*SILK_BIQUAD_ALT_STRIDE2_IMPL[(arch)&OPUS_ARCHMASK])(in, B_Q28, A_Q28, S, out, len)) +# elif defined(OPUS_ARM_PRESUME_NEON_INTR) +# define OVERRIDE_silk_biquad_alt_stride2 (1) +# define silk_biquad_alt_stride2(in, B_Q28, A_Q28, S, out, len, arch) ((void)(arch), silk_biquad_alt_stride2_neon(in, B_Q28, A_Q28, S, out, len)) +# endif +# endif + +#endif /* end SILK_BIQUAD_ALT_ARM_H */ diff --git a/media/libopus/silk/arm/biquad_alt_neon_intr.c b/media/libopus/silk/arm/biquad_alt_neon_intr.c new file mode 100644 index 0000000000..9715733185 --- /dev/null +++ b/media/libopus/silk/arm/biquad_alt_neon_intr.c @@ -0,0 +1,156 @@ +/*********************************************************************** +Copyright (c) 2017 Google Inc. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include <arm_neon.h> +#ifdef OPUS_CHECK_ASM +# include <string.h> +# include "stack_alloc.h" +#endif +#include "SigProc_FIX.h" + +static inline void silk_biquad_alt_stride2_kernel( const int32x4_t A_L_s32x4, const int32x4_t A_U_s32x4, const int32x4_t B_Q28_s32x4, const int32x2_t t_s32x2, const int32x4_t in_s32x4, int32x4_t *S_s32x4, int32x2_t *out32_Q14_s32x2 ) +{ + int32x4_t t_s32x4, out32_Q14_s32x4; + + *out32_Q14_s32x2 = vadd_s32( vget_low_s32( *S_s32x4 ), t_s32x2 ); /* silk_SMLAWB( S{0,1}, B_Q28[ 0 ], in{0,1} ) */ + *S_s32x4 = vcombine_s32( vget_high_s32( *S_s32x4 ), vdup_n_s32( 0 ) ); /* S{0,1} = S{2,3}; S{2,3} = 0; */ + *out32_Q14_s32x2 = vshl_n_s32( *out32_Q14_s32x2, 2 ); /* out32_Q14_{0,1} = silk_LSHIFT( silk_SMLAWB( S{0,1}, B_Q28[ 0 ], in{0,1} ), 2 ); */ + out32_Q14_s32x4 = vcombine_s32( *out32_Q14_s32x2, *out32_Q14_s32x2 ); /* out32_Q14_{0,1,0,1} */ + t_s32x4 = vqdmulhq_s32( out32_Q14_s32x4, A_L_s32x4 ); /* silk_SMULWB( out32_Q14_{0,1,0,1}, A{0,0,1,1}_L_Q28 ) */ + *S_s32x4 = vrsraq_n_s32( *S_s32x4, t_s32x4, 14 ); /* S{0,1} = S{2,3} + silk_RSHIFT_ROUND(); S{2,3} = silk_RSHIFT_ROUND(); */ + t_s32x4 = vqdmulhq_s32( out32_Q14_s32x4, A_U_s32x4 ); /* silk_SMULWB( out32_Q14_{0,1,0,1}, A{0,0,1,1}_U_Q28 ) */ + *S_s32x4 = vaddq_s32( *S_s32x4, t_s32x4 ); /* S0 = silk_SMLAWB( S{0,1,2,3}, out32_Q14_{0,1,0,1}, A{0,0,1,1}_U_Q28 ); */ + t_s32x4 = vqdmulhq_s32( in_s32x4, B_Q28_s32x4 ); /* silk_SMULWB( B_Q28[ {1,1,2,2} ], in{0,1,0,1} ) */ + *S_s32x4 = vaddq_s32( *S_s32x4, t_s32x4 ); /* S0 = silk_SMLAWB( S0, B_Q28[ {1,1,2,2} ], in{0,1,0,1} ); */ +} + +void silk_biquad_alt_stride2_neon( + const opus_int16 *in, /* I input signal */ + const opus_int32 *B_Q28, /* I MA coefficients [3] */ + const opus_int32 *A_Q28, /* I AR coefficients [2] */ + opus_int32 *S, /* I/O State vector [4] */ + opus_int16 *out, /* O output signal */ + const opus_int32 len /* I signal length (must be even) */ +) +{ + /* DIRECT FORM II TRANSPOSED (uses 2 element state vector) */ + opus_int k = 0; + const int32x2_t offset_s32x2 = vdup_n_s32( (1<<14) - 1 ); + const int32x4_t offset_s32x4 = vcombine_s32( offset_s32x2, offset_s32x2 ); + int16x4_t in_s16x4 = vdup_n_s16( 0 ); + int16x4_t out_s16x4; + int32x2_t A_Q28_s32x2, A_L_s32x2, A_U_s32x2, B_Q28_s32x2, t_s32x2; + int32x4_t A_L_s32x4, A_U_s32x4, B_Q28_s32x4, S_s32x4, out32_Q14_s32x4; + int32x2x2_t t0_s32x2x2, t1_s32x2x2, t2_s32x2x2, S_s32x2x2; + +#ifdef OPUS_CHECK_ASM + opus_int32 S_c[ 4 ]; + VARDECL( opus_int16, out_c ); + SAVE_STACK; + ALLOC( out_c, 2 * len, opus_int16 ); + + silk_memcpy( &S_c, S, sizeof( S_c ) ); + silk_biquad_alt_stride2_c( in, B_Q28, A_Q28, S_c, out_c, len ); +#endif + + /* Negate A_Q28 values and split in two parts */ + A_Q28_s32x2 = vld1_s32( A_Q28 ); + A_Q28_s32x2 = vneg_s32( A_Q28_s32x2 ); + A_L_s32x2 = vshl_n_s32( A_Q28_s32x2, 18 ); /* ( -A_Q28[] & 0x00003FFF ) << 18 */ + A_L_s32x2 = vreinterpret_s32_u32( vshr_n_u32( vreinterpret_u32_s32( A_L_s32x2 ), 3 ) ); /* ( -A_Q28[] & 0x00003FFF ) << 15 */ + A_U_s32x2 = vshr_n_s32( A_Q28_s32x2, 14 ); /* silk_RSHIFT( -A_Q28[], 14 ) */ + A_U_s32x2 = vshl_n_s32( A_U_s32x2, 16 ); /* silk_RSHIFT( -A_Q28[], 14 ) << 16 (Clip two leading bits to conform to C function.) */ + A_U_s32x2 = vshr_n_s32( A_U_s32x2, 1 ); /* silk_RSHIFT( -A_Q28[], 14 ) << 15 */ + + B_Q28_s32x2 = vld1_s32( B_Q28 ); + t_s32x2 = vld1_s32( B_Q28 + 1 ); + t0_s32x2x2 = vzip_s32( A_L_s32x2, A_L_s32x2 ); + t1_s32x2x2 = vzip_s32( A_U_s32x2, A_U_s32x2 ); + t2_s32x2x2 = vzip_s32( t_s32x2, t_s32x2 ); + A_L_s32x4 = vcombine_s32( t0_s32x2x2.val[ 0 ], t0_s32x2x2.val[ 1 ] ); /* A{0,0,1,1}_L_Q28 */ + A_U_s32x4 = vcombine_s32( t1_s32x2x2.val[ 0 ], t1_s32x2x2.val[ 1 ] ); /* A{0,0,1,1}_U_Q28 */ + B_Q28_s32x4 = vcombine_s32( t2_s32x2x2.val[ 0 ], t2_s32x2x2.val[ 1 ] ); /* B_Q28[ {1,1,2,2} ] */ + S_s32x4 = vld1q_s32( S ); /* S0 = S[ 0 ]; S3 = S[ 3 ]; */ + S_s32x2x2 = vtrn_s32( vget_low_s32( S_s32x4 ), vget_high_s32( S_s32x4 ) ); /* S2 = S[ 1 ]; S1 = S[ 2 ]; */ + S_s32x4 = vcombine_s32( S_s32x2x2.val[ 0 ], S_s32x2x2.val[ 1 ] ); + + for( ; k < len - 1; k += 2 ) { + int32x4_t in_s32x4[ 2 ], t_s32x4; + int32x2_t out32_Q14_s32x2[ 2 ]; + + /* S[ 2 * i + 0 ], S[ 2 * i + 1 ], S[ 2 * i + 2 ], S[ 2 * i + 3 ]: Q12 */ + in_s16x4 = vld1_s16( &in[ 2 * k ] ); /* in{0,1,2,3} = in[ 2 * k + {0,1,2,3} ]; */ + in_s32x4[ 0 ] = vshll_n_s16( in_s16x4, 15 ); /* in{0,1,2,3} << 15 */ + t_s32x4 = vqdmulhq_lane_s32( in_s32x4[ 0 ], B_Q28_s32x2, 0 ); /* silk_SMULWB( B_Q28[ 0 ], in{0,1,2,3} ) */ + in_s32x4[ 1 ] = vcombine_s32( vget_high_s32( in_s32x4[ 0 ] ), vget_high_s32( in_s32x4[ 0 ] ) ); /* in{2,3,2,3} << 15 */ + in_s32x4[ 0 ] = vcombine_s32( vget_low_s32 ( in_s32x4[ 0 ] ), vget_low_s32 ( in_s32x4[ 0 ] ) ); /* in{0,1,0,1} << 15 */ + silk_biquad_alt_stride2_kernel( A_L_s32x4, A_U_s32x4, B_Q28_s32x4, vget_low_s32 ( t_s32x4 ), in_s32x4[ 0 ], &S_s32x4, &out32_Q14_s32x2[ 0 ] ); + silk_biquad_alt_stride2_kernel( A_L_s32x4, A_U_s32x4, B_Q28_s32x4, vget_high_s32( t_s32x4 ), in_s32x4[ 1 ], &S_s32x4, &out32_Q14_s32x2[ 1 ] ); + + /* Scale back to Q0 and saturate */ + out32_Q14_s32x4 = vcombine_s32( out32_Q14_s32x2[ 0 ], out32_Q14_s32x2[ 1 ] ); /* out32_Q14_{0,1,2,3} */ + out32_Q14_s32x4 = vaddq_s32( out32_Q14_s32x4, offset_s32x4 ); /* out32_Q14_{0,1,2,3} + (1<<14) - 1 */ + out_s16x4 = vqshrn_n_s32( out32_Q14_s32x4, 14 ); /* (opus_int16)silk_SAT16( silk_RSHIFT( out32_Q14_{0,1,2,3} + (1<<14) - 1, 14 ) ) */ + vst1_s16( &out[ 2 * k ], out_s16x4 ); /* out[ 2 * k + {0,1,2,3} ] = (opus_int16)silk_SAT16( silk_RSHIFT( out32_Q14_{0,1,2,3} + (1<<14) - 1, 14 ) ); */ + } + + /* Process leftover. */ + if( k < len ) { + int32x4_t in_s32x4; + int32x2_t out32_Q14_s32x2; + + /* S[ 2 * i + 0 ], S[ 2 * i + 1 ]: Q12 */ + in_s16x4 = vld1_lane_s16( &in[ 2 * k + 0 ], in_s16x4, 0 ); /* in{0,1} = in[ 2 * k + {0,1} ]; */ + in_s16x4 = vld1_lane_s16( &in[ 2 * k + 1 ], in_s16x4, 1 ); /* in{0,1} = in[ 2 * k + {0,1} ]; */ + in_s32x4 = vshll_n_s16( in_s16x4, 15 ); /* in{0,1} << 15 */ + t_s32x2 = vqdmulh_lane_s32( vget_low_s32( in_s32x4 ), B_Q28_s32x2, 0 ); /* silk_SMULWB( B_Q28[ 0 ], in{0,1} ) */ + in_s32x4 = vcombine_s32( vget_low_s32( in_s32x4 ), vget_low_s32( in_s32x4 ) ); /* in{0,1,0,1} << 15 */ + silk_biquad_alt_stride2_kernel( A_L_s32x4, A_U_s32x4, B_Q28_s32x4, t_s32x2, in_s32x4, &S_s32x4, &out32_Q14_s32x2 ); + + /* Scale back to Q0 and saturate */ + out32_Q14_s32x2 = vadd_s32( out32_Q14_s32x2, offset_s32x2 ); /* out32_Q14_{0,1} + (1<<14) - 1 */ + out32_Q14_s32x4 = vcombine_s32( out32_Q14_s32x2, out32_Q14_s32x2 ); /* out32_Q14_{0,1,0,1} + (1<<14) - 1 */ + out_s16x4 = vqshrn_n_s32( out32_Q14_s32x4, 14 ); /* (opus_int16)silk_SAT16( silk_RSHIFT( out32_Q14_{0,1,0,1} + (1<<14) - 1, 14 ) ) */ + vst1_lane_s16( &out[ 2 * k + 0 ], out_s16x4, 0 ); /* out[ 2 * k + 0 ] = (opus_int16)silk_SAT16( silk_RSHIFT( out32_Q14_0 + (1<<14) - 1, 14 ) ); */ + vst1_lane_s16( &out[ 2 * k + 1 ], out_s16x4, 1 ); /* out[ 2 * k + 1 ] = (opus_int16)silk_SAT16( silk_RSHIFT( out32_Q14_1 + (1<<14) - 1, 14 ) ); */ + } + + vst1q_lane_s32( &S[ 0 ], S_s32x4, 0 ); /* S[ 0 ] = S0; */ + vst1q_lane_s32( &S[ 1 ], S_s32x4, 2 ); /* S[ 1 ] = S2; */ + vst1q_lane_s32( &S[ 2 ], S_s32x4, 1 ); /* S[ 2 ] = S1; */ + vst1q_lane_s32( &S[ 3 ], S_s32x4, 3 ); /* S[ 3 ] = S3; */ + +#ifdef OPUS_CHECK_ASM + silk_assert( !memcmp( S_c, S, sizeof( S_c ) ) ); + silk_assert( !memcmp( out_c, out, 2 * len * sizeof( opus_int16 ) ) ); + RESTORE_STACK; +#endif +} diff --git a/media/libopus/silk/arm/macros_arm64.h b/media/libopus/silk/arm/macros_arm64.h new file mode 100644 index 0000000000..ed030413c5 --- /dev/null +++ b/media/libopus/silk/arm/macros_arm64.h @@ -0,0 +1,39 @@ +/*********************************************************************** +Copyright (C) 2015 Vidyo +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef SILK_MACROS_ARM64_H +#define SILK_MACROS_ARM64_H + +#include <arm_neon.h> + +#undef silk_ADD_SAT32 +#define silk_ADD_SAT32(a, b) (vqadds_s32((a), (b))) + +#undef silk_SUB_SAT32 +#define silk_SUB_SAT32(a, b) (vqsubs_s32((a), (b))) + +#endif /* SILK_MACROS_ARM64_H */ diff --git a/media/libopus/silk/arm/macros_armv4.h b/media/libopus/silk/arm/macros_armv4.h new file mode 100644 index 0000000000..877eb18dd5 --- /dev/null +++ b/media/libopus/silk/arm/macros_armv4.h @@ -0,0 +1,110 @@ +/*********************************************************************** +Copyright (C) 2013 Xiph.Org Foundation and contributors. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef SILK_MACROS_ARMv4_H +#define SILK_MACROS_ARMv4_H + +/* This macro only avoids the undefined behaviour from a left shift of + a negative value. It should only be used in macros that can't include + SigProc_FIX.h. In other cases, use silk_LSHIFT32(). */ +#define SAFE_SHL(a,b) ((opus_int32)((opus_uint32)(a) << (b))) + +/* (a32 * (opus_int32)((opus_int16)(b32))) >> 16 output have to be 32bit int */ +#undef silk_SMULWB +static OPUS_INLINE opus_int32 silk_SMULWB_armv4(opus_int32 a, opus_int16 b) +{ + unsigned rd_lo; + int rd_hi; + __asm__( + "#silk_SMULWB\n\t" + "smull %0, %1, %2, %3\n\t" + : "=&r"(rd_lo), "=&r"(rd_hi) + : "%r"(a), "r"(SAFE_SHL(b,16)) + ); + return rd_hi; +} +#define silk_SMULWB(a, b) (silk_SMULWB_armv4(a, b)) + +/* a32 + (b32 * (opus_int32)((opus_int16)(c32))) >> 16 output have to be 32bit int */ +#undef silk_SMLAWB +#define silk_SMLAWB(a, b, c) ((a) + silk_SMULWB(b, c)) + +/* (a32 * (b32 >> 16)) >> 16 */ +#undef silk_SMULWT +static OPUS_INLINE opus_int32 silk_SMULWT_armv4(opus_int32 a, opus_int32 b) +{ + unsigned rd_lo; + int rd_hi; + __asm__( + "#silk_SMULWT\n\t" + "smull %0, %1, %2, %3\n\t" + : "=&r"(rd_lo), "=&r"(rd_hi) + : "%r"(a), "r"(b&~0xFFFF) + ); + return rd_hi; +} +#define silk_SMULWT(a, b) (silk_SMULWT_armv4(a, b)) + +/* a32 + (b32 * (c32 >> 16)) >> 16 */ +#undef silk_SMLAWT +#define silk_SMLAWT(a, b, c) ((a) + silk_SMULWT(b, c)) + +/* (a32 * b32) >> 16 */ +#undef silk_SMULWW +static OPUS_INLINE opus_int32 silk_SMULWW_armv4(opus_int32 a, opus_int32 b) +{ + unsigned rd_lo; + int rd_hi; + __asm__( + "#silk_SMULWW\n\t" + "smull %0, %1, %2, %3\n\t" + : "=&r"(rd_lo), "=&r"(rd_hi) + : "%r"(a), "r"(b) + ); + return SAFE_SHL(rd_hi,16)+(rd_lo>>16); +} +#define silk_SMULWW(a, b) (silk_SMULWW_armv4(a, b)) + +#undef silk_SMLAWW +static OPUS_INLINE opus_int32 silk_SMLAWW_armv4(opus_int32 a, opus_int32 b, + opus_int32 c) +{ + unsigned rd_lo; + int rd_hi; + __asm__( + "#silk_SMLAWW\n\t" + "smull %0, %1, %2, %3\n\t" + : "=&r"(rd_lo), "=&r"(rd_hi) + : "%r"(b), "r"(c) + ); + return a+SAFE_SHL(rd_hi,16)+(rd_lo>>16); +} +#define silk_SMLAWW(a, b, c) (silk_SMLAWW_armv4(a, b, c)) + +#undef SAFE_SHL + +#endif /* SILK_MACROS_ARMv4_H */ diff --git a/media/libopus/silk/arm/macros_armv5e.h b/media/libopus/silk/arm/macros_armv5e.h new file mode 100644 index 0000000000..b14ec65ddb --- /dev/null +++ b/media/libopus/silk/arm/macros_armv5e.h @@ -0,0 +1,220 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Copyright (c) 2013 Parrot +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef SILK_MACROS_ARMv5E_H +#define SILK_MACROS_ARMv5E_H + +/* This macro only avoids the undefined behaviour from a left shift of + a negative value. It should only be used in macros that can't include + SigProc_FIX.h. In other cases, use silk_LSHIFT32(). */ +#define SAFE_SHL(a,b) ((opus_int32)((opus_uint32)(a) << (b))) + +/* (a32 * (opus_int32)((opus_int16)(b32))) >> 16 output have to be 32bit int */ +#undef silk_SMULWB +static OPUS_INLINE opus_int32 silk_SMULWB_armv5e(opus_int32 a, opus_int16 b) +{ + int res; + __asm__( + "#silk_SMULWB\n\t" + "smulwb %0, %1, %2\n\t" + : "=r"(res) + : "r"(a), "r"(b) + ); + return res; +} +#define silk_SMULWB(a, b) (silk_SMULWB_armv5e(a, b)) + +/* a32 + (b32 * (opus_int32)((opus_int16)(c32))) >> 16 output have to be 32bit int */ +#undef silk_SMLAWB +static OPUS_INLINE opus_int32 silk_SMLAWB_armv5e(opus_int32 a, opus_int32 b, + opus_int16 c) +{ + int res; + __asm__( + "#silk_SMLAWB\n\t" + "smlawb %0, %1, %2, %3\n\t" + : "=r"(res) + : "r"(b), "r"(c), "r"(a) + ); + return res; +} +#define silk_SMLAWB(a, b, c) (silk_SMLAWB_armv5e(a, b, c)) + +/* (a32 * (b32 >> 16)) >> 16 */ +#undef silk_SMULWT +static OPUS_INLINE opus_int32 silk_SMULWT_armv5e(opus_int32 a, opus_int32 b) +{ + int res; + __asm__( + "#silk_SMULWT\n\t" + "smulwt %0, %1, %2\n\t" + : "=r"(res) + : "r"(a), "r"(b) + ); + return res; +} +#define silk_SMULWT(a, b) (silk_SMULWT_armv5e(a, b)) + +/* a32 + (b32 * (c32 >> 16)) >> 16 */ +#undef silk_SMLAWT +static OPUS_INLINE opus_int32 silk_SMLAWT_armv5e(opus_int32 a, opus_int32 b, + opus_int32 c) +{ + int res; + __asm__( + "#silk_SMLAWT\n\t" + "smlawt %0, %1, %2, %3\n\t" + : "=r"(res) + : "r"(b), "r"(c), "r"(a) + ); + return res; +} +#define silk_SMLAWT(a, b, c) (silk_SMLAWT_armv5e(a, b, c)) + +/* (opus_int32)((opus_int16)(a3))) * (opus_int32)((opus_int16)(b32)) output have to be 32bit int */ +#undef silk_SMULBB +static OPUS_INLINE opus_int32 silk_SMULBB_armv5e(opus_int32 a, opus_int32 b) +{ + int res; + __asm__( + "#silk_SMULBB\n\t" + "smulbb %0, %1, %2\n\t" + : "=r"(res) + : "%r"(a), "r"(b) + ); + return res; +} +#define silk_SMULBB(a, b) (silk_SMULBB_armv5e(a, b)) + +/* a32 + (opus_int32)((opus_int16)(b32)) * (opus_int32)((opus_int16)(c32)) output have to be 32bit int */ +#undef silk_SMLABB +static OPUS_INLINE opus_int32 silk_SMLABB_armv5e(opus_int32 a, opus_int32 b, + opus_int32 c) +{ + int res; + __asm__( + "#silk_SMLABB\n\t" + "smlabb %0, %1, %2, %3\n\t" + : "=r"(res) + : "%r"(b), "r"(c), "r"(a) + ); + return res; +} +#define silk_SMLABB(a, b, c) (silk_SMLABB_armv5e(a, b, c)) + +/* (opus_int32)((opus_int16)(a32)) * (b32 >> 16) */ +#undef silk_SMULBT +static OPUS_INLINE opus_int32 silk_SMULBT_armv5e(opus_int32 a, opus_int32 b) +{ + int res; + __asm__( + "#silk_SMULBT\n\t" + "smulbt %0, %1, %2\n\t" + : "=r"(res) + : "r"(a), "r"(b) + ); + return res; +} +#define silk_SMULBT(a, b) (silk_SMULBT_armv5e(a, b)) + +/* a32 + (opus_int32)((opus_int16)(b32)) * (c32 >> 16) */ +#undef silk_SMLABT +static OPUS_INLINE opus_int32 silk_SMLABT_armv5e(opus_int32 a, opus_int32 b, + opus_int32 c) +{ + int res; + __asm__( + "#silk_SMLABT\n\t" + "smlabt %0, %1, %2, %3\n\t" + : "=r"(res) + : "r"(b), "r"(c), "r"(a) + ); + return res; +} +#define silk_SMLABT(a, b, c) (silk_SMLABT_armv5e(a, b, c)) + +/* add/subtract with output saturated */ +#undef silk_ADD_SAT32 +static OPUS_INLINE opus_int32 silk_ADD_SAT32_armv5e(opus_int32 a, opus_int32 b) +{ + int res; + __asm__( + "#silk_ADD_SAT32\n\t" + "qadd %0, %1, %2\n\t" + : "=r"(res) + : "%r"(a), "r"(b) + ); + return res; +} +#define silk_ADD_SAT32(a, b) (silk_ADD_SAT32_armv5e(a, b)) + +#undef silk_SUB_SAT32 +static OPUS_INLINE opus_int32 silk_SUB_SAT32_armv5e(opus_int32 a, opus_int32 b) +{ + int res; + __asm__( + "#silk_SUB_SAT32\n\t" + "qsub %0, %1, %2\n\t" + : "=r"(res) + : "r"(a), "r"(b) + ); + return res; +} +#define silk_SUB_SAT32(a, b) (silk_SUB_SAT32_armv5e(a, b)) + +#undef silk_CLZ16 +static OPUS_INLINE opus_int32 silk_CLZ16_armv5(opus_int16 in16) +{ + int res; + __asm__( + "#silk_CLZ16\n\t" + "clz %0, %1;\n" + : "=r"(res) + : "r"(SAFE_SHL(in16,16)|0x8000) + ); + return res; +} +#define silk_CLZ16(in16) (silk_CLZ16_armv5(in16)) + +#undef silk_CLZ32 +static OPUS_INLINE opus_int32 silk_CLZ32_armv5(opus_int32 in32) +{ + int res; + __asm__( + "#silk_CLZ32\n\t" + "clz %0, %1\n\t" + : "=r"(res) + : "r"(in32) + ); + return res; +} +#define silk_CLZ32(in32) (silk_CLZ32_armv5(in32)) + +#undef SAFE_SHL + +#endif /* SILK_MACROS_ARMv5E_H */ diff --git a/media/libopus/silk/biquad_alt.c b/media/libopus/silk/biquad_alt.c new file mode 100644 index 0000000000..54566a43c0 --- /dev/null +++ b/media/libopus/silk/biquad_alt.c @@ -0,0 +1,121 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +/* * + * silk_biquad_alt.c * + * * + * Second order ARMA filter * + * Can handle slowly varying filter coefficients * + * */ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "SigProc_FIX.h" + +/* Second order ARMA filter, alternative implementation */ +void silk_biquad_alt_stride1( + const opus_int16 *in, /* I input signal */ + const opus_int32 *B_Q28, /* I MA coefficients [3] */ + const opus_int32 *A_Q28, /* I AR coefficients [2] */ + opus_int32 *S, /* I/O State vector [2] */ + opus_int16 *out, /* O output signal */ + const opus_int32 len /* I signal length (must be even) */ +) +{ + /* DIRECT FORM II TRANSPOSED (uses 2 element state vector) */ + opus_int k; + opus_int32 inval, A0_U_Q28, A0_L_Q28, A1_U_Q28, A1_L_Q28, out32_Q14; + + /* Negate A_Q28 values and split in two parts */ + A0_L_Q28 = ( -A_Q28[ 0 ] ) & 0x00003FFF; /* lower part */ + A0_U_Q28 = silk_RSHIFT( -A_Q28[ 0 ], 14 ); /* upper part */ + A1_L_Q28 = ( -A_Q28[ 1 ] ) & 0x00003FFF; /* lower part */ + A1_U_Q28 = silk_RSHIFT( -A_Q28[ 1 ], 14 ); /* upper part */ + + for( k = 0; k < len; k++ ) { + /* S[ 0 ], S[ 1 ]: Q12 */ + inval = in[ k ]; + out32_Q14 = silk_LSHIFT( silk_SMLAWB( S[ 0 ], B_Q28[ 0 ], inval ), 2 ); + + S[ 0 ] = S[1] + silk_RSHIFT_ROUND( silk_SMULWB( out32_Q14, A0_L_Q28 ), 14 ); + S[ 0 ] = silk_SMLAWB( S[ 0 ], out32_Q14, A0_U_Q28 ); + S[ 0 ] = silk_SMLAWB( S[ 0 ], B_Q28[ 1 ], inval); + + S[ 1 ] = silk_RSHIFT_ROUND( silk_SMULWB( out32_Q14, A1_L_Q28 ), 14 ); + S[ 1 ] = silk_SMLAWB( S[ 1 ], out32_Q14, A1_U_Q28 ); + S[ 1 ] = silk_SMLAWB( S[ 1 ], B_Q28[ 2 ], inval ); + + /* Scale back to Q0 and saturate */ + out[ k ] = (opus_int16)silk_SAT16( silk_RSHIFT( out32_Q14 + (1<<14) - 1, 14 ) ); + } +} + +void silk_biquad_alt_stride2_c( + const opus_int16 *in, /* I input signal */ + const opus_int32 *B_Q28, /* I MA coefficients [3] */ + const opus_int32 *A_Q28, /* I AR coefficients [2] */ + opus_int32 *S, /* I/O State vector [4] */ + opus_int16 *out, /* O output signal */ + const opus_int32 len /* I signal length (must be even) */ +) +{ + /* DIRECT FORM II TRANSPOSED (uses 2 element state vector) */ + opus_int k; + opus_int32 A0_U_Q28, A0_L_Q28, A1_U_Q28, A1_L_Q28, out32_Q14[ 2 ]; + + /* Negate A_Q28 values and split in two parts */ + A0_L_Q28 = ( -A_Q28[ 0 ] ) & 0x00003FFF; /* lower part */ + A0_U_Q28 = silk_RSHIFT( -A_Q28[ 0 ], 14 ); /* upper part */ + A1_L_Q28 = ( -A_Q28[ 1 ] ) & 0x00003FFF; /* lower part */ + A1_U_Q28 = silk_RSHIFT( -A_Q28[ 1 ], 14 ); /* upper part */ + + for( k = 0; k < len; k++ ) { + /* S[ 0 ], S[ 1 ], S[ 2 ], S[ 3 ]: Q12 */ + out32_Q14[ 0 ] = silk_LSHIFT( silk_SMLAWB( S[ 0 ], B_Q28[ 0 ], in[ 2 * k + 0 ] ), 2 ); + out32_Q14[ 1 ] = silk_LSHIFT( silk_SMLAWB( S[ 2 ], B_Q28[ 0 ], in[ 2 * k + 1 ] ), 2 ); + + S[ 0 ] = S[ 1 ] + silk_RSHIFT_ROUND( silk_SMULWB( out32_Q14[ 0 ], A0_L_Q28 ), 14 ); + S[ 2 ] = S[ 3 ] + silk_RSHIFT_ROUND( silk_SMULWB( out32_Q14[ 1 ], A0_L_Q28 ), 14 ); + S[ 0 ] = silk_SMLAWB( S[ 0 ], out32_Q14[ 0 ], A0_U_Q28 ); + S[ 2 ] = silk_SMLAWB( S[ 2 ], out32_Q14[ 1 ], A0_U_Q28 ); + S[ 0 ] = silk_SMLAWB( S[ 0 ], B_Q28[ 1 ], in[ 2 * k + 0 ] ); + S[ 2 ] = silk_SMLAWB( S[ 2 ], B_Q28[ 1 ], in[ 2 * k + 1 ] ); + + S[ 1 ] = silk_RSHIFT_ROUND( silk_SMULWB( out32_Q14[ 0 ], A1_L_Q28 ), 14 ); + S[ 3 ] = silk_RSHIFT_ROUND( silk_SMULWB( out32_Q14[ 1 ], A1_L_Q28 ), 14 ); + S[ 1 ] = silk_SMLAWB( S[ 1 ], out32_Q14[ 0 ], A1_U_Q28 ); + S[ 3 ] = silk_SMLAWB( S[ 3 ], out32_Q14[ 1 ], A1_U_Q28 ); + S[ 1 ] = silk_SMLAWB( S[ 1 ], B_Q28[ 2 ], in[ 2 * k + 0 ] ); + S[ 3 ] = silk_SMLAWB( S[ 3 ], B_Q28[ 2 ], in[ 2 * k + 1 ] ); + + /* Scale back to Q0 and saturate */ + out[ 2 * k + 0 ] = (opus_int16)silk_SAT16( silk_RSHIFT( out32_Q14[ 0 ] + (1<<14) - 1, 14 ) ); + out[ 2 * k + 1 ] = (opus_int16)silk_SAT16( silk_RSHIFT( out32_Q14[ 1 ] + (1<<14) - 1, 14 ) ); + } +} diff --git a/media/libopus/silk/bwexpander.c b/media/libopus/silk/bwexpander.c new file mode 100644 index 0000000000..afa97907ec --- /dev/null +++ b/media/libopus/silk/bwexpander.c @@ -0,0 +1,51 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "SigProc_FIX.h" + +/* Chirp (bandwidth expand) LP AR filter */ +void silk_bwexpander( + opus_int16 *ar, /* I/O AR filter to be expanded (without leading 1) */ + const opus_int d, /* I Length of ar */ + opus_int32 chirp_Q16 /* I Chirp factor (typically in the range 0 to 1) */ +) +{ + opus_int i; + opus_int32 chirp_minus_one_Q16 = chirp_Q16 - 65536; + + /* NB: Dont use silk_SMULWB, instead of silk_RSHIFT_ROUND( silk_MUL(), 16 ), below. */ + /* Bias in silk_SMULWB can lead to unstable filters */ + for( i = 0; i < d - 1; i++ ) { + ar[ i ] = (opus_int16)silk_RSHIFT_ROUND( silk_MUL( chirp_Q16, ar[ i ] ), 16 ); + chirp_Q16 += silk_RSHIFT_ROUND( silk_MUL( chirp_Q16, chirp_minus_one_Q16 ), 16 ); + } + ar[ d - 1 ] = (opus_int16)silk_RSHIFT_ROUND( silk_MUL( chirp_Q16, ar[ d - 1 ] ), 16 ); +} diff --git a/media/libopus/silk/bwexpander_32.c b/media/libopus/silk/bwexpander_32.c new file mode 100644 index 0000000000..0f32b9df1a --- /dev/null +++ b/media/libopus/silk/bwexpander_32.c @@ -0,0 +1,51 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "SigProc_FIX.h" + +/* Chirp (bandwidth expand) LP AR filter. + This logic is reused in _celt_lpc(). Any bug fixes should also be applied there. */ +void silk_bwexpander_32( + opus_int32 *ar, /* I/O AR filter to be expanded (without leading 1) */ + const opus_int d, /* I Length of ar */ + opus_int32 chirp_Q16 /* I Chirp factor in Q16 */ +) +{ + opus_int i; + opus_int32 chirp_minus_one_Q16 = chirp_Q16 - 65536; + + for( i = 0; i < d - 1; i++ ) { + ar[ i ] = silk_SMULWW( chirp_Q16, ar[ i ] ); + chirp_Q16 += silk_RSHIFT_ROUND( silk_MUL( chirp_Q16, chirp_minus_one_Q16 ), 16 ); + } + ar[ d - 1 ] = silk_SMULWW( chirp_Q16, ar[ d - 1 ] ); +} + diff --git a/media/libopus/silk/check_control_input.c b/media/libopus/silk/check_control_input.c new file mode 100644 index 0000000000..739fb01f1e --- /dev/null +++ b/media/libopus/silk/check_control_input.c @@ -0,0 +1,106 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" +#include "control.h" +#include "errors.h" + +/* Check encoder control struct */ +opus_int check_control_input( + silk_EncControlStruct *encControl /* I Control structure */ +) +{ + celt_assert( encControl != NULL ); + + if( ( ( encControl->API_sampleRate != 8000 ) && + ( encControl->API_sampleRate != 12000 ) && + ( encControl->API_sampleRate != 16000 ) && + ( encControl->API_sampleRate != 24000 ) && + ( encControl->API_sampleRate != 32000 ) && + ( encControl->API_sampleRate != 44100 ) && + ( encControl->API_sampleRate != 48000 ) ) || + ( ( encControl->desiredInternalSampleRate != 8000 ) && + ( encControl->desiredInternalSampleRate != 12000 ) && + ( encControl->desiredInternalSampleRate != 16000 ) ) || + ( ( encControl->maxInternalSampleRate != 8000 ) && + ( encControl->maxInternalSampleRate != 12000 ) && + ( encControl->maxInternalSampleRate != 16000 ) ) || + ( ( encControl->minInternalSampleRate != 8000 ) && + ( encControl->minInternalSampleRate != 12000 ) && + ( encControl->minInternalSampleRate != 16000 ) ) || + ( encControl->minInternalSampleRate > encControl->desiredInternalSampleRate ) || + ( encControl->maxInternalSampleRate < encControl->desiredInternalSampleRate ) || + ( encControl->minInternalSampleRate > encControl->maxInternalSampleRate ) ) { + celt_assert( 0 ); + return SILK_ENC_FS_NOT_SUPPORTED; + } + if( encControl->payloadSize_ms != 10 && + encControl->payloadSize_ms != 20 && + encControl->payloadSize_ms != 40 && + encControl->payloadSize_ms != 60 ) { + celt_assert( 0 ); + return SILK_ENC_PACKET_SIZE_NOT_SUPPORTED; + } + if( encControl->packetLossPercentage < 0 || encControl->packetLossPercentage > 100 ) { + celt_assert( 0 ); + return SILK_ENC_INVALID_LOSS_RATE; + } + if( encControl->useDTX < 0 || encControl->useDTX > 1 ) { + celt_assert( 0 ); + return SILK_ENC_INVALID_DTX_SETTING; + } + if( encControl->useCBR < 0 || encControl->useCBR > 1 ) { + celt_assert( 0 ); + return SILK_ENC_INVALID_CBR_SETTING; + } + if( encControl->useInBandFEC < 0 || encControl->useInBandFEC > 1 ) { + celt_assert( 0 ); + return SILK_ENC_INVALID_INBAND_FEC_SETTING; + } + if( encControl->nChannelsAPI < 1 || encControl->nChannelsAPI > ENCODER_NUM_CHANNELS ) { + celt_assert( 0 ); + return SILK_ENC_INVALID_NUMBER_OF_CHANNELS_ERROR; + } + if( encControl->nChannelsInternal < 1 || encControl->nChannelsInternal > ENCODER_NUM_CHANNELS ) { + celt_assert( 0 ); + return SILK_ENC_INVALID_NUMBER_OF_CHANNELS_ERROR; + } + if( encControl->nChannelsInternal > encControl->nChannelsAPI ) { + celt_assert( 0 ); + return SILK_ENC_INVALID_NUMBER_OF_CHANNELS_ERROR; + } + if( encControl->complexity < 0 || encControl->complexity > 10 ) { + celt_assert( 0 ); + return SILK_ENC_INVALID_COMPLEXITY_SETTING; + } + + return SILK_NO_ERROR; +} diff --git a/media/libopus/silk/code_signs.c b/media/libopus/silk/code_signs.c new file mode 100644 index 0000000000..dfd1dca9a1 --- /dev/null +++ b/media/libopus/silk/code_signs.c @@ -0,0 +1,115 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" + +/*#define silk_enc_map(a) ((a) > 0 ? 1 : 0)*/ +/*#define silk_dec_map(a) ((a) > 0 ? 1 : -1)*/ +/* shifting avoids if-statement */ +#define silk_enc_map(a) ( silk_RSHIFT( (a), 15 ) + 1 ) +#define silk_dec_map(a) ( silk_LSHIFT( (a), 1 ) - 1 ) + +/* Encodes signs of excitation */ +void silk_encode_signs( + ec_enc *psRangeEnc, /* I/O Compressor data structure */ + const opus_int8 pulses[], /* I pulse signal */ + opus_int length, /* I length of input */ + const opus_int signalType, /* I Signal type */ + const opus_int quantOffsetType, /* I Quantization offset type */ + const opus_int sum_pulses[ MAX_NB_SHELL_BLOCKS ] /* I Sum of absolute pulses per block */ +) +{ + opus_int i, j, p; + opus_uint8 icdf[ 2 ]; + const opus_int8 *q_ptr; + const opus_uint8 *icdf_ptr; + + icdf[ 1 ] = 0; + q_ptr = pulses; + i = silk_SMULBB( 7, silk_ADD_LSHIFT( quantOffsetType, signalType, 1 ) ); + icdf_ptr = &silk_sign_iCDF[ i ]; + length = silk_RSHIFT( length + SHELL_CODEC_FRAME_LENGTH/2, LOG2_SHELL_CODEC_FRAME_LENGTH ); + for( i = 0; i < length; i++ ) { + p = sum_pulses[ i ]; + if( p > 0 ) { + icdf[ 0 ] = icdf_ptr[ silk_min( p & 0x1F, 6 ) ]; + for( j = 0; j < SHELL_CODEC_FRAME_LENGTH; j++ ) { + if( q_ptr[ j ] != 0 ) { + ec_enc_icdf( psRangeEnc, silk_enc_map( q_ptr[ j ]), icdf, 8 ); + } + } + } + q_ptr += SHELL_CODEC_FRAME_LENGTH; + } +} + +/* Decodes signs of excitation */ +void silk_decode_signs( + ec_dec *psRangeDec, /* I/O Compressor data structure */ + opus_int16 pulses[], /* I/O pulse signal */ + opus_int length, /* I length of input */ + const opus_int signalType, /* I Signal type */ + const opus_int quantOffsetType, /* I Quantization offset type */ + const opus_int sum_pulses[ MAX_NB_SHELL_BLOCKS ] /* I Sum of absolute pulses per block */ +) +{ + opus_int i, j, p; + opus_uint8 icdf[ 2 ]; + opus_int16 *q_ptr; + const opus_uint8 *icdf_ptr; + + icdf[ 1 ] = 0; + q_ptr = pulses; + i = silk_SMULBB( 7, silk_ADD_LSHIFT( quantOffsetType, signalType, 1 ) ); + icdf_ptr = &silk_sign_iCDF[ i ]; + length = silk_RSHIFT( length + SHELL_CODEC_FRAME_LENGTH/2, LOG2_SHELL_CODEC_FRAME_LENGTH ); + for( i = 0; i < length; i++ ) { + p = sum_pulses[ i ]; + if( p > 0 ) { + icdf[ 0 ] = icdf_ptr[ silk_min( p & 0x1F, 6 ) ]; + for( j = 0; j < SHELL_CODEC_FRAME_LENGTH; j++ ) { + if( q_ptr[ j ] > 0 ) { + /* attach sign */ +#if 0 + /* conditional implementation */ + if( ec_dec_icdf( psRangeDec, icdf, 8 ) == 0 ) { + q_ptr[ j ] = -q_ptr[ j ]; + } +#else + /* implementation with shift, subtraction, multiplication */ + q_ptr[ j ] *= silk_dec_map( ec_dec_icdf( psRangeDec, icdf, 8 ) ); +#endif + } + } + } + q_ptr += SHELL_CODEC_FRAME_LENGTH; + } +} diff --git a/media/libopus/silk/control.h b/media/libopus/silk/control.h new file mode 100644 index 0000000000..b76ec33cd6 --- /dev/null +++ b/media/libopus/silk/control.h @@ -0,0 +1,150 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef SILK_CONTROL_H +#define SILK_CONTROL_H + +#include "typedef.h" + +#ifdef __cplusplus +extern "C" +{ +#endif + +/* Decoder API flags */ +#define FLAG_DECODE_NORMAL 0 +#define FLAG_PACKET_LOST 1 +#define FLAG_DECODE_LBRR 2 + +/***********************************************/ +/* Structure for controlling encoder operation */ +/***********************************************/ +typedef struct { + /* I: Number of channels; 1/2 */ + opus_int32 nChannelsAPI; + + /* I: Number of channels; 1/2 */ + opus_int32 nChannelsInternal; + + /* I: Input signal sampling rate in Hertz; 8000/12000/16000/24000/32000/44100/48000 */ + opus_int32 API_sampleRate; + + /* I: Maximum internal sampling rate in Hertz; 8000/12000/16000 */ + opus_int32 maxInternalSampleRate; + + /* I: Minimum internal sampling rate in Hertz; 8000/12000/16000 */ + opus_int32 minInternalSampleRate; + + /* I: Soft request for internal sampling rate in Hertz; 8000/12000/16000 */ + opus_int32 desiredInternalSampleRate; + + /* I: Number of samples per packet in milliseconds; 10/20/40/60 */ + opus_int payloadSize_ms; + + /* I: Bitrate during active speech in bits/second; internally limited */ + opus_int32 bitRate; + + /* I: Uplink packet loss in percent (0-100) */ + opus_int packetLossPercentage; + + /* I: Complexity mode; 0 is lowest, 10 is highest complexity */ + opus_int complexity; + + /* I: Flag to enable in-band Forward Error Correction (FEC); 0/1 */ + opus_int useInBandFEC; + + /* I: Flag to actually code in-band Forward Error Correction (FEC) in the current packet; 0/1 */ + opus_int LBRR_coded; + + /* I: Flag to enable discontinuous transmission (DTX); 0/1 */ + opus_int useDTX; + + /* I: Flag to use constant bitrate */ + opus_int useCBR; + + /* I: Maximum number of bits allowed for the frame */ + opus_int maxBits; + + /* I: Causes a smooth downmix to mono */ + opus_int toMono; + + /* I: Opus encoder is allowing us to switch bandwidth */ + opus_int opusCanSwitch; + + /* I: Make frames as independent as possible (but still use LPC) */ + opus_int reducedDependency; + + /* O: Internal sampling rate used, in Hertz; 8000/12000/16000 */ + opus_int32 internalSampleRate; + + /* O: Flag that bandwidth switching is allowed (because low voice activity) */ + opus_int allowBandwidthSwitch; + + /* O: Flag that SILK runs in WB mode without variable LP filter (use for switching between WB/SWB/FB) */ + opus_int inWBmodeWithoutVariableLP; + + /* O: Stereo width */ + opus_int stereoWidth_Q14; + + /* O: Tells the Opus encoder we're ready to switch */ + opus_int switchReady; + + /* O: SILK Signal type */ + opus_int signalType; + + /* O: SILK offset (dithering) */ + opus_int offset; +} silk_EncControlStruct; + +/**************************************************************************/ +/* Structure for controlling decoder operation and reading decoder status */ +/**************************************************************************/ +typedef struct { + /* I: Number of channels; 1/2 */ + opus_int32 nChannelsAPI; + + /* I: Number of channels; 1/2 */ + opus_int32 nChannelsInternal; + + /* I: Output signal sampling rate in Hertz; 8000/12000/16000/24000/32000/44100/48000 */ + opus_int32 API_sampleRate; + + /* I: Internal sampling rate used, in Hertz; 8000/12000/16000 */ + opus_int32 internalSampleRate; + + /* I: Number of samples per packet in milliseconds; 10/20/40/60 */ + opus_int payloadSize_ms; + + /* O: Pitch lag of previous frame (0 if unvoiced), measured in samples at 48 kHz */ + opus_int prevPitchLag; +} silk_DecControlStruct; + +#ifdef __cplusplus +} +#endif + +#endif diff --git a/media/libopus/silk/control_SNR.c b/media/libopus/silk/control_SNR.c new file mode 100644 index 0000000000..9a6db27543 --- /dev/null +++ b/media/libopus/silk/control_SNR.c @@ -0,0 +1,113 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" +#include "tuning_parameters.h" + +/* These tables hold SNR values divided by 21 (so they fit in 8 bits) + for different target bitrates spaced at 400 bps interval. The first + 10 values are omitted (0-4 kb/s) because they're all zeros. + These tables were obtained by running different SNRs through the + encoder and measuring the active bitrate. */ +static const unsigned char silk_TargetRate_NB_21[117 - 10] = { + 0, 15, 39, 52, 61, 68, + 74, 79, 84, 88, 92, 95, 99,102,105,108,111,114,117,119,122,124, + 126,129,131,133,135,137,139,142,143,145,147,149,151,153,155,157, + 158,160,162,163,165,167,168,170,171,173,174,176,177,179,180,182, + 183,185,186,187,189,190,192,193,194,196,197,199,200,201,203,204, + 205,207,208,209,211,212,213,215,216,217,219,220,221,223,224,225, + 227,228,230,231,232,234,235,236,238,239,241,242,243,245,246,248, + 249,250,252,253,255 +}; + +static const unsigned char silk_TargetRate_MB_21[165 - 10] = { + 0, 0, 28, 43, 52, 59, + 65, 70, 74, 78, 81, 85, 87, 90, 93, 95, 98,100,102,105,107,109, + 111,113,115,116,118,120,122,123,125,127,128,130,131,133,134,136, + 137,138,140,141,143,144,145,147,148,149,151,152,153,154,156,157, + 158,159,160,162,163,164,165,166,167,168,169,171,172,173,174,175, + 176,177,178,179,180,181,182,183,184,185,186,187,188,188,189,190, + 191,192,193,194,195,196,197,198,199,200,201,202,203,203,204,205, + 206,207,208,209,210,211,212,213,214,214,215,216,217,218,219,220, + 221,222,223,224,224,225,226,227,228,229,230,231,232,233,234,235, + 236,236,237,238,239,240,241,242,243,244,245,246,247,248,249,250, + 251,252,253,254,255 +}; + +static const unsigned char silk_TargetRate_WB_21[201 - 10] = { + 0, 0, 0, 8, 29, 41, + 49, 56, 62, 66, 70, 74, 77, 80, 83, 86, 88, 91, 93, 95, 97, 99, + 101,103,105,107,108,110,112,113,115,116,118,119,121,122,123,125, + 126,127,129,130,131,132,134,135,136,137,138,140,141,142,143,144, + 145,146,147,148,149,150,151,152,153,154,156,157,158,159,159,160, + 161,162,163,164,165,166,167,168,169,170,171,171,172,173,174,175, + 176,177,177,178,179,180,181,181,182,183,184,185,185,186,187,188, + 189,189,190,191,192,192,193,194,195,195,196,197,198,198,199,200, + 200,201,202,203,203,204,205,206,206,207,208,209,209,210,211,211, + 212,213,214,214,215,216,216,217,218,219,219,220,221,221,222,223, + 224,224,225,226,226,227,228,229,229,230,231,232,232,233,234,234, + 235,236,237,237,238,239,240,240,241,242,243,243,244,245,246,246, + 247,248,249,249,250,251,252,253,255 +}; + +/* Control SNR of redidual quantizer */ +opus_int silk_control_SNR( + silk_encoder_state *psEncC, /* I/O Pointer to Silk encoder state */ + opus_int32 TargetRate_bps /* I Target max bitrate (bps) */ +) +{ + int id; + int bound; + const unsigned char *snr_table; + + psEncC->TargetRate_bps = TargetRate_bps; + if( psEncC->nb_subfr == 2 ) { + TargetRate_bps -= 2000 + psEncC->fs_kHz/16; + } + if( psEncC->fs_kHz == 8 ) { + bound = sizeof(silk_TargetRate_NB_21); + snr_table = silk_TargetRate_NB_21; + } else if( psEncC->fs_kHz == 12 ) { + bound = sizeof(silk_TargetRate_MB_21); + snr_table = silk_TargetRate_MB_21; + } else { + bound = sizeof(silk_TargetRate_WB_21); + snr_table = silk_TargetRate_WB_21; + } + id = (TargetRate_bps+200)/400; + id = silk_min(id - 10, bound-1); + if( id <= 0 ) { + psEncC->SNR_dB_Q7 = 0; + } else { + psEncC->SNR_dB_Q7 = snr_table[id]*21; + } + return SILK_NO_ERROR; +} diff --git a/media/libopus/silk/control_audio_bandwidth.c b/media/libopus/silk/control_audio_bandwidth.c new file mode 100644 index 0000000000..f6d22d8395 --- /dev/null +++ b/media/libopus/silk/control_audio_bandwidth.c @@ -0,0 +1,132 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" +#include "tuning_parameters.h" + +/* Control internal sampling rate */ +opus_int silk_control_audio_bandwidth( + silk_encoder_state *psEncC, /* I/O Pointer to Silk encoder state */ + silk_EncControlStruct *encControl /* I Control structure */ +) +{ + opus_int fs_kHz; + opus_int orig_kHz; + opus_int32 fs_Hz; + + orig_kHz = psEncC->fs_kHz; + /* Handle a bandwidth-switching reset where we need to be aware what the last sampling rate was. */ + if( orig_kHz == 0 ) { + orig_kHz = psEncC->sLP.saved_fs_kHz; + } + fs_kHz = orig_kHz; + fs_Hz = silk_SMULBB( fs_kHz, 1000 ); + if( fs_Hz == 0 ) { + /* Encoder has just been initialized */ + fs_Hz = silk_min( psEncC->desiredInternal_fs_Hz, psEncC->API_fs_Hz ); + fs_kHz = silk_DIV32_16( fs_Hz, 1000 ); + } else if( fs_Hz > psEncC->API_fs_Hz || fs_Hz > psEncC->maxInternal_fs_Hz || fs_Hz < psEncC->minInternal_fs_Hz ) { + /* Make sure internal rate is not higher than external rate or maximum allowed, or lower than minimum allowed */ + fs_Hz = psEncC->API_fs_Hz; + fs_Hz = silk_min( fs_Hz, psEncC->maxInternal_fs_Hz ); + fs_Hz = silk_max( fs_Hz, psEncC->minInternal_fs_Hz ); + fs_kHz = silk_DIV32_16( fs_Hz, 1000 ); + } else { + /* State machine for the internal sampling rate switching */ + if( psEncC->sLP.transition_frame_no >= TRANSITION_FRAMES ) { + /* Stop transition phase */ + psEncC->sLP.mode = 0; + } + if( psEncC->allow_bandwidth_switch || encControl->opusCanSwitch ) { + /* Check if we should switch down */ + if( silk_SMULBB( orig_kHz, 1000 ) > psEncC->desiredInternal_fs_Hz ) + { + /* Switch down */ + if( psEncC->sLP.mode == 0 ) { + /* New transition */ + psEncC->sLP.transition_frame_no = TRANSITION_FRAMES; + + /* Reset transition filter state */ + silk_memset( psEncC->sLP.In_LP_State, 0, sizeof( psEncC->sLP.In_LP_State ) ); + } + if( encControl->opusCanSwitch ) { + /* Stop transition phase */ + psEncC->sLP.mode = 0; + + /* Switch to a lower sample frequency */ + fs_kHz = orig_kHz == 16 ? 12 : 8; + } else { + if( psEncC->sLP.transition_frame_no <= 0 ) { + encControl->switchReady = 1; + /* Make room for redundancy */ + encControl->maxBits -= encControl->maxBits * 5 / ( encControl->payloadSize_ms + 5 ); + } else { + /* Direction: down (at double speed) */ + psEncC->sLP.mode = -2; + } + } + } + else + /* Check if we should switch up */ + if( silk_SMULBB( orig_kHz, 1000 ) < psEncC->desiredInternal_fs_Hz ) + { + /* Switch up */ + if( encControl->opusCanSwitch ) { + /* Switch to a higher sample frequency */ + fs_kHz = orig_kHz == 8 ? 12 : 16; + + /* New transition */ + psEncC->sLP.transition_frame_no = 0; + + /* Reset transition filter state */ + silk_memset( psEncC->sLP.In_LP_State, 0, sizeof( psEncC->sLP.In_LP_State ) ); + + /* Direction: up */ + psEncC->sLP.mode = 1; + } else { + if( psEncC->sLP.mode == 0 ) { + encControl->switchReady = 1; + /* Make room for redundancy */ + encControl->maxBits -= encControl->maxBits * 5 / ( encControl->payloadSize_ms + 5 ); + } else { + /* Direction: up */ + psEncC->sLP.mode = 1; + } + } + } else { + if (psEncC->sLP.mode<0) + psEncC->sLP.mode = 1; + } + } + } + + return fs_kHz; +} diff --git a/media/libopus/silk/control_codec.c b/media/libopus/silk/control_codec.c new file mode 100644 index 0000000000..784ffe66d6 --- /dev/null +++ b/media/libopus/silk/control_codec.c @@ -0,0 +1,423 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif +#ifdef FIXED_POINT +#include "main_FIX.h" +#define silk_encoder_state_Fxx silk_encoder_state_FIX +#else +#include "main_FLP.h" +#define silk_encoder_state_Fxx silk_encoder_state_FLP +#endif +#include "stack_alloc.h" +#include "tuning_parameters.h" +#include "pitch_est_defines.h" + +static opus_int silk_setup_resamplers( + silk_encoder_state_Fxx *psEnc, /* I/O */ + opus_int fs_kHz /* I */ +); + +static opus_int silk_setup_fs( + silk_encoder_state_Fxx *psEnc, /* I/O */ + opus_int fs_kHz, /* I */ + opus_int PacketSize_ms /* I */ +); + +static opus_int silk_setup_complexity( + silk_encoder_state *psEncC, /* I/O */ + opus_int Complexity /* I */ +); + +static OPUS_INLINE opus_int silk_setup_LBRR( + silk_encoder_state *psEncC, /* I/O */ + const silk_EncControlStruct *encControl /* I */ +); + + +/* Control encoder */ +opus_int silk_control_encoder( + silk_encoder_state_Fxx *psEnc, /* I/O Pointer to Silk encoder state */ + silk_EncControlStruct *encControl, /* I Control structure */ + const opus_int allow_bw_switch, /* I Flag to allow switching audio bandwidth */ + const opus_int channelNb, /* I Channel number */ + const opus_int force_fs_kHz +) +{ + opus_int fs_kHz, ret = 0; + + psEnc->sCmn.useDTX = encControl->useDTX; + psEnc->sCmn.useCBR = encControl->useCBR; + psEnc->sCmn.API_fs_Hz = encControl->API_sampleRate; + psEnc->sCmn.maxInternal_fs_Hz = encControl->maxInternalSampleRate; + psEnc->sCmn.minInternal_fs_Hz = encControl->minInternalSampleRate; + psEnc->sCmn.desiredInternal_fs_Hz = encControl->desiredInternalSampleRate; + psEnc->sCmn.useInBandFEC = encControl->useInBandFEC; + psEnc->sCmn.nChannelsAPI = encControl->nChannelsAPI; + psEnc->sCmn.nChannelsInternal = encControl->nChannelsInternal; + psEnc->sCmn.allow_bandwidth_switch = allow_bw_switch; + psEnc->sCmn.channelNb = channelNb; + + if( psEnc->sCmn.controlled_since_last_payload != 0 && psEnc->sCmn.prefillFlag == 0 ) { + if( psEnc->sCmn.API_fs_Hz != psEnc->sCmn.prev_API_fs_Hz && psEnc->sCmn.fs_kHz > 0 ) { + /* Change in API sampling rate in the middle of encoding a packet */ + ret += silk_setup_resamplers( psEnc, psEnc->sCmn.fs_kHz ); + } + return ret; + } + + /* Beyond this point we know that there are no previously coded frames in the payload buffer */ + + /********************************************/ + /* Determine internal sampling rate */ + /********************************************/ + fs_kHz = silk_control_audio_bandwidth( &psEnc->sCmn, encControl ); + if( force_fs_kHz ) { + fs_kHz = force_fs_kHz; + } + /********************************************/ + /* Prepare resampler and buffered data */ + /********************************************/ + ret += silk_setup_resamplers( psEnc, fs_kHz ); + + /********************************************/ + /* Set internal sampling frequency */ + /********************************************/ + ret += silk_setup_fs( psEnc, fs_kHz, encControl->payloadSize_ms ); + + /********************************************/ + /* Set encoding complexity */ + /********************************************/ + ret += silk_setup_complexity( &psEnc->sCmn, encControl->complexity ); + + /********************************************/ + /* Set packet loss rate measured by farend */ + /********************************************/ + psEnc->sCmn.PacketLoss_perc = encControl->packetLossPercentage; + + /********************************************/ + /* Set LBRR usage */ + /********************************************/ + ret += silk_setup_LBRR( &psEnc->sCmn, encControl ); + + psEnc->sCmn.controlled_since_last_payload = 1; + + return ret; +} + +static opus_int silk_setup_resamplers( + silk_encoder_state_Fxx *psEnc, /* I/O */ + opus_int fs_kHz /* I */ +) +{ + opus_int ret = SILK_NO_ERROR; + SAVE_STACK; + + if( psEnc->sCmn.fs_kHz != fs_kHz || psEnc->sCmn.prev_API_fs_Hz != psEnc->sCmn.API_fs_Hz ) + { + if( psEnc->sCmn.fs_kHz == 0 ) { + /* Initialize the resampler for enc_API.c preparing resampling from API_fs_Hz to fs_kHz */ + ret += silk_resampler_init( &psEnc->sCmn.resampler_state, psEnc->sCmn.API_fs_Hz, fs_kHz * 1000, 1 ); + } else { + VARDECL( opus_int16, x_buf_API_fs_Hz ); + VARDECL( silk_resampler_state_struct, temp_resampler_state ); +#ifdef FIXED_POINT + opus_int16 *x_bufFIX = psEnc->x_buf; +#else + VARDECL( opus_int16, x_bufFIX ); + opus_int32 new_buf_samples; +#endif + opus_int32 api_buf_samples; + opus_int32 old_buf_samples; + opus_int32 buf_length_ms; + + buf_length_ms = silk_LSHIFT( psEnc->sCmn.nb_subfr * 5, 1 ) + LA_SHAPE_MS; + old_buf_samples = buf_length_ms * psEnc->sCmn.fs_kHz; + +#ifndef FIXED_POINT + new_buf_samples = buf_length_ms * fs_kHz; + ALLOC( x_bufFIX, silk_max( old_buf_samples, new_buf_samples ), + opus_int16 ); + silk_float2short_array( x_bufFIX, psEnc->x_buf, old_buf_samples ); +#endif + + /* Initialize resampler for temporary resampling of x_buf data to API_fs_Hz */ + ALLOC( temp_resampler_state, 1, silk_resampler_state_struct ); + ret += silk_resampler_init( temp_resampler_state, silk_SMULBB( psEnc->sCmn.fs_kHz, 1000 ), psEnc->sCmn.API_fs_Hz, 0 ); + + /* Calculate number of samples to temporarily upsample */ + api_buf_samples = buf_length_ms * silk_DIV32_16( psEnc->sCmn.API_fs_Hz, 1000 ); + + /* Temporary resampling of x_buf data to API_fs_Hz */ + ALLOC( x_buf_API_fs_Hz, api_buf_samples, opus_int16 ); + ret += silk_resampler( temp_resampler_state, x_buf_API_fs_Hz, x_bufFIX, old_buf_samples ); + + /* Initialize the resampler for enc_API.c preparing resampling from API_fs_Hz to fs_kHz */ + ret += silk_resampler_init( &psEnc->sCmn.resampler_state, psEnc->sCmn.API_fs_Hz, silk_SMULBB( fs_kHz, 1000 ), 1 ); + + /* Correct resampler state by resampling buffered data from API_fs_Hz to fs_kHz */ + ret += silk_resampler( &psEnc->sCmn.resampler_state, x_bufFIX, x_buf_API_fs_Hz, api_buf_samples ); + +#ifndef FIXED_POINT + silk_short2float_array( psEnc->x_buf, x_bufFIX, new_buf_samples); +#endif + } + } + + psEnc->sCmn.prev_API_fs_Hz = psEnc->sCmn.API_fs_Hz; + + RESTORE_STACK; + return ret; +} + +static opus_int silk_setup_fs( + silk_encoder_state_Fxx *psEnc, /* I/O */ + opus_int fs_kHz, /* I */ + opus_int PacketSize_ms /* I */ +) +{ + opus_int ret = SILK_NO_ERROR; + + /* Set packet size */ + if( PacketSize_ms != psEnc->sCmn.PacketSize_ms ) { + if( ( PacketSize_ms != 10 ) && + ( PacketSize_ms != 20 ) && + ( PacketSize_ms != 40 ) && + ( PacketSize_ms != 60 ) ) { + ret = SILK_ENC_PACKET_SIZE_NOT_SUPPORTED; + } + if( PacketSize_ms <= 10 ) { + psEnc->sCmn.nFramesPerPacket = 1; + psEnc->sCmn.nb_subfr = PacketSize_ms == 10 ? 2 : 1; + psEnc->sCmn.frame_length = silk_SMULBB( PacketSize_ms, fs_kHz ); + psEnc->sCmn.pitch_LPC_win_length = silk_SMULBB( FIND_PITCH_LPC_WIN_MS_2_SF, fs_kHz ); + if( psEnc->sCmn.fs_kHz == 8 ) { + psEnc->sCmn.pitch_contour_iCDF = silk_pitch_contour_10_ms_NB_iCDF; + } else { + psEnc->sCmn.pitch_contour_iCDF = silk_pitch_contour_10_ms_iCDF; + } + } else { + psEnc->sCmn.nFramesPerPacket = silk_DIV32_16( PacketSize_ms, MAX_FRAME_LENGTH_MS ); + psEnc->sCmn.nb_subfr = MAX_NB_SUBFR; + psEnc->sCmn.frame_length = silk_SMULBB( 20, fs_kHz ); + psEnc->sCmn.pitch_LPC_win_length = silk_SMULBB( FIND_PITCH_LPC_WIN_MS, fs_kHz ); + if( psEnc->sCmn.fs_kHz == 8 ) { + psEnc->sCmn.pitch_contour_iCDF = silk_pitch_contour_NB_iCDF; + } else { + psEnc->sCmn.pitch_contour_iCDF = silk_pitch_contour_iCDF; + } + } + psEnc->sCmn.PacketSize_ms = PacketSize_ms; + psEnc->sCmn.TargetRate_bps = 0; /* trigger new SNR computation */ + } + + /* Set internal sampling frequency */ + celt_assert( fs_kHz == 8 || fs_kHz == 12 || fs_kHz == 16 ); + celt_assert( psEnc->sCmn.nb_subfr == 2 || psEnc->sCmn.nb_subfr == 4 ); + if( psEnc->sCmn.fs_kHz != fs_kHz ) { + /* reset part of the state */ + silk_memset( &psEnc->sShape, 0, sizeof( psEnc->sShape ) ); + silk_memset( &psEnc->sCmn.sNSQ, 0, sizeof( psEnc->sCmn.sNSQ ) ); + silk_memset( psEnc->sCmn.prev_NLSFq_Q15, 0, sizeof( psEnc->sCmn.prev_NLSFq_Q15 ) ); + silk_memset( &psEnc->sCmn.sLP.In_LP_State, 0, sizeof( psEnc->sCmn.sLP.In_LP_State ) ); + psEnc->sCmn.inputBufIx = 0; + psEnc->sCmn.nFramesEncoded = 0; + psEnc->sCmn.TargetRate_bps = 0; /* trigger new SNR computation */ + + /* Initialize non-zero parameters */ + psEnc->sCmn.prevLag = 100; + psEnc->sCmn.first_frame_after_reset = 1; + psEnc->sShape.LastGainIndex = 10; + psEnc->sCmn.sNSQ.lagPrev = 100; + psEnc->sCmn.sNSQ.prev_gain_Q16 = 65536; + psEnc->sCmn.prevSignalType = TYPE_NO_VOICE_ACTIVITY; + + psEnc->sCmn.fs_kHz = fs_kHz; + if( psEnc->sCmn.fs_kHz == 8 ) { + if( psEnc->sCmn.nb_subfr == MAX_NB_SUBFR ) { + psEnc->sCmn.pitch_contour_iCDF = silk_pitch_contour_NB_iCDF; + } else { + psEnc->sCmn.pitch_contour_iCDF = silk_pitch_contour_10_ms_NB_iCDF; + } + } else { + if( psEnc->sCmn.nb_subfr == MAX_NB_SUBFR ) { + psEnc->sCmn.pitch_contour_iCDF = silk_pitch_contour_iCDF; + } else { + psEnc->sCmn.pitch_contour_iCDF = silk_pitch_contour_10_ms_iCDF; + } + } + if( psEnc->sCmn.fs_kHz == 8 || psEnc->sCmn.fs_kHz == 12 ) { + psEnc->sCmn.predictLPCOrder = MIN_LPC_ORDER; + psEnc->sCmn.psNLSF_CB = &silk_NLSF_CB_NB_MB; + } else { + psEnc->sCmn.predictLPCOrder = MAX_LPC_ORDER; + psEnc->sCmn.psNLSF_CB = &silk_NLSF_CB_WB; + } + psEnc->sCmn.subfr_length = SUB_FRAME_LENGTH_MS * fs_kHz; + psEnc->sCmn.frame_length = silk_SMULBB( psEnc->sCmn.subfr_length, psEnc->sCmn.nb_subfr ); + psEnc->sCmn.ltp_mem_length = silk_SMULBB( LTP_MEM_LENGTH_MS, fs_kHz ); + psEnc->sCmn.la_pitch = silk_SMULBB( LA_PITCH_MS, fs_kHz ); + psEnc->sCmn.max_pitch_lag = silk_SMULBB( 18, fs_kHz ); + if( psEnc->sCmn.nb_subfr == MAX_NB_SUBFR ) { + psEnc->sCmn.pitch_LPC_win_length = silk_SMULBB( FIND_PITCH_LPC_WIN_MS, fs_kHz ); + } else { + psEnc->sCmn.pitch_LPC_win_length = silk_SMULBB( FIND_PITCH_LPC_WIN_MS_2_SF, fs_kHz ); + } + if( psEnc->sCmn.fs_kHz == 16 ) { + psEnc->sCmn.pitch_lag_low_bits_iCDF = silk_uniform8_iCDF; + } else if( psEnc->sCmn.fs_kHz == 12 ) { + psEnc->sCmn.pitch_lag_low_bits_iCDF = silk_uniform6_iCDF; + } else { + psEnc->sCmn.pitch_lag_low_bits_iCDF = silk_uniform4_iCDF; + } + } + + /* Check that settings are valid */ + celt_assert( ( psEnc->sCmn.subfr_length * psEnc->sCmn.nb_subfr ) == psEnc->sCmn.frame_length ); + + return ret; +} + +static opus_int silk_setup_complexity( + silk_encoder_state *psEncC, /* I/O */ + opus_int Complexity /* I */ +) +{ + opus_int ret = 0; + + /* Set encoding complexity */ + celt_assert( Complexity >= 0 && Complexity <= 10 ); + if( Complexity < 1 ) { + psEncC->pitchEstimationComplexity = SILK_PE_MIN_COMPLEX; + psEncC->pitchEstimationThreshold_Q16 = SILK_FIX_CONST( 0.8, 16 ); + psEncC->pitchEstimationLPCOrder = 6; + psEncC->shapingLPCOrder = 12; + psEncC->la_shape = 3 * psEncC->fs_kHz; + psEncC->nStatesDelayedDecision = 1; + psEncC->useInterpolatedNLSFs = 0; + psEncC->NLSF_MSVQ_Survivors = 2; + psEncC->warping_Q16 = 0; + } else if( Complexity < 2 ) { + psEncC->pitchEstimationComplexity = SILK_PE_MID_COMPLEX; + psEncC->pitchEstimationThreshold_Q16 = SILK_FIX_CONST( 0.76, 16 ); + psEncC->pitchEstimationLPCOrder = 8; + psEncC->shapingLPCOrder = 14; + psEncC->la_shape = 5 * psEncC->fs_kHz; + psEncC->nStatesDelayedDecision = 1; + psEncC->useInterpolatedNLSFs = 0; + psEncC->NLSF_MSVQ_Survivors = 3; + psEncC->warping_Q16 = 0; + } else if( Complexity < 3 ) { + psEncC->pitchEstimationComplexity = SILK_PE_MIN_COMPLEX; + psEncC->pitchEstimationThreshold_Q16 = SILK_FIX_CONST( 0.8, 16 ); + psEncC->pitchEstimationLPCOrder = 6; + psEncC->shapingLPCOrder = 12; + psEncC->la_shape = 3 * psEncC->fs_kHz; + psEncC->nStatesDelayedDecision = 2; + psEncC->useInterpolatedNLSFs = 0; + psEncC->NLSF_MSVQ_Survivors = 2; + psEncC->warping_Q16 = 0; + } else if( Complexity < 4 ) { + psEncC->pitchEstimationComplexity = SILK_PE_MID_COMPLEX; + psEncC->pitchEstimationThreshold_Q16 = SILK_FIX_CONST( 0.76, 16 ); + psEncC->pitchEstimationLPCOrder = 8; + psEncC->shapingLPCOrder = 14; + psEncC->la_shape = 5 * psEncC->fs_kHz; + psEncC->nStatesDelayedDecision = 2; + psEncC->useInterpolatedNLSFs = 0; + psEncC->NLSF_MSVQ_Survivors = 4; + psEncC->warping_Q16 = 0; + } else if( Complexity < 6 ) { + psEncC->pitchEstimationComplexity = SILK_PE_MID_COMPLEX; + psEncC->pitchEstimationThreshold_Q16 = SILK_FIX_CONST( 0.74, 16 ); + psEncC->pitchEstimationLPCOrder = 10; + psEncC->shapingLPCOrder = 16; + psEncC->la_shape = 5 * psEncC->fs_kHz; + psEncC->nStatesDelayedDecision = 2; + psEncC->useInterpolatedNLSFs = 1; + psEncC->NLSF_MSVQ_Survivors = 6; + psEncC->warping_Q16 = psEncC->fs_kHz * SILK_FIX_CONST( WARPING_MULTIPLIER, 16 ); + } else if( Complexity < 8 ) { + psEncC->pitchEstimationComplexity = SILK_PE_MID_COMPLEX; + psEncC->pitchEstimationThreshold_Q16 = SILK_FIX_CONST( 0.72, 16 ); + psEncC->pitchEstimationLPCOrder = 12; + psEncC->shapingLPCOrder = 20; + psEncC->la_shape = 5 * psEncC->fs_kHz; + psEncC->nStatesDelayedDecision = 3; + psEncC->useInterpolatedNLSFs = 1; + psEncC->NLSF_MSVQ_Survivors = 8; + psEncC->warping_Q16 = psEncC->fs_kHz * SILK_FIX_CONST( WARPING_MULTIPLIER, 16 ); + } else { + psEncC->pitchEstimationComplexity = SILK_PE_MAX_COMPLEX; + psEncC->pitchEstimationThreshold_Q16 = SILK_FIX_CONST( 0.7, 16 ); + psEncC->pitchEstimationLPCOrder = 16; + psEncC->shapingLPCOrder = 24; + psEncC->la_shape = 5 * psEncC->fs_kHz; + psEncC->nStatesDelayedDecision = MAX_DEL_DEC_STATES; + psEncC->useInterpolatedNLSFs = 1; + psEncC->NLSF_MSVQ_Survivors = 16; + psEncC->warping_Q16 = psEncC->fs_kHz * SILK_FIX_CONST( WARPING_MULTIPLIER, 16 ); + } + + /* Do not allow higher pitch estimation LPC order than predict LPC order */ + psEncC->pitchEstimationLPCOrder = silk_min_int( psEncC->pitchEstimationLPCOrder, psEncC->predictLPCOrder ); + psEncC->shapeWinLength = SUB_FRAME_LENGTH_MS * psEncC->fs_kHz + 2 * psEncC->la_shape; + psEncC->Complexity = Complexity; + + celt_assert( psEncC->pitchEstimationLPCOrder <= MAX_FIND_PITCH_LPC_ORDER ); + celt_assert( psEncC->shapingLPCOrder <= MAX_SHAPE_LPC_ORDER ); + celt_assert( psEncC->nStatesDelayedDecision <= MAX_DEL_DEC_STATES ); + celt_assert( psEncC->warping_Q16 <= 32767 ); + celt_assert( psEncC->la_shape <= LA_SHAPE_MAX ); + celt_assert( psEncC->shapeWinLength <= SHAPE_LPC_WIN_MAX ); + + return ret; +} + +static OPUS_INLINE opus_int silk_setup_LBRR( + silk_encoder_state *psEncC, /* I/O */ + const silk_EncControlStruct *encControl /* I */ +) +{ + opus_int LBRR_in_previous_packet, ret = SILK_NO_ERROR; + + LBRR_in_previous_packet = psEncC->LBRR_enabled; + psEncC->LBRR_enabled = encControl->LBRR_coded; + if( psEncC->LBRR_enabled ) { + /* Set gain increase for coding LBRR excitation */ + if( LBRR_in_previous_packet == 0 ) { + /* Previous packet did not have LBRR, and was therefore coded at a higher bitrate */ + psEncC->LBRR_GainIncreases = 7; + } else { + psEncC->LBRR_GainIncreases = silk_max_int( 7 - silk_SMULWB( (opus_int32)psEncC->PacketLoss_perc, SILK_FIX_CONST( 0.2, 16 ) ), 3 ); + } + } + + return ret; +} diff --git a/media/libopus/silk/debug.c b/media/libopus/silk/debug.c new file mode 100644 index 0000000000..46a24a47e4 --- /dev/null +++ b/media/libopus/silk/debug.c @@ -0,0 +1,174 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +typedef int prevent_empty_translation_unit_warning; + +#include "debug.h" + +#if SILK_DEBUG || SILK_TIC_TOC +#include "SigProc_FIX.h" +#endif + +#if SILK_TIC_TOC + +#if (defined(_WIN32) || defined(_WINCE)) +#include <windows.h> /* timer */ +#else /* Linux or Mac*/ +#include <sys/time.h> +#endif + +#ifdef _WIN32 +unsigned long silk_GetHighResolutionTime(void) /* O time in usec*/ +{ + /* Returns a time counter in microsec */ + /* the resolution is platform dependent */ + /* but is typically 1.62 us resolution */ + LARGE_INTEGER lpPerformanceCount; + LARGE_INTEGER lpFrequency; + QueryPerformanceCounter(&lpPerformanceCount); + QueryPerformanceFrequency(&lpFrequency); + return (unsigned long)((1000000*(lpPerformanceCount.QuadPart)) / lpFrequency.QuadPart); +} +#else /* Linux or Mac*/ +unsigned long GetHighResolutionTime(void) /* O time in usec*/ +{ + struct timeval tv; + gettimeofday(&tv, 0); + return((tv.tv_sec*1000000)+(tv.tv_usec)); +} +#endif + +int silk_Timer_nTimers = 0; +int silk_Timer_depth_ctr = 0; +char silk_Timer_tags[silk_NUM_TIMERS_MAX][silk_NUM_TIMERS_MAX_TAG_LEN]; +#ifdef _WIN32 +LARGE_INTEGER silk_Timer_start[silk_NUM_TIMERS_MAX]; +#else +unsigned long silk_Timer_start[silk_NUM_TIMERS_MAX]; +#endif +unsigned int silk_Timer_cnt[silk_NUM_TIMERS_MAX]; +opus_int64 silk_Timer_min[silk_NUM_TIMERS_MAX]; +opus_int64 silk_Timer_sum[silk_NUM_TIMERS_MAX]; +opus_int64 silk_Timer_max[silk_NUM_TIMERS_MAX]; +opus_int64 silk_Timer_depth[silk_NUM_TIMERS_MAX]; + +#ifdef _WIN32 +void silk_TimerSave(char *file_name) +{ + if( silk_Timer_nTimers > 0 ) + { + int k; + FILE *fp; + LARGE_INTEGER lpFrequency; + LARGE_INTEGER lpPerformanceCount1, lpPerformanceCount2; + int del = 0x7FFFFFFF; + double avg, sum_avg; + /* estimate overhead of calling performance counters */ + for( k = 0; k < 1000; k++ ) { + QueryPerformanceCounter(&lpPerformanceCount1); + QueryPerformanceCounter(&lpPerformanceCount2); + lpPerformanceCount2.QuadPart -= lpPerformanceCount1.QuadPart; + if( (int)lpPerformanceCount2.LowPart < del ) + del = lpPerformanceCount2.LowPart; + } + QueryPerformanceFrequency(&lpFrequency); + /* print results to file */ + sum_avg = 0.0f; + for( k = 0; k < silk_Timer_nTimers; k++ ) { + if (silk_Timer_depth[k] == 0) { + sum_avg += (1e6 * silk_Timer_sum[k] / silk_Timer_cnt[k] - del) / lpFrequency.QuadPart * silk_Timer_cnt[k]; + } + } + fp = fopen(file_name, "w"); + fprintf(fp, " min avg %% max count\n"); + for( k = 0; k < silk_Timer_nTimers; k++ ) { + if (silk_Timer_depth[k] == 0) { + fprintf(fp, "%-28s", silk_Timer_tags[k]); + } else if (silk_Timer_depth[k] == 1) { + fprintf(fp, " %-27s", silk_Timer_tags[k]); + } else if (silk_Timer_depth[k] == 2) { + fprintf(fp, " %-26s", silk_Timer_tags[k]); + } else if (silk_Timer_depth[k] == 3) { + fprintf(fp, " %-25s", silk_Timer_tags[k]); + } else { + fprintf(fp, " %-24s", silk_Timer_tags[k]); + } + avg = (1e6 * silk_Timer_sum[k] / silk_Timer_cnt[k] - del) / lpFrequency.QuadPart; + fprintf(fp, "%8.2f", (1e6 * (silk_max_64(silk_Timer_min[k] - del, 0))) / lpFrequency.QuadPart); + fprintf(fp, "%12.2f %6.2f", avg, 100.0 * avg / sum_avg * silk_Timer_cnt[k]); + fprintf(fp, "%12.2f", (1e6 * (silk_max_64(silk_Timer_max[k] - del, 0))) / lpFrequency.QuadPart); + fprintf(fp, "%10d\n", silk_Timer_cnt[k]); + } + fprintf(fp, " microseconds\n"); + fclose(fp); + } +} +#else +void silk_TimerSave(char *file_name) +{ + if( silk_Timer_nTimers > 0 ) + { + int k; + FILE *fp; + /* print results to file */ + fp = fopen(file_name, "w"); + fprintf(fp, " min avg max count\n"); + for( k = 0; k < silk_Timer_nTimers; k++ ) + { + if (silk_Timer_depth[k] == 0) { + fprintf(fp, "%-28s", silk_Timer_tags[k]); + } else if (silk_Timer_depth[k] == 1) { + fprintf(fp, " %-27s", silk_Timer_tags[k]); + } else if (silk_Timer_depth[k] == 2) { + fprintf(fp, " %-26s", silk_Timer_tags[k]); + } else if (silk_Timer_depth[k] == 3) { + fprintf(fp, " %-25s", silk_Timer_tags[k]); + } else { + fprintf(fp, " %-24s", silk_Timer_tags[k]); + } + fprintf(fp, "%d ", silk_Timer_min[k]); + fprintf(fp, "%f ", (double)silk_Timer_sum[k] / (double)silk_Timer_cnt[k]); + fprintf(fp, "%d ", silk_Timer_max[k]); + fprintf(fp, "%10d\n", silk_Timer_cnt[k]); + } + fprintf(fp, " microseconds\n"); + fclose(fp); + } +} +#endif + +#endif /* SILK_TIC_TOC */ + +#if SILK_DEBUG +FILE *silk_debug_store_fp[ silk_NUM_STORES_MAX ]; +int silk_debug_store_count = 0; +#endif /* SILK_DEBUG */ + diff --git a/media/libopus/silk/debug.h b/media/libopus/silk/debug.h new file mode 100644 index 0000000000..36163e478d --- /dev/null +++ b/media/libopus/silk/debug.h @@ -0,0 +1,267 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef SILK_DEBUG_H +#define SILK_DEBUG_H + +/* Set to 1 to enable DEBUG_STORE_DATA() macros for dumping + * intermediate signals from the codec. + */ +#define SILK_DEBUG 0 + +/* Flag for using timers */ +#define SILK_TIC_TOC 0 + +#if SILK_DEBUG || SILK_TIC_TOC +#include "typedef.h" +#include <string.h> /* strcpy, strcmp */ +#include <stdio.h> /* file writing */ +#endif + +#ifdef __cplusplus +extern "C" +{ +#endif + +#if SILK_TIC_TOC + +unsigned long GetHighResolutionTime(void); /* O time in usec*/ + +#if (defined(_WIN32) || defined(_WINCE)) +#include <windows.h> /* timer */ +#else /* Linux or Mac*/ +#include <sys/time.h> +#endif + +/*********************************/ +/* timer functions for profiling */ +/*********************************/ +/* example: */ +/* */ +/* TIC(LPC) */ +/* do_LPC(in_vec, order, acoef); // do LPC analysis */ +/* TOC(LPC) */ +/* */ +/* and call the following just before exiting (from main) */ +/* */ +/* silk_TimerSave("silk_TimingData.txt"); */ +/* */ +/* results are now in silk_TimingData.txt */ + +void silk_TimerSave(char *file_name); + +/* max number of timers (in different locations) */ +#define silk_NUM_TIMERS_MAX 50 +/* max length of name tags in TIC(..), TOC(..) */ +#define silk_NUM_TIMERS_MAX_TAG_LEN 30 + +extern int silk_Timer_nTimers; +extern int silk_Timer_depth_ctr; +extern char silk_Timer_tags[silk_NUM_TIMERS_MAX][silk_NUM_TIMERS_MAX_TAG_LEN]; +#ifdef _WIN32 +extern LARGE_INTEGER silk_Timer_start[silk_NUM_TIMERS_MAX]; +#else +extern unsigned long silk_Timer_start[silk_NUM_TIMERS_MAX]; +#endif +extern unsigned int silk_Timer_cnt[silk_NUM_TIMERS_MAX]; +extern opus_int64 silk_Timer_sum[silk_NUM_TIMERS_MAX]; +extern opus_int64 silk_Timer_max[silk_NUM_TIMERS_MAX]; +extern opus_int64 silk_Timer_min[silk_NUM_TIMERS_MAX]; +extern opus_int64 silk_Timer_depth[silk_NUM_TIMERS_MAX]; + +/* WARNING: TIC()/TOC can measure only up to 0.1 seconds at a time */ +#ifdef _WIN32 +#define TIC(TAG_NAME) { \ + static int init = 0; \ + static int ID = -1; \ + if( init == 0 ) \ + { \ + int k; \ + init = 1; \ + for( k = 0; k < silk_Timer_nTimers; k++ ) { \ + if( strcmp(silk_Timer_tags[k], #TAG_NAME) == 0 ) { \ + ID = k; \ + break; \ + } \ + } \ + if (ID == -1) { \ + ID = silk_Timer_nTimers; \ + silk_Timer_nTimers++; \ + silk_Timer_depth[ID] = silk_Timer_depth_ctr; \ + strcpy(silk_Timer_tags[ID], #TAG_NAME); \ + silk_Timer_cnt[ID] = 0; \ + silk_Timer_sum[ID] = 0; \ + silk_Timer_min[ID] = 0xFFFFFFFF; \ + silk_Timer_max[ID] = 0; \ + } \ + } \ + silk_Timer_depth_ctr++; \ + QueryPerformanceCounter(&silk_Timer_start[ID]); \ +} +#else +#define TIC(TAG_NAME) { \ + static int init = 0; \ + static int ID = -1; \ + if( init == 0 ) \ + { \ + int k; \ + init = 1; \ + for( k = 0; k < silk_Timer_nTimers; k++ ) { \ + if( strcmp(silk_Timer_tags[k], #TAG_NAME) == 0 ) { \ + ID = k; \ + break; \ + } \ + } \ + if (ID == -1) { \ + ID = silk_Timer_nTimers; \ + silk_Timer_nTimers++; \ + silk_Timer_depth[ID] = silk_Timer_depth_ctr; \ + strcpy(silk_Timer_tags[ID], #TAG_NAME); \ + silk_Timer_cnt[ID] = 0; \ + silk_Timer_sum[ID] = 0; \ + silk_Timer_min[ID] = 0xFFFFFFFF; \ + silk_Timer_max[ID] = 0; \ + } \ + } \ + silk_Timer_depth_ctr++; \ + silk_Timer_start[ID] = GetHighResolutionTime(); \ +} +#endif + +#ifdef _WIN32 +#define TOC(TAG_NAME) { \ + LARGE_INTEGER lpPerformanceCount; \ + static int init = 0; \ + static int ID = 0; \ + if( init == 0 ) \ + { \ + int k; \ + init = 1; \ + for( k = 0; k < silk_Timer_nTimers; k++ ) { \ + if( strcmp(silk_Timer_tags[k], #TAG_NAME) == 0 ) { \ + ID = k; \ + break; \ + } \ + } \ + } \ + QueryPerformanceCounter(&lpPerformanceCount); \ + lpPerformanceCount.QuadPart -= silk_Timer_start[ID].QuadPart; \ + if((lpPerformanceCount.QuadPart < 100000000) && \ + (lpPerformanceCount.QuadPart >= 0)) { \ + silk_Timer_cnt[ID]++; \ + silk_Timer_sum[ID] += lpPerformanceCount.QuadPart; \ + if( lpPerformanceCount.QuadPart > silk_Timer_max[ID] ) \ + silk_Timer_max[ID] = lpPerformanceCount.QuadPart; \ + if( lpPerformanceCount.QuadPart < silk_Timer_min[ID] ) \ + silk_Timer_min[ID] = lpPerformanceCount.QuadPart; \ + } \ + silk_Timer_depth_ctr--; \ +} +#else +#define TOC(TAG_NAME) { \ + unsigned long endTime; \ + static int init = 0; \ + static int ID = 0; \ + if( init == 0 ) \ + { \ + int k; \ + init = 1; \ + for( k = 0; k < silk_Timer_nTimers; k++ ) { \ + if( strcmp(silk_Timer_tags[k], #TAG_NAME) == 0 ) { \ + ID = k; \ + break; \ + } \ + } \ + } \ + endTime = GetHighResolutionTime(); \ + endTime -= silk_Timer_start[ID]; \ + if((endTime < 100000000) && \ + (endTime >= 0)) { \ + silk_Timer_cnt[ID]++; \ + silk_Timer_sum[ID] += endTime; \ + if( endTime > silk_Timer_max[ID] ) \ + silk_Timer_max[ID] = endTime; \ + if( endTime < silk_Timer_min[ID] ) \ + silk_Timer_min[ID] = endTime; \ + } \ + silk_Timer_depth_ctr--; \ +} +#endif + +#else /* SILK_TIC_TOC */ + +/* define macros as empty strings */ +#define TIC(TAG_NAME) +#define TOC(TAG_NAME) +#define silk_TimerSave(FILE_NAME) + +#endif /* SILK_TIC_TOC */ + + +#if SILK_DEBUG +/************************************/ +/* write data to file for debugging */ +/************************************/ +/* Example: DEBUG_STORE_DATA(testfile.pcm, &RIN[0], 160*sizeof(opus_int16)); */ + +#define silk_NUM_STORES_MAX 100 +extern FILE *silk_debug_store_fp[ silk_NUM_STORES_MAX ]; +extern int silk_debug_store_count; + +/* Faster way of storing the data */ +#define DEBUG_STORE_DATA( FILE_NAME, DATA_PTR, N_BYTES ) { \ + static opus_int init = 0, cnt = 0; \ + static FILE **fp; \ + if (init == 0) { \ + init = 1; \ + cnt = silk_debug_store_count++; \ + silk_debug_store_fp[ cnt ] = fopen(#FILE_NAME, "wb"); \ + } \ + fwrite((DATA_PTR), (N_BYTES), 1, silk_debug_store_fp[ cnt ]); \ +} + +/* Call this at the end of main() */ +#define SILK_DEBUG_STORE_CLOSE_FILES { \ + opus_int i; \ + for( i = 0; i < silk_debug_store_count; i++ ) { \ + fclose( silk_debug_store_fp[ i ] ); \ + } \ +} + +#else /* SILK_DEBUG */ + +/* define macros as empty strings */ +#define DEBUG_STORE_DATA(FILE_NAME, DATA_PTR, N_BYTES) +#define SILK_DEBUG_STORE_CLOSE_FILES + +#endif /* SILK_DEBUG */ + +#ifdef __cplusplus +} +#endif + +#endif /* SILK_DEBUG_H */ diff --git a/media/libopus/silk/dec_API.c b/media/libopus/silk/dec_API.c new file mode 100644 index 0000000000..7d5ca7fb9f --- /dev/null +++ b/media/libopus/silk/dec_API.c @@ -0,0 +1,419 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif +#include "API.h" +#include "main.h" +#include "stack_alloc.h" +#include "os_support.h" + +/************************/ +/* Decoder Super Struct */ +/************************/ +typedef struct { + silk_decoder_state channel_state[ DECODER_NUM_CHANNELS ]; + stereo_dec_state sStereo; + opus_int nChannelsAPI; + opus_int nChannelsInternal; + opus_int prev_decode_only_middle; +} silk_decoder; + +/*********************/ +/* Decoder functions */ +/*********************/ + +opus_int silk_Get_Decoder_Size( /* O Returns error code */ + opus_int *decSizeBytes /* O Number of bytes in SILK decoder state */ +) +{ + opus_int ret = SILK_NO_ERROR; + + *decSizeBytes = sizeof( silk_decoder ); + + return ret; +} + +/* Reset decoder state */ +opus_int silk_InitDecoder( /* O Returns error code */ + void *decState /* I/O State */ +) +{ + opus_int n, ret = SILK_NO_ERROR; + silk_decoder_state *channel_state = ((silk_decoder *)decState)->channel_state; + + for( n = 0; n < DECODER_NUM_CHANNELS; n++ ) { + ret = silk_init_decoder( &channel_state[ n ] ); + } + silk_memset(&((silk_decoder *)decState)->sStereo, 0, sizeof(((silk_decoder *)decState)->sStereo)); + /* Not strictly needed, but it's cleaner that way */ + ((silk_decoder *)decState)->prev_decode_only_middle = 0; + + return ret; +} + +/* Decode a frame */ +opus_int silk_Decode( /* O Returns error code */ + void* decState, /* I/O State */ + silk_DecControlStruct* decControl, /* I/O Control Structure */ + opus_int lostFlag, /* I 0: no loss, 1 loss, 2 decode fec */ + opus_int newPacketFlag, /* I Indicates first decoder call for this packet */ + ec_dec *psRangeDec, /* I/O Compressor data structure */ + opus_int16 *samplesOut, /* O Decoded output speech vector */ + opus_int32 *nSamplesOut, /* O Number of samples decoded */ + int arch /* I Run-time architecture */ +) +{ + opus_int i, n, decode_only_middle = 0, ret = SILK_NO_ERROR; + opus_int32 nSamplesOutDec, LBRR_symbol; + opus_int16 *samplesOut1_tmp[ 2 ]; + VARDECL( opus_int16, samplesOut1_tmp_storage1 ); + VARDECL( opus_int16, samplesOut1_tmp_storage2 ); + VARDECL( opus_int16, samplesOut2_tmp ); + opus_int32 MS_pred_Q13[ 2 ] = { 0 }; + opus_int16 *resample_out_ptr; + silk_decoder *psDec = ( silk_decoder * )decState; + silk_decoder_state *channel_state = psDec->channel_state; + opus_int has_side; + opus_int stereo_to_mono; + int delay_stack_alloc; + SAVE_STACK; + + celt_assert( decControl->nChannelsInternal == 1 || decControl->nChannelsInternal == 2 ); + + /**********************************/ + /* Test if first frame in payload */ + /**********************************/ + if( newPacketFlag ) { + for( n = 0; n < decControl->nChannelsInternal; n++ ) { + channel_state[ n ].nFramesDecoded = 0; /* Used to count frames in packet */ + } + } + + /* If Mono -> Stereo transition in bitstream: init state of second channel */ + if( decControl->nChannelsInternal > psDec->nChannelsInternal ) { + ret += silk_init_decoder( &channel_state[ 1 ] ); + } + + stereo_to_mono = decControl->nChannelsInternal == 1 && psDec->nChannelsInternal == 2 && + ( decControl->internalSampleRate == 1000*channel_state[ 0 ].fs_kHz ); + + if( channel_state[ 0 ].nFramesDecoded == 0 ) { + for( n = 0; n < decControl->nChannelsInternal; n++ ) { + opus_int fs_kHz_dec; + if( decControl->payloadSize_ms == 0 ) { + /* Assuming packet loss, use 10 ms */ + channel_state[ n ].nFramesPerPacket = 1; + channel_state[ n ].nb_subfr = 2; + } else if( decControl->payloadSize_ms == 10 ) { + channel_state[ n ].nFramesPerPacket = 1; + channel_state[ n ].nb_subfr = 2; + } else if( decControl->payloadSize_ms == 20 ) { + channel_state[ n ].nFramesPerPacket = 1; + channel_state[ n ].nb_subfr = 4; + } else if( decControl->payloadSize_ms == 40 ) { + channel_state[ n ].nFramesPerPacket = 2; + channel_state[ n ].nb_subfr = 4; + } else if( decControl->payloadSize_ms == 60 ) { + channel_state[ n ].nFramesPerPacket = 3; + channel_state[ n ].nb_subfr = 4; + } else { + celt_assert( 0 ); + RESTORE_STACK; + return SILK_DEC_INVALID_FRAME_SIZE; + } + fs_kHz_dec = ( decControl->internalSampleRate >> 10 ) + 1; + if( fs_kHz_dec != 8 && fs_kHz_dec != 12 && fs_kHz_dec != 16 ) { + celt_assert( 0 ); + RESTORE_STACK; + return SILK_DEC_INVALID_SAMPLING_FREQUENCY; + } + ret += silk_decoder_set_fs( &channel_state[ n ], fs_kHz_dec, decControl->API_sampleRate ); + } + } + + if( decControl->nChannelsAPI == 2 && decControl->nChannelsInternal == 2 && ( psDec->nChannelsAPI == 1 || psDec->nChannelsInternal == 1 ) ) { + silk_memset( psDec->sStereo.pred_prev_Q13, 0, sizeof( psDec->sStereo.pred_prev_Q13 ) ); + silk_memset( psDec->sStereo.sSide, 0, sizeof( psDec->sStereo.sSide ) ); + silk_memcpy( &channel_state[ 1 ].resampler_state, &channel_state[ 0 ].resampler_state, sizeof( silk_resampler_state_struct ) ); + } + psDec->nChannelsAPI = decControl->nChannelsAPI; + psDec->nChannelsInternal = decControl->nChannelsInternal; + + if( decControl->API_sampleRate > (opus_int32)MAX_API_FS_KHZ * 1000 || decControl->API_sampleRate < 8000 ) { + ret = SILK_DEC_INVALID_SAMPLING_FREQUENCY; + RESTORE_STACK; + return( ret ); + } + + if( lostFlag != FLAG_PACKET_LOST && channel_state[ 0 ].nFramesDecoded == 0 ) { + /* First decoder call for this payload */ + /* Decode VAD flags and LBRR flag */ + for( n = 0; n < decControl->nChannelsInternal; n++ ) { + for( i = 0; i < channel_state[ n ].nFramesPerPacket; i++ ) { + channel_state[ n ].VAD_flags[ i ] = ec_dec_bit_logp(psRangeDec, 1); + } + channel_state[ n ].LBRR_flag = ec_dec_bit_logp(psRangeDec, 1); + } + /* Decode LBRR flags */ + for( n = 0; n < decControl->nChannelsInternal; n++ ) { + silk_memset( channel_state[ n ].LBRR_flags, 0, sizeof( channel_state[ n ].LBRR_flags ) ); + if( channel_state[ n ].LBRR_flag ) { + if( channel_state[ n ].nFramesPerPacket == 1 ) { + channel_state[ n ].LBRR_flags[ 0 ] = 1; + } else { + LBRR_symbol = ec_dec_icdf( psRangeDec, silk_LBRR_flags_iCDF_ptr[ channel_state[ n ].nFramesPerPacket - 2 ], 8 ) + 1; + for( i = 0; i < channel_state[ n ].nFramesPerPacket; i++ ) { + channel_state[ n ].LBRR_flags[ i ] = silk_RSHIFT( LBRR_symbol, i ) & 1; + } + } + } + } + + if( lostFlag == FLAG_DECODE_NORMAL ) { + /* Regular decoding: skip all LBRR data */ + for( i = 0; i < channel_state[ 0 ].nFramesPerPacket; i++ ) { + for( n = 0; n < decControl->nChannelsInternal; n++ ) { + if( channel_state[ n ].LBRR_flags[ i ] ) { + opus_int16 pulses[ MAX_FRAME_LENGTH ]; + opus_int condCoding; + + if( decControl->nChannelsInternal == 2 && n == 0 ) { + silk_stereo_decode_pred( psRangeDec, MS_pred_Q13 ); + if( channel_state[ 1 ].LBRR_flags[ i ] == 0 ) { + silk_stereo_decode_mid_only( psRangeDec, &decode_only_middle ); + } + } + /* Use conditional coding if previous frame available */ + if( i > 0 && channel_state[ n ].LBRR_flags[ i - 1 ] ) { + condCoding = CODE_CONDITIONALLY; + } else { + condCoding = CODE_INDEPENDENTLY; + } + silk_decode_indices( &channel_state[ n ], psRangeDec, i, 1, condCoding ); + silk_decode_pulses( psRangeDec, pulses, channel_state[ n ].indices.signalType, + channel_state[ n ].indices.quantOffsetType, channel_state[ n ].frame_length ); + } + } + } + } + } + + /* Get MS predictor index */ + if( decControl->nChannelsInternal == 2 ) { + if( lostFlag == FLAG_DECODE_NORMAL || + ( lostFlag == FLAG_DECODE_LBRR && channel_state[ 0 ].LBRR_flags[ channel_state[ 0 ].nFramesDecoded ] == 1 ) ) + { + silk_stereo_decode_pred( psRangeDec, MS_pred_Q13 ); + /* For LBRR data, decode mid-only flag only if side-channel's LBRR flag is false */ + if( ( lostFlag == FLAG_DECODE_NORMAL && channel_state[ 1 ].VAD_flags[ channel_state[ 0 ].nFramesDecoded ] == 0 ) || + ( lostFlag == FLAG_DECODE_LBRR && channel_state[ 1 ].LBRR_flags[ channel_state[ 0 ].nFramesDecoded ] == 0 ) ) + { + silk_stereo_decode_mid_only( psRangeDec, &decode_only_middle ); + } else { + decode_only_middle = 0; + } + } else { + for( n = 0; n < 2; n++ ) { + MS_pred_Q13[ n ] = psDec->sStereo.pred_prev_Q13[ n ]; + } + } + } + + /* Reset side channel decoder prediction memory for first frame with side coding */ + if( decControl->nChannelsInternal == 2 && decode_only_middle == 0 && psDec->prev_decode_only_middle == 1 ) { + silk_memset( psDec->channel_state[ 1 ].outBuf, 0, sizeof(psDec->channel_state[ 1 ].outBuf) ); + silk_memset( psDec->channel_state[ 1 ].sLPC_Q14_buf, 0, sizeof(psDec->channel_state[ 1 ].sLPC_Q14_buf) ); + psDec->channel_state[ 1 ].lagPrev = 100; + psDec->channel_state[ 1 ].LastGainIndex = 10; + psDec->channel_state[ 1 ].prevSignalType = TYPE_NO_VOICE_ACTIVITY; + psDec->channel_state[ 1 ].first_frame_after_reset = 1; + } + + /* Check if the temp buffer fits into the output PCM buffer. If it fits, + we can delay allocating the temp buffer until after the SILK peak stack + usage. We need to use a < and not a <= because of the two extra samples. */ + delay_stack_alloc = decControl->internalSampleRate*decControl->nChannelsInternal + < decControl->API_sampleRate*decControl->nChannelsAPI; + ALLOC( samplesOut1_tmp_storage1, delay_stack_alloc ? ALLOC_NONE + : decControl->nChannelsInternal*(channel_state[ 0 ].frame_length + 2 ), + opus_int16 ); + if ( delay_stack_alloc ) + { + samplesOut1_tmp[ 0 ] = samplesOut; + samplesOut1_tmp[ 1 ] = samplesOut + channel_state[ 0 ].frame_length + 2; + } else { + samplesOut1_tmp[ 0 ] = samplesOut1_tmp_storage1; + samplesOut1_tmp[ 1 ] = samplesOut1_tmp_storage1 + channel_state[ 0 ].frame_length + 2; + } + + if( lostFlag == FLAG_DECODE_NORMAL ) { + has_side = !decode_only_middle; + } else { + has_side = !psDec->prev_decode_only_middle + || (decControl->nChannelsInternal == 2 && lostFlag == FLAG_DECODE_LBRR && channel_state[1].LBRR_flags[ channel_state[1].nFramesDecoded ] == 1 ); + } + /* Call decoder for one frame */ + for( n = 0; n < decControl->nChannelsInternal; n++ ) { + if( n == 0 || has_side ) { + opus_int FrameIndex; + opus_int condCoding; + + FrameIndex = channel_state[ 0 ].nFramesDecoded - n; + /* Use independent coding if no previous frame available */ + if( FrameIndex <= 0 ) { + condCoding = CODE_INDEPENDENTLY; + } else if( lostFlag == FLAG_DECODE_LBRR ) { + condCoding = channel_state[ n ].LBRR_flags[ FrameIndex - 1 ] ? CODE_CONDITIONALLY : CODE_INDEPENDENTLY; + } else if( n > 0 && psDec->prev_decode_only_middle ) { + /* If we skipped a side frame in this packet, we don't + need LTP scaling; the LTP state is well-defined. */ + condCoding = CODE_INDEPENDENTLY_NO_LTP_SCALING; + } else { + condCoding = CODE_CONDITIONALLY; + } + ret += silk_decode_frame( &channel_state[ n ], psRangeDec, &samplesOut1_tmp[ n ][ 2 ], &nSamplesOutDec, lostFlag, condCoding, arch); + } else { + silk_memset( &samplesOut1_tmp[ n ][ 2 ], 0, nSamplesOutDec * sizeof( opus_int16 ) ); + } + channel_state[ n ].nFramesDecoded++; + } + + if( decControl->nChannelsAPI == 2 && decControl->nChannelsInternal == 2 ) { + /* Convert Mid/Side to Left/Right */ + silk_stereo_MS_to_LR( &psDec->sStereo, samplesOut1_tmp[ 0 ], samplesOut1_tmp[ 1 ], MS_pred_Q13, channel_state[ 0 ].fs_kHz, nSamplesOutDec ); + } else { + /* Buffering */ + silk_memcpy( samplesOut1_tmp[ 0 ], psDec->sStereo.sMid, 2 * sizeof( opus_int16 ) ); + silk_memcpy( psDec->sStereo.sMid, &samplesOut1_tmp[ 0 ][ nSamplesOutDec ], 2 * sizeof( opus_int16 ) ); + } + + /* Number of output samples */ + *nSamplesOut = silk_DIV32( nSamplesOutDec * decControl->API_sampleRate, silk_SMULBB( channel_state[ 0 ].fs_kHz, 1000 ) ); + + /* Set up pointers to temp buffers */ + ALLOC( samplesOut2_tmp, + decControl->nChannelsAPI == 2 ? *nSamplesOut : ALLOC_NONE, opus_int16 ); + if( decControl->nChannelsAPI == 2 ) { + resample_out_ptr = samplesOut2_tmp; + } else { + resample_out_ptr = samplesOut; + } + + ALLOC( samplesOut1_tmp_storage2, delay_stack_alloc + ? decControl->nChannelsInternal*(channel_state[ 0 ].frame_length + 2 ) + : ALLOC_NONE, + opus_int16 ); + if ( delay_stack_alloc ) { + OPUS_COPY(samplesOut1_tmp_storage2, samplesOut, decControl->nChannelsInternal*(channel_state[ 0 ].frame_length + 2)); + samplesOut1_tmp[ 0 ] = samplesOut1_tmp_storage2; + samplesOut1_tmp[ 1 ] = samplesOut1_tmp_storage2 + channel_state[ 0 ].frame_length + 2; + } + for( n = 0; n < silk_min( decControl->nChannelsAPI, decControl->nChannelsInternal ); n++ ) { + + /* Resample decoded signal to API_sampleRate */ + ret += silk_resampler( &channel_state[ n ].resampler_state, resample_out_ptr, &samplesOut1_tmp[ n ][ 1 ], nSamplesOutDec ); + + /* Interleave if stereo output and stereo stream */ + if( decControl->nChannelsAPI == 2 ) { + for( i = 0; i < *nSamplesOut; i++ ) { + samplesOut[ n + 2 * i ] = resample_out_ptr[ i ]; + } + } + } + + /* Create two channel output from mono stream */ + if( decControl->nChannelsAPI == 2 && decControl->nChannelsInternal == 1 ) { + if ( stereo_to_mono ){ + /* Resample right channel for newly collapsed stereo just in case + we weren't doing collapsing when switching to mono */ + ret += silk_resampler( &channel_state[ 1 ].resampler_state, resample_out_ptr, &samplesOut1_tmp[ 0 ][ 1 ], nSamplesOutDec ); + + for( i = 0; i < *nSamplesOut; i++ ) { + samplesOut[ 1 + 2 * i ] = resample_out_ptr[ i ]; + } + } else { + for( i = 0; i < *nSamplesOut; i++ ) { + samplesOut[ 1 + 2 * i ] = samplesOut[ 0 + 2 * i ]; + } + } + } + + /* Export pitch lag, measured at 48 kHz sampling rate */ + if( channel_state[ 0 ].prevSignalType == TYPE_VOICED ) { + int mult_tab[ 3 ] = { 6, 4, 3 }; + decControl->prevPitchLag = channel_state[ 0 ].lagPrev * mult_tab[ ( channel_state[ 0 ].fs_kHz - 8 ) >> 2 ]; + } else { + decControl->prevPitchLag = 0; + } + + if( lostFlag == FLAG_PACKET_LOST ) { + /* On packet loss, remove the gain clamping to prevent having the energy "bounce back" + if we lose packets when the energy is going down */ + for ( i = 0; i < psDec->nChannelsInternal; i++ ) + psDec->channel_state[ i ].LastGainIndex = 10; + } else { + psDec->prev_decode_only_middle = decode_only_middle; + } + RESTORE_STACK; + return ret; +} + +#if 0 +/* Getting table of contents for a packet */ +opus_int silk_get_TOC( + const opus_uint8 *payload, /* I Payload data */ + const opus_int nBytesIn, /* I Number of input bytes */ + const opus_int nFramesPerPayload, /* I Number of SILK frames per payload */ + silk_TOC_struct *Silk_TOC /* O Type of content */ +) +{ + opus_int i, flags, ret = SILK_NO_ERROR; + + if( nBytesIn < 1 ) { + return -1; + } + if( nFramesPerPayload < 0 || nFramesPerPayload > 3 ) { + return -1; + } + + silk_memset( Silk_TOC, 0, sizeof( *Silk_TOC ) ); + + /* For stereo, extract the flags for the mid channel */ + flags = silk_RSHIFT( payload[ 0 ], 7 - nFramesPerPayload ) & ( silk_LSHIFT( 1, nFramesPerPayload + 1 ) - 1 ); + + Silk_TOC->inbandFECFlag = flags & 1; + for( i = nFramesPerPayload - 1; i >= 0 ; i-- ) { + flags = silk_RSHIFT( flags, 1 ); + Silk_TOC->VADFlags[ i ] = flags & 1; + Silk_TOC->VADFlag |= flags & 1; + } + + return ret; +} +#endif diff --git a/media/libopus/silk/decode_core.c b/media/libopus/silk/decode_core.c new file mode 100644 index 0000000000..1c352a6522 --- /dev/null +++ b/media/libopus/silk/decode_core.c @@ -0,0 +1,237 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" +#include "stack_alloc.h" + +/**********************************************************/ +/* Core decoder. Performs inverse NSQ operation LTP + LPC */ +/**********************************************************/ +void silk_decode_core( + silk_decoder_state *psDec, /* I/O Decoder state */ + silk_decoder_control *psDecCtrl, /* I Decoder control */ + opus_int16 xq[], /* O Decoded speech */ + const opus_int16 pulses[ MAX_FRAME_LENGTH ], /* I Pulse signal */ + int arch /* I Run-time architecture */ +) +{ + opus_int i, k, lag = 0, start_idx, sLTP_buf_idx, NLSF_interpolation_flag, signalType; + opus_int16 *A_Q12, *B_Q14, *pxq, A_Q12_tmp[ MAX_LPC_ORDER ]; + VARDECL( opus_int16, sLTP ); + VARDECL( opus_int32, sLTP_Q15 ); + opus_int32 LTP_pred_Q13, LPC_pred_Q10, Gain_Q10, inv_gain_Q31, gain_adj_Q16, rand_seed, offset_Q10; + opus_int32 *pred_lag_ptr, *pexc_Q14, *pres_Q14; + VARDECL( opus_int32, res_Q14 ); + VARDECL( opus_int32, sLPC_Q14 ); + SAVE_STACK; + + silk_assert( psDec->prev_gain_Q16 != 0 ); + + ALLOC( sLTP, psDec->ltp_mem_length, opus_int16 ); + ALLOC( sLTP_Q15, psDec->ltp_mem_length + psDec->frame_length, opus_int32 ); + ALLOC( res_Q14, psDec->subfr_length, opus_int32 ); + ALLOC( sLPC_Q14, psDec->subfr_length + MAX_LPC_ORDER, opus_int32 ); + + offset_Q10 = silk_Quantization_Offsets_Q10[ psDec->indices.signalType >> 1 ][ psDec->indices.quantOffsetType ]; + + if( psDec->indices.NLSFInterpCoef_Q2 < 1 << 2 ) { + NLSF_interpolation_flag = 1; + } else { + NLSF_interpolation_flag = 0; + } + + /* Decode excitation */ + rand_seed = psDec->indices.Seed; + for( i = 0; i < psDec->frame_length; i++ ) { + rand_seed = silk_RAND( rand_seed ); + psDec->exc_Q14[ i ] = silk_LSHIFT( (opus_int32)pulses[ i ], 14 ); + if( psDec->exc_Q14[ i ] > 0 ) { + psDec->exc_Q14[ i ] -= QUANT_LEVEL_ADJUST_Q10 << 4; + } else + if( psDec->exc_Q14[ i ] < 0 ) { + psDec->exc_Q14[ i ] += QUANT_LEVEL_ADJUST_Q10 << 4; + } + psDec->exc_Q14[ i ] += offset_Q10 << 4; + if( rand_seed < 0 ) { + psDec->exc_Q14[ i ] = -psDec->exc_Q14[ i ]; + } + + rand_seed = silk_ADD32_ovflw( rand_seed, pulses[ i ] ); + } + + /* Copy LPC state */ + silk_memcpy( sLPC_Q14, psDec->sLPC_Q14_buf, MAX_LPC_ORDER * sizeof( opus_int32 ) ); + + pexc_Q14 = psDec->exc_Q14; + pxq = xq; + sLTP_buf_idx = psDec->ltp_mem_length; + /* Loop over subframes */ + for( k = 0; k < psDec->nb_subfr; k++ ) { + pres_Q14 = res_Q14; + A_Q12 = psDecCtrl->PredCoef_Q12[ k >> 1 ]; + + /* Preload LPC coeficients to array on stack. Gives small performance gain */ + silk_memcpy( A_Q12_tmp, A_Q12, psDec->LPC_order * sizeof( opus_int16 ) ); + B_Q14 = &psDecCtrl->LTPCoef_Q14[ k * LTP_ORDER ]; + signalType = psDec->indices.signalType; + + Gain_Q10 = silk_RSHIFT( psDecCtrl->Gains_Q16[ k ], 6 ); + inv_gain_Q31 = silk_INVERSE32_varQ( psDecCtrl->Gains_Q16[ k ], 47 ); + + /* Calculate gain adjustment factor */ + if( psDecCtrl->Gains_Q16[ k ] != psDec->prev_gain_Q16 ) { + gain_adj_Q16 = silk_DIV32_varQ( psDec->prev_gain_Q16, psDecCtrl->Gains_Q16[ k ], 16 ); + + /* Scale short term state */ + for( i = 0; i < MAX_LPC_ORDER; i++ ) { + sLPC_Q14[ i ] = silk_SMULWW( gain_adj_Q16, sLPC_Q14[ i ] ); + } + } else { + gain_adj_Q16 = (opus_int32)1 << 16; + } + + /* Save inv_gain */ + silk_assert( inv_gain_Q31 != 0 ); + psDec->prev_gain_Q16 = psDecCtrl->Gains_Q16[ k ]; + + /* Avoid abrupt transition from voiced PLC to unvoiced normal decoding */ + if( psDec->lossCnt && psDec->prevSignalType == TYPE_VOICED && + psDec->indices.signalType != TYPE_VOICED && k < MAX_NB_SUBFR/2 ) { + + silk_memset( B_Q14, 0, LTP_ORDER * sizeof( opus_int16 ) ); + B_Q14[ LTP_ORDER/2 ] = SILK_FIX_CONST( 0.25, 14 ); + + signalType = TYPE_VOICED; + psDecCtrl->pitchL[ k ] = psDec->lagPrev; + } + + if( signalType == TYPE_VOICED ) { + /* Voiced */ + lag = psDecCtrl->pitchL[ k ]; + + /* Re-whitening */ + if( k == 0 || ( k == 2 && NLSF_interpolation_flag ) ) { + /* Rewhiten with new A coefs */ + start_idx = psDec->ltp_mem_length - lag - psDec->LPC_order - LTP_ORDER / 2; + celt_assert( start_idx > 0 ); + + if( k == 2 ) { + silk_memcpy( &psDec->outBuf[ psDec->ltp_mem_length ], xq, 2 * psDec->subfr_length * sizeof( opus_int16 ) ); + } + + silk_LPC_analysis_filter( &sLTP[ start_idx ], &psDec->outBuf[ start_idx + k * psDec->subfr_length ], + A_Q12, psDec->ltp_mem_length - start_idx, psDec->LPC_order, arch ); + + /* After rewhitening the LTP state is unscaled */ + if( k == 0 ) { + /* Do LTP downscaling to reduce inter-packet dependency */ + inv_gain_Q31 = silk_LSHIFT( silk_SMULWB( inv_gain_Q31, psDecCtrl->LTP_scale_Q14 ), 2 ); + } + for( i = 0; i < lag + LTP_ORDER/2; i++ ) { + sLTP_Q15[ sLTP_buf_idx - i - 1 ] = silk_SMULWB( inv_gain_Q31, sLTP[ psDec->ltp_mem_length - i - 1 ] ); + } + } else { + /* Update LTP state when Gain changes */ + if( gain_adj_Q16 != (opus_int32)1 << 16 ) { + for( i = 0; i < lag + LTP_ORDER/2; i++ ) { + sLTP_Q15[ sLTP_buf_idx - i - 1 ] = silk_SMULWW( gain_adj_Q16, sLTP_Q15[ sLTP_buf_idx - i - 1 ] ); + } + } + } + } + + /* Long-term prediction */ + if( signalType == TYPE_VOICED ) { + /* Set up pointer */ + pred_lag_ptr = &sLTP_Q15[ sLTP_buf_idx - lag + LTP_ORDER / 2 ]; + for( i = 0; i < psDec->subfr_length; i++ ) { + /* Unrolled loop */ + /* Avoids introducing a bias because silk_SMLAWB() always rounds to -inf */ + LTP_pred_Q13 = 2; + LTP_pred_Q13 = silk_SMLAWB( LTP_pred_Q13, pred_lag_ptr[ 0 ], B_Q14[ 0 ] ); + LTP_pred_Q13 = silk_SMLAWB( LTP_pred_Q13, pred_lag_ptr[ -1 ], B_Q14[ 1 ] ); + LTP_pred_Q13 = silk_SMLAWB( LTP_pred_Q13, pred_lag_ptr[ -2 ], B_Q14[ 2 ] ); + LTP_pred_Q13 = silk_SMLAWB( LTP_pred_Q13, pred_lag_ptr[ -3 ], B_Q14[ 3 ] ); + LTP_pred_Q13 = silk_SMLAWB( LTP_pred_Q13, pred_lag_ptr[ -4 ], B_Q14[ 4 ] ); + pred_lag_ptr++; + + /* Generate LPC excitation */ + pres_Q14[ i ] = silk_ADD_LSHIFT32( pexc_Q14[ i ], LTP_pred_Q13, 1 ); + + /* Update states */ + sLTP_Q15[ sLTP_buf_idx ] = silk_LSHIFT( pres_Q14[ i ], 1 ); + sLTP_buf_idx++; + } + } else { + pres_Q14 = pexc_Q14; + } + + for( i = 0; i < psDec->subfr_length; i++ ) { + /* Short-term prediction */ + celt_assert( psDec->LPC_order == 10 || psDec->LPC_order == 16 ); + /* Avoids introducing a bias because silk_SMLAWB() always rounds to -inf */ + LPC_pred_Q10 = silk_RSHIFT( psDec->LPC_order, 1 ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, sLPC_Q14[ MAX_LPC_ORDER + i - 1 ], A_Q12_tmp[ 0 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, sLPC_Q14[ MAX_LPC_ORDER + i - 2 ], A_Q12_tmp[ 1 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, sLPC_Q14[ MAX_LPC_ORDER + i - 3 ], A_Q12_tmp[ 2 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, sLPC_Q14[ MAX_LPC_ORDER + i - 4 ], A_Q12_tmp[ 3 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, sLPC_Q14[ MAX_LPC_ORDER + i - 5 ], A_Q12_tmp[ 4 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, sLPC_Q14[ MAX_LPC_ORDER + i - 6 ], A_Q12_tmp[ 5 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, sLPC_Q14[ MAX_LPC_ORDER + i - 7 ], A_Q12_tmp[ 6 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, sLPC_Q14[ MAX_LPC_ORDER + i - 8 ], A_Q12_tmp[ 7 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, sLPC_Q14[ MAX_LPC_ORDER + i - 9 ], A_Q12_tmp[ 8 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, sLPC_Q14[ MAX_LPC_ORDER + i - 10 ], A_Q12_tmp[ 9 ] ); + if( psDec->LPC_order == 16 ) { + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, sLPC_Q14[ MAX_LPC_ORDER + i - 11 ], A_Q12_tmp[ 10 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, sLPC_Q14[ MAX_LPC_ORDER + i - 12 ], A_Q12_tmp[ 11 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, sLPC_Q14[ MAX_LPC_ORDER + i - 13 ], A_Q12_tmp[ 12 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, sLPC_Q14[ MAX_LPC_ORDER + i - 14 ], A_Q12_tmp[ 13 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, sLPC_Q14[ MAX_LPC_ORDER + i - 15 ], A_Q12_tmp[ 14 ] ); + LPC_pred_Q10 = silk_SMLAWB( LPC_pred_Q10, sLPC_Q14[ MAX_LPC_ORDER + i - 16 ], A_Q12_tmp[ 15 ] ); + } + + /* Add prediction to LPC excitation */ + sLPC_Q14[ MAX_LPC_ORDER + i ] = silk_ADD_SAT32( pres_Q14[ i ], silk_LSHIFT_SAT32( LPC_pred_Q10, 4 ) ); + + /* Scale with gain */ + pxq[ i ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( silk_SMULWW( sLPC_Q14[ MAX_LPC_ORDER + i ], Gain_Q10 ), 8 ) ); + } + + /* Update LPC filter state */ + silk_memcpy( sLPC_Q14, &sLPC_Q14[ psDec->subfr_length ], MAX_LPC_ORDER * sizeof( opus_int32 ) ); + pexc_Q14 += psDec->subfr_length; + pxq += psDec->subfr_length; + } + + /* Save LPC state */ + silk_memcpy( psDec->sLPC_Q14_buf, sLPC_Q14, MAX_LPC_ORDER * sizeof( opus_int32 ) ); + RESTORE_STACK; +} diff --git a/media/libopus/silk/decode_frame.c b/media/libopus/silk/decode_frame.c new file mode 100644 index 0000000000..4f36f854c2 --- /dev/null +++ b/media/libopus/silk/decode_frame.c @@ -0,0 +1,129 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" +#include "stack_alloc.h" +#include "PLC.h" + +/****************/ +/* Decode frame */ +/****************/ +opus_int silk_decode_frame( + silk_decoder_state *psDec, /* I/O Pointer to Silk decoder state */ + ec_dec *psRangeDec, /* I/O Compressor data structure */ + opus_int16 pOut[], /* O Pointer to output speech frame */ + opus_int32 *pN, /* O Pointer to size of output frame */ + opus_int lostFlag, /* I 0: no loss, 1 loss, 2 decode fec */ + opus_int condCoding, /* I The type of conditional coding to use */ + int arch /* I Run-time architecture */ +) +{ + VARDECL( silk_decoder_control, psDecCtrl ); + opus_int L, mv_len, ret = 0; + SAVE_STACK; + + L = psDec->frame_length; + ALLOC( psDecCtrl, 1, silk_decoder_control ); + psDecCtrl->LTP_scale_Q14 = 0; + + /* Safety checks */ + celt_assert( L > 0 && L <= MAX_FRAME_LENGTH ); + + if( lostFlag == FLAG_DECODE_NORMAL || + ( lostFlag == FLAG_DECODE_LBRR && psDec->LBRR_flags[ psDec->nFramesDecoded ] == 1 ) ) + { + VARDECL( opus_int16, pulses ); + ALLOC( pulses, (L + SHELL_CODEC_FRAME_LENGTH - 1) & + ~(SHELL_CODEC_FRAME_LENGTH - 1), opus_int16 ); + /*********************************************/ + /* Decode quantization indices of side info */ + /*********************************************/ + silk_decode_indices( psDec, psRangeDec, psDec->nFramesDecoded, lostFlag, condCoding ); + + /*********************************************/ + /* Decode quantization indices of excitation */ + /*********************************************/ + silk_decode_pulses( psRangeDec, pulses, psDec->indices.signalType, + psDec->indices.quantOffsetType, psDec->frame_length ); + + /********************************************/ + /* Decode parameters and pulse signal */ + /********************************************/ + silk_decode_parameters( psDec, psDecCtrl, condCoding ); + + /********************************************************/ + /* Run inverse NSQ */ + /********************************************************/ + silk_decode_core( psDec, psDecCtrl, pOut, pulses, arch ); + + /********************************************************/ + /* Update PLC state */ + /********************************************************/ + silk_PLC( psDec, psDecCtrl, pOut, 0, arch ); + + psDec->lossCnt = 0; + psDec->prevSignalType = psDec->indices.signalType; + celt_assert( psDec->prevSignalType >= 0 && psDec->prevSignalType <= 2 ); + + /* A frame has been decoded without errors */ + psDec->first_frame_after_reset = 0; + } else { + /* Handle packet loss by extrapolation */ + silk_PLC( psDec, psDecCtrl, pOut, 1, arch ); + } + + /*************************/ + /* Update output buffer. */ + /*************************/ + celt_assert( psDec->ltp_mem_length >= psDec->frame_length ); + mv_len = psDec->ltp_mem_length - psDec->frame_length; + silk_memmove( psDec->outBuf, &psDec->outBuf[ psDec->frame_length ], mv_len * sizeof(opus_int16) ); + silk_memcpy( &psDec->outBuf[ mv_len ], pOut, psDec->frame_length * sizeof( opus_int16 ) ); + + /************************************************/ + /* Comfort noise generation / estimation */ + /************************************************/ + silk_CNG( psDec, psDecCtrl, pOut, L ); + + /****************************************************************/ + /* Ensure smooth connection of extrapolated and good frames */ + /****************************************************************/ + silk_PLC_glue_frames( psDec, pOut, L ); + + /* Update some decoder state variables */ + psDec->lagPrev = psDecCtrl->pitchL[ psDec->nb_subfr - 1 ]; + + /* Set output frame length */ + *pN = L; + + RESTORE_STACK; + return ret; +} diff --git a/media/libopus/silk/decode_indices.c b/media/libopus/silk/decode_indices.c new file mode 100644 index 0000000000..0bb4a997a5 --- /dev/null +++ b/media/libopus/silk/decode_indices.c @@ -0,0 +1,151 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" + +/* Decode side-information parameters from payload */ +void silk_decode_indices( + silk_decoder_state *psDec, /* I/O State */ + ec_dec *psRangeDec, /* I/O Compressor data structure */ + opus_int FrameIndex, /* I Frame number */ + opus_int decode_LBRR, /* I Flag indicating LBRR data is being decoded */ + opus_int condCoding /* I The type of conditional coding to use */ +) +{ + opus_int i, k, Ix; + opus_int decode_absolute_lagIndex, delta_lagIndex; + opus_int16 ec_ix[ MAX_LPC_ORDER ]; + opus_uint8 pred_Q8[ MAX_LPC_ORDER ]; + + /*******************************************/ + /* Decode signal type and quantizer offset */ + /*******************************************/ + if( decode_LBRR || psDec->VAD_flags[ FrameIndex ] ) { + Ix = ec_dec_icdf( psRangeDec, silk_type_offset_VAD_iCDF, 8 ) + 2; + } else { + Ix = ec_dec_icdf( psRangeDec, silk_type_offset_no_VAD_iCDF, 8 ); + } + psDec->indices.signalType = (opus_int8)silk_RSHIFT( Ix, 1 ); + psDec->indices.quantOffsetType = (opus_int8)( Ix & 1 ); + + /****************/ + /* Decode gains */ + /****************/ + /* First subframe */ + if( condCoding == CODE_CONDITIONALLY ) { + /* Conditional coding */ + psDec->indices.GainsIndices[ 0 ] = (opus_int8)ec_dec_icdf( psRangeDec, silk_delta_gain_iCDF, 8 ); + } else { + /* Independent coding, in two stages: MSB bits followed by 3 LSBs */ + psDec->indices.GainsIndices[ 0 ] = (opus_int8)silk_LSHIFT( ec_dec_icdf( psRangeDec, silk_gain_iCDF[ psDec->indices.signalType ], 8 ), 3 ); + psDec->indices.GainsIndices[ 0 ] += (opus_int8)ec_dec_icdf( psRangeDec, silk_uniform8_iCDF, 8 ); + } + + /* Remaining subframes */ + for( i = 1; i < psDec->nb_subfr; i++ ) { + psDec->indices.GainsIndices[ i ] = (opus_int8)ec_dec_icdf( psRangeDec, silk_delta_gain_iCDF, 8 ); + } + + /**********************/ + /* Decode LSF Indices */ + /**********************/ + psDec->indices.NLSFIndices[ 0 ] = (opus_int8)ec_dec_icdf( psRangeDec, &psDec->psNLSF_CB->CB1_iCDF[ ( psDec->indices.signalType >> 1 ) * psDec->psNLSF_CB->nVectors ], 8 ); + silk_NLSF_unpack( ec_ix, pred_Q8, psDec->psNLSF_CB, psDec->indices.NLSFIndices[ 0 ] ); + celt_assert( psDec->psNLSF_CB->order == psDec->LPC_order ); + for( i = 0; i < psDec->psNLSF_CB->order; i++ ) { + Ix = ec_dec_icdf( psRangeDec, &psDec->psNLSF_CB->ec_iCDF[ ec_ix[ i ] ], 8 ); + if( Ix == 0 ) { + Ix -= ec_dec_icdf( psRangeDec, silk_NLSF_EXT_iCDF, 8 ); + } else if( Ix == 2 * NLSF_QUANT_MAX_AMPLITUDE ) { + Ix += ec_dec_icdf( psRangeDec, silk_NLSF_EXT_iCDF, 8 ); + } + psDec->indices.NLSFIndices[ i+1 ] = (opus_int8)( Ix - NLSF_QUANT_MAX_AMPLITUDE ); + } + + /* Decode LSF interpolation factor */ + if( psDec->nb_subfr == MAX_NB_SUBFR ) { + psDec->indices.NLSFInterpCoef_Q2 = (opus_int8)ec_dec_icdf( psRangeDec, silk_NLSF_interpolation_factor_iCDF, 8 ); + } else { + psDec->indices.NLSFInterpCoef_Q2 = 4; + } + + if( psDec->indices.signalType == TYPE_VOICED ) + { + /*********************/ + /* Decode pitch lags */ + /*********************/ + /* Get lag index */ + decode_absolute_lagIndex = 1; + if( condCoding == CODE_CONDITIONALLY && psDec->ec_prevSignalType == TYPE_VOICED ) { + /* Decode Delta index */ + delta_lagIndex = (opus_int16)ec_dec_icdf( psRangeDec, silk_pitch_delta_iCDF, 8 ); + if( delta_lagIndex > 0 ) { + delta_lagIndex = delta_lagIndex - 9; + psDec->indices.lagIndex = (opus_int16)( psDec->ec_prevLagIndex + delta_lagIndex ); + decode_absolute_lagIndex = 0; + } + } + if( decode_absolute_lagIndex ) { + /* Absolute decoding */ + psDec->indices.lagIndex = (opus_int16)ec_dec_icdf( psRangeDec, silk_pitch_lag_iCDF, 8 ) * silk_RSHIFT( psDec->fs_kHz, 1 ); + psDec->indices.lagIndex += (opus_int16)ec_dec_icdf( psRangeDec, psDec->pitch_lag_low_bits_iCDF, 8 ); + } + psDec->ec_prevLagIndex = psDec->indices.lagIndex; + + /* Get countour index */ + psDec->indices.contourIndex = (opus_int8)ec_dec_icdf( psRangeDec, psDec->pitch_contour_iCDF, 8 ); + + /********************/ + /* Decode LTP gains */ + /********************/ + /* Decode PERIndex value */ + psDec->indices.PERIndex = (opus_int8)ec_dec_icdf( psRangeDec, silk_LTP_per_index_iCDF, 8 ); + + for( k = 0; k < psDec->nb_subfr; k++ ) { + psDec->indices.LTPIndex[ k ] = (opus_int8)ec_dec_icdf( psRangeDec, silk_LTP_gain_iCDF_ptrs[ psDec->indices.PERIndex ], 8 ); + } + + /**********************/ + /* Decode LTP scaling */ + /**********************/ + if( condCoding == CODE_INDEPENDENTLY ) { + psDec->indices.LTP_scaleIndex = (opus_int8)ec_dec_icdf( psRangeDec, silk_LTPscale_iCDF, 8 ); + } else { + psDec->indices.LTP_scaleIndex = 0; + } + } + psDec->ec_prevSignalType = psDec->indices.signalType; + + /***************/ + /* Decode seed */ + /***************/ + psDec->indices.Seed = (opus_int8)ec_dec_icdf( psRangeDec, silk_uniform4_iCDF, 8 ); +} diff --git a/media/libopus/silk/decode_parameters.c b/media/libopus/silk/decode_parameters.c new file mode 100644 index 0000000000..a56a409858 --- /dev/null +++ b/media/libopus/silk/decode_parameters.c @@ -0,0 +1,115 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" + +/* Decode parameters from payload */ +void silk_decode_parameters( + silk_decoder_state *psDec, /* I/O State */ + silk_decoder_control *psDecCtrl, /* I/O Decoder control */ + opus_int condCoding /* I The type of conditional coding to use */ +) +{ + opus_int i, k, Ix; + opus_int16 pNLSF_Q15[ MAX_LPC_ORDER ], pNLSF0_Q15[ MAX_LPC_ORDER ]; + const opus_int8 *cbk_ptr_Q7; + + /* Dequant Gains */ + silk_gains_dequant( psDecCtrl->Gains_Q16, psDec->indices.GainsIndices, + &psDec->LastGainIndex, condCoding == CODE_CONDITIONALLY, psDec->nb_subfr ); + + /****************/ + /* Decode NLSFs */ + /****************/ + silk_NLSF_decode( pNLSF_Q15, psDec->indices.NLSFIndices, psDec->psNLSF_CB ); + + /* Convert NLSF parameters to AR prediction filter coefficients */ + silk_NLSF2A( psDecCtrl->PredCoef_Q12[ 1 ], pNLSF_Q15, psDec->LPC_order, psDec->arch ); + + /* If just reset, e.g., because internal Fs changed, do not allow interpolation */ + /* improves the case of packet loss in the first frame after a switch */ + if( psDec->first_frame_after_reset == 1 ) { + psDec->indices.NLSFInterpCoef_Q2 = 4; + } + + if( psDec->indices.NLSFInterpCoef_Q2 < 4 ) { + /* Calculation of the interpolated NLSF0 vector from the interpolation factor, */ + /* the previous NLSF1, and the current NLSF1 */ + for( i = 0; i < psDec->LPC_order; i++ ) { + pNLSF0_Q15[ i ] = psDec->prevNLSF_Q15[ i ] + silk_RSHIFT( silk_MUL( psDec->indices.NLSFInterpCoef_Q2, + pNLSF_Q15[ i ] - psDec->prevNLSF_Q15[ i ] ), 2 ); + } + + /* Convert NLSF parameters to AR prediction filter coefficients */ + silk_NLSF2A( psDecCtrl->PredCoef_Q12[ 0 ], pNLSF0_Q15, psDec->LPC_order, psDec->arch ); + } else { + /* Copy LPC coefficients for first half from second half */ + silk_memcpy( psDecCtrl->PredCoef_Q12[ 0 ], psDecCtrl->PredCoef_Q12[ 1 ], psDec->LPC_order * sizeof( opus_int16 ) ); + } + + silk_memcpy( psDec->prevNLSF_Q15, pNLSF_Q15, psDec->LPC_order * sizeof( opus_int16 ) ); + + /* After a packet loss do BWE of LPC coefs */ + if( psDec->lossCnt ) { + silk_bwexpander( psDecCtrl->PredCoef_Q12[ 0 ], psDec->LPC_order, BWE_AFTER_LOSS_Q16 ); + silk_bwexpander( psDecCtrl->PredCoef_Q12[ 1 ], psDec->LPC_order, BWE_AFTER_LOSS_Q16 ); + } + + if( psDec->indices.signalType == TYPE_VOICED ) { + /*********************/ + /* Decode pitch lags */ + /*********************/ + + /* Decode pitch values */ + silk_decode_pitch( psDec->indices.lagIndex, psDec->indices.contourIndex, psDecCtrl->pitchL, psDec->fs_kHz, psDec->nb_subfr ); + + /* Decode Codebook Index */ + cbk_ptr_Q7 = silk_LTP_vq_ptrs_Q7[ psDec->indices.PERIndex ]; /* set pointer to start of codebook */ + + for( k = 0; k < psDec->nb_subfr; k++ ) { + Ix = psDec->indices.LTPIndex[ k ]; + for( i = 0; i < LTP_ORDER; i++ ) { + psDecCtrl->LTPCoef_Q14[ k * LTP_ORDER + i ] = silk_LSHIFT( cbk_ptr_Q7[ Ix * LTP_ORDER + i ], 7 ); + } + } + + /**********************/ + /* Decode LTP scaling */ + /**********************/ + Ix = psDec->indices.LTP_scaleIndex; + psDecCtrl->LTP_scale_Q14 = silk_LTPScales_table_Q14[ Ix ]; + } else { + silk_memset( psDecCtrl->pitchL, 0, psDec->nb_subfr * sizeof( opus_int ) ); + silk_memset( psDecCtrl->LTPCoef_Q14, 0, LTP_ORDER * psDec->nb_subfr * sizeof( opus_int16 ) ); + psDec->indices.PERIndex = 0; + psDecCtrl->LTP_scale_Q14 = 0; + } +} diff --git a/media/libopus/silk/decode_pitch.c b/media/libopus/silk/decode_pitch.c new file mode 100644 index 0000000000..fd1b6bf551 --- /dev/null +++ b/media/libopus/silk/decode_pitch.c @@ -0,0 +1,77 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +/*********************************************************** +* Pitch analyser function +********************************************************** */ +#include "SigProc_FIX.h" +#include "pitch_est_defines.h" + +void silk_decode_pitch( + opus_int16 lagIndex, /* I */ + opus_int8 contourIndex, /* O */ + opus_int pitch_lags[], /* O 4 pitch values */ + const opus_int Fs_kHz, /* I sampling frequency (kHz) */ + const opus_int nb_subfr /* I number of sub frames */ +) +{ + opus_int lag, k, min_lag, max_lag, cbk_size; + const opus_int8 *Lag_CB_ptr; + + if( Fs_kHz == 8 ) { + if( nb_subfr == PE_MAX_NB_SUBFR ) { + Lag_CB_ptr = &silk_CB_lags_stage2[ 0 ][ 0 ]; + cbk_size = PE_NB_CBKS_STAGE2_EXT; + } else { + celt_assert( nb_subfr == PE_MAX_NB_SUBFR >> 1 ); + Lag_CB_ptr = &silk_CB_lags_stage2_10_ms[ 0 ][ 0 ]; + cbk_size = PE_NB_CBKS_STAGE2_10MS; + } + } else { + if( nb_subfr == PE_MAX_NB_SUBFR ) { + Lag_CB_ptr = &silk_CB_lags_stage3[ 0 ][ 0 ]; + cbk_size = PE_NB_CBKS_STAGE3_MAX; + } else { + celt_assert( nb_subfr == PE_MAX_NB_SUBFR >> 1 ); + Lag_CB_ptr = &silk_CB_lags_stage3_10_ms[ 0 ][ 0 ]; + cbk_size = PE_NB_CBKS_STAGE3_10MS; + } + } + + min_lag = silk_SMULBB( PE_MIN_LAG_MS, Fs_kHz ); + max_lag = silk_SMULBB( PE_MAX_LAG_MS, Fs_kHz ); + lag = min_lag + lagIndex; + + for( k = 0; k < nb_subfr; k++ ) { + pitch_lags[ k ] = lag + matrix_ptr( Lag_CB_ptr, k, contourIndex, cbk_size ); + pitch_lags[ k ] = silk_LIMIT( pitch_lags[ k ], min_lag, max_lag ); + } +} diff --git a/media/libopus/silk/decode_pulses.c b/media/libopus/silk/decode_pulses.c new file mode 100644 index 0000000000..a56d2d3074 --- /dev/null +++ b/media/libopus/silk/decode_pulses.c @@ -0,0 +1,115 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" + +/*********************************************/ +/* Decode quantization indices of excitation */ +/*********************************************/ +void silk_decode_pulses( + ec_dec *psRangeDec, /* I/O Compressor data structure */ + opus_int16 pulses[], /* O Excitation signal */ + const opus_int signalType, /* I Sigtype */ + const opus_int quantOffsetType, /* I quantOffsetType */ + const opus_int frame_length /* I Frame length */ +) +{ + opus_int i, j, k, iter, abs_q, nLS, RateLevelIndex; + opus_int sum_pulses[ MAX_NB_SHELL_BLOCKS ], nLshifts[ MAX_NB_SHELL_BLOCKS ]; + opus_int16 *pulses_ptr; + const opus_uint8 *cdf_ptr; + + /*********************/ + /* Decode rate level */ + /*********************/ + RateLevelIndex = ec_dec_icdf( psRangeDec, silk_rate_levels_iCDF[ signalType >> 1 ], 8 ); + + /* Calculate number of shell blocks */ + silk_assert( 1 << LOG2_SHELL_CODEC_FRAME_LENGTH == SHELL_CODEC_FRAME_LENGTH ); + iter = silk_RSHIFT( frame_length, LOG2_SHELL_CODEC_FRAME_LENGTH ); + if( iter * SHELL_CODEC_FRAME_LENGTH < frame_length ) { + celt_assert( frame_length == 12 * 10 ); /* Make sure only happens for 10 ms @ 12 kHz */ + iter++; + } + + /***************************************************/ + /* Sum-Weighted-Pulses Decoding */ + /***************************************************/ + cdf_ptr = silk_pulses_per_block_iCDF[ RateLevelIndex ]; + for( i = 0; i < iter; i++ ) { + nLshifts[ i ] = 0; + sum_pulses[ i ] = ec_dec_icdf( psRangeDec, cdf_ptr, 8 ); + + /* LSB indication */ + while( sum_pulses[ i ] == SILK_MAX_PULSES + 1 ) { + nLshifts[ i ]++; + /* When we've already got 10 LSBs, we shift the table to not allow (SILK_MAX_PULSES + 1) */ + sum_pulses[ i ] = ec_dec_icdf( psRangeDec, + silk_pulses_per_block_iCDF[ N_RATE_LEVELS - 1] + ( nLshifts[ i ] == 10 ), 8 ); + } + } + + /***************************************************/ + /* Shell decoding */ + /***************************************************/ + for( i = 0; i < iter; i++ ) { + if( sum_pulses[ i ] > 0 ) { + silk_shell_decoder( &pulses[ silk_SMULBB( i, SHELL_CODEC_FRAME_LENGTH ) ], psRangeDec, sum_pulses[ i ] ); + } else { + silk_memset( &pulses[ silk_SMULBB( i, SHELL_CODEC_FRAME_LENGTH ) ], 0, SHELL_CODEC_FRAME_LENGTH * sizeof( pulses[0] ) ); + } + } + + /***************************************************/ + /* LSB Decoding */ + /***************************************************/ + for( i = 0; i < iter; i++ ) { + if( nLshifts[ i ] > 0 ) { + nLS = nLshifts[ i ]; + pulses_ptr = &pulses[ silk_SMULBB( i, SHELL_CODEC_FRAME_LENGTH ) ]; + for( k = 0; k < SHELL_CODEC_FRAME_LENGTH; k++ ) { + abs_q = pulses_ptr[ k ]; + for( j = 0; j < nLS; j++ ) { + abs_q = silk_LSHIFT( abs_q, 1 ); + abs_q += ec_dec_icdf( psRangeDec, silk_lsb_iCDF, 8 ); + } + pulses_ptr[ k ] = abs_q; + } + /* Mark the number of pulses non-zero for sign decoding. */ + sum_pulses[ i ] |= nLS << 5; + } + } + + /****************************************/ + /* Decode and add signs to pulse signal */ + /****************************************/ + silk_decode_signs( psRangeDec, pulses, frame_length, signalType, quantOffsetType, sum_pulses ); +} diff --git a/media/libopus/silk/decoder_set_fs.c b/media/libopus/silk/decoder_set_fs.c new file mode 100644 index 0000000000..d9a13d0f0c --- /dev/null +++ b/media/libopus/silk/decoder_set_fs.c @@ -0,0 +1,108 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" + +/* Set decoder sampling rate */ +opus_int silk_decoder_set_fs( + silk_decoder_state *psDec, /* I/O Decoder state pointer */ + opus_int fs_kHz, /* I Sampling frequency (kHz) */ + opus_int32 fs_API_Hz /* I API Sampling frequency (Hz) */ +) +{ + opus_int frame_length, ret = 0; + + celt_assert( fs_kHz == 8 || fs_kHz == 12 || fs_kHz == 16 ); + celt_assert( psDec->nb_subfr == MAX_NB_SUBFR || psDec->nb_subfr == MAX_NB_SUBFR/2 ); + + /* New (sub)frame length */ + psDec->subfr_length = silk_SMULBB( SUB_FRAME_LENGTH_MS, fs_kHz ); + frame_length = silk_SMULBB( psDec->nb_subfr, psDec->subfr_length ); + + /* Initialize resampler when switching internal or external sampling frequency */ + if( psDec->fs_kHz != fs_kHz || psDec->fs_API_hz != fs_API_Hz ) { + /* Initialize the resampler for dec_API.c preparing resampling from fs_kHz to API_fs_Hz */ + ret += silk_resampler_init( &psDec->resampler_state, silk_SMULBB( fs_kHz, 1000 ), fs_API_Hz, 0 ); + + psDec->fs_API_hz = fs_API_Hz; + } + + if( psDec->fs_kHz != fs_kHz || frame_length != psDec->frame_length ) { + if( fs_kHz == 8 ) { + if( psDec->nb_subfr == MAX_NB_SUBFR ) { + psDec->pitch_contour_iCDF = silk_pitch_contour_NB_iCDF; + } else { + psDec->pitch_contour_iCDF = silk_pitch_contour_10_ms_NB_iCDF; + } + } else { + if( psDec->nb_subfr == MAX_NB_SUBFR ) { + psDec->pitch_contour_iCDF = silk_pitch_contour_iCDF; + } else { + psDec->pitch_contour_iCDF = silk_pitch_contour_10_ms_iCDF; + } + } + if( psDec->fs_kHz != fs_kHz ) { + psDec->ltp_mem_length = silk_SMULBB( LTP_MEM_LENGTH_MS, fs_kHz ); + if( fs_kHz == 8 || fs_kHz == 12 ) { + psDec->LPC_order = MIN_LPC_ORDER; + psDec->psNLSF_CB = &silk_NLSF_CB_NB_MB; + } else { + psDec->LPC_order = MAX_LPC_ORDER; + psDec->psNLSF_CB = &silk_NLSF_CB_WB; + } + if( fs_kHz == 16 ) { + psDec->pitch_lag_low_bits_iCDF = silk_uniform8_iCDF; + } else if( fs_kHz == 12 ) { + psDec->pitch_lag_low_bits_iCDF = silk_uniform6_iCDF; + } else if( fs_kHz == 8 ) { + psDec->pitch_lag_low_bits_iCDF = silk_uniform4_iCDF; + } else { + /* unsupported sampling rate */ + celt_assert( 0 ); + } + psDec->first_frame_after_reset = 1; + psDec->lagPrev = 100; + psDec->LastGainIndex = 10; + psDec->prevSignalType = TYPE_NO_VOICE_ACTIVITY; + silk_memset( psDec->outBuf, 0, sizeof(psDec->outBuf)); + silk_memset( psDec->sLPC_Q14_buf, 0, sizeof(psDec->sLPC_Q14_buf) ); + } + + psDec->fs_kHz = fs_kHz; + psDec->frame_length = frame_length; + } + + /* Check that settings are valid */ + celt_assert( psDec->frame_length > 0 && psDec->frame_length <= MAX_FRAME_LENGTH ); + + return ret; +} + diff --git a/media/libopus/silk/define.h b/media/libopus/silk/define.h new file mode 100644 index 0000000000..491c86f33e --- /dev/null +++ b/media/libopus/silk/define.h @@ -0,0 +1,235 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef SILK_DEFINE_H +#define SILK_DEFINE_H + +#include "errors.h" +#include "typedef.h" + +#ifdef __cplusplus +extern "C" +{ +#endif + +/* Max number of encoder channels (1/2) */ +#define ENCODER_NUM_CHANNELS 2 +/* Number of decoder channels (1/2) */ +#define DECODER_NUM_CHANNELS 2 + +#define MAX_FRAMES_PER_PACKET 3 + +/* Limits on bitrate */ +#define MIN_TARGET_RATE_BPS 5000 +#define MAX_TARGET_RATE_BPS 80000 + +/* LBRR thresholds */ +#define LBRR_NB_MIN_RATE_BPS 12000 +#define LBRR_MB_MIN_RATE_BPS 14000 +#define LBRR_WB_MIN_RATE_BPS 16000 + +/* DTX settings */ +#define NB_SPEECH_FRAMES_BEFORE_DTX 10 /* eq 200 ms */ +#define MAX_CONSECUTIVE_DTX 20 /* eq 400 ms */ +#define DTX_ACTIVITY_THRESHOLD 0.1f + +/* VAD decision */ +#define VAD_NO_DECISION -1 +#define VAD_NO_ACTIVITY 0 +#define VAD_ACTIVITY 1 + +/* Maximum sampling frequency */ +#define MAX_FS_KHZ 16 +#define MAX_API_FS_KHZ 48 + +/* Signal types */ +#define TYPE_NO_VOICE_ACTIVITY 0 +#define TYPE_UNVOICED 1 +#define TYPE_VOICED 2 + +/* Conditional coding types */ +#define CODE_INDEPENDENTLY 0 +#define CODE_INDEPENDENTLY_NO_LTP_SCALING 1 +#define CODE_CONDITIONALLY 2 + +/* Settings for stereo processing */ +#define STEREO_QUANT_TAB_SIZE 16 +#define STEREO_QUANT_SUB_STEPS 5 +#define STEREO_INTERP_LEN_MS 8 /* must be even */ +#define STEREO_RATIO_SMOOTH_COEF 0.01 /* smoothing coef for signal norms and stereo width */ + +/* Range of pitch lag estimates */ +#define PITCH_EST_MIN_LAG_MS 2 /* 2 ms -> 500 Hz */ +#define PITCH_EST_MAX_LAG_MS 18 /* 18 ms -> 56 Hz */ + +/* Maximum number of subframes */ +#define MAX_NB_SUBFR 4 + +/* Number of samples per frame */ +#define LTP_MEM_LENGTH_MS 20 +#define SUB_FRAME_LENGTH_MS 5 +#define MAX_SUB_FRAME_LENGTH ( SUB_FRAME_LENGTH_MS * MAX_FS_KHZ ) +#define MAX_FRAME_LENGTH_MS ( SUB_FRAME_LENGTH_MS * MAX_NB_SUBFR ) +#define MAX_FRAME_LENGTH ( MAX_FRAME_LENGTH_MS * MAX_FS_KHZ ) + +/* Milliseconds of lookahead for pitch analysis */ +#define LA_PITCH_MS 2 +#define LA_PITCH_MAX ( LA_PITCH_MS * MAX_FS_KHZ ) + +/* Order of LPC used in find pitch */ +#define MAX_FIND_PITCH_LPC_ORDER 16 + +/* Length of LPC window used in find pitch */ +#define FIND_PITCH_LPC_WIN_MS ( 20 + (LA_PITCH_MS << 1) ) +#define FIND_PITCH_LPC_WIN_MS_2_SF ( 10 + (LA_PITCH_MS << 1) ) +#define FIND_PITCH_LPC_WIN_MAX ( FIND_PITCH_LPC_WIN_MS * MAX_FS_KHZ ) + +/* Milliseconds of lookahead for noise shape analysis */ +#define LA_SHAPE_MS 5 +#define LA_SHAPE_MAX ( LA_SHAPE_MS * MAX_FS_KHZ ) + +/* Maximum length of LPC window used in noise shape analysis */ +#define SHAPE_LPC_WIN_MAX ( 15 * MAX_FS_KHZ ) + +/* dB level of lowest gain quantization level */ +#define MIN_QGAIN_DB 2 +/* dB level of highest gain quantization level */ +#define MAX_QGAIN_DB 88 +/* Number of gain quantization levels */ +#define N_LEVELS_QGAIN 64 +/* Max increase in gain quantization index */ +#define MAX_DELTA_GAIN_QUANT 36 +/* Max decrease in gain quantization index */ +#define MIN_DELTA_GAIN_QUANT -4 + +/* Quantization offsets (multiples of 4) */ +#define OFFSET_VL_Q10 32 +#define OFFSET_VH_Q10 100 +#define OFFSET_UVL_Q10 100 +#define OFFSET_UVH_Q10 240 + +#define QUANT_LEVEL_ADJUST_Q10 80 + +/* Maximum numbers of iterations used to stabilize an LPC vector */ +#define MAX_LPC_STABILIZE_ITERATIONS 16 +#define MAX_PREDICTION_POWER_GAIN 1e4f +#define MAX_PREDICTION_POWER_GAIN_AFTER_RESET 1e2f + +#define MAX_LPC_ORDER 16 +#define MIN_LPC_ORDER 10 + +/* Find Pred Coef defines */ +#define LTP_ORDER 5 + +/* LTP quantization settings */ +#define NB_LTP_CBKS 3 + +/* Flag to use harmonic noise shaping */ +#define USE_HARM_SHAPING 1 + +/* Max LPC order of noise shaping filters */ +#define MAX_SHAPE_LPC_ORDER 24 + +#define HARM_SHAPE_FIR_TAPS 3 + +/* Maximum number of delayed decision states */ +#define MAX_DEL_DEC_STATES 4 + +#define LTP_BUF_LENGTH 512 +#define LTP_MASK ( LTP_BUF_LENGTH - 1 ) + +#define DECISION_DELAY 40 + +/* Number of subframes for excitation entropy coding */ +#define SHELL_CODEC_FRAME_LENGTH 16 +#define LOG2_SHELL_CODEC_FRAME_LENGTH 4 +#define MAX_NB_SHELL_BLOCKS ( MAX_FRAME_LENGTH / SHELL_CODEC_FRAME_LENGTH ) + +/* Number of rate levels, for entropy coding of excitation */ +#define N_RATE_LEVELS 10 + +/* Maximum sum of pulses per shell coding frame */ +#define SILK_MAX_PULSES 16 + +#define MAX_MATRIX_SIZE MAX_LPC_ORDER /* Max of LPC Order and LTP order */ + +# define NSQ_LPC_BUF_LENGTH MAX_LPC_ORDER + +/***************************/ +/* Voice activity detector */ +/***************************/ +#define VAD_N_BANDS 4 + +#define VAD_INTERNAL_SUBFRAMES_LOG2 2 +#define VAD_INTERNAL_SUBFRAMES ( 1 << VAD_INTERNAL_SUBFRAMES_LOG2 ) + +#define VAD_NOISE_LEVEL_SMOOTH_COEF_Q16 1024 /* Must be < 4096 */ +#define VAD_NOISE_LEVELS_BIAS 50 + +/* Sigmoid settings */ +#define VAD_NEGATIVE_OFFSET_Q5 128 /* sigmoid is 0 at -128 */ +#define VAD_SNR_FACTOR_Q16 45000 + +/* smoothing for SNR measurement */ +#define VAD_SNR_SMOOTH_COEF_Q18 4096 + +/* Size of the piecewise linear cosine approximation table for the LSFs */ +#define LSF_COS_TAB_SZ_FIX 128 + +/******************/ +/* NLSF quantizer */ +/******************/ +#define NLSF_W_Q 2 +#define NLSF_VQ_MAX_VECTORS 32 +#define NLSF_QUANT_MAX_AMPLITUDE 4 +#define NLSF_QUANT_MAX_AMPLITUDE_EXT 10 +#define NLSF_QUANT_LEVEL_ADJ 0.1 +#define NLSF_QUANT_DEL_DEC_STATES_LOG2 2 +#define NLSF_QUANT_DEL_DEC_STATES ( 1 << NLSF_QUANT_DEL_DEC_STATES_LOG2 ) + +/* Transition filtering for mode switching */ +#define TRANSITION_TIME_MS 5120 /* 5120 = 64 * FRAME_LENGTH_MS * ( TRANSITION_INT_NUM - 1 ) = 64*(20*4)*/ +#define TRANSITION_NB 3 /* Hardcoded in tables */ +#define TRANSITION_NA 2 /* Hardcoded in tables */ +#define TRANSITION_INT_NUM 5 /* Hardcoded in tables */ +#define TRANSITION_FRAMES ( TRANSITION_TIME_MS / MAX_FRAME_LENGTH_MS ) +#define TRANSITION_INT_STEPS ( TRANSITION_FRAMES / ( TRANSITION_INT_NUM - 1 ) ) + +/* BWE factors to apply after packet loss */ +#define BWE_AFTER_LOSS_Q16 63570 + +/* Defines for CN generation */ +#define CNG_BUF_MASK_MAX 255 /* 2^floor(log2(MAX_FRAME_LENGTH))-1 */ +#define CNG_GAIN_SMTH_Q16 4634 /* 0.25^(1/4) */ +#define CNG_GAIN_SMTH_THRESHOLD_Q16 46396 /* -3 dB */ +#define CNG_NLSF_SMTH_Q16 16348 /* 0.25 */ + +#ifdef __cplusplus +} +#endif + +#endif diff --git a/media/libopus/silk/enc_API.c b/media/libopus/silk/enc_API.c new file mode 100644 index 0000000000..548e07364d --- /dev/null +++ b/media/libopus/silk/enc_API.c @@ -0,0 +1,587 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif +#include "define.h" +#include "API.h" +#include "control.h" +#include "typedef.h" +#include "stack_alloc.h" +#include "structs.h" +#include "tuning_parameters.h" +#ifdef FIXED_POINT +#include "main_FIX.h" +#else +#include "main_FLP.h" +#endif + +/***************************************/ +/* Read control structure from encoder */ +/***************************************/ +static opus_int silk_QueryEncoder( /* O Returns error code */ + const void *encState, /* I State */ + silk_EncControlStruct *encStatus /* O Encoder Status */ +); + +/****************************************/ +/* Encoder functions */ +/****************************************/ + +opus_int silk_Get_Encoder_Size( /* O Returns error code */ + opus_int *encSizeBytes /* O Number of bytes in SILK encoder state */ +) +{ + opus_int ret = SILK_NO_ERROR; + + *encSizeBytes = sizeof( silk_encoder ); + + return ret; +} + +/*************************/ +/* Init or Reset encoder */ +/*************************/ +opus_int silk_InitEncoder( /* O Returns error code */ + void *encState, /* I/O State */ + int arch, /* I Run-time architecture */ + silk_EncControlStruct *encStatus /* O Encoder Status */ +) +{ + silk_encoder *psEnc; + opus_int n, ret = SILK_NO_ERROR; + + psEnc = (silk_encoder *)encState; + + /* Reset encoder */ + silk_memset( psEnc, 0, sizeof( silk_encoder ) ); + for( n = 0; n < ENCODER_NUM_CHANNELS; n++ ) { + if( ret += silk_init_encoder( &psEnc->state_Fxx[ n ], arch ) ) { + celt_assert( 0 ); + } + } + + psEnc->nChannelsAPI = 1; + psEnc->nChannelsInternal = 1; + + /* Read control structure */ + if( ret += silk_QueryEncoder( encState, encStatus ) ) { + celt_assert( 0 ); + } + + return ret; +} + +/***************************************/ +/* Read control structure from encoder */ +/***************************************/ +static opus_int silk_QueryEncoder( /* O Returns error code */ + const void *encState, /* I State */ + silk_EncControlStruct *encStatus /* O Encoder Status */ +) +{ + opus_int ret = SILK_NO_ERROR; + silk_encoder_state_Fxx *state_Fxx; + silk_encoder *psEnc = (silk_encoder *)encState; + + state_Fxx = psEnc->state_Fxx; + + encStatus->nChannelsAPI = psEnc->nChannelsAPI; + encStatus->nChannelsInternal = psEnc->nChannelsInternal; + encStatus->API_sampleRate = state_Fxx[ 0 ].sCmn.API_fs_Hz; + encStatus->maxInternalSampleRate = state_Fxx[ 0 ].sCmn.maxInternal_fs_Hz; + encStatus->minInternalSampleRate = state_Fxx[ 0 ].sCmn.minInternal_fs_Hz; + encStatus->desiredInternalSampleRate = state_Fxx[ 0 ].sCmn.desiredInternal_fs_Hz; + encStatus->payloadSize_ms = state_Fxx[ 0 ].sCmn.PacketSize_ms; + encStatus->bitRate = state_Fxx[ 0 ].sCmn.TargetRate_bps; + encStatus->packetLossPercentage = state_Fxx[ 0 ].sCmn.PacketLoss_perc; + encStatus->complexity = state_Fxx[ 0 ].sCmn.Complexity; + encStatus->useInBandFEC = state_Fxx[ 0 ].sCmn.useInBandFEC; + encStatus->useDTX = state_Fxx[ 0 ].sCmn.useDTX; + encStatus->useCBR = state_Fxx[ 0 ].sCmn.useCBR; + encStatus->internalSampleRate = silk_SMULBB( state_Fxx[ 0 ].sCmn.fs_kHz, 1000 ); + encStatus->allowBandwidthSwitch = state_Fxx[ 0 ].sCmn.allow_bandwidth_switch; + encStatus->inWBmodeWithoutVariableLP = state_Fxx[ 0 ].sCmn.fs_kHz == 16 && state_Fxx[ 0 ].sCmn.sLP.mode == 0; + + return ret; +} + + +/**************************/ +/* Encode frame with Silk */ +/**************************/ +/* Note: if prefillFlag is set, the input must contain 10 ms of audio, irrespective of what */ +/* encControl->payloadSize_ms is set to */ +opus_int silk_Encode( /* O Returns error code */ + void *encState, /* I/O State */ + silk_EncControlStruct *encControl, /* I Control status */ + const opus_int16 *samplesIn, /* I Speech sample input vector */ + opus_int nSamplesIn, /* I Number of samples in input vector */ + ec_enc *psRangeEnc, /* I/O Compressor data structure */ + opus_int32 *nBytesOut, /* I/O Number of bytes in payload (input: Max bytes) */ + const opus_int prefillFlag, /* I Flag to indicate prefilling buffers no coding */ + opus_int activity /* I Decision of Opus voice activity detector */ +) +{ + opus_int n, i, nBits, flags, tmp_payloadSize_ms = 0, tmp_complexity = 0, ret = 0; + opus_int nSamplesToBuffer, nSamplesToBufferMax, nBlocksOf10ms; + opus_int nSamplesFromInput = 0, nSamplesFromInputMax; + opus_int speech_act_thr_for_switch_Q8; + opus_int32 TargetRate_bps, MStargetRates_bps[ 2 ], channelRate_bps, LBRR_symbol, sum; + silk_encoder *psEnc = ( silk_encoder * )encState; + VARDECL( opus_int16, buf ); + opus_int transition, curr_block, tot_blocks; + SAVE_STACK; + + if (encControl->reducedDependency) + { + psEnc->state_Fxx[0].sCmn.first_frame_after_reset = 1; + psEnc->state_Fxx[1].sCmn.first_frame_after_reset = 1; + } + psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded = psEnc->state_Fxx[ 1 ].sCmn.nFramesEncoded = 0; + + /* Check values in encoder control structure */ + if( ( ret = check_control_input( encControl ) ) != 0 ) { + celt_assert( 0 ); + RESTORE_STACK; + return ret; + } + + encControl->switchReady = 0; + + if( encControl->nChannelsInternal > psEnc->nChannelsInternal ) { + /* Mono -> Stereo transition: init state of second channel and stereo state */ + ret += silk_init_encoder( &psEnc->state_Fxx[ 1 ], psEnc->state_Fxx[ 0 ].sCmn.arch ); + silk_memset( psEnc->sStereo.pred_prev_Q13, 0, sizeof( psEnc->sStereo.pred_prev_Q13 ) ); + silk_memset( psEnc->sStereo.sSide, 0, sizeof( psEnc->sStereo.sSide ) ); + psEnc->sStereo.mid_side_amp_Q0[ 0 ] = 0; + psEnc->sStereo.mid_side_amp_Q0[ 1 ] = 1; + psEnc->sStereo.mid_side_amp_Q0[ 2 ] = 0; + psEnc->sStereo.mid_side_amp_Q0[ 3 ] = 1; + psEnc->sStereo.width_prev_Q14 = 0; + psEnc->sStereo.smth_width_Q14 = SILK_FIX_CONST( 1, 14 ); + if( psEnc->nChannelsAPI == 2 ) { + silk_memcpy( &psEnc->state_Fxx[ 1 ].sCmn.resampler_state, &psEnc->state_Fxx[ 0 ].sCmn.resampler_state, sizeof( silk_resampler_state_struct ) ); + silk_memcpy( &psEnc->state_Fxx[ 1 ].sCmn.In_HP_State, &psEnc->state_Fxx[ 0 ].sCmn.In_HP_State, sizeof( psEnc->state_Fxx[ 1 ].sCmn.In_HP_State ) ); + } + } + + transition = (encControl->payloadSize_ms != psEnc->state_Fxx[ 0 ].sCmn.PacketSize_ms) || (psEnc->nChannelsInternal != encControl->nChannelsInternal); + + psEnc->nChannelsAPI = encControl->nChannelsAPI; + psEnc->nChannelsInternal = encControl->nChannelsInternal; + + nBlocksOf10ms = silk_DIV32( 100 * nSamplesIn, encControl->API_sampleRate ); + tot_blocks = ( nBlocksOf10ms > 1 ) ? nBlocksOf10ms >> 1 : 1; + curr_block = 0; + if( prefillFlag ) { + silk_LP_state save_LP; + /* Only accept input length of 10 ms */ + if( nBlocksOf10ms != 1 ) { + celt_assert( 0 ); + RESTORE_STACK; + return SILK_ENC_INPUT_INVALID_NO_OF_SAMPLES; + } + if ( prefillFlag == 2 ) { + save_LP = psEnc->state_Fxx[ 0 ].sCmn.sLP; + /* Save the sampling rate so the bandwidth switching code can keep handling transitions. */ + save_LP.saved_fs_kHz = psEnc->state_Fxx[ 0 ].sCmn.fs_kHz; + } + /* Reset Encoder */ + for( n = 0; n < encControl->nChannelsInternal; n++ ) { + ret = silk_init_encoder( &psEnc->state_Fxx[ n ], psEnc->state_Fxx[ n ].sCmn.arch ); + /* Restore the variable LP state. */ + if ( prefillFlag == 2 ) { + psEnc->state_Fxx[ n ].sCmn.sLP = save_LP; + } + celt_assert( !ret ); + } + tmp_payloadSize_ms = encControl->payloadSize_ms; + encControl->payloadSize_ms = 10; + tmp_complexity = encControl->complexity; + encControl->complexity = 0; + for( n = 0; n < encControl->nChannelsInternal; n++ ) { + psEnc->state_Fxx[ n ].sCmn.controlled_since_last_payload = 0; + psEnc->state_Fxx[ n ].sCmn.prefillFlag = 1; + } + } else { + /* Only accept input lengths that are a multiple of 10 ms */ + if( nBlocksOf10ms * encControl->API_sampleRate != 100 * nSamplesIn || nSamplesIn < 0 ) { + celt_assert( 0 ); + RESTORE_STACK; + return SILK_ENC_INPUT_INVALID_NO_OF_SAMPLES; + } + /* Make sure no more than one packet can be produced */ + if( 1000 * (opus_int32)nSamplesIn > encControl->payloadSize_ms * encControl->API_sampleRate ) { + celt_assert( 0 ); + RESTORE_STACK; + return SILK_ENC_INPUT_INVALID_NO_OF_SAMPLES; + } + } + + for( n = 0; n < encControl->nChannelsInternal; n++ ) { + /* Force the side channel to the same rate as the mid */ + opus_int force_fs_kHz = (n==1) ? psEnc->state_Fxx[0].sCmn.fs_kHz : 0; + if( ( ret = silk_control_encoder( &psEnc->state_Fxx[ n ], encControl, psEnc->allowBandwidthSwitch, n, force_fs_kHz ) ) != 0 ) { + silk_assert( 0 ); + RESTORE_STACK; + return ret; + } + if( psEnc->state_Fxx[n].sCmn.first_frame_after_reset || transition ) { + for( i = 0; i < psEnc->state_Fxx[ 0 ].sCmn.nFramesPerPacket; i++ ) { + psEnc->state_Fxx[ n ].sCmn.LBRR_flags[ i ] = 0; + } + } + psEnc->state_Fxx[ n ].sCmn.inDTX = psEnc->state_Fxx[ n ].sCmn.useDTX; + } + celt_assert( encControl->nChannelsInternal == 1 || psEnc->state_Fxx[ 0 ].sCmn.fs_kHz == psEnc->state_Fxx[ 1 ].sCmn.fs_kHz ); + + /* Input buffering/resampling and encoding */ + nSamplesToBufferMax = + 10 * nBlocksOf10ms * psEnc->state_Fxx[ 0 ].sCmn.fs_kHz; + nSamplesFromInputMax = + silk_DIV32_16( nSamplesToBufferMax * + psEnc->state_Fxx[ 0 ].sCmn.API_fs_Hz, + psEnc->state_Fxx[ 0 ].sCmn.fs_kHz * 1000 ); + ALLOC( buf, nSamplesFromInputMax, opus_int16 ); + while( 1 ) { + int curr_nBitsUsedLBRR = 0; + nSamplesToBuffer = psEnc->state_Fxx[ 0 ].sCmn.frame_length - psEnc->state_Fxx[ 0 ].sCmn.inputBufIx; + nSamplesToBuffer = silk_min( nSamplesToBuffer, nSamplesToBufferMax ); + nSamplesFromInput = silk_DIV32_16( nSamplesToBuffer * psEnc->state_Fxx[ 0 ].sCmn.API_fs_Hz, psEnc->state_Fxx[ 0 ].sCmn.fs_kHz * 1000 ); + /* Resample and write to buffer */ + if( encControl->nChannelsAPI == 2 && encControl->nChannelsInternal == 2 ) { + opus_int id = psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded; + for( n = 0; n < nSamplesFromInput; n++ ) { + buf[ n ] = samplesIn[ 2 * n ]; + } + /* Making sure to start both resamplers from the same state when switching from mono to stereo */ + if( psEnc->nPrevChannelsInternal == 1 && id==0 ) { + silk_memcpy( &psEnc->state_Fxx[ 1 ].sCmn.resampler_state, &psEnc->state_Fxx[ 0 ].sCmn.resampler_state, sizeof(psEnc->state_Fxx[ 1 ].sCmn.resampler_state)); + } + + ret += silk_resampler( &psEnc->state_Fxx[ 0 ].sCmn.resampler_state, + &psEnc->state_Fxx[ 0 ].sCmn.inputBuf[ psEnc->state_Fxx[ 0 ].sCmn.inputBufIx + 2 ], buf, nSamplesFromInput ); + psEnc->state_Fxx[ 0 ].sCmn.inputBufIx += nSamplesToBuffer; + + nSamplesToBuffer = psEnc->state_Fxx[ 1 ].sCmn.frame_length - psEnc->state_Fxx[ 1 ].sCmn.inputBufIx; + nSamplesToBuffer = silk_min( nSamplesToBuffer, 10 * nBlocksOf10ms * psEnc->state_Fxx[ 1 ].sCmn.fs_kHz ); + for( n = 0; n < nSamplesFromInput; n++ ) { + buf[ n ] = samplesIn[ 2 * n + 1 ]; + } + ret += silk_resampler( &psEnc->state_Fxx[ 1 ].sCmn.resampler_state, + &psEnc->state_Fxx[ 1 ].sCmn.inputBuf[ psEnc->state_Fxx[ 1 ].sCmn.inputBufIx + 2 ], buf, nSamplesFromInput ); + + psEnc->state_Fxx[ 1 ].sCmn.inputBufIx += nSamplesToBuffer; + } else if( encControl->nChannelsAPI == 2 && encControl->nChannelsInternal == 1 ) { + /* Combine left and right channels before resampling */ + for( n = 0; n < nSamplesFromInput; n++ ) { + sum = samplesIn[ 2 * n ] + samplesIn[ 2 * n + 1 ]; + buf[ n ] = (opus_int16)silk_RSHIFT_ROUND( sum, 1 ); + } + ret += silk_resampler( &psEnc->state_Fxx[ 0 ].sCmn.resampler_state, + &psEnc->state_Fxx[ 0 ].sCmn.inputBuf[ psEnc->state_Fxx[ 0 ].sCmn.inputBufIx + 2 ], buf, nSamplesFromInput ); + /* On the first mono frame, average the results for the two resampler states */ + if( psEnc->nPrevChannelsInternal == 2 && psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded == 0 ) { + ret += silk_resampler( &psEnc->state_Fxx[ 1 ].sCmn.resampler_state, + &psEnc->state_Fxx[ 1 ].sCmn.inputBuf[ psEnc->state_Fxx[ 1 ].sCmn.inputBufIx + 2 ], buf, nSamplesFromInput ); + for( n = 0; n < psEnc->state_Fxx[ 0 ].sCmn.frame_length; n++ ) { + psEnc->state_Fxx[ 0 ].sCmn.inputBuf[ psEnc->state_Fxx[ 0 ].sCmn.inputBufIx+n+2 ] = + silk_RSHIFT(psEnc->state_Fxx[ 0 ].sCmn.inputBuf[ psEnc->state_Fxx[ 0 ].sCmn.inputBufIx+n+2 ] + + psEnc->state_Fxx[ 1 ].sCmn.inputBuf[ psEnc->state_Fxx[ 1 ].sCmn.inputBufIx+n+2 ], 1); + } + } + psEnc->state_Fxx[ 0 ].sCmn.inputBufIx += nSamplesToBuffer; + } else { + celt_assert( encControl->nChannelsAPI == 1 && encControl->nChannelsInternal == 1 ); + silk_memcpy(buf, samplesIn, nSamplesFromInput*sizeof(opus_int16)); + ret += silk_resampler( &psEnc->state_Fxx[ 0 ].sCmn.resampler_state, + &psEnc->state_Fxx[ 0 ].sCmn.inputBuf[ psEnc->state_Fxx[ 0 ].sCmn.inputBufIx + 2 ], buf, nSamplesFromInput ); + psEnc->state_Fxx[ 0 ].sCmn.inputBufIx += nSamplesToBuffer; + } + + samplesIn += nSamplesFromInput * encControl->nChannelsAPI; + nSamplesIn -= nSamplesFromInput; + + /* Default */ + psEnc->allowBandwidthSwitch = 0; + + /* Silk encoder */ + if( psEnc->state_Fxx[ 0 ].sCmn.inputBufIx >= psEnc->state_Fxx[ 0 ].sCmn.frame_length ) { + /* Enough data in input buffer, so encode */ + celt_assert( psEnc->state_Fxx[ 0 ].sCmn.inputBufIx == psEnc->state_Fxx[ 0 ].sCmn.frame_length ); + celt_assert( encControl->nChannelsInternal == 1 || psEnc->state_Fxx[ 1 ].sCmn.inputBufIx == psEnc->state_Fxx[ 1 ].sCmn.frame_length ); + + /* Deal with LBRR data */ + if( psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded == 0 && !prefillFlag ) { + /* Create space at start of payload for VAD and FEC flags */ + opus_uint8 iCDF[ 2 ] = { 0, 0 }; + iCDF[ 0 ] = 256 - silk_RSHIFT( 256, ( psEnc->state_Fxx[ 0 ].sCmn.nFramesPerPacket + 1 ) * encControl->nChannelsInternal ); + ec_enc_icdf( psRangeEnc, 0, iCDF, 8 ); + curr_nBitsUsedLBRR = ec_tell( psRangeEnc ); + + /* Encode any LBRR data from previous packet */ + /* Encode LBRR flags */ + for( n = 0; n < encControl->nChannelsInternal; n++ ) { + LBRR_symbol = 0; + for( i = 0; i < psEnc->state_Fxx[ n ].sCmn.nFramesPerPacket; i++ ) { + LBRR_symbol |= silk_LSHIFT( psEnc->state_Fxx[ n ].sCmn.LBRR_flags[ i ], i ); + } + psEnc->state_Fxx[ n ].sCmn.LBRR_flag = LBRR_symbol > 0 ? 1 : 0; + if( LBRR_symbol && psEnc->state_Fxx[ n ].sCmn.nFramesPerPacket > 1 ) { + ec_enc_icdf( psRangeEnc, LBRR_symbol - 1, silk_LBRR_flags_iCDF_ptr[ psEnc->state_Fxx[ n ].sCmn.nFramesPerPacket - 2 ], 8 ); + } + } + + /* Code LBRR indices and excitation signals */ + for( i = 0; i < psEnc->state_Fxx[ 0 ].sCmn.nFramesPerPacket; i++ ) { + for( n = 0; n < encControl->nChannelsInternal; n++ ) { + if( psEnc->state_Fxx[ n ].sCmn.LBRR_flags[ i ] ) { + opus_int condCoding; + + if( encControl->nChannelsInternal == 2 && n == 0 ) { + silk_stereo_encode_pred( psRangeEnc, psEnc->sStereo.predIx[ i ] ); + /* For LBRR data there's no need to code the mid-only flag if the side-channel LBRR flag is set */ + if( psEnc->state_Fxx[ 1 ].sCmn.LBRR_flags[ i ] == 0 ) { + silk_stereo_encode_mid_only( psRangeEnc, psEnc->sStereo.mid_only_flags[ i ] ); + } + } + /* Use conditional coding if previous frame available */ + if( i > 0 && psEnc->state_Fxx[ n ].sCmn.LBRR_flags[ i - 1 ] ) { + condCoding = CODE_CONDITIONALLY; + } else { + condCoding = CODE_INDEPENDENTLY; + } + silk_encode_indices( &psEnc->state_Fxx[ n ].sCmn, psRangeEnc, i, 1, condCoding ); + silk_encode_pulses( psRangeEnc, psEnc->state_Fxx[ n ].sCmn.indices_LBRR[i].signalType, psEnc->state_Fxx[ n ].sCmn.indices_LBRR[i].quantOffsetType, + psEnc->state_Fxx[ n ].sCmn.pulses_LBRR[ i ], psEnc->state_Fxx[ n ].sCmn.frame_length ); + } + } + } + + /* Reset LBRR flags */ + for( n = 0; n < encControl->nChannelsInternal; n++ ) { + silk_memset( psEnc->state_Fxx[ n ].sCmn.LBRR_flags, 0, sizeof( psEnc->state_Fxx[ n ].sCmn.LBRR_flags ) ); + } + curr_nBitsUsedLBRR = ec_tell( psRangeEnc ) - curr_nBitsUsedLBRR; + } + + silk_HP_variable_cutoff( psEnc->state_Fxx ); + + /* Total target bits for packet */ + nBits = silk_DIV32_16( silk_MUL( encControl->bitRate, encControl->payloadSize_ms ), 1000 ); + /* Subtract bits used for LBRR */ + if( !prefillFlag ) { + /* psEnc->nBitsUsedLBRR is an exponential moving average of the LBRR usage, + except that for the first LBRR frame it does no averaging and for the first + frame after after LBRR, it goes back to zero immediately. */ + if ( curr_nBitsUsedLBRR < 10 ) { + psEnc->nBitsUsedLBRR = 0; + } else if ( psEnc->nBitsUsedLBRR < 10) { + psEnc->nBitsUsedLBRR = curr_nBitsUsedLBRR; + } else { + psEnc->nBitsUsedLBRR = ( psEnc->nBitsUsedLBRR + curr_nBitsUsedLBRR ) / 2; + } + nBits -= psEnc->nBitsUsedLBRR; + } + /* Divide by number of uncoded frames left in packet */ + nBits = silk_DIV32_16( nBits, psEnc->state_Fxx[ 0 ].sCmn.nFramesPerPacket ); + /* Convert to bits/second */ + if( encControl->payloadSize_ms == 10 ) { + TargetRate_bps = silk_SMULBB( nBits, 100 ); + } else { + TargetRate_bps = silk_SMULBB( nBits, 50 ); + } + /* Subtract fraction of bits in excess of target in previous frames and packets */ + TargetRate_bps -= silk_DIV32_16( silk_MUL( psEnc->nBitsExceeded, 1000 ), BITRESERVOIR_DECAY_TIME_MS ); + if( !prefillFlag && psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded > 0 ) { + /* Compare actual vs target bits so far in this packet */ + opus_int32 bitsBalance = ec_tell( psRangeEnc ) - psEnc->nBitsUsedLBRR - nBits * psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded; + TargetRate_bps -= silk_DIV32_16( silk_MUL( bitsBalance, 1000 ), BITRESERVOIR_DECAY_TIME_MS ); + } + /* Never exceed input bitrate */ + TargetRate_bps = silk_LIMIT( TargetRate_bps, encControl->bitRate, 5000 ); + + /* Convert Left/Right to Mid/Side */ + if( encControl->nChannelsInternal == 2 ) { + silk_stereo_LR_to_MS( &psEnc->sStereo, &psEnc->state_Fxx[ 0 ].sCmn.inputBuf[ 2 ], &psEnc->state_Fxx[ 1 ].sCmn.inputBuf[ 2 ], + psEnc->sStereo.predIx[ psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded ], &psEnc->sStereo.mid_only_flags[ psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded ], + MStargetRates_bps, TargetRate_bps, psEnc->state_Fxx[ 0 ].sCmn.speech_activity_Q8, encControl->toMono, + psEnc->state_Fxx[ 0 ].sCmn.fs_kHz, psEnc->state_Fxx[ 0 ].sCmn.frame_length ); + if( psEnc->sStereo.mid_only_flags[ psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded ] == 0 ) { + /* Reset side channel encoder memory for first frame with side coding */ + if( psEnc->prev_decode_only_middle == 1 ) { + silk_memset( &psEnc->state_Fxx[ 1 ].sShape, 0, sizeof( psEnc->state_Fxx[ 1 ].sShape ) ); + silk_memset( &psEnc->state_Fxx[ 1 ].sCmn.sNSQ, 0, sizeof( psEnc->state_Fxx[ 1 ].sCmn.sNSQ ) ); + silk_memset( psEnc->state_Fxx[ 1 ].sCmn.prev_NLSFq_Q15, 0, sizeof( psEnc->state_Fxx[ 1 ].sCmn.prev_NLSFq_Q15 ) ); + silk_memset( &psEnc->state_Fxx[ 1 ].sCmn.sLP.In_LP_State, 0, sizeof( psEnc->state_Fxx[ 1 ].sCmn.sLP.In_LP_State ) ); + psEnc->state_Fxx[ 1 ].sCmn.prevLag = 100; + psEnc->state_Fxx[ 1 ].sCmn.sNSQ.lagPrev = 100; + psEnc->state_Fxx[ 1 ].sShape.LastGainIndex = 10; + psEnc->state_Fxx[ 1 ].sCmn.prevSignalType = TYPE_NO_VOICE_ACTIVITY; + psEnc->state_Fxx[ 1 ].sCmn.sNSQ.prev_gain_Q16 = 65536; + psEnc->state_Fxx[ 1 ].sCmn.first_frame_after_reset = 1; + } + silk_encode_do_VAD_Fxx( &psEnc->state_Fxx[ 1 ], activity ); + } else { + psEnc->state_Fxx[ 1 ].sCmn.VAD_flags[ psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded ] = 0; + } + if( !prefillFlag ) { + silk_stereo_encode_pred( psRangeEnc, psEnc->sStereo.predIx[ psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded ] ); + if( psEnc->state_Fxx[ 1 ].sCmn.VAD_flags[ psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded ] == 0 ) { + silk_stereo_encode_mid_only( psRangeEnc, psEnc->sStereo.mid_only_flags[ psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded ] ); + } + } + } else { + /* Buffering */ + silk_memcpy( psEnc->state_Fxx[ 0 ].sCmn.inputBuf, psEnc->sStereo.sMid, 2 * sizeof( opus_int16 ) ); + silk_memcpy( psEnc->sStereo.sMid, &psEnc->state_Fxx[ 0 ].sCmn.inputBuf[ psEnc->state_Fxx[ 0 ].sCmn.frame_length ], 2 * sizeof( opus_int16 ) ); + } + silk_encode_do_VAD_Fxx( &psEnc->state_Fxx[ 0 ], activity ); + + /* Encode */ + for( n = 0; n < encControl->nChannelsInternal; n++ ) { + opus_int maxBits, useCBR; + + /* Handling rate constraints */ + maxBits = encControl->maxBits; + if( tot_blocks == 2 && curr_block == 0 ) { + maxBits = maxBits * 3 / 5; + } else if( tot_blocks == 3 ) { + if( curr_block == 0 ) { + maxBits = maxBits * 2 / 5; + } else if( curr_block == 1 ) { + maxBits = maxBits * 3 / 4; + } + } + useCBR = encControl->useCBR && curr_block == tot_blocks - 1; + + if( encControl->nChannelsInternal == 1 ) { + channelRate_bps = TargetRate_bps; + } else { + channelRate_bps = MStargetRates_bps[ n ]; + if( n == 0 && MStargetRates_bps[ 1 ] > 0 ) { + useCBR = 0; + /* Give mid up to 1/2 of the max bits for that frame */ + maxBits -= encControl->maxBits / ( tot_blocks * 2 ); + } + } + + if( channelRate_bps > 0 ) { + opus_int condCoding; + + silk_control_SNR( &psEnc->state_Fxx[ n ].sCmn, channelRate_bps ); + + /* Use independent coding if no previous frame available */ + if( psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded - n <= 0 ) { + condCoding = CODE_INDEPENDENTLY; + } else if( n > 0 && psEnc->prev_decode_only_middle ) { + /* If we skipped a side frame in this packet, we don't + need LTP scaling; the LTP state is well-defined. */ + condCoding = CODE_INDEPENDENTLY_NO_LTP_SCALING; + } else { + condCoding = CODE_CONDITIONALLY; + } + if( ( ret = silk_encode_frame_Fxx( &psEnc->state_Fxx[ n ], nBytesOut, psRangeEnc, condCoding, maxBits, useCBR ) ) != 0 ) { + silk_assert( 0 ); + } + } + psEnc->state_Fxx[ n ].sCmn.controlled_since_last_payload = 0; + psEnc->state_Fxx[ n ].sCmn.inputBufIx = 0; + psEnc->state_Fxx[ n ].sCmn.nFramesEncoded++; + } + psEnc->prev_decode_only_middle = psEnc->sStereo.mid_only_flags[ psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded - 1 ]; + + /* Insert VAD and FEC flags at beginning of bitstream */ + if( *nBytesOut > 0 && psEnc->state_Fxx[ 0 ].sCmn.nFramesEncoded == psEnc->state_Fxx[ 0 ].sCmn.nFramesPerPacket) { + flags = 0; + for( n = 0; n < encControl->nChannelsInternal; n++ ) { + for( i = 0; i < psEnc->state_Fxx[ n ].sCmn.nFramesPerPacket; i++ ) { + flags = silk_LSHIFT( flags, 1 ); + flags |= psEnc->state_Fxx[ n ].sCmn.VAD_flags[ i ]; + } + flags = silk_LSHIFT( flags, 1 ); + flags |= psEnc->state_Fxx[ n ].sCmn.LBRR_flag; + } + if( !prefillFlag ) { + ec_enc_patch_initial_bits( psRangeEnc, flags, ( psEnc->state_Fxx[ 0 ].sCmn.nFramesPerPacket + 1 ) * encControl->nChannelsInternal ); + } + + /* Return zero bytes if all channels DTXed */ + if( psEnc->state_Fxx[ 0 ].sCmn.inDTX && ( encControl->nChannelsInternal == 1 || psEnc->state_Fxx[ 1 ].sCmn.inDTX ) ) { + *nBytesOut = 0; + } + + psEnc->nBitsExceeded += *nBytesOut * 8; + psEnc->nBitsExceeded -= silk_DIV32_16( silk_MUL( encControl->bitRate, encControl->payloadSize_ms ), 1000 ); + psEnc->nBitsExceeded = silk_LIMIT( psEnc->nBitsExceeded, 0, 10000 ); + + /* Update flag indicating if bandwidth switching is allowed */ + speech_act_thr_for_switch_Q8 = silk_SMLAWB( SILK_FIX_CONST( SPEECH_ACTIVITY_DTX_THRES, 8 ), + SILK_FIX_CONST( ( 1 - SPEECH_ACTIVITY_DTX_THRES ) / MAX_BANDWIDTH_SWITCH_DELAY_MS, 16 + 8 ), psEnc->timeSinceSwitchAllowed_ms ); + if( psEnc->state_Fxx[ 0 ].sCmn.speech_activity_Q8 < speech_act_thr_for_switch_Q8 ) { + psEnc->allowBandwidthSwitch = 1; + psEnc->timeSinceSwitchAllowed_ms = 0; + } else { + psEnc->allowBandwidthSwitch = 0; + psEnc->timeSinceSwitchAllowed_ms += encControl->payloadSize_ms; + } + } + + if( nSamplesIn == 0 ) { + break; + } + } else { + break; + } + curr_block++; + } + + psEnc->nPrevChannelsInternal = encControl->nChannelsInternal; + + encControl->allowBandwidthSwitch = psEnc->allowBandwidthSwitch; + encControl->inWBmodeWithoutVariableLP = psEnc->state_Fxx[ 0 ].sCmn.fs_kHz == 16 && psEnc->state_Fxx[ 0 ].sCmn.sLP.mode == 0; + encControl->internalSampleRate = silk_SMULBB( psEnc->state_Fxx[ 0 ].sCmn.fs_kHz, 1000 ); + encControl->stereoWidth_Q14 = encControl->toMono ? 0 : psEnc->sStereo.smth_width_Q14; + if( prefillFlag ) { + encControl->payloadSize_ms = tmp_payloadSize_ms; + encControl->complexity = tmp_complexity; + for( n = 0; n < encControl->nChannelsInternal; n++ ) { + psEnc->state_Fxx[ n ].sCmn.controlled_since_last_payload = 0; + psEnc->state_Fxx[ n ].sCmn.prefillFlag = 0; + } + } + + encControl->signalType = psEnc->state_Fxx[0].sCmn.indices.signalType; + encControl->offset = silk_Quantization_Offsets_Q10 + [ psEnc->state_Fxx[0].sCmn.indices.signalType >> 1 ] + [ psEnc->state_Fxx[0].sCmn.indices.quantOffsetType ]; + RESTORE_STACK; + return ret; +} + diff --git a/media/libopus/silk/encode_indices.c b/media/libopus/silk/encode_indices.c new file mode 100644 index 0000000000..4bcbc3347b --- /dev/null +++ b/media/libopus/silk/encode_indices.c @@ -0,0 +1,181 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" + +/* Encode side-information parameters to payload */ +void silk_encode_indices( + silk_encoder_state *psEncC, /* I/O Encoder state */ + ec_enc *psRangeEnc, /* I/O Compressor data structure */ + opus_int FrameIndex, /* I Frame number */ + opus_int encode_LBRR, /* I Flag indicating LBRR data is being encoded */ + opus_int condCoding /* I The type of conditional coding to use */ +) +{ + opus_int i, k, typeOffset; + opus_int encode_absolute_lagIndex, delta_lagIndex; + opus_int16 ec_ix[ MAX_LPC_ORDER ]; + opus_uint8 pred_Q8[ MAX_LPC_ORDER ]; + const SideInfoIndices *psIndices; + + if( encode_LBRR ) { + psIndices = &psEncC->indices_LBRR[ FrameIndex ]; + } else { + psIndices = &psEncC->indices; + } + + /*******************************************/ + /* Encode signal type and quantizer offset */ + /*******************************************/ + typeOffset = 2 * psIndices->signalType + psIndices->quantOffsetType; + celt_assert( typeOffset >= 0 && typeOffset < 6 ); + celt_assert( encode_LBRR == 0 || typeOffset >= 2 ); + if( encode_LBRR || typeOffset >= 2 ) { + ec_enc_icdf( psRangeEnc, typeOffset - 2, silk_type_offset_VAD_iCDF, 8 ); + } else { + ec_enc_icdf( psRangeEnc, typeOffset, silk_type_offset_no_VAD_iCDF, 8 ); + } + + /****************/ + /* Encode gains */ + /****************/ + /* first subframe */ + if( condCoding == CODE_CONDITIONALLY ) { + /* conditional coding */ + silk_assert( psIndices->GainsIndices[ 0 ] >= 0 && psIndices->GainsIndices[ 0 ] < MAX_DELTA_GAIN_QUANT - MIN_DELTA_GAIN_QUANT + 1 ); + ec_enc_icdf( psRangeEnc, psIndices->GainsIndices[ 0 ], silk_delta_gain_iCDF, 8 ); + } else { + /* independent coding, in two stages: MSB bits followed by 3 LSBs */ + silk_assert( psIndices->GainsIndices[ 0 ] >= 0 && psIndices->GainsIndices[ 0 ] < N_LEVELS_QGAIN ); + ec_enc_icdf( psRangeEnc, silk_RSHIFT( psIndices->GainsIndices[ 0 ], 3 ), silk_gain_iCDF[ psIndices->signalType ], 8 ); + ec_enc_icdf( psRangeEnc, psIndices->GainsIndices[ 0 ] & 7, silk_uniform8_iCDF, 8 ); + } + + /* remaining subframes */ + for( i = 1; i < psEncC->nb_subfr; i++ ) { + silk_assert( psIndices->GainsIndices[ i ] >= 0 && psIndices->GainsIndices[ i ] < MAX_DELTA_GAIN_QUANT - MIN_DELTA_GAIN_QUANT + 1 ); + ec_enc_icdf( psRangeEnc, psIndices->GainsIndices[ i ], silk_delta_gain_iCDF, 8 ); + } + + /****************/ + /* Encode NLSFs */ + /****************/ + ec_enc_icdf( psRangeEnc, psIndices->NLSFIndices[ 0 ], &psEncC->psNLSF_CB->CB1_iCDF[ ( psIndices->signalType >> 1 ) * psEncC->psNLSF_CB->nVectors ], 8 ); + silk_NLSF_unpack( ec_ix, pred_Q8, psEncC->psNLSF_CB, psIndices->NLSFIndices[ 0 ] ); + celt_assert( psEncC->psNLSF_CB->order == psEncC->predictLPCOrder ); + for( i = 0; i < psEncC->psNLSF_CB->order; i++ ) { + if( psIndices->NLSFIndices[ i+1 ] >= NLSF_QUANT_MAX_AMPLITUDE ) { + ec_enc_icdf( psRangeEnc, 2 * NLSF_QUANT_MAX_AMPLITUDE, &psEncC->psNLSF_CB->ec_iCDF[ ec_ix[ i ] ], 8 ); + ec_enc_icdf( psRangeEnc, psIndices->NLSFIndices[ i+1 ] - NLSF_QUANT_MAX_AMPLITUDE, silk_NLSF_EXT_iCDF, 8 ); + } else if( psIndices->NLSFIndices[ i+1 ] <= -NLSF_QUANT_MAX_AMPLITUDE ) { + ec_enc_icdf( psRangeEnc, 0, &psEncC->psNLSF_CB->ec_iCDF[ ec_ix[ i ] ], 8 ); + ec_enc_icdf( psRangeEnc, -psIndices->NLSFIndices[ i+1 ] - NLSF_QUANT_MAX_AMPLITUDE, silk_NLSF_EXT_iCDF, 8 ); + } else { + ec_enc_icdf( psRangeEnc, psIndices->NLSFIndices[ i+1 ] + NLSF_QUANT_MAX_AMPLITUDE, &psEncC->psNLSF_CB->ec_iCDF[ ec_ix[ i ] ], 8 ); + } + } + + /* Encode NLSF interpolation factor */ + if( psEncC->nb_subfr == MAX_NB_SUBFR ) { + silk_assert( psIndices->NLSFInterpCoef_Q2 >= 0 && psIndices->NLSFInterpCoef_Q2 < 5 ); + ec_enc_icdf( psRangeEnc, psIndices->NLSFInterpCoef_Q2, silk_NLSF_interpolation_factor_iCDF, 8 ); + } + + if( psIndices->signalType == TYPE_VOICED ) + { + /*********************/ + /* Encode pitch lags */ + /*********************/ + /* lag index */ + encode_absolute_lagIndex = 1; + if( condCoding == CODE_CONDITIONALLY && psEncC->ec_prevSignalType == TYPE_VOICED ) { + /* Delta Encoding */ + delta_lagIndex = psIndices->lagIndex - psEncC->ec_prevLagIndex; + if( delta_lagIndex < -8 || delta_lagIndex > 11 ) { + delta_lagIndex = 0; + } else { + delta_lagIndex = delta_lagIndex + 9; + encode_absolute_lagIndex = 0; /* Only use delta */ + } + silk_assert( delta_lagIndex >= 0 && delta_lagIndex < 21 ); + ec_enc_icdf( psRangeEnc, delta_lagIndex, silk_pitch_delta_iCDF, 8 ); + } + if( encode_absolute_lagIndex ) { + /* Absolute encoding */ + opus_int32 pitch_high_bits, pitch_low_bits; + pitch_high_bits = silk_DIV32_16( psIndices->lagIndex, silk_RSHIFT( psEncC->fs_kHz, 1 ) ); + pitch_low_bits = psIndices->lagIndex - silk_SMULBB( pitch_high_bits, silk_RSHIFT( psEncC->fs_kHz, 1 ) ); + silk_assert( pitch_low_bits < psEncC->fs_kHz / 2 ); + silk_assert( pitch_high_bits < 32 ); + ec_enc_icdf( psRangeEnc, pitch_high_bits, silk_pitch_lag_iCDF, 8 ); + ec_enc_icdf( psRangeEnc, pitch_low_bits, psEncC->pitch_lag_low_bits_iCDF, 8 ); + } + psEncC->ec_prevLagIndex = psIndices->lagIndex; + + /* Countour index */ + silk_assert( psIndices->contourIndex >= 0 ); + silk_assert( ( psIndices->contourIndex < 34 && psEncC->fs_kHz > 8 && psEncC->nb_subfr == 4 ) || + ( psIndices->contourIndex < 11 && psEncC->fs_kHz == 8 && psEncC->nb_subfr == 4 ) || + ( psIndices->contourIndex < 12 && psEncC->fs_kHz > 8 && psEncC->nb_subfr == 2 ) || + ( psIndices->contourIndex < 3 && psEncC->fs_kHz == 8 && psEncC->nb_subfr == 2 ) ); + ec_enc_icdf( psRangeEnc, psIndices->contourIndex, psEncC->pitch_contour_iCDF, 8 ); + + /********************/ + /* Encode LTP gains */ + /********************/ + /* PERIndex value */ + silk_assert( psIndices->PERIndex >= 0 && psIndices->PERIndex < 3 ); + ec_enc_icdf( psRangeEnc, psIndices->PERIndex, silk_LTP_per_index_iCDF, 8 ); + + /* Codebook Indices */ + for( k = 0; k < psEncC->nb_subfr; k++ ) { + silk_assert( psIndices->LTPIndex[ k ] >= 0 && psIndices->LTPIndex[ k ] < ( 8 << psIndices->PERIndex ) ); + ec_enc_icdf( psRangeEnc, psIndices->LTPIndex[ k ], silk_LTP_gain_iCDF_ptrs[ psIndices->PERIndex ], 8 ); + } + + /**********************/ + /* Encode LTP scaling */ + /**********************/ + if( condCoding == CODE_INDEPENDENTLY ) { + silk_assert( psIndices->LTP_scaleIndex >= 0 && psIndices->LTP_scaleIndex < 3 ); + ec_enc_icdf( psRangeEnc, psIndices->LTP_scaleIndex, silk_LTPscale_iCDF, 8 ); + } + silk_assert( !condCoding || psIndices->LTP_scaleIndex == 0 ); + } + + psEncC->ec_prevSignalType = psIndices->signalType; + + /***************/ + /* Encode seed */ + /***************/ + silk_assert( psIndices->Seed >= 0 && psIndices->Seed < 4 ); + ec_enc_icdf( psRangeEnc, psIndices->Seed, silk_uniform4_iCDF, 8 ); +} diff --git a/media/libopus/silk/encode_pulses.c b/media/libopus/silk/encode_pulses.c new file mode 100644 index 0000000000..8a1999138b --- /dev/null +++ b/media/libopus/silk/encode_pulses.c @@ -0,0 +1,206 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" +#include "stack_alloc.h" + +/*********************************************/ +/* Encode quantization indices of excitation */ +/*********************************************/ + +static OPUS_INLINE opus_int combine_and_check( /* return ok */ + opus_int *pulses_comb, /* O */ + const opus_int *pulses_in, /* I */ + opus_int max_pulses, /* I max value for sum of pulses */ + opus_int len /* I number of output values */ +) +{ + opus_int k, sum; + + for( k = 0; k < len; k++ ) { + sum = pulses_in[ 2 * k ] + pulses_in[ 2 * k + 1 ]; + if( sum > max_pulses ) { + return 1; + } + pulses_comb[ k ] = sum; + } + + return 0; +} + +/* Encode quantization indices of excitation */ +void silk_encode_pulses( + ec_enc *psRangeEnc, /* I/O compressor data structure */ + const opus_int signalType, /* I Signal type */ + const opus_int quantOffsetType, /* I quantOffsetType */ + opus_int8 pulses[], /* I quantization indices */ + const opus_int frame_length /* I Frame length */ +) +{ + opus_int i, k, j, iter, bit, nLS, scale_down, RateLevelIndex = 0; + opus_int32 abs_q, minSumBits_Q5, sumBits_Q5; + VARDECL( opus_int, abs_pulses ); + VARDECL( opus_int, sum_pulses ); + VARDECL( opus_int, nRshifts ); + opus_int pulses_comb[ 8 ]; + opus_int *abs_pulses_ptr; + const opus_int8 *pulses_ptr; + const opus_uint8 *cdf_ptr; + const opus_uint8 *nBits_ptr; + SAVE_STACK; + + silk_memset( pulses_comb, 0, 8 * sizeof( opus_int ) ); /* Fixing Valgrind reported problem*/ + + /****************************/ + /* Prepare for shell coding */ + /****************************/ + /* Calculate number of shell blocks */ + silk_assert( 1 << LOG2_SHELL_CODEC_FRAME_LENGTH == SHELL_CODEC_FRAME_LENGTH ); + iter = silk_RSHIFT( frame_length, LOG2_SHELL_CODEC_FRAME_LENGTH ); + if( iter * SHELL_CODEC_FRAME_LENGTH < frame_length ) { + celt_assert( frame_length == 12 * 10 ); /* Make sure only happens for 10 ms @ 12 kHz */ + iter++; + silk_memset( &pulses[ frame_length ], 0, SHELL_CODEC_FRAME_LENGTH * sizeof(opus_int8)); + } + + /* Take the absolute value of the pulses */ + ALLOC( abs_pulses, iter * SHELL_CODEC_FRAME_LENGTH, opus_int ); + silk_assert( !( SHELL_CODEC_FRAME_LENGTH & 3 ) ); + for( i = 0; i < iter * SHELL_CODEC_FRAME_LENGTH; i+=4 ) { + abs_pulses[i+0] = ( opus_int )silk_abs( pulses[ i + 0 ] ); + abs_pulses[i+1] = ( opus_int )silk_abs( pulses[ i + 1 ] ); + abs_pulses[i+2] = ( opus_int )silk_abs( pulses[ i + 2 ] ); + abs_pulses[i+3] = ( opus_int )silk_abs( pulses[ i + 3 ] ); + } + + /* Calc sum pulses per shell code frame */ + ALLOC( sum_pulses, iter, opus_int ); + ALLOC( nRshifts, iter, opus_int ); + abs_pulses_ptr = abs_pulses; + for( i = 0; i < iter; i++ ) { + nRshifts[ i ] = 0; + + while( 1 ) { + /* 1+1 -> 2 */ + scale_down = combine_and_check( pulses_comb, abs_pulses_ptr, silk_max_pulses_table[ 0 ], 8 ); + /* 2+2 -> 4 */ + scale_down += combine_and_check( pulses_comb, pulses_comb, silk_max_pulses_table[ 1 ], 4 ); + /* 4+4 -> 8 */ + scale_down += combine_and_check( pulses_comb, pulses_comb, silk_max_pulses_table[ 2 ], 2 ); + /* 8+8 -> 16 */ + scale_down += combine_and_check( &sum_pulses[ i ], pulses_comb, silk_max_pulses_table[ 3 ], 1 ); + + if( scale_down ) { + /* We need to downscale the quantization signal */ + nRshifts[ i ]++; + for( k = 0; k < SHELL_CODEC_FRAME_LENGTH; k++ ) { + abs_pulses_ptr[ k ] = silk_RSHIFT( abs_pulses_ptr[ k ], 1 ); + } + } else { + /* Jump out of while(1) loop and go to next shell coding frame */ + break; + } + } + abs_pulses_ptr += SHELL_CODEC_FRAME_LENGTH; + } + + /**************/ + /* Rate level */ + /**************/ + /* find rate level that leads to fewest bits for coding of pulses per block info */ + minSumBits_Q5 = silk_int32_MAX; + for( k = 0; k < N_RATE_LEVELS - 1; k++ ) { + nBits_ptr = silk_pulses_per_block_BITS_Q5[ k ]; + sumBits_Q5 = silk_rate_levels_BITS_Q5[ signalType >> 1 ][ k ]; + for( i = 0; i < iter; i++ ) { + if( nRshifts[ i ] > 0 ) { + sumBits_Q5 += nBits_ptr[ SILK_MAX_PULSES + 1 ]; + } else { + sumBits_Q5 += nBits_ptr[ sum_pulses[ i ] ]; + } + } + if( sumBits_Q5 < minSumBits_Q5 ) { + minSumBits_Q5 = sumBits_Q5; + RateLevelIndex = k; + } + } + ec_enc_icdf( psRangeEnc, RateLevelIndex, silk_rate_levels_iCDF[ signalType >> 1 ], 8 ); + + /***************************************************/ + /* Sum-Weighted-Pulses Encoding */ + /***************************************************/ + cdf_ptr = silk_pulses_per_block_iCDF[ RateLevelIndex ]; + for( i = 0; i < iter; i++ ) { + if( nRshifts[ i ] == 0 ) { + ec_enc_icdf( psRangeEnc, sum_pulses[ i ], cdf_ptr, 8 ); + } else { + ec_enc_icdf( psRangeEnc, SILK_MAX_PULSES + 1, cdf_ptr, 8 ); + for( k = 0; k < nRshifts[ i ] - 1; k++ ) { + ec_enc_icdf( psRangeEnc, SILK_MAX_PULSES + 1, silk_pulses_per_block_iCDF[ N_RATE_LEVELS - 1 ], 8 ); + } + ec_enc_icdf( psRangeEnc, sum_pulses[ i ], silk_pulses_per_block_iCDF[ N_RATE_LEVELS - 1 ], 8 ); + } + } + + /******************/ + /* Shell Encoding */ + /******************/ + for( i = 0; i < iter; i++ ) { + if( sum_pulses[ i ] > 0 ) { + silk_shell_encoder( psRangeEnc, &abs_pulses[ i * SHELL_CODEC_FRAME_LENGTH ] ); + } + } + + /****************/ + /* LSB Encoding */ + /****************/ + for( i = 0; i < iter; i++ ) { + if( nRshifts[ i ] > 0 ) { + pulses_ptr = &pulses[ i * SHELL_CODEC_FRAME_LENGTH ]; + nLS = nRshifts[ i ] - 1; + for( k = 0; k < SHELL_CODEC_FRAME_LENGTH; k++ ) { + abs_q = (opus_int8)silk_abs( pulses_ptr[ k ] ); + for( j = nLS; j > 0; j-- ) { + bit = silk_RSHIFT( abs_q, j ) & 1; + ec_enc_icdf( psRangeEnc, bit, silk_lsb_iCDF, 8 ); + } + bit = abs_q & 1; + ec_enc_icdf( psRangeEnc, bit, silk_lsb_iCDF, 8 ); + } + } + } + + /****************/ + /* Encode signs */ + /****************/ + silk_encode_signs( psRangeEnc, pulses, frame_length, signalType, quantOffsetType, sum_pulses ); + RESTORE_STACK; +} diff --git a/media/libopus/silk/errors.h b/media/libopus/silk/errors.h new file mode 100644 index 0000000000..45070800f2 --- /dev/null +++ b/media/libopus/silk/errors.h @@ -0,0 +1,98 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef SILK_ERRORS_H +#define SILK_ERRORS_H + +#ifdef __cplusplus +extern "C" +{ +#endif + +/******************/ +/* Error messages */ +/******************/ +#define SILK_NO_ERROR 0 + +/**************************/ +/* Encoder error messages */ +/**************************/ + +/* Input length is not a multiple of 10 ms, or length is longer than the packet length */ +#define SILK_ENC_INPUT_INVALID_NO_OF_SAMPLES -101 + +/* Sampling frequency not 8000, 12000 or 16000 Hertz */ +#define SILK_ENC_FS_NOT_SUPPORTED -102 + +/* Packet size not 10, 20, 40, or 60 ms */ +#define SILK_ENC_PACKET_SIZE_NOT_SUPPORTED -103 + +/* Allocated payload buffer too short */ +#define SILK_ENC_PAYLOAD_BUF_TOO_SHORT -104 + +/* Loss rate not between 0 and 100 percent */ +#define SILK_ENC_INVALID_LOSS_RATE -105 + +/* Complexity setting not valid, use 0...10 */ +#define SILK_ENC_INVALID_COMPLEXITY_SETTING -106 + +/* Inband FEC setting not valid, use 0 or 1 */ +#define SILK_ENC_INVALID_INBAND_FEC_SETTING -107 + +/* DTX setting not valid, use 0 or 1 */ +#define SILK_ENC_INVALID_DTX_SETTING -108 + +/* CBR setting not valid, use 0 or 1 */ +#define SILK_ENC_INVALID_CBR_SETTING -109 + +/* Internal encoder error */ +#define SILK_ENC_INTERNAL_ERROR -110 + +/* Internal encoder error */ +#define SILK_ENC_INVALID_NUMBER_OF_CHANNELS_ERROR -111 + +/**************************/ +/* Decoder error messages */ +/**************************/ + +/* Output sampling frequency lower than internal decoded sampling frequency */ +#define SILK_DEC_INVALID_SAMPLING_FREQUENCY -200 + +/* Payload size exceeded the maximum allowed 1024 bytes */ +#define SILK_DEC_PAYLOAD_TOO_LARGE -201 + +/* Payload has bit errors */ +#define SILK_DEC_PAYLOAD_ERROR -202 + +/* Payload has bit errors */ +#define SILK_DEC_INVALID_FRAME_SIZE -203 + +#ifdef __cplusplus +} +#endif + +#endif diff --git a/media/libopus/silk/fixed/LTP_analysis_filter_FIX.c b/media/libopus/silk/fixed/LTP_analysis_filter_FIX.c new file mode 100644 index 0000000000..5574e7069f --- /dev/null +++ b/media/libopus/silk/fixed/LTP_analysis_filter_FIX.c @@ -0,0 +1,90 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main_FIX.h" + +void silk_LTP_analysis_filter_FIX( + opus_int16 *LTP_res, /* O LTP residual signal of length MAX_NB_SUBFR * ( pre_length + subfr_length ) */ + const opus_int16 *x, /* I Pointer to input signal with at least max( pitchL ) preceding samples */ + const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ],/* I LTP_ORDER LTP coefficients for each MAX_NB_SUBFR subframe */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lag, one for each subframe */ + const opus_int32 invGains_Q16[ MAX_NB_SUBFR ], /* I Inverse quantization gains, one for each subframe */ + const opus_int subfr_length, /* I Length of each subframe */ + const opus_int nb_subfr, /* I Number of subframes */ + const opus_int pre_length /* I Length of the preceding samples starting at &x[0] for each subframe */ +) +{ + const opus_int16 *x_ptr, *x_lag_ptr; + opus_int16 Btmp_Q14[ LTP_ORDER ]; + opus_int16 *LTP_res_ptr; + opus_int k, i; + opus_int32 LTP_est; + + x_ptr = x; + LTP_res_ptr = LTP_res; + for( k = 0; k < nb_subfr; k++ ) { + + x_lag_ptr = x_ptr - pitchL[ k ]; + + Btmp_Q14[ 0 ] = LTPCoef_Q14[ k * LTP_ORDER ]; + Btmp_Q14[ 1 ] = LTPCoef_Q14[ k * LTP_ORDER + 1 ]; + Btmp_Q14[ 2 ] = LTPCoef_Q14[ k * LTP_ORDER + 2 ]; + Btmp_Q14[ 3 ] = LTPCoef_Q14[ k * LTP_ORDER + 3 ]; + Btmp_Q14[ 4 ] = LTPCoef_Q14[ k * LTP_ORDER + 4 ]; + + /* LTP analysis FIR filter */ + for( i = 0; i < subfr_length + pre_length; i++ ) { + LTP_res_ptr[ i ] = x_ptr[ i ]; + + /* Long-term prediction */ + LTP_est = silk_SMULBB( x_lag_ptr[ LTP_ORDER / 2 ], Btmp_Q14[ 0 ] ); + LTP_est = silk_SMLABB_ovflw( LTP_est, x_lag_ptr[ 1 ], Btmp_Q14[ 1 ] ); + LTP_est = silk_SMLABB_ovflw( LTP_est, x_lag_ptr[ 0 ], Btmp_Q14[ 2 ] ); + LTP_est = silk_SMLABB_ovflw( LTP_est, x_lag_ptr[ -1 ], Btmp_Q14[ 3 ] ); + LTP_est = silk_SMLABB_ovflw( LTP_est, x_lag_ptr[ -2 ], Btmp_Q14[ 4 ] ); + + LTP_est = silk_RSHIFT_ROUND( LTP_est, 14 ); /* round and -> Q0*/ + + /* Subtract long-term prediction */ + LTP_res_ptr[ i ] = (opus_int16)silk_SAT16( (opus_int32)x_ptr[ i ] - LTP_est ); + + /* Scale residual */ + LTP_res_ptr[ i ] = silk_SMULWB( invGains_Q16[ k ], LTP_res_ptr[ i ] ); + + x_lag_ptr++; + } + + /* Update pointers */ + LTP_res_ptr += subfr_length + pre_length; + x_ptr += subfr_length; + } +} + diff --git a/media/libopus/silk/fixed/LTP_scale_ctrl_FIX.c b/media/libopus/silk/fixed/LTP_scale_ctrl_FIX.c new file mode 100644 index 0000000000..db1016e0b2 --- /dev/null +++ b/media/libopus/silk/fixed/LTP_scale_ctrl_FIX.c @@ -0,0 +1,58 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main_FIX.h" + +/* Calculation of LTP state scaling */ +void silk_LTP_scale_ctrl_FIX( + silk_encoder_state_FIX *psEnc, /* I/O encoder state */ + silk_encoder_control_FIX *psEncCtrl, /* I/O encoder control */ + opus_int condCoding /* I The type of conditional coding to use */ +) +{ + opus_int round_loss; + + if( condCoding == CODE_INDEPENDENTLY ) { + /* Only scale if first frame in packet */ + round_loss = psEnc->sCmn.PacketLoss_perc * psEnc->sCmn.nFramesPerPacket; + if ( psEnc->sCmn.LBRR_flag ) { + /* LBRR reduces the effective loss. In practice, it does not square the loss because + losses aren't independent, but that still seems to work best. We also never go below 2%. */ + round_loss = 2 + silk_SMULBB( round_loss, round_loss ) / 100; + } + psEnc->sCmn.indices.LTP_scaleIndex = silk_SMULBB( psEncCtrl->LTPredCodGain_Q7, round_loss ) > silk_log2lin( 128*7 + 2900-psEnc->sCmn.SNR_dB_Q7 ); + psEnc->sCmn.indices.LTP_scaleIndex += silk_SMULBB( psEncCtrl->LTPredCodGain_Q7, round_loss ) > silk_log2lin( 128*7 + 3900-psEnc->sCmn.SNR_dB_Q7 ); + } else { + /* Default is minimum scaling */ + psEnc->sCmn.indices.LTP_scaleIndex = 0; + } + psEncCtrl->LTP_scale_Q14 = silk_LTPScales_table_Q14[ psEnc->sCmn.indices.LTP_scaleIndex ]; +} diff --git a/media/libopus/silk/fixed/apply_sine_window_FIX.c b/media/libopus/silk/fixed/apply_sine_window_FIX.c new file mode 100644 index 0000000000..03e088a6de --- /dev/null +++ b/media/libopus/silk/fixed/apply_sine_window_FIX.c @@ -0,0 +1,101 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "SigProc_FIX.h" + +/* Apply sine window to signal vector. */ +/* Window types: */ +/* 1 -> sine window from 0 to pi/2 */ +/* 2 -> sine window from pi/2 to pi */ +/* Every other sample is linearly interpolated, for speed. */ +/* Window length must be between 16 and 120 (incl) and a multiple of 4. */ + +/* Matlab code for table: + for k=16:9*4:16+2*9*4, fprintf(' %7.d,', -round(65536*pi ./ (k:4:k+8*4))); fprintf('\n'); end +*/ +static const opus_int16 freq_table_Q16[ 27 ] = { + 12111, 9804, 8235, 7100, 6239, 5565, 5022, 4575, 4202, + 3885, 3612, 3375, 3167, 2984, 2820, 2674, 2542, 2422, + 2313, 2214, 2123, 2038, 1961, 1889, 1822, 1760, 1702, +}; + +void silk_apply_sine_window( + opus_int16 px_win[], /* O Pointer to windowed signal */ + const opus_int16 px[], /* I Pointer to input signal */ + const opus_int win_type, /* I Selects a window type */ + const opus_int length /* I Window length, multiple of 4 */ +) +{ + opus_int k, f_Q16, c_Q16; + opus_int32 S0_Q16, S1_Q16; + + celt_assert( win_type == 1 || win_type == 2 ); + + /* Length must be in a range from 16 to 120 and a multiple of 4 */ + celt_assert( length >= 16 && length <= 120 ); + celt_assert( ( length & 3 ) == 0 ); + + /* Frequency */ + k = ( length >> 2 ) - 4; + celt_assert( k >= 0 && k <= 26 ); + f_Q16 = (opus_int)freq_table_Q16[ k ]; + + /* Factor used for cosine approximation */ + c_Q16 = silk_SMULWB( (opus_int32)f_Q16, -f_Q16 ); + silk_assert( c_Q16 >= -32768 ); + + /* initialize state */ + if( win_type == 1 ) { + /* start from 0 */ + S0_Q16 = 0; + /* approximation of sin(f) */ + S1_Q16 = f_Q16 + silk_RSHIFT( length, 3 ); + } else { + /* start from 1 */ + S0_Q16 = ( (opus_int32)1 << 16 ); + /* approximation of cos(f) */ + S1_Q16 = ( (opus_int32)1 << 16 ) + silk_RSHIFT( c_Q16, 1 ) + silk_RSHIFT( length, 4 ); + } + + /* Uses the recursive equation: sin(n*f) = 2 * cos(f) * sin((n-1)*f) - sin((n-2)*f) */ + /* 4 samples at a time */ + for( k = 0; k < length; k += 4 ) { + px_win[ k ] = (opus_int16)silk_SMULWB( silk_RSHIFT( S0_Q16 + S1_Q16, 1 ), px[ k ] ); + px_win[ k + 1 ] = (opus_int16)silk_SMULWB( S1_Q16, px[ k + 1] ); + S0_Q16 = silk_SMULWB( S1_Q16, c_Q16 ) + silk_LSHIFT( S1_Q16, 1 ) - S0_Q16 + 1; + S0_Q16 = silk_min( S0_Q16, ( (opus_int32)1 << 16 ) ); + + px_win[ k + 2 ] = (opus_int16)silk_SMULWB( silk_RSHIFT( S0_Q16 + S1_Q16, 1 ), px[ k + 2] ); + px_win[ k + 3 ] = (opus_int16)silk_SMULWB( S0_Q16, px[ k + 3 ] ); + S1_Q16 = silk_SMULWB( S0_Q16, c_Q16 ) + silk_LSHIFT( S0_Q16, 1 ) - S1_Q16; + S1_Q16 = silk_min( S1_Q16, ( (opus_int32)1 << 16 ) ); + } +} diff --git a/media/libopus/silk/fixed/arm/warped_autocorrelation_FIX_arm.h b/media/libopus/silk/fixed/arm/warped_autocorrelation_FIX_arm.h new file mode 100644 index 0000000000..1992e43288 --- /dev/null +++ b/media/libopus/silk/fixed/arm/warped_autocorrelation_FIX_arm.h @@ -0,0 +1,68 @@ +/*********************************************************************** +Copyright (c) 2017 Google Inc. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef SILK_WARPED_AUTOCORRELATION_FIX_ARM_H +# define SILK_WARPED_AUTOCORRELATION_FIX_ARM_H + +# include "celt/arm/armcpu.h" + +# if defined(FIXED_POINT) + +# if defined(OPUS_ARM_MAY_HAVE_NEON_INTR) +void silk_warped_autocorrelation_FIX_neon( + opus_int32 *corr, /* O Result [order + 1] */ + opus_int *scale, /* O Scaling of the correlation vector */ + const opus_int16 *input, /* I Input data to correlate */ + const opus_int warping_Q16, /* I Warping coefficient */ + const opus_int length, /* I Length of input */ + const opus_int order /* I Correlation order (even) */ +); + +# if !defined(OPUS_HAVE_RTCD) && defined(OPUS_ARM_PRESUME_NEON) +# define OVERRIDE_silk_warped_autocorrelation_FIX (1) +# define silk_warped_autocorrelation_FIX(corr, scale, input, warping_Q16, length, order, arch) \ + ((void)(arch), PRESUME_NEON(silk_warped_autocorrelation_FIX)(corr, scale, input, warping_Q16, length, order)) +# endif +# endif + +# if !defined(OVERRIDE_silk_warped_autocorrelation_FIX) +/*Is run-time CPU detection enabled on this platform?*/ +# if defined(OPUS_HAVE_RTCD) && (defined(OPUS_ARM_MAY_HAVE_NEON_INTR) && !defined(OPUS_ARM_PRESUME_NEON_INTR)) +extern void (*const SILK_WARPED_AUTOCORRELATION_FIX_IMPL[OPUS_ARCHMASK+1])(opus_int32*, opus_int*, const opus_int16*, const opus_int, const opus_int, const opus_int); +# define OVERRIDE_silk_warped_autocorrelation_FIX (1) +# define silk_warped_autocorrelation_FIX(corr, scale, input, warping_Q16, length, order, arch) \ + ((*SILK_WARPED_AUTOCORRELATION_FIX_IMPL[(arch)&OPUS_ARCHMASK])(corr, scale, input, warping_Q16, length, order)) +# elif defined(OPUS_ARM_PRESUME_NEON_INTR) +# define OVERRIDE_silk_warped_autocorrelation_FIX (1) +# define silk_warped_autocorrelation_FIX(corr, scale, input, warping_Q16, length, order, arch) \ + ((void)(arch), silk_warped_autocorrelation_FIX_neon(corr, scale, input, warping_Q16, length, order)) +# endif +# endif + +# endif /* end FIXED_POINT */ + +#endif /* end SILK_WARPED_AUTOCORRELATION_FIX_ARM_H */ diff --git a/media/libopus/silk/fixed/arm/warped_autocorrelation_FIX_neon_intr.c b/media/libopus/silk/fixed/arm/warped_autocorrelation_FIX_neon_intr.c new file mode 100644 index 0000000000..6f3be025cc --- /dev/null +++ b/media/libopus/silk/fixed/arm/warped_autocorrelation_FIX_neon_intr.c @@ -0,0 +1,265 @@ +/*********************************************************************** +Copyright (c) 2017 Google Inc., Jean-Marc Valin +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include <arm_neon.h> +#ifdef OPUS_CHECK_ASM +# include <string.h> +#endif +#include "stack_alloc.h" +#include "main_FIX.h" + +static OPUS_INLINE void calc_corr( const opus_int32 *const input_QS, opus_int64 *const corr_QC, const opus_int offset, const int32x4_t state_QS_s32x4 ) +{ + int64x2_t corr_QC_s64x2[ 2 ], t_s64x2[ 2 ]; + const int32x4_t input_QS_s32x4 = vld1q_s32( input_QS + offset ); + corr_QC_s64x2[ 0 ] = vld1q_s64( corr_QC + offset + 0 ); + corr_QC_s64x2[ 1 ] = vld1q_s64( corr_QC + offset + 2 ); + t_s64x2[ 0 ] = vmull_s32( vget_low_s32( state_QS_s32x4 ), vget_low_s32( input_QS_s32x4 ) ); + t_s64x2[ 1 ] = vmull_s32( vget_high_s32( state_QS_s32x4 ), vget_high_s32( input_QS_s32x4 ) ); + corr_QC_s64x2[ 0 ] = vsraq_n_s64( corr_QC_s64x2[ 0 ], t_s64x2[ 0 ], 2 * QS - QC ); + corr_QC_s64x2[ 1 ] = vsraq_n_s64( corr_QC_s64x2[ 1 ], t_s64x2[ 1 ], 2 * QS - QC ); + vst1q_s64( corr_QC + offset + 0, corr_QC_s64x2[ 0 ] ); + vst1q_s64( corr_QC + offset + 2, corr_QC_s64x2[ 1 ] ); +} + +static OPUS_INLINE int32x4_t calc_state( const int32x4_t state_QS0_s32x4, const int32x4_t state_QS0_1_s32x4, const int32x4_t state_QS1_1_s32x4, const int32x4_t warping_Q16_s32x4 ) +{ + int32x4_t t_s32x4 = vsubq_s32( state_QS0_s32x4, state_QS0_1_s32x4 ); + t_s32x4 = vqdmulhq_s32( t_s32x4, warping_Q16_s32x4 ); + return vaddq_s32( state_QS1_1_s32x4, t_s32x4 ); +} + +void silk_warped_autocorrelation_FIX_neon( + opus_int32 *corr, /* O Result [order + 1] */ + opus_int *scale, /* O Scaling of the correlation vector */ + const opus_int16 *input, /* I Input data to correlate */ + const opus_int warping_Q16, /* I Warping coefficient */ + const opus_int length, /* I Length of input */ + const opus_int order /* I Correlation order (even) */ +) +{ + if( ( MAX_SHAPE_LPC_ORDER > 24 ) || ( order < 6 ) ) { + silk_warped_autocorrelation_FIX_c( corr, scale, input, warping_Q16, length, order ); + } else { + opus_int n, i, lsh; + opus_int64 corr_QC[ MAX_SHAPE_LPC_ORDER + 1 ] = { 0 }; /* In reverse order */ + opus_int64 corr_QC_orderT; + int64x2_t lsh_s64x2; + const opus_int orderT = ( order + 3 ) & ~3; + opus_int64 *corr_QCT; + opus_int32 *input_QS; + VARDECL( opus_int32, input_QST ); + VARDECL( opus_int32, state ); + SAVE_STACK; + + /* Order must be even */ + silk_assert( ( order & 1 ) == 0 ); + silk_assert( 2 * QS - QC >= 0 ); + + /* The additional +4 is to ensure a later vld1q_s32 call does not overflow. */ + /* Strictly, only +3 is needed but +4 simplifies initialization using the 4x32 neon load. */ + ALLOC( input_QST, length + 2 * MAX_SHAPE_LPC_ORDER + 4, opus_int32 ); + + input_QS = input_QST; + /* input_QS has zero paddings in the beginning and end. */ + vst1q_s32( input_QS, vdupq_n_s32( 0 ) ); + input_QS += 4; + vst1q_s32( input_QS, vdupq_n_s32( 0 ) ); + input_QS += 4; + vst1q_s32( input_QS, vdupq_n_s32( 0 ) ); + input_QS += 4; + vst1q_s32( input_QS, vdupq_n_s32( 0 ) ); + input_QS += 4; + vst1q_s32( input_QS, vdupq_n_s32( 0 ) ); + input_QS += 4; + vst1q_s32( input_QS, vdupq_n_s32( 0 ) ); + input_QS += 4; + + /* Loop over samples */ + for( n = 0; n < length - 7; n += 8, input_QS += 8 ) { + const int16x8_t t0_s16x4 = vld1q_s16( input + n ); + vst1q_s32( input_QS + 0, vshll_n_s16( vget_low_s16( t0_s16x4 ), QS ) ); + vst1q_s32( input_QS + 4, vshll_n_s16( vget_high_s16( t0_s16x4 ), QS ) ); + } + for( ; n < length; n++, input_QS++ ) { + input_QS[ 0 ] = silk_LSHIFT32( (opus_int32)input[ n ], QS ); + } + vst1q_s32( input_QS, vdupq_n_s32( 0 ) ); + input_QS += 4; + vst1q_s32( input_QS, vdupq_n_s32( 0 ) ); + input_QS += 4; + vst1q_s32( input_QS, vdupq_n_s32( 0 ) ); + input_QS += 4; + vst1q_s32( input_QS, vdupq_n_s32( 0 ) ); + input_QS += 4; + vst1q_s32( input_QS, vdupq_n_s32( 0 ) ); + input_QS += 4; + vst1q_s32( input_QS, vdupq_n_s32( 0 ) ); + input_QS += 4; + vst1q_s32( input_QS, vdupq_n_s32( 0 ) ); + input_QS = input_QST + MAX_SHAPE_LPC_ORDER - orderT; + + /* The following loop runs ( length + order ) times, with ( order ) extra epilogues. */ + /* The zero paddings in input_QS guarantee corr_QC's correctness even with the extra epilogues. */ + /* The values of state_QS will be polluted by the extra epilogues, however they are temporary values. */ + + /* Keep the C code here to help understand the intrinsics optimization. */ + /* + { + opus_int32 state_QS[ 2 ][ MAX_SHAPE_LPC_ORDER + 1 ] = { 0 }; + opus_int32 *state_QST[ 3 ]; + state_QST[ 0 ] = state_QS[ 0 ]; + state_QST[ 1 ] = state_QS[ 1 ]; + for( n = 0; n < length + order; n++, input_QS++ ) { + state_QST[ 0 ][ orderT ] = input_QS[ orderT ]; + for( i = 0; i < orderT; i++ ) { + corr_QC[ i ] += silk_RSHIFT64( silk_SMULL( state_QST[ 0 ][ i ], input_QS[ i ] ), 2 * QS - QC ); + state_QST[ 1 ][ i ] = silk_SMLAWB( state_QST[ 1 ][ i + 1 ], state_QST[ 0 ][ i ] - state_QST[ 0 ][ i + 1 ], warping_Q16 ); + } + state_QST[ 2 ] = state_QST[ 0 ]; + state_QST[ 0 ] = state_QST[ 1 ]; + state_QST[ 1 ] = state_QST[ 2 ]; + } + } + */ + + { + const int32x4_t warping_Q16_s32x4 = vdupq_n_s32( warping_Q16 << 15 ); + const opus_int32 *in = input_QS + orderT; + opus_int o = orderT; + int32x4_t state_QS_s32x4[ 3 ][ 2 ]; + + /* The additional +4 is to ensure a later vld1q_s32 call does not overflow. */ + ALLOC( state, length + order + 4, opus_int32 ); + state_QS_s32x4[ 2 ][ 1 ] = vdupq_n_s32( 0 ); + + /* Calculate 8 taps of all inputs in each loop. */ + do { + state_QS_s32x4[ 0 ][ 0 ] = state_QS_s32x4[ 0 ][ 1 ] = + state_QS_s32x4[ 1 ][ 0 ] = state_QS_s32x4[ 1 ][ 1 ] = vdupq_n_s32( 0 ); + n = 0; + do { + calc_corr( input_QS + n, corr_QC, o - 8, state_QS_s32x4[ 0 ][ 0 ] ); + calc_corr( input_QS + n, corr_QC, o - 4, state_QS_s32x4[ 0 ][ 1 ] ); + state_QS_s32x4[ 2 ][ 1 ] = vld1q_s32( in + n ); + vst1q_lane_s32( state + n, state_QS_s32x4[ 0 ][ 0 ], 0 ); + state_QS_s32x4[ 2 ][ 0 ] = vextq_s32( state_QS_s32x4[ 0 ][ 0 ], state_QS_s32x4[ 0 ][ 1 ], 1 ); + state_QS_s32x4[ 2 ][ 1 ] = vextq_s32( state_QS_s32x4[ 0 ][ 1 ], state_QS_s32x4[ 2 ][ 1 ], 1 ); + state_QS_s32x4[ 0 ][ 0 ] = calc_state( state_QS_s32x4[ 0 ][ 0 ], state_QS_s32x4[ 2 ][ 0 ], state_QS_s32x4[ 1 ][ 0 ], warping_Q16_s32x4 ); + state_QS_s32x4[ 0 ][ 1 ] = calc_state( state_QS_s32x4[ 0 ][ 1 ], state_QS_s32x4[ 2 ][ 1 ], state_QS_s32x4[ 1 ][ 1 ], warping_Q16_s32x4 ); + state_QS_s32x4[ 1 ][ 0 ] = state_QS_s32x4[ 2 ][ 0 ]; + state_QS_s32x4[ 1 ][ 1 ] = state_QS_s32x4[ 2 ][ 1 ]; + } while( ++n < ( length + order ) ); + in = state; + o -= 8; + } while( o > 4 ); + + if( o ) { + /* Calculate the last 4 taps of all inputs. */ + opus_int32 *stateT = state; + silk_assert( o == 4 ); + state_QS_s32x4[ 0 ][ 0 ] = state_QS_s32x4[ 1 ][ 0 ] = vdupq_n_s32( 0 ); + n = length + order; + do { + calc_corr( input_QS, corr_QC, 0, state_QS_s32x4[ 0 ][ 0 ] ); + state_QS_s32x4[ 2 ][ 0 ] = vld1q_s32( stateT ); + vst1q_lane_s32( stateT, state_QS_s32x4[ 0 ][ 0 ], 0 ); + state_QS_s32x4[ 2 ][ 0 ] = vextq_s32( state_QS_s32x4[ 0 ][ 0 ], state_QS_s32x4[ 2 ][ 0 ], 1 ); + state_QS_s32x4[ 0 ][ 0 ] = calc_state( state_QS_s32x4[ 0 ][ 0 ], state_QS_s32x4[ 2 ][ 0 ], state_QS_s32x4[ 1 ][ 0 ], warping_Q16_s32x4 ); + state_QS_s32x4[ 1 ][ 0 ] = state_QS_s32x4[ 2 ][ 0 ]; + input_QS++; + stateT++; + } while( --n ); + } + } + + { + const opus_int16 *inputT = input; + int32x4_t t_s32x4; + int64x1_t t_s64x1; + int64x2_t t_s64x2 = vdupq_n_s64( 0 ); + for( n = 0; n <= length - 8; n += 8 ) { + int16x8_t input_s16x8 = vld1q_s16( inputT ); + t_s32x4 = vmull_s16( vget_low_s16( input_s16x8 ), vget_low_s16( input_s16x8 ) ); + t_s32x4 = vmlal_s16( t_s32x4, vget_high_s16( input_s16x8 ), vget_high_s16( input_s16x8 ) ); + t_s64x2 = vaddw_s32( t_s64x2, vget_low_s32( t_s32x4 ) ); + t_s64x2 = vaddw_s32( t_s64x2, vget_high_s32( t_s32x4 ) ); + inputT += 8; + } + t_s64x1 = vadd_s64( vget_low_s64( t_s64x2 ), vget_high_s64( t_s64x2 ) ); + corr_QC_orderT = vget_lane_s64( t_s64x1, 0 ); + for( ; n < length; n++ ) { + corr_QC_orderT += silk_SMULL( input[ n ], input[ n ] ); + } + corr_QC_orderT = silk_LSHIFT64( corr_QC_orderT, QC ); + corr_QC[ orderT ] = corr_QC_orderT; + } + + corr_QCT = corr_QC + orderT - order; + lsh = silk_CLZ64( corr_QC_orderT ) - 35; + lsh = silk_LIMIT( lsh, -12 - QC, 30 - QC ); + *scale = -( QC + lsh ); + silk_assert( *scale >= -30 && *scale <= 12 ); + lsh_s64x2 = vdupq_n_s64( lsh ); + for( i = 0; i <= order - 3; i += 4 ) { + int32x4_t corr_s32x4; + int64x2_t corr_QC0_s64x2, corr_QC1_s64x2; + corr_QC0_s64x2 = vld1q_s64( corr_QCT + i ); + corr_QC1_s64x2 = vld1q_s64( corr_QCT + i + 2 ); + corr_QC0_s64x2 = vshlq_s64( corr_QC0_s64x2, lsh_s64x2 ); + corr_QC1_s64x2 = vshlq_s64( corr_QC1_s64x2, lsh_s64x2 ); + corr_s32x4 = vcombine_s32( vmovn_s64( corr_QC1_s64x2 ), vmovn_s64( corr_QC0_s64x2 ) ); + corr_s32x4 = vrev64q_s32( corr_s32x4 ); + vst1q_s32( corr + order - i - 3, corr_s32x4 ); + } + if( lsh >= 0 ) { + for( ; i < order + 1; i++ ) { + corr[ order - i ] = (opus_int32)silk_CHECK_FIT32( silk_LSHIFT64( corr_QCT[ i ], lsh ) ); + } + } else { + for( ; i < order + 1; i++ ) { + corr[ order - i ] = (opus_int32)silk_CHECK_FIT32( silk_RSHIFT64( corr_QCT[ i ], -lsh ) ); + } + } + silk_assert( corr_QCT[ order ] >= 0 ); /* If breaking, decrease QC*/ + RESTORE_STACK; + } + +#ifdef OPUS_CHECK_ASM + { + opus_int32 corr_c[ MAX_SHAPE_LPC_ORDER + 1 ]; + opus_int scale_c; + silk_warped_autocorrelation_FIX_c( corr_c, &scale_c, input, warping_Q16, length, order ); + silk_assert( !memcmp( corr_c, corr, sizeof( corr_c[ 0 ] ) * ( order + 1 ) ) ); + silk_assert( scale_c == *scale ); + } +#endif +} diff --git a/media/libopus/silk/fixed/autocorr_FIX.c b/media/libopus/silk/fixed/autocorr_FIX.c new file mode 100644 index 0000000000..de95c98693 --- /dev/null +++ b/media/libopus/silk/fixed/autocorr_FIX.c @@ -0,0 +1,48 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "SigProc_FIX.h" +#include "celt_lpc.h" + +/* Compute autocorrelation */ +void silk_autocorr( + opus_int32 *results, /* O Result (length correlationCount) */ + opus_int *scale, /* O Scaling of the correlation vector */ + const opus_int16 *inputData, /* I Input data to correlate */ + const opus_int inputDataSize, /* I Length of input */ + const opus_int correlationCount, /* I Number of correlation taps to compute */ + int arch /* I Run-time architecture */ +) +{ + opus_int corrCount; + corrCount = silk_min_int( inputDataSize, correlationCount ); + *scale = _celt_autocorr(inputData, results, NULL, 0, corrCount-1, inputDataSize, arch); +} diff --git a/media/libopus/silk/fixed/burg_modified_FIX.c b/media/libopus/silk/fixed/burg_modified_FIX.c new file mode 100644 index 0000000000..185a12b178 --- /dev/null +++ b/media/libopus/silk/fixed/burg_modified_FIX.c @@ -0,0 +1,280 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "SigProc_FIX.h" +#include "define.h" +#include "tuning_parameters.h" +#include "pitch.h" + +#define MAX_FRAME_SIZE 384 /* subfr_length * nb_subfr = ( 0.005 * 16000 + 16 ) * 4 = 384 */ + +#define QA 25 +#define N_BITS_HEAD_ROOM 3 +#define MIN_RSHIFTS -16 +#define MAX_RSHIFTS (32 - QA) + +/* Compute reflection coefficients from input signal */ +void silk_burg_modified_c( + opus_int32 *res_nrg, /* O Residual energy */ + opus_int *res_nrg_Q, /* O Residual energy Q value */ + opus_int32 A_Q16[], /* O Prediction coefficients (length order) */ + const opus_int16 x[], /* I Input signal, length: nb_subfr * ( D + subfr_length ) */ + const opus_int32 minInvGain_Q30, /* I Inverse of max prediction gain */ + const opus_int subfr_length, /* I Input signal subframe length (incl. D preceding samples) */ + const opus_int nb_subfr, /* I Number of subframes stacked in x */ + const opus_int D, /* I Order */ + int arch /* I Run-time architecture */ +) +{ + opus_int k, n, s, lz, rshifts, reached_max_gain; + opus_int32 C0, num, nrg, rc_Q31, invGain_Q30, Atmp_QA, Atmp1, tmp1, tmp2, x1, x2; + const opus_int16 *x_ptr; + opus_int32 C_first_row[ SILK_MAX_ORDER_LPC ]; + opus_int32 C_last_row[ SILK_MAX_ORDER_LPC ]; + opus_int32 Af_QA[ SILK_MAX_ORDER_LPC ]; + opus_int32 CAf[ SILK_MAX_ORDER_LPC + 1 ]; + opus_int32 CAb[ SILK_MAX_ORDER_LPC + 1 ]; + opus_int32 xcorr[ SILK_MAX_ORDER_LPC ]; + opus_int64 C0_64; + + celt_assert( subfr_length * nb_subfr <= MAX_FRAME_SIZE ); + + /* Compute autocorrelations, added over subframes */ + C0_64 = silk_inner_prod16( x, x, subfr_length*nb_subfr, arch ); + lz = silk_CLZ64(C0_64); + rshifts = 32 + 1 + N_BITS_HEAD_ROOM - lz; + if (rshifts > MAX_RSHIFTS) rshifts = MAX_RSHIFTS; + if (rshifts < MIN_RSHIFTS) rshifts = MIN_RSHIFTS; + + if (rshifts > 0) { + C0 = (opus_int32)silk_RSHIFT64(C0_64, rshifts ); + } else { + C0 = silk_LSHIFT32((opus_int32)C0_64, -rshifts ); + } + + CAb[ 0 ] = CAf[ 0 ] = C0 + silk_SMMUL( SILK_FIX_CONST( FIND_LPC_COND_FAC, 32 ), C0 ) + 1; /* Q(-rshifts) */ + silk_memset( C_first_row, 0, SILK_MAX_ORDER_LPC * sizeof( opus_int32 ) ); + if( rshifts > 0 ) { + for( s = 0; s < nb_subfr; s++ ) { + x_ptr = x + s * subfr_length; + for( n = 1; n < D + 1; n++ ) { + C_first_row[ n - 1 ] += (opus_int32)silk_RSHIFT64( + silk_inner_prod16( x_ptr, x_ptr + n, subfr_length - n, arch ), rshifts ); + } + } + } else { + for( s = 0; s < nb_subfr; s++ ) { + int i; + opus_int32 d; + x_ptr = x + s * subfr_length; + celt_pitch_xcorr(x_ptr, x_ptr + 1, xcorr, subfr_length - D, D, arch ); + for( n = 1; n < D + 1; n++ ) { + for ( i = n + subfr_length - D, d = 0; i < subfr_length; i++ ) + d = MAC16_16( d, x_ptr[ i ], x_ptr[ i - n ] ); + xcorr[ n - 1 ] += d; + } + for( n = 1; n < D + 1; n++ ) { + C_first_row[ n - 1 ] += silk_LSHIFT32( xcorr[ n - 1 ], -rshifts ); + } + } + } + silk_memcpy( C_last_row, C_first_row, SILK_MAX_ORDER_LPC * sizeof( opus_int32 ) ); + + /* Initialize */ + CAb[ 0 ] = CAf[ 0 ] = C0 + silk_SMMUL( SILK_FIX_CONST( FIND_LPC_COND_FAC, 32 ), C0 ) + 1; /* Q(-rshifts) */ + + invGain_Q30 = (opus_int32)1 << 30; + reached_max_gain = 0; + for( n = 0; n < D; n++ ) { + /* Update first row of correlation matrix (without first element) */ + /* Update last row of correlation matrix (without last element, stored in reversed order) */ + /* Update C * Af */ + /* Update C * flipud(Af) (stored in reversed order) */ + if( rshifts > -2 ) { + for( s = 0; s < nb_subfr; s++ ) { + x_ptr = x + s * subfr_length; + x1 = -silk_LSHIFT32( (opus_int32)x_ptr[ n ], 16 - rshifts ); /* Q(16-rshifts) */ + x2 = -silk_LSHIFT32( (opus_int32)x_ptr[ subfr_length - n - 1 ], 16 - rshifts ); /* Q(16-rshifts) */ + tmp1 = silk_LSHIFT32( (opus_int32)x_ptr[ n ], QA - 16 ); /* Q(QA-16) */ + tmp2 = silk_LSHIFT32( (opus_int32)x_ptr[ subfr_length - n - 1 ], QA - 16 ); /* Q(QA-16) */ + for( k = 0; k < n; k++ ) { + C_first_row[ k ] = silk_SMLAWB( C_first_row[ k ], x1, x_ptr[ n - k - 1 ] ); /* Q( -rshifts ) */ + C_last_row[ k ] = silk_SMLAWB( C_last_row[ k ], x2, x_ptr[ subfr_length - n + k ] ); /* Q( -rshifts ) */ + Atmp_QA = Af_QA[ k ]; + tmp1 = silk_SMLAWB( tmp1, Atmp_QA, x_ptr[ n - k - 1 ] ); /* Q(QA-16) */ + tmp2 = silk_SMLAWB( tmp2, Atmp_QA, x_ptr[ subfr_length - n + k ] ); /* Q(QA-16) */ + } + tmp1 = silk_LSHIFT32( -tmp1, 32 - QA - rshifts ); /* Q(16-rshifts) */ + tmp2 = silk_LSHIFT32( -tmp2, 32 - QA - rshifts ); /* Q(16-rshifts) */ + for( k = 0; k <= n; k++ ) { + CAf[ k ] = silk_SMLAWB( CAf[ k ], tmp1, x_ptr[ n - k ] ); /* Q( -rshift ) */ + CAb[ k ] = silk_SMLAWB( CAb[ k ], tmp2, x_ptr[ subfr_length - n + k - 1 ] ); /* Q( -rshift ) */ + } + } + } else { + for( s = 0; s < nb_subfr; s++ ) { + x_ptr = x + s * subfr_length; + x1 = -silk_LSHIFT32( (opus_int32)x_ptr[ n ], -rshifts ); /* Q( -rshifts ) */ + x2 = -silk_LSHIFT32( (opus_int32)x_ptr[ subfr_length - n - 1 ], -rshifts ); /* Q( -rshifts ) */ + tmp1 = silk_LSHIFT32( (opus_int32)x_ptr[ n ], 17 ); /* Q17 */ + tmp2 = silk_LSHIFT32( (opus_int32)x_ptr[ subfr_length - n - 1 ], 17 ); /* Q17 */ + for( k = 0; k < n; k++ ) { + C_first_row[ k ] = silk_MLA( C_first_row[ k ], x1, x_ptr[ n - k - 1 ] ); /* Q( -rshifts ) */ + C_last_row[ k ] = silk_MLA( C_last_row[ k ], x2, x_ptr[ subfr_length - n + k ] ); /* Q( -rshifts ) */ + Atmp1 = silk_RSHIFT_ROUND( Af_QA[ k ], QA - 17 ); /* Q17 */ + /* We sometimes get overflows in the multiplications (even beyond +/- 2^32), + but they cancel each other and the real result seems to always fit in a 32-bit + signed integer. This was determined experimentally, not theoretically (unfortunately). */ + tmp1 = silk_MLA_ovflw( tmp1, x_ptr[ n - k - 1 ], Atmp1 ); /* Q17 */ + tmp2 = silk_MLA_ovflw( tmp2, x_ptr[ subfr_length - n + k ], Atmp1 ); /* Q17 */ + } + tmp1 = -tmp1; /* Q17 */ + tmp2 = -tmp2; /* Q17 */ + for( k = 0; k <= n; k++ ) { + CAf[ k ] = silk_SMLAWW( CAf[ k ], tmp1, + silk_LSHIFT32( (opus_int32)x_ptr[ n - k ], -rshifts - 1 ) ); /* Q( -rshift ) */ + CAb[ k ] = silk_SMLAWW( CAb[ k ], tmp2, + silk_LSHIFT32( (opus_int32)x_ptr[ subfr_length - n + k - 1 ], -rshifts - 1 ) ); /* Q( -rshift ) */ + } + } + } + + /* Calculate nominator and denominator for the next order reflection (parcor) coefficient */ + tmp1 = C_first_row[ n ]; /* Q( -rshifts ) */ + tmp2 = C_last_row[ n ]; /* Q( -rshifts ) */ + num = 0; /* Q( -rshifts ) */ + nrg = silk_ADD32( CAb[ 0 ], CAf[ 0 ] ); /* Q( 1-rshifts ) */ + for( k = 0; k < n; k++ ) { + Atmp_QA = Af_QA[ k ]; + lz = silk_CLZ32( silk_abs( Atmp_QA ) ) - 1; + lz = silk_min( 32 - QA, lz ); + Atmp1 = silk_LSHIFT32( Atmp_QA, lz ); /* Q( QA + lz ) */ + + tmp1 = silk_ADD_LSHIFT32( tmp1, silk_SMMUL( C_last_row[ n - k - 1 ], Atmp1 ), 32 - QA - lz ); /* Q( -rshifts ) */ + tmp2 = silk_ADD_LSHIFT32( tmp2, silk_SMMUL( C_first_row[ n - k - 1 ], Atmp1 ), 32 - QA - lz ); /* Q( -rshifts ) */ + num = silk_ADD_LSHIFT32( num, silk_SMMUL( CAb[ n - k ], Atmp1 ), 32 - QA - lz ); /* Q( -rshifts ) */ + nrg = silk_ADD_LSHIFT32( nrg, silk_SMMUL( silk_ADD32( CAb[ k + 1 ], CAf[ k + 1 ] ), + Atmp1 ), 32 - QA - lz ); /* Q( 1-rshifts ) */ + } + CAf[ n + 1 ] = tmp1; /* Q( -rshifts ) */ + CAb[ n + 1 ] = tmp2; /* Q( -rshifts ) */ + num = silk_ADD32( num, tmp2 ); /* Q( -rshifts ) */ + num = silk_LSHIFT32( -num, 1 ); /* Q( 1-rshifts ) */ + + /* Calculate the next order reflection (parcor) coefficient */ + if( silk_abs( num ) < nrg ) { + rc_Q31 = silk_DIV32_varQ( num, nrg, 31 ); + } else { + rc_Q31 = ( num > 0 ) ? silk_int32_MAX : silk_int32_MIN; + } + + /* Update inverse prediction gain */ + tmp1 = ( (opus_int32)1 << 30 ) - silk_SMMUL( rc_Q31, rc_Q31 ); + tmp1 = silk_LSHIFT( silk_SMMUL( invGain_Q30, tmp1 ), 2 ); + if( tmp1 <= minInvGain_Q30 ) { + /* Max prediction gain exceeded; set reflection coefficient such that max prediction gain is exactly hit */ + tmp2 = ( (opus_int32)1 << 30 ) - silk_DIV32_varQ( minInvGain_Q30, invGain_Q30, 30 ); /* Q30 */ + rc_Q31 = silk_SQRT_APPROX( tmp2 ); /* Q15 */ + if( rc_Q31 > 0 ) { + /* Newton-Raphson iteration */ + rc_Q31 = silk_RSHIFT32( rc_Q31 + silk_DIV32( tmp2, rc_Q31 ), 1 ); /* Q15 */ + rc_Q31 = silk_LSHIFT32( rc_Q31, 16 ); /* Q31 */ + if( num < 0 ) { + /* Ensure adjusted reflection coefficients has the original sign */ + rc_Q31 = -rc_Q31; + } + } + invGain_Q30 = minInvGain_Q30; + reached_max_gain = 1; + } else { + invGain_Q30 = tmp1; + } + + /* Update the AR coefficients */ + for( k = 0; k < (n + 1) >> 1; k++ ) { + tmp1 = Af_QA[ k ]; /* QA */ + tmp2 = Af_QA[ n - k - 1 ]; /* QA */ + Af_QA[ k ] = silk_ADD_LSHIFT32( tmp1, silk_SMMUL( tmp2, rc_Q31 ), 1 ); /* QA */ + Af_QA[ n - k - 1 ] = silk_ADD_LSHIFT32( tmp2, silk_SMMUL( tmp1, rc_Q31 ), 1 ); /* QA */ + } + Af_QA[ n ] = silk_RSHIFT32( rc_Q31, 31 - QA ); /* QA */ + + if( reached_max_gain ) { + /* Reached max prediction gain; set remaining coefficients to zero and exit loop */ + for( k = n + 1; k < D; k++ ) { + Af_QA[ k ] = 0; + } + break; + } + + /* Update C * Af and C * Ab */ + for( k = 0; k <= n + 1; k++ ) { + tmp1 = CAf[ k ]; /* Q( -rshifts ) */ + tmp2 = CAb[ n - k + 1 ]; /* Q( -rshifts ) */ + CAf[ k ] = silk_ADD_LSHIFT32( tmp1, silk_SMMUL( tmp2, rc_Q31 ), 1 ); /* Q( -rshifts ) */ + CAb[ n - k + 1 ] = silk_ADD_LSHIFT32( tmp2, silk_SMMUL( tmp1, rc_Q31 ), 1 ); /* Q( -rshifts ) */ + } + } + + if( reached_max_gain ) { + for( k = 0; k < D; k++ ) { + /* Scale coefficients */ + A_Q16[ k ] = -silk_RSHIFT_ROUND( Af_QA[ k ], QA - 16 ); + } + /* Subtract energy of preceding samples from C0 */ + if( rshifts > 0 ) { + for( s = 0; s < nb_subfr; s++ ) { + x_ptr = x + s * subfr_length; + C0 -= (opus_int32)silk_RSHIFT64( silk_inner_prod16( x_ptr, x_ptr, D, arch ), rshifts ); + } + } else { + for( s = 0; s < nb_subfr; s++ ) { + x_ptr = x + s * subfr_length; + C0 -= silk_LSHIFT32( silk_inner_prod_aligned( x_ptr, x_ptr, D, arch), -rshifts); + } + } + /* Approximate residual energy */ + *res_nrg = silk_LSHIFT( silk_SMMUL( invGain_Q30, C0 ), 2 ); + *res_nrg_Q = -rshifts; + } else { + /* Return residual energy */ + nrg = CAf[ 0 ]; /* Q( -rshifts ) */ + tmp1 = (opus_int32)1 << 16; /* Q16 */ + for( k = 0; k < D; k++ ) { + Atmp1 = silk_RSHIFT_ROUND( Af_QA[ k ], QA - 16 ); /* Q16 */ + nrg = silk_SMLAWW( nrg, CAf[ k + 1 ], Atmp1 ); /* Q( -rshifts ) */ + tmp1 = silk_SMLAWW( tmp1, Atmp1, Atmp1 ); /* Q16 */ + A_Q16[ k ] = -Atmp1; + } + *res_nrg = silk_SMLAWW( nrg, silk_SMMUL( SILK_FIX_CONST( FIND_LPC_COND_FAC, 32 ), C0 ), -tmp1 );/* Q( -rshifts ) */ + *res_nrg_Q = -rshifts; + } +} diff --git a/media/libopus/silk/fixed/corrMatrix_FIX.c b/media/libopus/silk/fixed/corrMatrix_FIX.c new file mode 100644 index 0000000000..1b4a29c232 --- /dev/null +++ b/media/libopus/silk/fixed/corrMatrix_FIX.c @@ -0,0 +1,150 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +/********************************************************************** + * Correlation Matrix Computations for LS estimate. + **********************************************************************/ + +#include "main_FIX.h" + +/* Calculates correlation vector X'*t */ +void silk_corrVector_FIX( + const opus_int16 *x, /* I x vector [L + order - 1] used to form data matrix X */ + const opus_int16 *t, /* I Target vector [L] */ + const opus_int L, /* I Length of vectors */ + const opus_int order, /* I Max lag for correlation */ + opus_int32 *Xt, /* O Pointer to X'*t correlation vector [order] */ + const opus_int rshifts, /* I Right shifts of correlations */ + int arch /* I Run-time architecture */ +) +{ + opus_int lag, i; + const opus_int16 *ptr1, *ptr2; + opus_int32 inner_prod; + + ptr1 = &x[ order - 1 ]; /* Points to first sample of column 0 of X: X[:,0] */ + ptr2 = t; + /* Calculate X'*t */ + if( rshifts > 0 ) { + /* Right shifting used */ + for( lag = 0; lag < order; lag++ ) { + inner_prod = 0; + for( i = 0; i < L; i++ ) { + inner_prod = silk_ADD_RSHIFT32( inner_prod, silk_SMULBB( ptr1[ i ], ptr2[i] ), rshifts ); + } + Xt[ lag ] = inner_prod; /* X[:,lag]'*t */ + ptr1--; /* Go to next column of X */ + } + } else { + silk_assert( rshifts == 0 ); + for( lag = 0; lag < order; lag++ ) { + Xt[ lag ] = silk_inner_prod_aligned( ptr1, ptr2, L, arch ); /* X[:,lag]'*t */ + ptr1--; /* Go to next column of X */ + } + } +} + +/* Calculates correlation matrix X'*X */ +void silk_corrMatrix_FIX( + const opus_int16 *x, /* I x vector [L + order - 1] used to form data matrix X */ + const opus_int L, /* I Length of vectors */ + const opus_int order, /* I Max lag for correlation */ + opus_int32 *XX, /* O Pointer to X'*X correlation matrix [ order x order ] */ + opus_int32 *nrg, /* O Energy of x vector */ + opus_int *rshifts, /* O Right shifts of correlations and energy */ + int arch /* I Run-time architecture */ +) +{ + opus_int i, j, lag; + opus_int32 energy; + const opus_int16 *ptr1, *ptr2; + + /* Calculate energy to find shift used to fit in 32 bits */ + silk_sum_sqr_shift( nrg, rshifts, x, L + order - 1 ); + energy = *nrg; + + /* Calculate energy of first column (0) of X: X[:,0]'*X[:,0] */ + /* Remove contribution of first order - 1 samples */ + for( i = 0; i < order - 1; i++ ) { + energy -= silk_RSHIFT32( silk_SMULBB( x[ i ], x[ i ] ), *rshifts ); + } + + /* Calculate energy of remaining columns of X: X[:,j]'*X[:,j] */ + /* Fill out the diagonal of the correlation matrix */ + matrix_ptr( XX, 0, 0, order ) = energy; + silk_assert( energy >= 0 ); + ptr1 = &x[ order - 1 ]; /* First sample of column 0 of X */ + for( j = 1; j < order; j++ ) { + energy = silk_SUB32( energy, silk_RSHIFT32( silk_SMULBB( ptr1[ L - j ], ptr1[ L - j ] ), *rshifts ) ); + energy = silk_ADD32( energy, silk_RSHIFT32( silk_SMULBB( ptr1[ -j ], ptr1[ -j ] ), *rshifts ) ); + matrix_ptr( XX, j, j, order ) = energy; + silk_assert( energy >= 0 ); + } + + ptr2 = &x[ order - 2 ]; /* First sample of column 1 of X */ + /* Calculate the remaining elements of the correlation matrix */ + if( *rshifts > 0 ) { + /* Right shifting used */ + for( lag = 1; lag < order; lag++ ) { + /* Inner product of column 0 and column lag: X[:,0]'*X[:,lag] */ + energy = 0; + for( i = 0; i < L; i++ ) { + energy += silk_RSHIFT32( silk_SMULBB( ptr1[ i ], ptr2[i] ), *rshifts ); + } + /* Calculate remaining off diagonal: X[:,j]'*X[:,j + lag] */ + matrix_ptr( XX, lag, 0, order ) = energy; + matrix_ptr( XX, 0, lag, order ) = energy; + for( j = 1; j < ( order - lag ); j++ ) { + energy = silk_SUB32( energy, silk_RSHIFT32( silk_SMULBB( ptr1[ L - j ], ptr2[ L - j ] ), *rshifts ) ); + energy = silk_ADD32( energy, silk_RSHIFT32( silk_SMULBB( ptr1[ -j ], ptr2[ -j ] ), *rshifts ) ); + matrix_ptr( XX, lag + j, j, order ) = energy; + matrix_ptr( XX, j, lag + j, order ) = energy; + } + ptr2--; /* Update pointer to first sample of next column (lag) in X */ + } + } else { + for( lag = 1; lag < order; lag++ ) { + /* Inner product of column 0 and column lag: X[:,0]'*X[:,lag] */ + energy = silk_inner_prod_aligned( ptr1, ptr2, L, arch ); + matrix_ptr( XX, lag, 0, order ) = energy; + matrix_ptr( XX, 0, lag, order ) = energy; + /* Calculate remaining off diagonal: X[:,j]'*X[:,j + lag] */ + for( j = 1; j < ( order - lag ); j++ ) { + energy = silk_SUB32( energy, silk_SMULBB( ptr1[ L - j ], ptr2[ L - j ] ) ); + energy = silk_SMLABB( energy, ptr1[ -j ], ptr2[ -j ] ); + matrix_ptr( XX, lag + j, j, order ) = energy; + matrix_ptr( XX, j, lag + j, order ) = energy; + } + ptr2--;/* Update pointer to first sample of next column (lag) in X */ + } + } +} + diff --git a/media/libopus/silk/fixed/encode_frame_FIX.c b/media/libopus/silk/fixed/encode_frame_FIX.c new file mode 100644 index 0000000000..a02bf87dbb --- /dev/null +++ b/media/libopus/silk/fixed/encode_frame_FIX.c @@ -0,0 +1,448 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include <stdlib.h> +#include "main_FIX.h" +#include "stack_alloc.h" +#include "tuning_parameters.h" + +/* Low Bitrate Redundancy (LBRR) encoding. Reuse all parameters but encode with lower bitrate */ +static OPUS_INLINE void silk_LBRR_encode_FIX( + silk_encoder_state_FIX *psEnc, /* I/O Pointer to Silk FIX encoder state */ + silk_encoder_control_FIX *psEncCtrl, /* I/O Pointer to Silk FIX encoder control struct */ + const opus_int16 x16[], /* I Input signal */ + opus_int condCoding /* I The type of conditional coding used so far for this frame */ +); + +void silk_encode_do_VAD_FIX( + silk_encoder_state_FIX *psEnc, /* I/O Pointer to Silk FIX encoder state */ + opus_int activity /* I Decision of Opus voice activity detector */ +) +{ + const opus_int activity_threshold = SILK_FIX_CONST( SPEECH_ACTIVITY_DTX_THRES, 8 ); + + /****************************/ + /* Voice Activity Detection */ + /****************************/ + silk_VAD_GetSA_Q8( &psEnc->sCmn, psEnc->sCmn.inputBuf + 1, psEnc->sCmn.arch ); + /* If Opus VAD is inactive and Silk VAD is active: lower Silk VAD to just under the threshold */ + if( activity == VAD_NO_ACTIVITY && psEnc->sCmn.speech_activity_Q8 >= activity_threshold ) { + psEnc->sCmn.speech_activity_Q8 = activity_threshold - 1; + } + + /**************************************************/ + /* Convert speech activity into VAD and DTX flags */ + /**************************************************/ + if( psEnc->sCmn.speech_activity_Q8 < activity_threshold ) { + psEnc->sCmn.indices.signalType = TYPE_NO_VOICE_ACTIVITY; + psEnc->sCmn.noSpeechCounter++; + if( psEnc->sCmn.noSpeechCounter <= NB_SPEECH_FRAMES_BEFORE_DTX ) { + psEnc->sCmn.inDTX = 0; + } else if( psEnc->sCmn.noSpeechCounter > MAX_CONSECUTIVE_DTX + NB_SPEECH_FRAMES_BEFORE_DTX ) { + psEnc->sCmn.noSpeechCounter = NB_SPEECH_FRAMES_BEFORE_DTX; + psEnc->sCmn.inDTX = 0; + } + psEnc->sCmn.VAD_flags[ psEnc->sCmn.nFramesEncoded ] = 0; + } else { + psEnc->sCmn.noSpeechCounter = 0; + psEnc->sCmn.inDTX = 0; + psEnc->sCmn.indices.signalType = TYPE_UNVOICED; + psEnc->sCmn.VAD_flags[ psEnc->sCmn.nFramesEncoded ] = 1; + } +} + +/****************/ +/* Encode frame */ +/****************/ +opus_int silk_encode_frame_FIX( + silk_encoder_state_FIX *psEnc, /* I/O Pointer to Silk FIX encoder state */ + opus_int32 *pnBytesOut, /* O Pointer to number of payload bytes; */ + ec_enc *psRangeEnc, /* I/O compressor data structure */ + opus_int condCoding, /* I The type of conditional coding to use */ + opus_int maxBits, /* I If > 0: maximum number of output bits */ + opus_int useCBR /* I Flag to force constant-bitrate operation */ +) +{ + silk_encoder_control_FIX sEncCtrl; + opus_int i, iter, maxIter, found_upper, found_lower, ret = 0; + opus_int16 *x_frame; + ec_enc sRangeEnc_copy, sRangeEnc_copy2; + silk_nsq_state sNSQ_copy, sNSQ_copy2; + opus_int32 seed_copy, nBits, nBits_lower, nBits_upper, gainMult_lower, gainMult_upper; + opus_int32 gainsID, gainsID_lower, gainsID_upper; + opus_int16 gainMult_Q8; + opus_int16 ec_prevLagIndex_copy; + opus_int ec_prevSignalType_copy; + opus_int8 LastGainIndex_copy2; + opus_int gain_lock[ MAX_NB_SUBFR ] = {0}; + opus_int16 best_gain_mult[ MAX_NB_SUBFR ]; + opus_int best_sum[ MAX_NB_SUBFR ]; + SAVE_STACK; + + /* This is totally unnecessary but many compilers (including gcc) are too dumb to realise it */ + LastGainIndex_copy2 = nBits_lower = nBits_upper = gainMult_lower = gainMult_upper = 0; + + psEnc->sCmn.indices.Seed = psEnc->sCmn.frameCounter++ & 3; + + /**************************************************************/ + /* Set up Input Pointers, and insert frame in input buffer */ + /*************************************************************/ + /* start of frame to encode */ + x_frame = psEnc->x_buf + psEnc->sCmn.ltp_mem_length; + + /***************************************/ + /* Ensure smooth bandwidth transitions */ + /***************************************/ + silk_LP_variable_cutoff( &psEnc->sCmn.sLP, psEnc->sCmn.inputBuf + 1, psEnc->sCmn.frame_length ); + + /*******************************************/ + /* Copy new frame to front of input buffer */ + /*******************************************/ + silk_memcpy( x_frame + LA_SHAPE_MS * psEnc->sCmn.fs_kHz, psEnc->sCmn.inputBuf + 1, psEnc->sCmn.frame_length * sizeof( opus_int16 ) ); + + if( !psEnc->sCmn.prefillFlag ) { + VARDECL( opus_int16, res_pitch ); + VARDECL( opus_uint8, ec_buf_copy ); + opus_int16 *res_pitch_frame; + + ALLOC( res_pitch, + psEnc->sCmn.la_pitch + psEnc->sCmn.frame_length + + psEnc->sCmn.ltp_mem_length, opus_int16 ); + /* start of pitch LPC residual frame */ + res_pitch_frame = res_pitch + psEnc->sCmn.ltp_mem_length; + + /*****************************************/ + /* Find pitch lags, initial LPC analysis */ + /*****************************************/ + silk_find_pitch_lags_FIX( psEnc, &sEncCtrl, res_pitch, x_frame - psEnc->sCmn.ltp_mem_length, psEnc->sCmn.arch ); + + /************************/ + /* Noise shape analysis */ + /************************/ + silk_noise_shape_analysis_FIX( psEnc, &sEncCtrl, res_pitch_frame, x_frame, psEnc->sCmn.arch ); + + /***************************************************/ + /* Find linear prediction coefficients (LPC + LTP) */ + /***************************************************/ + silk_find_pred_coefs_FIX( psEnc, &sEncCtrl, res_pitch_frame, x_frame, condCoding ); + + /****************************************/ + /* Process gains */ + /****************************************/ + silk_process_gains_FIX( psEnc, &sEncCtrl, condCoding ); + + /****************************************/ + /* Low Bitrate Redundant Encoding */ + /****************************************/ + silk_LBRR_encode_FIX( psEnc, &sEncCtrl, x_frame, condCoding ); + + /* Loop over quantizer and entropy coding to control bitrate */ + maxIter = 6; + gainMult_Q8 = SILK_FIX_CONST( 1, 8 ); + found_lower = 0; + found_upper = 0; + gainsID = silk_gains_ID( psEnc->sCmn.indices.GainsIndices, psEnc->sCmn.nb_subfr ); + gainsID_lower = -1; + gainsID_upper = -1; + /* Copy part of the input state */ + silk_memcpy( &sRangeEnc_copy, psRangeEnc, sizeof( ec_enc ) ); + silk_memcpy( &sNSQ_copy, &psEnc->sCmn.sNSQ, sizeof( silk_nsq_state ) ); + seed_copy = psEnc->sCmn.indices.Seed; + ec_prevLagIndex_copy = psEnc->sCmn.ec_prevLagIndex; + ec_prevSignalType_copy = psEnc->sCmn.ec_prevSignalType; + ALLOC( ec_buf_copy, 1275, opus_uint8 ); + for( iter = 0; ; iter++ ) { + if( gainsID == gainsID_lower ) { + nBits = nBits_lower; + } else if( gainsID == gainsID_upper ) { + nBits = nBits_upper; + } else { + /* Restore part of the input state */ + if( iter > 0 ) { + silk_memcpy( psRangeEnc, &sRangeEnc_copy, sizeof( ec_enc ) ); + silk_memcpy( &psEnc->sCmn.sNSQ, &sNSQ_copy, sizeof( silk_nsq_state ) ); + psEnc->sCmn.indices.Seed = seed_copy; + psEnc->sCmn.ec_prevLagIndex = ec_prevLagIndex_copy; + psEnc->sCmn.ec_prevSignalType = ec_prevSignalType_copy; + } + + /*****************************************/ + /* Noise shaping quantization */ + /*****************************************/ + if( psEnc->sCmn.nStatesDelayedDecision > 1 || psEnc->sCmn.warping_Q16 > 0 ) { + silk_NSQ_del_dec( &psEnc->sCmn, &psEnc->sCmn.sNSQ, &psEnc->sCmn.indices, x_frame, psEnc->sCmn.pulses, + sEncCtrl.PredCoef_Q12[ 0 ], sEncCtrl.LTPCoef_Q14, sEncCtrl.AR_Q13, sEncCtrl.HarmShapeGain_Q14, + sEncCtrl.Tilt_Q14, sEncCtrl.LF_shp_Q14, sEncCtrl.Gains_Q16, sEncCtrl.pitchL, sEncCtrl.Lambda_Q10, sEncCtrl.LTP_scale_Q14, + psEnc->sCmn.arch ); + } else { + silk_NSQ( &psEnc->sCmn, &psEnc->sCmn.sNSQ, &psEnc->sCmn.indices, x_frame, psEnc->sCmn.pulses, + sEncCtrl.PredCoef_Q12[ 0 ], sEncCtrl.LTPCoef_Q14, sEncCtrl.AR_Q13, sEncCtrl.HarmShapeGain_Q14, + sEncCtrl.Tilt_Q14, sEncCtrl.LF_shp_Q14, sEncCtrl.Gains_Q16, sEncCtrl.pitchL, sEncCtrl.Lambda_Q10, sEncCtrl.LTP_scale_Q14, + psEnc->sCmn.arch); + } + + if ( iter == maxIter && !found_lower ) { + silk_memcpy( &sRangeEnc_copy2, psRangeEnc, sizeof( ec_enc ) ); + } + + /****************************************/ + /* Encode Parameters */ + /****************************************/ + silk_encode_indices( &psEnc->sCmn, psRangeEnc, psEnc->sCmn.nFramesEncoded, 0, condCoding ); + + /****************************************/ + /* Encode Excitation Signal */ + /****************************************/ + silk_encode_pulses( psRangeEnc, psEnc->sCmn.indices.signalType, psEnc->sCmn.indices.quantOffsetType, + psEnc->sCmn.pulses, psEnc->sCmn.frame_length ); + + nBits = ec_tell( psRangeEnc ); + + /* If we still bust after the last iteration, do some damage control. */ + if ( iter == maxIter && !found_lower && nBits > maxBits ) { + silk_memcpy( psRangeEnc, &sRangeEnc_copy2, sizeof( ec_enc ) ); + + /* Keep gains the same as the last frame. */ + psEnc->sShape.LastGainIndex = sEncCtrl.lastGainIndexPrev; + for ( i = 0; i < psEnc->sCmn.nb_subfr; i++ ) { + psEnc->sCmn.indices.GainsIndices[ i ] = 4; + } + if (condCoding != CODE_CONDITIONALLY) { + psEnc->sCmn.indices.GainsIndices[ 0 ] = sEncCtrl.lastGainIndexPrev; + } + psEnc->sCmn.ec_prevLagIndex = ec_prevLagIndex_copy; + psEnc->sCmn.ec_prevSignalType = ec_prevSignalType_copy; + /* Clear all pulses. */ + for ( i = 0; i < psEnc->sCmn.frame_length; i++ ) { + psEnc->sCmn.pulses[ i ] = 0; + } + + silk_encode_indices( &psEnc->sCmn, psRangeEnc, psEnc->sCmn.nFramesEncoded, 0, condCoding ); + + silk_encode_pulses( psRangeEnc, psEnc->sCmn.indices.signalType, psEnc->sCmn.indices.quantOffsetType, + psEnc->sCmn.pulses, psEnc->sCmn.frame_length ); + + nBits = ec_tell( psRangeEnc ); + } + + if( useCBR == 0 && iter == 0 && nBits <= maxBits ) { + break; + } + } + + if( iter == maxIter ) { + if( found_lower && ( gainsID == gainsID_lower || nBits > maxBits ) ) { + /* Restore output state from earlier iteration that did meet the bitrate budget */ + silk_memcpy( psRangeEnc, &sRangeEnc_copy2, sizeof( ec_enc ) ); + celt_assert( sRangeEnc_copy2.offs <= 1275 ); + silk_memcpy( psRangeEnc->buf, ec_buf_copy, sRangeEnc_copy2.offs ); + silk_memcpy( &psEnc->sCmn.sNSQ, &sNSQ_copy2, sizeof( silk_nsq_state ) ); + psEnc->sShape.LastGainIndex = LastGainIndex_copy2; + } + break; + } + + if( nBits > maxBits ) { + if( found_lower == 0 && iter >= 2 ) { + /* Adjust the quantizer's rate/distortion tradeoff and discard previous "upper" results */ + sEncCtrl.Lambda_Q10 = silk_ADD_RSHIFT32( sEncCtrl.Lambda_Q10, sEncCtrl.Lambda_Q10, 1 ); + found_upper = 0; + gainsID_upper = -1; + } else { + found_upper = 1; + nBits_upper = nBits; + gainMult_upper = gainMult_Q8; + gainsID_upper = gainsID; + } + } else if( nBits < maxBits - 5 ) { + found_lower = 1; + nBits_lower = nBits; + gainMult_lower = gainMult_Q8; + if( gainsID != gainsID_lower ) { + gainsID_lower = gainsID; + /* Copy part of the output state */ + silk_memcpy( &sRangeEnc_copy2, psRangeEnc, sizeof( ec_enc ) ); + celt_assert( psRangeEnc->offs <= 1275 ); + silk_memcpy( ec_buf_copy, psRangeEnc->buf, psRangeEnc->offs ); + silk_memcpy( &sNSQ_copy2, &psEnc->sCmn.sNSQ, sizeof( silk_nsq_state ) ); + LastGainIndex_copy2 = psEnc->sShape.LastGainIndex; + } + } else { + /* Within 5 bits of budget: close enough */ + break; + } + + if ( !found_lower && nBits > maxBits ) { + int j; + for ( i = 0; i < psEnc->sCmn.nb_subfr; i++ ) { + int sum=0; + for ( j = i*psEnc->sCmn.subfr_length; j < (i+1)*psEnc->sCmn.subfr_length; j++ ) { + sum += abs( psEnc->sCmn.pulses[j] ); + } + if ( iter == 0 || (sum < best_sum[i] && !gain_lock[i]) ) { + best_sum[i] = sum; + best_gain_mult[i] = gainMult_Q8; + } else { + gain_lock[i] = 1; + } + } + } + if( ( found_lower & found_upper ) == 0 ) { + /* Adjust gain according to high-rate rate/distortion curve */ + if( nBits > maxBits ) { + if (gainMult_Q8 < 16384) { + gainMult_Q8 *= 2; + } else { + gainMult_Q8 = 32767; + } + } else { + opus_int32 gain_factor_Q16; + gain_factor_Q16 = silk_log2lin( silk_LSHIFT( nBits - maxBits, 7 ) / psEnc->sCmn.frame_length + SILK_FIX_CONST( 16, 7 ) ); + gainMult_Q8 = silk_SMULWB( gain_factor_Q16, gainMult_Q8 ); + } + + } else { + /* Adjust gain by interpolating */ + gainMult_Q8 = gainMult_lower + silk_DIV32_16( silk_MUL( gainMult_upper - gainMult_lower, maxBits - nBits_lower ), nBits_upper - nBits_lower ); + /* New gain multplier must be between 25% and 75% of old range (note that gainMult_upper < gainMult_lower) */ + if( gainMult_Q8 > silk_ADD_RSHIFT32( gainMult_lower, gainMult_upper - gainMult_lower, 2 ) ) { + gainMult_Q8 = silk_ADD_RSHIFT32( gainMult_lower, gainMult_upper - gainMult_lower, 2 ); + } else + if( gainMult_Q8 < silk_SUB_RSHIFT32( gainMult_upper, gainMult_upper - gainMult_lower, 2 ) ) { + gainMult_Q8 = silk_SUB_RSHIFT32( gainMult_upper, gainMult_upper - gainMult_lower, 2 ); + } + } + + for( i = 0; i < psEnc->sCmn.nb_subfr; i++ ) { + opus_int16 tmp; + if ( gain_lock[i] ) { + tmp = best_gain_mult[i]; + } else { + tmp = gainMult_Q8; + } + sEncCtrl.Gains_Q16[ i ] = silk_LSHIFT_SAT32( silk_SMULWB( sEncCtrl.GainsUnq_Q16[ i ], tmp ), 8 ); + } + + /* Quantize gains */ + psEnc->sShape.LastGainIndex = sEncCtrl.lastGainIndexPrev; + silk_gains_quant( psEnc->sCmn.indices.GainsIndices, sEncCtrl.Gains_Q16, + &psEnc->sShape.LastGainIndex, condCoding == CODE_CONDITIONALLY, psEnc->sCmn.nb_subfr ); + + /* Unique identifier of gains vector */ + gainsID = silk_gains_ID( psEnc->sCmn.indices.GainsIndices, psEnc->sCmn.nb_subfr ); + } + } + + /* Update input buffer */ + silk_memmove( psEnc->x_buf, &psEnc->x_buf[ psEnc->sCmn.frame_length ], + ( psEnc->sCmn.ltp_mem_length + LA_SHAPE_MS * psEnc->sCmn.fs_kHz ) * sizeof( opus_int16 ) ); + + /* Exit without entropy coding */ + if( psEnc->sCmn.prefillFlag ) { + /* No payload */ + *pnBytesOut = 0; + RESTORE_STACK; + return ret; + } + + /* Parameters needed for next frame */ + psEnc->sCmn.prevLag = sEncCtrl.pitchL[ psEnc->sCmn.nb_subfr - 1 ]; + psEnc->sCmn.prevSignalType = psEnc->sCmn.indices.signalType; + + /****************************************/ + /* Finalize payload */ + /****************************************/ + psEnc->sCmn.first_frame_after_reset = 0; + /* Payload size */ + *pnBytesOut = silk_RSHIFT( ec_tell( psRangeEnc ) + 7, 3 ); + + RESTORE_STACK; + return ret; +} + +/* Low-Bitrate Redundancy (LBRR) encoding. Reuse all parameters but encode excitation at lower bitrate */ +static OPUS_INLINE void silk_LBRR_encode_FIX( + silk_encoder_state_FIX *psEnc, /* I/O Pointer to Silk FIX encoder state */ + silk_encoder_control_FIX *psEncCtrl, /* I/O Pointer to Silk FIX encoder control struct */ + const opus_int16 x16[], /* I Input signal */ + opus_int condCoding /* I The type of conditional coding used so far for this frame */ +) +{ + opus_int32 TempGains_Q16[ MAX_NB_SUBFR ]; + SideInfoIndices *psIndices_LBRR = &psEnc->sCmn.indices_LBRR[ psEnc->sCmn.nFramesEncoded ]; + silk_nsq_state sNSQ_LBRR; + + /*******************************************/ + /* Control use of inband LBRR */ + /*******************************************/ + if( psEnc->sCmn.LBRR_enabled && psEnc->sCmn.speech_activity_Q8 > SILK_FIX_CONST( LBRR_SPEECH_ACTIVITY_THRES, 8 ) ) { + psEnc->sCmn.LBRR_flags[ psEnc->sCmn.nFramesEncoded ] = 1; + + /* Copy noise shaping quantizer state and quantization indices from regular encoding */ + silk_memcpy( &sNSQ_LBRR, &psEnc->sCmn.sNSQ, sizeof( silk_nsq_state ) ); + silk_memcpy( psIndices_LBRR, &psEnc->sCmn.indices, sizeof( SideInfoIndices ) ); + + /* Save original gains */ + silk_memcpy( TempGains_Q16, psEncCtrl->Gains_Q16, psEnc->sCmn.nb_subfr * sizeof( opus_int32 ) ); + + if( psEnc->sCmn.nFramesEncoded == 0 || psEnc->sCmn.LBRR_flags[ psEnc->sCmn.nFramesEncoded - 1 ] == 0 ) { + /* First frame in packet or previous frame not LBRR coded */ + psEnc->sCmn.LBRRprevLastGainIndex = psEnc->sShape.LastGainIndex; + + /* Increase Gains to get target LBRR rate */ + psIndices_LBRR->GainsIndices[ 0 ] = psIndices_LBRR->GainsIndices[ 0 ] + psEnc->sCmn.LBRR_GainIncreases; + psIndices_LBRR->GainsIndices[ 0 ] = silk_min_int( psIndices_LBRR->GainsIndices[ 0 ], N_LEVELS_QGAIN - 1 ); + } + + /* Decode to get gains in sync with decoder */ + /* Overwrite unquantized gains with quantized gains */ + silk_gains_dequant( psEncCtrl->Gains_Q16, psIndices_LBRR->GainsIndices, + &psEnc->sCmn.LBRRprevLastGainIndex, condCoding == CODE_CONDITIONALLY, psEnc->sCmn.nb_subfr ); + + /*****************************************/ + /* Noise shaping quantization */ + /*****************************************/ + if( psEnc->sCmn.nStatesDelayedDecision > 1 || psEnc->sCmn.warping_Q16 > 0 ) { + silk_NSQ_del_dec( &psEnc->sCmn, &sNSQ_LBRR, psIndices_LBRR, x16, + psEnc->sCmn.pulses_LBRR[ psEnc->sCmn.nFramesEncoded ], psEncCtrl->PredCoef_Q12[ 0 ], psEncCtrl->LTPCoef_Q14, + psEncCtrl->AR_Q13, psEncCtrl->HarmShapeGain_Q14, psEncCtrl->Tilt_Q14, psEncCtrl->LF_shp_Q14, + psEncCtrl->Gains_Q16, psEncCtrl->pitchL, psEncCtrl->Lambda_Q10, psEncCtrl->LTP_scale_Q14, psEnc->sCmn.arch ); + } else { + silk_NSQ( &psEnc->sCmn, &sNSQ_LBRR, psIndices_LBRR, x16, + psEnc->sCmn.pulses_LBRR[ psEnc->sCmn.nFramesEncoded ], psEncCtrl->PredCoef_Q12[ 0 ], psEncCtrl->LTPCoef_Q14, + psEncCtrl->AR_Q13, psEncCtrl->HarmShapeGain_Q14, psEncCtrl->Tilt_Q14, psEncCtrl->LF_shp_Q14, + psEncCtrl->Gains_Q16, psEncCtrl->pitchL, psEncCtrl->Lambda_Q10, psEncCtrl->LTP_scale_Q14, psEnc->sCmn.arch ); + } + + /* Restore original gains */ + silk_memcpy( psEncCtrl->Gains_Q16, TempGains_Q16, psEnc->sCmn.nb_subfr * sizeof( opus_int32 ) ); + } +} diff --git a/media/libopus/silk/fixed/find_LPC_FIX.c b/media/libopus/silk/fixed/find_LPC_FIX.c new file mode 100644 index 0000000000..c762a0f2a2 --- /dev/null +++ b/media/libopus/silk/fixed/find_LPC_FIX.c @@ -0,0 +1,151 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main_FIX.h" +#include "stack_alloc.h" +#include "tuning_parameters.h" + +/* Finds LPC vector from correlations, and converts to NLSF */ +void silk_find_LPC_FIX( + silk_encoder_state *psEncC, /* I/O Encoder state */ + opus_int16 NLSF_Q15[], /* O NLSFs */ + const opus_int16 x[], /* I Input signal */ + const opus_int32 minInvGain_Q30 /* I Inverse of max prediction gain */ +) +{ + opus_int k, subfr_length; + opus_int32 a_Q16[ MAX_LPC_ORDER ]; + opus_int isInterpLower, shift; + opus_int32 res_nrg0, res_nrg1; + opus_int rshift0, rshift1; + + /* Used only for LSF interpolation */ + opus_int32 a_tmp_Q16[ MAX_LPC_ORDER ], res_nrg_interp, res_nrg, res_tmp_nrg; + opus_int res_nrg_interp_Q, res_nrg_Q, res_tmp_nrg_Q; + opus_int16 a_tmp_Q12[ MAX_LPC_ORDER ]; + opus_int16 NLSF0_Q15[ MAX_LPC_ORDER ]; + SAVE_STACK; + + subfr_length = psEncC->subfr_length + psEncC->predictLPCOrder; + + /* Default: no interpolation */ + psEncC->indices.NLSFInterpCoef_Q2 = 4; + + /* Burg AR analysis for the full frame */ + silk_burg_modified( &res_nrg, &res_nrg_Q, a_Q16, x, minInvGain_Q30, subfr_length, psEncC->nb_subfr, psEncC->predictLPCOrder, psEncC->arch ); + + if( psEncC->useInterpolatedNLSFs && !psEncC->first_frame_after_reset && psEncC->nb_subfr == MAX_NB_SUBFR ) { + VARDECL( opus_int16, LPC_res ); + + /* Optimal solution for last 10 ms */ + silk_burg_modified( &res_tmp_nrg, &res_tmp_nrg_Q, a_tmp_Q16, x + 2 * subfr_length, minInvGain_Q30, subfr_length, 2, psEncC->predictLPCOrder, psEncC->arch ); + + /* subtract residual energy here, as that's easier than adding it to the */ + /* residual energy of the first 10 ms in each iteration of the search below */ + shift = res_tmp_nrg_Q - res_nrg_Q; + if( shift >= 0 ) { + if( shift < 32 ) { + res_nrg = res_nrg - silk_RSHIFT( res_tmp_nrg, shift ); + } + } else { + silk_assert( shift > -32 ); + res_nrg = silk_RSHIFT( res_nrg, -shift ) - res_tmp_nrg; + res_nrg_Q = res_tmp_nrg_Q; + } + + /* Convert to NLSFs */ + silk_A2NLSF( NLSF_Q15, a_tmp_Q16, psEncC->predictLPCOrder ); + + ALLOC( LPC_res, 2 * subfr_length, opus_int16 ); + + /* Search over interpolation indices to find the one with lowest residual energy */ + for( k = 3; k >= 0; k-- ) { + /* Interpolate NLSFs for first half */ + silk_interpolate( NLSF0_Q15, psEncC->prev_NLSFq_Q15, NLSF_Q15, k, psEncC->predictLPCOrder ); + + /* Convert to LPC for residual energy evaluation */ + silk_NLSF2A( a_tmp_Q12, NLSF0_Q15, psEncC->predictLPCOrder, psEncC->arch ); + + /* Calculate residual energy with NLSF interpolation */ + silk_LPC_analysis_filter( LPC_res, x, a_tmp_Q12, 2 * subfr_length, psEncC->predictLPCOrder, psEncC->arch ); + + silk_sum_sqr_shift( &res_nrg0, &rshift0, LPC_res + psEncC->predictLPCOrder, subfr_length - psEncC->predictLPCOrder ); + silk_sum_sqr_shift( &res_nrg1, &rshift1, LPC_res + psEncC->predictLPCOrder + subfr_length, subfr_length - psEncC->predictLPCOrder ); + + /* Add subframe energies from first half frame */ + shift = rshift0 - rshift1; + if( shift >= 0 ) { + res_nrg1 = silk_RSHIFT( res_nrg1, shift ); + res_nrg_interp_Q = -rshift0; + } else { + res_nrg0 = silk_RSHIFT( res_nrg0, -shift ); + res_nrg_interp_Q = -rshift1; + } + res_nrg_interp = silk_ADD32( res_nrg0, res_nrg1 ); + + /* Compare with first half energy without NLSF interpolation, or best interpolated value so far */ + shift = res_nrg_interp_Q - res_nrg_Q; + if( shift >= 0 ) { + if( silk_RSHIFT( res_nrg_interp, shift ) < res_nrg ) { + isInterpLower = silk_TRUE; + } else { + isInterpLower = silk_FALSE; + } + } else { + if( -shift < 32 ) { + if( res_nrg_interp < silk_RSHIFT( res_nrg, -shift ) ) { + isInterpLower = silk_TRUE; + } else { + isInterpLower = silk_FALSE; + } + } else { + isInterpLower = silk_FALSE; + } + } + + /* Determine whether current interpolated NLSFs are best so far */ + if( isInterpLower == silk_TRUE ) { + /* Interpolation has lower residual energy */ + res_nrg = res_nrg_interp; + res_nrg_Q = res_nrg_interp_Q; + psEncC->indices.NLSFInterpCoef_Q2 = (opus_int8)k; + } + } + } + + if( psEncC->indices.NLSFInterpCoef_Q2 == 4 ) { + /* NLSF interpolation is currently inactive, calculate NLSFs from full frame AR coefficients */ + silk_A2NLSF( NLSF_Q15, a_Q16, psEncC->predictLPCOrder ); + } + + celt_assert( psEncC->indices.NLSFInterpCoef_Q2 == 4 || ( psEncC->useInterpolatedNLSFs && !psEncC->first_frame_after_reset && psEncC->nb_subfr == MAX_NB_SUBFR ) ); + RESTORE_STACK; +} diff --git a/media/libopus/silk/fixed/find_LTP_FIX.c b/media/libopus/silk/fixed/find_LTP_FIX.c new file mode 100644 index 0000000000..62d4afb250 --- /dev/null +++ b/media/libopus/silk/fixed/find_LTP_FIX.c @@ -0,0 +1,99 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main_FIX.h" +#include "tuning_parameters.h" + +void silk_find_LTP_FIX( + opus_int32 XXLTP_Q17[ MAX_NB_SUBFR * LTP_ORDER * LTP_ORDER ], /* O Correlation matrix */ + opus_int32 xXLTP_Q17[ MAX_NB_SUBFR * LTP_ORDER ], /* O Correlation vector */ + const opus_int16 r_ptr[], /* I Residual signal after LPC */ + const opus_int lag[ MAX_NB_SUBFR ], /* I LTP lags */ + const opus_int subfr_length, /* I Subframe length */ + const opus_int nb_subfr, /* I Number of subframes */ + int arch /* I Run-time architecture */ +) +{ + opus_int i, k, extra_shifts; + opus_int xx_shifts, xX_shifts, XX_shifts; + const opus_int16 *lag_ptr; + opus_int32 *XXLTP_Q17_ptr, *xXLTP_Q17_ptr; + opus_int32 xx, nrg, temp; + + xXLTP_Q17_ptr = xXLTP_Q17; + XXLTP_Q17_ptr = XXLTP_Q17; + for( k = 0; k < nb_subfr; k++ ) { + lag_ptr = r_ptr - ( lag[ k ] + LTP_ORDER / 2 ); + + silk_sum_sqr_shift( &xx, &xx_shifts, r_ptr, subfr_length + LTP_ORDER ); /* xx in Q( -xx_shifts ) */ + silk_corrMatrix_FIX( lag_ptr, subfr_length, LTP_ORDER, XXLTP_Q17_ptr, &nrg, &XX_shifts, arch ); /* XXLTP_Q17_ptr and nrg in Q( -XX_shifts ) */ + extra_shifts = xx_shifts - XX_shifts; + if( extra_shifts > 0 ) { + /* Shift XX */ + xX_shifts = xx_shifts; + for( i = 0; i < LTP_ORDER * LTP_ORDER; i++ ) { + XXLTP_Q17_ptr[ i ] = silk_RSHIFT32( XXLTP_Q17_ptr[ i ], extra_shifts ); /* Q( -xX_shifts ) */ + } + nrg = silk_RSHIFT32( nrg, extra_shifts ); /* Q( -xX_shifts ) */ + } else if( extra_shifts < 0 ) { + /* Shift xx */ + xX_shifts = XX_shifts; + xx = silk_RSHIFT32( xx, -extra_shifts ); /* Q( -xX_shifts ) */ + } else { + xX_shifts = xx_shifts; + } + silk_corrVector_FIX( lag_ptr, r_ptr, subfr_length, LTP_ORDER, xXLTP_Q17_ptr, xX_shifts, arch ); /* xXLTP_Q17_ptr in Q( -xX_shifts ) */ + + /* At this point all correlations are in Q(-xX_shifts) */ + temp = silk_SMLAWB( 1, nrg, SILK_FIX_CONST( LTP_CORR_INV_MAX, 16 ) ); + temp = silk_max( temp, xx ); +TIC(div) +#if 0 + for( i = 0; i < LTP_ORDER * LTP_ORDER; i++ ) { + XXLTP_Q17_ptr[ i ] = silk_DIV32_varQ( XXLTP_Q17_ptr[ i ], temp, 17 ); + } + for( i = 0; i < LTP_ORDER; i++ ) { + xXLTP_Q17_ptr[ i ] = silk_DIV32_varQ( xXLTP_Q17_ptr[ i ], temp, 17 ); + } +#else + for( i = 0; i < LTP_ORDER * LTP_ORDER; i++ ) { + XXLTP_Q17_ptr[ i ] = (opus_int32)( silk_LSHIFT64( (opus_int64)XXLTP_Q17_ptr[ i ], 17 ) / temp ); + } + for( i = 0; i < LTP_ORDER; i++ ) { + xXLTP_Q17_ptr[ i ] = (opus_int32)( silk_LSHIFT64( (opus_int64)xXLTP_Q17_ptr[ i ], 17 ) / temp ); + } +#endif +TOC(div) + r_ptr += subfr_length; + XXLTP_Q17_ptr += LTP_ORDER * LTP_ORDER; + xXLTP_Q17_ptr += LTP_ORDER; + } +} diff --git a/media/libopus/silk/fixed/find_pitch_lags_FIX.c b/media/libopus/silk/fixed/find_pitch_lags_FIX.c new file mode 100644 index 0000000000..6c3379f2bb --- /dev/null +++ b/media/libopus/silk/fixed/find_pitch_lags_FIX.c @@ -0,0 +1,143 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main_FIX.h" +#include "stack_alloc.h" +#include "tuning_parameters.h" + +/* Find pitch lags */ +void silk_find_pitch_lags_FIX( + silk_encoder_state_FIX *psEnc, /* I/O encoder state */ + silk_encoder_control_FIX *psEncCtrl, /* I/O encoder control */ + opus_int16 res[], /* O residual */ + const opus_int16 x[], /* I Speech signal */ + int arch /* I Run-time architecture */ +) +{ + opus_int buf_len, i, scale; + opus_int32 thrhld_Q13, res_nrg; + const opus_int16 *x_ptr; + VARDECL( opus_int16, Wsig ); + opus_int16 *Wsig_ptr; + opus_int32 auto_corr[ MAX_FIND_PITCH_LPC_ORDER + 1 ]; + opus_int16 rc_Q15[ MAX_FIND_PITCH_LPC_ORDER ]; + opus_int32 A_Q24[ MAX_FIND_PITCH_LPC_ORDER ]; + opus_int16 A_Q12[ MAX_FIND_PITCH_LPC_ORDER ]; + SAVE_STACK; + + /******************************************/ + /* Set up buffer lengths etc based on Fs */ + /******************************************/ + buf_len = psEnc->sCmn.la_pitch + psEnc->sCmn.frame_length + psEnc->sCmn.ltp_mem_length; + + /* Safety check */ + celt_assert( buf_len >= psEnc->sCmn.pitch_LPC_win_length ); + + /*************************************/ + /* Estimate LPC AR coefficients */ + /*************************************/ + + /* Calculate windowed signal */ + + ALLOC( Wsig, psEnc->sCmn.pitch_LPC_win_length, opus_int16 ); + + /* First LA_LTP samples */ + x_ptr = x + buf_len - psEnc->sCmn.pitch_LPC_win_length; + Wsig_ptr = Wsig; + silk_apply_sine_window( Wsig_ptr, x_ptr, 1, psEnc->sCmn.la_pitch ); + + /* Middle un - windowed samples */ + Wsig_ptr += psEnc->sCmn.la_pitch; + x_ptr += psEnc->sCmn.la_pitch; + silk_memcpy( Wsig_ptr, x_ptr, ( psEnc->sCmn.pitch_LPC_win_length - silk_LSHIFT( psEnc->sCmn.la_pitch, 1 ) ) * sizeof( opus_int16 ) ); + + /* Last LA_LTP samples */ + Wsig_ptr += psEnc->sCmn.pitch_LPC_win_length - silk_LSHIFT( psEnc->sCmn.la_pitch, 1 ); + x_ptr += psEnc->sCmn.pitch_LPC_win_length - silk_LSHIFT( psEnc->sCmn.la_pitch, 1 ); + silk_apply_sine_window( Wsig_ptr, x_ptr, 2, psEnc->sCmn.la_pitch ); + + /* Calculate autocorrelation sequence */ + silk_autocorr( auto_corr, &scale, Wsig, psEnc->sCmn.pitch_LPC_win_length, psEnc->sCmn.pitchEstimationLPCOrder + 1, arch ); + + /* Add white noise, as fraction of energy */ + auto_corr[ 0 ] = silk_SMLAWB( auto_corr[ 0 ], auto_corr[ 0 ], SILK_FIX_CONST( FIND_PITCH_WHITE_NOISE_FRACTION, 16 ) ) + 1; + + /* Calculate the reflection coefficients using schur */ + res_nrg = silk_schur( rc_Q15, auto_corr, psEnc->sCmn.pitchEstimationLPCOrder ); + + /* Prediction gain */ + psEncCtrl->predGain_Q16 = silk_DIV32_varQ( auto_corr[ 0 ], silk_max_int( res_nrg, 1 ), 16 ); + + /* Convert reflection coefficients to prediction coefficients */ + silk_k2a( A_Q24, rc_Q15, psEnc->sCmn.pitchEstimationLPCOrder ); + + /* Convert From 32 bit Q24 to 16 bit Q12 coefs */ + for( i = 0; i < psEnc->sCmn.pitchEstimationLPCOrder; i++ ) { + A_Q12[ i ] = (opus_int16)silk_SAT16( silk_RSHIFT( A_Q24[ i ], 12 ) ); + } + + /* Do BWE */ + silk_bwexpander( A_Q12, psEnc->sCmn.pitchEstimationLPCOrder, SILK_FIX_CONST( FIND_PITCH_BANDWIDTH_EXPANSION, 16 ) ); + + /*****************************************/ + /* LPC analysis filtering */ + /*****************************************/ + silk_LPC_analysis_filter( res, x, A_Q12, buf_len, psEnc->sCmn.pitchEstimationLPCOrder, psEnc->sCmn.arch ); + + if( psEnc->sCmn.indices.signalType != TYPE_NO_VOICE_ACTIVITY && psEnc->sCmn.first_frame_after_reset == 0 ) { + /* Threshold for pitch estimator */ + thrhld_Q13 = SILK_FIX_CONST( 0.6, 13 ); + thrhld_Q13 = silk_SMLABB( thrhld_Q13, SILK_FIX_CONST( -0.004, 13 ), psEnc->sCmn.pitchEstimationLPCOrder ); + thrhld_Q13 = silk_SMLAWB( thrhld_Q13, SILK_FIX_CONST( -0.1, 21 ), psEnc->sCmn.speech_activity_Q8 ); + thrhld_Q13 = silk_SMLABB( thrhld_Q13, SILK_FIX_CONST( -0.15, 13 ), silk_RSHIFT( psEnc->sCmn.prevSignalType, 1 ) ); + thrhld_Q13 = silk_SMLAWB( thrhld_Q13, SILK_FIX_CONST( -0.1, 14 ), psEnc->sCmn.input_tilt_Q15 ); + thrhld_Q13 = silk_SAT16( thrhld_Q13 ); + + /*****************************************/ + /* Call pitch estimator */ + /*****************************************/ + if( silk_pitch_analysis_core( res, psEncCtrl->pitchL, &psEnc->sCmn.indices.lagIndex, &psEnc->sCmn.indices.contourIndex, + &psEnc->LTPCorr_Q15, psEnc->sCmn.prevLag, psEnc->sCmn.pitchEstimationThreshold_Q16, + (opus_int)thrhld_Q13, psEnc->sCmn.fs_kHz, psEnc->sCmn.pitchEstimationComplexity, psEnc->sCmn.nb_subfr, + psEnc->sCmn.arch) == 0 ) + { + psEnc->sCmn.indices.signalType = TYPE_VOICED; + } else { + psEnc->sCmn.indices.signalType = TYPE_UNVOICED; + } + } else { + silk_memset( psEncCtrl->pitchL, 0, sizeof( psEncCtrl->pitchL ) ); + psEnc->sCmn.indices.lagIndex = 0; + psEnc->sCmn.indices.contourIndex = 0; + psEnc->LTPCorr_Q15 = 0; + } + RESTORE_STACK; +} diff --git a/media/libopus/silk/fixed/find_pred_coefs_FIX.c b/media/libopus/silk/fixed/find_pred_coefs_FIX.c new file mode 100644 index 0000000000..ad363fb7bc --- /dev/null +++ b/media/libopus/silk/fixed/find_pred_coefs_FIX.c @@ -0,0 +1,146 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main_FIX.h" +#include "stack_alloc.h" + +void silk_find_pred_coefs_FIX( + silk_encoder_state_FIX *psEnc, /* I/O encoder state */ + silk_encoder_control_FIX *psEncCtrl, /* I/O encoder control */ + const opus_int16 res_pitch[], /* I Residual from pitch analysis */ + const opus_int16 x[], /* I Speech signal */ + opus_int condCoding /* I The type of conditional coding to use */ +) +{ + opus_int i; + opus_int32 invGains_Q16[ MAX_NB_SUBFR ], local_gains[ MAX_NB_SUBFR ]; + /* Set to NLSF_Q15 to zero so we don't copy junk to the state. */ + opus_int16 NLSF_Q15[ MAX_LPC_ORDER ]={0}; + const opus_int16 *x_ptr; + opus_int16 *x_pre_ptr; + VARDECL( opus_int16, LPC_in_pre ); + opus_int32 min_gain_Q16, minInvGain_Q30; + SAVE_STACK; + + /* weighting for weighted least squares */ + min_gain_Q16 = silk_int32_MAX >> 6; + for( i = 0; i < psEnc->sCmn.nb_subfr; i++ ) { + min_gain_Q16 = silk_min( min_gain_Q16, psEncCtrl->Gains_Q16[ i ] ); + } + for( i = 0; i < psEnc->sCmn.nb_subfr; i++ ) { + /* Divide to Q16 */ + silk_assert( psEncCtrl->Gains_Q16[ i ] > 0 ); + /* Invert and normalize gains, and ensure that maximum invGains_Q16 is within range of a 16 bit int */ + invGains_Q16[ i ] = silk_DIV32_varQ( min_gain_Q16, psEncCtrl->Gains_Q16[ i ], 16 - 2 ); + + /* Limit inverse */ + invGains_Q16[ i ] = silk_max( invGains_Q16[ i ], 100 ); + + /* Square the inverted gains */ + silk_assert( invGains_Q16[ i ] == silk_SAT16( invGains_Q16[ i ] ) ); + + /* Invert the inverted and normalized gains */ + local_gains[ i ] = silk_DIV32( ( (opus_int32)1 << 16 ), invGains_Q16[ i ] ); + } + + ALLOC( LPC_in_pre, + psEnc->sCmn.nb_subfr * psEnc->sCmn.predictLPCOrder + + psEnc->sCmn.frame_length, opus_int16 ); + if( psEnc->sCmn.indices.signalType == TYPE_VOICED ) { + VARDECL( opus_int32, xXLTP_Q17 ); + VARDECL( opus_int32, XXLTP_Q17 ); + + /**********/ + /* VOICED */ + /**********/ + celt_assert( psEnc->sCmn.ltp_mem_length - psEnc->sCmn.predictLPCOrder >= psEncCtrl->pitchL[ 0 ] + LTP_ORDER / 2 ); + + ALLOC( xXLTP_Q17, psEnc->sCmn.nb_subfr * LTP_ORDER, opus_int32 ); + ALLOC( XXLTP_Q17, psEnc->sCmn.nb_subfr * LTP_ORDER * LTP_ORDER, opus_int32 ); + + /* LTP analysis */ + silk_find_LTP_FIX( XXLTP_Q17, xXLTP_Q17, res_pitch, + psEncCtrl->pitchL, psEnc->sCmn.subfr_length, psEnc->sCmn.nb_subfr, psEnc->sCmn.arch ); + + /* Quantize LTP gain parameters */ + silk_quant_LTP_gains( psEncCtrl->LTPCoef_Q14, psEnc->sCmn.indices.LTPIndex, &psEnc->sCmn.indices.PERIndex, + &psEnc->sCmn.sum_log_gain_Q7, &psEncCtrl->LTPredCodGain_Q7, XXLTP_Q17, xXLTP_Q17, psEnc->sCmn.subfr_length, psEnc->sCmn.nb_subfr, psEnc->sCmn.arch ); + + /* Control LTP scaling */ + silk_LTP_scale_ctrl_FIX( psEnc, psEncCtrl, condCoding ); + + /* Create LTP residual */ + silk_LTP_analysis_filter_FIX( LPC_in_pre, x - psEnc->sCmn.predictLPCOrder, psEncCtrl->LTPCoef_Q14, + psEncCtrl->pitchL, invGains_Q16, psEnc->sCmn.subfr_length, psEnc->sCmn.nb_subfr, psEnc->sCmn.predictLPCOrder ); + + } else { + /************/ + /* UNVOICED */ + /************/ + /* Create signal with prepended subframes, scaled by inverse gains */ + x_ptr = x - psEnc->sCmn.predictLPCOrder; + x_pre_ptr = LPC_in_pre; + for( i = 0; i < psEnc->sCmn.nb_subfr; i++ ) { + silk_scale_copy_vector16( x_pre_ptr, x_ptr, invGains_Q16[ i ], + psEnc->sCmn.subfr_length + psEnc->sCmn.predictLPCOrder ); + x_pre_ptr += psEnc->sCmn.subfr_length + psEnc->sCmn.predictLPCOrder; + x_ptr += psEnc->sCmn.subfr_length; + } + + silk_memset( psEncCtrl->LTPCoef_Q14, 0, psEnc->sCmn.nb_subfr * LTP_ORDER * sizeof( opus_int16 ) ); + psEncCtrl->LTPredCodGain_Q7 = 0; + psEnc->sCmn.sum_log_gain_Q7 = 0; + } + + /* Limit on total predictive coding gain */ + if( psEnc->sCmn.first_frame_after_reset ) { + minInvGain_Q30 = SILK_FIX_CONST( 1.0f / MAX_PREDICTION_POWER_GAIN_AFTER_RESET, 30 ); + } else { + minInvGain_Q30 = silk_log2lin( silk_SMLAWB( 16 << 7, (opus_int32)psEncCtrl->LTPredCodGain_Q7, SILK_FIX_CONST( 1.0 / 3, 16 ) ) ); /* Q16 */ + minInvGain_Q30 = silk_DIV32_varQ( minInvGain_Q30, + silk_SMULWW( SILK_FIX_CONST( MAX_PREDICTION_POWER_GAIN, 0 ), + silk_SMLAWB( SILK_FIX_CONST( 0.25, 18 ), SILK_FIX_CONST( 0.75, 18 ), psEncCtrl->coding_quality_Q14 ) ), 14 ); + } + + /* LPC_in_pre contains the LTP-filtered input for voiced, and the unfiltered input for unvoiced */ + silk_find_LPC_FIX( &psEnc->sCmn, NLSF_Q15, LPC_in_pre, minInvGain_Q30 ); + + /* Quantize LSFs */ + silk_process_NLSFs( &psEnc->sCmn, psEncCtrl->PredCoef_Q12, NLSF_Q15, psEnc->sCmn.prev_NLSFq_Q15 ); + + /* Calculate residual energy using quantized LPC coefficients */ + silk_residual_energy_FIX( psEncCtrl->ResNrg, psEncCtrl->ResNrgQ, LPC_in_pre, psEncCtrl->PredCoef_Q12, local_gains, + psEnc->sCmn.subfr_length, psEnc->sCmn.nb_subfr, psEnc->sCmn.predictLPCOrder, psEnc->sCmn.arch ); + + /* Copy to prediction struct for use in next frame for interpolation */ + silk_memcpy( psEnc->sCmn.prev_NLSFq_Q15, NLSF_Q15, sizeof( psEnc->sCmn.prev_NLSFq_Q15 ) ); + RESTORE_STACK; +} diff --git a/media/libopus/silk/fixed/k2a_FIX.c b/media/libopus/silk/fixed/k2a_FIX.c new file mode 100644 index 0000000000..549f6eadaa --- /dev/null +++ b/media/libopus/silk/fixed/k2a_FIX.c @@ -0,0 +1,54 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "SigProc_FIX.h" + +/* Step up function, converts reflection coefficients to prediction coefficients */ +void silk_k2a( + opus_int32 *A_Q24, /* O Prediction coefficients [order] Q24 */ + const opus_int16 *rc_Q15, /* I Reflection coefficients [order] Q15 */ + const opus_int32 order /* I Prediction order */ +) +{ + opus_int k, n; + opus_int32 rc, tmp1, tmp2; + + for( k = 0; k < order; k++ ) { + rc = rc_Q15[ k ]; + for( n = 0; n < (k + 1) >> 1; n++ ) { + tmp1 = A_Q24[ n ]; + tmp2 = A_Q24[ k - n - 1 ]; + A_Q24[ n ] = silk_SMLAWB( tmp1, silk_LSHIFT( tmp2, 1 ), rc ); + A_Q24[ k - n - 1 ] = silk_SMLAWB( tmp2, silk_LSHIFT( tmp1, 1 ), rc ); + } + A_Q24[ k ] = -silk_LSHIFT( (opus_int32)rc_Q15[ k ], 9 ); + } +} diff --git a/media/libopus/silk/fixed/k2a_Q16_FIX.c b/media/libopus/silk/fixed/k2a_Q16_FIX.c new file mode 100644 index 0000000000..1595aa6212 --- /dev/null +++ b/media/libopus/silk/fixed/k2a_Q16_FIX.c @@ -0,0 +1,54 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "SigProc_FIX.h" + +/* Step up function, converts reflection coefficients to prediction coefficients */ +void silk_k2a_Q16( + opus_int32 *A_Q24, /* O Prediction coefficients [order] Q24 */ + const opus_int32 *rc_Q16, /* I Reflection coefficients [order] Q16 */ + const opus_int32 order /* I Prediction order */ +) +{ + opus_int k, n; + opus_int32 rc, tmp1, tmp2; + + for( k = 0; k < order; k++ ) { + rc = rc_Q16[ k ]; + for( n = 0; n < (k + 1) >> 1; n++ ) { + tmp1 = A_Q24[ n ]; + tmp2 = A_Q24[ k - n - 1 ]; + A_Q24[ n ] = silk_SMLAWW( tmp1, tmp2, rc ); + A_Q24[ k - n - 1 ] = silk_SMLAWW( tmp2, tmp1, rc ); + } + A_Q24[ k ] = -silk_LSHIFT( rc, 8 ); + } +} diff --git a/media/libopus/silk/fixed/main_FIX.h b/media/libopus/silk/fixed/main_FIX.h new file mode 100644 index 0000000000..6d2112e511 --- /dev/null +++ b/media/libopus/silk/fixed/main_FIX.h @@ -0,0 +1,244 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef SILK_MAIN_FIX_H +#define SILK_MAIN_FIX_H + +#include "SigProc_FIX.h" +#include "structs_FIX.h" +#include "control.h" +#include "main.h" +#include "PLC.h" +#include "debug.h" +#include "entenc.h" + +#if ((defined(OPUS_ARM_ASM) && defined(FIXED_POINT)) \ + || defined(OPUS_ARM_MAY_HAVE_NEON_INTR)) +#include "fixed/arm/warped_autocorrelation_FIX_arm.h" +#endif + +#ifndef FORCE_CPP_BUILD +#ifdef __cplusplus +extern "C" +{ +#endif +#endif + +#define silk_encoder_state_Fxx silk_encoder_state_FIX +#define silk_encode_do_VAD_Fxx silk_encode_do_VAD_FIX +#define silk_encode_frame_Fxx silk_encode_frame_FIX + +#define QC 10 +#define QS 13 + +/*********************/ +/* Encoder Functions */ +/*********************/ + +/* High-pass filter with cutoff frequency adaptation based on pitch lag statistics */ +void silk_HP_variable_cutoff( + silk_encoder_state_Fxx state_Fxx[] /* I/O Encoder states */ +); + +/* Encoder main function */ +void silk_encode_do_VAD_FIX( + silk_encoder_state_FIX *psEnc, /* I/O Pointer to Silk FIX encoder state */ + opus_int activity /* I Decision of Opus voice activity detector */ +); + +/* Encoder main function */ +opus_int silk_encode_frame_FIX( + silk_encoder_state_FIX *psEnc, /* I/O Pointer to Silk FIX encoder state */ + opus_int32 *pnBytesOut, /* O Pointer to number of payload bytes; */ + ec_enc *psRangeEnc, /* I/O compressor data structure */ + opus_int condCoding, /* I The type of conditional coding to use */ + opus_int maxBits, /* I If > 0: maximum number of output bits */ + opus_int useCBR /* I Flag to force constant-bitrate operation */ +); + +/* Initializes the Silk encoder state */ +opus_int silk_init_encoder( + silk_encoder_state_Fxx *psEnc, /* I/O Pointer to Silk FIX encoder state */ + int arch /* I Run-time architecture */ +); + +/* Control the Silk encoder */ +opus_int silk_control_encoder( + silk_encoder_state_Fxx *psEnc, /* I/O Pointer to Silk encoder state */ + silk_EncControlStruct *encControl, /* I Control structure */ + const opus_int allow_bw_switch, /* I Flag to allow switching audio bandwidth */ + const opus_int channelNb, /* I Channel number */ + const opus_int force_fs_kHz +); + +/**************************/ +/* Noise shaping analysis */ +/**************************/ +/* Compute noise shaping coefficients and initial gain values */ +void silk_noise_shape_analysis_FIX( + silk_encoder_state_FIX *psEnc, /* I/O Encoder state FIX */ + silk_encoder_control_FIX *psEncCtrl, /* I/O Encoder control FIX */ + const opus_int16 *pitch_res, /* I LPC residual from pitch analysis */ + const opus_int16 *x, /* I Input signal [ frame_length + la_shape ] */ + int arch /* I Run-time architecture */ +); + +/* Autocorrelations for a warped frequency axis */ +void silk_warped_autocorrelation_FIX_c( + opus_int32 *corr, /* O Result [order + 1] */ + opus_int *scale, /* O Scaling of the correlation vector */ + const opus_int16 *input, /* I Input data to correlate */ + const opus_int warping_Q16, /* I Warping coefficient */ + const opus_int length, /* I Length of input */ + const opus_int order /* I Correlation order (even) */ +); + +#if !defined(OVERRIDE_silk_warped_autocorrelation_FIX) +#define silk_warped_autocorrelation_FIX(corr, scale, input, warping_Q16, length, order, arch) \ + ((void)(arch), silk_warped_autocorrelation_FIX_c(corr, scale, input, warping_Q16, length, order)) +#endif + +/* Calculation of LTP state scaling */ +void silk_LTP_scale_ctrl_FIX( + silk_encoder_state_FIX *psEnc, /* I/O encoder state */ + silk_encoder_control_FIX *psEncCtrl, /* I/O encoder control */ + opus_int condCoding /* I The type of conditional coding to use */ +); + +/**********************************************/ +/* Prediction Analysis */ +/**********************************************/ +/* Find pitch lags */ +void silk_find_pitch_lags_FIX( + silk_encoder_state_FIX *psEnc, /* I/O encoder state */ + silk_encoder_control_FIX *psEncCtrl, /* I/O encoder control */ + opus_int16 res[], /* O residual */ + const opus_int16 x[], /* I Speech signal */ + int arch /* I Run-time architecture */ +); + +/* Find LPC and LTP coefficients */ +void silk_find_pred_coefs_FIX( + silk_encoder_state_FIX *psEnc, /* I/O encoder state */ + silk_encoder_control_FIX *psEncCtrl, /* I/O encoder control */ + const opus_int16 res_pitch[], /* I Residual from pitch analysis */ + const opus_int16 x[], /* I Speech signal */ + opus_int condCoding /* I The type of conditional coding to use */ +); + +/* LPC analysis */ +void silk_find_LPC_FIX( + silk_encoder_state *psEncC, /* I/O Encoder state */ + opus_int16 NLSF_Q15[], /* O NLSFs */ + const opus_int16 x[], /* I Input signal */ + const opus_int32 minInvGain_Q30 /* I Inverse of max prediction gain */ +); + +/* LTP analysis */ +void silk_find_LTP_FIX( + opus_int32 XXLTP_Q17[ MAX_NB_SUBFR * LTP_ORDER * LTP_ORDER ], /* O Correlation matrix */ + opus_int32 xXLTP_Q17[ MAX_NB_SUBFR * LTP_ORDER ], /* O Correlation vector */ + const opus_int16 r_lpc[], /* I Residual signal after LPC */ + const opus_int lag[ MAX_NB_SUBFR ], /* I LTP lags */ + const opus_int subfr_length, /* I Subframe length */ + const opus_int nb_subfr, /* I Number of subframes */ + int arch /* I Run-time architecture */ +); + +void silk_LTP_analysis_filter_FIX( + opus_int16 *LTP_res, /* O LTP residual signal of length MAX_NB_SUBFR * ( pre_length + subfr_length ) */ + const opus_int16 *x, /* I Pointer to input signal with at least max( pitchL ) preceding samples */ + const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ],/* I LTP_ORDER LTP coefficients for each MAX_NB_SUBFR subframe */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lag, one for each subframe */ + const opus_int32 invGains_Q16[ MAX_NB_SUBFR ], /* I Inverse quantization gains, one for each subframe */ + const opus_int subfr_length, /* I Length of each subframe */ + const opus_int nb_subfr, /* I Number of subframes */ + const opus_int pre_length /* I Length of the preceding samples starting at &x[0] for each subframe */ +); + +/* Calculates residual energies of input subframes where all subframes have LPC_order */ +/* of preceding samples */ +void silk_residual_energy_FIX( + opus_int32 nrgs[ MAX_NB_SUBFR ], /* O Residual energy per subframe */ + opus_int nrgsQ[ MAX_NB_SUBFR ], /* O Q value per subframe */ + const opus_int16 x[], /* I Input signal */ + opus_int16 a_Q12[ 2 ][ MAX_LPC_ORDER ], /* I AR coefs for each frame half */ + const opus_int32 gains[ MAX_NB_SUBFR ], /* I Quantization gains */ + const opus_int subfr_length, /* I Subframe length */ + const opus_int nb_subfr, /* I Number of subframes */ + const opus_int LPC_order, /* I LPC order */ + int arch /* I Run-time architecture */ +); + +/* Residual energy: nrg = wxx - 2 * wXx * c + c' * wXX * c */ +opus_int32 silk_residual_energy16_covar_FIX( + const opus_int16 *c, /* I Prediction vector */ + const opus_int32 *wXX, /* I Correlation matrix */ + const opus_int32 *wXx, /* I Correlation vector */ + opus_int32 wxx, /* I Signal energy */ + opus_int D, /* I Dimension */ + opus_int cQ /* I Q value for c vector 0 - 15 */ +); + +/* Processing of gains */ +void silk_process_gains_FIX( + silk_encoder_state_FIX *psEnc, /* I/O Encoder state */ + silk_encoder_control_FIX *psEncCtrl, /* I/O Encoder control */ + opus_int condCoding /* I The type of conditional coding to use */ +); + +/******************/ +/* Linear Algebra */ +/******************/ +/* Calculates correlation matrix X'*X */ +void silk_corrMatrix_FIX( + const opus_int16 *x, /* I x vector [L + order - 1] used to form data matrix X */ + const opus_int L, /* I Length of vectors */ + const opus_int order, /* I Max lag for correlation */ + opus_int32 *XX, /* O Pointer to X'*X correlation matrix [ order x order ] */ + opus_int32 *nrg, /* O Energy of x vector */ + opus_int *rshifts, /* O Right shifts of correlations */ + int arch /* I Run-time architecture */ +); + +/* Calculates correlation vector X'*t */ +void silk_corrVector_FIX( + const opus_int16 *x, /* I x vector [L + order - 1] used to form data matrix X */ + const opus_int16 *t, /* I Target vector [L] */ + const opus_int L, /* I Length of vectors */ + const opus_int order, /* I Max lag for correlation */ + opus_int32 *Xt, /* O Pointer to X'*t correlation vector [order] */ + const opus_int rshifts, /* I Right shifts of correlations */ + int arch /* I Run-time architecture */ +); + +#ifndef FORCE_CPP_BUILD +#ifdef __cplusplus +} +#endif /* __cplusplus */ +#endif /* FORCE_CPP_BUILD */ +#endif /* SILK_MAIN_FIX_H */ diff --git a/media/libopus/silk/fixed/mips/noise_shape_analysis_FIX_mipsr1.h b/media/libopus/silk/fixed/mips/noise_shape_analysis_FIX_mipsr1.h new file mode 100644 index 0000000000..3999b5bd09 --- /dev/null +++ b/media/libopus/silk/fixed/mips/noise_shape_analysis_FIX_mipsr1.h @@ -0,0 +1,336 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + + +/**************************************************************/ +/* Compute noise shaping coefficients and initial gain values */ +/**************************************************************/ +#define OVERRIDE_silk_noise_shape_analysis_FIX + +void silk_noise_shape_analysis_FIX( + silk_encoder_state_FIX *psEnc, /* I/O Encoder state FIX */ + silk_encoder_control_FIX *psEncCtrl, /* I/O Encoder control FIX */ + const opus_int16 *pitch_res, /* I LPC residual from pitch analysis */ + const opus_int16 *x, /* I Input signal [ frame_length + la_shape ] */ + int arch /* I Run-time architecture */ +) +{ + silk_shape_state_FIX *psShapeSt = &psEnc->sShape; + opus_int k, i, nSamples, Qnrg, b_Q14, warping_Q16, scale = 0; + opus_int32 SNR_adj_dB_Q7, HarmBoost_Q16, HarmShapeGain_Q16, Tilt_Q16, tmp32; + opus_int32 nrg, pre_nrg_Q30, log_energy_Q7, log_energy_prev_Q7, energy_variation_Q7; + opus_int32 delta_Q16, BWExp1_Q16, BWExp2_Q16, gain_mult_Q16, gain_add_Q16, strength_Q16, b_Q8; + opus_int32 auto_corr[ MAX_SHAPE_LPC_ORDER + 1 ]; + opus_int32 refl_coef_Q16[ MAX_SHAPE_LPC_ORDER ]; + opus_int32 AR1_Q24[ MAX_SHAPE_LPC_ORDER ]; + opus_int32 AR2_Q24[ MAX_SHAPE_LPC_ORDER ]; + VARDECL( opus_int16, x_windowed ); + const opus_int16 *x_ptr, *pitch_res_ptr; + SAVE_STACK; + + /* Point to start of first LPC analysis block */ + x_ptr = x - psEnc->sCmn.la_shape; + + /****************/ + /* GAIN CONTROL */ + /****************/ + SNR_adj_dB_Q7 = psEnc->sCmn.SNR_dB_Q7; + + /* Input quality is the average of the quality in the lowest two VAD bands */ + psEncCtrl->input_quality_Q14 = ( opus_int )silk_RSHIFT( (opus_int32)psEnc->sCmn.input_quality_bands_Q15[ 0 ] + + psEnc->sCmn.input_quality_bands_Q15[ 1 ], 2 ); + + /* Coding quality level, between 0.0_Q0 and 1.0_Q0, but in Q14 */ + psEncCtrl->coding_quality_Q14 = silk_RSHIFT( silk_sigm_Q15( silk_RSHIFT_ROUND( SNR_adj_dB_Q7 - + SILK_FIX_CONST( 20.0, 7 ), 4 ) ), 1 ); + + /* Reduce coding SNR during low speech activity */ + if( psEnc->sCmn.useCBR == 0 ) { + b_Q8 = SILK_FIX_CONST( 1.0, 8 ) - psEnc->sCmn.speech_activity_Q8; + b_Q8 = silk_SMULWB( silk_LSHIFT( b_Q8, 8 ), b_Q8 ); + SNR_adj_dB_Q7 = silk_SMLAWB( SNR_adj_dB_Q7, + silk_SMULBB( SILK_FIX_CONST( -BG_SNR_DECR_dB, 7 ) >> ( 4 + 1 ), b_Q8 ), /* Q11*/ + silk_SMULWB( SILK_FIX_CONST( 1.0, 14 ) + psEncCtrl->input_quality_Q14, psEncCtrl->coding_quality_Q14 ) ); /* Q12*/ + } + + if( psEnc->sCmn.indices.signalType == TYPE_VOICED ) { + /* Reduce gains for periodic signals */ + SNR_adj_dB_Q7 = silk_SMLAWB( SNR_adj_dB_Q7, SILK_FIX_CONST( HARM_SNR_INCR_dB, 8 ), psEnc->LTPCorr_Q15 ); + } else { + /* For unvoiced signals and low-quality input, adjust the quality slower than SNR_dB setting */ + SNR_adj_dB_Q7 = silk_SMLAWB( SNR_adj_dB_Q7, + silk_SMLAWB( SILK_FIX_CONST( 6.0, 9 ), -SILK_FIX_CONST( 0.4, 18 ), psEnc->sCmn.SNR_dB_Q7 ), + SILK_FIX_CONST( 1.0, 14 ) - psEncCtrl->input_quality_Q14 ); + } + + /*************************/ + /* SPARSENESS PROCESSING */ + /*************************/ + /* Set quantizer offset */ + if( psEnc->sCmn.indices.signalType == TYPE_VOICED ) { + /* Initially set to 0; may be overruled in process_gains(..) */ + psEnc->sCmn.indices.quantOffsetType = 0; + psEncCtrl->sparseness_Q8 = 0; + } else { + /* Sparseness measure, based on relative fluctuations of energy per 2 milliseconds */ + nSamples = silk_LSHIFT( psEnc->sCmn.fs_kHz, 1 ); + energy_variation_Q7 = 0; + log_energy_prev_Q7 = 0; + pitch_res_ptr = pitch_res; + for( k = 0; k < silk_SMULBB( SUB_FRAME_LENGTH_MS, psEnc->sCmn.nb_subfr ) / 2; k++ ) { + silk_sum_sqr_shift( &nrg, &scale, pitch_res_ptr, nSamples ); + nrg += silk_RSHIFT( nSamples, scale ); /* Q(-scale)*/ + + log_energy_Q7 = silk_lin2log( nrg ); + if( k > 0 ) { + energy_variation_Q7 += silk_abs( log_energy_Q7 - log_energy_prev_Q7 ); + } + log_energy_prev_Q7 = log_energy_Q7; + pitch_res_ptr += nSamples; + } + + psEncCtrl->sparseness_Q8 = silk_RSHIFT( silk_sigm_Q15( silk_SMULWB( energy_variation_Q7 - + SILK_FIX_CONST( 5.0, 7 ), SILK_FIX_CONST( 0.1, 16 ) ) ), 7 ); + + /* Set quantization offset depending on sparseness measure */ + if( psEncCtrl->sparseness_Q8 > SILK_FIX_CONST( SPARSENESS_THRESHOLD_QNT_OFFSET, 8 ) ) { + psEnc->sCmn.indices.quantOffsetType = 0; + } else { + psEnc->sCmn.indices.quantOffsetType = 1; + } + + /* Increase coding SNR for sparse signals */ + SNR_adj_dB_Q7 = silk_SMLAWB( SNR_adj_dB_Q7, SILK_FIX_CONST( SPARSE_SNR_INCR_dB, 15 ), psEncCtrl->sparseness_Q8 - SILK_FIX_CONST( 0.5, 8 ) ); + } + + /*******************************/ + /* Control bandwidth expansion */ + /*******************************/ + /* More BWE for signals with high prediction gain */ + strength_Q16 = silk_SMULWB( psEncCtrl->predGain_Q16, SILK_FIX_CONST( FIND_PITCH_WHITE_NOISE_FRACTION, 16 ) ); + BWExp1_Q16 = BWExp2_Q16 = silk_DIV32_varQ( SILK_FIX_CONST( BANDWIDTH_EXPANSION, 16 ), + silk_SMLAWW( SILK_FIX_CONST( 1.0, 16 ), strength_Q16, strength_Q16 ), 16 ); + delta_Q16 = silk_SMULWB( SILK_FIX_CONST( 1.0, 16 ) - silk_SMULBB( 3, psEncCtrl->coding_quality_Q14 ), + SILK_FIX_CONST( LOW_RATE_BANDWIDTH_EXPANSION_DELTA, 16 ) ); + BWExp1_Q16 = silk_SUB32( BWExp1_Q16, delta_Q16 ); + BWExp2_Q16 = silk_ADD32( BWExp2_Q16, delta_Q16 ); + /* BWExp1 will be applied after BWExp2, so make it relative */ + BWExp1_Q16 = silk_DIV32_16( silk_LSHIFT( BWExp1_Q16, 14 ), silk_RSHIFT( BWExp2_Q16, 2 ) ); + + if( psEnc->sCmn.warping_Q16 > 0 ) { + /* Slightly more warping in analysis will move quantization noise up in frequency, where it's better masked */ + warping_Q16 = silk_SMLAWB( psEnc->sCmn.warping_Q16, (opus_int32)psEncCtrl->coding_quality_Q14, SILK_FIX_CONST( 0.01, 18 ) ); + } else { + warping_Q16 = 0; + } + + /********************************************/ + /* Compute noise shaping AR coefs and gains */ + /********************************************/ + ALLOC( x_windowed, psEnc->sCmn.shapeWinLength, opus_int16 ); + for( k = 0; k < psEnc->sCmn.nb_subfr; k++ ) { + /* Apply window: sine slope followed by flat part followed by cosine slope */ + opus_int shift, slope_part, flat_part; + flat_part = psEnc->sCmn.fs_kHz * 3; + slope_part = silk_RSHIFT( psEnc->sCmn.shapeWinLength - flat_part, 1 ); + + silk_apply_sine_window( x_windowed, x_ptr, 1, slope_part ); + shift = slope_part; + silk_memcpy( x_windowed + shift, x_ptr + shift, flat_part * sizeof(opus_int16) ); + shift += flat_part; + silk_apply_sine_window( x_windowed + shift, x_ptr + shift, 2, slope_part ); + + /* Update pointer: next LPC analysis block */ + x_ptr += psEnc->sCmn.subfr_length; + + if( psEnc->sCmn.warping_Q16 > 0 ) { + /* Calculate warped auto correlation */ + silk_warped_autocorrelation_FIX( auto_corr, &scale, x_windowed, warping_Q16, psEnc->sCmn.shapeWinLength, psEnc->sCmn.shapingLPCOrder, arch ); + } else { + /* Calculate regular auto correlation */ + silk_autocorr( auto_corr, &scale, x_windowed, psEnc->sCmn.shapeWinLength, psEnc->sCmn.shapingLPCOrder + 1, arch ); + } + + /* Add white noise, as a fraction of energy */ + auto_corr[0] = silk_ADD32( auto_corr[0], silk_max_32( silk_SMULWB( silk_RSHIFT( auto_corr[ 0 ], 4 ), + SILK_FIX_CONST( SHAPE_WHITE_NOISE_FRACTION, 20 ) ), 1 ) ); + + /* Calculate the reflection coefficients using schur */ + nrg = silk_schur64( refl_coef_Q16, auto_corr, psEnc->sCmn.shapingLPCOrder ); + silk_assert( nrg >= 0 ); + + /* Convert reflection coefficients to prediction coefficients */ + silk_k2a_Q16( AR2_Q24, refl_coef_Q16, psEnc->sCmn.shapingLPCOrder ); + + Qnrg = -scale; /* range: -12...30*/ + silk_assert( Qnrg >= -12 ); + silk_assert( Qnrg <= 30 ); + + /* Make sure that Qnrg is an even number */ + if( Qnrg & 1 ) { + Qnrg -= 1; + nrg >>= 1; + } + + tmp32 = silk_SQRT_APPROX( nrg ); + Qnrg >>= 1; /* range: -6...15*/ + + psEncCtrl->Gains_Q16[ k ] = (silk_LSHIFT32( silk_LIMIT( (tmp32), silk_RSHIFT32( silk_int32_MIN, (16 - Qnrg) ), \ + silk_RSHIFT32( silk_int32_MAX, (16 - Qnrg) ) ), (16 - Qnrg) )); + + if( psEnc->sCmn.warping_Q16 > 0 ) { + /* Adjust gain for warping */ + gain_mult_Q16 = warped_gain( AR2_Q24, warping_Q16, psEnc->sCmn.shapingLPCOrder ); + silk_assert( psEncCtrl->Gains_Q16[ k ] >= 0 ); + if ( silk_SMULWW( silk_RSHIFT_ROUND( psEncCtrl->Gains_Q16[ k ], 1 ), gain_mult_Q16 ) >= ( silk_int32_MAX >> 1 ) ) { + psEncCtrl->Gains_Q16[ k ] = silk_int32_MAX; + } else { + psEncCtrl->Gains_Q16[ k ] = silk_SMULWW( psEncCtrl->Gains_Q16[ k ], gain_mult_Q16 ); + } + } + + /* Bandwidth expansion for synthesis filter shaping */ + silk_bwexpander_32( AR2_Q24, psEnc->sCmn.shapingLPCOrder, BWExp2_Q16 ); + + /* Compute noise shaping filter coefficients */ + silk_memcpy( AR1_Q24, AR2_Q24, psEnc->sCmn.shapingLPCOrder * sizeof( opus_int32 ) ); + + /* Bandwidth expansion for analysis filter shaping */ + silk_assert( BWExp1_Q16 <= SILK_FIX_CONST( 1.0, 16 ) ); + silk_bwexpander_32( AR1_Q24, psEnc->sCmn.shapingLPCOrder, BWExp1_Q16 ); + + /* Ratio of prediction gains, in energy domain */ + pre_nrg_Q30 = silk_LPC_inverse_pred_gain_Q24( AR2_Q24, psEnc->sCmn.shapingLPCOrder, arch ); + nrg = silk_LPC_inverse_pred_gain_Q24( AR1_Q24, psEnc->sCmn.shapingLPCOrder, arch ); + + /*psEncCtrl->GainsPre[ k ] = 1.0f - 0.7f * ( 1.0f - pre_nrg / nrg ) = 0.3f + 0.7f * pre_nrg / nrg;*/ + pre_nrg_Q30 = silk_LSHIFT32( silk_SMULWB( pre_nrg_Q30, SILK_FIX_CONST( 0.7, 15 ) ), 1 ); + psEncCtrl->GainsPre_Q14[ k ] = ( opus_int ) SILK_FIX_CONST( 0.3, 14 ) + silk_DIV32_varQ( pre_nrg_Q30, nrg, 14 ); + + /* Convert to monic warped prediction coefficients and limit absolute values */ + limit_warped_coefs( AR2_Q24, AR1_Q24, warping_Q16, SILK_FIX_CONST( 3.999, 24 ), psEnc->sCmn.shapingLPCOrder ); + + /* Convert from Q24 to Q13 and store in int16 */ + for( i = 0; i < psEnc->sCmn.shapingLPCOrder; i++ ) { + psEncCtrl->AR1_Q13[ k * MAX_SHAPE_LPC_ORDER + i ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( AR1_Q24[ i ], 11 ) ); + psEncCtrl->AR2_Q13[ k * MAX_SHAPE_LPC_ORDER + i ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( AR2_Q24[ i ], 11 ) ); + } + } + + /*****************/ + /* Gain tweaking */ + /*****************/ + /* Increase gains during low speech activity and put lower limit on gains */ + gain_mult_Q16 = silk_log2lin( -silk_SMLAWB( -SILK_FIX_CONST( 16.0, 7 ), SNR_adj_dB_Q7, SILK_FIX_CONST( 0.16, 16 ) ) ); + gain_add_Q16 = silk_log2lin( silk_SMLAWB( SILK_FIX_CONST( 16.0, 7 ), SILK_FIX_CONST( MIN_QGAIN_DB, 7 ), SILK_FIX_CONST( 0.16, 16 ) ) ); + silk_assert( gain_mult_Q16 > 0 ); + for( k = 0; k < psEnc->sCmn.nb_subfr; k++ ) { + psEncCtrl->Gains_Q16[ k ] = silk_SMULWW( psEncCtrl->Gains_Q16[ k ], gain_mult_Q16 ); + silk_assert( psEncCtrl->Gains_Q16[ k ] >= 0 ); + psEncCtrl->Gains_Q16[ k ] = silk_ADD_POS_SAT32( psEncCtrl->Gains_Q16[ k ], gain_add_Q16 ); + } + + gain_mult_Q16 = SILK_FIX_CONST( 1.0, 16 ) + silk_RSHIFT_ROUND( silk_MLA( SILK_FIX_CONST( INPUT_TILT, 26 ), + psEncCtrl->coding_quality_Q14, SILK_FIX_CONST( HIGH_RATE_INPUT_TILT, 12 ) ), 10 ); + for( k = 0; k < psEnc->sCmn.nb_subfr; k++ ) { + psEncCtrl->GainsPre_Q14[ k ] = silk_SMULWB( gain_mult_Q16, psEncCtrl->GainsPre_Q14[ k ] ); + } + + /************************************************/ + /* Control low-frequency shaping and noise tilt */ + /************************************************/ + /* Less low frequency shaping for noisy inputs */ + strength_Q16 = silk_MUL( SILK_FIX_CONST( LOW_FREQ_SHAPING, 4 ), silk_SMLAWB( SILK_FIX_CONST( 1.0, 12 ), + SILK_FIX_CONST( LOW_QUALITY_LOW_FREQ_SHAPING_DECR, 13 ), psEnc->sCmn.input_quality_bands_Q15[ 0 ] - SILK_FIX_CONST( 1.0, 15 ) ) ); + strength_Q16 = silk_RSHIFT( silk_MUL( strength_Q16, psEnc->sCmn.speech_activity_Q8 ), 8 ); + if( psEnc->sCmn.indices.signalType == TYPE_VOICED ) { + /* Reduce low frequencies quantization noise for periodic signals, depending on pitch lag */ + /*f = 400; freqz([1, -0.98 + 2e-4 * f], [1, -0.97 + 7e-4 * f], 2^12, Fs); axis([0, 1000, -10, 1])*/ + opus_int fs_kHz_inv = silk_DIV32_16( SILK_FIX_CONST( 0.2, 14 ), psEnc->sCmn.fs_kHz ); + for( k = 0; k < psEnc->sCmn.nb_subfr; k++ ) { + b_Q14 = fs_kHz_inv + silk_DIV32_16( SILK_FIX_CONST( 3.0, 14 ), psEncCtrl->pitchL[ k ] ); + /* Pack two coefficients in one int32 */ + psEncCtrl->LF_shp_Q14[ k ] = silk_LSHIFT( SILK_FIX_CONST( 1.0, 14 ) - b_Q14 - silk_SMULWB( strength_Q16, b_Q14 ), 16 ); + psEncCtrl->LF_shp_Q14[ k ] |= (opus_uint16)( b_Q14 - SILK_FIX_CONST( 1.0, 14 ) ); + } + silk_assert( SILK_FIX_CONST( HARM_HP_NOISE_COEF, 24 ) < SILK_FIX_CONST( 0.5, 24 ) ); /* Guarantees that second argument to SMULWB() is within range of an opus_int16*/ + Tilt_Q16 = - SILK_FIX_CONST( HP_NOISE_COEF, 16 ) - + silk_SMULWB( SILK_FIX_CONST( 1.0, 16 ) - SILK_FIX_CONST( HP_NOISE_COEF, 16 ), + silk_SMULWB( SILK_FIX_CONST( HARM_HP_NOISE_COEF, 24 ), psEnc->sCmn.speech_activity_Q8 ) ); + } else { + b_Q14 = silk_DIV32_16( 21299, psEnc->sCmn.fs_kHz ); /* 1.3_Q0 = 21299_Q14*/ + /* Pack two coefficients in one int32 */ + psEncCtrl->LF_shp_Q14[ 0 ] = silk_LSHIFT( SILK_FIX_CONST( 1.0, 14 ) - b_Q14 - + silk_SMULWB( strength_Q16, silk_SMULWB( SILK_FIX_CONST( 0.6, 16 ), b_Q14 ) ), 16 ); + psEncCtrl->LF_shp_Q14[ 0 ] |= (opus_uint16)( b_Q14 - SILK_FIX_CONST( 1.0, 14 ) ); + for( k = 1; k < psEnc->sCmn.nb_subfr; k++ ) { + psEncCtrl->LF_shp_Q14[ k ] = psEncCtrl->LF_shp_Q14[ 0 ]; + } + Tilt_Q16 = -SILK_FIX_CONST( HP_NOISE_COEF, 16 ); + } + + /****************************/ + /* HARMONIC SHAPING CONTROL */ + /****************************/ + /* Control boosting of harmonic frequencies */ + HarmBoost_Q16 = silk_SMULWB( silk_SMULWB( SILK_FIX_CONST( 1.0, 17 ) - silk_LSHIFT( psEncCtrl->coding_quality_Q14, 3 ), + psEnc->LTPCorr_Q15 ), SILK_FIX_CONST( LOW_RATE_HARMONIC_BOOST, 16 ) ); + + /* More harmonic boost for noisy input signals */ + HarmBoost_Q16 = silk_SMLAWB( HarmBoost_Q16, + SILK_FIX_CONST( 1.0, 16 ) - silk_LSHIFT( psEncCtrl->input_quality_Q14, 2 ), SILK_FIX_CONST( LOW_INPUT_QUALITY_HARMONIC_BOOST, 16 ) ); + + if( USE_HARM_SHAPING && psEnc->sCmn.indices.signalType == TYPE_VOICED ) { + /* More harmonic noise shaping for high bitrates or noisy input */ + HarmShapeGain_Q16 = silk_SMLAWB( SILK_FIX_CONST( HARMONIC_SHAPING, 16 ), + SILK_FIX_CONST( 1.0, 16 ) - silk_SMULWB( SILK_FIX_CONST( 1.0, 18 ) - silk_LSHIFT( psEncCtrl->coding_quality_Q14, 4 ), + psEncCtrl->input_quality_Q14 ), SILK_FIX_CONST( HIGH_RATE_OR_LOW_QUALITY_HARMONIC_SHAPING, 16 ) ); + + /* Less harmonic noise shaping for less periodic signals */ + HarmShapeGain_Q16 = silk_SMULWB( silk_LSHIFT( HarmShapeGain_Q16, 1 ), + silk_SQRT_APPROX( silk_LSHIFT( psEnc->LTPCorr_Q15, 15 ) ) ); + } else { + HarmShapeGain_Q16 = 0; + } + + /*************************/ + /* Smooth over subframes */ + /*************************/ + for( k = 0; k < MAX_NB_SUBFR; k++ ) { + psShapeSt->HarmBoost_smth_Q16 = + silk_SMLAWB( psShapeSt->HarmBoost_smth_Q16, HarmBoost_Q16 - psShapeSt->HarmBoost_smth_Q16, SILK_FIX_CONST( SUBFR_SMTH_COEF, 16 ) ); + psShapeSt->HarmShapeGain_smth_Q16 = + silk_SMLAWB( psShapeSt->HarmShapeGain_smth_Q16, HarmShapeGain_Q16 - psShapeSt->HarmShapeGain_smth_Q16, SILK_FIX_CONST( SUBFR_SMTH_COEF, 16 ) ); + psShapeSt->Tilt_smth_Q16 = + silk_SMLAWB( psShapeSt->Tilt_smth_Q16, Tilt_Q16 - psShapeSt->Tilt_smth_Q16, SILK_FIX_CONST( SUBFR_SMTH_COEF, 16 ) ); + + psEncCtrl->HarmBoost_Q14[ k ] = ( opus_int )silk_RSHIFT_ROUND( psShapeSt->HarmBoost_smth_Q16, 2 ); + psEncCtrl->HarmShapeGain_Q14[ k ] = ( opus_int )silk_RSHIFT_ROUND( psShapeSt->HarmShapeGain_smth_Q16, 2 ); + psEncCtrl->Tilt_Q14[ k ] = ( opus_int )silk_RSHIFT_ROUND( psShapeSt->Tilt_smth_Q16, 2 ); + } + RESTORE_STACK; +} diff --git a/media/libopus/silk/fixed/mips/warped_autocorrelation_FIX_mipsr1.h b/media/libopus/silk/fixed/mips/warped_autocorrelation_FIX_mipsr1.h new file mode 100644 index 0000000000..66eb2ed26d --- /dev/null +++ b/media/libopus/silk/fixed/mips/warped_autocorrelation_FIX_mipsr1.h @@ -0,0 +1,165 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef __WARPED_AUTOCORRELATION_FIX_MIPSR1_H__ +#define __WARPED_AUTOCORRELATION_FIX_MIPSR1_H__ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main_FIX.h" + +#undef QC +#define QC 10 + +#undef QS +#define QS 14 + +/* Autocorrelations for a warped frequency axis */ +#define OVERRIDE_silk_warped_autocorrelation_FIX_c +void silk_warped_autocorrelation_FIX_c( + opus_int32 *corr, /* O Result [order + 1] */ + opus_int *scale, /* O Scaling of the correlation vector */ + const opus_int16 *input, /* I Input data to correlate */ + const opus_int warping_Q16, /* I Warping coefficient */ + const opus_int length, /* I Length of input */ + const opus_int order /* I Correlation order (even) */ +) +{ + opus_int n, i, lsh; + opus_int32 tmp1_QS=0, tmp2_QS=0, tmp3_QS=0, tmp4_QS=0, tmp5_QS=0, tmp6_QS=0, tmp7_QS=0, tmp8_QS=0, start_1=0, start_2=0, start_3=0; + opus_int32 state_QS[ MAX_SHAPE_LPC_ORDER + 1 ] = { 0 }; + opus_int64 corr_QC[ MAX_SHAPE_LPC_ORDER + 1 ] = { 0 }; + opus_int64 temp64; + + opus_int32 val; + val = 2 * QS - QC; + + /* Order must be even */ + silk_assert( ( order & 1 ) == 0 ); + silk_assert( 2 * QS - QC >= 0 ); + + /* Loop over samples */ + for( n = 0; n < length; n=n+4 ) { + + tmp1_QS = silk_LSHIFT32( (opus_int32)input[ n ], QS ); + start_1 = tmp1_QS; + tmp3_QS = silk_LSHIFT32( (opus_int32)input[ n+1], QS ); + start_2 = tmp3_QS; + tmp5_QS = silk_LSHIFT32( (opus_int32)input[ n+2], QS ); + start_3 = tmp5_QS; + tmp7_QS = silk_LSHIFT32( (opus_int32)input[ n+3], QS ); + + /* Loop over allpass sections */ + for( i = 0; i < order; i += 2 ) { + /* Output of allpass section */ + tmp2_QS = silk_SMLAWB( state_QS[ i ], state_QS[ i + 1 ] - tmp1_QS, warping_Q16 ); + corr_QC[ i ] = __builtin_mips_madd( corr_QC[ i ], tmp1_QS, start_1); + + tmp4_QS = silk_SMLAWB( tmp1_QS, tmp2_QS - tmp3_QS, warping_Q16 ); + corr_QC[ i ] = __builtin_mips_madd( corr_QC[ i ], tmp3_QS, start_2); + + tmp6_QS = silk_SMLAWB( tmp3_QS, tmp4_QS - tmp5_QS, warping_Q16 ); + corr_QC[ i ] = __builtin_mips_madd( corr_QC[ i ], tmp5_QS, start_3); + + tmp8_QS = silk_SMLAWB( tmp5_QS, tmp6_QS - tmp7_QS, warping_Q16 ); + state_QS[ i ] = tmp7_QS; + corr_QC[ i ] = __builtin_mips_madd( corr_QC[ i ], tmp7_QS, state_QS[0]); + + /* Output of allpass section */ + tmp1_QS = silk_SMLAWB( state_QS[ i + 1 ], state_QS[ i + 2 ] - tmp2_QS, warping_Q16 ); + corr_QC[ i+1 ] = __builtin_mips_madd( corr_QC[ i+1 ], tmp2_QS, start_1); + + tmp3_QS = silk_SMLAWB( tmp2_QS, tmp1_QS - tmp4_QS, warping_Q16 ); + corr_QC[ i+1 ] = __builtin_mips_madd( corr_QC[ i+1 ], tmp4_QS, start_2); + + tmp5_QS = silk_SMLAWB( tmp4_QS, tmp3_QS - tmp6_QS, warping_Q16 ); + corr_QC[ i+1 ] = __builtin_mips_madd( corr_QC[ i+1 ], tmp6_QS, start_3); + + tmp7_QS = silk_SMLAWB( tmp6_QS, tmp5_QS - tmp8_QS, warping_Q16 ); + state_QS[ i + 1 ] = tmp8_QS; + corr_QC[ i+1 ] = __builtin_mips_madd( corr_QC[ i+1 ], tmp8_QS, state_QS[ 0 ]); + + } + state_QS[ order ] = tmp7_QS; + + corr_QC[ order ] = __builtin_mips_madd( corr_QC[ order ], tmp1_QS, start_1); + corr_QC[ order ] = __builtin_mips_madd( corr_QC[ order ], tmp3_QS, start_2); + corr_QC[ order ] = __builtin_mips_madd( corr_QC[ order ], tmp5_QS, start_3); + corr_QC[ order ] = __builtin_mips_madd( corr_QC[ order ], tmp7_QS, state_QS[ 0 ]); + } + + for(;n< length; n++ ) { + + tmp1_QS = silk_LSHIFT32( (opus_int32)input[ n ], QS ); + + /* Loop over allpass sections */ + for( i = 0; i < order; i += 2 ) { + + /* Output of allpass section */ + tmp2_QS = silk_SMLAWB( state_QS[ i ], state_QS[ i + 1 ] - tmp1_QS, warping_Q16 ); + state_QS[ i ] = tmp1_QS; + corr_QC[ i ] = __builtin_mips_madd( corr_QC[ i ], tmp1_QS, state_QS[ 0 ]); + + /* Output of allpass section */ + tmp1_QS = silk_SMLAWB( state_QS[ i + 1 ], state_QS[ i + 2 ] - tmp2_QS, warping_Q16 ); + state_QS[ i + 1 ] = tmp2_QS; + corr_QC[ i+1 ] = __builtin_mips_madd( corr_QC[ i+1 ], tmp2_QS, state_QS[ 0 ]); + } + state_QS[ order ] = tmp1_QS; + corr_QC[ order ] = __builtin_mips_madd( corr_QC[ order ], tmp1_QS, state_QS[ 0 ]); + } + + temp64 = corr_QC[ 0 ]; + temp64 = __builtin_mips_shilo(temp64, val); + + lsh = silk_CLZ64( temp64 ) - 35; + lsh = silk_LIMIT( lsh, -12 - QC, 30 - QC ); + *scale = -( QC + lsh ); + silk_assert( *scale >= -30 && *scale <= 12 ); + if( lsh >= 0 ) { + for( i = 0; i < order + 1; i++ ) { + temp64 = corr_QC[ i ]; + //temp64 = __builtin_mips_shilo(temp64, val); + temp64 = (val >= 0) ? (temp64 >> val) : (temp64 << -val); + corr[ i ] = (opus_int32)silk_CHECK_FIT32( __builtin_mips_shilo( temp64, -lsh ) ); + } + } else { + for( i = 0; i < order + 1; i++ ) { + temp64 = corr_QC[ i ]; + //temp64 = __builtin_mips_shilo(temp64, val); + temp64 = (val >= 0) ? (temp64 >> val) : (temp64 << -val); + corr[ i ] = (opus_int32)silk_CHECK_FIT32( __builtin_mips_shilo( temp64, -lsh ) ); + } + } + + corr_QC[ 0 ] = __builtin_mips_shilo(corr_QC[ 0 ], val); + + silk_assert( corr_QC[ 0 ] >= 0 ); /* If breaking, decrease QC*/ +} +#endif /* __WARPED_AUTOCORRELATION_FIX_MIPSR1_H__ */ diff --git a/media/libopus/silk/fixed/noise_shape_analysis_FIX.c b/media/libopus/silk/fixed/noise_shape_analysis_FIX.c new file mode 100644 index 0000000000..85fea0bf09 --- /dev/null +++ b/media/libopus/silk/fixed/noise_shape_analysis_FIX.c @@ -0,0 +1,407 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main_FIX.h" +#include "stack_alloc.h" +#include "tuning_parameters.h" + +/* Compute gain to make warped filter coefficients have a zero mean log frequency response on a */ +/* non-warped frequency scale. (So that it can be implemented with a minimum-phase monic filter.) */ +/* Note: A monic filter is one with the first coefficient equal to 1.0. In Silk we omit the first */ +/* coefficient in an array of coefficients, for monic filters. */ +static OPUS_INLINE opus_int32 warped_gain( /* gain in Q16*/ + const opus_int32 *coefs_Q24, + opus_int lambda_Q16, + opus_int order +) { + opus_int i; + opus_int32 gain_Q24; + + lambda_Q16 = -lambda_Q16; + gain_Q24 = coefs_Q24[ order - 1 ]; + for( i = order - 2; i >= 0; i-- ) { + gain_Q24 = silk_SMLAWB( coefs_Q24[ i ], gain_Q24, lambda_Q16 ); + } + gain_Q24 = silk_SMLAWB( SILK_FIX_CONST( 1.0, 24 ), gain_Q24, -lambda_Q16 ); + return silk_INVERSE32_varQ( gain_Q24, 40 ); +} + +/* Convert warped filter coefficients to monic pseudo-warped coefficients and limit maximum */ +/* amplitude of monic warped coefficients by using bandwidth expansion on the true coefficients */ +static OPUS_INLINE void limit_warped_coefs( + opus_int32 *coefs_Q24, + opus_int lambda_Q16, + opus_int32 limit_Q24, + opus_int order +) { + opus_int i, iter, ind = 0; + opus_int32 tmp, maxabs_Q24, chirp_Q16, gain_Q16; + opus_int32 nom_Q16, den_Q24; + opus_int32 limit_Q20, maxabs_Q20; + + /* Convert to monic coefficients */ + lambda_Q16 = -lambda_Q16; + for( i = order - 1; i > 0; i-- ) { + coefs_Q24[ i - 1 ] = silk_SMLAWB( coefs_Q24[ i - 1 ], coefs_Q24[ i ], lambda_Q16 ); + } + lambda_Q16 = -lambda_Q16; + nom_Q16 = silk_SMLAWB( SILK_FIX_CONST( 1.0, 16 ), -(opus_int32)lambda_Q16, lambda_Q16 ); + den_Q24 = silk_SMLAWB( SILK_FIX_CONST( 1.0, 24 ), coefs_Q24[ 0 ], lambda_Q16 ); + gain_Q16 = silk_DIV32_varQ( nom_Q16, den_Q24, 24 ); + for( i = 0; i < order; i++ ) { + coefs_Q24[ i ] = silk_SMULWW( gain_Q16, coefs_Q24[ i ] ); + } + limit_Q20 = silk_RSHIFT(limit_Q24, 4); + for( iter = 0; iter < 10; iter++ ) { + /* Find maximum absolute value */ + maxabs_Q24 = -1; + for( i = 0; i < order; i++ ) { + tmp = silk_abs_int32( coefs_Q24[ i ] ); + if( tmp > maxabs_Q24 ) { + maxabs_Q24 = tmp; + ind = i; + } + } + /* Use Q20 to avoid any overflow when multiplying by (ind + 1) later. */ + maxabs_Q20 = silk_RSHIFT(maxabs_Q24, 4); + if( maxabs_Q20 <= limit_Q20 ) { + /* Coefficients are within range - done */ + return; + } + + /* Convert back to true warped coefficients */ + for( i = 1; i < order; i++ ) { + coefs_Q24[ i - 1 ] = silk_SMLAWB( coefs_Q24[ i - 1 ], coefs_Q24[ i ], lambda_Q16 ); + } + gain_Q16 = silk_INVERSE32_varQ( gain_Q16, 32 ); + for( i = 0; i < order; i++ ) { + coefs_Q24[ i ] = silk_SMULWW( gain_Q16, coefs_Q24[ i ] ); + } + + /* Apply bandwidth expansion */ + chirp_Q16 = SILK_FIX_CONST( 0.99, 16 ) - silk_DIV32_varQ( + silk_SMULWB( maxabs_Q20 - limit_Q20, silk_SMLABB( SILK_FIX_CONST( 0.8, 10 ), SILK_FIX_CONST( 0.1, 10 ), iter ) ), + silk_MUL( maxabs_Q20, ind + 1 ), 22 ); + silk_bwexpander_32( coefs_Q24, order, chirp_Q16 ); + + /* Convert to monic warped coefficients */ + lambda_Q16 = -lambda_Q16; + for( i = order - 1; i > 0; i-- ) { + coefs_Q24[ i - 1 ] = silk_SMLAWB( coefs_Q24[ i - 1 ], coefs_Q24[ i ], lambda_Q16 ); + } + lambda_Q16 = -lambda_Q16; + nom_Q16 = silk_SMLAWB( SILK_FIX_CONST( 1.0, 16 ), -(opus_int32)lambda_Q16, lambda_Q16 ); + den_Q24 = silk_SMLAWB( SILK_FIX_CONST( 1.0, 24 ), coefs_Q24[ 0 ], lambda_Q16 ); + gain_Q16 = silk_DIV32_varQ( nom_Q16, den_Q24, 24 ); + for( i = 0; i < order; i++ ) { + coefs_Q24[ i ] = silk_SMULWW( gain_Q16, coefs_Q24[ i ] ); + } + } + silk_assert( 0 ); +} + +/* Disable MIPS version until it's updated. */ +#if 0 && defined(MIPSr1_ASM) +#include "mips/noise_shape_analysis_FIX_mipsr1.h" +#endif + +/**************************************************************/ +/* Compute noise shaping coefficients and initial gain values */ +/**************************************************************/ +#ifndef OVERRIDE_silk_noise_shape_analysis_FIX +void silk_noise_shape_analysis_FIX( + silk_encoder_state_FIX *psEnc, /* I/O Encoder state FIX */ + silk_encoder_control_FIX *psEncCtrl, /* I/O Encoder control FIX */ + const opus_int16 *pitch_res, /* I LPC residual from pitch analysis */ + const opus_int16 *x, /* I Input signal [ frame_length + la_shape ] */ + int arch /* I Run-time architecture */ +) +{ + silk_shape_state_FIX *psShapeSt = &psEnc->sShape; + opus_int k, i, nSamples, nSegs, Qnrg, b_Q14, warping_Q16, scale = 0; + opus_int32 SNR_adj_dB_Q7, HarmShapeGain_Q16, Tilt_Q16, tmp32; + opus_int32 nrg, log_energy_Q7, log_energy_prev_Q7, energy_variation_Q7; + opus_int32 BWExp_Q16, gain_mult_Q16, gain_add_Q16, strength_Q16, b_Q8; + opus_int32 auto_corr[ MAX_SHAPE_LPC_ORDER + 1 ]; + opus_int32 refl_coef_Q16[ MAX_SHAPE_LPC_ORDER ]; + opus_int32 AR_Q24[ MAX_SHAPE_LPC_ORDER ]; + VARDECL( opus_int16, x_windowed ); + const opus_int16 *x_ptr, *pitch_res_ptr; + SAVE_STACK; + + /* Point to start of first LPC analysis block */ + x_ptr = x - psEnc->sCmn.la_shape; + + /****************/ + /* GAIN CONTROL */ + /****************/ + SNR_adj_dB_Q7 = psEnc->sCmn.SNR_dB_Q7; + + /* Input quality is the average of the quality in the lowest two VAD bands */ + psEncCtrl->input_quality_Q14 = ( opus_int )silk_RSHIFT( (opus_int32)psEnc->sCmn.input_quality_bands_Q15[ 0 ] + + psEnc->sCmn.input_quality_bands_Q15[ 1 ], 2 ); + + /* Coding quality level, between 0.0_Q0 and 1.0_Q0, but in Q14 */ + psEncCtrl->coding_quality_Q14 = silk_RSHIFT( silk_sigm_Q15( silk_RSHIFT_ROUND( SNR_adj_dB_Q7 - + SILK_FIX_CONST( 20.0, 7 ), 4 ) ), 1 ); + + /* Reduce coding SNR during low speech activity */ + if( psEnc->sCmn.useCBR == 0 ) { + b_Q8 = SILK_FIX_CONST( 1.0, 8 ) - psEnc->sCmn.speech_activity_Q8; + b_Q8 = silk_SMULWB( silk_LSHIFT( b_Q8, 8 ), b_Q8 ); + SNR_adj_dB_Q7 = silk_SMLAWB( SNR_adj_dB_Q7, + silk_SMULBB( SILK_FIX_CONST( -BG_SNR_DECR_dB, 7 ) >> ( 4 + 1 ), b_Q8 ), /* Q11*/ + silk_SMULWB( SILK_FIX_CONST( 1.0, 14 ) + psEncCtrl->input_quality_Q14, psEncCtrl->coding_quality_Q14 ) ); /* Q12*/ + } + + if( psEnc->sCmn.indices.signalType == TYPE_VOICED ) { + /* Reduce gains for periodic signals */ + SNR_adj_dB_Q7 = silk_SMLAWB( SNR_adj_dB_Q7, SILK_FIX_CONST( HARM_SNR_INCR_dB, 8 ), psEnc->LTPCorr_Q15 ); + } else { + /* For unvoiced signals and low-quality input, adjust the quality slower than SNR_dB setting */ + SNR_adj_dB_Q7 = silk_SMLAWB( SNR_adj_dB_Q7, + silk_SMLAWB( SILK_FIX_CONST( 6.0, 9 ), -SILK_FIX_CONST( 0.4, 18 ), psEnc->sCmn.SNR_dB_Q7 ), + SILK_FIX_CONST( 1.0, 14 ) - psEncCtrl->input_quality_Q14 ); + } + + /*************************/ + /* SPARSENESS PROCESSING */ + /*************************/ + /* Set quantizer offset */ + if( psEnc->sCmn.indices.signalType == TYPE_VOICED ) { + /* Initially set to 0; may be overruled in process_gains(..) */ + psEnc->sCmn.indices.quantOffsetType = 0; + } else { + /* Sparseness measure, based on relative fluctuations of energy per 2 milliseconds */ + nSamples = silk_LSHIFT( psEnc->sCmn.fs_kHz, 1 ); + energy_variation_Q7 = 0; + log_energy_prev_Q7 = 0; + pitch_res_ptr = pitch_res; + nSegs = silk_SMULBB( SUB_FRAME_LENGTH_MS, psEnc->sCmn.nb_subfr ) / 2; + for( k = 0; k < nSegs; k++ ) { + silk_sum_sqr_shift( &nrg, &scale, pitch_res_ptr, nSamples ); + nrg += silk_RSHIFT( nSamples, scale ); /* Q(-scale)*/ + + log_energy_Q7 = silk_lin2log( nrg ); + if( k > 0 ) { + energy_variation_Q7 += silk_abs( log_energy_Q7 - log_energy_prev_Q7 ); + } + log_energy_prev_Q7 = log_energy_Q7; + pitch_res_ptr += nSamples; + } + + /* Set quantization offset depending on sparseness measure */ + if( energy_variation_Q7 > SILK_FIX_CONST( ENERGY_VARIATION_THRESHOLD_QNT_OFFSET, 7 ) * (nSegs-1) ) { + psEnc->sCmn.indices.quantOffsetType = 0; + } else { + psEnc->sCmn.indices.quantOffsetType = 1; + } + } + + /*******************************/ + /* Control bandwidth expansion */ + /*******************************/ + /* More BWE for signals with high prediction gain */ + strength_Q16 = silk_SMULWB( psEncCtrl->predGain_Q16, SILK_FIX_CONST( FIND_PITCH_WHITE_NOISE_FRACTION, 16 ) ); + BWExp_Q16 = silk_DIV32_varQ( SILK_FIX_CONST( BANDWIDTH_EXPANSION, 16 ), + silk_SMLAWW( SILK_FIX_CONST( 1.0, 16 ), strength_Q16, strength_Q16 ), 16 ); + + if( psEnc->sCmn.warping_Q16 > 0 ) { + /* Slightly more warping in analysis will move quantization noise up in frequency, where it's better masked */ + warping_Q16 = silk_SMLAWB( psEnc->sCmn.warping_Q16, (opus_int32)psEncCtrl->coding_quality_Q14, SILK_FIX_CONST( 0.01, 18 ) ); + } else { + warping_Q16 = 0; + } + + /********************************************/ + /* Compute noise shaping AR coefs and gains */ + /********************************************/ + ALLOC( x_windowed, psEnc->sCmn.shapeWinLength, opus_int16 ); + for( k = 0; k < psEnc->sCmn.nb_subfr; k++ ) { + /* Apply window: sine slope followed by flat part followed by cosine slope */ + opus_int shift, slope_part, flat_part; + flat_part = psEnc->sCmn.fs_kHz * 3; + slope_part = silk_RSHIFT( psEnc->sCmn.shapeWinLength - flat_part, 1 ); + + silk_apply_sine_window( x_windowed, x_ptr, 1, slope_part ); + shift = slope_part; + silk_memcpy( x_windowed + shift, x_ptr + shift, flat_part * sizeof(opus_int16) ); + shift += flat_part; + silk_apply_sine_window( x_windowed + shift, x_ptr + shift, 2, slope_part ); + + /* Update pointer: next LPC analysis block */ + x_ptr += psEnc->sCmn.subfr_length; + + if( psEnc->sCmn.warping_Q16 > 0 ) { + /* Calculate warped auto correlation */ + silk_warped_autocorrelation_FIX( auto_corr, &scale, x_windowed, warping_Q16, psEnc->sCmn.shapeWinLength, psEnc->sCmn.shapingLPCOrder, arch ); + } else { + /* Calculate regular auto correlation */ + silk_autocorr( auto_corr, &scale, x_windowed, psEnc->sCmn.shapeWinLength, psEnc->sCmn.shapingLPCOrder + 1, arch ); + } + + /* Add white noise, as a fraction of energy */ + auto_corr[0] = silk_ADD32( auto_corr[0], silk_max_32( silk_SMULWB( silk_RSHIFT( auto_corr[ 0 ], 4 ), + SILK_FIX_CONST( SHAPE_WHITE_NOISE_FRACTION, 20 ) ), 1 ) ); + + /* Calculate the reflection coefficients using schur */ + nrg = silk_schur64( refl_coef_Q16, auto_corr, psEnc->sCmn.shapingLPCOrder ); + silk_assert( nrg >= 0 ); + + /* Convert reflection coefficients to prediction coefficients */ + silk_k2a_Q16( AR_Q24, refl_coef_Q16, psEnc->sCmn.shapingLPCOrder ); + + Qnrg = -scale; /* range: -12...30*/ + silk_assert( Qnrg >= -12 ); + silk_assert( Qnrg <= 30 ); + + /* Make sure that Qnrg is an even number */ + if( Qnrg & 1 ) { + Qnrg -= 1; + nrg >>= 1; + } + + tmp32 = silk_SQRT_APPROX( nrg ); + Qnrg >>= 1; /* range: -6...15*/ + + psEncCtrl->Gains_Q16[ k ] = silk_LSHIFT_SAT32( tmp32, 16 - Qnrg ); + + if( psEnc->sCmn.warping_Q16 > 0 ) { + /* Adjust gain for warping */ + gain_mult_Q16 = warped_gain( AR_Q24, warping_Q16, psEnc->sCmn.shapingLPCOrder ); + silk_assert( psEncCtrl->Gains_Q16[ k ] > 0 ); + if( psEncCtrl->Gains_Q16[ k ] < SILK_FIX_CONST( 0.25, 16 ) ) { + psEncCtrl->Gains_Q16[ k ] = silk_SMULWW( psEncCtrl->Gains_Q16[ k ], gain_mult_Q16 ); + } else { + psEncCtrl->Gains_Q16[ k ] = silk_SMULWW( silk_RSHIFT_ROUND( psEncCtrl->Gains_Q16[ k ], 1 ), gain_mult_Q16 ); + if ( psEncCtrl->Gains_Q16[ k ] >= ( silk_int32_MAX >> 1 ) ) { + psEncCtrl->Gains_Q16[ k ] = silk_int32_MAX; + } else { + psEncCtrl->Gains_Q16[ k ] = silk_LSHIFT32( psEncCtrl->Gains_Q16[ k ], 1 ); + } + } + silk_assert( psEncCtrl->Gains_Q16[ k ] > 0 ); + } + + /* Bandwidth expansion */ + silk_bwexpander_32( AR_Q24, psEnc->sCmn.shapingLPCOrder, BWExp_Q16 ); + + if( psEnc->sCmn.warping_Q16 > 0 ) { + /* Convert to monic warped prediction coefficients and limit absolute values */ + limit_warped_coefs( AR_Q24, warping_Q16, SILK_FIX_CONST( 3.999, 24 ), psEnc->sCmn.shapingLPCOrder ); + + /* Convert from Q24 to Q13 and store in int16 */ + for( i = 0; i < psEnc->sCmn.shapingLPCOrder; i++ ) { + psEncCtrl->AR_Q13[ k * MAX_SHAPE_LPC_ORDER + i ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( AR_Q24[ i ], 11 ) ); + } + } else { + silk_LPC_fit( &psEncCtrl->AR_Q13[ k * MAX_SHAPE_LPC_ORDER ], AR_Q24, 13, 24, psEnc->sCmn.shapingLPCOrder ); + } + } + + /*****************/ + /* Gain tweaking */ + /*****************/ + /* Increase gains during low speech activity and put lower limit on gains */ + gain_mult_Q16 = silk_log2lin( -silk_SMLAWB( -SILK_FIX_CONST( 16.0, 7 ), SNR_adj_dB_Q7, SILK_FIX_CONST( 0.16, 16 ) ) ); + gain_add_Q16 = silk_log2lin( silk_SMLAWB( SILK_FIX_CONST( 16.0, 7 ), SILK_FIX_CONST( MIN_QGAIN_DB, 7 ), SILK_FIX_CONST( 0.16, 16 ) ) ); + silk_assert( gain_mult_Q16 > 0 ); + for( k = 0; k < psEnc->sCmn.nb_subfr; k++ ) { + psEncCtrl->Gains_Q16[ k ] = silk_SMULWW( psEncCtrl->Gains_Q16[ k ], gain_mult_Q16 ); + silk_assert( psEncCtrl->Gains_Q16[ k ] >= 0 ); + psEncCtrl->Gains_Q16[ k ] = silk_ADD_POS_SAT32( psEncCtrl->Gains_Q16[ k ], gain_add_Q16 ); + } + + + /************************************************/ + /* Control low-frequency shaping and noise tilt */ + /************************************************/ + /* Less low frequency shaping for noisy inputs */ + strength_Q16 = silk_MUL( SILK_FIX_CONST( LOW_FREQ_SHAPING, 4 ), silk_SMLAWB( SILK_FIX_CONST( 1.0, 12 ), + SILK_FIX_CONST( LOW_QUALITY_LOW_FREQ_SHAPING_DECR, 13 ), psEnc->sCmn.input_quality_bands_Q15[ 0 ] - SILK_FIX_CONST( 1.0, 15 ) ) ); + strength_Q16 = silk_RSHIFT( silk_MUL( strength_Q16, psEnc->sCmn.speech_activity_Q8 ), 8 ); + if( psEnc->sCmn.indices.signalType == TYPE_VOICED ) { + /* Reduce low frequencies quantization noise for periodic signals, depending on pitch lag */ + /*f = 400; freqz([1, -0.98 + 2e-4 * f], [1, -0.97 + 7e-4 * f], 2^12, Fs); axis([0, 1000, -10, 1])*/ + opus_int fs_kHz_inv = silk_DIV32_16( SILK_FIX_CONST( 0.2, 14 ), psEnc->sCmn.fs_kHz ); + for( k = 0; k < psEnc->sCmn.nb_subfr; k++ ) { + b_Q14 = fs_kHz_inv + silk_DIV32_16( SILK_FIX_CONST( 3.0, 14 ), psEncCtrl->pitchL[ k ] ); + /* Pack two coefficients in one int32 */ + psEncCtrl->LF_shp_Q14[ k ] = silk_LSHIFT( SILK_FIX_CONST( 1.0, 14 ) - b_Q14 - silk_SMULWB( strength_Q16, b_Q14 ), 16 ); + psEncCtrl->LF_shp_Q14[ k ] |= (opus_uint16)( b_Q14 - SILK_FIX_CONST( 1.0, 14 ) ); + } + silk_assert( SILK_FIX_CONST( HARM_HP_NOISE_COEF, 24 ) < SILK_FIX_CONST( 0.5, 24 ) ); /* Guarantees that second argument to SMULWB() is within range of an opus_int16*/ + Tilt_Q16 = - SILK_FIX_CONST( HP_NOISE_COEF, 16 ) - + silk_SMULWB( SILK_FIX_CONST( 1.0, 16 ) - SILK_FIX_CONST( HP_NOISE_COEF, 16 ), + silk_SMULWB( SILK_FIX_CONST( HARM_HP_NOISE_COEF, 24 ), psEnc->sCmn.speech_activity_Q8 ) ); + } else { + b_Q14 = silk_DIV32_16( 21299, psEnc->sCmn.fs_kHz ); /* 1.3_Q0 = 21299_Q14*/ + /* Pack two coefficients in one int32 */ + psEncCtrl->LF_shp_Q14[ 0 ] = silk_LSHIFT( SILK_FIX_CONST( 1.0, 14 ) - b_Q14 - + silk_SMULWB( strength_Q16, silk_SMULWB( SILK_FIX_CONST( 0.6, 16 ), b_Q14 ) ), 16 ); + psEncCtrl->LF_shp_Q14[ 0 ] |= (opus_uint16)( b_Q14 - SILK_FIX_CONST( 1.0, 14 ) ); + for( k = 1; k < psEnc->sCmn.nb_subfr; k++ ) { + psEncCtrl->LF_shp_Q14[ k ] = psEncCtrl->LF_shp_Q14[ 0 ]; + } + Tilt_Q16 = -SILK_FIX_CONST( HP_NOISE_COEF, 16 ); + } + + /****************************/ + /* HARMONIC SHAPING CONTROL */ + /****************************/ + if( USE_HARM_SHAPING && psEnc->sCmn.indices.signalType == TYPE_VOICED ) { + /* More harmonic noise shaping for high bitrates or noisy input */ + HarmShapeGain_Q16 = silk_SMLAWB( SILK_FIX_CONST( HARMONIC_SHAPING, 16 ), + SILK_FIX_CONST( 1.0, 16 ) - silk_SMULWB( SILK_FIX_CONST( 1.0, 18 ) - silk_LSHIFT( psEncCtrl->coding_quality_Q14, 4 ), + psEncCtrl->input_quality_Q14 ), SILK_FIX_CONST( HIGH_RATE_OR_LOW_QUALITY_HARMONIC_SHAPING, 16 ) ); + + /* Less harmonic noise shaping for less periodic signals */ + HarmShapeGain_Q16 = silk_SMULWB( silk_LSHIFT( HarmShapeGain_Q16, 1 ), + silk_SQRT_APPROX( silk_LSHIFT( psEnc->LTPCorr_Q15, 15 ) ) ); + } else { + HarmShapeGain_Q16 = 0; + } + + /*************************/ + /* Smooth over subframes */ + /*************************/ + for( k = 0; k < MAX_NB_SUBFR; k++ ) { + psShapeSt->HarmShapeGain_smth_Q16 = + silk_SMLAWB( psShapeSt->HarmShapeGain_smth_Q16, HarmShapeGain_Q16 - psShapeSt->HarmShapeGain_smth_Q16, SILK_FIX_CONST( SUBFR_SMTH_COEF, 16 ) ); + psShapeSt->Tilt_smth_Q16 = + silk_SMLAWB( psShapeSt->Tilt_smth_Q16, Tilt_Q16 - psShapeSt->Tilt_smth_Q16, SILK_FIX_CONST( SUBFR_SMTH_COEF, 16 ) ); + + psEncCtrl->HarmShapeGain_Q14[ k ] = ( opus_int )silk_RSHIFT_ROUND( psShapeSt->HarmShapeGain_smth_Q16, 2 ); + psEncCtrl->Tilt_Q14[ k ] = ( opus_int )silk_RSHIFT_ROUND( psShapeSt->Tilt_smth_Q16, 2 ); + } + RESTORE_STACK; +} +#endif /* OVERRIDE_silk_noise_shape_analysis_FIX */ diff --git a/media/libopus/silk/fixed/pitch_analysis_core_FIX.c b/media/libopus/silk/fixed/pitch_analysis_core_FIX.c new file mode 100644 index 0000000000..14729046d2 --- /dev/null +++ b/media/libopus/silk/fixed/pitch_analysis_core_FIX.c @@ -0,0 +1,721 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +/*********************************************************** +* Pitch analyser function +********************************************************** */ +#include "SigProc_FIX.h" +#include "pitch_est_defines.h" +#include "stack_alloc.h" +#include "debug.h" +#include "pitch.h" + +#define SCRATCH_SIZE 22 +#define SF_LENGTH_4KHZ ( PE_SUBFR_LENGTH_MS * 4 ) +#define SF_LENGTH_8KHZ ( PE_SUBFR_LENGTH_MS * 8 ) +#define MIN_LAG_4KHZ ( PE_MIN_LAG_MS * 4 ) +#define MIN_LAG_8KHZ ( PE_MIN_LAG_MS * 8 ) +#define MAX_LAG_4KHZ ( PE_MAX_LAG_MS * 4 ) +#define MAX_LAG_8KHZ ( PE_MAX_LAG_MS * 8 - 1 ) +#define CSTRIDE_4KHZ ( MAX_LAG_4KHZ + 1 - MIN_LAG_4KHZ ) +#define CSTRIDE_8KHZ ( MAX_LAG_8KHZ + 3 - ( MIN_LAG_8KHZ - 2 ) ) +#define D_COMP_MIN ( MIN_LAG_8KHZ - 3 ) +#define D_COMP_MAX ( MAX_LAG_8KHZ + 4 ) +#define D_COMP_STRIDE ( D_COMP_MAX - D_COMP_MIN ) + +typedef opus_int32 silk_pe_stage3_vals[ PE_NB_STAGE3_LAGS ]; + +/************************************************************/ +/* Internally used functions */ +/************************************************************/ +static void silk_P_Ana_calc_corr_st3( + silk_pe_stage3_vals cross_corr_st3[], /* O 3 DIM correlation array */ + const opus_int16 frame[], /* I vector to correlate */ + opus_int start_lag, /* I lag offset to search around */ + opus_int sf_length, /* I length of a 5 ms subframe */ + opus_int nb_subfr, /* I number of subframes */ + opus_int complexity, /* I Complexity setting */ + int arch /* I Run-time architecture */ +); + +static void silk_P_Ana_calc_energy_st3( + silk_pe_stage3_vals energies_st3[], /* O 3 DIM energy array */ + const opus_int16 frame[], /* I vector to calc energy in */ + opus_int start_lag, /* I lag offset to search around */ + opus_int sf_length, /* I length of one 5 ms subframe */ + opus_int nb_subfr, /* I number of subframes */ + opus_int complexity, /* I Complexity setting */ + int arch /* I Run-time architecture */ +); + +/*************************************************************/ +/* FIXED POINT CORE PITCH ANALYSIS FUNCTION */ +/*************************************************************/ +opus_int silk_pitch_analysis_core( /* O Voicing estimate: 0 voiced, 1 unvoiced */ + const opus_int16 *frame_unscaled, /* I Signal of length PE_FRAME_LENGTH_MS*Fs_kHz */ + opus_int *pitch_out, /* O 4 pitch lag values */ + opus_int16 *lagIndex, /* O Lag Index */ + opus_int8 *contourIndex, /* O Pitch contour Index */ + opus_int *LTPCorr_Q15, /* I/O Normalized correlation; input: value from previous frame */ + opus_int prevLag, /* I Last lag of previous frame; set to zero is unvoiced */ + const opus_int32 search_thres1_Q16, /* I First stage threshold for lag candidates 0 - 1 */ + const opus_int search_thres2_Q13, /* I Final threshold for lag candidates 0 - 1 */ + const opus_int Fs_kHz, /* I Sample frequency (kHz) */ + const opus_int complexity, /* I Complexity setting, 0-2, where 2 is highest */ + const opus_int nb_subfr, /* I number of 5 ms subframes */ + int arch /* I Run-time architecture */ +) +{ + VARDECL( opus_int16, frame_8kHz_buf ); + VARDECL( opus_int16, frame_4kHz ); + VARDECL( opus_int16, frame_scaled ); + opus_int32 filt_state[ 6 ]; + const opus_int16 *frame, *frame_8kHz; + opus_int i, k, d, j; + VARDECL( opus_int16, C ); + VARDECL( opus_int32, xcorr32 ); + const opus_int16 *target_ptr, *basis_ptr; + opus_int32 cross_corr, normalizer, energy, energy_basis, energy_target; + opus_int d_srch[ PE_D_SRCH_LENGTH ], Cmax, length_d_srch, length_d_comp, shift; + VARDECL( opus_int16, d_comp ); + opus_int32 sum, threshold, lag_counter; + opus_int CBimax, CBimax_new, CBimax_old, lag, start_lag, end_lag, lag_new; + opus_int32 CC[ PE_NB_CBKS_STAGE2_EXT ], CCmax, CCmax_b, CCmax_new_b, CCmax_new; + VARDECL( silk_pe_stage3_vals, energies_st3 ); + VARDECL( silk_pe_stage3_vals, cross_corr_st3 ); + opus_int frame_length, frame_length_8kHz, frame_length_4kHz; + opus_int sf_length; + opus_int min_lag; + opus_int max_lag; + opus_int32 contour_bias_Q15, diff; + opus_int nb_cbk_search, cbk_size; + opus_int32 delta_lag_log2_sqr_Q7, lag_log2_Q7, prevLag_log2_Q7, prev_lag_bias_Q13; + const opus_int8 *Lag_CB_ptr; + SAVE_STACK; + + /* Check for valid sampling frequency */ + celt_assert( Fs_kHz == 8 || Fs_kHz == 12 || Fs_kHz == 16 ); + + /* Check for valid complexity setting */ + celt_assert( complexity >= SILK_PE_MIN_COMPLEX ); + celt_assert( complexity <= SILK_PE_MAX_COMPLEX ); + + silk_assert( search_thres1_Q16 >= 0 && search_thres1_Q16 <= (1<<16) ); + silk_assert( search_thres2_Q13 >= 0 && search_thres2_Q13 <= (1<<13) ); + + /* Set up frame lengths max / min lag for the sampling frequency */ + frame_length = ( PE_LTP_MEM_LENGTH_MS + nb_subfr * PE_SUBFR_LENGTH_MS ) * Fs_kHz; + frame_length_4kHz = ( PE_LTP_MEM_LENGTH_MS + nb_subfr * PE_SUBFR_LENGTH_MS ) * 4; + frame_length_8kHz = ( PE_LTP_MEM_LENGTH_MS + nb_subfr * PE_SUBFR_LENGTH_MS ) * 8; + sf_length = PE_SUBFR_LENGTH_MS * Fs_kHz; + min_lag = PE_MIN_LAG_MS * Fs_kHz; + max_lag = PE_MAX_LAG_MS * Fs_kHz - 1; + + /* Downscale input if necessary */ + silk_sum_sqr_shift( &energy, &shift, frame_unscaled, frame_length ); + shift += 3 - silk_CLZ32( energy ); /* at least two bits headroom */ + ALLOC( frame_scaled, frame_length, opus_int16 ); + if( shift > 0 ) { + shift = silk_RSHIFT( shift + 1, 1 ); + for( i = 0; i < frame_length; i++ ) { + frame_scaled[ i ] = silk_RSHIFT( frame_unscaled[ i ], shift ); + } + frame = frame_scaled; + } else { + frame = frame_unscaled; + } + + ALLOC( frame_8kHz_buf, ( Fs_kHz == 8 ) ? 1 : frame_length_8kHz, opus_int16 ); + /* Resample from input sampled at Fs_kHz to 8 kHz */ + if( Fs_kHz == 16 ) { + silk_memset( filt_state, 0, 2 * sizeof( opus_int32 ) ); + silk_resampler_down2( filt_state, frame_8kHz_buf, frame, frame_length ); + frame_8kHz = frame_8kHz_buf; + } else if( Fs_kHz == 12 ) { + silk_memset( filt_state, 0, 6 * sizeof( opus_int32 ) ); + silk_resampler_down2_3( filt_state, frame_8kHz_buf, frame, frame_length ); + frame_8kHz = frame_8kHz_buf; + } else { + celt_assert( Fs_kHz == 8 ); + frame_8kHz = frame; + } + + /* Decimate again to 4 kHz */ + silk_memset( filt_state, 0, 2 * sizeof( opus_int32 ) );/* Set state to zero */ + ALLOC( frame_4kHz, frame_length_4kHz, opus_int16 ); + silk_resampler_down2( filt_state, frame_4kHz, frame_8kHz, frame_length_8kHz ); + + /* Low-pass filter */ + for( i = frame_length_4kHz - 1; i > 0; i-- ) { + frame_4kHz[ i ] = silk_ADD_SAT16( frame_4kHz[ i ], frame_4kHz[ i - 1 ] ); + } + + + /****************************************************************************** + * FIRST STAGE, operating in 4 khz + ******************************************************************************/ + ALLOC( C, nb_subfr * CSTRIDE_8KHZ, opus_int16 ); + ALLOC( xcorr32, MAX_LAG_4KHZ-MIN_LAG_4KHZ+1, opus_int32 ); + silk_memset( C, 0, (nb_subfr >> 1) * CSTRIDE_4KHZ * sizeof( opus_int16 ) ); + target_ptr = &frame_4kHz[ silk_LSHIFT( SF_LENGTH_4KHZ, 2 ) ]; + for( k = 0; k < nb_subfr >> 1; k++ ) { + /* Check that we are within range of the array */ + celt_assert( target_ptr >= frame_4kHz ); + celt_assert( target_ptr + SF_LENGTH_8KHZ <= frame_4kHz + frame_length_4kHz ); + + basis_ptr = target_ptr - MIN_LAG_4KHZ; + + /* Check that we are within range of the array */ + celt_assert( basis_ptr >= frame_4kHz ); + celt_assert( basis_ptr + SF_LENGTH_8KHZ <= frame_4kHz + frame_length_4kHz ); + + celt_pitch_xcorr( target_ptr, target_ptr - MAX_LAG_4KHZ, xcorr32, SF_LENGTH_8KHZ, MAX_LAG_4KHZ - MIN_LAG_4KHZ + 1, arch ); + + /* Calculate first vector products before loop */ + cross_corr = xcorr32[ MAX_LAG_4KHZ - MIN_LAG_4KHZ ]; + normalizer = silk_inner_prod_aligned( target_ptr, target_ptr, SF_LENGTH_8KHZ, arch ); + normalizer = silk_ADD32( normalizer, silk_inner_prod_aligned( basis_ptr, basis_ptr, SF_LENGTH_8KHZ, arch ) ); + normalizer = silk_ADD32( normalizer, silk_SMULBB( SF_LENGTH_8KHZ, 4000 ) ); + + matrix_ptr( C, k, 0, CSTRIDE_4KHZ ) = + (opus_int16)silk_DIV32_varQ( cross_corr, normalizer, 13 + 1 ); /* Q13 */ + + /* From now on normalizer is computed recursively */ + for( d = MIN_LAG_4KHZ + 1; d <= MAX_LAG_4KHZ; d++ ) { + basis_ptr--; + + /* Check that we are within range of the array */ + silk_assert( basis_ptr >= frame_4kHz ); + silk_assert( basis_ptr + SF_LENGTH_8KHZ <= frame_4kHz + frame_length_4kHz ); + + cross_corr = xcorr32[ MAX_LAG_4KHZ - d ]; + + /* Add contribution of new sample and remove contribution from oldest sample */ + normalizer = silk_ADD32( normalizer, + silk_SMULBB( basis_ptr[ 0 ], basis_ptr[ 0 ] ) - + silk_SMULBB( basis_ptr[ SF_LENGTH_8KHZ ], basis_ptr[ SF_LENGTH_8KHZ ] ) ); + + matrix_ptr( C, k, d - MIN_LAG_4KHZ, CSTRIDE_4KHZ) = + (opus_int16)silk_DIV32_varQ( cross_corr, normalizer, 13 + 1 ); /* Q13 */ + } + /* Update target pointer */ + target_ptr += SF_LENGTH_8KHZ; + } + + /* Combine two subframes into single correlation measure and apply short-lag bias */ + if( nb_subfr == PE_MAX_NB_SUBFR ) { + for( i = MAX_LAG_4KHZ; i >= MIN_LAG_4KHZ; i-- ) { + sum = (opus_int32)matrix_ptr( C, 0, i - MIN_LAG_4KHZ, CSTRIDE_4KHZ ) + + (opus_int32)matrix_ptr( C, 1, i - MIN_LAG_4KHZ, CSTRIDE_4KHZ ); /* Q14 */ + sum = silk_SMLAWB( sum, sum, silk_LSHIFT( -i, 4 ) ); /* Q14 */ + C[ i - MIN_LAG_4KHZ ] = (opus_int16)sum; /* Q14 */ + } + } else { + /* Only short-lag bias */ + for( i = MAX_LAG_4KHZ; i >= MIN_LAG_4KHZ; i-- ) { + sum = silk_LSHIFT( (opus_int32)C[ i - MIN_LAG_4KHZ ], 1 ); /* Q14 */ + sum = silk_SMLAWB( sum, sum, silk_LSHIFT( -i, 4 ) ); /* Q14 */ + C[ i - MIN_LAG_4KHZ ] = (opus_int16)sum; /* Q14 */ + } + } + + /* Sort */ + length_d_srch = silk_ADD_LSHIFT32( 4, complexity, 1 ); + celt_assert( 3 * length_d_srch <= PE_D_SRCH_LENGTH ); + silk_insertion_sort_decreasing_int16( C, d_srch, CSTRIDE_4KHZ, + length_d_srch ); + + /* Escape if correlation is very low already here */ + Cmax = (opus_int)C[ 0 ]; /* Q14 */ + if( Cmax < SILK_FIX_CONST( 0.2, 14 ) ) { + silk_memset( pitch_out, 0, nb_subfr * sizeof( opus_int ) ); + *LTPCorr_Q15 = 0; + *lagIndex = 0; + *contourIndex = 0; + RESTORE_STACK; + return 1; + } + + threshold = silk_SMULWB( search_thres1_Q16, Cmax ); + for( i = 0; i < length_d_srch; i++ ) { + /* Convert to 8 kHz indices for the sorted correlation that exceeds the threshold */ + if( C[ i ] > threshold ) { + d_srch[ i ] = silk_LSHIFT( d_srch[ i ] + MIN_LAG_4KHZ, 1 ); + } else { + length_d_srch = i; + break; + } + } + celt_assert( length_d_srch > 0 ); + + ALLOC( d_comp, D_COMP_STRIDE, opus_int16 ); + for( i = D_COMP_MIN; i < D_COMP_MAX; i++ ) { + d_comp[ i - D_COMP_MIN ] = 0; + } + for( i = 0; i < length_d_srch; i++ ) { + d_comp[ d_srch[ i ] - D_COMP_MIN ] = 1; + } + + /* Convolution */ + for( i = D_COMP_MAX - 1; i >= MIN_LAG_8KHZ; i-- ) { + d_comp[ i - D_COMP_MIN ] += + d_comp[ i - 1 - D_COMP_MIN ] + d_comp[ i - 2 - D_COMP_MIN ]; + } + + length_d_srch = 0; + for( i = MIN_LAG_8KHZ; i < MAX_LAG_8KHZ + 1; i++ ) { + if( d_comp[ i + 1 - D_COMP_MIN ] > 0 ) { + d_srch[ length_d_srch ] = i; + length_d_srch++; + } + } + + /* Convolution */ + for( i = D_COMP_MAX - 1; i >= MIN_LAG_8KHZ; i-- ) { + d_comp[ i - D_COMP_MIN ] += d_comp[ i - 1 - D_COMP_MIN ] + + d_comp[ i - 2 - D_COMP_MIN ] + d_comp[ i - 3 - D_COMP_MIN ]; + } + + length_d_comp = 0; + for( i = MIN_LAG_8KHZ; i < D_COMP_MAX; i++ ) { + if( d_comp[ i - D_COMP_MIN ] > 0 ) { + d_comp[ length_d_comp ] = i - 2; + length_d_comp++; + } + } + + /********************************************************************************** + ** SECOND STAGE, operating at 8 kHz, on lag sections with high correlation + *************************************************************************************/ + + /********************************************************************************* + * Find energy of each subframe projected onto its history, for a range of delays + *********************************************************************************/ + silk_memset( C, 0, nb_subfr * CSTRIDE_8KHZ * sizeof( opus_int16 ) ); + + target_ptr = &frame_8kHz[ PE_LTP_MEM_LENGTH_MS * 8 ]; + for( k = 0; k < nb_subfr; k++ ) { + + /* Check that we are within range of the array */ + celt_assert( target_ptr >= frame_8kHz ); + celt_assert( target_ptr + SF_LENGTH_8KHZ <= frame_8kHz + frame_length_8kHz ); + + energy_target = silk_ADD32( silk_inner_prod_aligned( target_ptr, target_ptr, SF_LENGTH_8KHZ, arch ), 1 ); + for( j = 0; j < length_d_comp; j++ ) { + d = d_comp[ j ]; + basis_ptr = target_ptr - d; + + /* Check that we are within range of the array */ + silk_assert( basis_ptr >= frame_8kHz ); + silk_assert( basis_ptr + SF_LENGTH_8KHZ <= frame_8kHz + frame_length_8kHz ); + + cross_corr = silk_inner_prod_aligned( target_ptr, basis_ptr, SF_LENGTH_8KHZ, arch ); + if( cross_corr > 0 ) { + energy_basis = silk_inner_prod_aligned( basis_ptr, basis_ptr, SF_LENGTH_8KHZ, arch ); + matrix_ptr( C, k, d - ( MIN_LAG_8KHZ - 2 ), CSTRIDE_8KHZ ) = + (opus_int16)silk_DIV32_varQ( cross_corr, + silk_ADD32( energy_target, + energy_basis ), + 13 + 1 ); /* Q13 */ + } else { + matrix_ptr( C, k, d - ( MIN_LAG_8KHZ - 2 ), CSTRIDE_8KHZ ) = 0; + } + } + target_ptr += SF_LENGTH_8KHZ; + } + + /* search over lag range and lags codebook */ + /* scale factor for lag codebook, as a function of center lag */ + + CCmax = silk_int32_MIN; + CCmax_b = silk_int32_MIN; + + CBimax = 0; /* To avoid returning undefined lag values */ + lag = -1; /* To check if lag with strong enough correlation has been found */ + + if( prevLag > 0 ) { + if( Fs_kHz == 12 ) { + prevLag = silk_DIV32_16( silk_LSHIFT( prevLag, 1 ), 3 ); + } else if( Fs_kHz == 16 ) { + prevLag = silk_RSHIFT( prevLag, 1 ); + } + prevLag_log2_Q7 = silk_lin2log( (opus_int32)prevLag ); + } else { + prevLag_log2_Q7 = 0; + } + silk_assert( search_thres2_Q13 == silk_SAT16( search_thres2_Q13 ) ); + /* Set up stage 2 codebook based on number of subframes */ + if( nb_subfr == PE_MAX_NB_SUBFR ) { + cbk_size = PE_NB_CBKS_STAGE2_EXT; + Lag_CB_ptr = &silk_CB_lags_stage2[ 0 ][ 0 ]; + if( Fs_kHz == 8 && complexity > SILK_PE_MIN_COMPLEX ) { + /* If input is 8 khz use a larger codebook here because it is last stage */ + nb_cbk_search = PE_NB_CBKS_STAGE2_EXT; + } else { + nb_cbk_search = PE_NB_CBKS_STAGE2; + } + } else { + cbk_size = PE_NB_CBKS_STAGE2_10MS; + Lag_CB_ptr = &silk_CB_lags_stage2_10_ms[ 0 ][ 0 ]; + nb_cbk_search = PE_NB_CBKS_STAGE2_10MS; + } + + for( k = 0; k < length_d_srch; k++ ) { + d = d_srch[ k ]; + for( j = 0; j < nb_cbk_search; j++ ) { + CC[ j ] = 0; + for( i = 0; i < nb_subfr; i++ ) { + opus_int d_subfr; + /* Try all codebooks */ + d_subfr = d + matrix_ptr( Lag_CB_ptr, i, j, cbk_size ); + CC[ j ] = CC[ j ] + + (opus_int32)matrix_ptr( C, i, + d_subfr - ( MIN_LAG_8KHZ - 2 ), + CSTRIDE_8KHZ ); + } + } + /* Find best codebook */ + CCmax_new = silk_int32_MIN; + CBimax_new = 0; + for( i = 0; i < nb_cbk_search; i++ ) { + if( CC[ i ] > CCmax_new ) { + CCmax_new = CC[ i ]; + CBimax_new = i; + } + } + + /* Bias towards shorter lags */ + lag_log2_Q7 = silk_lin2log( d ); /* Q7 */ + silk_assert( lag_log2_Q7 == silk_SAT16( lag_log2_Q7 ) ); + silk_assert( nb_subfr * SILK_FIX_CONST( PE_SHORTLAG_BIAS, 13 ) == silk_SAT16( nb_subfr * SILK_FIX_CONST( PE_SHORTLAG_BIAS, 13 ) ) ); + CCmax_new_b = CCmax_new - silk_RSHIFT( silk_SMULBB( nb_subfr * SILK_FIX_CONST( PE_SHORTLAG_BIAS, 13 ), lag_log2_Q7 ), 7 ); /* Q13 */ + + /* Bias towards previous lag */ + silk_assert( nb_subfr * SILK_FIX_CONST( PE_PREVLAG_BIAS, 13 ) == silk_SAT16( nb_subfr * SILK_FIX_CONST( PE_PREVLAG_BIAS, 13 ) ) ); + if( prevLag > 0 ) { + delta_lag_log2_sqr_Q7 = lag_log2_Q7 - prevLag_log2_Q7; + silk_assert( delta_lag_log2_sqr_Q7 == silk_SAT16( delta_lag_log2_sqr_Q7 ) ); + delta_lag_log2_sqr_Q7 = silk_RSHIFT( silk_SMULBB( delta_lag_log2_sqr_Q7, delta_lag_log2_sqr_Q7 ), 7 ); + prev_lag_bias_Q13 = silk_RSHIFT( silk_SMULBB( nb_subfr * SILK_FIX_CONST( PE_PREVLAG_BIAS, 13 ), *LTPCorr_Q15 ), 15 ); /* Q13 */ + prev_lag_bias_Q13 = silk_DIV32( silk_MUL( prev_lag_bias_Q13, delta_lag_log2_sqr_Q7 ), delta_lag_log2_sqr_Q7 + SILK_FIX_CONST( 0.5, 7 ) ); + CCmax_new_b -= prev_lag_bias_Q13; /* Q13 */ + } + + if( CCmax_new_b > CCmax_b && /* Find maximum biased correlation */ + CCmax_new > silk_SMULBB( nb_subfr, search_thres2_Q13 ) && /* Correlation needs to be high enough to be voiced */ + silk_CB_lags_stage2[ 0 ][ CBimax_new ] <= MIN_LAG_8KHZ /* Lag must be in range */ + ) { + CCmax_b = CCmax_new_b; + CCmax = CCmax_new; + lag = d; + CBimax = CBimax_new; + } + } + + if( lag == -1 ) { + /* No suitable candidate found */ + silk_memset( pitch_out, 0, nb_subfr * sizeof( opus_int ) ); + *LTPCorr_Q15 = 0; + *lagIndex = 0; + *contourIndex = 0; + RESTORE_STACK; + return 1; + } + + /* Output normalized correlation */ + *LTPCorr_Q15 = (opus_int)silk_LSHIFT( silk_DIV32_16( CCmax, nb_subfr ), 2 ); + silk_assert( *LTPCorr_Q15 >= 0 ); + + if( Fs_kHz > 8 ) { + /* Search in original signal */ + + CBimax_old = CBimax; + /* Compensate for decimation */ + silk_assert( lag == silk_SAT16( lag ) ); + if( Fs_kHz == 12 ) { + lag = silk_RSHIFT( silk_SMULBB( lag, 3 ), 1 ); + } else if( Fs_kHz == 16 ) { + lag = silk_LSHIFT( lag, 1 ); + } else { + lag = silk_SMULBB( lag, 3 ); + } + + lag = silk_LIMIT_int( lag, min_lag, max_lag ); + start_lag = silk_max_int( lag - 2, min_lag ); + end_lag = silk_min_int( lag + 2, max_lag ); + lag_new = lag; /* to avoid undefined lag */ + CBimax = 0; /* to avoid undefined lag */ + + CCmax = silk_int32_MIN; + /* pitch lags according to second stage */ + for( k = 0; k < nb_subfr; k++ ) { + pitch_out[ k ] = lag + 2 * silk_CB_lags_stage2[ k ][ CBimax_old ]; + } + + /* Set up codebook parameters according to complexity setting and frame length */ + if( nb_subfr == PE_MAX_NB_SUBFR ) { + nb_cbk_search = (opus_int)silk_nb_cbk_searchs_stage3[ complexity ]; + cbk_size = PE_NB_CBKS_STAGE3_MAX; + Lag_CB_ptr = &silk_CB_lags_stage3[ 0 ][ 0 ]; + } else { + nb_cbk_search = PE_NB_CBKS_STAGE3_10MS; + cbk_size = PE_NB_CBKS_STAGE3_10MS; + Lag_CB_ptr = &silk_CB_lags_stage3_10_ms[ 0 ][ 0 ]; + } + + /* Calculate the correlations and energies needed in stage 3 */ + ALLOC( energies_st3, nb_subfr * nb_cbk_search, silk_pe_stage3_vals ); + ALLOC( cross_corr_st3, nb_subfr * nb_cbk_search, silk_pe_stage3_vals ); + silk_P_Ana_calc_corr_st3( cross_corr_st3, frame, start_lag, sf_length, nb_subfr, complexity, arch ); + silk_P_Ana_calc_energy_st3( energies_st3, frame, start_lag, sf_length, nb_subfr, complexity, arch ); + + lag_counter = 0; + silk_assert( lag == silk_SAT16( lag ) ); + contour_bias_Q15 = silk_DIV32_16( SILK_FIX_CONST( PE_FLATCONTOUR_BIAS, 15 ), lag ); + + target_ptr = &frame[ PE_LTP_MEM_LENGTH_MS * Fs_kHz ]; + energy_target = silk_ADD32( silk_inner_prod_aligned( target_ptr, target_ptr, nb_subfr * sf_length, arch ), 1 ); + for( d = start_lag; d <= end_lag; d++ ) { + for( j = 0; j < nb_cbk_search; j++ ) { + cross_corr = 0; + energy = energy_target; + for( k = 0; k < nb_subfr; k++ ) { + cross_corr = silk_ADD32( cross_corr, + matrix_ptr( cross_corr_st3, k, j, + nb_cbk_search )[ lag_counter ] ); + energy = silk_ADD32( energy, + matrix_ptr( energies_st3, k, j, + nb_cbk_search )[ lag_counter ] ); + silk_assert( energy >= 0 ); + } + if( cross_corr > 0 ) { + CCmax_new = silk_DIV32_varQ( cross_corr, energy, 13 + 1 ); /* Q13 */ + /* Reduce depending on flatness of contour */ + diff = silk_int16_MAX - silk_MUL( contour_bias_Q15, j ); /* Q15 */ + silk_assert( diff == silk_SAT16( diff ) ); + CCmax_new = silk_SMULWB( CCmax_new, diff ); /* Q14 */ + } else { + CCmax_new = 0; + } + + if( CCmax_new > CCmax && ( d + silk_CB_lags_stage3[ 0 ][ j ] ) <= max_lag ) { + CCmax = CCmax_new; + lag_new = d; + CBimax = j; + } + } + lag_counter++; + } + + for( k = 0; k < nb_subfr; k++ ) { + pitch_out[ k ] = lag_new + matrix_ptr( Lag_CB_ptr, k, CBimax, cbk_size ); + pitch_out[ k ] = silk_LIMIT( pitch_out[ k ], min_lag, PE_MAX_LAG_MS * Fs_kHz ); + } + *lagIndex = (opus_int16)( lag_new - min_lag); + *contourIndex = (opus_int8)CBimax; + } else { /* Fs_kHz == 8 */ + /* Save Lags */ + for( k = 0; k < nb_subfr; k++ ) { + pitch_out[ k ] = lag + matrix_ptr( Lag_CB_ptr, k, CBimax, cbk_size ); + pitch_out[ k ] = silk_LIMIT( pitch_out[ k ], MIN_LAG_8KHZ, PE_MAX_LAG_MS * 8 ); + } + *lagIndex = (opus_int16)( lag - MIN_LAG_8KHZ ); + *contourIndex = (opus_int8)CBimax; + } + celt_assert( *lagIndex >= 0 ); + /* return as voiced */ + RESTORE_STACK; + return 0; +} + +/*********************************************************************** + * Calculates the correlations used in stage 3 search. In order to cover + * the whole lag codebook for all the searched offset lags (lag +- 2), + * the following correlations are needed in each sub frame: + * + * sf1: lag range [-8,...,7] total 16 correlations + * sf2: lag range [-4,...,4] total 9 correlations + * sf3: lag range [-3,....4] total 8 correltions + * sf4: lag range [-6,....8] total 15 correlations + * + * In total 48 correlations. The direct implementation computed in worst + * case 4*12*5 = 240 correlations, but more likely around 120. + ***********************************************************************/ +static void silk_P_Ana_calc_corr_st3( + silk_pe_stage3_vals cross_corr_st3[], /* O 3 DIM correlation array */ + const opus_int16 frame[], /* I vector to correlate */ + opus_int start_lag, /* I lag offset to search around */ + opus_int sf_length, /* I length of a 5 ms subframe */ + opus_int nb_subfr, /* I number of subframes */ + opus_int complexity, /* I Complexity setting */ + int arch /* I Run-time architecture */ +) +{ + const opus_int16 *target_ptr; + opus_int i, j, k, lag_counter, lag_low, lag_high; + opus_int nb_cbk_search, delta, idx, cbk_size; + VARDECL( opus_int32, scratch_mem ); + VARDECL( opus_int32, xcorr32 ); + const opus_int8 *Lag_range_ptr, *Lag_CB_ptr; + SAVE_STACK; + + celt_assert( complexity >= SILK_PE_MIN_COMPLEX ); + celt_assert( complexity <= SILK_PE_MAX_COMPLEX ); + + if( nb_subfr == PE_MAX_NB_SUBFR ) { + Lag_range_ptr = &silk_Lag_range_stage3[ complexity ][ 0 ][ 0 ]; + Lag_CB_ptr = &silk_CB_lags_stage3[ 0 ][ 0 ]; + nb_cbk_search = silk_nb_cbk_searchs_stage3[ complexity ]; + cbk_size = PE_NB_CBKS_STAGE3_MAX; + } else { + celt_assert( nb_subfr == PE_MAX_NB_SUBFR >> 1); + Lag_range_ptr = &silk_Lag_range_stage3_10_ms[ 0 ][ 0 ]; + Lag_CB_ptr = &silk_CB_lags_stage3_10_ms[ 0 ][ 0 ]; + nb_cbk_search = PE_NB_CBKS_STAGE3_10MS; + cbk_size = PE_NB_CBKS_STAGE3_10MS; + } + ALLOC( scratch_mem, SCRATCH_SIZE, opus_int32 ); + ALLOC( xcorr32, SCRATCH_SIZE, opus_int32 ); + + target_ptr = &frame[ silk_LSHIFT( sf_length, 2 ) ]; /* Pointer to middle of frame */ + for( k = 0; k < nb_subfr; k++ ) { + lag_counter = 0; + + /* Calculate the correlations for each subframe */ + lag_low = matrix_ptr( Lag_range_ptr, k, 0, 2 ); + lag_high = matrix_ptr( Lag_range_ptr, k, 1, 2 ); + celt_assert(lag_high-lag_low+1 <= SCRATCH_SIZE); + celt_pitch_xcorr( target_ptr, target_ptr - start_lag - lag_high, xcorr32, sf_length, lag_high - lag_low + 1, arch ); + for( j = lag_low; j <= lag_high; j++ ) { + silk_assert( lag_counter < SCRATCH_SIZE ); + scratch_mem[ lag_counter ] = xcorr32[ lag_high - j ]; + lag_counter++; + } + + delta = matrix_ptr( Lag_range_ptr, k, 0, 2 ); + for( i = 0; i < nb_cbk_search; i++ ) { + /* Fill out the 3 dim array that stores the correlations for */ + /* each code_book vector for each start lag */ + idx = matrix_ptr( Lag_CB_ptr, k, i, cbk_size ) - delta; + for( j = 0; j < PE_NB_STAGE3_LAGS; j++ ) { + silk_assert( idx + j < SCRATCH_SIZE ); + silk_assert( idx + j < lag_counter ); + matrix_ptr( cross_corr_st3, k, i, nb_cbk_search )[ j ] = + scratch_mem[ idx + j ]; + } + } + target_ptr += sf_length; + } + RESTORE_STACK; +} + +/********************************************************************/ +/* Calculate the energies for first two subframes. The energies are */ +/* calculated recursively. */ +/********************************************************************/ +static void silk_P_Ana_calc_energy_st3( + silk_pe_stage3_vals energies_st3[], /* O 3 DIM energy array */ + const opus_int16 frame[], /* I vector to calc energy in */ + opus_int start_lag, /* I lag offset to search around */ + opus_int sf_length, /* I length of one 5 ms subframe */ + opus_int nb_subfr, /* I number of subframes */ + opus_int complexity, /* I Complexity setting */ + int arch /* I Run-time architecture */ +) +{ + const opus_int16 *target_ptr, *basis_ptr; + opus_int32 energy; + opus_int k, i, j, lag_counter; + opus_int nb_cbk_search, delta, idx, cbk_size, lag_diff; + VARDECL( opus_int32, scratch_mem ); + const opus_int8 *Lag_range_ptr, *Lag_CB_ptr; + SAVE_STACK; + + celt_assert( complexity >= SILK_PE_MIN_COMPLEX ); + celt_assert( complexity <= SILK_PE_MAX_COMPLEX ); + + if( nb_subfr == PE_MAX_NB_SUBFR ) { + Lag_range_ptr = &silk_Lag_range_stage3[ complexity ][ 0 ][ 0 ]; + Lag_CB_ptr = &silk_CB_lags_stage3[ 0 ][ 0 ]; + nb_cbk_search = silk_nb_cbk_searchs_stage3[ complexity ]; + cbk_size = PE_NB_CBKS_STAGE3_MAX; + } else { + celt_assert( nb_subfr == PE_MAX_NB_SUBFR >> 1); + Lag_range_ptr = &silk_Lag_range_stage3_10_ms[ 0 ][ 0 ]; + Lag_CB_ptr = &silk_CB_lags_stage3_10_ms[ 0 ][ 0 ]; + nb_cbk_search = PE_NB_CBKS_STAGE3_10MS; + cbk_size = PE_NB_CBKS_STAGE3_10MS; + } + ALLOC( scratch_mem, SCRATCH_SIZE, opus_int32 ); + + target_ptr = &frame[ silk_LSHIFT( sf_length, 2 ) ]; + for( k = 0; k < nb_subfr; k++ ) { + lag_counter = 0; + + /* Calculate the energy for first lag */ + basis_ptr = target_ptr - ( start_lag + matrix_ptr( Lag_range_ptr, k, 0, 2 ) ); + energy = silk_inner_prod_aligned( basis_ptr, basis_ptr, sf_length, arch ); + silk_assert( energy >= 0 ); + scratch_mem[ lag_counter ] = energy; + lag_counter++; + + lag_diff = ( matrix_ptr( Lag_range_ptr, k, 1, 2 ) - matrix_ptr( Lag_range_ptr, k, 0, 2 ) + 1 ); + for( i = 1; i < lag_diff; i++ ) { + /* remove part outside new window */ + energy -= silk_SMULBB( basis_ptr[ sf_length - i ], basis_ptr[ sf_length - i ] ); + silk_assert( energy >= 0 ); + + /* add part that comes into window */ + energy = silk_ADD_SAT32( energy, silk_SMULBB( basis_ptr[ -i ], basis_ptr[ -i ] ) ); + silk_assert( energy >= 0 ); + silk_assert( lag_counter < SCRATCH_SIZE ); + scratch_mem[ lag_counter ] = energy; + lag_counter++; + } + + delta = matrix_ptr( Lag_range_ptr, k, 0, 2 ); + for( i = 0; i < nb_cbk_search; i++ ) { + /* Fill out the 3 dim array that stores the correlations for */ + /* each code_book vector for each start lag */ + idx = matrix_ptr( Lag_CB_ptr, k, i, cbk_size ) - delta; + for( j = 0; j < PE_NB_STAGE3_LAGS; j++ ) { + silk_assert( idx + j < SCRATCH_SIZE ); + silk_assert( idx + j < lag_counter ); + matrix_ptr( energies_st3, k, i, nb_cbk_search )[ j ] = + scratch_mem[ idx + j ]; + silk_assert( + matrix_ptr( energies_st3, k, i, nb_cbk_search )[ j ] >= 0 ); + } + } + target_ptr += sf_length; + } + RESTORE_STACK; +} diff --git a/media/libopus/silk/fixed/process_gains_FIX.c b/media/libopus/silk/fixed/process_gains_FIX.c new file mode 100644 index 0000000000..05aba31788 --- /dev/null +++ b/media/libopus/silk/fixed/process_gains_FIX.c @@ -0,0 +1,117 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main_FIX.h" +#include "tuning_parameters.h" + +/* Processing of gains */ +void silk_process_gains_FIX( + silk_encoder_state_FIX *psEnc, /* I/O Encoder state */ + silk_encoder_control_FIX *psEncCtrl, /* I/O Encoder control */ + opus_int condCoding /* I The type of conditional coding to use */ +) +{ + silk_shape_state_FIX *psShapeSt = &psEnc->sShape; + opus_int k; + opus_int32 s_Q16, InvMaxSqrVal_Q16, gain, gain_squared, ResNrg, ResNrgPart, quant_offset_Q10; + + /* Gain reduction when LTP coding gain is high */ + if( psEnc->sCmn.indices.signalType == TYPE_VOICED ) { + /*s = -0.5f * silk_sigmoid( 0.25f * ( psEncCtrl->LTPredCodGain - 12.0f ) ); */ + s_Q16 = -silk_sigm_Q15( silk_RSHIFT_ROUND( psEncCtrl->LTPredCodGain_Q7 - SILK_FIX_CONST( 12.0, 7 ), 4 ) ); + for( k = 0; k < psEnc->sCmn.nb_subfr; k++ ) { + psEncCtrl->Gains_Q16[ k ] = silk_SMLAWB( psEncCtrl->Gains_Q16[ k ], psEncCtrl->Gains_Q16[ k ], s_Q16 ); + } + } + + /* Limit the quantized signal */ + /* InvMaxSqrVal = pow( 2.0f, 0.33f * ( 21.0f - SNR_dB ) ) / subfr_length; */ + InvMaxSqrVal_Q16 = silk_DIV32_16( silk_log2lin( + silk_SMULWB( SILK_FIX_CONST( 21 + 16 / 0.33, 7 ) - psEnc->sCmn.SNR_dB_Q7, SILK_FIX_CONST( 0.33, 16 ) ) ), psEnc->sCmn.subfr_length ); + + for( k = 0; k < psEnc->sCmn.nb_subfr; k++ ) { + /* Soft limit on ratio residual energy and squared gains */ + ResNrg = psEncCtrl->ResNrg[ k ]; + ResNrgPart = silk_SMULWW( ResNrg, InvMaxSqrVal_Q16 ); + if( psEncCtrl->ResNrgQ[ k ] > 0 ) { + ResNrgPart = silk_RSHIFT_ROUND( ResNrgPart, psEncCtrl->ResNrgQ[ k ] ); + } else { + if( ResNrgPart >= silk_RSHIFT( silk_int32_MAX, -psEncCtrl->ResNrgQ[ k ] ) ) { + ResNrgPart = silk_int32_MAX; + } else { + ResNrgPart = silk_LSHIFT( ResNrgPart, -psEncCtrl->ResNrgQ[ k ] ); + } + } + gain = psEncCtrl->Gains_Q16[ k ]; + gain_squared = silk_ADD_SAT32( ResNrgPart, silk_SMMUL( gain, gain ) ); + if( gain_squared < silk_int16_MAX ) { + /* recalculate with higher precision */ + gain_squared = silk_SMLAWW( silk_LSHIFT( ResNrgPart, 16 ), gain, gain ); + silk_assert( gain_squared > 0 ); + gain = silk_SQRT_APPROX( gain_squared ); /* Q8 */ + gain = silk_min( gain, silk_int32_MAX >> 8 ); + psEncCtrl->Gains_Q16[ k ] = silk_LSHIFT_SAT32( gain, 8 ); /* Q16 */ + } else { + gain = silk_SQRT_APPROX( gain_squared ); /* Q0 */ + gain = silk_min( gain, silk_int32_MAX >> 16 ); + psEncCtrl->Gains_Q16[ k ] = silk_LSHIFT_SAT32( gain, 16 ); /* Q16 */ + } + } + + /* Save unquantized gains and gain Index */ + silk_memcpy( psEncCtrl->GainsUnq_Q16, psEncCtrl->Gains_Q16, psEnc->sCmn.nb_subfr * sizeof( opus_int32 ) ); + psEncCtrl->lastGainIndexPrev = psShapeSt->LastGainIndex; + + /* Quantize gains */ + silk_gains_quant( psEnc->sCmn.indices.GainsIndices, psEncCtrl->Gains_Q16, + &psShapeSt->LastGainIndex, condCoding == CODE_CONDITIONALLY, psEnc->sCmn.nb_subfr ); + + /* Set quantizer offset for voiced signals. Larger offset when LTP coding gain is low or tilt is high (ie low-pass) */ + if( psEnc->sCmn.indices.signalType == TYPE_VOICED ) { + if( psEncCtrl->LTPredCodGain_Q7 + silk_RSHIFT( psEnc->sCmn.input_tilt_Q15, 8 ) > SILK_FIX_CONST( 1.0, 7 ) ) { + psEnc->sCmn.indices.quantOffsetType = 0; + } else { + psEnc->sCmn.indices.quantOffsetType = 1; + } + } + + /* Quantizer boundary adjustment */ + quant_offset_Q10 = silk_Quantization_Offsets_Q10[ psEnc->sCmn.indices.signalType >> 1 ][ psEnc->sCmn.indices.quantOffsetType ]; + psEncCtrl->Lambda_Q10 = SILK_FIX_CONST( LAMBDA_OFFSET, 10 ) + + silk_SMULBB( SILK_FIX_CONST( LAMBDA_DELAYED_DECISIONS, 10 ), psEnc->sCmn.nStatesDelayedDecision ) + + silk_SMULWB( SILK_FIX_CONST( LAMBDA_SPEECH_ACT, 18 ), psEnc->sCmn.speech_activity_Q8 ) + + silk_SMULWB( SILK_FIX_CONST( LAMBDA_INPUT_QUALITY, 12 ), psEncCtrl->input_quality_Q14 ) + + silk_SMULWB( SILK_FIX_CONST( LAMBDA_CODING_QUALITY, 12 ), psEncCtrl->coding_quality_Q14 ) + + silk_SMULWB( SILK_FIX_CONST( LAMBDA_QUANT_OFFSET, 16 ), quant_offset_Q10 ); + + silk_assert( psEncCtrl->Lambda_Q10 > 0 ); + silk_assert( psEncCtrl->Lambda_Q10 < SILK_FIX_CONST( 2, 10 ) ); +} diff --git a/media/libopus/silk/fixed/regularize_correlations_FIX.c b/media/libopus/silk/fixed/regularize_correlations_FIX.c new file mode 100644 index 0000000000..a2836b05f4 --- /dev/null +++ b/media/libopus/silk/fixed/regularize_correlations_FIX.c @@ -0,0 +1,47 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main_FIX.h" + +/* Add noise to matrix diagonal */ +void silk_regularize_correlations_FIX( + opus_int32 *XX, /* I/O Correlation matrices */ + opus_int32 *xx, /* I/O Correlation values */ + opus_int32 noise, /* I Noise to add */ + opus_int D /* I Dimension of XX */ +) +{ + opus_int i; + for( i = 0; i < D; i++ ) { + matrix_ptr( &XX[ 0 ], i, i, D ) = silk_ADD32( matrix_ptr( &XX[ 0 ], i, i, D ), noise ); + } + xx[ 0 ] += noise; +} diff --git a/media/libopus/silk/fixed/residual_energy16_FIX.c b/media/libopus/silk/fixed/residual_energy16_FIX.c new file mode 100644 index 0000000000..7f130f3d3d --- /dev/null +++ b/media/libopus/silk/fixed/residual_energy16_FIX.c @@ -0,0 +1,103 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main_FIX.h" + +/* Residual energy: nrg = wxx - 2 * wXx * c + c' * wXX * c */ +opus_int32 silk_residual_energy16_covar_FIX( + const opus_int16 *c, /* I Prediction vector */ + const opus_int32 *wXX, /* I Correlation matrix */ + const opus_int32 *wXx, /* I Correlation vector */ + opus_int32 wxx, /* I Signal energy */ + opus_int D, /* I Dimension */ + opus_int cQ /* I Q value for c vector 0 - 15 */ +) +{ + opus_int i, j, lshifts, Qxtra; + opus_int32 c_max, w_max, tmp, tmp2, nrg; + opus_int cn[ MAX_MATRIX_SIZE ]; + const opus_int32 *pRow; + + /* Safety checks */ + celt_assert( D >= 0 ); + celt_assert( D <= 16 ); + celt_assert( cQ > 0 ); + celt_assert( cQ < 16 ); + + lshifts = 16 - cQ; + Qxtra = lshifts; + + c_max = 0; + for( i = 0; i < D; i++ ) { + c_max = silk_max_32( c_max, silk_abs( (opus_int32)c[ i ] ) ); + } + Qxtra = silk_min_int( Qxtra, silk_CLZ32( c_max ) - 17 ); + + w_max = silk_max_32( wXX[ 0 ], wXX[ D * D - 1 ] ); + Qxtra = silk_min_int( Qxtra, silk_CLZ32( silk_MUL( D, silk_RSHIFT( silk_SMULWB( w_max, c_max ), 4 ) ) ) - 5 ); + Qxtra = silk_max_int( Qxtra, 0 ); + for( i = 0; i < D; i++ ) { + cn[ i ] = silk_LSHIFT( ( opus_int )c[ i ], Qxtra ); + silk_assert( silk_abs(cn[i]) <= ( silk_int16_MAX + 1 ) ); /* Check that silk_SMLAWB can be used */ + } + lshifts -= Qxtra; + + /* Compute wxx - 2 * wXx * c */ + tmp = 0; + for( i = 0; i < D; i++ ) { + tmp = silk_SMLAWB( tmp, wXx[ i ], cn[ i ] ); + } + nrg = silk_RSHIFT( wxx, 1 + lshifts ) - tmp; /* Q: -lshifts - 1 */ + + /* Add c' * wXX * c, assuming wXX is symmetric */ + tmp2 = 0; + for( i = 0; i < D; i++ ) { + tmp = 0; + pRow = &wXX[ i * D ]; + for( j = i + 1; j < D; j++ ) { + tmp = silk_SMLAWB( tmp, pRow[ j ], cn[ j ] ); + } + tmp = silk_SMLAWB( tmp, silk_RSHIFT( pRow[ i ], 1 ), cn[ i ] ); + tmp2 = silk_SMLAWB( tmp2, tmp, cn[ i ] ); + } + nrg = silk_ADD_LSHIFT32( nrg, tmp2, lshifts ); /* Q: -lshifts - 1 */ + + /* Keep one bit free always, because we add them for LSF interpolation */ + if( nrg < 1 ) { + nrg = 1; + } else if( nrg > silk_RSHIFT( silk_int32_MAX, lshifts + 2 ) ) { + nrg = silk_int32_MAX >> 1; + } else { + nrg = silk_LSHIFT( nrg, lshifts + 1 ); /* Q0 */ + } + return nrg; + +} diff --git a/media/libopus/silk/fixed/residual_energy_FIX.c b/media/libopus/silk/fixed/residual_energy_FIX.c new file mode 100644 index 0000000000..6c7cade9a0 --- /dev/null +++ b/media/libopus/silk/fixed/residual_energy_FIX.c @@ -0,0 +1,98 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main_FIX.h" +#include "stack_alloc.h" + +/* Calculates residual energies of input subframes where all subframes have LPC_order */ +/* of preceding samples */ +void silk_residual_energy_FIX( + opus_int32 nrgs[ MAX_NB_SUBFR ], /* O Residual energy per subframe */ + opus_int nrgsQ[ MAX_NB_SUBFR ], /* O Q value per subframe */ + const opus_int16 x[], /* I Input signal */ + opus_int16 a_Q12[ 2 ][ MAX_LPC_ORDER ], /* I AR coefs for each frame half */ + const opus_int32 gains[ MAX_NB_SUBFR ], /* I Quantization gains */ + const opus_int subfr_length, /* I Subframe length */ + const opus_int nb_subfr, /* I Number of subframes */ + const opus_int LPC_order, /* I LPC order */ + int arch /* I Run-time architecture */ +) +{ + opus_int offset, i, j, rshift, lz1, lz2; + opus_int16 *LPC_res_ptr; + VARDECL( opus_int16, LPC_res ); + const opus_int16 *x_ptr; + opus_int32 tmp32; + SAVE_STACK; + + x_ptr = x; + offset = LPC_order + subfr_length; + + /* Filter input to create the LPC residual for each frame half, and measure subframe energies */ + ALLOC( LPC_res, ( MAX_NB_SUBFR >> 1 ) * offset, opus_int16 ); + celt_assert( ( nb_subfr >> 1 ) * ( MAX_NB_SUBFR >> 1 ) == nb_subfr ); + for( i = 0; i < nb_subfr >> 1; i++ ) { + /* Calculate half frame LPC residual signal including preceding samples */ + silk_LPC_analysis_filter( LPC_res, x_ptr, a_Q12[ i ], ( MAX_NB_SUBFR >> 1 ) * offset, LPC_order, arch ); + + /* Point to first subframe of the just calculated LPC residual signal */ + LPC_res_ptr = LPC_res + LPC_order; + for( j = 0; j < ( MAX_NB_SUBFR >> 1 ); j++ ) { + /* Measure subframe energy */ + silk_sum_sqr_shift( &nrgs[ i * ( MAX_NB_SUBFR >> 1 ) + j ], &rshift, LPC_res_ptr, subfr_length ); + + /* Set Q values for the measured energy */ + nrgsQ[ i * ( MAX_NB_SUBFR >> 1 ) + j ] = -rshift; + + /* Move to next subframe */ + LPC_res_ptr += offset; + } + /* Move to next frame half */ + x_ptr += ( MAX_NB_SUBFR >> 1 ) * offset; + } + + /* Apply the squared subframe gains */ + for( i = 0; i < nb_subfr; i++ ) { + /* Fully upscale gains and energies */ + lz1 = silk_CLZ32( nrgs[ i ] ) - 1; + lz2 = silk_CLZ32( gains[ i ] ) - 1; + + tmp32 = silk_LSHIFT32( gains[ i ], lz2 ); + + /* Find squared gains */ + tmp32 = silk_SMMUL( tmp32, tmp32 ); /* Q( 2 * lz2 - 32 )*/ + + /* Scale energies */ + nrgs[ i ] = silk_SMMUL( tmp32, silk_LSHIFT32( nrgs[ i ], lz1 ) ); /* Q( nrgsQ[ i ] + lz1 + 2 * lz2 - 32 - 32 )*/ + nrgsQ[ i ] += lz1 + 2 * lz2 - 32 - 32; + } + RESTORE_STACK; +} diff --git a/media/libopus/silk/fixed/schur64_FIX.c b/media/libopus/silk/fixed/schur64_FIX.c new file mode 100644 index 0000000000..4b7e19ea59 --- /dev/null +++ b/media/libopus/silk/fixed/schur64_FIX.c @@ -0,0 +1,93 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "SigProc_FIX.h" + +/* Slower than schur(), but more accurate. */ +/* Uses SMULL(), available on armv4 */ +opus_int32 silk_schur64( /* O returns residual energy */ + opus_int32 rc_Q16[], /* O Reflection coefficients [order] Q16 */ + const opus_int32 c[], /* I Correlations [order+1] */ + opus_int32 order /* I Prediction order */ +) +{ + opus_int k, n; + opus_int32 C[ SILK_MAX_ORDER_LPC + 1 ][ 2 ]; + opus_int32 Ctmp1_Q30, Ctmp2_Q30, rc_tmp_Q31; + + celt_assert( order >= 0 && order <= SILK_MAX_ORDER_LPC ); + + /* Check for invalid input */ + if( c[ 0 ] <= 0 ) { + silk_memset( rc_Q16, 0, order * sizeof( opus_int32 ) ); + return 0; + } + + k = 0; + do { + C[ k ][ 0 ] = C[ k ][ 1 ] = c[ k ]; + } while( ++k <= order ); + + for( k = 0; k < order; k++ ) { + /* Check that we won't be getting an unstable rc, otherwise stop here. */ + if (silk_abs_int32(C[ k + 1 ][ 0 ]) >= C[ 0 ][ 1 ]) { + if ( C[ k + 1 ][ 0 ] > 0 ) { + rc_Q16[ k ] = -SILK_FIX_CONST( .99f, 16 ); + } else { + rc_Q16[ k ] = SILK_FIX_CONST( .99f, 16 ); + } + k++; + break; + } + + /* Get reflection coefficient: divide two Q30 values and get result in Q31 */ + rc_tmp_Q31 = silk_DIV32_varQ( -C[ k + 1 ][ 0 ], C[ 0 ][ 1 ], 31 ); + + /* Save the output */ + rc_Q16[ k ] = silk_RSHIFT_ROUND( rc_tmp_Q31, 15 ); + + /* Update correlations */ + for( n = 0; n < order - k; n++ ) { + Ctmp1_Q30 = C[ n + k + 1 ][ 0 ]; + Ctmp2_Q30 = C[ n ][ 1 ]; + + /* Multiply and add the highest int32 */ + C[ n + k + 1 ][ 0 ] = Ctmp1_Q30 + silk_SMMUL( silk_LSHIFT( Ctmp2_Q30, 1 ), rc_tmp_Q31 ); + C[ n ][ 1 ] = Ctmp2_Q30 + silk_SMMUL( silk_LSHIFT( Ctmp1_Q30, 1 ), rc_tmp_Q31 ); + } + } + + for(; k < order; k++ ) { + rc_Q16[ k ] = 0; + } + + return silk_max_32( 1, C[ 0 ][ 1 ] ); +} diff --git a/media/libopus/silk/fixed/schur_FIX.c b/media/libopus/silk/fixed/schur_FIX.c new file mode 100644 index 0000000000..2840f6b1aa --- /dev/null +++ b/media/libopus/silk/fixed/schur_FIX.c @@ -0,0 +1,107 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "SigProc_FIX.h" + +/* Faster than schur64(), but much less accurate. */ +/* uses SMLAWB(), requiring armv5E and higher. */ +opus_int32 silk_schur( /* O Returns residual energy */ + opus_int16 *rc_Q15, /* O reflection coefficients [order] Q15 */ + const opus_int32 *c, /* I correlations [order+1] */ + const opus_int32 order /* I prediction order */ +) +{ + opus_int k, n, lz; + opus_int32 C[ SILK_MAX_ORDER_LPC + 1 ][ 2 ]; + opus_int32 Ctmp1, Ctmp2, rc_tmp_Q15; + + celt_assert( order >= 0 && order <= SILK_MAX_ORDER_LPC ); + + /* Get number of leading zeros */ + lz = silk_CLZ32( c[ 0 ] ); + + /* Copy correlations and adjust level to Q30 */ + k = 0; + if( lz < 2 ) { + /* lz must be 1, so shift one to the right */ + do { + C[ k ][ 0 ] = C[ k ][ 1 ] = silk_RSHIFT( c[ k ], 1 ); + } while( ++k <= order ); + } else if( lz > 2 ) { + /* Shift to the left */ + lz -= 2; + do { + C[ k ][ 0 ] = C[ k ][ 1 ] = silk_LSHIFT( c[ k ], lz ); + } while( ++k <= order ); + } else { + /* No need to shift */ + do { + C[ k ][ 0 ] = C[ k ][ 1 ] = c[ k ]; + } while( ++k <= order ); + } + + for( k = 0; k < order; k++ ) { + /* Check that we won't be getting an unstable rc, otherwise stop here. */ + if (silk_abs_int32(C[ k + 1 ][ 0 ]) >= C[ 0 ][ 1 ]) { + if ( C[ k + 1 ][ 0 ] > 0 ) { + rc_Q15[ k ] = -SILK_FIX_CONST( .99f, 15 ); + } else { + rc_Q15[ k ] = SILK_FIX_CONST( .99f, 15 ); + } + k++; + break; + } + + /* Get reflection coefficient */ + rc_tmp_Q15 = -silk_DIV32_16( C[ k + 1 ][ 0 ], silk_max_32( silk_RSHIFT( C[ 0 ][ 1 ], 15 ), 1 ) ); + + /* Clip (shouldn't happen for properly conditioned inputs) */ + rc_tmp_Q15 = silk_SAT16( rc_tmp_Q15 ); + + /* Store */ + rc_Q15[ k ] = (opus_int16)rc_tmp_Q15; + + /* Update correlations */ + for( n = 0; n < order - k; n++ ) { + Ctmp1 = C[ n + k + 1 ][ 0 ]; + Ctmp2 = C[ n ][ 1 ]; + C[ n + k + 1 ][ 0 ] = silk_SMLAWB( Ctmp1, silk_LSHIFT( Ctmp2, 1 ), rc_tmp_Q15 ); + C[ n ][ 1 ] = silk_SMLAWB( Ctmp2, silk_LSHIFT( Ctmp1, 1 ), rc_tmp_Q15 ); + } + } + + for(; k < order; k++ ) { + rc_Q15[ k ] = 0; + } + + /* return residual energy */ + return silk_max_32( 1, C[ 0 ][ 1 ] ); +} diff --git a/media/libopus/silk/fixed/structs_FIX.h b/media/libopus/silk/fixed/structs_FIX.h new file mode 100644 index 0000000000..2774a97b24 --- /dev/null +++ b/media/libopus/silk/fixed/structs_FIX.h @@ -0,0 +1,116 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef SILK_STRUCTS_FIX_H +#define SILK_STRUCTS_FIX_H + +#include "typedef.h" +#include "main.h" +#include "structs.h" + +#ifdef __cplusplus +extern "C" +{ +#endif + +/********************************/ +/* Noise shaping analysis state */ +/********************************/ +typedef struct { + opus_int8 LastGainIndex; + opus_int32 HarmBoost_smth_Q16; + opus_int32 HarmShapeGain_smth_Q16; + opus_int32 Tilt_smth_Q16; +} silk_shape_state_FIX; + +/********************************/ +/* Encoder state FIX */ +/********************************/ +typedef struct { + silk_encoder_state sCmn; /* Common struct, shared with floating-point code */ + silk_shape_state_FIX sShape; /* Shape state */ + + /* Buffer for find pitch and noise shape analysis */ + silk_DWORD_ALIGN opus_int16 x_buf[ 2 * MAX_FRAME_LENGTH + LA_SHAPE_MAX ];/* Buffer for find pitch and noise shape analysis */ + opus_int LTPCorr_Q15; /* Normalized correlation from pitch lag estimator */ + opus_int32 resNrgSmth; +} silk_encoder_state_FIX; + +/************************/ +/* Encoder control FIX */ +/************************/ +typedef struct { + /* Prediction and coding parameters */ + opus_int32 Gains_Q16[ MAX_NB_SUBFR ]; + silk_DWORD_ALIGN opus_int16 PredCoef_Q12[ 2 ][ MAX_LPC_ORDER ]; + opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ]; + opus_int LTP_scale_Q14; + opus_int pitchL[ MAX_NB_SUBFR ]; + + /* Noise shaping parameters */ + /* Testing */ + silk_DWORD_ALIGN opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ]; + opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ]; /* Packs two int16 coefficients per int32 value */ + opus_int Tilt_Q14[ MAX_NB_SUBFR ]; + opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ]; + opus_int Lambda_Q10; + opus_int input_quality_Q14; + opus_int coding_quality_Q14; + + /* measures */ + opus_int32 predGain_Q16; + opus_int LTPredCodGain_Q7; + opus_int32 ResNrg[ MAX_NB_SUBFR ]; /* Residual energy per subframe */ + opus_int ResNrgQ[ MAX_NB_SUBFR ]; /* Q domain for the residual energy > 0 */ + + /* Parameters for CBR mode */ + opus_int32 GainsUnq_Q16[ MAX_NB_SUBFR ]; + opus_int8 lastGainIndexPrev; +} silk_encoder_control_FIX; + +/************************/ +/* Encoder Super Struct */ +/************************/ +typedef struct { + silk_encoder_state_FIX state_Fxx[ ENCODER_NUM_CHANNELS ]; + stereo_enc_state sStereo; + opus_int32 nBitsUsedLBRR; + opus_int32 nBitsExceeded; + opus_int nChannelsAPI; + opus_int nChannelsInternal; + opus_int nPrevChannelsInternal; + opus_int timeSinceSwitchAllowed_ms; + opus_int allowBandwidthSwitch; + opus_int prev_decode_only_middle; +} silk_encoder; + + +#ifdef __cplusplus +} +#endif + +#endif diff --git a/media/libopus/silk/fixed/vector_ops_FIX.c b/media/libopus/silk/fixed/vector_ops_FIX.c new file mode 100644 index 0000000000..dcf84070a6 --- /dev/null +++ b/media/libopus/silk/fixed/vector_ops_FIX.c @@ -0,0 +1,102 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "SigProc_FIX.h" +#include "pitch.h" + +/* Copy and multiply a vector by a constant */ +void silk_scale_copy_vector16( + opus_int16 *data_out, + const opus_int16 *data_in, + opus_int32 gain_Q16, /* I Gain in Q16 */ + const opus_int dataSize /* I Length */ +) +{ + opus_int i; + opus_int32 tmp32; + + for( i = 0; i < dataSize; i++ ) { + tmp32 = silk_SMULWB( gain_Q16, data_in[ i ] ); + data_out[ i ] = (opus_int16)silk_CHECK_FIT16( tmp32 ); + } +} + +/* Multiply a vector by a constant */ +void silk_scale_vector32_Q26_lshift_18( + opus_int32 *data1, /* I/O Q0/Q18 */ + opus_int32 gain_Q26, /* I Q26 */ + opus_int dataSize /* I length */ +) +{ + opus_int i; + + for( i = 0; i < dataSize; i++ ) { + data1[ i ] = (opus_int32)silk_CHECK_FIT32( silk_RSHIFT64( silk_SMULL( data1[ i ], gain_Q26 ), 8 ) ); /* OUTPUT: Q18 */ + } +} + +/* sum = for(i=0;i<len;i++)inVec1[i]*inVec2[i]; --- inner product */ +/* Note for ARM asm: */ +/* * inVec1 and inVec2 should be at least 2 byte aligned. */ +/* * len should be positive 16bit integer. */ +/* * only when len>6, memory access can be reduced by half. */ +opus_int32 silk_inner_prod_aligned( + const opus_int16 *const inVec1, /* I input vector 1 */ + const opus_int16 *const inVec2, /* I input vector 2 */ + const opus_int len, /* I vector lengths */ + int arch /* I Run-time architecture */ +) +{ +#ifdef FIXED_POINT + return celt_inner_prod(inVec1, inVec2, len, arch); +#else + opus_int i; + opus_int32 sum = 0; + for( i = 0; i < len; i++ ) { + sum = silk_SMLABB( sum, inVec1[ i ], inVec2[ i ] ); + } + return sum; +#endif +} + +opus_int64 silk_inner_prod16_c( + const opus_int16 *inVec1, /* I input vector 1 */ + const opus_int16 *inVec2, /* I input vector 2 */ + const opus_int len /* I vector lengths */ +) +{ + opus_int i; + opus_int64 sum = 0; + for( i = 0; i < len; i++ ) { + sum = silk_SMLALBB( sum, inVec1[ i ], inVec2[ i ] ); + } + return sum; +} diff --git a/media/libopus/silk/fixed/warped_autocorrelation_FIX.c b/media/libopus/silk/fixed/warped_autocorrelation_FIX.c new file mode 100644 index 0000000000..5c79553bc0 --- /dev/null +++ b/media/libopus/silk/fixed/warped_autocorrelation_FIX.c @@ -0,0 +1,92 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main_FIX.h" + +#if defined(MIPSr1_ASM) +#include "mips/warped_autocorrelation_FIX_mipsr1.h" +#endif + + +/* Autocorrelations for a warped frequency axis */ +#ifndef OVERRIDE_silk_warped_autocorrelation_FIX_c +void silk_warped_autocorrelation_FIX_c( + opus_int32 *corr, /* O Result [order + 1] */ + opus_int *scale, /* O Scaling of the correlation vector */ + const opus_int16 *input, /* I Input data to correlate */ + const opus_int warping_Q16, /* I Warping coefficient */ + const opus_int length, /* I Length of input */ + const opus_int order /* I Correlation order (even) */ +) +{ + opus_int n, i, lsh; + opus_int32 tmp1_QS, tmp2_QS; + opus_int32 state_QS[ MAX_SHAPE_LPC_ORDER + 1 ] = { 0 }; + opus_int64 corr_QC[ MAX_SHAPE_LPC_ORDER + 1 ] = { 0 }; + + /* Order must be even */ + celt_assert( ( order & 1 ) == 0 ); + silk_assert( 2 * QS - QC >= 0 ); + + /* Loop over samples */ + for( n = 0; n < length; n++ ) { + tmp1_QS = silk_LSHIFT32( (opus_int32)input[ n ], QS ); + /* Loop over allpass sections */ + for( i = 0; i < order; i += 2 ) { + /* Output of allpass section */ + tmp2_QS = silk_SMLAWB( state_QS[ i ], state_QS[ i + 1 ] - tmp1_QS, warping_Q16 ); + state_QS[ i ] = tmp1_QS; + corr_QC[ i ] += silk_RSHIFT64( silk_SMULL( tmp1_QS, state_QS[ 0 ] ), 2 * QS - QC ); + /* Output of allpass section */ + tmp1_QS = silk_SMLAWB( state_QS[ i + 1 ], state_QS[ i + 2 ] - tmp2_QS, warping_Q16 ); + state_QS[ i + 1 ] = tmp2_QS; + corr_QC[ i + 1 ] += silk_RSHIFT64( silk_SMULL( tmp2_QS, state_QS[ 0 ] ), 2 * QS - QC ); + } + state_QS[ order ] = tmp1_QS; + corr_QC[ order ] += silk_RSHIFT64( silk_SMULL( tmp1_QS, state_QS[ 0 ] ), 2 * QS - QC ); + } + + lsh = silk_CLZ64( corr_QC[ 0 ] ) - 35; + lsh = silk_LIMIT( lsh, -12 - QC, 30 - QC ); + *scale = -( QC + lsh ); + silk_assert( *scale >= -30 && *scale <= 12 ); + if( lsh >= 0 ) { + for( i = 0; i < order + 1; i++ ) { + corr[ i ] = (opus_int32)silk_CHECK_FIT32( silk_LSHIFT64( corr_QC[ i ], lsh ) ); + } + } else { + for( i = 0; i < order + 1; i++ ) { + corr[ i ] = (opus_int32)silk_CHECK_FIT32( silk_RSHIFT64( corr_QC[ i ], -lsh ) ); + } + } + silk_assert( corr_QC[ 0 ] >= 0 ); /* If breaking, decrease QC*/ +} +#endif /* OVERRIDE_silk_warped_autocorrelation_FIX_c */ diff --git a/media/libopus/silk/fixed/x86/burg_modified_FIX_sse4_1.c b/media/libopus/silk/fixed/x86/burg_modified_FIX_sse4_1.c new file mode 100644 index 0000000000..e58bf079e5 --- /dev/null +++ b/media/libopus/silk/fixed/x86/burg_modified_FIX_sse4_1.c @@ -0,0 +1,400 @@ +/* Copyright (c) 2014-2020, Cisco Systems, INC + Written by XiangMingZhu WeiZhou MinPeng YanWang FrancisQuiers + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include <xmmintrin.h> +#include <emmintrin.h> +#include <smmintrin.h> + +#include "SigProc_FIX.h" +#include "define.h" +#include "tuning_parameters.h" +#include "pitch.h" +#include "celt/x86/x86cpu.h" + +#define MAX_FRAME_SIZE 384 /* subfr_length * nb_subfr = ( 0.005 * 16000 + 16 ) * 4 = 384 */ + +#define QA 25 +#define N_BITS_HEAD_ROOM 3 +#define MIN_RSHIFTS -16 +#define MAX_RSHIFTS (32 - QA) + +/* Compute reflection coefficients from input signal */ +void silk_burg_modified_sse4_1( + opus_int32 *res_nrg, /* O Residual energy */ + opus_int *res_nrg_Q, /* O Residual energy Q value */ + opus_int32 A_Q16[], /* O Prediction coefficients (length order) */ + const opus_int16 x[], /* I Input signal, length: nb_subfr * ( D + subfr_length ) */ + const opus_int32 minInvGain_Q30, /* I Inverse of max prediction gain */ + const opus_int subfr_length, /* I Input signal subframe length (incl. D preceding samples) */ + const opus_int nb_subfr, /* I Number of subframes stacked in x */ + const opus_int D, /* I Order */ + int arch /* I Run-time architecture */ +) +{ + opus_int k, n, s, lz, rshifts, reached_max_gain; + opus_int32 C0, num, nrg, rc_Q31, invGain_Q30, Atmp_QA, Atmp1, tmp1, tmp2, x1, x2; + const opus_int16 *x_ptr; + opus_int32 C_first_row[ SILK_MAX_ORDER_LPC ]; + opus_int32 C_last_row[ SILK_MAX_ORDER_LPC ]; + opus_int32 Af_QA[ SILK_MAX_ORDER_LPC ]; + opus_int32 CAf[ SILK_MAX_ORDER_LPC + 1 ]; + opus_int32 CAb[ SILK_MAX_ORDER_LPC + 1 ]; + opus_int32 xcorr[ SILK_MAX_ORDER_LPC ]; + opus_int64 C0_64; + + __m128i FIRST_3210, LAST_3210, ATMP_3210, TMP1_3210, TMP2_3210, T1_3210, T2_3210, PTR_3210, SUBFR_3210, X1_3210, X2_3210; + __m128i CONST1 = _mm_set1_epi32(1); + + celt_assert( subfr_length * nb_subfr <= MAX_FRAME_SIZE ); + + /* Compute autocorrelations, added over subframes */ + C0_64 = silk_inner_prod16( x, x, subfr_length*nb_subfr, arch ); + lz = silk_CLZ64(C0_64); + rshifts = 32 + 1 + N_BITS_HEAD_ROOM - lz; + if (rshifts > MAX_RSHIFTS) rshifts = MAX_RSHIFTS; + if (rshifts < MIN_RSHIFTS) rshifts = MIN_RSHIFTS; + + if (rshifts > 0) { + C0 = (opus_int32)silk_RSHIFT64(C0_64, rshifts ); + } else { + C0 = silk_LSHIFT32((opus_int32)C0_64, -rshifts ); + } + + CAb[ 0 ] = CAf[ 0 ] = C0 + silk_SMMUL( SILK_FIX_CONST( FIND_LPC_COND_FAC, 32 ), C0 ) + 1; /* Q(-rshifts) */ + silk_memset( C_first_row, 0, SILK_MAX_ORDER_LPC * sizeof( opus_int32 ) ); + if( rshifts > 0 ) { + for( s = 0; s < nb_subfr; s++ ) { + x_ptr = x + s * subfr_length; + for( n = 1; n < D + 1; n++ ) { + C_first_row[ n - 1 ] += (opus_int32)silk_RSHIFT64( + silk_inner_prod16( x_ptr, x_ptr + n, subfr_length - n, arch ), rshifts ); + } + } + } else { + for( s = 0; s < nb_subfr; s++ ) { + int i; + opus_int32 d; + x_ptr = x + s * subfr_length; + celt_pitch_xcorr(x_ptr, x_ptr + 1, xcorr, subfr_length - D, D, arch ); + for( n = 1; n < D + 1; n++ ) { + for ( i = n + subfr_length - D, d = 0; i < subfr_length; i++ ) + d = MAC16_16( d, x_ptr[ i ], x_ptr[ i - n ] ); + xcorr[ n - 1 ] += d; + } + for( n = 1; n < D + 1; n++ ) { + C_first_row[ n - 1 ] += silk_LSHIFT32( xcorr[ n - 1 ], -rshifts ); + } + } + } + silk_memcpy( C_last_row, C_first_row, SILK_MAX_ORDER_LPC * sizeof( opus_int32 ) ); + + /* Initialize */ + CAb[ 0 ] = CAf[ 0 ] = C0 + silk_SMMUL( SILK_FIX_CONST( FIND_LPC_COND_FAC, 32 ), C0 ) + 1; /* Q(-rshifts) */ + + invGain_Q30 = (opus_int32)1 << 30; + reached_max_gain = 0; + for( n = 0; n < D; n++ ) { + /* Update first row of correlation matrix (without first element) */ + /* Update last row of correlation matrix (without last element, stored in reversed order) */ + /* Update C * Af */ + /* Update C * flipud(Af) (stored in reversed order) */ + if( rshifts > -2 ) { + for( s = 0; s < nb_subfr; s++ ) { + x_ptr = x + s * subfr_length; + x1 = -silk_LSHIFT32( (opus_int32)x_ptr[ n ], 16 - rshifts ); /* Q(16-rshifts) */ + x2 = -silk_LSHIFT32( (opus_int32)x_ptr[ subfr_length - n - 1 ], 16 - rshifts ); /* Q(16-rshifts) */ + tmp1 = silk_LSHIFT32( (opus_int32)x_ptr[ n ], QA - 16 ); /* Q(QA-16) */ + tmp2 = silk_LSHIFT32( (opus_int32)x_ptr[ subfr_length - n - 1 ], QA - 16 ); /* Q(QA-16) */ + for( k = 0; k < n; k++ ) { + C_first_row[ k ] = silk_SMLAWB( C_first_row[ k ], x1, x_ptr[ n - k - 1 ] ); /* Q( -rshifts ) */ + C_last_row[ k ] = silk_SMLAWB( C_last_row[ k ], x2, x_ptr[ subfr_length - n + k ] ); /* Q( -rshifts ) */ + Atmp_QA = Af_QA[ k ]; + tmp1 = silk_SMLAWB( tmp1, Atmp_QA, x_ptr[ n - k - 1 ] ); /* Q(QA-16) */ + tmp2 = silk_SMLAWB( tmp2, Atmp_QA, x_ptr[ subfr_length - n + k ] ); /* Q(QA-16) */ + } + tmp1 = silk_LSHIFT32( -tmp1, 32 - QA - rshifts ); /* Q(16-rshifts) */ + tmp2 = silk_LSHIFT32( -tmp2, 32 - QA - rshifts ); /* Q(16-rshifts) */ + for( k = 0; k <= n; k++ ) { + CAf[ k ] = silk_SMLAWB( CAf[ k ], tmp1, x_ptr[ n - k ] ); /* Q( -rshift ) */ + CAb[ k ] = silk_SMLAWB( CAb[ k ], tmp2, x_ptr[ subfr_length - n + k - 1 ] ); /* Q( -rshift ) */ + } + } + } else { + for( s = 0; s < nb_subfr; s++ ) { + x_ptr = x + s * subfr_length; + x1 = -silk_LSHIFT32( (opus_int32)x_ptr[ n ], -rshifts ); /* Q( -rshifts ) */ + x2 = -silk_LSHIFT32( (opus_int32)x_ptr[ subfr_length - n - 1 ], -rshifts ); /* Q( -rshifts ) */ + tmp1 = silk_LSHIFT32( (opus_int32)x_ptr[ n ], 17 ); /* Q17 */ + tmp2 = silk_LSHIFT32( (opus_int32)x_ptr[ subfr_length - n - 1 ], 17 ); /* Q17 */ + + X1_3210 = _mm_set1_epi32( x1 ); + X2_3210 = _mm_set1_epi32( x2 ); + TMP1_3210 = _mm_setzero_si128(); + TMP2_3210 = _mm_setzero_si128(); + for( k = 0; k < n - 3; k += 4 ) { + PTR_3210 = OP_CVTEPI16_EPI32_M64( &x_ptr[ n - k - 1 - 3 ] ); + SUBFR_3210 = OP_CVTEPI16_EPI32_M64( &x_ptr[ subfr_length - n + k ] ); + FIRST_3210 = _mm_loadu_si128( (__m128i *)&C_first_row[ k ] ); + PTR_3210 = _mm_shuffle_epi32( PTR_3210, _MM_SHUFFLE( 0, 1, 2, 3 ) ); + LAST_3210 = _mm_loadu_si128( (__m128i *)&C_last_row[ k ] ); + ATMP_3210 = _mm_loadu_si128( (__m128i *)&Af_QA[ k ] ); + + T1_3210 = _mm_mullo_epi32( PTR_3210, X1_3210 ); + T2_3210 = _mm_mullo_epi32( SUBFR_3210, X2_3210 ); + + ATMP_3210 = _mm_srai_epi32( ATMP_3210, 7 ); + ATMP_3210 = _mm_add_epi32( ATMP_3210, CONST1 ); + ATMP_3210 = _mm_srai_epi32( ATMP_3210, 1 ); + + FIRST_3210 = _mm_add_epi32( FIRST_3210, T1_3210 ); + LAST_3210 = _mm_add_epi32( LAST_3210, T2_3210 ); + + PTR_3210 = _mm_mullo_epi32( ATMP_3210, PTR_3210 ); + SUBFR_3210 = _mm_mullo_epi32( ATMP_3210, SUBFR_3210 ); + + _mm_storeu_si128( (__m128i *)&C_first_row[ k ], FIRST_3210 ); + _mm_storeu_si128( (__m128i *)&C_last_row[ k ], LAST_3210 ); + + TMP1_3210 = _mm_add_epi32( TMP1_3210, PTR_3210 ); + TMP2_3210 = _mm_add_epi32( TMP2_3210, SUBFR_3210 ); + } + + TMP1_3210 = _mm_add_epi32( TMP1_3210, _mm_unpackhi_epi64(TMP1_3210, TMP1_3210 ) ); + TMP2_3210 = _mm_add_epi32( TMP2_3210, _mm_unpackhi_epi64(TMP2_3210, TMP2_3210 ) ); + TMP1_3210 = _mm_add_epi32( TMP1_3210, _mm_shufflelo_epi16(TMP1_3210, 0x0E ) ); + TMP2_3210 = _mm_add_epi32( TMP2_3210, _mm_shufflelo_epi16(TMP2_3210, 0x0E ) ); + + tmp1 += _mm_cvtsi128_si32( TMP1_3210 ); + tmp2 += _mm_cvtsi128_si32( TMP2_3210 ); + + for( ; k < n; k++ ) { + C_first_row[ k ] = silk_MLA( C_first_row[ k ], x1, x_ptr[ n - k - 1 ] ); /* Q( -rshifts ) */ + C_last_row[ k ] = silk_MLA( C_last_row[ k ], x2, x_ptr[ subfr_length - n + k ] ); /* Q( -rshifts ) */ + Atmp1 = silk_RSHIFT_ROUND( Af_QA[ k ], QA - 17 ); /* Q17 */ + /* We sometimes get overflows in the multiplications (even beyond +/- 2^32), + but they cancel each other and the real result seems to always fit in a 32-bit + signed integer. This was determined experimentally, not theoretically (unfortunately). */ + tmp1 = silk_MLA_ovflw( tmp1, x_ptr[ n - k - 1 ], Atmp1 ); /* Q17 */ + tmp2 = silk_MLA_ovflw( tmp2, x_ptr[ subfr_length - n + k ], Atmp1 ); /* Q17 */ + } + + tmp1 = -tmp1; /* Q17 */ + tmp2 = -tmp2; /* Q17 */ + + { + __m128i xmm_tmp1, xmm_tmp2; + __m128i xmm_x_ptr_n_k_x2x0, xmm_x_ptr_n_k_x3x1; + __m128i xmm_x_ptr_sub_x2x0, xmm_x_ptr_sub_x3x1; + + xmm_tmp1 = _mm_set1_epi32( tmp1 ); + xmm_tmp2 = _mm_set1_epi32( tmp2 ); + + for( k = 0; k <= n - 3; k += 4 ) { + xmm_x_ptr_n_k_x2x0 = OP_CVTEPI16_EPI32_M64( &x_ptr[ n - k - 3 ] ); + xmm_x_ptr_sub_x2x0 = OP_CVTEPI16_EPI32_M64( &x_ptr[ subfr_length - n + k - 1 ] ); + + xmm_x_ptr_n_k_x2x0 = _mm_shuffle_epi32( xmm_x_ptr_n_k_x2x0, _MM_SHUFFLE( 0, 1, 2, 3 ) ); + + xmm_x_ptr_n_k_x2x0 = _mm_slli_epi32( xmm_x_ptr_n_k_x2x0, -rshifts - 1 ); + xmm_x_ptr_sub_x2x0 = _mm_slli_epi32( xmm_x_ptr_sub_x2x0, -rshifts - 1 ); + + /* equal shift right 4 bytes, xmm_x_ptr_n_k_x3x1 = _mm_srli_si128(xmm_x_ptr_n_k_x2x0, 4)*/ + xmm_x_ptr_n_k_x3x1 = _mm_shuffle_epi32( xmm_x_ptr_n_k_x2x0, _MM_SHUFFLE( 0, 3, 2, 1 ) ); + xmm_x_ptr_sub_x3x1 = _mm_shuffle_epi32( xmm_x_ptr_sub_x2x0, _MM_SHUFFLE( 0, 3, 2, 1 ) ); + + xmm_x_ptr_n_k_x2x0 = _mm_mul_epi32( xmm_x_ptr_n_k_x2x0, xmm_tmp1 ); + xmm_x_ptr_n_k_x3x1 = _mm_mul_epi32( xmm_x_ptr_n_k_x3x1, xmm_tmp1 ); + xmm_x_ptr_sub_x2x0 = _mm_mul_epi32( xmm_x_ptr_sub_x2x0, xmm_tmp2 ); + xmm_x_ptr_sub_x3x1 = _mm_mul_epi32( xmm_x_ptr_sub_x3x1, xmm_tmp2 ); + + xmm_x_ptr_n_k_x2x0 = _mm_srli_epi64( xmm_x_ptr_n_k_x2x0, 16 ); + xmm_x_ptr_n_k_x3x1 = _mm_slli_epi64( xmm_x_ptr_n_k_x3x1, 16 ); + xmm_x_ptr_sub_x2x0 = _mm_srli_epi64( xmm_x_ptr_sub_x2x0, 16 ); + xmm_x_ptr_sub_x3x1 = _mm_slli_epi64( xmm_x_ptr_sub_x3x1, 16 ); + + xmm_x_ptr_n_k_x2x0 = _mm_blend_epi16( xmm_x_ptr_n_k_x2x0, xmm_x_ptr_n_k_x3x1, 0xCC ); + xmm_x_ptr_sub_x2x0 = _mm_blend_epi16( xmm_x_ptr_sub_x2x0, xmm_x_ptr_sub_x3x1, 0xCC ); + + X1_3210 = _mm_loadu_si128( (__m128i *)&CAf[ k ] ); + PTR_3210 = _mm_loadu_si128( (__m128i *)&CAb[ k ] ); + + X1_3210 = _mm_add_epi32( X1_3210, xmm_x_ptr_n_k_x2x0 ); + PTR_3210 = _mm_add_epi32( PTR_3210, xmm_x_ptr_sub_x2x0 ); + + _mm_storeu_si128( (__m128i *)&CAf[ k ], X1_3210 ); + _mm_storeu_si128( (__m128i *)&CAb[ k ], PTR_3210 ); + } + + for( ; k <= n; k++ ) { + CAf[ k ] = silk_SMLAWW( CAf[ k ], tmp1, + silk_LSHIFT32( (opus_int32)x_ptr[ n - k ], -rshifts - 1 ) ); /* Q( -rshift ) */ + CAb[ k ] = silk_SMLAWW( CAb[ k ], tmp2, + silk_LSHIFT32( (opus_int32)x_ptr[ subfr_length - n + k - 1 ], -rshifts - 1 ) ); /* Q( -rshift ) */ + } + } + } + } + + /* Calculate nominator and denominator for the next order reflection (parcor) coefficient */ + tmp1 = C_first_row[ n ]; /* Q( -rshifts ) */ + tmp2 = C_last_row[ n ]; /* Q( -rshifts ) */ + num = 0; /* Q( -rshifts ) */ + nrg = silk_ADD32( CAb[ 0 ], CAf[ 0 ] ); /* Q( 1-rshifts ) */ + for( k = 0; k < n; k++ ) { + Atmp_QA = Af_QA[ k ]; + lz = silk_CLZ32( silk_abs( Atmp_QA ) ) - 1; + lz = silk_min( 32 - QA, lz ); + Atmp1 = silk_LSHIFT32( Atmp_QA, lz ); /* Q( QA + lz ) */ + + tmp1 = silk_ADD_LSHIFT32( tmp1, silk_SMMUL( C_last_row[ n - k - 1 ], Atmp1 ), 32 - QA - lz ); /* Q( -rshifts ) */ + tmp2 = silk_ADD_LSHIFT32( tmp2, silk_SMMUL( C_first_row[ n - k - 1 ], Atmp1 ), 32 - QA - lz ); /* Q( -rshifts ) */ + num = silk_ADD_LSHIFT32( num, silk_SMMUL( CAb[ n - k ], Atmp1 ), 32 - QA - lz ); /* Q( -rshifts ) */ + nrg = silk_ADD_LSHIFT32( nrg, silk_SMMUL( silk_ADD32( CAb[ k + 1 ], CAf[ k + 1 ] ), + Atmp1 ), 32 - QA - lz ); /* Q( 1-rshifts ) */ + } + CAf[ n + 1 ] = tmp1; /* Q( -rshifts ) */ + CAb[ n + 1 ] = tmp2; /* Q( -rshifts ) */ + num = silk_ADD32( num, tmp2 ); /* Q( -rshifts ) */ + num = silk_LSHIFT32( -num, 1 ); /* Q( 1-rshifts ) */ + + /* Calculate the next order reflection (parcor) coefficient */ + if( silk_abs( num ) < nrg ) { + rc_Q31 = silk_DIV32_varQ( num, nrg, 31 ); + } else { + rc_Q31 = ( num > 0 ) ? silk_int32_MAX : silk_int32_MIN; + } + + /* Update inverse prediction gain */ + tmp1 = ( (opus_int32)1 << 30 ) - silk_SMMUL( rc_Q31, rc_Q31 ); + tmp1 = silk_LSHIFT( silk_SMMUL( invGain_Q30, tmp1 ), 2 ); + if( tmp1 <= minInvGain_Q30 ) { + /* Max prediction gain exceeded; set reflection coefficient such that max prediction gain is exactly hit */ + tmp2 = ( (opus_int32)1 << 30 ) - silk_DIV32_varQ( minInvGain_Q30, invGain_Q30, 30 ); /* Q30 */ + rc_Q31 = silk_SQRT_APPROX( tmp2 ); /* Q15 */ + if( rc_Q31 > 0 ) { + /* Newton-Raphson iteration */ + rc_Q31 = silk_RSHIFT32( rc_Q31 + silk_DIV32( tmp2, rc_Q31 ), 1 ); /* Q15 */ + rc_Q31 = silk_LSHIFT32( rc_Q31, 16 ); /* Q31 */ + if( num < 0 ) { + /* Ensure adjusted reflection coefficients has the original sign */ + rc_Q31 = -rc_Q31; + } + } + invGain_Q30 = minInvGain_Q30; + reached_max_gain = 1; + } else { + invGain_Q30 = tmp1; + } + + /* Update the AR coefficients */ + for( k = 0; k < (n + 1) >> 1; k++ ) { + tmp1 = Af_QA[ k ]; /* QA */ + tmp2 = Af_QA[ n - k - 1 ]; /* QA */ + Af_QA[ k ] = silk_ADD_LSHIFT32( tmp1, silk_SMMUL( tmp2, rc_Q31 ), 1 ); /* QA */ + Af_QA[ n - k - 1 ] = silk_ADD_LSHIFT32( tmp2, silk_SMMUL( tmp1, rc_Q31 ), 1 ); /* QA */ + } + Af_QA[ n ] = silk_RSHIFT32( rc_Q31, 31 - QA ); /* QA */ + + if( reached_max_gain ) { + /* Reached max prediction gain; set remaining coefficients to zero and exit loop */ + for( k = n + 1; k < D; k++ ) { + Af_QA[ k ] = 0; + } + break; + } + + /* Update C * Af and C * Ab */ + for( k = 0; k <= n + 1; k++ ) { + tmp1 = CAf[ k ]; /* Q( -rshifts ) */ + tmp2 = CAb[ n - k + 1 ]; /* Q( -rshifts ) */ + CAf[ k ] = silk_ADD_LSHIFT32( tmp1, silk_SMMUL( tmp2, rc_Q31 ), 1 ); /* Q( -rshifts ) */ + CAb[ n - k + 1 ] = silk_ADD_LSHIFT32( tmp2, silk_SMMUL( tmp1, rc_Q31 ), 1 ); /* Q( -rshifts ) */ + } + } + + if( reached_max_gain ) { + for( k = 0; k < D; k++ ) { + /* Scale coefficients */ + A_Q16[ k ] = -silk_RSHIFT_ROUND( Af_QA[ k ], QA - 16 ); + } + /* Subtract energy of preceding samples from C0 */ + if( rshifts > 0 ) { + for( s = 0; s < nb_subfr; s++ ) { + x_ptr = x + s * subfr_length; + C0 -= (opus_int32)silk_RSHIFT64( silk_inner_prod16( x_ptr, x_ptr, D, arch ), rshifts ); + } + } else { + for( s = 0; s < nb_subfr; s++ ) { + x_ptr = x + s * subfr_length; + C0 -= silk_LSHIFT32( silk_inner_prod_aligned( x_ptr, x_ptr, D, arch ), -rshifts ); + } + } + /* Approximate residual energy */ + *res_nrg = silk_LSHIFT( silk_SMMUL( invGain_Q30, C0 ), 2 ); + *res_nrg_Q = -rshifts; + } else { + /* Return residual energy */ + nrg = CAf[ 0 ]; /* Q( -rshifts ) */ + tmp1 = (opus_int32)1 << 16; /* Q16 */ + for( k = 0; k < D; k++ ) { + Atmp1 = silk_RSHIFT_ROUND( Af_QA[ k ], QA - 16 ); /* Q16 */ + nrg = silk_SMLAWW( nrg, CAf[ k + 1 ], Atmp1 ); /* Q( -rshifts ) */ + tmp1 = silk_SMLAWW( tmp1, Atmp1, Atmp1 ); /* Q16 */ + A_Q16[ k ] = -Atmp1; + } + *res_nrg = silk_SMLAWW( nrg, silk_SMMUL( SILK_FIX_CONST( FIND_LPC_COND_FAC, 32 ), C0 ), -tmp1 );/* Q( -rshifts ) */ + *res_nrg_Q = -rshifts; + } + +#ifdef OPUS_CHECK_ASM + { + opus_int32 res_nrg_c = 0; + opus_int res_nrg_Q_c = 0; + opus_int32 A_Q16_c[ MAX_LPC_ORDER ] = {0}; + + silk_burg_modified_c( + &res_nrg_c, + &res_nrg_Q_c, + A_Q16_c, + x, + minInvGain_Q30, + subfr_length, + nb_subfr, + D, + 0 + ); + + silk_assert( *res_nrg == res_nrg_c ); + silk_assert( *res_nrg_Q == res_nrg_Q_c ); + silk_assert( !memcmp( A_Q16, A_Q16_c, D * sizeof( *A_Q16 ) ) ); + } +#endif +} diff --git a/media/libopus/silk/fixed/x86/vector_ops_FIX_sse4_1.c b/media/libopus/silk/fixed/x86/vector_ops_FIX_sse4_1.c new file mode 100644 index 0000000000..a46289bbc4 --- /dev/null +++ b/media/libopus/silk/fixed/x86/vector_ops_FIX_sse4_1.c @@ -0,0 +1,93 @@ +/* Copyright (c) 2014, Cisco Systems, INC + Written by XiangMingZhu WeiZhou MinPeng YanWang + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include <xmmintrin.h> +#include <emmintrin.h> +#include <smmintrin.h> +#include "main.h" + +#include "SigProc_FIX.h" +#include "pitch.h" +#include "celt/x86/x86cpu.h" + +opus_int64 silk_inner_prod16_sse4_1( + const opus_int16 *inVec1, /* I input vector 1 */ + const opus_int16 *inVec2, /* I input vector 2 */ + const opus_int len /* I vector lengths */ +) +{ + opus_int i, dataSize4; + opus_int64 sum; + + __m128i xmm_prod_20, xmm_prod_31; + __m128i inVec1_3210, acc1; + __m128i inVec2_3210, acc2; + + sum = 0; + dataSize4 = len & ~3; + + acc1 = _mm_setzero_si128(); + acc2 = _mm_setzero_si128(); + + for( i = 0; i < dataSize4; i += 4 ) { + inVec1_3210 = OP_CVTEPI16_EPI32_M64( &inVec1[i + 0] ); + inVec2_3210 = OP_CVTEPI16_EPI32_M64( &inVec2[i + 0] ); + xmm_prod_20 = _mm_mul_epi32( inVec1_3210, inVec2_3210 ); + + inVec1_3210 = _mm_shuffle_epi32( inVec1_3210, _MM_SHUFFLE( 0, 3, 2, 1 ) ); + inVec2_3210 = _mm_shuffle_epi32( inVec2_3210, _MM_SHUFFLE( 0, 3, 2, 1 ) ); + xmm_prod_31 = _mm_mul_epi32( inVec1_3210, inVec2_3210 ); + + acc1 = _mm_add_epi64( acc1, xmm_prod_20 ); + acc2 = _mm_add_epi64( acc2, xmm_prod_31 ); + } + + acc1 = _mm_add_epi64( acc1, acc2 ); + + /* equal shift right 8 bytes */ + acc2 = _mm_shuffle_epi32( acc1, _MM_SHUFFLE( 0, 0, 3, 2 ) ); + acc1 = _mm_add_epi64( acc1, acc2 ); + + _mm_storel_epi64( (__m128i *)&sum, acc1 ); + + for( ; i < len; i++ ) { + sum = silk_SMLALBB( sum, inVec1[ i ], inVec2[ i ] ); + } + +#ifdef OPUS_CHECK_ASM + { + opus_int64 sum_c = silk_inner_prod16_c( inVec1, inVec2, len ); + silk_assert( sum == sum_c ); + } +#endif + + return sum; +} diff --git a/media/libopus/silk/float/LPC_analysis_filter_FLP.c b/media/libopus/silk/float/LPC_analysis_filter_FLP.c new file mode 100644 index 0000000000..0e1a1fed0f --- /dev/null +++ b/media/libopus/silk/float/LPC_analysis_filter_FLP.c @@ -0,0 +1,249 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include <stdlib.h> +#include "main_FLP.h" + +/************************************************/ +/* LPC analysis filter */ +/* NB! State is kept internally and the */ +/* filter always starts with zero state */ +/* first Order output samples are set to zero */ +/************************************************/ + +/* 16th order LPC analysis filter, does not write first 16 samples */ +static OPUS_INLINE void silk_LPC_analysis_filter16_FLP( + silk_float r_LPC[], /* O LPC residual signal */ + const silk_float PredCoef[], /* I LPC coefficients */ + const silk_float s[], /* I Input signal */ + const opus_int length /* I Length of input signal */ +) +{ + opus_int ix; + silk_float LPC_pred; + const silk_float *s_ptr; + + for( ix = 16; ix < length; ix++ ) { + s_ptr = &s[ix - 1]; + + /* short-term prediction */ + LPC_pred = s_ptr[ 0 ] * PredCoef[ 0 ] + + s_ptr[ -1 ] * PredCoef[ 1 ] + + s_ptr[ -2 ] * PredCoef[ 2 ] + + s_ptr[ -3 ] * PredCoef[ 3 ] + + s_ptr[ -4 ] * PredCoef[ 4 ] + + s_ptr[ -5 ] * PredCoef[ 5 ] + + s_ptr[ -6 ] * PredCoef[ 6 ] + + s_ptr[ -7 ] * PredCoef[ 7 ] + + s_ptr[ -8 ] * PredCoef[ 8 ] + + s_ptr[ -9 ] * PredCoef[ 9 ] + + s_ptr[ -10 ] * PredCoef[ 10 ] + + s_ptr[ -11 ] * PredCoef[ 11 ] + + s_ptr[ -12 ] * PredCoef[ 12 ] + + s_ptr[ -13 ] * PredCoef[ 13 ] + + s_ptr[ -14 ] * PredCoef[ 14 ] + + s_ptr[ -15 ] * PredCoef[ 15 ]; + + /* prediction error */ + r_LPC[ix] = s_ptr[ 1 ] - LPC_pred; + } +} + +/* 12th order LPC analysis filter, does not write first 12 samples */ +static OPUS_INLINE void silk_LPC_analysis_filter12_FLP( + silk_float r_LPC[], /* O LPC residual signal */ + const silk_float PredCoef[], /* I LPC coefficients */ + const silk_float s[], /* I Input signal */ + const opus_int length /* I Length of input signal */ +) +{ + opus_int ix; + silk_float LPC_pred; + const silk_float *s_ptr; + + for( ix = 12; ix < length; ix++ ) { + s_ptr = &s[ix - 1]; + + /* short-term prediction */ + LPC_pred = s_ptr[ 0 ] * PredCoef[ 0 ] + + s_ptr[ -1 ] * PredCoef[ 1 ] + + s_ptr[ -2 ] * PredCoef[ 2 ] + + s_ptr[ -3 ] * PredCoef[ 3 ] + + s_ptr[ -4 ] * PredCoef[ 4 ] + + s_ptr[ -5 ] * PredCoef[ 5 ] + + s_ptr[ -6 ] * PredCoef[ 6 ] + + s_ptr[ -7 ] * PredCoef[ 7 ] + + s_ptr[ -8 ] * PredCoef[ 8 ] + + s_ptr[ -9 ] * PredCoef[ 9 ] + + s_ptr[ -10 ] * PredCoef[ 10 ] + + s_ptr[ -11 ] * PredCoef[ 11 ]; + + /* prediction error */ + r_LPC[ix] = s_ptr[ 1 ] - LPC_pred; + } +} + +/* 10th order LPC analysis filter, does not write first 10 samples */ +static OPUS_INLINE void silk_LPC_analysis_filter10_FLP( + silk_float r_LPC[], /* O LPC residual signal */ + const silk_float PredCoef[], /* I LPC coefficients */ + const silk_float s[], /* I Input signal */ + const opus_int length /* I Length of input signal */ +) +{ + opus_int ix; + silk_float LPC_pred; + const silk_float *s_ptr; + + for( ix = 10; ix < length; ix++ ) { + s_ptr = &s[ix - 1]; + + /* short-term prediction */ + LPC_pred = s_ptr[ 0 ] * PredCoef[ 0 ] + + s_ptr[ -1 ] * PredCoef[ 1 ] + + s_ptr[ -2 ] * PredCoef[ 2 ] + + s_ptr[ -3 ] * PredCoef[ 3 ] + + s_ptr[ -4 ] * PredCoef[ 4 ] + + s_ptr[ -5 ] * PredCoef[ 5 ] + + s_ptr[ -6 ] * PredCoef[ 6 ] + + s_ptr[ -7 ] * PredCoef[ 7 ] + + s_ptr[ -8 ] * PredCoef[ 8 ] + + s_ptr[ -9 ] * PredCoef[ 9 ]; + + /* prediction error */ + r_LPC[ix] = s_ptr[ 1 ] - LPC_pred; + } +} + +/* 8th order LPC analysis filter, does not write first 8 samples */ +static OPUS_INLINE void silk_LPC_analysis_filter8_FLP( + silk_float r_LPC[], /* O LPC residual signal */ + const silk_float PredCoef[], /* I LPC coefficients */ + const silk_float s[], /* I Input signal */ + const opus_int length /* I Length of input signal */ +) +{ + opus_int ix; + silk_float LPC_pred; + const silk_float *s_ptr; + + for( ix = 8; ix < length; ix++ ) { + s_ptr = &s[ix - 1]; + + /* short-term prediction */ + LPC_pred = s_ptr[ 0 ] * PredCoef[ 0 ] + + s_ptr[ -1 ] * PredCoef[ 1 ] + + s_ptr[ -2 ] * PredCoef[ 2 ] + + s_ptr[ -3 ] * PredCoef[ 3 ] + + s_ptr[ -4 ] * PredCoef[ 4 ] + + s_ptr[ -5 ] * PredCoef[ 5 ] + + s_ptr[ -6 ] * PredCoef[ 6 ] + + s_ptr[ -7 ] * PredCoef[ 7 ]; + + /* prediction error */ + r_LPC[ix] = s_ptr[ 1 ] - LPC_pred; + } +} + +/* 6th order LPC analysis filter, does not write first 6 samples */ +static OPUS_INLINE void silk_LPC_analysis_filter6_FLP( + silk_float r_LPC[], /* O LPC residual signal */ + const silk_float PredCoef[], /* I LPC coefficients */ + const silk_float s[], /* I Input signal */ + const opus_int length /* I Length of input signal */ +) +{ + opus_int ix; + silk_float LPC_pred; + const silk_float *s_ptr; + + for( ix = 6; ix < length; ix++ ) { + s_ptr = &s[ix - 1]; + + /* short-term prediction */ + LPC_pred = s_ptr[ 0 ] * PredCoef[ 0 ] + + s_ptr[ -1 ] * PredCoef[ 1 ] + + s_ptr[ -2 ] * PredCoef[ 2 ] + + s_ptr[ -3 ] * PredCoef[ 3 ] + + s_ptr[ -4 ] * PredCoef[ 4 ] + + s_ptr[ -5 ] * PredCoef[ 5 ]; + + /* prediction error */ + r_LPC[ix] = s_ptr[ 1 ] - LPC_pred; + } +} + +/************************************************/ +/* LPC analysis filter */ +/* NB! State is kept internally and the */ +/* filter always starts with zero state */ +/* first Order output samples are set to zero */ +/************************************************/ +void silk_LPC_analysis_filter_FLP( + silk_float r_LPC[], /* O LPC residual signal */ + const silk_float PredCoef[], /* I LPC coefficients */ + const silk_float s[], /* I Input signal */ + const opus_int length, /* I Length of input signal */ + const opus_int Order /* I LPC order */ +) +{ + celt_assert( Order <= length ); + + switch( Order ) { + case 6: + silk_LPC_analysis_filter6_FLP( r_LPC, PredCoef, s, length ); + break; + + case 8: + silk_LPC_analysis_filter8_FLP( r_LPC, PredCoef, s, length ); + break; + + case 10: + silk_LPC_analysis_filter10_FLP( r_LPC, PredCoef, s, length ); + break; + + case 12: + silk_LPC_analysis_filter12_FLP( r_LPC, PredCoef, s, length ); + break; + + case 16: + silk_LPC_analysis_filter16_FLP( r_LPC, PredCoef, s, length ); + break; + + default: + celt_assert( 0 ); + break; + } + + /* Set first Order output samples to zero */ + silk_memset( r_LPC, 0, Order * sizeof( silk_float ) ); +} + diff --git a/media/libopus/silk/float/LPC_inv_pred_gain_FLP.c b/media/libopus/silk/float/LPC_inv_pred_gain_FLP.c new file mode 100644 index 0000000000..2be2122d61 --- /dev/null +++ b/media/libopus/silk/float/LPC_inv_pred_gain_FLP.c @@ -0,0 +1,73 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "SigProc_FIX.h" +#include "SigProc_FLP.h" +#include "define.h" + +/* compute inverse of LPC prediction gain, and */ +/* test if LPC coefficients are stable (all poles within unit circle) */ +/* this code is based on silk_a2k_FLP() */ +silk_float silk_LPC_inverse_pred_gain_FLP( /* O return inverse prediction gain, energy domain */ + const silk_float *A, /* I prediction coefficients [order] */ + opus_int32 order /* I prediction order */ +) +{ + opus_int k, n; + double invGain, rc, rc_mult1, rc_mult2, tmp1, tmp2; + silk_float Atmp[ SILK_MAX_ORDER_LPC ]; + + silk_memcpy( Atmp, A, order * sizeof(silk_float) ); + + invGain = 1.0; + for( k = order - 1; k > 0; k-- ) { + rc = -Atmp[ k ]; + rc_mult1 = 1.0f - rc * rc; + invGain *= rc_mult1; + if( invGain * MAX_PREDICTION_POWER_GAIN < 1.0f ) { + return 0.0f; + } + rc_mult2 = 1.0f / rc_mult1; + for( n = 0; n < (k + 1) >> 1; n++ ) { + tmp1 = Atmp[ n ]; + tmp2 = Atmp[ k - n - 1 ]; + Atmp[ n ] = (silk_float)( ( tmp1 - tmp2 * rc ) * rc_mult2 ); + Atmp[ k - n - 1 ] = (silk_float)( ( tmp2 - tmp1 * rc ) * rc_mult2 ); + } + } + rc = -Atmp[ 0 ]; + rc_mult1 = 1.0f - rc * rc; + invGain *= rc_mult1; + if( invGain * MAX_PREDICTION_POWER_GAIN < 1.0f ) { + return 0.0f; + } + return (silk_float)invGain; +} diff --git a/media/libopus/silk/float/LTP_analysis_filter_FLP.c b/media/libopus/silk/float/LTP_analysis_filter_FLP.c new file mode 100644 index 0000000000..849b7c1c52 --- /dev/null +++ b/media/libopus/silk/float/LTP_analysis_filter_FLP.c @@ -0,0 +1,75 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main_FLP.h" + +void silk_LTP_analysis_filter_FLP( + silk_float *LTP_res, /* O LTP res MAX_NB_SUBFR*(pre_lgth+subfr_lngth) */ + const silk_float *x, /* I Input signal, with preceding samples */ + const silk_float B[ LTP_ORDER * MAX_NB_SUBFR ], /* I LTP coefficients for each subframe */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ + const silk_float invGains[ MAX_NB_SUBFR ], /* I Inverse quantization gains */ + const opus_int subfr_length, /* I Length of each subframe */ + const opus_int nb_subfr, /* I number of subframes */ + const opus_int pre_length /* I Preceding samples for each subframe */ +) +{ + const silk_float *x_ptr, *x_lag_ptr; + silk_float Btmp[ LTP_ORDER ]; + silk_float *LTP_res_ptr; + silk_float inv_gain; + opus_int k, i, j; + + x_ptr = x; + LTP_res_ptr = LTP_res; + for( k = 0; k < nb_subfr; k++ ) { + x_lag_ptr = x_ptr - pitchL[ k ]; + inv_gain = invGains[ k ]; + for( i = 0; i < LTP_ORDER; i++ ) { + Btmp[ i ] = B[ k * LTP_ORDER + i ]; + } + + /* LTP analysis FIR filter */ + for( i = 0; i < subfr_length + pre_length; i++ ) { + LTP_res_ptr[ i ] = x_ptr[ i ]; + /* Subtract long-term prediction */ + for( j = 0; j < LTP_ORDER; j++ ) { + LTP_res_ptr[ i ] -= Btmp[ j ] * x_lag_ptr[ LTP_ORDER / 2 - j ]; + } + LTP_res_ptr[ i ] *= inv_gain; + x_lag_ptr++; + } + + /* Update pointers */ + LTP_res_ptr += subfr_length + pre_length; + x_ptr += subfr_length; + } +} diff --git a/media/libopus/silk/float/LTP_scale_ctrl_FLP.c b/media/libopus/silk/float/LTP_scale_ctrl_FLP.c new file mode 100644 index 0000000000..6f30ff095f --- /dev/null +++ b/media/libopus/silk/float/LTP_scale_ctrl_FLP.c @@ -0,0 +1,58 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main_FLP.h" + +void silk_LTP_scale_ctrl_FLP( + silk_encoder_state_FLP *psEnc, /* I/O Encoder state FLP */ + silk_encoder_control_FLP *psEncCtrl, /* I/O Encoder control FLP */ + opus_int condCoding /* I The type of conditional coding to use */ +) +{ + opus_int round_loss; + + if( condCoding == CODE_INDEPENDENTLY ) { + /* Only scale if first frame in packet */ + round_loss = psEnc->sCmn.PacketLoss_perc * psEnc->sCmn.nFramesPerPacket; + if ( psEnc->sCmn.LBRR_flag ) { + /* LBRR reduces the effective loss. In practice, it does not square the loss because + losses aren't independent, but that still seems to work best. We also never go below 2%. */ + round_loss = 2 + silk_SMULBB( round_loss, round_loss) / 100; + } + psEnc->sCmn.indices.LTP_scaleIndex = silk_SMULBB( psEncCtrl->LTPredCodGain, round_loss ) > silk_log2lin( 2900 - psEnc->sCmn.SNR_dB_Q7 ); + psEnc->sCmn.indices.LTP_scaleIndex += silk_SMULBB( psEncCtrl->LTPredCodGain, round_loss ) > silk_log2lin( 3900 - psEnc->sCmn.SNR_dB_Q7 ); + } else { + /* Default is minimum scaling */ + psEnc->sCmn.indices.LTP_scaleIndex = 0; + } + + psEncCtrl->LTP_scale = (silk_float)silk_LTPScales_table_Q14[ psEnc->sCmn.indices.LTP_scaleIndex ] / 16384.0f; +} diff --git a/media/libopus/silk/float/SigProc_FLP.h b/media/libopus/silk/float/SigProc_FLP.h new file mode 100644 index 0000000000..953de8b09e --- /dev/null +++ b/media/libopus/silk/float/SigProc_FLP.h @@ -0,0 +1,197 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef SILK_SIGPROC_FLP_H +#define SILK_SIGPROC_FLP_H + +#include "SigProc_FIX.h" +#include "float_cast.h" +#include <math.h> + +#ifdef __cplusplus +extern "C" +{ +#endif + +/********************************************************************/ +/* SIGNAL PROCESSING FUNCTIONS */ +/********************************************************************/ + +/* Chirp (bw expand) LP AR filter */ +void silk_bwexpander_FLP( + silk_float *ar, /* I/O AR filter to be expanded (without leading 1) */ + const opus_int d, /* I length of ar */ + const silk_float chirp /* I chirp factor (typically in range (0..1) ) */ +); + +/* compute inverse of LPC prediction gain, and */ +/* test if LPC coefficients are stable (all poles within unit circle) */ +/* this code is based on silk_FLP_a2k() */ +silk_float silk_LPC_inverse_pred_gain_FLP( /* O return inverse prediction gain, energy domain */ + const silk_float *A, /* I prediction coefficients [order] */ + opus_int32 order /* I prediction order */ +); + +silk_float silk_schur_FLP( /* O returns residual energy */ + silk_float refl_coef[], /* O reflection coefficients (length order) */ + const silk_float auto_corr[], /* I autocorrelation sequence (length order+1) */ + opus_int order /* I order */ +); + +void silk_k2a_FLP( + silk_float *A, /* O prediction coefficients [order] */ + const silk_float *rc, /* I reflection coefficients [order] */ + opus_int32 order /* I prediction order */ +); + +/* compute autocorrelation */ +void silk_autocorrelation_FLP( + silk_float *results, /* O result (length correlationCount) */ + const silk_float *inputData, /* I input data to correlate */ + opus_int inputDataSize, /* I length of input */ + opus_int correlationCount /* I number of correlation taps to compute */ +); + +opus_int silk_pitch_analysis_core_FLP( /* O Voicing estimate: 0 voiced, 1 unvoiced */ + const silk_float *frame, /* I Signal of length PE_FRAME_LENGTH_MS*Fs_kHz */ + opus_int *pitch_out, /* O Pitch lag values [nb_subfr] */ + opus_int16 *lagIndex, /* O Lag Index */ + opus_int8 *contourIndex, /* O Pitch contour Index */ + silk_float *LTPCorr, /* I/O Normalized correlation; input: value from previous frame */ + opus_int prevLag, /* I Last lag of previous frame; set to zero is unvoiced */ + const silk_float search_thres1, /* I First stage threshold for lag candidates 0 - 1 */ + const silk_float search_thres2, /* I Final threshold for lag candidates 0 - 1 */ + const opus_int Fs_kHz, /* I sample frequency (kHz) */ + const opus_int complexity, /* I Complexity setting, 0-2, where 2 is highest */ + const opus_int nb_subfr, /* I Number of 5 ms subframes */ + int arch /* I Run-time architecture */ +); + +void silk_insertion_sort_decreasing_FLP( + silk_float *a, /* I/O Unsorted / Sorted vector */ + opus_int *idx, /* O Index vector for the sorted elements */ + const opus_int L, /* I Vector length */ + const opus_int K /* I Number of correctly sorted positions */ +); + +/* Compute reflection coefficients from input signal */ +silk_float silk_burg_modified_FLP( /* O returns residual energy */ + silk_float A[], /* O prediction coefficients (length order) */ + const silk_float x[], /* I input signal, length: nb_subfr*(D+L_sub) */ + const silk_float minInvGain, /* I minimum inverse prediction gain */ + const opus_int subfr_length, /* I input signal subframe length (incl. D preceding samples) */ + const opus_int nb_subfr, /* I number of subframes stacked in x */ + const opus_int D /* I order */ +); + +/* multiply a vector by a constant */ +void silk_scale_vector_FLP( + silk_float *data1, + silk_float gain, + opus_int dataSize +); + +/* copy and multiply a vector by a constant */ +void silk_scale_copy_vector_FLP( + silk_float *data_out, + const silk_float *data_in, + silk_float gain, + opus_int dataSize +); + +/* inner product of two silk_float arrays, with result as double */ +double silk_inner_product_FLP( + const silk_float *data1, + const silk_float *data2, + opus_int dataSize +); + +/* sum of squares of a silk_float array, with result as double */ +double silk_energy_FLP( + const silk_float *data, + opus_int dataSize +); + +/********************************************************************/ +/* MACROS */ +/********************************************************************/ + +#define PI (3.1415926536f) + +#define silk_min_float( a, b ) (((a) < (b)) ? (a) : (b)) +#define silk_max_float( a, b ) (((a) > (b)) ? (a) : (b)) +#define silk_abs_float( a ) ((silk_float)fabs(a)) + +/* sigmoid function */ +static OPUS_INLINE silk_float silk_sigmoid( silk_float x ) +{ + return (silk_float)(1.0 / (1.0 + exp(-x))); +} + +/* floating-point to integer conversion (rounding) */ +static OPUS_INLINE opus_int32 silk_float2int( silk_float x ) +{ + return (opus_int32)float2int( x ); +} + +/* floating-point to integer conversion (rounding) */ +static OPUS_INLINE void silk_float2short_array( + opus_int16 *out, + const silk_float *in, + opus_int32 length +) +{ + opus_int32 k; + for( k = length - 1; k >= 0; k-- ) { + out[k] = silk_SAT16( (opus_int32)float2int( in[k] ) ); + } +} + +/* integer to floating-point conversion */ +static OPUS_INLINE void silk_short2float_array( + silk_float *out, + const opus_int16 *in, + opus_int32 length +) +{ + opus_int32 k; + for( k = length - 1; k >= 0; k-- ) { + out[k] = (silk_float)in[k]; + } +} + +/* using log2() helps the fixed-point conversion */ +static OPUS_INLINE silk_float silk_log2( double x ) +{ + return ( silk_float )( 3.32192809488736 * log10( x ) ); +} + +#ifdef __cplusplus +} +#endif + +#endif /* SILK_SIGPROC_FLP_H */ diff --git a/media/libopus/silk/float/apply_sine_window_FLP.c b/media/libopus/silk/float/apply_sine_window_FLP.c new file mode 100644 index 0000000000..e49e717991 --- /dev/null +++ b/media/libopus/silk/float/apply_sine_window_FLP.c @@ -0,0 +1,81 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main_FLP.h" + +/* Apply sine window to signal vector */ +/* Window types: */ +/* 1 -> sine window from 0 to pi/2 */ +/* 2 -> sine window from pi/2 to pi */ +void silk_apply_sine_window_FLP( + silk_float px_win[], /* O Pointer to windowed signal */ + const silk_float px[], /* I Pointer to input signal */ + const opus_int win_type, /* I Selects a window type */ + const opus_int length /* I Window length, multiple of 4 */ +) +{ + opus_int k; + silk_float freq, c, S0, S1; + + celt_assert( win_type == 1 || win_type == 2 ); + + /* Length must be multiple of 4 */ + celt_assert( ( length & 3 ) == 0 ); + + freq = PI / ( length + 1 ); + + /* Approximation of 2 * cos(f) */ + c = 2.0f - freq * freq; + + /* Initialize state */ + if( win_type < 2 ) { + /* Start from 0 */ + S0 = 0.0f; + /* Approximation of sin(f) */ + S1 = freq; + } else { + /* Start from 1 */ + S0 = 1.0f; + /* Approximation of cos(f) */ + S1 = 0.5f * c; + } + + /* Uses the recursive equation: sin(n*f) = 2 * cos(f) * sin((n-1)*f) - sin((n-2)*f) */ + /* 4 samples at a time */ + for( k = 0; k < length; k += 4 ) { + px_win[ k + 0 ] = px[ k + 0 ] * 0.5f * ( S0 + S1 ); + px_win[ k + 1 ] = px[ k + 1 ] * S1; + S0 = c * S1 - S0; + px_win[ k + 2 ] = px[ k + 2 ] * 0.5f * ( S1 + S0 ); + px_win[ k + 3 ] = px[ k + 3 ] * S0; + S1 = c * S0 - S1; + } +} diff --git a/media/libopus/silk/float/autocorrelation_FLP.c b/media/libopus/silk/float/autocorrelation_FLP.c new file mode 100644 index 0000000000..8b8a9e659a --- /dev/null +++ b/media/libopus/silk/float/autocorrelation_FLP.c @@ -0,0 +1,52 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "typedef.h" +#include "SigProc_FLP.h" + +/* compute autocorrelation */ +void silk_autocorrelation_FLP( + silk_float *results, /* O result (length correlationCount) */ + const silk_float *inputData, /* I input data to correlate */ + opus_int inputDataSize, /* I length of input */ + opus_int correlationCount /* I number of correlation taps to compute */ +) +{ + opus_int i; + + if( correlationCount > inputDataSize ) { + correlationCount = inputDataSize; + } + + for( i = 0; i < correlationCount; i++ ) { + results[ i ] = (silk_float)silk_inner_product_FLP( inputData, inputData + i, inputDataSize - i ); + } +} diff --git a/media/libopus/silk/float/burg_modified_FLP.c b/media/libopus/silk/float/burg_modified_FLP.c new file mode 100644 index 0000000000..756b76a35b --- /dev/null +++ b/media/libopus/silk/float/burg_modified_FLP.c @@ -0,0 +1,186 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "SigProc_FLP.h" +#include "tuning_parameters.h" +#include "define.h" + +#define MAX_FRAME_SIZE 384 /* subfr_length * nb_subfr = ( 0.005 * 16000 + 16 ) * 4 = 384*/ + +/* Compute reflection coefficients from input signal */ +silk_float silk_burg_modified_FLP( /* O returns residual energy */ + silk_float A[], /* O prediction coefficients (length order) */ + const silk_float x[], /* I input signal, length: nb_subfr*(D+L_sub) */ + const silk_float minInvGain, /* I minimum inverse prediction gain */ + const opus_int subfr_length, /* I input signal subframe length (incl. D preceding samples) */ + const opus_int nb_subfr, /* I number of subframes stacked in x */ + const opus_int D /* I order */ +) +{ + opus_int k, n, s, reached_max_gain; + double C0, invGain, num, nrg_f, nrg_b, rc, Atmp, tmp1, tmp2; + const silk_float *x_ptr; + double C_first_row[ SILK_MAX_ORDER_LPC ], C_last_row[ SILK_MAX_ORDER_LPC ]; + double CAf[ SILK_MAX_ORDER_LPC + 1 ], CAb[ SILK_MAX_ORDER_LPC + 1 ]; + double Af[ SILK_MAX_ORDER_LPC ]; + + celt_assert( subfr_length * nb_subfr <= MAX_FRAME_SIZE ); + + /* Compute autocorrelations, added over subframes */ + C0 = silk_energy_FLP( x, nb_subfr * subfr_length ); + silk_memset( C_first_row, 0, SILK_MAX_ORDER_LPC * sizeof( double ) ); + for( s = 0; s < nb_subfr; s++ ) { + x_ptr = x + s * subfr_length; + for( n = 1; n < D + 1; n++ ) { + C_first_row[ n - 1 ] += silk_inner_product_FLP( x_ptr, x_ptr + n, subfr_length - n ); + } + } + silk_memcpy( C_last_row, C_first_row, SILK_MAX_ORDER_LPC * sizeof( double ) ); + + /* Initialize */ + CAb[ 0 ] = CAf[ 0 ] = C0 + FIND_LPC_COND_FAC * C0 + 1e-9f; + invGain = 1.0f; + reached_max_gain = 0; + for( n = 0; n < D; n++ ) { + /* Update first row of correlation matrix (without first element) */ + /* Update last row of correlation matrix (without last element, stored in reversed order) */ + /* Update C * Af */ + /* Update C * flipud(Af) (stored in reversed order) */ + for( s = 0; s < nb_subfr; s++ ) { + x_ptr = x + s * subfr_length; + tmp1 = x_ptr[ n ]; + tmp2 = x_ptr[ subfr_length - n - 1 ]; + for( k = 0; k < n; k++ ) { + C_first_row[ k ] -= x_ptr[ n ] * x_ptr[ n - k - 1 ]; + C_last_row[ k ] -= x_ptr[ subfr_length - n - 1 ] * x_ptr[ subfr_length - n + k ]; + Atmp = Af[ k ]; + tmp1 += x_ptr[ n - k - 1 ] * Atmp; + tmp2 += x_ptr[ subfr_length - n + k ] * Atmp; + } + for( k = 0; k <= n; k++ ) { + CAf[ k ] -= tmp1 * x_ptr[ n - k ]; + CAb[ k ] -= tmp2 * x_ptr[ subfr_length - n + k - 1 ]; + } + } + tmp1 = C_first_row[ n ]; + tmp2 = C_last_row[ n ]; + for( k = 0; k < n; k++ ) { + Atmp = Af[ k ]; + tmp1 += C_last_row[ n - k - 1 ] * Atmp; + tmp2 += C_first_row[ n - k - 1 ] * Atmp; + } + CAf[ n + 1 ] = tmp1; + CAb[ n + 1 ] = tmp2; + + /* Calculate nominator and denominator for the next order reflection (parcor) coefficient */ + num = CAb[ n + 1 ]; + nrg_b = CAb[ 0 ]; + nrg_f = CAf[ 0 ]; + for( k = 0; k < n; k++ ) { + Atmp = Af[ k ]; + num += CAb[ n - k ] * Atmp; + nrg_b += CAb[ k + 1 ] * Atmp; + nrg_f += CAf[ k + 1 ] * Atmp; + } + silk_assert( nrg_f > 0.0 ); + silk_assert( nrg_b > 0.0 ); + + /* Calculate the next order reflection (parcor) coefficient */ + rc = -2.0 * num / ( nrg_f + nrg_b ); + silk_assert( rc > -1.0 && rc < 1.0 ); + + /* Update inverse prediction gain */ + tmp1 = invGain * ( 1.0 - rc * rc ); + if( tmp1 <= minInvGain ) { + /* Max prediction gain exceeded; set reflection coefficient such that max prediction gain is exactly hit */ + rc = sqrt( 1.0 - minInvGain / invGain ); + if( num > 0 ) { + /* Ensure adjusted reflection coefficients has the original sign */ + rc = -rc; + } + invGain = minInvGain; + reached_max_gain = 1; + } else { + invGain = tmp1; + } + + /* Update the AR coefficients */ + for( k = 0; k < (n + 1) >> 1; k++ ) { + tmp1 = Af[ k ]; + tmp2 = Af[ n - k - 1 ]; + Af[ k ] = tmp1 + rc * tmp2; + Af[ n - k - 1 ] = tmp2 + rc * tmp1; + } + Af[ n ] = rc; + + if( reached_max_gain ) { + /* Reached max prediction gain; set remaining coefficients to zero and exit loop */ + for( k = n + 1; k < D; k++ ) { + Af[ k ] = 0.0; + } + break; + } + + /* Update C * Af and C * Ab */ + for( k = 0; k <= n + 1; k++ ) { + tmp1 = CAf[ k ]; + CAf[ k ] += rc * CAb[ n - k + 1 ]; + CAb[ n - k + 1 ] += rc * tmp1; + } + } + + if( reached_max_gain ) { + /* Convert to silk_float */ + for( k = 0; k < D; k++ ) { + A[ k ] = (silk_float)( -Af[ k ] ); + } + /* Subtract energy of preceding samples from C0 */ + for( s = 0; s < nb_subfr; s++ ) { + C0 -= silk_energy_FLP( x + s * subfr_length, D ); + } + /* Approximate residual energy */ + nrg_f = C0 * invGain; + } else { + /* Compute residual energy and store coefficients as silk_float */ + nrg_f = CAf[ 0 ]; + tmp1 = 1.0; + for( k = 0; k < D; k++ ) { + Atmp = Af[ k ]; + nrg_f += CAf[ k + 1 ] * Atmp; + tmp1 += Atmp * Atmp; + A[ k ] = (silk_float)(-Atmp); + } + nrg_f -= FIND_LPC_COND_FAC * C0 * tmp1; + } + + /* Return residual energy */ + return (silk_float)nrg_f; +} diff --git a/media/libopus/silk/float/bwexpander_FLP.c b/media/libopus/silk/float/bwexpander_FLP.c new file mode 100644 index 0000000000..d55a4d79ab --- /dev/null +++ b/media/libopus/silk/float/bwexpander_FLP.c @@ -0,0 +1,49 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "SigProc_FLP.h" + +/* Chirp (bw expand) LP AR filter */ +void silk_bwexpander_FLP( + silk_float *ar, /* I/O AR filter to be expanded (without leading 1) */ + const opus_int d, /* I length of ar */ + const silk_float chirp /* I chirp factor (typically in range (0..1) ) */ +) +{ + opus_int i; + silk_float cfac = chirp; + + for( i = 0; i < d - 1; i++ ) { + ar[ i ] *= cfac; + cfac *= chirp; + } + ar[ d - 1 ] *= cfac; +} diff --git a/media/libopus/silk/float/corrMatrix_FLP.c b/media/libopus/silk/float/corrMatrix_FLP.c new file mode 100644 index 0000000000..eae6a1cfca --- /dev/null +++ b/media/libopus/silk/float/corrMatrix_FLP.c @@ -0,0 +1,93 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +/********************************************************************** + * Correlation matrix computations for LS estimate. + **********************************************************************/ + +#include "main_FLP.h" + +/* Calculates correlation vector X'*t */ +void silk_corrVector_FLP( + const silk_float *x, /* I x vector [L+order-1] used to create X */ + const silk_float *t, /* I Target vector [L] */ + const opus_int L, /* I Length of vecors */ + const opus_int Order, /* I Max lag for correlation */ + silk_float *Xt /* O X'*t correlation vector [order] */ +) +{ + opus_int lag; + const silk_float *ptr1; + + ptr1 = &x[ Order - 1 ]; /* Points to first sample of column 0 of X: X[:,0] */ + for( lag = 0; lag < Order; lag++ ) { + /* Calculate X[:,lag]'*t */ + Xt[ lag ] = (silk_float)silk_inner_product_FLP( ptr1, t, L ); + ptr1--; /* Next column of X */ + } +} + +/* Calculates correlation matrix X'*X */ +void silk_corrMatrix_FLP( + const silk_float *x, /* I x vector [ L+order-1 ] used to create X */ + const opus_int L, /* I Length of vectors */ + const opus_int Order, /* I Max lag for correlation */ + silk_float *XX /* O X'*X correlation matrix [order x order] */ +) +{ + opus_int j, lag; + double energy; + const silk_float *ptr1, *ptr2; + + ptr1 = &x[ Order - 1 ]; /* First sample of column 0 of X */ + energy = silk_energy_FLP( ptr1, L ); /* X[:,0]'*X[:,0] */ + matrix_ptr( XX, 0, 0, Order ) = ( silk_float )energy; + for( j = 1; j < Order; j++ ) { + /* Calculate X[:,j]'*X[:,j] */ + energy += ptr1[ -j ] * ptr1[ -j ] - ptr1[ L - j ] * ptr1[ L - j ]; + matrix_ptr( XX, j, j, Order ) = ( silk_float )energy; + } + + ptr2 = &x[ Order - 2 ]; /* First sample of column 1 of X */ + for( lag = 1; lag < Order; lag++ ) { + /* Calculate X[:,0]'*X[:,lag] */ + energy = silk_inner_product_FLP( ptr1, ptr2, L ); + matrix_ptr( XX, lag, 0, Order ) = ( silk_float )energy; + matrix_ptr( XX, 0, lag, Order ) = ( silk_float )energy; + /* Calculate X[:,j]'*X[:,j + lag] */ + for( j = 1; j < ( Order - lag ); j++ ) { + energy += ptr1[ -j ] * ptr2[ -j ] - ptr1[ L - j ] * ptr2[ L - j ]; + matrix_ptr( XX, lag + j, j, Order ) = ( silk_float )energy; + matrix_ptr( XX, j, lag + j, Order ) = ( silk_float )energy; + } + ptr2--; /* Next column of X */ + } +} diff --git a/media/libopus/silk/float/encode_frame_FLP.c b/media/libopus/silk/float/encode_frame_FLP.c new file mode 100644 index 0000000000..b029c3f5ca --- /dev/null +++ b/media/libopus/silk/float/encode_frame_FLP.c @@ -0,0 +1,435 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include <stdlib.h> +#include "main_FLP.h" +#include "tuning_parameters.h" + +/* Low Bitrate Redundancy (LBRR) encoding. Reuse all parameters but encode with lower bitrate */ +static OPUS_INLINE void silk_LBRR_encode_FLP( + silk_encoder_state_FLP *psEnc, /* I/O Encoder state FLP */ + silk_encoder_control_FLP *psEncCtrl, /* I/O Encoder control FLP */ + const silk_float xfw[], /* I Input signal */ + opus_int condCoding /* I The type of conditional coding used so far for this frame */ +); + +void silk_encode_do_VAD_FLP( + silk_encoder_state_FLP *psEnc, /* I/O Encoder state FLP */ + opus_int activity /* I Decision of Opus voice activity detector */ +) +{ + const opus_int activity_threshold = SILK_FIX_CONST( SPEECH_ACTIVITY_DTX_THRES, 8 ); + + /****************************/ + /* Voice Activity Detection */ + /****************************/ + silk_VAD_GetSA_Q8( &psEnc->sCmn, psEnc->sCmn.inputBuf + 1, psEnc->sCmn.arch ); + /* If Opus VAD is inactive and Silk VAD is active: lower Silk VAD to just under the threshold */ + if( activity == VAD_NO_ACTIVITY && psEnc->sCmn.speech_activity_Q8 >= activity_threshold ) { + psEnc->sCmn.speech_activity_Q8 = activity_threshold - 1; + } + + /**************************************************/ + /* Convert speech activity into VAD and DTX flags */ + /**************************************************/ + if( psEnc->sCmn.speech_activity_Q8 < activity_threshold ) { + psEnc->sCmn.indices.signalType = TYPE_NO_VOICE_ACTIVITY; + psEnc->sCmn.noSpeechCounter++; + if( psEnc->sCmn.noSpeechCounter <= NB_SPEECH_FRAMES_BEFORE_DTX ) { + psEnc->sCmn.inDTX = 0; + } else if( psEnc->sCmn.noSpeechCounter > MAX_CONSECUTIVE_DTX + NB_SPEECH_FRAMES_BEFORE_DTX ) { + psEnc->sCmn.noSpeechCounter = NB_SPEECH_FRAMES_BEFORE_DTX; + psEnc->sCmn.inDTX = 0; + } + psEnc->sCmn.VAD_flags[ psEnc->sCmn.nFramesEncoded ] = 0; + } else { + psEnc->sCmn.noSpeechCounter = 0; + psEnc->sCmn.inDTX = 0; + psEnc->sCmn.indices.signalType = TYPE_UNVOICED; + psEnc->sCmn.VAD_flags[ psEnc->sCmn.nFramesEncoded ] = 1; + } +} + +/****************/ +/* Encode frame */ +/****************/ +opus_int silk_encode_frame_FLP( + silk_encoder_state_FLP *psEnc, /* I/O Encoder state FLP */ + opus_int32 *pnBytesOut, /* O Number of payload bytes; */ + ec_enc *psRangeEnc, /* I/O compressor data structure */ + opus_int condCoding, /* I The type of conditional coding to use */ + opus_int maxBits, /* I If > 0: maximum number of output bits */ + opus_int useCBR /* I Flag to force constant-bitrate operation */ +) +{ + silk_encoder_control_FLP sEncCtrl; + opus_int i, iter, maxIter, found_upper, found_lower, ret = 0; + silk_float *x_frame, *res_pitch_frame; + silk_float res_pitch[ 2 * MAX_FRAME_LENGTH + LA_PITCH_MAX ]; + ec_enc sRangeEnc_copy, sRangeEnc_copy2; + silk_nsq_state sNSQ_copy, sNSQ_copy2; + opus_int32 seed_copy, nBits, nBits_lower, nBits_upper, gainMult_lower, gainMult_upper; + opus_int32 gainsID, gainsID_lower, gainsID_upper; + opus_int16 gainMult_Q8; + opus_int16 ec_prevLagIndex_copy; + opus_int ec_prevSignalType_copy; + opus_int8 LastGainIndex_copy2; + opus_int32 pGains_Q16[ MAX_NB_SUBFR ]; + opus_uint8 ec_buf_copy[ 1275 ]; + opus_int gain_lock[ MAX_NB_SUBFR ] = {0}; + opus_int16 best_gain_mult[ MAX_NB_SUBFR ]; + opus_int best_sum[ MAX_NB_SUBFR ]; + + /* This is totally unnecessary but many compilers (including gcc) are too dumb to realise it */ + LastGainIndex_copy2 = nBits_lower = nBits_upper = gainMult_lower = gainMult_upper = 0; + + psEnc->sCmn.indices.Seed = psEnc->sCmn.frameCounter++ & 3; + + /**************************************************************/ + /* Set up Input Pointers, and insert frame in input buffer */ + /**************************************************************/ + /* pointers aligned with start of frame to encode */ + x_frame = psEnc->x_buf + psEnc->sCmn.ltp_mem_length; /* start of frame to encode */ + res_pitch_frame = res_pitch + psEnc->sCmn.ltp_mem_length; /* start of pitch LPC residual frame */ + + /***************************************/ + /* Ensure smooth bandwidth transitions */ + /***************************************/ + silk_LP_variable_cutoff( &psEnc->sCmn.sLP, psEnc->sCmn.inputBuf + 1, psEnc->sCmn.frame_length ); + + /*******************************************/ + /* Copy new frame to front of input buffer */ + /*******************************************/ + silk_short2float_array( x_frame + LA_SHAPE_MS * psEnc->sCmn.fs_kHz, psEnc->sCmn.inputBuf + 1, psEnc->sCmn.frame_length ); + + /* Add tiny signal to avoid high CPU load from denormalized floating point numbers */ + for( i = 0; i < 8; i++ ) { + x_frame[ LA_SHAPE_MS * psEnc->sCmn.fs_kHz + i * ( psEnc->sCmn.frame_length >> 3 ) ] += ( 1 - ( i & 2 ) ) * 1e-6f; + } + + if( !psEnc->sCmn.prefillFlag ) { + /*****************************************/ + /* Find pitch lags, initial LPC analysis */ + /*****************************************/ + silk_find_pitch_lags_FLP( psEnc, &sEncCtrl, res_pitch, x_frame, psEnc->sCmn.arch ); + + /************************/ + /* Noise shape analysis */ + /************************/ + silk_noise_shape_analysis_FLP( psEnc, &sEncCtrl, res_pitch_frame, x_frame ); + + /***************************************************/ + /* Find linear prediction coefficients (LPC + LTP) */ + /***************************************************/ + silk_find_pred_coefs_FLP( psEnc, &sEncCtrl, res_pitch_frame, x_frame, condCoding ); + + /****************************************/ + /* Process gains */ + /****************************************/ + silk_process_gains_FLP( psEnc, &sEncCtrl, condCoding ); + + /****************************************/ + /* Low Bitrate Redundant Encoding */ + /****************************************/ + silk_LBRR_encode_FLP( psEnc, &sEncCtrl, x_frame, condCoding ); + + /* Loop over quantizer and entroy coding to control bitrate */ + maxIter = 6; + gainMult_Q8 = SILK_FIX_CONST( 1, 8 ); + found_lower = 0; + found_upper = 0; + gainsID = silk_gains_ID( psEnc->sCmn.indices.GainsIndices, psEnc->sCmn.nb_subfr ); + gainsID_lower = -1; + gainsID_upper = -1; + /* Copy part of the input state */ + silk_memcpy( &sRangeEnc_copy, psRangeEnc, sizeof( ec_enc ) ); + silk_memcpy( &sNSQ_copy, &psEnc->sCmn.sNSQ, sizeof( silk_nsq_state ) ); + seed_copy = psEnc->sCmn.indices.Seed; + ec_prevLagIndex_copy = psEnc->sCmn.ec_prevLagIndex; + ec_prevSignalType_copy = psEnc->sCmn.ec_prevSignalType; + for( iter = 0; ; iter++ ) { + if( gainsID == gainsID_lower ) { + nBits = nBits_lower; + } else if( gainsID == gainsID_upper ) { + nBits = nBits_upper; + } else { + /* Restore part of the input state */ + if( iter > 0 ) { + silk_memcpy( psRangeEnc, &sRangeEnc_copy, sizeof( ec_enc ) ); + silk_memcpy( &psEnc->sCmn.sNSQ, &sNSQ_copy, sizeof( silk_nsq_state ) ); + psEnc->sCmn.indices.Seed = seed_copy; + psEnc->sCmn.ec_prevLagIndex = ec_prevLagIndex_copy; + psEnc->sCmn.ec_prevSignalType = ec_prevSignalType_copy; + } + + /*****************************************/ + /* Noise shaping quantization */ + /*****************************************/ + silk_NSQ_wrapper_FLP( psEnc, &sEncCtrl, &psEnc->sCmn.indices, &psEnc->sCmn.sNSQ, psEnc->sCmn.pulses, x_frame ); + + if ( iter == maxIter && !found_lower ) { + silk_memcpy( &sRangeEnc_copy2, psRangeEnc, sizeof( ec_enc ) ); + } + + /****************************************/ + /* Encode Parameters */ + /****************************************/ + silk_encode_indices( &psEnc->sCmn, psRangeEnc, psEnc->sCmn.nFramesEncoded, 0, condCoding ); + + /****************************************/ + /* Encode Excitation Signal */ + /****************************************/ + silk_encode_pulses( psRangeEnc, psEnc->sCmn.indices.signalType, psEnc->sCmn.indices.quantOffsetType, + psEnc->sCmn.pulses, psEnc->sCmn.frame_length ); + + nBits = ec_tell( psRangeEnc ); + + /* If we still bust after the last iteration, do some damage control. */ + if ( iter == maxIter && !found_lower && nBits > maxBits ) { + silk_memcpy( psRangeEnc, &sRangeEnc_copy2, sizeof( ec_enc ) ); + + /* Keep gains the same as the last frame. */ + psEnc->sShape.LastGainIndex = sEncCtrl.lastGainIndexPrev; + for ( i = 0; i < psEnc->sCmn.nb_subfr; i++ ) { + psEnc->sCmn.indices.GainsIndices[ i ] = 4; + } + if (condCoding != CODE_CONDITIONALLY) { + psEnc->sCmn.indices.GainsIndices[ 0 ] = sEncCtrl.lastGainIndexPrev; + } + psEnc->sCmn.ec_prevLagIndex = ec_prevLagIndex_copy; + psEnc->sCmn.ec_prevSignalType = ec_prevSignalType_copy; + /* Clear all pulses. */ + for ( i = 0; i < psEnc->sCmn.frame_length; i++ ) { + psEnc->sCmn.pulses[ i ] = 0; + } + + silk_encode_indices( &psEnc->sCmn, psRangeEnc, psEnc->sCmn.nFramesEncoded, 0, condCoding ); + + silk_encode_pulses( psRangeEnc, psEnc->sCmn.indices.signalType, psEnc->sCmn.indices.quantOffsetType, + psEnc->sCmn.pulses, psEnc->sCmn.frame_length ); + + nBits = ec_tell( psRangeEnc ); + } + + if( useCBR == 0 && iter == 0 && nBits <= maxBits ) { + break; + } + } + + if( iter == maxIter ) { + if( found_lower && ( gainsID == gainsID_lower || nBits > maxBits ) ) { + /* Restore output state from earlier iteration that did meet the bitrate budget */ + silk_memcpy( psRangeEnc, &sRangeEnc_copy2, sizeof( ec_enc ) ); + celt_assert( sRangeEnc_copy2.offs <= 1275 ); + silk_memcpy( psRangeEnc->buf, ec_buf_copy, sRangeEnc_copy2.offs ); + silk_memcpy( &psEnc->sCmn.sNSQ, &sNSQ_copy2, sizeof( silk_nsq_state ) ); + psEnc->sShape.LastGainIndex = LastGainIndex_copy2; + } + break; + } + + if( nBits > maxBits ) { + if( found_lower == 0 && iter >= 2 ) { + /* Adjust the quantizer's rate/distortion tradeoff and discard previous "upper" results */ + sEncCtrl.Lambda = silk_max_float(sEncCtrl.Lambda*1.5f, 1.5f); + /* Reducing dithering can help us hit the target. */ + psEnc->sCmn.indices.quantOffsetType = 0; + found_upper = 0; + gainsID_upper = -1; + } else { + found_upper = 1; + nBits_upper = nBits; + gainMult_upper = gainMult_Q8; + gainsID_upper = gainsID; + } + } else if( nBits < maxBits - 5 ) { + found_lower = 1; + nBits_lower = nBits; + gainMult_lower = gainMult_Q8; + if( gainsID != gainsID_lower ) { + gainsID_lower = gainsID; + /* Copy part of the output state */ + silk_memcpy( &sRangeEnc_copy2, psRangeEnc, sizeof( ec_enc ) ); + celt_assert( psRangeEnc->offs <= 1275 ); + silk_memcpy( ec_buf_copy, psRangeEnc->buf, psRangeEnc->offs ); + silk_memcpy( &sNSQ_copy2, &psEnc->sCmn.sNSQ, sizeof( silk_nsq_state ) ); + LastGainIndex_copy2 = psEnc->sShape.LastGainIndex; + } + } else { + /* Within 5 bits of budget: close enough */ + break; + } + + if ( !found_lower && nBits > maxBits ) { + int j; + for ( i = 0; i < psEnc->sCmn.nb_subfr; i++ ) { + int sum=0; + for ( j = i*psEnc->sCmn.subfr_length; j < (i+1)*psEnc->sCmn.subfr_length; j++ ) { + sum += abs( psEnc->sCmn.pulses[j] ); + } + if ( iter == 0 || (sum < best_sum[i] && !gain_lock[i]) ) { + best_sum[i] = sum; + best_gain_mult[i] = gainMult_Q8; + } else { + gain_lock[i] = 1; + } + } + } + if( ( found_lower & found_upper ) == 0 ) { + /* Adjust gain according to high-rate rate/distortion curve */ + if( nBits > maxBits ) { + if (gainMult_Q8 < 16384) { + gainMult_Q8 *= 2; + } else { + gainMult_Q8 = 32767; + } + } else { + opus_int32 gain_factor_Q16; + gain_factor_Q16 = silk_log2lin( silk_LSHIFT( nBits - maxBits, 7 ) / psEnc->sCmn.frame_length + SILK_FIX_CONST( 16, 7 ) ); + gainMult_Q8 = silk_SMULWB( gain_factor_Q16, gainMult_Q8 ); + } + } else { + /* Adjust gain by interpolating */ + gainMult_Q8 = gainMult_lower + ( ( gainMult_upper - gainMult_lower ) * ( maxBits - nBits_lower ) ) / ( nBits_upper - nBits_lower ); + /* New gain multplier must be between 25% and 75% of old range (note that gainMult_upper < gainMult_lower) */ + if( gainMult_Q8 > silk_ADD_RSHIFT32( gainMult_lower, gainMult_upper - gainMult_lower, 2 ) ) { + gainMult_Q8 = silk_ADD_RSHIFT32( gainMult_lower, gainMult_upper - gainMult_lower, 2 ); + } else + if( gainMult_Q8 < silk_SUB_RSHIFT32( gainMult_upper, gainMult_upper - gainMult_lower, 2 ) ) { + gainMult_Q8 = silk_SUB_RSHIFT32( gainMult_upper, gainMult_upper - gainMult_lower, 2 ); + } + } + + for( i = 0; i < psEnc->sCmn.nb_subfr; i++ ) { + opus_int16 tmp; + if ( gain_lock[i] ) { + tmp = best_gain_mult[i]; + } else { + tmp = gainMult_Q8; + } + pGains_Q16[ i ] = silk_LSHIFT_SAT32( silk_SMULWB( sEncCtrl.GainsUnq_Q16[ i ], tmp ), 8 ); + } + + /* Quantize gains */ + psEnc->sShape.LastGainIndex = sEncCtrl.lastGainIndexPrev; + silk_gains_quant( psEnc->sCmn.indices.GainsIndices, pGains_Q16, + &psEnc->sShape.LastGainIndex, condCoding == CODE_CONDITIONALLY, psEnc->sCmn.nb_subfr ); + + /* Unique identifier of gains vector */ + gainsID = silk_gains_ID( psEnc->sCmn.indices.GainsIndices, psEnc->sCmn.nb_subfr ); + + /* Overwrite unquantized gains with quantized gains and convert back to Q0 from Q16 */ + for( i = 0; i < psEnc->sCmn.nb_subfr; i++ ) { + sEncCtrl.Gains[ i ] = pGains_Q16[ i ] / 65536.0f; + } + } + } + + /* Update input buffer */ + silk_memmove( psEnc->x_buf, &psEnc->x_buf[ psEnc->sCmn.frame_length ], + ( psEnc->sCmn.ltp_mem_length + LA_SHAPE_MS * psEnc->sCmn.fs_kHz ) * sizeof( silk_float ) ); + + /* Exit without entropy coding */ + if( psEnc->sCmn.prefillFlag ) { + /* No payload */ + *pnBytesOut = 0; + return ret; + } + + /* Parameters needed for next frame */ + psEnc->sCmn.prevLag = sEncCtrl.pitchL[ psEnc->sCmn.nb_subfr - 1 ]; + psEnc->sCmn.prevSignalType = psEnc->sCmn.indices.signalType; + + /****************************************/ + /* Finalize payload */ + /****************************************/ + psEnc->sCmn.first_frame_after_reset = 0; + /* Payload size */ + *pnBytesOut = silk_RSHIFT( ec_tell( psRangeEnc ) + 7, 3 ); + + return ret; +} + +/* Low-Bitrate Redundancy (LBRR) encoding. Reuse all parameters but encode excitation at lower bitrate */ +static OPUS_INLINE void silk_LBRR_encode_FLP( + silk_encoder_state_FLP *psEnc, /* I/O Encoder state FLP */ + silk_encoder_control_FLP *psEncCtrl, /* I/O Encoder control FLP */ + const silk_float xfw[], /* I Input signal */ + opus_int condCoding /* I The type of conditional coding used so far for this frame */ +) +{ + opus_int k; + opus_int32 Gains_Q16[ MAX_NB_SUBFR ]; + silk_float TempGains[ MAX_NB_SUBFR ]; + SideInfoIndices *psIndices_LBRR = &psEnc->sCmn.indices_LBRR[ psEnc->sCmn.nFramesEncoded ]; + silk_nsq_state sNSQ_LBRR; + + /*******************************************/ + /* Control use of inband LBRR */ + /*******************************************/ + if( psEnc->sCmn.LBRR_enabled && psEnc->sCmn.speech_activity_Q8 > SILK_FIX_CONST( LBRR_SPEECH_ACTIVITY_THRES, 8 ) ) { + psEnc->sCmn.LBRR_flags[ psEnc->sCmn.nFramesEncoded ] = 1; + + /* Copy noise shaping quantizer state and quantization indices from regular encoding */ + silk_memcpy( &sNSQ_LBRR, &psEnc->sCmn.sNSQ, sizeof( silk_nsq_state ) ); + silk_memcpy( psIndices_LBRR, &psEnc->sCmn.indices, sizeof( SideInfoIndices ) ); + + /* Save original gains */ + silk_memcpy( TempGains, psEncCtrl->Gains, psEnc->sCmn.nb_subfr * sizeof( silk_float ) ); + + if( psEnc->sCmn.nFramesEncoded == 0 || psEnc->sCmn.LBRR_flags[ psEnc->sCmn.nFramesEncoded - 1 ] == 0 ) { + /* First frame in packet or previous frame not LBRR coded */ + psEnc->sCmn.LBRRprevLastGainIndex = psEnc->sShape.LastGainIndex; + + /* Increase Gains to get target LBRR rate */ + psIndices_LBRR->GainsIndices[ 0 ] += psEnc->sCmn.LBRR_GainIncreases; + psIndices_LBRR->GainsIndices[ 0 ] = silk_min_int( psIndices_LBRR->GainsIndices[ 0 ], N_LEVELS_QGAIN - 1 ); + } + + /* Decode to get gains in sync with decoder */ + silk_gains_dequant( Gains_Q16, psIndices_LBRR->GainsIndices, + &psEnc->sCmn.LBRRprevLastGainIndex, condCoding == CODE_CONDITIONALLY, psEnc->sCmn.nb_subfr ); + + /* Overwrite unquantized gains with quantized gains and convert back to Q0 from Q16 */ + for( k = 0; k < psEnc->sCmn.nb_subfr; k++ ) { + psEncCtrl->Gains[ k ] = Gains_Q16[ k ] * ( 1.0f / 65536.0f ); + } + + /*****************************************/ + /* Noise shaping quantization */ + /*****************************************/ + silk_NSQ_wrapper_FLP( psEnc, psEncCtrl, psIndices_LBRR, &sNSQ_LBRR, + psEnc->sCmn.pulses_LBRR[ psEnc->sCmn.nFramesEncoded ], xfw ); + + /* Restore original gains */ + silk_memcpy( psEncCtrl->Gains, TempGains, psEnc->sCmn.nb_subfr * sizeof( silk_float ) ); + } +} diff --git a/media/libopus/silk/float/energy_FLP.c b/media/libopus/silk/float/energy_FLP.c new file mode 100644 index 0000000000..7bc7173c9c --- /dev/null +++ b/media/libopus/silk/float/energy_FLP.c @@ -0,0 +1,59 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "SigProc_FLP.h" + +/* sum of squares of a silk_float array, with result as double */ +double silk_energy_FLP( + const silk_float *data, + opus_int dataSize +) +{ + opus_int i; + double result; + + /* 4x unrolled loop */ + result = 0.0; + for( i = 0; i < dataSize - 3; i += 4 ) { + result += data[ i + 0 ] * (double)data[ i + 0 ] + + data[ i + 1 ] * (double)data[ i + 1 ] + + data[ i + 2 ] * (double)data[ i + 2 ] + + data[ i + 3 ] * (double)data[ i + 3 ]; + } + + /* add any remaining products */ + for( ; i < dataSize; i++ ) { + result += data[ i ] * (double)data[ i ]; + } + + silk_assert( result >= 0.0 ); + return result; +} diff --git a/media/libopus/silk/float/find_LPC_FLP.c b/media/libopus/silk/float/find_LPC_FLP.c new file mode 100644 index 0000000000..fa3ffe7f8b --- /dev/null +++ b/media/libopus/silk/float/find_LPC_FLP.c @@ -0,0 +1,104 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "define.h" +#include "main_FLP.h" +#include "tuning_parameters.h" + +/* LPC analysis */ +void silk_find_LPC_FLP( + silk_encoder_state *psEncC, /* I/O Encoder state */ + opus_int16 NLSF_Q15[], /* O NLSFs */ + const silk_float x[], /* I Input signal */ + const silk_float minInvGain /* I Inverse of max prediction gain */ +) +{ + opus_int k, subfr_length; + silk_float a[ MAX_LPC_ORDER ]; + + /* Used only for NLSF interpolation */ + silk_float res_nrg, res_nrg_2nd, res_nrg_interp; + opus_int16 NLSF0_Q15[ MAX_LPC_ORDER ]; + silk_float a_tmp[ MAX_LPC_ORDER ]; + silk_float LPC_res[ MAX_FRAME_LENGTH + MAX_NB_SUBFR * MAX_LPC_ORDER ]; + + subfr_length = psEncC->subfr_length + psEncC->predictLPCOrder; + + /* Default: No interpolation */ + psEncC->indices.NLSFInterpCoef_Q2 = 4; + + /* Burg AR analysis for the full frame */ + res_nrg = silk_burg_modified_FLP( a, x, minInvGain, subfr_length, psEncC->nb_subfr, psEncC->predictLPCOrder ); + + if( psEncC->useInterpolatedNLSFs && !psEncC->first_frame_after_reset && psEncC->nb_subfr == MAX_NB_SUBFR ) { + /* Optimal solution for last 10 ms; subtract residual energy here, as that's easier than */ + /* adding it to the residual energy of the first 10 ms in each iteration of the search below */ + res_nrg -= silk_burg_modified_FLP( a_tmp, x + ( MAX_NB_SUBFR / 2 ) * subfr_length, minInvGain, subfr_length, MAX_NB_SUBFR / 2, psEncC->predictLPCOrder ); + + /* Convert to NLSFs */ + silk_A2NLSF_FLP( NLSF_Q15, a_tmp, psEncC->predictLPCOrder ); + + /* Search over interpolation indices to find the one with lowest residual energy */ + res_nrg_2nd = silk_float_MAX; + for( k = 3; k >= 0; k-- ) { + /* Interpolate NLSFs for first half */ + silk_interpolate( NLSF0_Q15, psEncC->prev_NLSFq_Q15, NLSF_Q15, k, psEncC->predictLPCOrder ); + + /* Convert to LPC for residual energy evaluation */ + silk_NLSF2A_FLP( a_tmp, NLSF0_Q15, psEncC->predictLPCOrder, psEncC->arch ); + + /* Calculate residual energy with LSF interpolation */ + silk_LPC_analysis_filter_FLP( LPC_res, a_tmp, x, 2 * subfr_length, psEncC->predictLPCOrder ); + res_nrg_interp = (silk_float)( + silk_energy_FLP( LPC_res + psEncC->predictLPCOrder, subfr_length - psEncC->predictLPCOrder ) + + silk_energy_FLP( LPC_res + psEncC->predictLPCOrder + subfr_length, subfr_length - psEncC->predictLPCOrder ) ); + + /* Determine whether current interpolated NLSFs are best so far */ + if( res_nrg_interp < res_nrg ) { + /* Interpolation has lower residual energy */ + res_nrg = res_nrg_interp; + psEncC->indices.NLSFInterpCoef_Q2 = (opus_int8)k; + } else if( res_nrg_interp > res_nrg_2nd ) { + /* No reason to continue iterating - residual energies will continue to climb */ + break; + } + res_nrg_2nd = res_nrg_interp; + } + } + + if( psEncC->indices.NLSFInterpCoef_Q2 == 4 ) { + /* NLSF interpolation is currently inactive, calculate NLSFs from full frame AR coefficients */ + silk_A2NLSF_FLP( NLSF_Q15, a, psEncC->predictLPCOrder ); + } + + celt_assert( psEncC->indices.NLSFInterpCoef_Q2 == 4 || + ( psEncC->useInterpolatedNLSFs && !psEncC->first_frame_after_reset && psEncC->nb_subfr == MAX_NB_SUBFR ) ); +} diff --git a/media/libopus/silk/float/find_LTP_FLP.c b/media/libopus/silk/float/find_LTP_FLP.c new file mode 100644 index 0000000000..f97064930e --- /dev/null +++ b/media/libopus/silk/float/find_LTP_FLP.c @@ -0,0 +1,64 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main_FLP.h" +#include "tuning_parameters.h" + +void silk_find_LTP_FLP( + silk_float XX[ MAX_NB_SUBFR * LTP_ORDER * LTP_ORDER ], /* O Weight for LTP quantization */ + silk_float xX[ MAX_NB_SUBFR * LTP_ORDER ], /* O Weight for LTP quantization */ + const silk_float r_ptr[], /* I LPC residual */ + const opus_int lag[ MAX_NB_SUBFR ], /* I LTP lags */ + const opus_int subfr_length, /* I Subframe length */ + const opus_int nb_subfr /* I number of subframes */ +) +{ + opus_int k; + silk_float *xX_ptr, *XX_ptr; + const silk_float *lag_ptr; + silk_float xx, temp; + + xX_ptr = xX; + XX_ptr = XX; + for( k = 0; k < nb_subfr; k++ ) { + lag_ptr = r_ptr - ( lag[ k ] + LTP_ORDER / 2 ); + silk_corrMatrix_FLP( lag_ptr, subfr_length, LTP_ORDER, XX_ptr ); + silk_corrVector_FLP( lag_ptr, r_ptr, subfr_length, LTP_ORDER, xX_ptr ); + xx = ( silk_float )silk_energy_FLP( r_ptr, subfr_length + LTP_ORDER ); + temp = 1.0f / silk_max( xx, LTP_CORR_INV_MAX * 0.5f * ( XX_ptr[ 0 ] + XX_ptr[ 24 ] ) + 1.0f ); + silk_scale_vector_FLP( XX_ptr, temp, LTP_ORDER * LTP_ORDER ); + silk_scale_vector_FLP( xX_ptr, temp, LTP_ORDER ); + + r_ptr += subfr_length; + XX_ptr += LTP_ORDER * LTP_ORDER; + xX_ptr += LTP_ORDER; + } +} diff --git a/media/libopus/silk/float/find_pitch_lags_FLP.c b/media/libopus/silk/float/find_pitch_lags_FLP.c new file mode 100644 index 0000000000..dedbcd2836 --- /dev/null +++ b/media/libopus/silk/float/find_pitch_lags_FLP.c @@ -0,0 +1,132 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include <stdlib.h> +#include "main_FLP.h" +#include "tuning_parameters.h" + +void silk_find_pitch_lags_FLP( + silk_encoder_state_FLP *psEnc, /* I/O Encoder state FLP */ + silk_encoder_control_FLP *psEncCtrl, /* I/O Encoder control FLP */ + silk_float res[], /* O Residual */ + const silk_float x[], /* I Speech signal */ + int arch /* I Run-time architecture */ +) +{ + opus_int buf_len; + silk_float thrhld, res_nrg; + const silk_float *x_buf_ptr, *x_buf; + silk_float auto_corr[ MAX_FIND_PITCH_LPC_ORDER + 1 ]; + silk_float A[ MAX_FIND_PITCH_LPC_ORDER ]; + silk_float refl_coef[ MAX_FIND_PITCH_LPC_ORDER ]; + silk_float Wsig[ FIND_PITCH_LPC_WIN_MAX ]; + silk_float *Wsig_ptr; + + /******************************************/ + /* Set up buffer lengths etc based on Fs */ + /******************************************/ + buf_len = psEnc->sCmn.la_pitch + psEnc->sCmn.frame_length + psEnc->sCmn.ltp_mem_length; + + /* Safety check */ + celt_assert( buf_len >= psEnc->sCmn.pitch_LPC_win_length ); + + x_buf = x - psEnc->sCmn.ltp_mem_length; + + /******************************************/ + /* Estimate LPC AR coeficients */ + /******************************************/ + + /* Calculate windowed signal */ + + /* First LA_LTP samples */ + x_buf_ptr = x_buf + buf_len - psEnc->sCmn.pitch_LPC_win_length; + Wsig_ptr = Wsig; + silk_apply_sine_window_FLP( Wsig_ptr, x_buf_ptr, 1, psEnc->sCmn.la_pitch ); + + /* Middle non-windowed samples */ + Wsig_ptr += psEnc->sCmn.la_pitch; + x_buf_ptr += psEnc->sCmn.la_pitch; + silk_memcpy( Wsig_ptr, x_buf_ptr, ( psEnc->sCmn.pitch_LPC_win_length - ( psEnc->sCmn.la_pitch << 1 ) ) * sizeof( silk_float ) ); + + /* Last LA_LTP samples */ + Wsig_ptr += psEnc->sCmn.pitch_LPC_win_length - ( psEnc->sCmn.la_pitch << 1 ); + x_buf_ptr += psEnc->sCmn.pitch_LPC_win_length - ( psEnc->sCmn.la_pitch << 1 ); + silk_apply_sine_window_FLP( Wsig_ptr, x_buf_ptr, 2, psEnc->sCmn.la_pitch ); + + /* Calculate autocorrelation sequence */ + silk_autocorrelation_FLP( auto_corr, Wsig, psEnc->sCmn.pitch_LPC_win_length, psEnc->sCmn.pitchEstimationLPCOrder + 1 ); + + /* Add white noise, as a fraction of the energy */ + auto_corr[ 0 ] += auto_corr[ 0 ] * FIND_PITCH_WHITE_NOISE_FRACTION + 1; + + /* Calculate the reflection coefficients using Schur */ + res_nrg = silk_schur_FLP( refl_coef, auto_corr, psEnc->sCmn.pitchEstimationLPCOrder ); + + /* Prediction gain */ + psEncCtrl->predGain = auto_corr[ 0 ] / silk_max_float( res_nrg, 1.0f ); + + /* Convert reflection coefficients to prediction coefficients */ + silk_k2a_FLP( A, refl_coef, psEnc->sCmn.pitchEstimationLPCOrder ); + + /* Bandwidth expansion */ + silk_bwexpander_FLP( A, psEnc->sCmn.pitchEstimationLPCOrder, FIND_PITCH_BANDWIDTH_EXPANSION ); + + /*****************************************/ + /* LPC analysis filtering */ + /*****************************************/ + silk_LPC_analysis_filter_FLP( res, A, x_buf, buf_len, psEnc->sCmn.pitchEstimationLPCOrder ); + + if( psEnc->sCmn.indices.signalType != TYPE_NO_VOICE_ACTIVITY && psEnc->sCmn.first_frame_after_reset == 0 ) { + /* Threshold for pitch estimator */ + thrhld = 0.6f; + thrhld -= 0.004f * psEnc->sCmn.pitchEstimationLPCOrder; + thrhld -= 0.1f * psEnc->sCmn.speech_activity_Q8 * ( 1.0f / 256.0f ); + thrhld -= 0.15f * (psEnc->sCmn.prevSignalType >> 1); + thrhld -= 0.1f * psEnc->sCmn.input_tilt_Q15 * ( 1.0f / 32768.0f ); + + /*****************************************/ + /* Call Pitch estimator */ + /*****************************************/ + if( silk_pitch_analysis_core_FLP( res, psEncCtrl->pitchL, &psEnc->sCmn.indices.lagIndex, + &psEnc->sCmn.indices.contourIndex, &psEnc->LTPCorr, psEnc->sCmn.prevLag, psEnc->sCmn.pitchEstimationThreshold_Q16 / 65536.0f, + thrhld, psEnc->sCmn.fs_kHz, psEnc->sCmn.pitchEstimationComplexity, psEnc->sCmn.nb_subfr, arch ) == 0 ) + { + psEnc->sCmn.indices.signalType = TYPE_VOICED; + } else { + psEnc->sCmn.indices.signalType = TYPE_UNVOICED; + } + } else { + silk_memset( psEncCtrl->pitchL, 0, sizeof( psEncCtrl->pitchL ) ); + psEnc->sCmn.indices.lagIndex = 0; + psEnc->sCmn.indices.contourIndex = 0; + psEnc->LTPCorr = 0; + } +} diff --git a/media/libopus/silk/float/find_pred_coefs_FLP.c b/media/libopus/silk/float/find_pred_coefs_FLP.c new file mode 100644 index 0000000000..6f79078893 --- /dev/null +++ b/media/libopus/silk/float/find_pred_coefs_FLP.c @@ -0,0 +1,117 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main_FLP.h" + +/* Find LPC and LTP coefficients */ +void silk_find_pred_coefs_FLP( + silk_encoder_state_FLP *psEnc, /* I/O Encoder state FLP */ + silk_encoder_control_FLP *psEncCtrl, /* I/O Encoder control FLP */ + const silk_float res_pitch[], /* I Residual from pitch analysis */ + const silk_float x[], /* I Speech signal */ + opus_int condCoding /* I The type of conditional coding to use */ +) +{ + opus_int i; + silk_float XXLTP[ MAX_NB_SUBFR * LTP_ORDER * LTP_ORDER ]; + silk_float xXLTP[ MAX_NB_SUBFR * LTP_ORDER ]; + silk_float invGains[ MAX_NB_SUBFR ]; + /* Set to NLSF_Q15 to zero so we don't copy junk to the state. */ + opus_int16 NLSF_Q15[ MAX_LPC_ORDER ]={0}; + const silk_float *x_ptr; + silk_float *x_pre_ptr, LPC_in_pre[ MAX_NB_SUBFR * MAX_LPC_ORDER + MAX_FRAME_LENGTH ]; + silk_float minInvGain; + + /* Weighting for weighted least squares */ + for( i = 0; i < psEnc->sCmn.nb_subfr; i++ ) { + silk_assert( psEncCtrl->Gains[ i ] > 0.0f ); + invGains[ i ] = 1.0f / psEncCtrl->Gains[ i ]; + } + + if( psEnc->sCmn.indices.signalType == TYPE_VOICED ) { + /**********/ + /* VOICED */ + /**********/ + celt_assert( psEnc->sCmn.ltp_mem_length - psEnc->sCmn.predictLPCOrder >= psEncCtrl->pitchL[ 0 ] + LTP_ORDER / 2 ); + + /* LTP analysis */ + silk_find_LTP_FLP( XXLTP, xXLTP, res_pitch, psEncCtrl->pitchL, psEnc->sCmn.subfr_length, psEnc->sCmn.nb_subfr ); + + /* Quantize LTP gain parameters */ + silk_quant_LTP_gains_FLP( psEncCtrl->LTPCoef, psEnc->sCmn.indices.LTPIndex, &psEnc->sCmn.indices.PERIndex, + &psEnc->sCmn.sum_log_gain_Q7, &psEncCtrl->LTPredCodGain, XXLTP, xXLTP, psEnc->sCmn.subfr_length, psEnc->sCmn.nb_subfr, psEnc->sCmn.arch ); + + /* Control LTP scaling */ + silk_LTP_scale_ctrl_FLP( psEnc, psEncCtrl, condCoding ); + + /* Create LTP residual */ + silk_LTP_analysis_filter_FLP( LPC_in_pre, x - psEnc->sCmn.predictLPCOrder, psEncCtrl->LTPCoef, + psEncCtrl->pitchL, invGains, psEnc->sCmn.subfr_length, psEnc->sCmn.nb_subfr, psEnc->sCmn.predictLPCOrder ); + } else { + /************/ + /* UNVOICED */ + /************/ + /* Create signal with prepended subframes, scaled by inverse gains */ + x_ptr = x - psEnc->sCmn.predictLPCOrder; + x_pre_ptr = LPC_in_pre; + for( i = 0; i < psEnc->sCmn.nb_subfr; i++ ) { + silk_scale_copy_vector_FLP( x_pre_ptr, x_ptr, invGains[ i ], + psEnc->sCmn.subfr_length + psEnc->sCmn.predictLPCOrder ); + x_pre_ptr += psEnc->sCmn.subfr_length + psEnc->sCmn.predictLPCOrder; + x_ptr += psEnc->sCmn.subfr_length; + } + silk_memset( psEncCtrl->LTPCoef, 0, psEnc->sCmn.nb_subfr * LTP_ORDER * sizeof( silk_float ) ); + psEncCtrl->LTPredCodGain = 0.0f; + psEnc->sCmn.sum_log_gain_Q7 = 0; + } + + /* Limit on total predictive coding gain */ + if( psEnc->sCmn.first_frame_after_reset ) { + minInvGain = 1.0f / MAX_PREDICTION_POWER_GAIN_AFTER_RESET; + } else { + minInvGain = (silk_float)pow( 2, psEncCtrl->LTPredCodGain / 3 ) / MAX_PREDICTION_POWER_GAIN; + minInvGain /= 0.25f + 0.75f * psEncCtrl->coding_quality; + } + + /* LPC_in_pre contains the LTP-filtered input for voiced, and the unfiltered input for unvoiced */ + silk_find_LPC_FLP( &psEnc->sCmn, NLSF_Q15, LPC_in_pre, minInvGain ); + + /* Quantize LSFs */ + silk_process_NLSFs_FLP( &psEnc->sCmn, psEncCtrl->PredCoef, NLSF_Q15, psEnc->sCmn.prev_NLSFq_Q15 ); + + /* Calculate residual energy using quantized LPC coefficients */ + silk_residual_energy_FLP( psEncCtrl->ResNrg, LPC_in_pre, psEncCtrl->PredCoef, psEncCtrl->Gains, + psEnc->sCmn.subfr_length, psEnc->sCmn.nb_subfr, psEnc->sCmn.predictLPCOrder ); + + /* Copy to prediction struct for use in next frame for interpolation */ + silk_memcpy( psEnc->sCmn.prev_NLSFq_Q15, NLSF_Q15, sizeof( psEnc->sCmn.prev_NLSFq_Q15 ) ); +} + diff --git a/media/libopus/silk/float/inner_product_FLP.c b/media/libopus/silk/float/inner_product_FLP.c new file mode 100644 index 0000000000..cdd39d24ce --- /dev/null +++ b/media/libopus/silk/float/inner_product_FLP.c @@ -0,0 +1,59 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "SigProc_FLP.h" + +/* inner product of two silk_float arrays, with result as double */ +double silk_inner_product_FLP( + const silk_float *data1, + const silk_float *data2, + opus_int dataSize +) +{ + opus_int i; + double result; + + /* 4x unrolled loop */ + result = 0.0; + for( i = 0; i < dataSize - 3; i += 4 ) { + result += data1[ i + 0 ] * (double)data2[ i + 0 ] + + data1[ i + 1 ] * (double)data2[ i + 1 ] + + data1[ i + 2 ] * (double)data2[ i + 2 ] + + data1[ i + 3 ] * (double)data2[ i + 3 ]; + } + + /* add any remaining products */ + for( ; i < dataSize; i++ ) { + result += data1[ i ] * (double)data2[ i ]; + } + + return result; +} diff --git a/media/libopus/silk/float/k2a_FLP.c b/media/libopus/silk/float/k2a_FLP.c new file mode 100644 index 0000000000..1448008dbb --- /dev/null +++ b/media/libopus/silk/float/k2a_FLP.c @@ -0,0 +1,54 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "SigProc_FLP.h" + +/* step up function, converts reflection coefficients to prediction coefficients */ +void silk_k2a_FLP( + silk_float *A, /* O prediction coefficients [order] */ + const silk_float *rc, /* I reflection coefficients [order] */ + opus_int32 order /* I prediction order */ +) +{ + opus_int k, n; + silk_float rck, tmp1, tmp2; + + for( k = 0; k < order; k++ ) { + rck = rc[ k ]; + for( n = 0; n < (k + 1) >> 1; n++ ) { + tmp1 = A[ n ]; + tmp2 = A[ k - n - 1 ]; + A[ n ] = tmp1 + tmp2 * rck; + A[ k - n - 1 ] = tmp2 + tmp1 * rck; + } + A[ k ] = -rck; + } +} diff --git a/media/libopus/silk/float/main_FLP.h b/media/libopus/silk/float/main_FLP.h new file mode 100644 index 0000000000..5dc0ccf4a4 --- /dev/null +++ b/media/libopus/silk/float/main_FLP.h @@ -0,0 +1,286 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef SILK_MAIN_FLP_H +#define SILK_MAIN_FLP_H + +#include "SigProc_FLP.h" +#include "SigProc_FIX.h" +#include "structs_FLP.h" +#include "main.h" +#include "define.h" +#include "debug.h" +#include "entenc.h" + +#ifdef __cplusplus +extern "C" +{ +#endif + +#define silk_encoder_state_Fxx silk_encoder_state_FLP +#define silk_encode_do_VAD_Fxx silk_encode_do_VAD_FLP +#define silk_encode_frame_Fxx silk_encode_frame_FLP + +/*********************/ +/* Encoder Functions */ +/*********************/ + +/* High-pass filter with cutoff frequency adaptation based on pitch lag statistics */ +void silk_HP_variable_cutoff( + silk_encoder_state_Fxx state_Fxx[] /* I/O Encoder states */ +); + +/* Encoder main function */ +void silk_encode_do_VAD_FLP( + silk_encoder_state_FLP *psEnc, /* I/O Encoder state FLP */ + opus_int activity /* I Decision of Opus voice activity detector */ +); + +/* Encoder main function */ +opus_int silk_encode_frame_FLP( + silk_encoder_state_FLP *psEnc, /* I/O Encoder state FLP */ + opus_int32 *pnBytesOut, /* O Number of payload bytes; */ + ec_enc *psRangeEnc, /* I/O compressor data structure */ + opus_int condCoding, /* I The type of conditional coding to use */ + opus_int maxBits, /* I If > 0: maximum number of output bits */ + opus_int useCBR /* I Flag to force constant-bitrate operation */ +); + +/* Initializes the Silk encoder state */ +opus_int silk_init_encoder( + silk_encoder_state_FLP *psEnc, /* I/O Encoder state FLP */ + int arch /* I Run-tim architecture */ +); + +/* Control the Silk encoder */ +opus_int silk_control_encoder( + silk_encoder_state_FLP *psEnc, /* I/O Pointer to Silk encoder state FLP */ + silk_EncControlStruct *encControl, /* I Control structure */ + const opus_int allow_bw_switch, /* I Flag to allow switching audio bandwidth */ + const opus_int channelNb, /* I Channel number */ + const opus_int force_fs_kHz +); + +/**************************/ +/* Noise shaping analysis */ +/**************************/ +/* Compute noise shaping coefficients and initial gain values */ +void silk_noise_shape_analysis_FLP( + silk_encoder_state_FLP *psEnc, /* I/O Encoder state FLP */ + silk_encoder_control_FLP *psEncCtrl, /* I/O Encoder control FLP */ + const silk_float *pitch_res, /* I LPC residual from pitch analysis */ + const silk_float *x /* I Input signal [frame_length + la_shape] */ +); + +/* Autocorrelations for a warped frequency axis */ +void silk_warped_autocorrelation_FLP( + silk_float *corr, /* O Result [order + 1] */ + const silk_float *input, /* I Input data to correlate */ + const silk_float warping, /* I Warping coefficient */ + const opus_int length, /* I Length of input */ + const opus_int order /* I Correlation order (even) */ +); + +/* Calculation of LTP state scaling */ +void silk_LTP_scale_ctrl_FLP( + silk_encoder_state_FLP *psEnc, /* I/O Encoder state FLP */ + silk_encoder_control_FLP *psEncCtrl, /* I/O Encoder control FLP */ + opus_int condCoding /* I The type of conditional coding to use */ +); + +/**********************************************/ +/* Prediction Analysis */ +/**********************************************/ +/* Find pitch lags */ +void silk_find_pitch_lags_FLP( + silk_encoder_state_FLP *psEnc, /* I/O Encoder state FLP */ + silk_encoder_control_FLP *psEncCtrl, /* I/O Encoder control FLP */ + silk_float res[], /* O Residual */ + const silk_float x[], /* I Speech signal */ + int arch /* I Run-time architecture */ +); + +/* Find LPC and LTP coefficients */ +void silk_find_pred_coefs_FLP( + silk_encoder_state_FLP *psEnc, /* I/O Encoder state FLP */ + silk_encoder_control_FLP *psEncCtrl, /* I/O Encoder control FLP */ + const silk_float res_pitch[], /* I Residual from pitch analysis */ + const silk_float x[], /* I Speech signal */ + opus_int condCoding /* I The type of conditional coding to use */ +); + +/* LPC analysis */ +void silk_find_LPC_FLP( + silk_encoder_state *psEncC, /* I/O Encoder state */ + opus_int16 NLSF_Q15[], /* O NLSFs */ + const silk_float x[], /* I Input signal */ + const silk_float minInvGain /* I Prediction gain from LTP (dB) */ +); + +/* LTP analysis */ +void silk_find_LTP_FLP( + silk_float XX[ MAX_NB_SUBFR * LTP_ORDER * LTP_ORDER ], /* O Weight for LTP quantization */ + silk_float xX[ MAX_NB_SUBFR * LTP_ORDER ], /* O Weight for LTP quantization */ + const silk_float r_ptr[], /* I LPC residual */ + const opus_int lag[ MAX_NB_SUBFR ], /* I LTP lags */ + const opus_int subfr_length, /* I Subframe length */ + const opus_int nb_subfr /* I number of subframes */ +); + +void silk_LTP_analysis_filter_FLP( + silk_float *LTP_res, /* O LTP res MAX_NB_SUBFR*(pre_lgth+subfr_lngth) */ + const silk_float *x, /* I Input signal, with preceding samples */ + const silk_float B[ LTP_ORDER * MAX_NB_SUBFR ], /* I LTP coefficients for each subframe */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ + const silk_float invGains[ MAX_NB_SUBFR ], /* I Inverse quantization gains */ + const opus_int subfr_length, /* I Length of each subframe */ + const opus_int nb_subfr, /* I number of subframes */ + const opus_int pre_length /* I Preceding samples for each subframe */ +); + +/* Calculates residual energies of input subframes where all subframes have LPC_order */ +/* of preceding samples */ +void silk_residual_energy_FLP( + silk_float nrgs[ MAX_NB_SUBFR ], /* O Residual energy per subframe */ + const silk_float x[], /* I Input signal */ + silk_float a[ 2 ][ MAX_LPC_ORDER ], /* I AR coefs for each frame half */ + const silk_float gains[], /* I Quantization gains */ + const opus_int subfr_length, /* I Subframe length */ + const opus_int nb_subfr, /* I number of subframes */ + const opus_int LPC_order /* I LPC order */ +); + +/* 16th order LPC analysis filter */ +void silk_LPC_analysis_filter_FLP( + silk_float r_LPC[], /* O LPC residual signal */ + const silk_float PredCoef[], /* I LPC coefficients */ + const silk_float s[], /* I Input signal */ + const opus_int length, /* I Length of input signal */ + const opus_int Order /* I LPC order */ +); + +/* LTP tap quantizer */ +void silk_quant_LTP_gains_FLP( + silk_float B[ MAX_NB_SUBFR * LTP_ORDER ], /* O Quantized LTP gains */ + opus_int8 cbk_index[ MAX_NB_SUBFR ], /* O Codebook index */ + opus_int8 *periodicity_index, /* O Periodicity index */ + opus_int32 *sum_log_gain_Q7, /* I/O Cumulative max prediction gain */ + silk_float *pred_gain_dB, /* O LTP prediction gain */ + const silk_float XX[ MAX_NB_SUBFR * LTP_ORDER * LTP_ORDER ], /* I Correlation matrix */ + const silk_float xX[ MAX_NB_SUBFR * LTP_ORDER ], /* I Correlation vector */ + const opus_int subfr_len, /* I Number of samples per subframe */ + const opus_int nb_subfr, /* I Number of subframes */ + int arch /* I Run-time architecture */ +); + +/* Residual energy: nrg = wxx - 2 * wXx * c + c' * wXX * c */ +silk_float silk_residual_energy_covar_FLP( /* O Weighted residual energy */ + const silk_float *c, /* I Filter coefficients */ + silk_float *wXX, /* I/O Weighted correlation matrix, reg. out */ + const silk_float *wXx, /* I Weighted correlation vector */ + const silk_float wxx, /* I Weighted correlation value */ + const opus_int D /* I Dimension */ +); + +/* Processing of gains */ +void silk_process_gains_FLP( + silk_encoder_state_FLP *psEnc, /* I/O Encoder state FLP */ + silk_encoder_control_FLP *psEncCtrl, /* I/O Encoder control FLP */ + opus_int condCoding /* I The type of conditional coding to use */ +); + +/******************/ +/* Linear Algebra */ +/******************/ +/* Calculates correlation matrix X'*X */ +void silk_corrMatrix_FLP( + const silk_float *x, /* I x vector [ L+order-1 ] used to create X */ + const opus_int L, /* I Length of vectors */ + const opus_int Order, /* I Max lag for correlation */ + silk_float *XX /* O X'*X correlation matrix [order x order] */ +); + +/* Calculates correlation vector X'*t */ +void silk_corrVector_FLP( + const silk_float *x, /* I x vector [L+order-1] used to create X */ + const silk_float *t, /* I Target vector [L] */ + const opus_int L, /* I Length of vecors */ + const opus_int Order, /* I Max lag for correlation */ + silk_float *Xt /* O X'*t correlation vector [order] */ +); + +/* Apply sine window to signal vector. */ +/* Window types: */ +/* 1 -> sine window from 0 to pi/2 */ +/* 2 -> sine window from pi/2 to pi */ +void silk_apply_sine_window_FLP( + silk_float px_win[], /* O Pointer to windowed signal */ + const silk_float px[], /* I Pointer to input signal */ + const opus_int win_type, /* I Selects a window type */ + const opus_int length /* I Window length, multiple of 4 */ +); + +/* Wrapper functions. Call flp / fix code */ + +/* Convert AR filter coefficients to NLSF parameters */ +void silk_A2NLSF_FLP( + opus_int16 *NLSF_Q15, /* O NLSF vector [ LPC_order ] */ + const silk_float *pAR, /* I LPC coefficients [ LPC_order ] */ + const opus_int LPC_order /* I LPC order */ +); + +/* Convert NLSF parameters to AR prediction filter coefficients */ +void silk_NLSF2A_FLP( + silk_float *pAR, /* O LPC coefficients [ LPC_order ] */ + const opus_int16 *NLSF_Q15, /* I NLSF vector [ LPC_order ] */ + const opus_int LPC_order, /* I LPC order */ + int arch /* I Run-time architecture */ +); + +/* Limit, stabilize, and quantize NLSFs */ +void silk_process_NLSFs_FLP( + silk_encoder_state *psEncC, /* I/O Encoder state */ + silk_float PredCoef[ 2 ][ MAX_LPC_ORDER ], /* O Prediction coefficients */ + opus_int16 NLSF_Q15[ MAX_LPC_ORDER ], /* I/O Normalized LSFs (quant out) (0 - (2^15-1)) */ + const opus_int16 prev_NLSF_Q15[ MAX_LPC_ORDER ] /* I Previous Normalized LSFs (0 - (2^15-1)) */ +); + +/* Floating-point Silk NSQ wrapper */ +void silk_NSQ_wrapper_FLP( + silk_encoder_state_FLP *psEnc, /* I/O Encoder state FLP */ + silk_encoder_control_FLP *psEncCtrl, /* I/O Encoder control FLP */ + SideInfoIndices *psIndices, /* I/O Quantization indices */ + silk_nsq_state *psNSQ, /* I/O Noise Shaping Quantzation state */ + opus_int8 pulses[], /* O Quantized pulse signal */ + const silk_float x[] /* I Prefiltered input signal */ +); + +#ifdef __cplusplus +} +#endif + +#endif diff --git a/media/libopus/silk/float/noise_shape_analysis_FLP.c b/media/libopus/silk/float/noise_shape_analysis_FLP.c new file mode 100644 index 0000000000..cb3d8a50b7 --- /dev/null +++ b/media/libopus/silk/float/noise_shape_analysis_FLP.c @@ -0,0 +1,350 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main_FLP.h" +#include "tuning_parameters.h" + +/* Compute gain to make warped filter coefficients have a zero mean log frequency response on a */ +/* non-warped frequency scale. (So that it can be implemented with a minimum-phase monic filter.) */ +/* Note: A monic filter is one with the first coefficient equal to 1.0. In Silk we omit the first */ +/* coefficient in an array of coefficients, for monic filters. */ +static OPUS_INLINE silk_float warped_gain( + const silk_float *coefs, + silk_float lambda, + opus_int order +) { + opus_int i; + silk_float gain; + + lambda = -lambda; + gain = coefs[ order - 1 ]; + for( i = order - 2; i >= 0; i-- ) { + gain = lambda * gain + coefs[ i ]; + } + return (silk_float)( 1.0f / ( 1.0f - lambda * gain ) ); +} + +/* Convert warped filter coefficients to monic pseudo-warped coefficients and limit maximum */ +/* amplitude of monic warped coefficients by using bandwidth expansion on the true coefficients */ +static OPUS_INLINE void warped_true2monic_coefs( + silk_float *coefs, + silk_float lambda, + silk_float limit, + opus_int order +) { + opus_int i, iter, ind = 0; + silk_float tmp, maxabs, chirp, gain; + + /* Convert to monic coefficients */ + for( i = order - 1; i > 0; i-- ) { + coefs[ i - 1 ] -= lambda * coefs[ i ]; + } + gain = ( 1.0f - lambda * lambda ) / ( 1.0f + lambda * coefs[ 0 ] ); + for( i = 0; i < order; i++ ) { + coefs[ i ] *= gain; + } + + /* Limit */ + for( iter = 0; iter < 10; iter++ ) { + /* Find maximum absolute value */ + maxabs = -1.0f; + for( i = 0; i < order; i++ ) { + tmp = silk_abs_float( coefs[ i ] ); + if( tmp > maxabs ) { + maxabs = tmp; + ind = i; + } + } + if( maxabs <= limit ) { + /* Coefficients are within range - done */ + return; + } + + /* Convert back to true warped coefficients */ + for( i = 1; i < order; i++ ) { + coefs[ i - 1 ] += lambda * coefs[ i ]; + } + gain = 1.0f / gain; + for( i = 0; i < order; i++ ) { + coefs[ i ] *= gain; + } + + /* Apply bandwidth expansion */ + chirp = 0.99f - ( 0.8f + 0.1f * iter ) * ( maxabs - limit ) / ( maxabs * ( ind + 1 ) ); + silk_bwexpander_FLP( coefs, order, chirp ); + + /* Convert to monic warped coefficients */ + for( i = order - 1; i > 0; i-- ) { + coefs[ i - 1 ] -= lambda * coefs[ i ]; + } + gain = ( 1.0f - lambda * lambda ) / ( 1.0f + lambda * coefs[ 0 ] ); + for( i = 0; i < order; i++ ) { + coefs[ i ] *= gain; + } + } + silk_assert( 0 ); +} + +static OPUS_INLINE void limit_coefs( + silk_float *coefs, + silk_float limit, + opus_int order +) { + opus_int i, iter, ind = 0; + silk_float tmp, maxabs, chirp; + + for( iter = 0; iter < 10; iter++ ) { + /* Find maximum absolute value */ + maxabs = -1.0f; + for( i = 0; i < order; i++ ) { + tmp = silk_abs_float( coefs[ i ] ); + if( tmp > maxabs ) { + maxabs = tmp; + ind = i; + } + } + if( maxabs <= limit ) { + /* Coefficients are within range - done */ + return; + } + + /* Apply bandwidth expansion */ + chirp = 0.99f - ( 0.8f + 0.1f * iter ) * ( maxabs - limit ) / ( maxabs * ( ind + 1 ) ); + silk_bwexpander_FLP( coefs, order, chirp ); + } + silk_assert( 0 ); +} + +/* Compute noise shaping coefficients and initial gain values */ +void silk_noise_shape_analysis_FLP( + silk_encoder_state_FLP *psEnc, /* I/O Encoder state FLP */ + silk_encoder_control_FLP *psEncCtrl, /* I/O Encoder control FLP */ + const silk_float *pitch_res, /* I LPC residual from pitch analysis */ + const silk_float *x /* I Input signal [frame_length + la_shape] */ +) +{ + silk_shape_state_FLP *psShapeSt = &psEnc->sShape; + opus_int k, nSamples, nSegs; + silk_float SNR_adj_dB, HarmShapeGain, Tilt; + silk_float nrg, log_energy, log_energy_prev, energy_variation; + silk_float BWExp, gain_mult, gain_add, strength, b, warping; + silk_float x_windowed[ SHAPE_LPC_WIN_MAX ]; + silk_float auto_corr[ MAX_SHAPE_LPC_ORDER + 1 ]; + silk_float rc[ MAX_SHAPE_LPC_ORDER + 1 ]; + const silk_float *x_ptr, *pitch_res_ptr; + + /* Point to start of first LPC analysis block */ + x_ptr = x - psEnc->sCmn.la_shape; + + /****************/ + /* GAIN CONTROL */ + /****************/ + SNR_adj_dB = psEnc->sCmn.SNR_dB_Q7 * ( 1 / 128.0f ); + + /* Input quality is the average of the quality in the lowest two VAD bands */ + psEncCtrl->input_quality = 0.5f * ( psEnc->sCmn.input_quality_bands_Q15[ 0 ] + psEnc->sCmn.input_quality_bands_Q15[ 1 ] ) * ( 1.0f / 32768.0f ); + + /* Coding quality level, between 0.0 and 1.0 */ + psEncCtrl->coding_quality = silk_sigmoid( 0.25f * ( SNR_adj_dB - 20.0f ) ); + + if( psEnc->sCmn.useCBR == 0 ) { + /* Reduce coding SNR during low speech activity */ + b = 1.0f - psEnc->sCmn.speech_activity_Q8 * ( 1.0f / 256.0f ); + SNR_adj_dB -= BG_SNR_DECR_dB * psEncCtrl->coding_quality * ( 0.5f + 0.5f * psEncCtrl->input_quality ) * b * b; + } + + if( psEnc->sCmn.indices.signalType == TYPE_VOICED ) { + /* Reduce gains for periodic signals */ + SNR_adj_dB += HARM_SNR_INCR_dB * psEnc->LTPCorr; + } else { + /* For unvoiced signals and low-quality input, adjust the quality slower than SNR_dB setting */ + SNR_adj_dB += ( -0.4f * psEnc->sCmn.SNR_dB_Q7 * ( 1 / 128.0f ) + 6.0f ) * ( 1.0f - psEncCtrl->input_quality ); + } + + /*************************/ + /* SPARSENESS PROCESSING */ + /*************************/ + /* Set quantizer offset */ + if( psEnc->sCmn.indices.signalType == TYPE_VOICED ) { + /* Initially set to 0; may be overruled in process_gains(..) */ + psEnc->sCmn.indices.quantOffsetType = 0; + } else { + /* Sparseness measure, based on relative fluctuations of energy per 2 milliseconds */ + nSamples = 2 * psEnc->sCmn.fs_kHz; + energy_variation = 0.0f; + log_energy_prev = 0.0f; + pitch_res_ptr = pitch_res; + nSegs = silk_SMULBB( SUB_FRAME_LENGTH_MS, psEnc->sCmn.nb_subfr ) / 2; + for( k = 0; k < nSegs; k++ ) { + nrg = ( silk_float )nSamples + ( silk_float )silk_energy_FLP( pitch_res_ptr, nSamples ); + log_energy = silk_log2( nrg ); + if( k > 0 ) { + energy_variation += silk_abs_float( log_energy - log_energy_prev ); + } + log_energy_prev = log_energy; + pitch_res_ptr += nSamples; + } + + /* Set quantization offset depending on sparseness measure */ + if( energy_variation > ENERGY_VARIATION_THRESHOLD_QNT_OFFSET * (nSegs-1) ) { + psEnc->sCmn.indices.quantOffsetType = 0; + } else { + psEnc->sCmn.indices.quantOffsetType = 1; + } + } + + /*******************************/ + /* Control bandwidth expansion */ + /*******************************/ + /* More BWE for signals with high prediction gain */ + strength = FIND_PITCH_WHITE_NOISE_FRACTION * psEncCtrl->predGain; /* between 0.0 and 1.0 */ + BWExp = BANDWIDTH_EXPANSION / ( 1.0f + strength * strength ); + + /* Slightly more warping in analysis will move quantization noise up in frequency, where it's better masked */ + warping = (silk_float)psEnc->sCmn.warping_Q16 / 65536.0f + 0.01f * psEncCtrl->coding_quality; + + /********************************************/ + /* Compute noise shaping AR coefs and gains */ + /********************************************/ + for( k = 0; k < psEnc->sCmn.nb_subfr; k++ ) { + /* Apply window: sine slope followed by flat part followed by cosine slope */ + opus_int shift, slope_part, flat_part; + flat_part = psEnc->sCmn.fs_kHz * 3; + slope_part = ( psEnc->sCmn.shapeWinLength - flat_part ) / 2; + + silk_apply_sine_window_FLP( x_windowed, x_ptr, 1, slope_part ); + shift = slope_part; + silk_memcpy( x_windowed + shift, x_ptr + shift, flat_part * sizeof(silk_float) ); + shift += flat_part; + silk_apply_sine_window_FLP( x_windowed + shift, x_ptr + shift, 2, slope_part ); + + /* Update pointer: next LPC analysis block */ + x_ptr += psEnc->sCmn.subfr_length; + + if( psEnc->sCmn.warping_Q16 > 0 ) { + /* Calculate warped auto correlation */ + silk_warped_autocorrelation_FLP( auto_corr, x_windowed, warping, + psEnc->sCmn.shapeWinLength, psEnc->sCmn.shapingLPCOrder ); + } else { + /* Calculate regular auto correlation */ + silk_autocorrelation_FLP( auto_corr, x_windowed, psEnc->sCmn.shapeWinLength, psEnc->sCmn.shapingLPCOrder + 1 ); + } + + /* Add white noise, as a fraction of energy */ + auto_corr[ 0 ] += auto_corr[ 0 ] * SHAPE_WHITE_NOISE_FRACTION + 1.0f; + + /* Convert correlations to prediction coefficients, and compute residual energy */ + nrg = silk_schur_FLP( rc, auto_corr, psEnc->sCmn.shapingLPCOrder ); + silk_k2a_FLP( &psEncCtrl->AR[ k * MAX_SHAPE_LPC_ORDER ], rc, psEnc->sCmn.shapingLPCOrder ); + psEncCtrl->Gains[ k ] = ( silk_float )sqrt( nrg ); + + if( psEnc->sCmn.warping_Q16 > 0 ) { + /* Adjust gain for warping */ + psEncCtrl->Gains[ k ] *= warped_gain( &psEncCtrl->AR[ k * MAX_SHAPE_LPC_ORDER ], warping, psEnc->sCmn.shapingLPCOrder ); + } + + /* Bandwidth expansion for synthesis filter shaping */ + silk_bwexpander_FLP( &psEncCtrl->AR[ k * MAX_SHAPE_LPC_ORDER ], psEnc->sCmn.shapingLPCOrder, BWExp ); + + if( psEnc->sCmn.warping_Q16 > 0 ) { + /* Convert to monic warped prediction coefficients and limit absolute values */ + warped_true2monic_coefs( &psEncCtrl->AR[ k * MAX_SHAPE_LPC_ORDER ], warping, 3.999f, psEnc->sCmn.shapingLPCOrder ); + } else { + /* Limit absolute values */ + limit_coefs( &psEncCtrl->AR[ k * MAX_SHAPE_LPC_ORDER ], 3.999f, psEnc->sCmn.shapingLPCOrder ); + } + } + + /*****************/ + /* Gain tweaking */ + /*****************/ + /* Increase gains during low speech activity */ + gain_mult = (silk_float)pow( 2.0f, -0.16f * SNR_adj_dB ); + gain_add = (silk_float)pow( 2.0f, 0.16f * MIN_QGAIN_DB ); + for( k = 0; k < psEnc->sCmn.nb_subfr; k++ ) { + psEncCtrl->Gains[ k ] *= gain_mult; + psEncCtrl->Gains[ k ] += gain_add; + } + + /************************************************/ + /* Control low-frequency shaping and noise tilt */ + /************************************************/ + /* Less low frequency shaping for noisy inputs */ + strength = LOW_FREQ_SHAPING * ( 1.0f + LOW_QUALITY_LOW_FREQ_SHAPING_DECR * ( psEnc->sCmn.input_quality_bands_Q15[ 0 ] * ( 1.0f / 32768.0f ) - 1.0f ) ); + strength *= psEnc->sCmn.speech_activity_Q8 * ( 1.0f / 256.0f ); + if( psEnc->sCmn.indices.signalType == TYPE_VOICED ) { + /* Reduce low frequencies quantization noise for periodic signals, depending on pitch lag */ + /*f = 400; freqz([1, -0.98 + 2e-4 * f], [1, -0.97 + 7e-4 * f], 2^12, Fs); axis([0, 1000, -10, 1])*/ + for( k = 0; k < psEnc->sCmn.nb_subfr; k++ ) { + b = 0.2f / psEnc->sCmn.fs_kHz + 3.0f / psEncCtrl->pitchL[ k ]; + psEncCtrl->LF_MA_shp[ k ] = -1.0f + b; + psEncCtrl->LF_AR_shp[ k ] = 1.0f - b - b * strength; + } + Tilt = - HP_NOISE_COEF - + (1 - HP_NOISE_COEF) * HARM_HP_NOISE_COEF * psEnc->sCmn.speech_activity_Q8 * ( 1.0f / 256.0f ); + } else { + b = 1.3f / psEnc->sCmn.fs_kHz; + psEncCtrl->LF_MA_shp[ 0 ] = -1.0f + b; + psEncCtrl->LF_AR_shp[ 0 ] = 1.0f - b - b * strength * 0.6f; + for( k = 1; k < psEnc->sCmn.nb_subfr; k++ ) { + psEncCtrl->LF_MA_shp[ k ] = psEncCtrl->LF_MA_shp[ 0 ]; + psEncCtrl->LF_AR_shp[ k ] = psEncCtrl->LF_AR_shp[ 0 ]; + } + Tilt = -HP_NOISE_COEF; + } + + /****************************/ + /* HARMONIC SHAPING CONTROL */ + /****************************/ + if( USE_HARM_SHAPING && psEnc->sCmn.indices.signalType == TYPE_VOICED ) { + /* Harmonic noise shaping */ + HarmShapeGain = HARMONIC_SHAPING; + + /* More harmonic noise shaping for high bitrates or noisy input */ + HarmShapeGain += HIGH_RATE_OR_LOW_QUALITY_HARMONIC_SHAPING * + ( 1.0f - ( 1.0f - psEncCtrl->coding_quality ) * psEncCtrl->input_quality ); + + /* Less harmonic noise shaping for less periodic signals */ + HarmShapeGain *= ( silk_float )sqrt( psEnc->LTPCorr ); + } else { + HarmShapeGain = 0.0f; + } + + /*************************/ + /* Smooth over subframes */ + /*************************/ + for( k = 0; k < psEnc->sCmn.nb_subfr; k++ ) { + psShapeSt->HarmShapeGain_smth += SUBFR_SMTH_COEF * ( HarmShapeGain - psShapeSt->HarmShapeGain_smth ); + psEncCtrl->HarmShapeGain[ k ] = psShapeSt->HarmShapeGain_smth; + psShapeSt->Tilt_smth += SUBFR_SMTH_COEF * ( Tilt - psShapeSt->Tilt_smth ); + psEncCtrl->Tilt[ k ] = psShapeSt->Tilt_smth; + } +} diff --git a/media/libopus/silk/float/pitch_analysis_core_FLP.c b/media/libopus/silk/float/pitch_analysis_core_FLP.c new file mode 100644 index 0000000000..f351bc3718 --- /dev/null +++ b/media/libopus/silk/float/pitch_analysis_core_FLP.c @@ -0,0 +1,630 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +/***************************************************************************** +* Pitch analyser function +******************************************************************************/ +#include "SigProc_FLP.h" +#include "SigProc_FIX.h" +#include "pitch_est_defines.h" +#include "pitch.h" + +#define SCRATCH_SIZE 22 + +/************************************************************/ +/* Internally used functions */ +/************************************************************/ +static void silk_P_Ana_calc_corr_st3( + silk_float cross_corr_st3[ PE_MAX_NB_SUBFR ][ PE_NB_CBKS_STAGE3_MAX ][ PE_NB_STAGE3_LAGS ], /* O 3 DIM correlation array */ + const silk_float frame[], /* I vector to correlate */ + opus_int start_lag, /* I start lag */ + opus_int sf_length, /* I sub frame length */ + opus_int nb_subfr, /* I number of subframes */ + opus_int complexity, /* I Complexity setting */ + int arch /* I Run-time architecture */ +); + +static void silk_P_Ana_calc_energy_st3( + silk_float energies_st3[ PE_MAX_NB_SUBFR ][ PE_NB_CBKS_STAGE3_MAX ][ PE_NB_STAGE3_LAGS ], /* O 3 DIM correlation array */ + const silk_float frame[], /* I vector to correlate */ + opus_int start_lag, /* I start lag */ + opus_int sf_length, /* I sub frame length */ + opus_int nb_subfr, /* I number of subframes */ + opus_int complexity /* I Complexity setting */ +); + +/************************************************************/ +/* CORE PITCH ANALYSIS FUNCTION */ +/************************************************************/ +opus_int silk_pitch_analysis_core_FLP( /* O Voicing estimate: 0 voiced, 1 unvoiced */ + const silk_float *frame, /* I Signal of length PE_FRAME_LENGTH_MS*Fs_kHz */ + opus_int *pitch_out, /* O Pitch lag values [nb_subfr] */ + opus_int16 *lagIndex, /* O Lag Index */ + opus_int8 *contourIndex, /* O Pitch contour Index */ + silk_float *LTPCorr, /* I/O Normalized correlation; input: value from previous frame */ + opus_int prevLag, /* I Last lag of previous frame; set to zero is unvoiced */ + const silk_float search_thres1, /* I First stage threshold for lag candidates 0 - 1 */ + const silk_float search_thres2, /* I Final threshold for lag candidates 0 - 1 */ + const opus_int Fs_kHz, /* I sample frequency (kHz) */ + const opus_int complexity, /* I Complexity setting, 0-2, where 2 is highest */ + const opus_int nb_subfr, /* I Number of 5 ms subframes */ + int arch /* I Run-time architecture */ +) +{ + opus_int i, k, d, j; + silk_float frame_8kHz[ PE_MAX_FRAME_LENGTH_MS * 8 ]; + silk_float frame_4kHz[ PE_MAX_FRAME_LENGTH_MS * 4 ]; + opus_int16 frame_8_FIX[ PE_MAX_FRAME_LENGTH_MS * 8 ]; + opus_int16 frame_4_FIX[ PE_MAX_FRAME_LENGTH_MS * 4 ]; + opus_int32 filt_state[ 6 ]; + silk_float threshold, contour_bias; + silk_float C[ PE_MAX_NB_SUBFR][ (PE_MAX_LAG >> 1) + 5 ]; + opus_val32 xcorr[ PE_MAX_LAG_MS * 4 - PE_MIN_LAG_MS * 4 + 1 ]; + silk_float CC[ PE_NB_CBKS_STAGE2_EXT ]; + const silk_float *target_ptr, *basis_ptr; + double cross_corr, normalizer, energy, energy_tmp; + opus_int d_srch[ PE_D_SRCH_LENGTH ]; + opus_int16 d_comp[ (PE_MAX_LAG >> 1) + 5 ]; + opus_int length_d_srch, length_d_comp; + silk_float Cmax, CCmax, CCmax_b, CCmax_new_b, CCmax_new; + opus_int CBimax, CBimax_new, lag, start_lag, end_lag, lag_new; + opus_int cbk_size; + silk_float lag_log2, prevLag_log2, delta_lag_log2_sqr; + silk_float energies_st3[ PE_MAX_NB_SUBFR ][ PE_NB_CBKS_STAGE3_MAX ][ PE_NB_STAGE3_LAGS ]; + silk_float cross_corr_st3[ PE_MAX_NB_SUBFR ][ PE_NB_CBKS_STAGE3_MAX ][ PE_NB_STAGE3_LAGS ]; + opus_int lag_counter; + opus_int frame_length, frame_length_8kHz, frame_length_4kHz; + opus_int sf_length, sf_length_8kHz, sf_length_4kHz; + opus_int min_lag, min_lag_8kHz, min_lag_4kHz; + opus_int max_lag, max_lag_8kHz, max_lag_4kHz; + opus_int nb_cbk_search; + const opus_int8 *Lag_CB_ptr; + + /* Check for valid sampling frequency */ + celt_assert( Fs_kHz == 8 || Fs_kHz == 12 || Fs_kHz == 16 ); + + /* Check for valid complexity setting */ + celt_assert( complexity >= SILK_PE_MIN_COMPLEX ); + celt_assert( complexity <= SILK_PE_MAX_COMPLEX ); + + silk_assert( search_thres1 >= 0.0f && search_thres1 <= 1.0f ); + silk_assert( search_thres2 >= 0.0f && search_thres2 <= 1.0f ); + + /* Set up frame lengths max / min lag for the sampling frequency */ + frame_length = ( PE_LTP_MEM_LENGTH_MS + nb_subfr * PE_SUBFR_LENGTH_MS ) * Fs_kHz; + frame_length_4kHz = ( PE_LTP_MEM_LENGTH_MS + nb_subfr * PE_SUBFR_LENGTH_MS ) * 4; + frame_length_8kHz = ( PE_LTP_MEM_LENGTH_MS + nb_subfr * PE_SUBFR_LENGTH_MS ) * 8; + sf_length = PE_SUBFR_LENGTH_MS * Fs_kHz; + sf_length_4kHz = PE_SUBFR_LENGTH_MS * 4; + sf_length_8kHz = PE_SUBFR_LENGTH_MS * 8; + min_lag = PE_MIN_LAG_MS * Fs_kHz; + min_lag_4kHz = PE_MIN_LAG_MS * 4; + min_lag_8kHz = PE_MIN_LAG_MS * 8; + max_lag = PE_MAX_LAG_MS * Fs_kHz - 1; + max_lag_4kHz = PE_MAX_LAG_MS * 4; + max_lag_8kHz = PE_MAX_LAG_MS * 8 - 1; + + /* Resample from input sampled at Fs_kHz to 8 kHz */ + if( Fs_kHz == 16 ) { + /* Resample to 16 -> 8 khz */ + opus_int16 frame_16_FIX[ 16 * PE_MAX_FRAME_LENGTH_MS ]; + silk_float2short_array( frame_16_FIX, frame, frame_length ); + silk_memset( filt_state, 0, 2 * sizeof( opus_int32 ) ); + silk_resampler_down2( filt_state, frame_8_FIX, frame_16_FIX, frame_length ); + silk_short2float_array( frame_8kHz, frame_8_FIX, frame_length_8kHz ); + } else if( Fs_kHz == 12 ) { + /* Resample to 12 -> 8 khz */ + opus_int16 frame_12_FIX[ 12 * PE_MAX_FRAME_LENGTH_MS ]; + silk_float2short_array( frame_12_FIX, frame, frame_length ); + silk_memset( filt_state, 0, 6 * sizeof( opus_int32 ) ); + silk_resampler_down2_3( filt_state, frame_8_FIX, frame_12_FIX, frame_length ); + silk_short2float_array( frame_8kHz, frame_8_FIX, frame_length_8kHz ); + } else { + celt_assert( Fs_kHz == 8 ); + silk_float2short_array( frame_8_FIX, frame, frame_length_8kHz ); + } + + /* Decimate again to 4 kHz */ + silk_memset( filt_state, 0, 2 * sizeof( opus_int32 ) ); + silk_resampler_down2( filt_state, frame_4_FIX, frame_8_FIX, frame_length_8kHz ); + silk_short2float_array( frame_4kHz, frame_4_FIX, frame_length_4kHz ); + + /* Low-pass filter */ + for( i = frame_length_4kHz - 1; i > 0; i-- ) { + frame_4kHz[ i ] = silk_ADD_SAT16( frame_4kHz[ i ], frame_4kHz[ i - 1 ] ); + } + + /****************************************************************************** + * FIRST STAGE, operating in 4 khz + ******************************************************************************/ + silk_memset(C, 0, sizeof(silk_float) * nb_subfr * ((PE_MAX_LAG >> 1) + 5)); + target_ptr = &frame_4kHz[ silk_LSHIFT( sf_length_4kHz, 2 ) ]; + for( k = 0; k < nb_subfr >> 1; k++ ) { + /* Check that we are within range of the array */ + celt_assert( target_ptr >= frame_4kHz ); + celt_assert( target_ptr + sf_length_8kHz <= frame_4kHz + frame_length_4kHz ); + + basis_ptr = target_ptr - min_lag_4kHz; + + /* Check that we are within range of the array */ + celt_assert( basis_ptr >= frame_4kHz ); + celt_assert( basis_ptr + sf_length_8kHz <= frame_4kHz + frame_length_4kHz ); + + celt_pitch_xcorr( target_ptr, target_ptr-max_lag_4kHz, xcorr, sf_length_8kHz, max_lag_4kHz - min_lag_4kHz + 1, arch ); + + /* Calculate first vector products before loop */ + cross_corr = xcorr[ max_lag_4kHz - min_lag_4kHz ]; + normalizer = silk_energy_FLP( target_ptr, sf_length_8kHz ) + + silk_energy_FLP( basis_ptr, sf_length_8kHz ) + + sf_length_8kHz * 4000.0f; + + C[ 0 ][ min_lag_4kHz ] += (silk_float)( 2 * cross_corr / normalizer ); + + /* From now on normalizer is computed recursively */ + for( d = min_lag_4kHz + 1; d <= max_lag_4kHz; d++ ) { + basis_ptr--; + + /* Check that we are within range of the array */ + silk_assert( basis_ptr >= frame_4kHz ); + silk_assert( basis_ptr + sf_length_8kHz <= frame_4kHz + frame_length_4kHz ); + + cross_corr = xcorr[ max_lag_4kHz - d ]; + + /* Add contribution of new sample and remove contribution from oldest sample */ + normalizer += + basis_ptr[ 0 ] * (double)basis_ptr[ 0 ] - + basis_ptr[ sf_length_8kHz ] * (double)basis_ptr[ sf_length_8kHz ]; + C[ 0 ][ d ] += (silk_float)( 2 * cross_corr / normalizer ); + } + /* Update target pointer */ + target_ptr += sf_length_8kHz; + } + + /* Apply short-lag bias */ + for( i = max_lag_4kHz; i >= min_lag_4kHz; i-- ) { + C[ 0 ][ i ] -= C[ 0 ][ i ] * i / 4096.0f; + } + + /* Sort */ + length_d_srch = 4 + 2 * complexity; + celt_assert( 3 * length_d_srch <= PE_D_SRCH_LENGTH ); + silk_insertion_sort_decreasing_FLP( &C[ 0 ][ min_lag_4kHz ], d_srch, max_lag_4kHz - min_lag_4kHz + 1, length_d_srch ); + + /* Escape if correlation is very low already here */ + Cmax = C[ 0 ][ min_lag_4kHz ]; + if( Cmax < 0.2f ) { + silk_memset( pitch_out, 0, nb_subfr * sizeof( opus_int ) ); + *LTPCorr = 0.0f; + *lagIndex = 0; + *contourIndex = 0; + return 1; + } + + threshold = search_thres1 * Cmax; + for( i = 0; i < length_d_srch; i++ ) { + /* Convert to 8 kHz indices for the sorted correlation that exceeds the threshold */ + if( C[ 0 ][ min_lag_4kHz + i ] > threshold ) { + d_srch[ i ] = silk_LSHIFT( d_srch[ i ] + min_lag_4kHz, 1 ); + } else { + length_d_srch = i; + break; + } + } + celt_assert( length_d_srch > 0 ); + + for( i = min_lag_8kHz - 5; i < max_lag_8kHz + 5; i++ ) { + d_comp[ i ] = 0; + } + for( i = 0; i < length_d_srch; i++ ) { + d_comp[ d_srch[ i ] ] = 1; + } + + /* Convolution */ + for( i = max_lag_8kHz + 3; i >= min_lag_8kHz; i-- ) { + d_comp[ i ] += d_comp[ i - 1 ] + d_comp[ i - 2 ]; + } + + length_d_srch = 0; + for( i = min_lag_8kHz; i < max_lag_8kHz + 1; i++ ) { + if( d_comp[ i + 1 ] > 0 ) { + d_srch[ length_d_srch ] = i; + length_d_srch++; + } + } + + /* Convolution */ + for( i = max_lag_8kHz + 3; i >= min_lag_8kHz; i-- ) { + d_comp[ i ] += d_comp[ i - 1 ] + d_comp[ i - 2 ] + d_comp[ i - 3 ]; + } + + length_d_comp = 0; + for( i = min_lag_8kHz; i < max_lag_8kHz + 4; i++ ) { + if( d_comp[ i ] > 0 ) { + d_comp[ length_d_comp ] = (opus_int16)( i - 2 ); + length_d_comp++; + } + } + + /********************************************************************************** + ** SECOND STAGE, operating at 8 kHz, on lag sections with high correlation + *************************************************************************************/ + /********************************************************************************* + * Find energy of each subframe projected onto its history, for a range of delays + *********************************************************************************/ + silk_memset( C, 0, PE_MAX_NB_SUBFR*((PE_MAX_LAG >> 1) + 5) * sizeof(silk_float)); + + if( Fs_kHz == 8 ) { + target_ptr = &frame[ PE_LTP_MEM_LENGTH_MS * 8 ]; + } else { + target_ptr = &frame_8kHz[ PE_LTP_MEM_LENGTH_MS * 8 ]; + } + for( k = 0; k < nb_subfr; k++ ) { + energy_tmp = silk_energy_FLP( target_ptr, sf_length_8kHz ) + 1.0; + for( j = 0; j < length_d_comp; j++ ) { + d = d_comp[ j ]; + basis_ptr = target_ptr - d; + cross_corr = silk_inner_product_FLP( basis_ptr, target_ptr, sf_length_8kHz ); + if( cross_corr > 0.0f ) { + energy = silk_energy_FLP( basis_ptr, sf_length_8kHz ); + C[ k ][ d ] = (silk_float)( 2 * cross_corr / ( energy + energy_tmp ) ); + } else { + C[ k ][ d ] = 0.0f; + } + } + target_ptr += sf_length_8kHz; + } + + /* search over lag range and lags codebook */ + /* scale factor for lag codebook, as a function of center lag */ + + CCmax = 0.0f; /* This value doesn't matter */ + CCmax_b = -1000.0f; + + CBimax = 0; /* To avoid returning undefined lag values */ + lag = -1; /* To check if lag with strong enough correlation has been found */ + + if( prevLag > 0 ) { + if( Fs_kHz == 12 ) { + prevLag = silk_LSHIFT( prevLag, 1 ) / 3; + } else if( Fs_kHz == 16 ) { + prevLag = silk_RSHIFT( prevLag, 1 ); + } + prevLag_log2 = silk_log2( (silk_float)prevLag ); + } else { + prevLag_log2 = 0; + } + + /* Set up stage 2 codebook based on number of subframes */ + if( nb_subfr == PE_MAX_NB_SUBFR ) { + cbk_size = PE_NB_CBKS_STAGE2_EXT; + Lag_CB_ptr = &silk_CB_lags_stage2[ 0 ][ 0 ]; + if( Fs_kHz == 8 && complexity > SILK_PE_MIN_COMPLEX ) { + /* If input is 8 khz use a larger codebook here because it is last stage */ + nb_cbk_search = PE_NB_CBKS_STAGE2_EXT; + } else { + nb_cbk_search = PE_NB_CBKS_STAGE2; + } + } else { + cbk_size = PE_NB_CBKS_STAGE2_10MS; + Lag_CB_ptr = &silk_CB_lags_stage2_10_ms[ 0 ][ 0 ]; + nb_cbk_search = PE_NB_CBKS_STAGE2_10MS; + } + + for( k = 0; k < length_d_srch; k++ ) { + d = d_srch[ k ]; + for( j = 0; j < nb_cbk_search; j++ ) { + CC[j] = 0.0f; + for( i = 0; i < nb_subfr; i++ ) { + /* Try all codebooks */ + CC[ j ] += C[ i ][ d + matrix_ptr( Lag_CB_ptr, i, j, cbk_size )]; + } + } + /* Find best codebook */ + CCmax_new = -1000.0f; + CBimax_new = 0; + for( i = 0; i < nb_cbk_search; i++ ) { + if( CC[ i ] > CCmax_new ) { + CCmax_new = CC[ i ]; + CBimax_new = i; + } + } + + /* Bias towards shorter lags */ + lag_log2 = silk_log2( (silk_float)d ); + CCmax_new_b = CCmax_new - PE_SHORTLAG_BIAS * nb_subfr * lag_log2; + + /* Bias towards previous lag */ + if( prevLag > 0 ) { + delta_lag_log2_sqr = lag_log2 - prevLag_log2; + delta_lag_log2_sqr *= delta_lag_log2_sqr; + CCmax_new_b -= PE_PREVLAG_BIAS * nb_subfr * (*LTPCorr) * delta_lag_log2_sqr / ( delta_lag_log2_sqr + 0.5f ); + } + + if( CCmax_new_b > CCmax_b && /* Find maximum biased correlation */ + CCmax_new > nb_subfr * search_thres2 /* Correlation needs to be high enough to be voiced */ + ) { + CCmax_b = CCmax_new_b; + CCmax = CCmax_new; + lag = d; + CBimax = CBimax_new; + } + } + + if( lag == -1 ) { + /* No suitable candidate found */ + silk_memset( pitch_out, 0, PE_MAX_NB_SUBFR * sizeof(opus_int) ); + *LTPCorr = 0.0f; + *lagIndex = 0; + *contourIndex = 0; + return 1; + } + + /* Output normalized correlation */ + *LTPCorr = (silk_float)( CCmax / nb_subfr ); + silk_assert( *LTPCorr >= 0.0f ); + + if( Fs_kHz > 8 ) { + /* Search in original signal */ + + /* Compensate for decimation */ + silk_assert( lag == silk_SAT16( lag ) ); + if( Fs_kHz == 12 ) { + lag = silk_RSHIFT_ROUND( silk_SMULBB( lag, 3 ), 1 ); + } else { /* Fs_kHz == 16 */ + lag = silk_LSHIFT( lag, 1 ); + } + + lag = silk_LIMIT_int( lag, min_lag, max_lag ); + start_lag = silk_max_int( lag - 2, min_lag ); + end_lag = silk_min_int( lag + 2, max_lag ); + lag_new = lag; /* to avoid undefined lag */ + CBimax = 0; /* to avoid undefined lag */ + + CCmax = -1000.0f; + + /* Calculate the correlations and energies needed in stage 3 */ + silk_P_Ana_calc_corr_st3( cross_corr_st3, frame, start_lag, sf_length, nb_subfr, complexity, arch ); + silk_P_Ana_calc_energy_st3( energies_st3, frame, start_lag, sf_length, nb_subfr, complexity ); + + lag_counter = 0; + silk_assert( lag == silk_SAT16( lag ) ); + contour_bias = PE_FLATCONTOUR_BIAS / lag; + + /* Set up cbk parameters according to complexity setting and frame length */ + if( nb_subfr == PE_MAX_NB_SUBFR ) { + nb_cbk_search = (opus_int)silk_nb_cbk_searchs_stage3[ complexity ]; + cbk_size = PE_NB_CBKS_STAGE3_MAX; + Lag_CB_ptr = &silk_CB_lags_stage3[ 0 ][ 0 ]; + } else { + nb_cbk_search = PE_NB_CBKS_STAGE3_10MS; + cbk_size = PE_NB_CBKS_STAGE3_10MS; + Lag_CB_ptr = &silk_CB_lags_stage3_10_ms[ 0 ][ 0 ]; + } + + target_ptr = &frame[ PE_LTP_MEM_LENGTH_MS * Fs_kHz ]; + energy_tmp = silk_energy_FLP( target_ptr, nb_subfr * sf_length ) + 1.0; + for( d = start_lag; d <= end_lag; d++ ) { + for( j = 0; j < nb_cbk_search; j++ ) { + cross_corr = 0.0; + energy = energy_tmp; + for( k = 0; k < nb_subfr; k++ ) { + cross_corr += cross_corr_st3[ k ][ j ][ lag_counter ]; + energy += energies_st3[ k ][ j ][ lag_counter ]; + } + if( cross_corr > 0.0 ) { + CCmax_new = (silk_float)( 2 * cross_corr / energy ); + /* Reduce depending on flatness of contour */ + CCmax_new *= 1.0f - contour_bias * j; + } else { + CCmax_new = 0.0f; + } + + if( CCmax_new > CCmax && ( d + (opus_int)silk_CB_lags_stage3[ 0 ][ j ] ) <= max_lag ) { + CCmax = CCmax_new; + lag_new = d; + CBimax = j; + } + } + lag_counter++; + } + + for( k = 0; k < nb_subfr; k++ ) { + pitch_out[ k ] = lag_new + matrix_ptr( Lag_CB_ptr, k, CBimax, cbk_size ); + pitch_out[ k ] = silk_LIMIT( pitch_out[ k ], min_lag, PE_MAX_LAG_MS * Fs_kHz ); + } + *lagIndex = (opus_int16)( lag_new - min_lag ); + *contourIndex = (opus_int8)CBimax; + } else { /* Fs_kHz == 8 */ + /* Save Lags */ + for( k = 0; k < nb_subfr; k++ ) { + pitch_out[ k ] = lag + matrix_ptr( Lag_CB_ptr, k, CBimax, cbk_size ); + pitch_out[ k ] = silk_LIMIT( pitch_out[ k ], min_lag_8kHz, PE_MAX_LAG_MS * 8 ); + } + *lagIndex = (opus_int16)( lag - min_lag_8kHz ); + *contourIndex = (opus_int8)CBimax; + } + celt_assert( *lagIndex >= 0 ); + /* return as voiced */ + return 0; +} + +/*********************************************************************** + * Calculates the correlations used in stage 3 search. In order to cover + * the whole lag codebook for all the searched offset lags (lag +- 2), + * the following correlations are needed in each sub frame: + * + * sf1: lag range [-8,...,7] total 16 correlations + * sf2: lag range [-4,...,4] total 9 correlations + * sf3: lag range [-3,....4] total 8 correltions + * sf4: lag range [-6,....8] total 15 correlations + * + * In total 48 correlations. The direct implementation computed in worst + * case 4*12*5 = 240 correlations, but more likely around 120. + ***********************************************************************/ +static void silk_P_Ana_calc_corr_st3( + silk_float cross_corr_st3[ PE_MAX_NB_SUBFR ][ PE_NB_CBKS_STAGE3_MAX ][ PE_NB_STAGE3_LAGS ], /* O 3 DIM correlation array */ + const silk_float frame[], /* I vector to correlate */ + opus_int start_lag, /* I start lag */ + opus_int sf_length, /* I sub frame length */ + opus_int nb_subfr, /* I number of subframes */ + opus_int complexity, /* I Complexity setting */ + int arch /* I Run-time architecture */ +) +{ + const silk_float *target_ptr; + opus_int i, j, k, lag_counter, lag_low, lag_high; + opus_int nb_cbk_search, delta, idx, cbk_size; + silk_float scratch_mem[ SCRATCH_SIZE ]; + opus_val32 xcorr[ SCRATCH_SIZE ]; + const opus_int8 *Lag_range_ptr, *Lag_CB_ptr; + + celt_assert( complexity >= SILK_PE_MIN_COMPLEX ); + celt_assert( complexity <= SILK_PE_MAX_COMPLEX ); + + if( nb_subfr == PE_MAX_NB_SUBFR ) { + Lag_range_ptr = &silk_Lag_range_stage3[ complexity ][ 0 ][ 0 ]; + Lag_CB_ptr = &silk_CB_lags_stage3[ 0 ][ 0 ]; + nb_cbk_search = silk_nb_cbk_searchs_stage3[ complexity ]; + cbk_size = PE_NB_CBKS_STAGE3_MAX; + } else { + celt_assert( nb_subfr == PE_MAX_NB_SUBFR >> 1); + Lag_range_ptr = &silk_Lag_range_stage3_10_ms[ 0 ][ 0 ]; + Lag_CB_ptr = &silk_CB_lags_stage3_10_ms[ 0 ][ 0 ]; + nb_cbk_search = PE_NB_CBKS_STAGE3_10MS; + cbk_size = PE_NB_CBKS_STAGE3_10MS; + } + + target_ptr = &frame[ silk_LSHIFT( sf_length, 2 ) ]; /* Pointer to middle of frame */ + for( k = 0; k < nb_subfr; k++ ) { + lag_counter = 0; + + /* Calculate the correlations for each subframe */ + lag_low = matrix_ptr( Lag_range_ptr, k, 0, 2 ); + lag_high = matrix_ptr( Lag_range_ptr, k, 1, 2 ); + silk_assert(lag_high-lag_low+1 <= SCRATCH_SIZE); + celt_pitch_xcorr( target_ptr, target_ptr - start_lag - lag_high, xcorr, sf_length, lag_high - lag_low + 1, arch ); + for( j = lag_low; j <= lag_high; j++ ) { + silk_assert( lag_counter < SCRATCH_SIZE ); + scratch_mem[ lag_counter ] = xcorr[ lag_high - j ]; + lag_counter++; + } + + delta = matrix_ptr( Lag_range_ptr, k, 0, 2 ); + for( i = 0; i < nb_cbk_search; i++ ) { + /* Fill out the 3 dim array that stores the correlations for */ + /* each code_book vector for each start lag */ + idx = matrix_ptr( Lag_CB_ptr, k, i, cbk_size ) - delta; + for( j = 0; j < PE_NB_STAGE3_LAGS; j++ ) { + silk_assert( idx + j < SCRATCH_SIZE ); + silk_assert( idx + j < lag_counter ); + cross_corr_st3[ k ][ i ][ j ] = scratch_mem[ idx + j ]; + } + } + target_ptr += sf_length; + } +} + +/********************************************************************/ +/* Calculate the energies for first two subframes. The energies are */ +/* calculated recursively. */ +/********************************************************************/ +static void silk_P_Ana_calc_energy_st3( + silk_float energies_st3[ PE_MAX_NB_SUBFR ][ PE_NB_CBKS_STAGE3_MAX ][ PE_NB_STAGE3_LAGS ], /* O 3 DIM correlation array */ + const silk_float frame[], /* I vector to correlate */ + opus_int start_lag, /* I start lag */ + opus_int sf_length, /* I sub frame length */ + opus_int nb_subfr, /* I number of subframes */ + opus_int complexity /* I Complexity setting */ +) +{ + const silk_float *target_ptr, *basis_ptr; + double energy; + opus_int k, i, j, lag_counter; + opus_int nb_cbk_search, delta, idx, cbk_size, lag_diff; + silk_float scratch_mem[ SCRATCH_SIZE ]; + const opus_int8 *Lag_range_ptr, *Lag_CB_ptr; + + celt_assert( complexity >= SILK_PE_MIN_COMPLEX ); + celt_assert( complexity <= SILK_PE_MAX_COMPLEX ); + + if( nb_subfr == PE_MAX_NB_SUBFR ) { + Lag_range_ptr = &silk_Lag_range_stage3[ complexity ][ 0 ][ 0 ]; + Lag_CB_ptr = &silk_CB_lags_stage3[ 0 ][ 0 ]; + nb_cbk_search = silk_nb_cbk_searchs_stage3[ complexity ]; + cbk_size = PE_NB_CBKS_STAGE3_MAX; + } else { + celt_assert( nb_subfr == PE_MAX_NB_SUBFR >> 1); + Lag_range_ptr = &silk_Lag_range_stage3_10_ms[ 0 ][ 0 ]; + Lag_CB_ptr = &silk_CB_lags_stage3_10_ms[ 0 ][ 0 ]; + nb_cbk_search = PE_NB_CBKS_STAGE3_10MS; + cbk_size = PE_NB_CBKS_STAGE3_10MS; + } + + target_ptr = &frame[ silk_LSHIFT( sf_length, 2 ) ]; + for( k = 0; k < nb_subfr; k++ ) { + lag_counter = 0; + + /* Calculate the energy for first lag */ + basis_ptr = target_ptr - ( start_lag + matrix_ptr( Lag_range_ptr, k, 0, 2 ) ); + energy = silk_energy_FLP( basis_ptr, sf_length ) + 1e-3; + silk_assert( energy >= 0.0 ); + scratch_mem[lag_counter] = (silk_float)energy; + lag_counter++; + + lag_diff = ( matrix_ptr( Lag_range_ptr, k, 1, 2 ) - matrix_ptr( Lag_range_ptr, k, 0, 2 ) + 1 ); + for( i = 1; i < lag_diff; i++ ) { + /* remove part outside new window */ + energy -= basis_ptr[sf_length - i] * (double)basis_ptr[sf_length - i]; + silk_assert( energy >= 0.0 ); + + /* add part that comes into window */ + energy += basis_ptr[ -i ] * (double)basis_ptr[ -i ]; + silk_assert( energy >= 0.0 ); + silk_assert( lag_counter < SCRATCH_SIZE ); + scratch_mem[lag_counter] = (silk_float)energy; + lag_counter++; + } + + delta = matrix_ptr( Lag_range_ptr, k, 0, 2 ); + for( i = 0; i < nb_cbk_search; i++ ) { + /* Fill out the 3 dim array that stores the correlations for */ + /* each code_book vector for each start lag */ + idx = matrix_ptr( Lag_CB_ptr, k, i, cbk_size ) - delta; + for( j = 0; j < PE_NB_STAGE3_LAGS; j++ ) { + silk_assert( idx + j < SCRATCH_SIZE ); + silk_assert( idx + j < lag_counter ); + energies_st3[ k ][ i ][ j ] = scratch_mem[ idx + j ]; + silk_assert( energies_st3[ k ][ i ][ j ] >= 0.0f ); + } + } + target_ptr += sf_length; + } +} diff --git a/media/libopus/silk/float/process_gains_FLP.c b/media/libopus/silk/float/process_gains_FLP.c new file mode 100644 index 0000000000..c0da0dae44 --- /dev/null +++ b/media/libopus/silk/float/process_gains_FLP.c @@ -0,0 +1,103 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main_FLP.h" +#include "tuning_parameters.h" + +/* Processing of gains */ +void silk_process_gains_FLP( + silk_encoder_state_FLP *psEnc, /* I/O Encoder state FLP */ + silk_encoder_control_FLP *psEncCtrl, /* I/O Encoder control FLP */ + opus_int condCoding /* I The type of conditional coding to use */ +) +{ + silk_shape_state_FLP *psShapeSt = &psEnc->sShape; + opus_int k; + opus_int32 pGains_Q16[ MAX_NB_SUBFR ]; + silk_float s, InvMaxSqrVal, gain, quant_offset; + + /* Gain reduction when LTP coding gain is high */ + if( psEnc->sCmn.indices.signalType == TYPE_VOICED ) { + s = 1.0f - 0.5f * silk_sigmoid( 0.25f * ( psEncCtrl->LTPredCodGain - 12.0f ) ); + for( k = 0; k < psEnc->sCmn.nb_subfr; k++ ) { + psEncCtrl->Gains[ k ] *= s; + } + } + + /* Limit the quantized signal */ + InvMaxSqrVal = ( silk_float )( pow( 2.0f, 0.33f * ( 21.0f - psEnc->sCmn.SNR_dB_Q7 * ( 1 / 128.0f ) ) ) / psEnc->sCmn.subfr_length ); + + for( k = 0; k < psEnc->sCmn.nb_subfr; k++ ) { + /* Soft limit on ratio residual energy and squared gains */ + gain = psEncCtrl->Gains[ k ]; + gain = ( silk_float )sqrt( gain * gain + psEncCtrl->ResNrg[ k ] * InvMaxSqrVal ); + psEncCtrl->Gains[ k ] = silk_min_float( gain, 32767.0f ); + } + + /* Prepare gains for noise shaping quantization */ + for( k = 0; k < psEnc->sCmn.nb_subfr; k++ ) { + pGains_Q16[ k ] = (opus_int32)( psEncCtrl->Gains[ k ] * 65536.0f ); + } + + /* Save unquantized gains and gain Index */ + silk_memcpy( psEncCtrl->GainsUnq_Q16, pGains_Q16, psEnc->sCmn.nb_subfr * sizeof( opus_int32 ) ); + psEncCtrl->lastGainIndexPrev = psShapeSt->LastGainIndex; + + /* Quantize gains */ + silk_gains_quant( psEnc->sCmn.indices.GainsIndices, pGains_Q16, + &psShapeSt->LastGainIndex, condCoding == CODE_CONDITIONALLY, psEnc->sCmn.nb_subfr ); + + /* Overwrite unquantized gains with quantized gains and convert back to Q0 from Q16 */ + for( k = 0; k < psEnc->sCmn.nb_subfr; k++ ) { + psEncCtrl->Gains[ k ] = pGains_Q16[ k ] / 65536.0f; + } + + /* Set quantizer offset for voiced signals. Larger offset when LTP coding gain is low or tilt is high (ie low-pass) */ + if( psEnc->sCmn.indices.signalType == TYPE_VOICED ) { + if( psEncCtrl->LTPredCodGain + psEnc->sCmn.input_tilt_Q15 * ( 1.0f / 32768.0f ) > 1.0f ) { + psEnc->sCmn.indices.quantOffsetType = 0; + } else { + psEnc->sCmn.indices.quantOffsetType = 1; + } + } + + /* Quantizer boundary adjustment */ + quant_offset = silk_Quantization_Offsets_Q10[ psEnc->sCmn.indices.signalType >> 1 ][ psEnc->sCmn.indices.quantOffsetType ] / 1024.0f; + psEncCtrl->Lambda = LAMBDA_OFFSET + + LAMBDA_DELAYED_DECISIONS * psEnc->sCmn.nStatesDelayedDecision + + LAMBDA_SPEECH_ACT * psEnc->sCmn.speech_activity_Q8 * ( 1.0f / 256.0f ) + + LAMBDA_INPUT_QUALITY * psEncCtrl->input_quality + + LAMBDA_CODING_QUALITY * psEncCtrl->coding_quality + + LAMBDA_QUANT_OFFSET * quant_offset; + + silk_assert( psEncCtrl->Lambda > 0.0f ); + silk_assert( psEncCtrl->Lambda < 2.0f ); +} diff --git a/media/libopus/silk/float/regularize_correlations_FLP.c b/media/libopus/silk/float/regularize_correlations_FLP.c new file mode 100644 index 0000000000..df4612604c --- /dev/null +++ b/media/libopus/silk/float/regularize_correlations_FLP.c @@ -0,0 +1,48 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main_FLP.h" + +/* Add noise to matrix diagonal */ +void silk_regularize_correlations_FLP( + silk_float *XX, /* I/O Correlation matrices */ + silk_float *xx, /* I/O Correlation values */ + const silk_float noise, /* I Noise energy to add */ + const opus_int D /* I Dimension of XX */ +) +{ + opus_int i; + + for( i = 0; i < D; i++ ) { + matrix_ptr( &XX[ 0 ], i, i, D ) += noise; + } + xx[ 0 ] += noise; +} diff --git a/media/libopus/silk/float/residual_energy_FLP.c b/media/libopus/silk/float/residual_energy_FLP.c new file mode 100644 index 0000000000..1bd07b33a4 --- /dev/null +++ b/media/libopus/silk/float/residual_energy_FLP.c @@ -0,0 +1,117 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main_FLP.h" + +#define MAX_ITERATIONS_RESIDUAL_NRG 10 +#define REGULARIZATION_FACTOR 1e-8f + +/* Residual energy: nrg = wxx - 2 * wXx * c + c' * wXX * c */ +silk_float silk_residual_energy_covar_FLP( /* O Weighted residual energy */ + const silk_float *c, /* I Filter coefficients */ + silk_float *wXX, /* I/O Weighted correlation matrix, reg. out */ + const silk_float *wXx, /* I Weighted correlation vector */ + const silk_float wxx, /* I Weighted correlation value */ + const opus_int D /* I Dimension */ +) +{ + opus_int i, j, k; + silk_float tmp, nrg = 0.0f, regularization; + + /* Safety checks */ + celt_assert( D >= 0 ); + + regularization = REGULARIZATION_FACTOR * ( wXX[ 0 ] + wXX[ D * D - 1 ] ); + for( k = 0; k < MAX_ITERATIONS_RESIDUAL_NRG; k++ ) { + nrg = wxx; + + tmp = 0.0f; + for( i = 0; i < D; i++ ) { + tmp += wXx[ i ] * c[ i ]; + } + nrg -= 2.0f * tmp; + + /* compute c' * wXX * c, assuming wXX is symmetric */ + for( i = 0; i < D; i++ ) { + tmp = 0.0f; + for( j = i + 1; j < D; j++ ) { + tmp += matrix_c_ptr( wXX, i, j, D ) * c[ j ]; + } + nrg += c[ i ] * ( 2.0f * tmp + matrix_c_ptr( wXX, i, i, D ) * c[ i ] ); + } + if( nrg > 0 ) { + break; + } else { + /* Add white noise */ + for( i = 0; i < D; i++ ) { + matrix_c_ptr( wXX, i, i, D ) += regularization; + } + /* Increase noise for next run */ + regularization *= 2.0f; + } + } + if( k == MAX_ITERATIONS_RESIDUAL_NRG ) { + silk_assert( nrg == 0 ); + nrg = 1.0f; + } + + return nrg; +} + +/* Calculates residual energies of input subframes where all subframes have LPC_order */ +/* of preceding samples */ +void silk_residual_energy_FLP( + silk_float nrgs[ MAX_NB_SUBFR ], /* O Residual energy per subframe */ + const silk_float x[], /* I Input signal */ + silk_float a[ 2 ][ MAX_LPC_ORDER ], /* I AR coefs for each frame half */ + const silk_float gains[], /* I Quantization gains */ + const opus_int subfr_length, /* I Subframe length */ + const opus_int nb_subfr, /* I number of subframes */ + const opus_int LPC_order /* I LPC order */ +) +{ + opus_int shift; + silk_float *LPC_res_ptr, LPC_res[ ( MAX_FRAME_LENGTH + MAX_NB_SUBFR * MAX_LPC_ORDER ) / 2 ]; + + LPC_res_ptr = LPC_res + LPC_order; + shift = LPC_order + subfr_length; + + /* Filter input to create the LPC residual for each frame half, and measure subframe energies */ + silk_LPC_analysis_filter_FLP( LPC_res, a[ 0 ], x + 0 * shift, 2 * shift, LPC_order ); + nrgs[ 0 ] = ( silk_float )( gains[ 0 ] * gains[ 0 ] * silk_energy_FLP( LPC_res_ptr + 0 * shift, subfr_length ) ); + nrgs[ 1 ] = ( silk_float )( gains[ 1 ] * gains[ 1 ] * silk_energy_FLP( LPC_res_ptr + 1 * shift, subfr_length ) ); + + if( nb_subfr == MAX_NB_SUBFR ) { + silk_LPC_analysis_filter_FLP( LPC_res, a[ 1 ], x + 2 * shift, 2 * shift, LPC_order ); + nrgs[ 2 ] = ( silk_float )( gains[ 2 ] * gains[ 2 ] * silk_energy_FLP( LPC_res_ptr + 0 * shift, subfr_length ) ); + nrgs[ 3 ] = ( silk_float )( gains[ 3 ] * gains[ 3 ] * silk_energy_FLP( LPC_res_ptr + 1 * shift, subfr_length ) ); + } +} diff --git a/media/libopus/silk/float/scale_copy_vector_FLP.c b/media/libopus/silk/float/scale_copy_vector_FLP.c new file mode 100644 index 0000000000..20db32b3b1 --- /dev/null +++ b/media/libopus/silk/float/scale_copy_vector_FLP.c @@ -0,0 +1,57 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "SigProc_FLP.h" + +/* copy and multiply a vector by a constant */ +void silk_scale_copy_vector_FLP( + silk_float *data_out, + const silk_float *data_in, + silk_float gain, + opus_int dataSize +) +{ + opus_int i, dataSize4; + + /* 4x unrolled loop */ + dataSize4 = dataSize & 0xFFFC; + for( i = 0; i < dataSize4; i += 4 ) { + data_out[ i + 0 ] = gain * data_in[ i + 0 ]; + data_out[ i + 1 ] = gain * data_in[ i + 1 ]; + data_out[ i + 2 ] = gain * data_in[ i + 2 ]; + data_out[ i + 3 ] = gain * data_in[ i + 3 ]; + } + + /* any remaining elements */ + for( ; i < dataSize; i++ ) { + data_out[ i ] = gain * data_in[ i ]; + } +} diff --git a/media/libopus/silk/float/scale_vector_FLP.c b/media/libopus/silk/float/scale_vector_FLP.c new file mode 100644 index 0000000000..108fdcbed5 --- /dev/null +++ b/media/libopus/silk/float/scale_vector_FLP.c @@ -0,0 +1,56 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "SigProc_FLP.h" + +/* multiply a vector by a constant */ +void silk_scale_vector_FLP( + silk_float *data1, + silk_float gain, + opus_int dataSize +) +{ + opus_int i, dataSize4; + + /* 4x unrolled loop */ + dataSize4 = dataSize & 0xFFFC; + for( i = 0; i < dataSize4; i += 4 ) { + data1[ i + 0 ] *= gain; + data1[ i + 1 ] *= gain; + data1[ i + 2 ] *= gain; + data1[ i + 3 ] *= gain; + } + + /* any remaining elements */ + for( ; i < dataSize; i++ ) { + data1[ i ] *= gain; + } +} diff --git a/media/libopus/silk/float/schur_FLP.c b/media/libopus/silk/float/schur_FLP.c new file mode 100644 index 0000000000..8526c748d3 --- /dev/null +++ b/media/libopus/silk/float/schur_FLP.c @@ -0,0 +1,70 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "SigProc_FLP.h" + +silk_float silk_schur_FLP( /* O returns residual energy */ + silk_float refl_coef[], /* O reflection coefficients (length order) */ + const silk_float auto_corr[], /* I autocorrelation sequence (length order+1) */ + opus_int order /* I order */ +) +{ + opus_int k, n; + double C[ SILK_MAX_ORDER_LPC + 1 ][ 2 ]; + double Ctmp1, Ctmp2, rc_tmp; + + celt_assert( order >= 0 && order <= SILK_MAX_ORDER_LPC ); + + /* Copy correlations */ + k = 0; + do { + C[ k ][ 0 ] = C[ k ][ 1 ] = auto_corr[ k ]; + } while( ++k <= order ); + + for( k = 0; k < order; k++ ) { + /* Get reflection coefficient */ + rc_tmp = -C[ k + 1 ][ 0 ] / silk_max_float( C[ 0 ][ 1 ], 1e-9f ); + + /* Save the output */ + refl_coef[ k ] = (silk_float)rc_tmp; + + /* Update correlations */ + for( n = 0; n < order - k; n++ ) { + Ctmp1 = C[ n + k + 1 ][ 0 ]; + Ctmp2 = C[ n ][ 1 ]; + C[ n + k + 1 ][ 0 ] = Ctmp1 + Ctmp2 * rc_tmp; + C[ n ][ 1 ] = Ctmp2 + Ctmp1 * rc_tmp; + } + } + + /* Return residual energy */ + return (silk_float)C[ 0 ][ 1 ]; +} diff --git a/media/libopus/silk/float/sort_FLP.c b/media/libopus/silk/float/sort_FLP.c new file mode 100644 index 0000000000..0e18f31950 --- /dev/null +++ b/media/libopus/silk/float/sort_FLP.c @@ -0,0 +1,83 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +/* Insertion sort (fast for already almost sorted arrays): */ +/* Best case: O(n) for an already sorted array */ +/* Worst case: O(n^2) for an inversely sorted array */ + +#include "typedef.h" +#include "SigProc_FLP.h" + +void silk_insertion_sort_decreasing_FLP( + silk_float *a, /* I/O Unsorted / Sorted vector */ + opus_int *idx, /* O Index vector for the sorted elements */ + const opus_int L, /* I Vector length */ + const opus_int K /* I Number of correctly sorted positions */ +) +{ + silk_float value; + opus_int i, j; + + /* Safety checks */ + celt_assert( K > 0 ); + celt_assert( L > 0 ); + celt_assert( L >= K ); + + /* Write start indices in index vector */ + for( i = 0; i < K; i++ ) { + idx[ i ] = i; + } + + /* Sort vector elements by value, decreasing order */ + for( i = 1; i < K; i++ ) { + value = a[ i ]; + for( j = i - 1; ( j >= 0 ) && ( value > a[ j ] ); j-- ) { + a[ j + 1 ] = a[ j ]; /* Shift value */ + idx[ j + 1 ] = idx[ j ]; /* Shift index */ + } + a[ j + 1 ] = value; /* Write value */ + idx[ j + 1 ] = i; /* Write index */ + } + + /* If less than L values are asked check the remaining values, */ + /* but only spend CPU to ensure that the K first values are correct */ + for( i = K; i < L; i++ ) { + value = a[ i ]; + if( value > a[ K - 1 ] ) { + for( j = K - 2; ( j >= 0 ) && ( value > a[ j ] ); j-- ) { + a[ j + 1 ] = a[ j ]; /* Shift value */ + idx[ j + 1 ] = idx[ j ]; /* Shift index */ + } + a[ j + 1 ] = value; /* Write value */ + idx[ j + 1 ] = i; /* Write index */ + } + } +} diff --git a/media/libopus/silk/float/structs_FLP.h b/media/libopus/silk/float/structs_FLP.h new file mode 100644 index 0000000000..3150b386e4 --- /dev/null +++ b/media/libopus/silk/float/structs_FLP.h @@ -0,0 +1,112 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef SILK_STRUCTS_FLP_H +#define SILK_STRUCTS_FLP_H + +#include "typedef.h" +#include "main.h" +#include "structs.h" + +#ifdef __cplusplus +extern "C" +{ +#endif + +/********************************/ +/* Noise shaping analysis state */ +/********************************/ +typedef struct { + opus_int8 LastGainIndex; + silk_float HarmShapeGain_smth; + silk_float Tilt_smth; +} silk_shape_state_FLP; + +/********************************/ +/* Encoder state FLP */ +/********************************/ +typedef struct { + silk_encoder_state sCmn; /* Common struct, shared with fixed-point code */ + silk_shape_state_FLP sShape; /* Noise shaping state */ + + /* Buffer for find pitch and noise shape analysis */ + silk_float x_buf[ 2 * MAX_FRAME_LENGTH + LA_SHAPE_MAX ];/* Buffer for find pitch and noise shape analysis */ + silk_float LTPCorr; /* Normalized correlation from pitch lag estimator */ +} silk_encoder_state_FLP; + +/************************/ +/* Encoder control FLP */ +/************************/ +typedef struct { + /* Prediction and coding parameters */ + silk_float Gains[ MAX_NB_SUBFR ]; + silk_float PredCoef[ 2 ][ MAX_LPC_ORDER ]; /* holds interpolated and final coefficients */ + silk_float LTPCoef[LTP_ORDER * MAX_NB_SUBFR]; + silk_float LTP_scale; + opus_int pitchL[ MAX_NB_SUBFR ]; + + /* Noise shaping parameters */ + silk_float AR[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ]; + silk_float LF_MA_shp[ MAX_NB_SUBFR ]; + silk_float LF_AR_shp[ MAX_NB_SUBFR ]; + silk_float Tilt[ MAX_NB_SUBFR ]; + silk_float HarmShapeGain[ MAX_NB_SUBFR ]; + silk_float Lambda; + silk_float input_quality; + silk_float coding_quality; + + /* Measures */ + silk_float predGain; + silk_float LTPredCodGain; + silk_float ResNrg[ MAX_NB_SUBFR ]; /* Residual energy per subframe */ + + /* Parameters for CBR mode */ + opus_int32 GainsUnq_Q16[ MAX_NB_SUBFR ]; + opus_int8 lastGainIndexPrev; +} silk_encoder_control_FLP; + +/************************/ +/* Encoder Super Struct */ +/************************/ +typedef struct { + silk_encoder_state_FLP state_Fxx[ ENCODER_NUM_CHANNELS ]; + stereo_enc_state sStereo; + opus_int32 nBitsUsedLBRR; + opus_int32 nBitsExceeded; + opus_int nChannelsAPI; + opus_int nChannelsInternal; + opus_int nPrevChannelsInternal; + opus_int timeSinceSwitchAllowed_ms; + opus_int allowBandwidthSwitch; + opus_int prev_decode_only_middle; +} silk_encoder; + +#ifdef __cplusplus +} +#endif + +#endif diff --git a/media/libopus/silk/float/warped_autocorrelation_FLP.c b/media/libopus/silk/float/warped_autocorrelation_FLP.c new file mode 100644 index 0000000000..09186e73d4 --- /dev/null +++ b/media/libopus/silk/float/warped_autocorrelation_FLP.c @@ -0,0 +1,73 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main_FLP.h" + +/* Autocorrelations for a warped frequency axis */ +void silk_warped_autocorrelation_FLP( + silk_float *corr, /* O Result [order + 1] */ + const silk_float *input, /* I Input data to correlate */ + const silk_float warping, /* I Warping coefficient */ + const opus_int length, /* I Length of input */ + const opus_int order /* I Correlation order (even) */ +) +{ + opus_int n, i; + double tmp1, tmp2; + double state[ MAX_SHAPE_LPC_ORDER + 1 ] = { 0 }; + double C[ MAX_SHAPE_LPC_ORDER + 1 ] = { 0 }; + + /* Order must be even */ + celt_assert( ( order & 1 ) == 0 ); + + /* Loop over samples */ + for( n = 0; n < length; n++ ) { + tmp1 = input[ n ]; + /* Loop over allpass sections */ + for( i = 0; i < order; i += 2 ) { + /* Output of allpass section */ + tmp2 = state[ i ] + warping * ( state[ i + 1 ] - tmp1 ); + state[ i ] = tmp1; + C[ i ] += state[ 0 ] * tmp1; + /* Output of allpass section */ + tmp1 = state[ i + 1 ] + warping * ( state[ i + 2 ] - tmp2 ); + state[ i + 1 ] = tmp2; + C[ i + 1 ] += state[ 0 ] * tmp2; + } + state[ order ] = tmp1; + C[ order ] += state[ 0 ] * tmp1; + } + + /* Copy correlations in silk_float output format */ + for( i = 0; i < order + 1; i++ ) { + corr[ i ] = ( silk_float )C[ i ]; + } +} diff --git a/media/libopus/silk/float/wrappers_FLP.c b/media/libopus/silk/float/wrappers_FLP.c new file mode 100644 index 0000000000..c0c183e35a --- /dev/null +++ b/media/libopus/silk/float/wrappers_FLP.c @@ -0,0 +1,209 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main_FLP.h" + +/* Wrappers. Calls flp / fix code */ + +/* Convert AR filter coefficients to NLSF parameters */ +void silk_A2NLSF_FLP( + opus_int16 *NLSF_Q15, /* O NLSF vector [ LPC_order ] */ + const silk_float *pAR, /* I LPC coefficients [ LPC_order ] */ + const opus_int LPC_order /* I LPC order */ +) +{ + opus_int i; + opus_int32 a_fix_Q16[ MAX_LPC_ORDER ]; + + for( i = 0; i < LPC_order; i++ ) { + a_fix_Q16[ i ] = silk_float2int( pAR[ i ] * 65536.0f ); + } + + silk_A2NLSF( NLSF_Q15, a_fix_Q16, LPC_order ); +} + +/* Convert LSF parameters to AR prediction filter coefficients */ +void silk_NLSF2A_FLP( + silk_float *pAR, /* O LPC coefficients [ LPC_order ] */ + const opus_int16 *NLSF_Q15, /* I NLSF vector [ LPC_order ] */ + const opus_int LPC_order, /* I LPC order */ + int arch /* I Run-time architecture */ +) +{ + opus_int i; + opus_int16 a_fix_Q12[ MAX_LPC_ORDER ]; + + silk_NLSF2A( a_fix_Q12, NLSF_Q15, LPC_order, arch ); + + for( i = 0; i < LPC_order; i++ ) { + pAR[ i ] = ( silk_float )a_fix_Q12[ i ] * ( 1.0f / 4096.0f ); + } +} + +/******************************************/ +/* Floating-point NLSF processing wrapper */ +/******************************************/ +void silk_process_NLSFs_FLP( + silk_encoder_state *psEncC, /* I/O Encoder state */ + silk_float PredCoef[ 2 ][ MAX_LPC_ORDER ], /* O Prediction coefficients */ + opus_int16 NLSF_Q15[ MAX_LPC_ORDER ], /* I/O Normalized LSFs (quant out) (0 - (2^15-1)) */ + const opus_int16 prev_NLSF_Q15[ MAX_LPC_ORDER ] /* I Previous Normalized LSFs (0 - (2^15-1)) */ +) +{ + opus_int i, j; + opus_int16 PredCoef_Q12[ 2 ][ MAX_LPC_ORDER ]; + + silk_process_NLSFs( psEncC, PredCoef_Q12, NLSF_Q15, prev_NLSF_Q15); + + for( j = 0; j < 2; j++ ) { + for( i = 0; i < psEncC->predictLPCOrder; i++ ) { + PredCoef[ j ][ i ] = ( silk_float )PredCoef_Q12[ j ][ i ] * ( 1.0f / 4096.0f ); + } + } +} + +/****************************************/ +/* Floating-point Silk NSQ wrapper */ +/****************************************/ +void silk_NSQ_wrapper_FLP( + silk_encoder_state_FLP *psEnc, /* I/O Encoder state FLP */ + silk_encoder_control_FLP *psEncCtrl, /* I/O Encoder control FLP */ + SideInfoIndices *psIndices, /* I/O Quantization indices */ + silk_nsq_state *psNSQ, /* I/O Noise Shaping Quantzation state */ + opus_int8 pulses[], /* O Quantized pulse signal */ + const silk_float x[] /* I Prefiltered input signal */ +) +{ + opus_int i, j; + opus_int16 x16[ MAX_FRAME_LENGTH ]; + opus_int32 Gains_Q16[ MAX_NB_SUBFR ]; + silk_DWORD_ALIGN opus_int16 PredCoef_Q12[ 2 ][ MAX_LPC_ORDER ]; + opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ]; + opus_int LTP_scale_Q14; + + /* Noise shaping parameters */ + opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ]; + opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ]; /* Packs two int16 coefficients per int32 value */ + opus_int Lambda_Q10; + opus_int Tilt_Q14[ MAX_NB_SUBFR ]; + opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ]; + + /* Convert control struct to fix control struct */ + /* Noise shape parameters */ + for( i = 0; i < psEnc->sCmn.nb_subfr; i++ ) { + for( j = 0; j < psEnc->sCmn.shapingLPCOrder; j++ ) { + AR_Q13[ i * MAX_SHAPE_LPC_ORDER + j ] = silk_float2int( psEncCtrl->AR[ i * MAX_SHAPE_LPC_ORDER + j ] * 8192.0f ); + } + } + + for( i = 0; i < psEnc->sCmn.nb_subfr; i++ ) { + LF_shp_Q14[ i ] = silk_LSHIFT32( silk_float2int( psEncCtrl->LF_AR_shp[ i ] * 16384.0f ), 16 ) | + (opus_uint16)silk_float2int( psEncCtrl->LF_MA_shp[ i ] * 16384.0f ); + Tilt_Q14[ i ] = (opus_int)silk_float2int( psEncCtrl->Tilt[ i ] * 16384.0f ); + HarmShapeGain_Q14[ i ] = (opus_int)silk_float2int( psEncCtrl->HarmShapeGain[ i ] * 16384.0f ); + } + Lambda_Q10 = ( opus_int )silk_float2int( psEncCtrl->Lambda * 1024.0f ); + + /* prediction and coding parameters */ + for( i = 0; i < psEnc->sCmn.nb_subfr * LTP_ORDER; i++ ) { + LTPCoef_Q14[ i ] = (opus_int16)silk_float2int( psEncCtrl->LTPCoef[ i ] * 16384.0f ); + } + + for( j = 0; j < 2; j++ ) { + for( i = 0; i < psEnc->sCmn.predictLPCOrder; i++ ) { + PredCoef_Q12[ j ][ i ] = (opus_int16)silk_float2int( psEncCtrl->PredCoef[ j ][ i ] * 4096.0f ); + } + } + + for( i = 0; i < psEnc->sCmn.nb_subfr; i++ ) { + Gains_Q16[ i ] = silk_float2int( psEncCtrl->Gains[ i ] * 65536.0f ); + silk_assert( Gains_Q16[ i ] > 0 ); + } + + if( psIndices->signalType == TYPE_VOICED ) { + LTP_scale_Q14 = silk_LTPScales_table_Q14[ psIndices->LTP_scaleIndex ]; + } else { + LTP_scale_Q14 = 0; + } + + /* Convert input to fix */ + for( i = 0; i < psEnc->sCmn.frame_length; i++ ) { + x16[ i ] = silk_float2int( x[ i ] ); + } + + /* Call NSQ */ + if( psEnc->sCmn.nStatesDelayedDecision > 1 || psEnc->sCmn.warping_Q16 > 0 ) { + silk_NSQ_del_dec( &psEnc->sCmn, psNSQ, psIndices, x16, pulses, PredCoef_Q12[ 0 ], LTPCoef_Q14, + AR_Q13, HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, psEncCtrl->pitchL, Lambda_Q10, LTP_scale_Q14, psEnc->sCmn.arch ); + } else { + silk_NSQ( &psEnc->sCmn, psNSQ, psIndices, x16, pulses, PredCoef_Q12[ 0 ], LTPCoef_Q14, + AR_Q13, HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, psEncCtrl->pitchL, Lambda_Q10, LTP_scale_Q14, psEnc->sCmn.arch ); + } +} + +/***********************************************/ +/* Floating-point Silk LTP quantiation wrapper */ +/***********************************************/ +void silk_quant_LTP_gains_FLP( + silk_float B[ MAX_NB_SUBFR * LTP_ORDER ], /* O Quantized LTP gains */ + opus_int8 cbk_index[ MAX_NB_SUBFR ], /* O Codebook index */ + opus_int8 *periodicity_index, /* O Periodicity index */ + opus_int32 *sum_log_gain_Q7, /* I/O Cumulative max prediction gain */ + silk_float *pred_gain_dB, /* O LTP prediction gain */ + const silk_float XX[ MAX_NB_SUBFR * LTP_ORDER * LTP_ORDER ], /* I Correlation matrix */ + const silk_float xX[ MAX_NB_SUBFR * LTP_ORDER ], /* I Correlation vector */ + const opus_int subfr_len, /* I Number of samples per subframe */ + const opus_int nb_subfr, /* I Number of subframes */ + int arch /* I Run-time architecture */ +) +{ + opus_int i, pred_gain_dB_Q7; + opus_int16 B_Q14[ MAX_NB_SUBFR * LTP_ORDER ]; + opus_int32 XX_Q17[ MAX_NB_SUBFR * LTP_ORDER * LTP_ORDER ]; + opus_int32 xX_Q17[ MAX_NB_SUBFR * LTP_ORDER ]; + + i = 0; + do { + XX_Q17[ i ] = (opus_int32)silk_float2int( XX[ i ] * 131072.0f ); + } while ( ++i < nb_subfr * LTP_ORDER * LTP_ORDER ); + i = 0; + do { + xX_Q17[ i ] = (opus_int32)silk_float2int( xX[ i ] * 131072.0f ); + } while ( ++i < nb_subfr * LTP_ORDER ); + + silk_quant_LTP_gains( B_Q14, cbk_index, periodicity_index, sum_log_gain_Q7, &pred_gain_dB_Q7, XX_Q17, xX_Q17, subfr_len, nb_subfr, arch ); + + for( i = 0; i < nb_subfr * LTP_ORDER; i++ ) { + B[ i ] = (silk_float)B_Q14[ i ] * ( 1.0f / 16384.0f ); + } + + *pred_gain_dB = (silk_float)pred_gain_dB_Q7 * ( 1.0f / 128.0f ); +} diff --git a/media/libopus/silk/gain_quant.c b/media/libopus/silk/gain_quant.c new file mode 100644 index 0000000000..ee65245aa3 --- /dev/null +++ b/media/libopus/silk/gain_quant.c @@ -0,0 +1,142 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" + +#define OFFSET ( ( MIN_QGAIN_DB * 128 ) / 6 + 16 * 128 ) +#define SCALE_Q16 ( ( 65536 * ( N_LEVELS_QGAIN - 1 ) ) / ( ( ( MAX_QGAIN_DB - MIN_QGAIN_DB ) * 128 ) / 6 ) ) +#define INV_SCALE_Q16 ( ( 65536 * ( ( ( MAX_QGAIN_DB - MIN_QGAIN_DB ) * 128 ) / 6 ) ) / ( N_LEVELS_QGAIN - 1 ) ) + +/* Gain scalar quantization with hysteresis, uniform on log scale */ +void silk_gains_quant( + opus_int8 ind[ MAX_NB_SUBFR ], /* O gain indices */ + opus_int32 gain_Q16[ MAX_NB_SUBFR ], /* I/O gains (quantized out) */ + opus_int8 *prev_ind, /* I/O last index in previous frame */ + const opus_int conditional, /* I first gain is delta coded if 1 */ + const opus_int nb_subfr /* I number of subframes */ +) +{ + opus_int k, double_step_size_threshold; + + for( k = 0; k < nb_subfr; k++ ) { + /* Convert to log scale, scale, floor() */ + ind[ k ] = silk_SMULWB( SCALE_Q16, silk_lin2log( gain_Q16[ k ] ) - OFFSET ); + + /* Round towards previous quantized gain (hysteresis) */ + if( ind[ k ] < *prev_ind ) { + ind[ k ]++; + } + ind[ k ] = silk_LIMIT_int( ind[ k ], 0, N_LEVELS_QGAIN - 1 ); + + /* Compute delta indices and limit */ + if( k == 0 && conditional == 0 ) { + /* Full index */ + ind[ k ] = silk_LIMIT_int( ind[ k ], *prev_ind + MIN_DELTA_GAIN_QUANT, N_LEVELS_QGAIN - 1 ); + *prev_ind = ind[ k ]; + } else { + /* Delta index */ + ind[ k ] = ind[ k ] - *prev_ind; + + /* Double the quantization step size for large gain increases, so that the max gain level can be reached */ + double_step_size_threshold = 2 * MAX_DELTA_GAIN_QUANT - N_LEVELS_QGAIN + *prev_ind; + if( ind[ k ] > double_step_size_threshold ) { + ind[ k ] = double_step_size_threshold + silk_RSHIFT( ind[ k ] - double_step_size_threshold + 1, 1 ); + } + + ind[ k ] = silk_LIMIT_int( ind[ k ], MIN_DELTA_GAIN_QUANT, MAX_DELTA_GAIN_QUANT ); + + /* Accumulate deltas */ + if( ind[ k ] > double_step_size_threshold ) { + *prev_ind += silk_LSHIFT( ind[ k ], 1 ) - double_step_size_threshold; + *prev_ind = silk_min_int( *prev_ind, N_LEVELS_QGAIN - 1 ); + } else { + *prev_ind += ind[ k ]; + } + + /* Shift to make non-negative */ + ind[ k ] -= MIN_DELTA_GAIN_QUANT; + } + + /* Scale and convert to linear scale */ + gain_Q16[ k ] = silk_log2lin( silk_min_32( silk_SMULWB( INV_SCALE_Q16, *prev_ind ) + OFFSET, 3967 ) ); /* 3967 = 31 in Q7 */ + } +} + +/* Gains scalar dequantization, uniform on log scale */ +void silk_gains_dequant( + opus_int32 gain_Q16[ MAX_NB_SUBFR ], /* O quantized gains */ + const opus_int8 ind[ MAX_NB_SUBFR ], /* I gain indices */ + opus_int8 *prev_ind, /* I/O last index in previous frame */ + const opus_int conditional, /* I first gain is delta coded if 1 */ + const opus_int nb_subfr /* I number of subframes */ +) +{ + opus_int k, ind_tmp, double_step_size_threshold; + + for( k = 0; k < nb_subfr; k++ ) { + if( k == 0 && conditional == 0 ) { + /* Gain index is not allowed to go down more than 16 steps (~21.8 dB) */ + *prev_ind = silk_max_int( ind[ k ], *prev_ind - 16 ); + } else { + /* Delta index */ + ind_tmp = ind[ k ] + MIN_DELTA_GAIN_QUANT; + + /* Accumulate deltas */ + double_step_size_threshold = 2 * MAX_DELTA_GAIN_QUANT - N_LEVELS_QGAIN + *prev_ind; + if( ind_tmp > double_step_size_threshold ) { + *prev_ind += silk_LSHIFT( ind_tmp, 1 ) - double_step_size_threshold; + } else { + *prev_ind += ind_tmp; + } + } + *prev_ind = silk_LIMIT_int( *prev_ind, 0, N_LEVELS_QGAIN - 1 ); + + /* Scale and convert to linear scale */ + gain_Q16[ k ] = silk_log2lin( silk_min_32( silk_SMULWB( INV_SCALE_Q16, *prev_ind ) + OFFSET, 3967 ) ); /* 3967 = 31 in Q7 */ + } +} + +/* Compute unique identifier of gain indices vector */ +opus_int32 silk_gains_ID( /* O returns unique identifier of gains */ + const opus_int8 ind[ MAX_NB_SUBFR ], /* I gain indices */ + const opus_int nb_subfr /* I number of subframes */ +) +{ + opus_int k; + opus_int32 gainsID; + + gainsID = 0; + for( k = 0; k < nb_subfr; k++ ) { + gainsID = silk_ADD_LSHIFT32( ind[ k ], gainsID, 8 ); + } + + return gainsID; +} diff --git a/media/libopus/silk/init_decoder.c b/media/libopus/silk/init_decoder.c new file mode 100644 index 0000000000..16c03dcd1c --- /dev/null +++ b/media/libopus/silk/init_decoder.c @@ -0,0 +1,57 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" + +/************************/ +/* Init Decoder State */ +/************************/ +opus_int silk_init_decoder( + silk_decoder_state *psDec /* I/O Decoder state pointer */ +) +{ + /* Clear the entire encoder state, except anything copied */ + silk_memset( psDec, 0, sizeof( silk_decoder_state ) ); + + /* Used to deactivate LSF interpolation */ + psDec->first_frame_after_reset = 1; + psDec->prev_gain_Q16 = 65536; + psDec->arch = opus_select_arch(); + + /* Reset CNG state */ + silk_CNG_Reset( psDec ); + + /* Reset PLC state */ + silk_PLC_Reset( psDec ); + + return(0); +} + diff --git a/media/libopus/silk/init_encoder.c b/media/libopus/silk/init_encoder.c new file mode 100644 index 0000000000..65995c33fa --- /dev/null +++ b/media/libopus/silk/init_encoder.c @@ -0,0 +1,64 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif +#ifdef FIXED_POINT +#include "main_FIX.h" +#else +#include "main_FLP.h" +#endif +#include "tuning_parameters.h" +#include "cpu_support.h" + +/*********************************/ +/* Initialize Silk Encoder state */ +/*********************************/ +opus_int silk_init_encoder( + silk_encoder_state_Fxx *psEnc, /* I/O Pointer to Silk FIX encoder state */ + int arch /* I Run-time architecture */ +) +{ + opus_int ret = 0; + + /* Clear the entire encoder state */ + silk_memset( psEnc, 0, sizeof( silk_encoder_state_Fxx ) ); + + psEnc->sCmn.arch = arch; + + psEnc->sCmn.variable_HP_smth1_Q15 = silk_LSHIFT( silk_lin2log( SILK_FIX_CONST( VARIABLE_HP_MIN_CUTOFF_HZ, 16 ) ) - ( 16 << 7 ), 8 ); + psEnc->sCmn.variable_HP_smth2_Q15 = psEnc->sCmn.variable_HP_smth1_Q15; + + /* Used to deactivate LSF interpolation, pitch prediction */ + psEnc->sCmn.first_frame_after_reset = 1; + + /* Initialize Silk VAD */ + ret += silk_VAD_Init( &psEnc->sCmn.sVAD ); + + return ret; +} diff --git a/media/libopus/silk/inner_prod_aligned.c b/media/libopus/silk/inner_prod_aligned.c new file mode 100644 index 0000000000..257ae9e04e --- /dev/null +++ b/media/libopus/silk/inner_prod_aligned.c @@ -0,0 +1,47 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "SigProc_FIX.h" + +opus_int32 silk_inner_prod_aligned_scale( + const opus_int16 *const inVec1, /* I input vector 1 */ + const opus_int16 *const inVec2, /* I input vector 2 */ + const opus_int scale, /* I number of bits to shift */ + const opus_int len /* I vector lengths */ +) +{ + opus_int i; + opus_int32 sum = 0; + for( i = 0; i < len; i++ ) { + sum = silk_ADD_RSHIFT32( sum, silk_SMULBB( inVec1[ i ], inVec2[ i ] ), scale ); + } + return sum; +} diff --git a/media/libopus/silk/interpolate.c b/media/libopus/silk/interpolate.c new file mode 100644 index 0000000000..833c28ef8e --- /dev/null +++ b/media/libopus/silk/interpolate.c @@ -0,0 +1,51 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" + +/* Interpolate two vectors */ +void silk_interpolate( + opus_int16 xi[ MAX_LPC_ORDER ], /* O interpolated vector */ + const opus_int16 x0[ MAX_LPC_ORDER ], /* I first vector */ + const opus_int16 x1[ MAX_LPC_ORDER ], /* I second vector */ + const opus_int ifact_Q2, /* I interp. factor, weight on 2nd vector */ + const opus_int d /* I number of parameters */ +) +{ + opus_int i; + + celt_assert( ifact_Q2 >= 0 ); + celt_assert( ifact_Q2 <= 4 ); + + for( i = 0; i < d; i++ ) { + xi[ i ] = (opus_int16)silk_ADD_RSHIFT( x0[ i ], silk_SMULBB( x1[ i ] - x0[ i ], ifact_Q2 ), 2 ); + } +} diff --git a/media/libopus/silk/lin2log.c b/media/libopus/silk/lin2log.c new file mode 100644 index 0000000000..0d5155aa86 --- /dev/null +++ b/media/libopus/silk/lin2log.c @@ -0,0 +1,46 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "SigProc_FIX.h" +/* Approximation of 128 * log2() (very close inverse of silk_log2lin()) */ +/* Convert input to a log scale */ +opus_int32 silk_lin2log( + const opus_int32 inLin /* I input in linear scale */ +) +{ + opus_int32 lz, frac_Q7; + + silk_CLZ_FRAC( inLin, &lz, &frac_Q7 ); + + /* Piece-wise parabolic approximation */ + return silk_ADD_LSHIFT32( silk_SMLAWB( frac_Q7, silk_MUL( frac_Q7, 128 - frac_Q7 ), 179 ), 31 - lz, 7 ); +} + diff --git a/media/libopus/silk/log2lin.c b/media/libopus/silk/log2lin.c new file mode 100644 index 0000000000..b7c48e4740 --- /dev/null +++ b/media/libopus/silk/log2lin.c @@ -0,0 +1,58 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "SigProc_FIX.h" + +/* Approximation of 2^() (very close inverse of silk_lin2log()) */ +/* Convert input to a linear scale */ +opus_int32 silk_log2lin( + const opus_int32 inLog_Q7 /* I input on log scale */ +) +{ + opus_int32 out, frac_Q7; + + if( inLog_Q7 < 0 ) { + return 0; + } else if ( inLog_Q7 >= 3967 ) { + return silk_int32_MAX; + } + + out = silk_LSHIFT( 1, silk_RSHIFT( inLog_Q7, 7 ) ); + frac_Q7 = inLog_Q7 & 0x7F; + if( inLog_Q7 < 2048 ) { + /* Piece-wise parabolic approximation */ + out = silk_ADD_RSHIFT32( out, silk_MUL( out, silk_SMLAWB( frac_Q7, silk_SMULBB( frac_Q7, 128 - frac_Q7 ), -174 ) ), 7 ); + } else { + /* Piece-wise parabolic approximation */ + out = silk_MLA( out, silk_RSHIFT( out, 7 ), silk_SMLAWB( frac_Q7, silk_SMULBB( frac_Q7, 128 - frac_Q7 ), -174 ) ); + } + return out; +} diff --git a/media/libopus/silk/macros.h b/media/libopus/silk/macros.h new file mode 100644 index 0000000000..3c67b6e5d9 --- /dev/null +++ b/media/libopus/silk/macros.h @@ -0,0 +1,151 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef SILK_MACROS_H +#define SILK_MACROS_H + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "opus_types.h" +#include "opus_defines.h" +#include "arch.h" + +/* This is an OPUS_INLINE header file for general platform. */ + +/* (a32 * (opus_int32)((opus_int16)(b32))) >> 16 output have to be 32bit int */ +#if OPUS_FAST_INT64 +#define silk_SMULWB(a32, b32) ((opus_int32)(((a32) * (opus_int64)((opus_int16)(b32))) >> 16)) +#else +#define silk_SMULWB(a32, b32) ((((a32) >> 16) * (opus_int32)((opus_int16)(b32))) + ((((a32) & 0x0000FFFF) * (opus_int32)((opus_int16)(b32))) >> 16)) +#endif + +/* a32 + (b32 * (opus_int32)((opus_int16)(c32))) >> 16 output have to be 32bit int */ +#if OPUS_FAST_INT64 +#define silk_SMLAWB(a32, b32, c32) ((opus_int32)((a32) + (((b32) * (opus_int64)((opus_int16)(c32))) >> 16))) +#else +#define silk_SMLAWB(a32, b32, c32) ((a32) + ((((b32) >> 16) * (opus_int32)((opus_int16)(c32))) + ((((b32) & 0x0000FFFF) * (opus_int32)((opus_int16)(c32))) >> 16))) +#endif + +/* (a32 * (b32 >> 16)) >> 16 */ +#if OPUS_FAST_INT64 +#define silk_SMULWT(a32, b32) ((opus_int32)(((a32) * (opus_int64)((b32) >> 16)) >> 16)) +#else +#define silk_SMULWT(a32, b32) (((a32) >> 16) * ((b32) >> 16) + ((((a32) & 0x0000FFFF) * ((b32) >> 16)) >> 16)) +#endif + +/* a32 + (b32 * (c32 >> 16)) >> 16 */ +#if OPUS_FAST_INT64 +#define silk_SMLAWT(a32, b32, c32) ((opus_int32)((a32) + (((b32) * ((opus_int64)(c32) >> 16)) >> 16))) +#else +#define silk_SMLAWT(a32, b32, c32) ((a32) + (((b32) >> 16) * ((c32) >> 16)) + ((((b32) & 0x0000FFFF) * ((c32) >> 16)) >> 16)) +#endif + +/* (opus_int32)((opus_int16)(a3))) * (opus_int32)((opus_int16)(b32)) output have to be 32bit int */ +#define silk_SMULBB(a32, b32) ((opus_int32)((opus_int16)(a32)) * (opus_int32)((opus_int16)(b32))) + +/* a32 + (opus_int32)((opus_int16)(b32)) * (opus_int32)((opus_int16)(c32)) output have to be 32bit int */ +#define silk_SMLABB(a32, b32, c32) ((a32) + ((opus_int32)((opus_int16)(b32))) * (opus_int32)((opus_int16)(c32))) + +/* (opus_int32)((opus_int16)(a32)) * (b32 >> 16) */ +#define silk_SMULBT(a32, b32) ((opus_int32)((opus_int16)(a32)) * ((b32) >> 16)) + +/* a32 + (opus_int32)((opus_int16)(b32)) * (c32 >> 16) */ +#define silk_SMLABT(a32, b32, c32) ((a32) + ((opus_int32)((opus_int16)(b32))) * ((c32) >> 16)) + +/* a64 + (b32 * c32) */ +#define silk_SMLAL(a64, b32, c32) (silk_ADD64((a64), ((opus_int64)(b32) * (opus_int64)(c32)))) + +/* (a32 * b32) >> 16 */ +#if OPUS_FAST_INT64 +#define silk_SMULWW(a32, b32) ((opus_int32)(((opus_int64)(a32) * (b32)) >> 16)) +#else +#define silk_SMULWW(a32, b32) silk_MLA(silk_SMULWB((a32), (b32)), (a32), silk_RSHIFT_ROUND((b32), 16)) +#endif + +/* a32 + ((b32 * c32) >> 16) */ +#if OPUS_FAST_INT64 +#define silk_SMLAWW(a32, b32, c32) ((opus_int32)((a32) + (((opus_int64)(b32) * (c32)) >> 16))) +#else +#define silk_SMLAWW(a32, b32, c32) silk_MLA(silk_SMLAWB((a32), (b32), (c32)), (b32), silk_RSHIFT_ROUND((c32), 16)) +#endif + +/* add/subtract with output saturated */ +#define silk_ADD_SAT32(a, b) ((((opus_uint32)(a) + (opus_uint32)(b)) & 0x80000000) == 0 ? \ + ((((a) & (b)) & 0x80000000) != 0 ? silk_int32_MIN : (a)+(b)) : \ + ((((a) | (b)) & 0x80000000) == 0 ? silk_int32_MAX : (a)+(b)) ) + +#define silk_SUB_SAT32(a, b) ((((opus_uint32)(a)-(opus_uint32)(b)) & 0x80000000) == 0 ? \ + (( (a) & ((b)^0x80000000) & 0x80000000) ? silk_int32_MIN : (a)-(b)) : \ + ((((a)^0x80000000) & (b) & 0x80000000) ? silk_int32_MAX : (a)-(b)) ) + +#if defined(MIPSr1_ASM) +#include "mips/macros_mipsr1.h" +#endif + +#include "ecintrin.h" +#ifndef OVERRIDE_silk_CLZ16 +static OPUS_INLINE opus_int32 silk_CLZ16(opus_int16 in16) +{ + return 32 - EC_ILOG(in16<<16|0x8000); +} +#endif + +#ifndef OVERRIDE_silk_CLZ32 +static OPUS_INLINE opus_int32 silk_CLZ32(opus_int32 in32) +{ + return in32 ? 32 - EC_ILOG(in32) : 32; +} +#endif + +/* Row based */ +#define matrix_ptr(Matrix_base_adr, row, column, N) \ + (*((Matrix_base_adr) + ((row)*(N)+(column)))) +#define matrix_adr(Matrix_base_adr, row, column, N) \ + ((Matrix_base_adr) + ((row)*(N)+(column))) + +/* Column based */ +#ifndef matrix_c_ptr +# define matrix_c_ptr(Matrix_base_adr, row, column, M) \ + (*((Matrix_base_adr) + ((row)+(M)*(column)))) +#endif + +#ifdef OPUS_ARM_INLINE_ASM +#include "arm/macros_armv4.h" +#endif + +#ifdef OPUS_ARM_INLINE_EDSP +#include "arm/macros_armv5e.h" +#endif + +#ifdef OPUS_ARM_PRESUME_AARCH64_NEON_INTR +#include "arm/macros_arm64.h" +#endif + +#endif /* SILK_MACROS_H */ + diff --git a/media/libopus/silk/main.h b/media/libopus/silk/main.h new file mode 100644 index 0000000000..a5f568758f --- /dev/null +++ b/media/libopus/silk/main.h @@ -0,0 +1,476 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef SILK_MAIN_H +#define SILK_MAIN_H + +#include "SigProc_FIX.h" +#include "define.h" +#include "structs.h" +#include "tables.h" +#include "PLC.h" +#include "control.h" +#include "debug.h" +#include "entenc.h" +#include "entdec.h" + +#if defined(OPUS_X86_MAY_HAVE_SSE4_1) +#include "x86/main_sse.h" +#endif + +#if (defined(OPUS_ARM_ASM) || defined(OPUS_ARM_MAY_HAVE_NEON_INTR)) +#include "arm/NSQ_del_dec_arm.h" +#endif + +/* Convert Left/Right stereo signal to adaptive Mid/Side representation */ +void silk_stereo_LR_to_MS( + stereo_enc_state *state, /* I/O State */ + opus_int16 x1[], /* I/O Left input signal, becomes mid signal */ + opus_int16 x2[], /* I/O Right input signal, becomes side signal */ + opus_int8 ix[ 2 ][ 3 ], /* O Quantization indices */ + opus_int8 *mid_only_flag, /* O Flag: only mid signal coded */ + opus_int32 mid_side_rates_bps[], /* O Bitrates for mid and side signals */ + opus_int32 total_rate_bps, /* I Total bitrate */ + opus_int prev_speech_act_Q8, /* I Speech activity level in previous frame */ + opus_int toMono, /* I Last frame before a stereo->mono transition */ + opus_int fs_kHz, /* I Sample rate (kHz) */ + opus_int frame_length /* I Number of samples */ +); + +/* Convert adaptive Mid/Side representation to Left/Right stereo signal */ +void silk_stereo_MS_to_LR( + stereo_dec_state *state, /* I/O State */ + opus_int16 x1[], /* I/O Left input signal, becomes mid signal */ + opus_int16 x2[], /* I/O Right input signal, becomes side signal */ + const opus_int32 pred_Q13[], /* I Predictors */ + opus_int fs_kHz, /* I Samples rate (kHz) */ + opus_int frame_length /* I Number of samples */ +); + +/* Find least-squares prediction gain for one signal based on another and quantize it */ +opus_int32 silk_stereo_find_predictor( /* O Returns predictor in Q13 */ + opus_int32 *ratio_Q14, /* O Ratio of residual and mid energies */ + const opus_int16 x[], /* I Basis signal */ + const opus_int16 y[], /* I Target signal */ + opus_int32 mid_res_amp_Q0[], /* I/O Smoothed mid, residual norms */ + opus_int length, /* I Number of samples */ + opus_int smooth_coef_Q16 /* I Smoothing coefficient */ +); + +/* Quantize mid/side predictors */ +void silk_stereo_quant_pred( + opus_int32 pred_Q13[], /* I/O Predictors (out: quantized) */ + opus_int8 ix[ 2 ][ 3 ] /* O Quantization indices */ +); + +/* Entropy code the mid/side quantization indices */ +void silk_stereo_encode_pred( + ec_enc *psRangeEnc, /* I/O Compressor data structure */ + opus_int8 ix[ 2 ][ 3 ] /* I Quantization indices */ +); + +/* Entropy code the mid-only flag */ +void silk_stereo_encode_mid_only( + ec_enc *psRangeEnc, /* I/O Compressor data structure */ + opus_int8 mid_only_flag +); + +/* Decode mid/side predictors */ +void silk_stereo_decode_pred( + ec_dec *psRangeDec, /* I/O Compressor data structure */ + opus_int32 pred_Q13[] /* O Predictors */ +); + +/* Decode mid-only flag */ +void silk_stereo_decode_mid_only( + ec_dec *psRangeDec, /* I/O Compressor data structure */ + opus_int *decode_only_mid /* O Flag that only mid channel has been coded */ +); + +/* Encodes signs of excitation */ +void silk_encode_signs( + ec_enc *psRangeEnc, /* I/O Compressor data structure */ + const opus_int8 pulses[], /* I pulse signal */ + opus_int length, /* I length of input */ + const opus_int signalType, /* I Signal type */ + const opus_int quantOffsetType, /* I Quantization offset type */ + const opus_int sum_pulses[ MAX_NB_SHELL_BLOCKS ] /* I Sum of absolute pulses per block */ +); + +/* Decodes signs of excitation */ +void silk_decode_signs( + ec_dec *psRangeDec, /* I/O Compressor data structure */ + opus_int16 pulses[], /* I/O pulse signal */ + opus_int length, /* I length of input */ + const opus_int signalType, /* I Signal type */ + const opus_int quantOffsetType, /* I Quantization offset type */ + const opus_int sum_pulses[ MAX_NB_SHELL_BLOCKS ] /* I Sum of absolute pulses per block */ +); + +/* Check encoder control struct */ +opus_int check_control_input( + silk_EncControlStruct *encControl /* I Control structure */ +); + +/* Control internal sampling rate */ +opus_int silk_control_audio_bandwidth( + silk_encoder_state *psEncC, /* I/O Pointer to Silk encoder state */ + silk_EncControlStruct *encControl /* I Control structure */ +); + +/* Control SNR of redidual quantizer */ +opus_int silk_control_SNR( + silk_encoder_state *psEncC, /* I/O Pointer to Silk encoder state */ + opus_int32 TargetRate_bps /* I Target max bitrate (bps) */ +); + +/***************/ +/* Shell coder */ +/***************/ + +/* Encode quantization indices of excitation */ +void silk_encode_pulses( + ec_enc *psRangeEnc, /* I/O compressor data structure */ + const opus_int signalType, /* I Signal type */ + const opus_int quantOffsetType, /* I quantOffsetType */ + opus_int8 pulses[], /* I quantization indices */ + const opus_int frame_length /* I Frame length */ +); + +/* Shell encoder, operates on one shell code frame of 16 pulses */ +void silk_shell_encoder( + ec_enc *psRangeEnc, /* I/O compressor data structure */ + const opus_int *pulses0 /* I data: nonnegative pulse amplitudes */ +); + +/* Shell decoder, operates on one shell code frame of 16 pulses */ +void silk_shell_decoder( + opus_int16 *pulses0, /* O data: nonnegative pulse amplitudes */ + ec_dec *psRangeDec, /* I/O Compressor data structure */ + const opus_int pulses4 /* I number of pulses per pulse-subframe */ +); + +/* Gain scalar quantization with hysteresis, uniform on log scale */ +void silk_gains_quant( + opus_int8 ind[ MAX_NB_SUBFR ], /* O gain indices */ + opus_int32 gain_Q16[ MAX_NB_SUBFR ], /* I/O gains (quantized out) */ + opus_int8 *prev_ind, /* I/O last index in previous frame */ + const opus_int conditional, /* I first gain is delta coded if 1 */ + const opus_int nb_subfr /* I number of subframes */ +); + +/* Gains scalar dequantization, uniform on log scale */ +void silk_gains_dequant( + opus_int32 gain_Q16[ MAX_NB_SUBFR ], /* O quantized gains */ + const opus_int8 ind[ MAX_NB_SUBFR ], /* I gain indices */ + opus_int8 *prev_ind, /* I/O last index in previous frame */ + const opus_int conditional, /* I first gain is delta coded if 1 */ + const opus_int nb_subfr /* I number of subframes */ +); + +/* Compute unique identifier of gain indices vector */ +opus_int32 silk_gains_ID( /* O returns unique identifier of gains */ + const opus_int8 ind[ MAX_NB_SUBFR ], /* I gain indices */ + const opus_int nb_subfr /* I number of subframes */ +); + +/* Interpolate two vectors */ +void silk_interpolate( + opus_int16 xi[ MAX_LPC_ORDER ], /* O interpolated vector */ + const opus_int16 x0[ MAX_LPC_ORDER ], /* I first vector */ + const opus_int16 x1[ MAX_LPC_ORDER ], /* I second vector */ + const opus_int ifact_Q2, /* I interp. factor, weight on 2nd vector */ + const opus_int d /* I number of parameters */ +); + +/* LTP tap quantizer */ +void silk_quant_LTP_gains( + opus_int16 B_Q14[ MAX_NB_SUBFR * LTP_ORDER ], /* O Quantized LTP gains */ + opus_int8 cbk_index[ MAX_NB_SUBFR ], /* O Codebook Index */ + opus_int8 *periodicity_index, /* O Periodicity Index */ + opus_int32 *sum_gain_dB_Q7, /* I/O Cumulative max prediction gain */ + opus_int *pred_gain_dB_Q7, /* O LTP prediction gain */ + const opus_int32 XX_Q17[ MAX_NB_SUBFR*LTP_ORDER*LTP_ORDER ], /* I Correlation matrix in Q18 */ + const opus_int32 xX_Q17[ MAX_NB_SUBFR*LTP_ORDER ], /* I Correlation vector in Q18 */ + const opus_int subfr_len, /* I Number of samples per subframe */ + const opus_int nb_subfr, /* I Number of subframes */ + int arch /* I Run-time architecture */ +); + +/* Entropy constrained matrix-weighted VQ, for a single input data vector */ +void silk_VQ_WMat_EC_c( + opus_int8 *ind, /* O index of best codebook vector */ + opus_int32 *res_nrg_Q15, /* O best residual energy */ + opus_int32 *rate_dist_Q8, /* O best total bitrate */ + opus_int *gain_Q7, /* O sum of absolute LTP coefficients */ + const opus_int32 *XX_Q17, /* I correlation matrix */ + const opus_int32 *xX_Q17, /* I correlation vector */ + const opus_int8 *cb_Q7, /* I codebook */ + const opus_uint8 *cb_gain_Q7, /* I codebook effective gain */ + const opus_uint8 *cl_Q5, /* I code length for each codebook vector */ + const opus_int subfr_len, /* I number of samples per subframe */ + const opus_int32 max_gain_Q7, /* I maximum sum of absolute LTP coefficients */ + const opus_int L /* I number of vectors in codebook */ +); + +#if !defined(OVERRIDE_silk_VQ_WMat_EC) +#define silk_VQ_WMat_EC(ind, res_nrg_Q15, rate_dist_Q8, gain_Q7, XX_Q17, xX_Q17, cb_Q7, cb_gain_Q7, cl_Q5, subfr_len, max_gain_Q7, L, arch) \ + ((void)(arch),silk_VQ_WMat_EC_c(ind, res_nrg_Q15, rate_dist_Q8, gain_Q7, XX_Q17, xX_Q17, cb_Q7, cb_gain_Q7, cl_Q5, subfr_len, max_gain_Q7, L)) +#endif + +/************************************/ +/* Noise shaping quantization (NSQ) */ +/************************************/ + +void silk_NSQ_c( + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + SideInfoIndices *psIndices, /* I/O Quantization Indices */ + const opus_int16 x16[], /* I Input */ + opus_int8 pulses[], /* O Quantized pulse signal */ + const opus_int16 PredCoef_Q12[ 2 * MAX_LPC_ORDER ], /* I Short term prediction coefs */ + const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ + const opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ + const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ + const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ + const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ + const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ + const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ + const opus_int LTP_scale_Q14 /* I LTP state scaling */ +); + +#if !defined(OVERRIDE_silk_NSQ) +#define silk_NSQ(psEncC, NSQ, psIndices, x16, pulses, PredCoef_Q12, LTPCoef_Q14, AR_Q13, \ + HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14, arch) \ + ((void)(arch),silk_NSQ_c(psEncC, NSQ, psIndices, x16, pulses, PredCoef_Q12, LTPCoef_Q14, AR_Q13, \ + HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14)) +#endif + +/* Noise shaping using delayed decision */ +void silk_NSQ_del_dec_c( + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + SideInfoIndices *psIndices, /* I/O Quantization Indices */ + const opus_int16 x16[], /* I Input */ + opus_int8 pulses[], /* O Quantized pulse signal */ + const opus_int16 PredCoef_Q12[ 2 * MAX_LPC_ORDER ], /* I Short term prediction coefs */ + const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ + const opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ + const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ + const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ + const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ + const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ + const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ + const opus_int LTP_scale_Q14 /* I LTP state scaling */ +); + +#if !defined(OVERRIDE_silk_NSQ_del_dec) +#define silk_NSQ_del_dec(psEncC, NSQ, psIndices, x16, pulses, PredCoef_Q12, LTPCoef_Q14, AR_Q13, \ + HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14, arch) \ + ((void)(arch),silk_NSQ_del_dec_c(psEncC, NSQ, psIndices, x16, pulses, PredCoef_Q12, LTPCoef_Q14, AR_Q13, \ + HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14)) +#endif + +/************/ +/* Silk VAD */ +/************/ +/* Initialize the Silk VAD */ +opus_int silk_VAD_Init( /* O Return value, 0 if success */ + silk_VAD_state *psSilk_VAD /* I/O Pointer to Silk VAD state */ +); + +/* Get speech activity level in Q8 */ +opus_int silk_VAD_GetSA_Q8_c( /* O Return value, 0 if success */ + silk_encoder_state *psEncC, /* I/O Encoder state */ + const opus_int16 pIn[] /* I PCM input */ +); + +#if !defined(OVERRIDE_silk_VAD_GetSA_Q8) +#define silk_VAD_GetSA_Q8(psEnC, pIn, arch) ((void)(arch),silk_VAD_GetSA_Q8_c(psEnC, pIn)) +#endif + +/* Low-pass filter with variable cutoff frequency based on */ +/* piece-wise linear interpolation between elliptic filters */ +/* Start by setting transition_frame_no = 1; */ +void silk_LP_variable_cutoff( + silk_LP_state *psLP, /* I/O LP filter state */ + opus_int16 *frame, /* I/O Low-pass filtered output signal */ + const opus_int frame_length /* I Frame length */ +); + +/******************/ +/* NLSF Quantizer */ +/******************/ +/* Limit, stabilize, convert and quantize NLSFs */ +void silk_process_NLSFs( + silk_encoder_state *psEncC, /* I/O Encoder state */ + opus_int16 PredCoef_Q12[ 2 ][ MAX_LPC_ORDER ], /* O Prediction coefficients */ + opus_int16 pNLSF_Q15[ MAX_LPC_ORDER ], /* I/O Normalized LSFs (quant out) (0 - (2^15-1)) */ + const opus_int16 prev_NLSFq_Q15[ MAX_LPC_ORDER ] /* I Previous Normalized LSFs (0 - (2^15-1)) */ +); + +opus_int32 silk_NLSF_encode( /* O Returns RD value in Q25 */ + opus_int8 *NLSFIndices, /* I Codebook path vector [ LPC_ORDER + 1 ] */ + opus_int16 *pNLSF_Q15, /* I/O Quantized NLSF vector [ LPC_ORDER ] */ + const silk_NLSF_CB_struct *psNLSF_CB, /* I Codebook object */ + const opus_int16 *pW_QW, /* I NLSF weight vector [ LPC_ORDER ] */ + const opus_int NLSF_mu_Q20, /* I Rate weight for the RD optimization */ + const opus_int nSurvivors, /* I Max survivors after first stage */ + const opus_int signalType /* I Signal type: 0/1/2 */ +); + +/* Compute quantization errors for an LPC_order element input vector for a VQ codebook */ +void silk_NLSF_VQ( + opus_int32 err_Q26[], /* O Quantization errors [K] */ + const opus_int16 in_Q15[], /* I Input vectors to be quantized [LPC_order] */ + const opus_uint8 pCB_Q8[], /* I Codebook vectors [K*LPC_order] */ + const opus_int16 pWght_Q9[], /* I Codebook weights [K*LPC_order] */ + const opus_int K, /* I Number of codebook vectors */ + const opus_int LPC_order /* I Number of LPCs */ +); + +/* Delayed-decision quantizer for NLSF residuals */ +opus_int32 silk_NLSF_del_dec_quant( /* O Returns RD value in Q25 */ + opus_int8 indices[], /* O Quantization indices [ order ] */ + const opus_int16 x_Q10[], /* I Input [ order ] */ + const opus_int16 w_Q5[], /* I Weights [ order ] */ + const opus_uint8 pred_coef_Q8[], /* I Backward predictor coefs [ order ] */ + const opus_int16 ec_ix[], /* I Indices to entropy coding tables [ order ] */ + const opus_uint8 ec_rates_Q5[], /* I Rates [] */ + const opus_int quant_step_size_Q16, /* I Quantization step size */ + const opus_int16 inv_quant_step_size_Q6, /* I Inverse quantization step size */ + const opus_int32 mu_Q20, /* I R/D tradeoff */ + const opus_int16 order /* I Number of input values */ +); + +/* Unpack predictor values and indices for entropy coding tables */ +void silk_NLSF_unpack( + opus_int16 ec_ix[], /* O Indices to entropy tables [ LPC_ORDER ] */ + opus_uint8 pred_Q8[], /* O LSF predictor [ LPC_ORDER ] */ + const silk_NLSF_CB_struct *psNLSF_CB, /* I Codebook object */ + const opus_int CB1_index /* I Index of vector in first LSF codebook */ +); + +/***********************/ +/* NLSF vector decoder */ +/***********************/ +void silk_NLSF_decode( + opus_int16 *pNLSF_Q15, /* O Quantized NLSF vector [ LPC_ORDER ] */ + opus_int8 *NLSFIndices, /* I Codebook path vector [ LPC_ORDER + 1 ] */ + const silk_NLSF_CB_struct *psNLSF_CB /* I Codebook object */ +); + +/****************************************************/ +/* Decoder Functions */ +/****************************************************/ +opus_int silk_init_decoder( + silk_decoder_state *psDec /* I/O Decoder state pointer */ +); + +/* Set decoder sampling rate */ +opus_int silk_decoder_set_fs( + silk_decoder_state *psDec, /* I/O Decoder state pointer */ + opus_int fs_kHz, /* I Sampling frequency (kHz) */ + opus_int32 fs_API_Hz /* I API Sampling frequency (Hz) */ +); + +/****************/ +/* Decode frame */ +/****************/ +opus_int silk_decode_frame( + silk_decoder_state *psDec, /* I/O Pointer to Silk decoder state */ + ec_dec *psRangeDec, /* I/O Compressor data structure */ + opus_int16 pOut[], /* O Pointer to output speech frame */ + opus_int32 *pN, /* O Pointer to size of output frame */ + opus_int lostFlag, /* I 0: no loss, 1 loss, 2 decode fec */ + opus_int condCoding, /* I The type of conditional coding to use */ + int arch /* I Run-time architecture */ +); + +/* Decode indices from bitstream */ +void silk_decode_indices( + silk_decoder_state *psDec, /* I/O State */ + ec_dec *psRangeDec, /* I/O Compressor data structure */ + opus_int FrameIndex, /* I Frame number */ + opus_int decode_LBRR, /* I Flag indicating LBRR data is being decoded */ + opus_int condCoding /* I The type of conditional coding to use */ +); + +/* Decode parameters from payload */ +void silk_decode_parameters( + silk_decoder_state *psDec, /* I/O State */ + silk_decoder_control *psDecCtrl, /* I/O Decoder control */ + opus_int condCoding /* I The type of conditional coding to use */ +); + +/* Core decoder. Performs inverse NSQ operation LTP + LPC */ +void silk_decode_core( + silk_decoder_state *psDec, /* I/O Decoder state */ + silk_decoder_control *psDecCtrl, /* I Decoder control */ + opus_int16 xq[], /* O Decoded speech */ + const opus_int16 pulses[ MAX_FRAME_LENGTH ], /* I Pulse signal */ + int arch /* I Run-time architecture */ +); + +/* Decode quantization indices of excitation (Shell coding) */ +void silk_decode_pulses( + ec_dec *psRangeDec, /* I/O Compressor data structure */ + opus_int16 pulses[], /* O Excitation signal */ + const opus_int signalType, /* I Sigtype */ + const opus_int quantOffsetType, /* I quantOffsetType */ + const opus_int frame_length /* I Frame length */ +); + +/******************/ +/* CNG */ +/******************/ + +/* Reset CNG */ +void silk_CNG_Reset( + silk_decoder_state *psDec /* I/O Decoder state */ +); + +/* Updates CNG estimate, and applies the CNG when packet was lost */ +void silk_CNG( + silk_decoder_state *psDec, /* I/O Decoder state */ + silk_decoder_control *psDecCtrl, /* I/O Decoder control */ + opus_int16 frame[], /* I/O Signal */ + opus_int length /* I Length of residual */ +); + +/* Encoding of various parameters */ +void silk_encode_indices( + silk_encoder_state *psEncC, /* I/O Encoder state */ + ec_enc *psRangeEnc, /* I/O Compressor data structure */ + opus_int FrameIndex, /* I Frame number */ + opus_int encode_LBRR, /* I Flag indicating LBRR data is being encoded */ + opus_int condCoding /* I The type of conditional coding to use */ +); + +#endif diff --git a/media/libopus/silk/mips/NSQ_del_dec_mipsr1.h b/media/libopus/silk/mips/NSQ_del_dec_mipsr1.h new file mode 100644 index 0000000000..cd70713a8f --- /dev/null +++ b/media/libopus/silk/mips/NSQ_del_dec_mipsr1.h @@ -0,0 +1,410 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef __NSQ_DEL_DEC_MIPSR1_H__ +#define __NSQ_DEL_DEC_MIPSR1_H__ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" +#include "stack_alloc.h" + +#define OVERRIDE_silk_noise_shape_quantizer_del_dec +static inline void silk_noise_shape_quantizer_del_dec( + silk_nsq_state *NSQ, /* I/O NSQ state */ + NSQ_del_dec_struct psDelDec[], /* I/O Delayed decision states */ + opus_int signalType, /* I Signal type */ + const opus_int32 x_Q10[], /* I */ + opus_int8 pulses[], /* O */ + opus_int16 xq[], /* O */ + opus_int32 sLTP_Q15[], /* I/O LTP filter state */ + opus_int32 delayedGain_Q10[], /* I/O Gain delay buffer */ + const opus_int16 a_Q12[], /* I Short term prediction coefs */ + const opus_int16 b_Q14[], /* I Long term prediction coefs */ + const opus_int16 AR_shp_Q13[], /* I Noise shaping coefs */ + opus_int lag, /* I Pitch lag */ + opus_int32 HarmShapeFIRPacked_Q14, /* I */ + opus_int Tilt_Q14, /* I Spectral tilt */ + opus_int32 LF_shp_Q14, /* I */ + opus_int32 Gain_Q16, /* I */ + opus_int Lambda_Q10, /* I */ + opus_int offset_Q10, /* I */ + opus_int length, /* I Input length */ + opus_int subfr, /* I Subframe number */ + opus_int shapingLPCOrder, /* I Shaping LPC filter order */ + opus_int predictLPCOrder, /* I Prediction filter order */ + opus_int warping_Q16, /* I */ + opus_int nStatesDelayedDecision, /* I Number of states in decision tree */ + opus_int *smpl_buf_idx, /* I/O Index to newest samples in buffers */ + opus_int decisionDelay, /* I */ + int arch /* I */ +) +{ + opus_int i, j, k, Winner_ind, RDmin_ind, RDmax_ind, last_smple_idx; + opus_int32 Winner_rand_state; + opus_int32 LTP_pred_Q14, LPC_pred_Q14, n_AR_Q14, n_LTP_Q14; + opus_int32 n_LF_Q14, r_Q10, rr_Q10, rd1_Q10, rd2_Q10, RDmin_Q10, RDmax_Q10; + opus_int32 q1_Q0, q1_Q10, q2_Q10, exc_Q14, LPC_exc_Q14, xq_Q14, Gain_Q10; + opus_int32 tmp1, tmp2, sLF_AR_shp_Q14; + opus_int32 *pred_lag_ptr, *shp_lag_ptr, *psLPC_Q14; + NSQ_sample_struct psSampleState[ MAX_DEL_DEC_STATES ][ 2 ]; + NSQ_del_dec_struct *psDD; + NSQ_sample_struct *psSS; + opus_int16 b_Q14_0, b_Q14_1, b_Q14_2, b_Q14_3, b_Q14_4; + opus_int16 a_Q12_0, a_Q12_1, a_Q12_2, a_Q12_3, a_Q12_4, a_Q12_5, a_Q12_6; + opus_int16 a_Q12_7, a_Q12_8, a_Q12_9, a_Q12_10, a_Q12_11, a_Q12_12, a_Q12_13; + opus_int16 a_Q12_14, a_Q12_15; + + opus_int32 cur, prev, next; + + /*Unused.*/ + (void)arch; + + //Intialize b_Q14 variables + b_Q14_0 = b_Q14[ 0 ]; + b_Q14_1 = b_Q14[ 1 ]; + b_Q14_2 = b_Q14[ 2 ]; + b_Q14_3 = b_Q14[ 3 ]; + b_Q14_4 = b_Q14[ 4 ]; + + //Intialize a_Q12 variables + a_Q12_0 = a_Q12[0]; + a_Q12_1 = a_Q12[1]; + a_Q12_2 = a_Q12[2]; + a_Q12_3 = a_Q12[3]; + a_Q12_4 = a_Q12[4]; + a_Q12_5 = a_Q12[5]; + a_Q12_6 = a_Q12[6]; + a_Q12_7 = a_Q12[7]; + a_Q12_8 = a_Q12[8]; + a_Q12_9 = a_Q12[9]; + a_Q12_10 = a_Q12[10]; + a_Q12_11 = a_Q12[11]; + a_Q12_12 = a_Q12[12]; + a_Q12_13 = a_Q12[13]; + a_Q12_14 = a_Q12[14]; + a_Q12_15 = a_Q12[15]; + + long long temp64; + + silk_assert( nStatesDelayedDecision > 0 ); + + shp_lag_ptr = &NSQ->sLTP_shp_Q14[ NSQ->sLTP_shp_buf_idx - lag + HARM_SHAPE_FIR_TAPS / 2 ]; + pred_lag_ptr = &sLTP_Q15[ NSQ->sLTP_buf_idx - lag + LTP_ORDER / 2 ]; + Gain_Q10 = silk_RSHIFT( Gain_Q16, 6 ); + + for( i = 0; i < length; i++ ) { + /* Perform common calculations used in all states */ + + /* Long-term prediction */ + if( signalType == TYPE_VOICED ) { + /* Unrolled loop */ + /* Avoids introducing a bias because silk_SMLAWB() always rounds to -inf */ + temp64 = __builtin_mips_mult(pred_lag_ptr[ 0 ], b_Q14_0 ); + temp64 = __builtin_mips_madd( temp64, pred_lag_ptr[ -1 ], b_Q14_1 ); + temp64 = __builtin_mips_madd( temp64, pred_lag_ptr[ -2 ], b_Q14_2 ); + temp64 = __builtin_mips_madd( temp64, pred_lag_ptr[ -3 ], b_Q14_3 ); + temp64 = __builtin_mips_madd( temp64, pred_lag_ptr[ -4 ], b_Q14_4 ); + temp64 += 32768; + LTP_pred_Q14 = __builtin_mips_extr_w(temp64, 16); + LTP_pred_Q14 = silk_LSHIFT( LTP_pred_Q14, 1 ); /* Q13 -> Q14 */ + pred_lag_ptr++; + } else { + LTP_pred_Q14 = 0; + } + + /* Long-term shaping */ + if( lag > 0 ) { + /* Symmetric, packed FIR coefficients */ + n_LTP_Q14 = silk_SMULWB( silk_ADD32( shp_lag_ptr[ 0 ], shp_lag_ptr[ -2 ] ), HarmShapeFIRPacked_Q14 ); + n_LTP_Q14 = silk_SMLAWT( n_LTP_Q14, shp_lag_ptr[ -1 ], HarmShapeFIRPacked_Q14 ); + n_LTP_Q14 = silk_SUB_LSHIFT32( LTP_pred_Q14, n_LTP_Q14, 2 ); /* Q12 -> Q14 */ + shp_lag_ptr++; + } else { + n_LTP_Q14 = 0; + } + + for( k = 0; k < nStatesDelayedDecision; k++ ) { + /* Delayed decision state */ + psDD = &psDelDec[ k ]; + + /* Sample state */ + psSS = psSampleState[ k ]; + + /* Generate dither */ + psDD->Seed = silk_RAND( psDD->Seed ); + + /* Pointer used in short term prediction and shaping */ + psLPC_Q14 = &psDD->sLPC_Q14[ NSQ_LPC_BUF_LENGTH - 1 + i ]; + /* Short-term prediction */ + silk_assert( predictLPCOrder == 10 || predictLPCOrder == 16 ); + temp64 = __builtin_mips_mult(psLPC_Q14[ 0 ], a_Q12_0 ); + temp64 = __builtin_mips_madd( temp64, psLPC_Q14[ -1 ], a_Q12_1 ); + temp64 = __builtin_mips_madd( temp64, psLPC_Q14[ -2 ], a_Q12_2 ); + temp64 = __builtin_mips_madd( temp64, psLPC_Q14[ -3 ], a_Q12_3 ); + temp64 = __builtin_mips_madd( temp64, psLPC_Q14[ -4 ], a_Q12_4 ); + temp64 = __builtin_mips_madd( temp64, psLPC_Q14[ -5 ], a_Q12_5 ); + temp64 = __builtin_mips_madd( temp64, psLPC_Q14[ -6 ], a_Q12_6 ); + temp64 = __builtin_mips_madd( temp64, psLPC_Q14[ -7 ], a_Q12_7 ); + temp64 = __builtin_mips_madd( temp64, psLPC_Q14[ -8 ], a_Q12_8 ); + temp64 = __builtin_mips_madd( temp64, psLPC_Q14[ -9 ], a_Q12_9 ); + if( predictLPCOrder == 16 ) { + temp64 = __builtin_mips_madd( temp64, psLPC_Q14[ -10 ], a_Q12_10 ); + temp64 = __builtin_mips_madd( temp64, psLPC_Q14[ -11 ], a_Q12_11 ); + temp64 = __builtin_mips_madd( temp64, psLPC_Q14[ -12 ], a_Q12_12 ); + temp64 = __builtin_mips_madd( temp64, psLPC_Q14[ -13 ], a_Q12_13 ); + temp64 = __builtin_mips_madd( temp64, psLPC_Q14[ -14 ], a_Q12_14 ); + temp64 = __builtin_mips_madd( temp64, psLPC_Q14[ -15 ], a_Q12_15 ); + } + temp64 += 32768; + LPC_pred_Q14 = __builtin_mips_extr_w(temp64, 16); + + LPC_pred_Q14 = silk_LSHIFT( LPC_pred_Q14, 4 ); /* Q10 -> Q14 */ + + /* Noise shape feedback */ + silk_assert( ( shapingLPCOrder & 1 ) == 0 ); /* check that order is even */ + /* Output of lowpass section */ + tmp2 = silk_SMLAWB( psLPC_Q14[ 0 ], psDD->sAR2_Q14[ 0 ], warping_Q16 ); + /* Output of allpass section */ + tmp1 = silk_SMLAWB( psDD->sAR2_Q14[ 0 ], psDD->sAR2_Q14[ 1 ] - tmp2, warping_Q16 ); + psDD->sAR2_Q14[ 0 ] = tmp2; + + temp64 = __builtin_mips_mult(tmp2, AR_shp_Q13[ 0 ] ); + + prev = psDD->sAR2_Q14[ 1 ]; + + /* Loop over allpass sections */ + for( j = 2; j < shapingLPCOrder; j += 2 ) { + cur = psDD->sAR2_Q14[ j ]; + next = psDD->sAR2_Q14[ j+1 ]; + /* Output of allpass section */ + tmp2 = silk_SMLAWB( prev, cur - tmp1, warping_Q16 ); + psDD->sAR2_Q14[ j - 1 ] = tmp1; + temp64 = __builtin_mips_madd( temp64, tmp1, AR_shp_Q13[ j - 1 ] ); + temp64 = __builtin_mips_madd( temp64, tmp2, AR_shp_Q13[ j ] ); + /* Output of allpass section */ + tmp1 = silk_SMLAWB( cur, next - tmp2, warping_Q16 ); + psDD->sAR2_Q14[ j + 0 ] = tmp2; + prev = next; + } + psDD->sAR2_Q14[ shapingLPCOrder - 1 ] = tmp1; + temp64 = __builtin_mips_madd( temp64, tmp1, AR_shp_Q13[ shapingLPCOrder - 1 ] ); + temp64 += 32768; + n_AR_Q14 = __builtin_mips_extr_w(temp64, 16); + n_AR_Q14 = silk_LSHIFT( n_AR_Q14, 1 ); /* Q11 -> Q12 */ + n_AR_Q14 = silk_SMLAWB( n_AR_Q14, psDD->LF_AR_Q14, Tilt_Q14 ); /* Q12 */ + n_AR_Q14 = silk_LSHIFT( n_AR_Q14, 2 ); /* Q12 -> Q14 */ + + n_LF_Q14 = silk_SMULWB( psDD->Shape_Q14[ *smpl_buf_idx ], LF_shp_Q14 ); /* Q12 */ + n_LF_Q14 = silk_SMLAWT( n_LF_Q14, psDD->LF_AR_Q14, LF_shp_Q14 ); /* Q12 */ + n_LF_Q14 = silk_LSHIFT( n_LF_Q14, 2 ); /* Q12 -> Q14 */ + + /* Input minus prediction plus noise feedback */ + /* r = x[ i ] - LTP_pred - LPC_pred + n_AR + n_Tilt + n_LF + n_LTP */ + tmp1 = silk_ADD32( n_AR_Q14, n_LF_Q14 ); /* Q14 */ + tmp2 = silk_ADD32( n_LTP_Q14, LPC_pred_Q14 ); /* Q13 */ + tmp1 = silk_SUB32( tmp2, tmp1 ); /* Q13 */ + tmp1 = silk_RSHIFT_ROUND( tmp1, 4 ); /* Q10 */ + + r_Q10 = silk_SUB32( x_Q10[ i ], tmp1 ); /* residual error Q10 */ + + /* Flip sign depending on dither */ + if ( psDD->Seed < 0 ) { + r_Q10 = -r_Q10; + } + r_Q10 = silk_LIMIT_32( r_Q10, -(31 << 10), 30 << 10 ); + + /* Find two quantization level candidates and measure their rate-distortion */ + q1_Q10 = silk_SUB32( r_Q10, offset_Q10 ); + q1_Q0 = silk_RSHIFT( q1_Q10, 10 ); + if( q1_Q0 > 0 ) { + q1_Q10 = silk_SUB32( silk_LSHIFT( q1_Q0, 10 ), QUANT_LEVEL_ADJUST_Q10 ); + q1_Q10 = silk_ADD32( q1_Q10, offset_Q10 ); + q2_Q10 = silk_ADD32( q1_Q10, 1024 ); + rd1_Q10 = silk_SMULBB( q1_Q10, Lambda_Q10 ); + rd2_Q10 = silk_SMULBB( q2_Q10, Lambda_Q10 ); + } else if( q1_Q0 == 0 ) { + q1_Q10 = offset_Q10; + q2_Q10 = silk_ADD32( q1_Q10, 1024 - QUANT_LEVEL_ADJUST_Q10 ); + rd1_Q10 = silk_SMULBB( q1_Q10, Lambda_Q10 ); + rd2_Q10 = silk_SMULBB( q2_Q10, Lambda_Q10 ); + } else if( q1_Q0 == -1 ) { + q2_Q10 = offset_Q10; + q1_Q10 = silk_SUB32( q2_Q10, 1024 - QUANT_LEVEL_ADJUST_Q10 ); + rd1_Q10 = silk_SMULBB( -q1_Q10, Lambda_Q10 ); + rd2_Q10 = silk_SMULBB( q2_Q10, Lambda_Q10 ); + } else { /* q1_Q0 < -1 */ + q1_Q10 = silk_ADD32( silk_LSHIFT( q1_Q0, 10 ), QUANT_LEVEL_ADJUST_Q10 ); + q1_Q10 = silk_ADD32( q1_Q10, offset_Q10 ); + q2_Q10 = silk_ADD32( q1_Q10, 1024 ); + rd1_Q10 = silk_SMULBB( -q1_Q10, Lambda_Q10 ); + rd2_Q10 = silk_SMULBB( -q2_Q10, Lambda_Q10 ); + } + rr_Q10 = silk_SUB32( r_Q10, q1_Q10 ); + rd1_Q10 = silk_RSHIFT( silk_SMLABB( rd1_Q10, rr_Q10, rr_Q10 ), 10 ); + rr_Q10 = silk_SUB32( r_Q10, q2_Q10 ); + rd2_Q10 = silk_RSHIFT( silk_SMLABB( rd2_Q10, rr_Q10, rr_Q10 ), 10 ); + + if( rd1_Q10 < rd2_Q10 ) { + psSS[ 0 ].RD_Q10 = silk_ADD32( psDD->RD_Q10, rd1_Q10 ); + psSS[ 1 ].RD_Q10 = silk_ADD32( psDD->RD_Q10, rd2_Q10 ); + psSS[ 0 ].Q_Q10 = q1_Q10; + psSS[ 1 ].Q_Q10 = q2_Q10; + } else { + psSS[ 0 ].RD_Q10 = silk_ADD32( psDD->RD_Q10, rd2_Q10 ); + psSS[ 1 ].RD_Q10 = silk_ADD32( psDD->RD_Q10, rd1_Q10 ); + psSS[ 0 ].Q_Q10 = q2_Q10; + psSS[ 1 ].Q_Q10 = q1_Q10; + } + + /* Update states for best quantization */ + + /* Quantized excitation */ + exc_Q14 = silk_LSHIFT32( psSS[ 0 ].Q_Q10, 4 ); + if ( psDD->Seed < 0 ) { + exc_Q14 = -exc_Q14; + } + + /* Add predictions */ + LPC_exc_Q14 = silk_ADD32( exc_Q14, LTP_pred_Q14 ); + xq_Q14 = silk_ADD32( LPC_exc_Q14, LPC_pred_Q14 ); + + /* Update states */ + sLF_AR_shp_Q14 = silk_SUB32( xq_Q14, n_AR_Q14 ); + psSS[ 0 ].sLTP_shp_Q14 = silk_SUB32( sLF_AR_shp_Q14, n_LF_Q14 ); + psSS[ 0 ].LF_AR_Q14 = sLF_AR_shp_Q14; + psSS[ 0 ].LPC_exc_Q14 = LPC_exc_Q14; + psSS[ 0 ].xq_Q14 = xq_Q14; + + /* Update states for second best quantization */ + + /* Quantized excitation */ + exc_Q14 = silk_LSHIFT32( psSS[ 1 ].Q_Q10, 4 ); + if ( psDD->Seed < 0 ) { + exc_Q14 = -exc_Q14; + } + + + /* Add predictions */ + LPC_exc_Q14 = silk_ADD32( exc_Q14, LTP_pred_Q14 ); + xq_Q14 = silk_ADD32( LPC_exc_Q14, LPC_pred_Q14 ); + + /* Update states */ + sLF_AR_shp_Q14 = silk_SUB32( xq_Q14, n_AR_Q14 ); + psSS[ 1 ].sLTP_shp_Q14 = silk_SUB32( sLF_AR_shp_Q14, n_LF_Q14 ); + psSS[ 1 ].LF_AR_Q14 = sLF_AR_shp_Q14; + psSS[ 1 ].LPC_exc_Q14 = LPC_exc_Q14; + psSS[ 1 ].xq_Q14 = xq_Q14; + } + + *smpl_buf_idx = ( *smpl_buf_idx - 1 ) % DECISION_DELAY; + if( *smpl_buf_idx < 0 ) *smpl_buf_idx += DECISION_DELAY; + last_smple_idx = ( *smpl_buf_idx + decisionDelay ) % DECISION_DELAY; + + /* Find winner */ + RDmin_Q10 = psSampleState[ 0 ][ 0 ].RD_Q10; + Winner_ind = 0; + for( k = 1; k < nStatesDelayedDecision; k++ ) { + if( psSampleState[ k ][ 0 ].RD_Q10 < RDmin_Q10 ) { + RDmin_Q10 = psSampleState[ k ][ 0 ].RD_Q10; + Winner_ind = k; + } + } + + /* Increase RD values of expired states */ + Winner_rand_state = psDelDec[ Winner_ind ].RandState[ last_smple_idx ]; + for( k = 0; k < nStatesDelayedDecision; k++ ) { + if( psDelDec[ k ].RandState[ last_smple_idx ] != Winner_rand_state ) { + psSampleState[ k ][ 0 ].RD_Q10 = silk_ADD32( psSampleState[ k ][ 0 ].RD_Q10, silk_int32_MAX >> 4 ); + psSampleState[ k ][ 1 ].RD_Q10 = silk_ADD32( psSampleState[ k ][ 1 ].RD_Q10, silk_int32_MAX >> 4 ); + silk_assert( psSampleState[ k ][ 0 ].RD_Q10 >= 0 ); + } + } + + /* Find worst in first set and best in second set */ + RDmax_Q10 = psSampleState[ 0 ][ 0 ].RD_Q10; + RDmin_Q10 = psSampleState[ 0 ][ 1 ].RD_Q10; + RDmax_ind = 0; + RDmin_ind = 0; + for( k = 1; k < nStatesDelayedDecision; k++ ) { + /* find worst in first set */ + if( psSampleState[ k ][ 0 ].RD_Q10 > RDmax_Q10 ) { + RDmax_Q10 = psSampleState[ k ][ 0 ].RD_Q10; + RDmax_ind = k; + } + /* find best in second set */ + if( psSampleState[ k ][ 1 ].RD_Q10 < RDmin_Q10 ) { + RDmin_Q10 = psSampleState[ k ][ 1 ].RD_Q10; + RDmin_ind = k; + } + } + + /* Replace a state if best from second set outperforms worst in first set */ + if( RDmin_Q10 < RDmax_Q10 ) { + silk_memcpy( ( (opus_int32 *)&psDelDec[ RDmax_ind ] ) + i, + ( (opus_int32 *)&psDelDec[ RDmin_ind ] ) + i, sizeof( NSQ_del_dec_struct ) - i * sizeof( opus_int32) ); + silk_memcpy( &psSampleState[ RDmax_ind ][ 0 ], &psSampleState[ RDmin_ind ][ 1 ], sizeof( NSQ_sample_struct ) ); + } + + /* Write samples from winner to output and long-term filter states */ + psDD = &psDelDec[ Winner_ind ]; + if( subfr > 0 || i >= decisionDelay ) { + pulses[ i - decisionDelay ] = (opus_int8)silk_RSHIFT_ROUND( psDD->Q_Q10[ last_smple_idx ], 10 ); + xq[ i - decisionDelay ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( + silk_SMULWW( psDD->Xq_Q14[ last_smple_idx ], delayedGain_Q10[ last_smple_idx ] ), 8 ) ); + NSQ->sLTP_shp_Q14[ NSQ->sLTP_shp_buf_idx - decisionDelay ] = psDD->Shape_Q14[ last_smple_idx ]; + sLTP_Q15[ NSQ->sLTP_buf_idx - decisionDelay ] = psDD->Pred_Q15[ last_smple_idx ]; + } + NSQ->sLTP_shp_buf_idx++; + NSQ->sLTP_buf_idx++; + + /* Update states */ + for( k = 0; k < nStatesDelayedDecision; k++ ) { + psDD = &psDelDec[ k ]; + psSS = &psSampleState[ k ][ 0 ]; + psDD->LF_AR_Q14 = psSS->LF_AR_Q14; + psDD->sLPC_Q14[ NSQ_LPC_BUF_LENGTH + i ] = psSS->xq_Q14; + psDD->Xq_Q14[ *smpl_buf_idx ] = psSS->xq_Q14; + psDD->Q_Q10[ *smpl_buf_idx ] = psSS->Q_Q10; + psDD->Pred_Q15[ *smpl_buf_idx ] = silk_LSHIFT32( psSS->LPC_exc_Q14, 1 ); + psDD->Shape_Q14[ *smpl_buf_idx ] = psSS->sLTP_shp_Q14; + psDD->Seed = silk_ADD32_ovflw( psDD->Seed, silk_RSHIFT_ROUND( psSS->Q_Q10, 10 ) ); + psDD->RandState[ *smpl_buf_idx ] = psDD->Seed; + psDD->RD_Q10 = psSS->RD_Q10; + } + delayedGain_Q10[ *smpl_buf_idx ] = Gain_Q10; + } + /* Update LPC states */ + for( k = 0; k < nStatesDelayedDecision; k++ ) { + psDD = &psDelDec[ k ]; + silk_memcpy( psDD->sLPC_Q14, &psDD->sLPC_Q14[ length ], NSQ_LPC_BUF_LENGTH * sizeof( opus_int32 ) ); + } +} + +#endif /* __NSQ_DEL_DEC_MIPSR1_H__ */ diff --git a/media/libopus/silk/mips/macros_mipsr1.h b/media/libopus/silk/mips/macros_mipsr1.h new file mode 100644 index 0000000000..12ed981a6e --- /dev/null +++ b/media/libopus/silk/mips/macros_mipsr1.h @@ -0,0 +1,92 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + + +#ifndef __SILK_MACROS_MIPSR1_H__ +#define __SILK_MACROS_MIPSR1_H__ + +#define mips_clz(x) __builtin_clz(x) + +#undef silk_SMULWB +static inline int silk_SMULWB(int a, int b) +{ + long long ac; + int c; + + ac = __builtin_mips_mult(a, (opus_int32)(opus_int16)b); + c = __builtin_mips_extr_w(ac, 16); + + return c; +} + +#undef silk_SMLAWB +#define silk_SMLAWB(a32, b32, c32) ((a32) + silk_SMULWB(b32, c32)) + +#undef silk_SMULWW +static inline int silk_SMULWW(int a, int b) +{ + long long ac; + int c; + + ac = __builtin_mips_mult(a, b); + c = __builtin_mips_extr_w(ac, 16); + + return c; +} + +#undef silk_SMLAWW +static inline int silk_SMLAWW(int a, int b, int c) +{ + long long ac; + int res; + + ac = __builtin_mips_mult(b, c); + res = __builtin_mips_extr_w(ac, 16); + res += a; + + return res; +} + +#define OVERRIDE_silk_CLZ16 +static inline opus_int32 silk_CLZ16(opus_int16 in16) +{ + int re32; + opus_int32 in32 = (opus_int32 )in16; + re32 = mips_clz(in32); + re32-=16; + return re32; +} + +#define OVERRIDE_silk_CLZ32 +static inline opus_int32 silk_CLZ32(opus_int32 in32) +{ + int re32; + re32 = mips_clz(in32); + return re32; +} + +#endif /* __SILK_MACROS_MIPSR1_H__ */ diff --git a/media/libopus/silk/mips/sigproc_fix_mipsr1.h b/media/libopus/silk/mips/sigproc_fix_mipsr1.h new file mode 100644 index 0000000000..51520c0a6f --- /dev/null +++ b/media/libopus/silk/mips/sigproc_fix_mipsr1.h @@ -0,0 +1,60 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef SILK_SIGPROC_FIX_MIPSR1_H +#define SILK_SIGPROC_FIX_MIPSR1_H + +#undef silk_SAT16 +static inline short int silk_SAT16(int a) +{ + int c; + c = __builtin_mips_shll_s_w(a, 16); + c = c>>16; + + return c; +} + +#undef silk_LSHIFT_SAT32 +static inline int silk_LSHIFT_SAT32(int a, int shift) +{ + int r; + + r = __builtin_mips_shll_s_w(a, shift); + + return r; +} + +#undef silk_RSHIFT_ROUND +static inline int silk_RSHIFT_ROUND(int a, int shift) +{ + int r; + + r = __builtin_mips_shra_r_w(a, shift); + return r; +} + +#endif /* SILK_SIGPROC_FIX_MIPSR1_H */ diff --git a/media/libopus/silk/pitch_est_defines.h b/media/libopus/silk/pitch_est_defines.h new file mode 100644 index 0000000000..e1e4b5d768 --- /dev/null +++ b/media/libopus/silk/pitch_est_defines.h @@ -0,0 +1,88 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef SILK_PE_DEFINES_H +#define SILK_PE_DEFINES_H + +#include "SigProc_FIX.h" + +/********************************************************/ +/* Definitions for pitch estimator */ +/********************************************************/ + +#define PE_MAX_FS_KHZ 16 /* Maximum sampling frequency used */ + +#define PE_MAX_NB_SUBFR 4 +#define PE_SUBFR_LENGTH_MS 5 /* 5 ms */ + +#define PE_LTP_MEM_LENGTH_MS ( 4 * PE_SUBFR_LENGTH_MS ) + +#define PE_MAX_FRAME_LENGTH_MS ( PE_LTP_MEM_LENGTH_MS + PE_MAX_NB_SUBFR * PE_SUBFR_LENGTH_MS ) +#define PE_MAX_FRAME_LENGTH ( PE_MAX_FRAME_LENGTH_MS * PE_MAX_FS_KHZ ) +#define PE_MAX_FRAME_LENGTH_ST_1 ( PE_MAX_FRAME_LENGTH >> 2 ) +#define PE_MAX_FRAME_LENGTH_ST_2 ( PE_MAX_FRAME_LENGTH >> 1 ) + +#define PE_MAX_LAG_MS 18 /* 18 ms -> 56 Hz */ +#define PE_MIN_LAG_MS 2 /* 2 ms -> 500 Hz */ +#define PE_MAX_LAG ( PE_MAX_LAG_MS * PE_MAX_FS_KHZ ) +#define PE_MIN_LAG ( PE_MIN_LAG_MS * PE_MAX_FS_KHZ ) + +#define PE_D_SRCH_LENGTH 24 + +#define PE_NB_STAGE3_LAGS 5 + +#define PE_NB_CBKS_STAGE2 3 +#define PE_NB_CBKS_STAGE2_EXT 11 + +#define PE_NB_CBKS_STAGE3_MAX 34 +#define PE_NB_CBKS_STAGE3_MID 24 +#define PE_NB_CBKS_STAGE3_MIN 16 + +#define PE_NB_CBKS_STAGE3_10MS 12 +#define PE_NB_CBKS_STAGE2_10MS 3 + +#define PE_SHORTLAG_BIAS 0.2f /* for logarithmic weighting */ +#define PE_PREVLAG_BIAS 0.2f /* for logarithmic weighting */ +#define PE_FLATCONTOUR_BIAS 0.05f + +#define SILK_PE_MIN_COMPLEX 0 +#define SILK_PE_MID_COMPLEX 1 +#define SILK_PE_MAX_COMPLEX 2 + +/* Tables for 20 ms frames */ +extern const opus_int8 silk_CB_lags_stage2[ PE_MAX_NB_SUBFR ][ PE_NB_CBKS_STAGE2_EXT ]; +extern const opus_int8 silk_CB_lags_stage3[ PE_MAX_NB_SUBFR ][ PE_NB_CBKS_STAGE3_MAX ]; +extern const opus_int8 silk_Lag_range_stage3[ SILK_PE_MAX_COMPLEX + 1 ] [ PE_MAX_NB_SUBFR ][ 2 ]; +extern const opus_int8 silk_nb_cbk_searchs_stage3[ SILK_PE_MAX_COMPLEX + 1 ]; + +/* Tables for 10 ms frames */ +extern const opus_int8 silk_CB_lags_stage2_10_ms[ PE_MAX_NB_SUBFR >> 1][ 3 ]; +extern const opus_int8 silk_CB_lags_stage3_10_ms[ PE_MAX_NB_SUBFR >> 1 ][ 12 ]; +extern const opus_int8 silk_Lag_range_stage3_10_ms[ PE_MAX_NB_SUBFR >> 1 ][ 2 ]; + +#endif + diff --git a/media/libopus/silk/pitch_est_tables.c b/media/libopus/silk/pitch_est_tables.c new file mode 100644 index 0000000000..81a8bacaca --- /dev/null +++ b/media/libopus/silk/pitch_est_tables.c @@ -0,0 +1,99 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "typedef.h" +#include "pitch_est_defines.h" + +const opus_int8 silk_CB_lags_stage2_10_ms[ PE_MAX_NB_SUBFR >> 1][ PE_NB_CBKS_STAGE2_10MS ] = +{ + {0, 1, 0}, + {0, 0, 1} +}; + +const opus_int8 silk_CB_lags_stage3_10_ms[ PE_MAX_NB_SUBFR >> 1 ][ PE_NB_CBKS_STAGE3_10MS ] = +{ + { 0, 0, 1,-1, 1,-1, 2,-2, 2,-2, 3,-3}, + { 0, 1, 0, 1,-1, 2,-1, 2,-2, 3,-2, 3} +}; + +const opus_int8 silk_Lag_range_stage3_10_ms[ PE_MAX_NB_SUBFR >> 1 ][ 2 ] = +{ + {-3, 7}, + {-2, 7} +}; + +const opus_int8 silk_CB_lags_stage2[ PE_MAX_NB_SUBFR ][ PE_NB_CBKS_STAGE2_EXT ] = +{ + {0, 2,-1,-1,-1, 0, 0, 1, 1, 0, 1}, + {0, 1, 0, 0, 0, 0, 0, 1, 0, 0, 0}, + {0, 0, 1, 0, 0, 0, 1, 0, 0, 0, 0}, + {0,-1, 2, 1, 0, 1, 1, 0, 0,-1,-1} +}; + +const opus_int8 silk_CB_lags_stage3[ PE_MAX_NB_SUBFR ][ PE_NB_CBKS_STAGE3_MAX ] = +{ + {0, 0, 1,-1, 0, 1,-1, 0,-1, 1,-2, 2,-2,-2, 2,-3, 2, 3,-3,-4, 3,-4, 4, 4,-5, 5,-6,-5, 6,-7, 6, 5, 8,-9}, + {0, 0, 1, 0, 0, 0, 0, 0, 0, 0,-1, 1, 0, 0, 1,-1, 0, 1,-1,-1, 1,-1, 2, 1,-1, 2,-2,-2, 2,-2, 2, 2, 3,-3}, + {0, 1, 0, 0, 0, 0, 0, 0, 1, 0, 1, 0, 0, 1,-1, 1, 0, 0, 2, 1,-1, 2,-1,-1, 2,-1, 2, 2,-1, 3,-2,-2,-2, 3}, + {0, 1, 0, 0, 1, 0, 1,-1, 2,-1, 2,-1, 2, 3,-2, 3,-2,-2, 4, 4,-3, 5,-3,-4, 6,-4, 6, 5,-5, 8,-6,-5,-7, 9} +}; + +const opus_int8 silk_Lag_range_stage3[ SILK_PE_MAX_COMPLEX + 1 ] [ PE_MAX_NB_SUBFR ][ 2 ] = +{ + /* Lags to search for low number of stage3 cbks */ + { + {-5,8}, + {-1,6}, + {-1,6}, + {-4,10} + }, + /* Lags to search for middle number of stage3 cbks */ + { + {-6,10}, + {-2,6}, + {-1,6}, + {-5,10} + }, + /* Lags to search for max number of stage3 cbks */ + { + {-9,12}, + {-3,7}, + {-2,7}, + {-7,13} + } +}; + +const opus_int8 silk_nb_cbk_searchs_stage3[ SILK_PE_MAX_COMPLEX + 1 ] = +{ + PE_NB_CBKS_STAGE3_MIN, + PE_NB_CBKS_STAGE3_MID, + PE_NB_CBKS_STAGE3_MAX +}; diff --git a/media/libopus/silk/process_NLSFs.c b/media/libopus/silk/process_NLSFs.c new file mode 100644 index 0000000000..d130809541 --- /dev/null +++ b/media/libopus/silk/process_NLSFs.c @@ -0,0 +1,107 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" + +/* Limit, stabilize, convert and quantize NLSFs */ +void silk_process_NLSFs( + silk_encoder_state *psEncC, /* I/O Encoder state */ + opus_int16 PredCoef_Q12[ 2 ][ MAX_LPC_ORDER ], /* O Prediction coefficients */ + opus_int16 pNLSF_Q15[ MAX_LPC_ORDER ], /* I/O Normalized LSFs (quant out) (0 - (2^15-1)) */ + const opus_int16 prev_NLSFq_Q15[ MAX_LPC_ORDER ] /* I Previous Normalized LSFs (0 - (2^15-1)) */ +) +{ + opus_int i, doInterpolate; + opus_int NLSF_mu_Q20; + opus_int16 i_sqr_Q15; + opus_int16 pNLSF0_temp_Q15[ MAX_LPC_ORDER ]; + opus_int16 pNLSFW_QW[ MAX_LPC_ORDER ]; + opus_int16 pNLSFW0_temp_QW[ MAX_LPC_ORDER ]; + + silk_assert( psEncC->speech_activity_Q8 >= 0 ); + silk_assert( psEncC->speech_activity_Q8 <= SILK_FIX_CONST( 1.0, 8 ) ); + celt_assert( psEncC->useInterpolatedNLSFs == 1 || psEncC->indices.NLSFInterpCoef_Q2 == ( 1 << 2 ) ); + + /***********************/ + /* Calculate mu values */ + /***********************/ + /* NLSF_mu = 0.003 - 0.0015 * psEnc->speech_activity; */ + NLSF_mu_Q20 = silk_SMLAWB( SILK_FIX_CONST( 0.003, 20 ), SILK_FIX_CONST( -0.001, 28 ), psEncC->speech_activity_Q8 ); + if( psEncC->nb_subfr == 2 ) { + /* Multiply by 1.5 for 10 ms packets */ + NLSF_mu_Q20 = silk_ADD_RSHIFT( NLSF_mu_Q20, NLSF_mu_Q20, 1 ); + } + + celt_assert( NLSF_mu_Q20 > 0 ); + silk_assert( NLSF_mu_Q20 <= SILK_FIX_CONST( 0.005, 20 ) ); + + /* Calculate NLSF weights */ + silk_NLSF_VQ_weights_laroia( pNLSFW_QW, pNLSF_Q15, psEncC->predictLPCOrder ); + + /* Update NLSF weights for interpolated NLSFs */ + doInterpolate = ( psEncC->useInterpolatedNLSFs == 1 ) && ( psEncC->indices.NLSFInterpCoef_Q2 < 4 ); + if( doInterpolate ) { + /* Calculate the interpolated NLSF vector for the first half */ + silk_interpolate( pNLSF0_temp_Q15, prev_NLSFq_Q15, pNLSF_Q15, + psEncC->indices.NLSFInterpCoef_Q2, psEncC->predictLPCOrder ); + + /* Calculate first half NLSF weights for the interpolated NLSFs */ + silk_NLSF_VQ_weights_laroia( pNLSFW0_temp_QW, pNLSF0_temp_Q15, psEncC->predictLPCOrder ); + + /* Update NLSF weights with contribution from first half */ + i_sqr_Q15 = silk_LSHIFT( silk_SMULBB( psEncC->indices.NLSFInterpCoef_Q2, psEncC->indices.NLSFInterpCoef_Q2 ), 11 ); + for( i = 0; i < psEncC->predictLPCOrder; i++ ) { + pNLSFW_QW[ i ] = silk_ADD16( silk_RSHIFT( pNLSFW_QW[ i ], 1 ), silk_RSHIFT( + silk_SMULBB( pNLSFW0_temp_QW[ i ], i_sqr_Q15 ), 16) ); + silk_assert( pNLSFW_QW[ i ] >= 1 ); + } + } + + silk_NLSF_encode( psEncC->indices.NLSFIndices, pNLSF_Q15, psEncC->psNLSF_CB, pNLSFW_QW, + NLSF_mu_Q20, psEncC->NLSF_MSVQ_Survivors, psEncC->indices.signalType ); + + /* Convert quantized NLSFs back to LPC coefficients */ + silk_NLSF2A( PredCoef_Q12[ 1 ], pNLSF_Q15, psEncC->predictLPCOrder, psEncC->arch ); + + if( doInterpolate ) { + /* Calculate the interpolated, quantized LSF vector for the first half */ + silk_interpolate( pNLSF0_temp_Q15, prev_NLSFq_Q15, pNLSF_Q15, + psEncC->indices.NLSFInterpCoef_Q2, psEncC->predictLPCOrder ); + + /* Convert back to LPC coefficients */ + silk_NLSF2A( PredCoef_Q12[ 0 ], pNLSF0_temp_Q15, psEncC->predictLPCOrder, psEncC->arch ); + + } else { + /* Copy LPC coefficients for first half from second half */ + celt_assert( psEncC->predictLPCOrder <= MAX_LPC_ORDER ); + silk_memcpy( PredCoef_Q12[ 0 ], PredCoef_Q12[ 1 ], psEncC->predictLPCOrder * sizeof( opus_int16 ) ); + } +} diff --git a/media/libopus/silk/quant_LTP_gains.c b/media/libopus/silk/quant_LTP_gains.c new file mode 100644 index 0000000000..d6b8eff8d1 --- /dev/null +++ b/media/libopus/silk/quant_LTP_gains.c @@ -0,0 +1,132 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" +#include "tuning_parameters.h" + +void silk_quant_LTP_gains( + opus_int16 B_Q14[ MAX_NB_SUBFR * LTP_ORDER ], /* O Quantized LTP gains */ + opus_int8 cbk_index[ MAX_NB_SUBFR ], /* O Codebook Index */ + opus_int8 *periodicity_index, /* O Periodicity Index */ + opus_int32 *sum_log_gain_Q7, /* I/O Cumulative max prediction gain */ + opus_int *pred_gain_dB_Q7, /* O LTP prediction gain */ + const opus_int32 XX_Q17[ MAX_NB_SUBFR*LTP_ORDER*LTP_ORDER ], /* I Correlation matrix in Q18 */ + const opus_int32 xX_Q17[ MAX_NB_SUBFR*LTP_ORDER ], /* I Correlation vector in Q18 */ + const opus_int subfr_len, /* I Number of samples per subframe */ + const opus_int nb_subfr, /* I Number of subframes */ + int arch /* I Run-time architecture */ +) +{ + opus_int j, k, cbk_size; + opus_int8 temp_idx[ MAX_NB_SUBFR ]; + const opus_uint8 *cl_ptr_Q5; + const opus_int8 *cbk_ptr_Q7; + const opus_uint8 *cbk_gain_ptr_Q7; + const opus_int32 *XX_Q17_ptr, *xX_Q17_ptr; + opus_int32 res_nrg_Q15_subfr, res_nrg_Q15, rate_dist_Q7_subfr, rate_dist_Q7, min_rate_dist_Q7; + opus_int32 sum_log_gain_tmp_Q7, best_sum_log_gain_Q7, max_gain_Q7; + opus_int gain_Q7; + + /***************************************************/ + /* iterate over different codebooks with different */ + /* rates/distortions, and choose best */ + /***************************************************/ + min_rate_dist_Q7 = silk_int32_MAX; + best_sum_log_gain_Q7 = 0; + for( k = 0; k < 3; k++ ) { + /* Safety margin for pitch gain control, to take into account factors + such as state rescaling/rewhitening. */ + opus_int32 gain_safety = SILK_FIX_CONST( 0.4, 7 ); + + cl_ptr_Q5 = silk_LTP_gain_BITS_Q5_ptrs[ k ]; + cbk_ptr_Q7 = silk_LTP_vq_ptrs_Q7[ k ]; + cbk_gain_ptr_Q7 = silk_LTP_vq_gain_ptrs_Q7[ k ]; + cbk_size = silk_LTP_vq_sizes[ k ]; + + /* Set up pointers to first subframe */ + XX_Q17_ptr = XX_Q17; + xX_Q17_ptr = xX_Q17; + + res_nrg_Q15 = 0; + rate_dist_Q7 = 0; + sum_log_gain_tmp_Q7 = *sum_log_gain_Q7; + for( j = 0; j < nb_subfr; j++ ) { + max_gain_Q7 = silk_log2lin( ( SILK_FIX_CONST( MAX_SUM_LOG_GAIN_DB / 6.0, 7 ) - sum_log_gain_tmp_Q7 ) + + SILK_FIX_CONST( 7, 7 ) ) - gain_safety; + silk_VQ_WMat_EC( + &temp_idx[ j ], /* O index of best codebook vector */ + &res_nrg_Q15_subfr, /* O residual energy */ + &rate_dist_Q7_subfr, /* O best weighted quantization error + mu * rate */ + &gain_Q7, /* O sum of absolute LTP coefficients */ + XX_Q17_ptr, /* I correlation matrix */ + xX_Q17_ptr, /* I correlation vector */ + cbk_ptr_Q7, /* I codebook */ + cbk_gain_ptr_Q7, /* I codebook effective gains */ + cl_ptr_Q5, /* I code length for each codebook vector */ + subfr_len, /* I number of samples per subframe */ + max_gain_Q7, /* I maximum sum of absolute LTP coefficients */ + cbk_size, /* I number of vectors in codebook */ + arch /* I Run-time architecture */ + ); + + res_nrg_Q15 = silk_ADD_POS_SAT32( res_nrg_Q15, res_nrg_Q15_subfr ); + rate_dist_Q7 = silk_ADD_POS_SAT32( rate_dist_Q7, rate_dist_Q7_subfr ); + sum_log_gain_tmp_Q7 = silk_max(0, sum_log_gain_tmp_Q7 + + silk_lin2log( gain_safety + gain_Q7 ) - SILK_FIX_CONST( 7, 7 )); + + XX_Q17_ptr += LTP_ORDER * LTP_ORDER; + xX_Q17_ptr += LTP_ORDER; + } + + if( rate_dist_Q7 <= min_rate_dist_Q7 ) { + min_rate_dist_Q7 = rate_dist_Q7; + *periodicity_index = (opus_int8)k; + silk_memcpy( cbk_index, temp_idx, nb_subfr * sizeof( opus_int8 ) ); + best_sum_log_gain_Q7 = sum_log_gain_tmp_Q7; + } + } + + cbk_ptr_Q7 = silk_LTP_vq_ptrs_Q7[ *periodicity_index ]; + for( j = 0; j < nb_subfr; j++ ) { + for( k = 0; k < LTP_ORDER; k++ ) { + B_Q14[ j * LTP_ORDER + k ] = silk_LSHIFT( cbk_ptr_Q7[ cbk_index[ j ] * LTP_ORDER + k ], 7 ); + } + } + + if( nb_subfr == 2 ) { + res_nrg_Q15 = silk_RSHIFT32( res_nrg_Q15, 1 ); + } else { + res_nrg_Q15 = silk_RSHIFT32( res_nrg_Q15, 2 ); + } + + *sum_log_gain_Q7 = best_sum_log_gain_Q7; + *pred_gain_dB_Q7 = (opus_int)silk_SMULBB( -3, silk_lin2log( res_nrg_Q15 ) - ( 15 << 7 ) ); +} diff --git a/media/libopus/silk/resampler.c b/media/libopus/silk/resampler.c new file mode 100644 index 0000000000..1f11e50891 --- /dev/null +++ b/media/libopus/silk/resampler.c @@ -0,0 +1,215 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +/* + * Matrix of resampling methods used: + * Fs_out (kHz) + * 8 12 16 24 48 + * + * 8 C UF U UF UF + * 12 AF C UF U UF + * Fs_in (kHz) 16 D AF C UF UF + * 24 AF D AF C U + * 48 AF AF AF D C + * + * C -> Copy (no resampling) + * D -> Allpass-based 2x downsampling + * U -> Allpass-based 2x upsampling + * UF -> Allpass-based 2x upsampling followed by FIR interpolation + * AF -> AR2 filter followed by FIR interpolation + */ + +#include "resampler_private.h" + +/* Tables with delay compensation values to equalize total delay for different modes */ +static const opus_int8 delay_matrix_enc[ 5 ][ 3 ] = { +/* in \ out 8 12 16 */ +/* 8 */ { 6, 0, 3 }, +/* 12 */ { 0, 7, 3 }, +/* 16 */ { 0, 1, 10 }, +/* 24 */ { 0, 2, 6 }, +/* 48 */ { 18, 10, 12 } +}; + +static const opus_int8 delay_matrix_dec[ 3 ][ 5 ] = { +/* in \ out 8 12 16 24 48 */ +/* 8 */ { 4, 0, 2, 0, 0 }, +/* 12 */ { 0, 9, 4, 7, 4 }, +/* 16 */ { 0, 3, 12, 7, 7 } +}; + +/* Simple way to make [8000, 12000, 16000, 24000, 48000] to [0, 1, 2, 3, 4] */ +#define rateID(R) ( ( ( ((R)>>12) - ((R)>16000) ) >> ((R)>24000) ) - 1 ) + +#define USE_silk_resampler_copy (0) +#define USE_silk_resampler_private_up2_HQ_wrapper (1) +#define USE_silk_resampler_private_IIR_FIR (2) +#define USE_silk_resampler_private_down_FIR (3) + +/* Initialize/reset the resampler state for a given pair of input/output sampling rates */ +opus_int silk_resampler_init( + silk_resampler_state_struct *S, /* I/O Resampler state */ + opus_int32 Fs_Hz_in, /* I Input sampling rate (Hz) */ + opus_int32 Fs_Hz_out, /* I Output sampling rate (Hz) */ + opus_int forEnc /* I If 1: encoder; if 0: decoder */ +) +{ + opus_int up2x; + + /* Clear state */ + silk_memset( S, 0, sizeof( silk_resampler_state_struct ) ); + + /* Input checking */ + if( forEnc ) { + if( ( Fs_Hz_in != 8000 && Fs_Hz_in != 12000 && Fs_Hz_in != 16000 && Fs_Hz_in != 24000 && Fs_Hz_in != 48000 ) || + ( Fs_Hz_out != 8000 && Fs_Hz_out != 12000 && Fs_Hz_out != 16000 ) ) { + celt_assert( 0 ); + return -1; + } + S->inputDelay = delay_matrix_enc[ rateID( Fs_Hz_in ) ][ rateID( Fs_Hz_out ) ]; + } else { + if( ( Fs_Hz_in != 8000 && Fs_Hz_in != 12000 && Fs_Hz_in != 16000 ) || + ( Fs_Hz_out != 8000 && Fs_Hz_out != 12000 && Fs_Hz_out != 16000 && Fs_Hz_out != 24000 && Fs_Hz_out != 48000 ) ) { + celt_assert( 0 ); + return -1; + } + S->inputDelay = delay_matrix_dec[ rateID( Fs_Hz_in ) ][ rateID( Fs_Hz_out ) ]; + } + + S->Fs_in_kHz = silk_DIV32_16( Fs_Hz_in, 1000 ); + S->Fs_out_kHz = silk_DIV32_16( Fs_Hz_out, 1000 ); + + /* Number of samples processed per batch */ + S->batchSize = S->Fs_in_kHz * RESAMPLER_MAX_BATCH_SIZE_MS; + + /* Find resampler with the right sampling ratio */ + up2x = 0; + if( Fs_Hz_out > Fs_Hz_in ) { + /* Upsample */ + if( Fs_Hz_out == silk_MUL( Fs_Hz_in, 2 ) ) { /* Fs_out : Fs_in = 2 : 1 */ + /* Special case: directly use 2x upsampler */ + S->resampler_function = USE_silk_resampler_private_up2_HQ_wrapper; + } else { + /* Default resampler */ + S->resampler_function = USE_silk_resampler_private_IIR_FIR; + up2x = 1; + } + } else if ( Fs_Hz_out < Fs_Hz_in ) { + /* Downsample */ + S->resampler_function = USE_silk_resampler_private_down_FIR; + if( silk_MUL( Fs_Hz_out, 4 ) == silk_MUL( Fs_Hz_in, 3 ) ) { /* Fs_out : Fs_in = 3 : 4 */ + S->FIR_Fracs = 3; + S->FIR_Order = RESAMPLER_DOWN_ORDER_FIR0; + S->Coefs = silk_Resampler_3_4_COEFS; + } else if( silk_MUL( Fs_Hz_out, 3 ) == silk_MUL( Fs_Hz_in, 2 ) ) { /* Fs_out : Fs_in = 2 : 3 */ + S->FIR_Fracs = 2; + S->FIR_Order = RESAMPLER_DOWN_ORDER_FIR0; + S->Coefs = silk_Resampler_2_3_COEFS; + } else if( silk_MUL( Fs_Hz_out, 2 ) == Fs_Hz_in ) { /* Fs_out : Fs_in = 1 : 2 */ + S->FIR_Fracs = 1; + S->FIR_Order = RESAMPLER_DOWN_ORDER_FIR1; + S->Coefs = silk_Resampler_1_2_COEFS; + } else if( silk_MUL( Fs_Hz_out, 3 ) == Fs_Hz_in ) { /* Fs_out : Fs_in = 1 : 3 */ + S->FIR_Fracs = 1; + S->FIR_Order = RESAMPLER_DOWN_ORDER_FIR2; + S->Coefs = silk_Resampler_1_3_COEFS; + } else if( silk_MUL( Fs_Hz_out, 4 ) == Fs_Hz_in ) { /* Fs_out : Fs_in = 1 : 4 */ + S->FIR_Fracs = 1; + S->FIR_Order = RESAMPLER_DOWN_ORDER_FIR2; + S->Coefs = silk_Resampler_1_4_COEFS; + } else if( silk_MUL( Fs_Hz_out, 6 ) == Fs_Hz_in ) { /* Fs_out : Fs_in = 1 : 6 */ + S->FIR_Fracs = 1; + S->FIR_Order = RESAMPLER_DOWN_ORDER_FIR2; + S->Coefs = silk_Resampler_1_6_COEFS; + } else { + /* None available */ + celt_assert( 0 ); + return -1; + } + } else { + /* Input and output sampling rates are equal: copy */ + S->resampler_function = USE_silk_resampler_copy; + } + + /* Ratio of input/output samples */ + S->invRatio_Q16 = silk_LSHIFT32( silk_DIV32( silk_LSHIFT32( Fs_Hz_in, 14 + up2x ), Fs_Hz_out ), 2 ); + /* Make sure the ratio is rounded up */ + while( silk_SMULWW( S->invRatio_Q16, Fs_Hz_out ) < silk_LSHIFT32( Fs_Hz_in, up2x ) ) { + S->invRatio_Q16++; + } + + return 0; +} + +/* Resampler: convert from one sampling rate to another */ +/* Input and output sampling rate are at most 48000 Hz */ +opus_int silk_resampler( + silk_resampler_state_struct *S, /* I/O Resampler state */ + opus_int16 out[], /* O Output signal */ + const opus_int16 in[], /* I Input signal */ + opus_int32 inLen /* I Number of input samples */ +) +{ + opus_int nSamples; + + /* Need at least 1 ms of input data */ + celt_assert( inLen >= S->Fs_in_kHz ); + /* Delay can't exceed the 1 ms of buffering */ + celt_assert( S->inputDelay <= S->Fs_in_kHz ); + + nSamples = S->Fs_in_kHz - S->inputDelay; + + /* Copy to delay buffer */ + silk_memcpy( &S->delayBuf[ S->inputDelay ], in, nSamples * sizeof( opus_int16 ) ); + + switch( S->resampler_function ) { + case USE_silk_resampler_private_up2_HQ_wrapper: + silk_resampler_private_up2_HQ_wrapper( S, out, S->delayBuf, S->Fs_in_kHz ); + silk_resampler_private_up2_HQ_wrapper( S, &out[ S->Fs_out_kHz ], &in[ nSamples ], inLen - S->Fs_in_kHz ); + break; + case USE_silk_resampler_private_IIR_FIR: + silk_resampler_private_IIR_FIR( S, out, S->delayBuf, S->Fs_in_kHz ); + silk_resampler_private_IIR_FIR( S, &out[ S->Fs_out_kHz ], &in[ nSamples ], inLen - S->Fs_in_kHz ); + break; + case USE_silk_resampler_private_down_FIR: + silk_resampler_private_down_FIR( S, out, S->delayBuf, S->Fs_in_kHz ); + silk_resampler_private_down_FIR( S, &out[ S->Fs_out_kHz ], &in[ nSamples ], inLen - S->Fs_in_kHz ); + break; + default: + silk_memcpy( out, S->delayBuf, S->Fs_in_kHz * sizeof( opus_int16 ) ); + silk_memcpy( &out[ S->Fs_out_kHz ], &in[ nSamples ], ( inLen - S->Fs_in_kHz ) * sizeof( opus_int16 ) ); + } + + /* Copy to delay buffer */ + silk_memcpy( S->delayBuf, &in[ inLen - S->inputDelay ], S->inputDelay * sizeof( opus_int16 ) ); + + return 0; +} diff --git a/media/libopus/silk/resampler_down2.c b/media/libopus/silk/resampler_down2.c new file mode 100644 index 0000000000..971d7bfd4a --- /dev/null +++ b/media/libopus/silk/resampler_down2.c @@ -0,0 +1,74 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "SigProc_FIX.h" +#include "resampler_rom.h" + +/* Downsample by a factor 2 */ +void silk_resampler_down2( + opus_int32 *S, /* I/O State vector [ 2 ] */ + opus_int16 *out, /* O Output signal [ floor(len/2) ] */ + const opus_int16 *in, /* I Input signal [ len ] */ + opus_int32 inLen /* I Number of input samples */ +) +{ + opus_int32 k, len2 = silk_RSHIFT32( inLen, 1 ); + opus_int32 in32, out32, Y, X; + + celt_assert( silk_resampler_down2_0 > 0 ); + celt_assert( silk_resampler_down2_1 < 0 ); + + /* Internal variables and state are in Q10 format */ + for( k = 0; k < len2; k++ ) { + /* Convert to Q10 */ + in32 = silk_LSHIFT( (opus_int32)in[ 2 * k ], 10 ); + + /* All-pass section for even input sample */ + Y = silk_SUB32( in32, S[ 0 ] ); + X = silk_SMLAWB( Y, Y, silk_resampler_down2_1 ); + out32 = silk_ADD32( S[ 0 ], X ); + S[ 0 ] = silk_ADD32( in32, X ); + + /* Convert to Q10 */ + in32 = silk_LSHIFT( (opus_int32)in[ 2 * k + 1 ], 10 ); + + /* All-pass section for odd input sample, and add to output of previous section */ + Y = silk_SUB32( in32, S[ 1 ] ); + X = silk_SMULWB( Y, silk_resampler_down2_0 ); + out32 = silk_ADD32( out32, S[ 1 ] ); + out32 = silk_ADD32( out32, X ); + S[ 1 ] = silk_ADD32( in32, X ); + + /* Add, convert back to int16 and store to output */ + out[ k ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( out32, 11 ) ); + } +} + diff --git a/media/libopus/silk/resampler_down2_3.c b/media/libopus/silk/resampler_down2_3.c new file mode 100644 index 0000000000..4342614dcc --- /dev/null +++ b/media/libopus/silk/resampler_down2_3.c @@ -0,0 +1,103 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "SigProc_FIX.h" +#include "resampler_private.h" +#include "stack_alloc.h" + +#define ORDER_FIR 4 + +/* Downsample by a factor 2/3, low quality */ +void silk_resampler_down2_3( + opus_int32 *S, /* I/O State vector [ 6 ] */ + opus_int16 *out, /* O Output signal [ floor(2*inLen/3) ] */ + const opus_int16 *in, /* I Input signal [ inLen ] */ + opus_int32 inLen /* I Number of input samples */ +) +{ + opus_int32 nSamplesIn, counter, res_Q6; + VARDECL( opus_int32, buf ); + opus_int32 *buf_ptr; + SAVE_STACK; + + ALLOC( buf, RESAMPLER_MAX_BATCH_SIZE_IN + ORDER_FIR, opus_int32 ); + + /* Copy buffered samples to start of buffer */ + silk_memcpy( buf, S, ORDER_FIR * sizeof( opus_int32 ) ); + + /* Iterate over blocks of frameSizeIn input samples */ + while( 1 ) { + nSamplesIn = silk_min( inLen, RESAMPLER_MAX_BATCH_SIZE_IN ); + + /* Second-order AR filter (output in Q8) */ + silk_resampler_private_AR2( &S[ ORDER_FIR ], &buf[ ORDER_FIR ], in, + silk_Resampler_2_3_COEFS_LQ, nSamplesIn ); + + /* Interpolate filtered signal */ + buf_ptr = buf; + counter = nSamplesIn; + while( counter > 2 ) { + /* Inner product */ + res_Q6 = silk_SMULWB( buf_ptr[ 0 ], silk_Resampler_2_3_COEFS_LQ[ 2 ] ); + res_Q6 = silk_SMLAWB( res_Q6, buf_ptr[ 1 ], silk_Resampler_2_3_COEFS_LQ[ 3 ] ); + res_Q6 = silk_SMLAWB( res_Q6, buf_ptr[ 2 ], silk_Resampler_2_3_COEFS_LQ[ 5 ] ); + res_Q6 = silk_SMLAWB( res_Q6, buf_ptr[ 3 ], silk_Resampler_2_3_COEFS_LQ[ 4 ] ); + + /* Scale down, saturate and store in output array */ + *out++ = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( res_Q6, 6 ) ); + + res_Q6 = silk_SMULWB( buf_ptr[ 1 ], silk_Resampler_2_3_COEFS_LQ[ 4 ] ); + res_Q6 = silk_SMLAWB( res_Q6, buf_ptr[ 2 ], silk_Resampler_2_3_COEFS_LQ[ 5 ] ); + res_Q6 = silk_SMLAWB( res_Q6, buf_ptr[ 3 ], silk_Resampler_2_3_COEFS_LQ[ 3 ] ); + res_Q6 = silk_SMLAWB( res_Q6, buf_ptr[ 4 ], silk_Resampler_2_3_COEFS_LQ[ 2 ] ); + + /* Scale down, saturate and store in output array */ + *out++ = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( res_Q6, 6 ) ); + + buf_ptr += 3; + counter -= 3; + } + + in += nSamplesIn; + inLen -= nSamplesIn; + + if( inLen > 0 ) { + /* More iterations to do; copy last part of filtered signal to beginning of buffer */ + silk_memcpy( buf, &buf[ nSamplesIn ], ORDER_FIR * sizeof( opus_int32 ) ); + } else { + break; + } + } + + /* Copy last part of filtered signal to the state for the next call */ + silk_memcpy( S, &buf[ nSamplesIn ], ORDER_FIR * sizeof( opus_int32 ) ); + RESTORE_STACK; +} diff --git a/media/libopus/silk/resampler_private.h b/media/libopus/silk/resampler_private.h new file mode 100644 index 0000000000..422a7d9d95 --- /dev/null +++ b/media/libopus/silk/resampler_private.h @@ -0,0 +1,88 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef SILK_RESAMPLER_PRIVATE_H +#define SILK_RESAMPLER_PRIVATE_H + +#ifdef __cplusplus +extern "C" { +#endif + +#include "SigProc_FIX.h" +#include "resampler_structs.h" +#include "resampler_rom.h" + +/* Number of input samples to process in the inner loop */ +#define RESAMPLER_MAX_BATCH_SIZE_MS 10 +#define RESAMPLER_MAX_FS_KHZ 48 +#define RESAMPLER_MAX_BATCH_SIZE_IN ( RESAMPLER_MAX_BATCH_SIZE_MS * RESAMPLER_MAX_FS_KHZ ) + +/* Description: Hybrid IIR/FIR polyphase implementation of resampling */ +void silk_resampler_private_IIR_FIR( + void *SS, /* I/O Resampler state */ + opus_int16 out[], /* O Output signal */ + const opus_int16 in[], /* I Input signal */ + opus_int32 inLen /* I Number of input samples */ +); + +/* Description: Hybrid IIR/FIR polyphase implementation of resampling */ +void silk_resampler_private_down_FIR( + void *SS, /* I/O Resampler state */ + opus_int16 out[], /* O Output signal */ + const opus_int16 in[], /* I Input signal */ + opus_int32 inLen /* I Number of input samples */ +); + +/* Upsample by a factor 2, high quality */ +void silk_resampler_private_up2_HQ_wrapper( + void *SS, /* I/O Resampler state (unused) */ + opus_int16 *out, /* O Output signal [ 2 * len ] */ + const opus_int16 *in, /* I Input signal [ len ] */ + opus_int32 len /* I Number of input samples */ +); + +/* Upsample by a factor 2, high quality */ +void silk_resampler_private_up2_HQ( + opus_int32 *S, /* I/O Resampler state [ 6 ] */ + opus_int16 *out, /* O Output signal [ 2 * len ] */ + const opus_int16 *in, /* I Input signal [ len ] */ + opus_int32 len /* I Number of input samples */ +); + +/* Second order AR filter */ +void silk_resampler_private_AR2( + opus_int32 S[], /* I/O State vector [ 2 ] */ + opus_int32 out_Q8[], /* O Output signal */ + const opus_int16 in[], /* I Input signal */ + const opus_int16 A_Q14[], /* I AR coefficients, Q14 */ + opus_int32 len /* I Signal length */ +); + +#ifdef __cplusplus +} +#endif +#endif /* SILK_RESAMPLER_PRIVATE_H */ diff --git a/media/libopus/silk/resampler_private_AR2.c b/media/libopus/silk/resampler_private_AR2.c new file mode 100644 index 0000000000..5fff23714f --- /dev/null +++ b/media/libopus/silk/resampler_private_AR2.c @@ -0,0 +1,55 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "SigProc_FIX.h" +#include "resampler_private.h" + +/* Second order AR filter with single delay elements */ +void silk_resampler_private_AR2( + opus_int32 S[], /* I/O State vector [ 2 ] */ + opus_int32 out_Q8[], /* O Output signal */ + const opus_int16 in[], /* I Input signal */ + const opus_int16 A_Q14[], /* I AR coefficients, Q14 */ + opus_int32 len /* I Signal length */ +) +{ + opus_int32 k; + opus_int32 out32; + + for( k = 0; k < len; k++ ) { + out32 = silk_ADD_LSHIFT32( S[ 0 ], (opus_int32)in[ k ], 8 ); + out_Q8[ k ] = out32; + out32 = silk_LSHIFT( out32, 2 ); + S[ 0 ] = silk_SMLAWB( S[ 1 ], out32, A_Q14[ 0 ] ); + S[ 1 ] = silk_SMULWB( out32, A_Q14[ 1 ] ); + } +} + diff --git a/media/libopus/silk/resampler_private_IIR_FIR.c b/media/libopus/silk/resampler_private_IIR_FIR.c new file mode 100644 index 0000000000..6b2b3a2e18 --- /dev/null +++ b/media/libopus/silk/resampler_private_IIR_FIR.c @@ -0,0 +1,107 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "SigProc_FIX.h" +#include "resampler_private.h" +#include "stack_alloc.h" + +static OPUS_INLINE opus_int16 *silk_resampler_private_IIR_FIR_INTERPOL( + opus_int16 *out, + opus_int16 *buf, + opus_int32 max_index_Q16, + opus_int32 index_increment_Q16 +) +{ + opus_int32 index_Q16, res_Q15; + opus_int16 *buf_ptr; + opus_int32 table_index; + + /* Interpolate upsampled signal and store in output array */ + for( index_Q16 = 0; index_Q16 < max_index_Q16; index_Q16 += index_increment_Q16 ) { + table_index = silk_SMULWB( index_Q16 & 0xFFFF, 12 ); + buf_ptr = &buf[ index_Q16 >> 16 ]; + + res_Q15 = silk_SMULBB( buf_ptr[ 0 ], silk_resampler_frac_FIR_12[ table_index ][ 0 ] ); + res_Q15 = silk_SMLABB( res_Q15, buf_ptr[ 1 ], silk_resampler_frac_FIR_12[ table_index ][ 1 ] ); + res_Q15 = silk_SMLABB( res_Q15, buf_ptr[ 2 ], silk_resampler_frac_FIR_12[ table_index ][ 2 ] ); + res_Q15 = silk_SMLABB( res_Q15, buf_ptr[ 3 ], silk_resampler_frac_FIR_12[ table_index ][ 3 ] ); + res_Q15 = silk_SMLABB( res_Q15, buf_ptr[ 4 ], silk_resampler_frac_FIR_12[ 11 - table_index ][ 3 ] ); + res_Q15 = silk_SMLABB( res_Q15, buf_ptr[ 5 ], silk_resampler_frac_FIR_12[ 11 - table_index ][ 2 ] ); + res_Q15 = silk_SMLABB( res_Q15, buf_ptr[ 6 ], silk_resampler_frac_FIR_12[ 11 - table_index ][ 1 ] ); + res_Q15 = silk_SMLABB( res_Q15, buf_ptr[ 7 ], silk_resampler_frac_FIR_12[ 11 - table_index ][ 0 ] ); + *out++ = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( res_Q15, 15 ) ); + } + return out; +} +/* Upsample using a combination of allpass-based 2x upsampling and FIR interpolation */ +void silk_resampler_private_IIR_FIR( + void *SS, /* I/O Resampler state */ + opus_int16 out[], /* O Output signal */ + const opus_int16 in[], /* I Input signal */ + opus_int32 inLen /* I Number of input samples */ +) +{ + silk_resampler_state_struct *S = (silk_resampler_state_struct *)SS; + opus_int32 nSamplesIn; + opus_int32 max_index_Q16, index_increment_Q16; + VARDECL( opus_int16, buf ); + SAVE_STACK; + + ALLOC( buf, 2 * S->batchSize + RESAMPLER_ORDER_FIR_12, opus_int16 ); + + /* Copy buffered samples to start of buffer */ + silk_memcpy( buf, S->sFIR.i16, RESAMPLER_ORDER_FIR_12 * sizeof( opus_int16 ) ); + + /* Iterate over blocks of frameSizeIn input samples */ + index_increment_Q16 = S->invRatio_Q16; + while( 1 ) { + nSamplesIn = silk_min( inLen, S->batchSize ); + + /* Upsample 2x */ + silk_resampler_private_up2_HQ( S->sIIR, &buf[ RESAMPLER_ORDER_FIR_12 ], in, nSamplesIn ); + + max_index_Q16 = silk_LSHIFT32( nSamplesIn, 16 + 1 ); /* + 1 because 2x upsampling */ + out = silk_resampler_private_IIR_FIR_INTERPOL( out, buf, max_index_Q16, index_increment_Q16 ); + in += nSamplesIn; + inLen -= nSamplesIn; + + if( inLen > 0 ) { + /* More iterations to do; copy last part of filtered signal to beginning of buffer */ + silk_memcpy( buf, &buf[ nSamplesIn << 1 ], RESAMPLER_ORDER_FIR_12 * sizeof( opus_int16 ) ); + } else { + break; + } + } + + /* Copy last part of filtered signal to the state for the next call */ + silk_memcpy( S->sFIR.i16, &buf[ nSamplesIn << 1 ], RESAMPLER_ORDER_FIR_12 * sizeof( opus_int16 ) ); + RESTORE_STACK; +} diff --git a/media/libopus/silk/resampler_private_down_FIR.c b/media/libopus/silk/resampler_private_down_FIR.c new file mode 100644 index 0000000000..3e8735a35a --- /dev/null +++ b/media/libopus/silk/resampler_private_down_FIR.c @@ -0,0 +1,194 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "SigProc_FIX.h" +#include "resampler_private.h" +#include "stack_alloc.h" + +static OPUS_INLINE opus_int16 *silk_resampler_private_down_FIR_INTERPOL( + opus_int16 *out, + opus_int32 *buf, + const opus_int16 *FIR_Coefs, + opus_int FIR_Order, + opus_int FIR_Fracs, + opus_int32 max_index_Q16, + opus_int32 index_increment_Q16 +) +{ + opus_int32 index_Q16, res_Q6; + opus_int32 *buf_ptr; + opus_int32 interpol_ind; + const opus_int16 *interpol_ptr; + + switch( FIR_Order ) { + case RESAMPLER_DOWN_ORDER_FIR0: + for( index_Q16 = 0; index_Q16 < max_index_Q16; index_Q16 += index_increment_Q16 ) { + /* Integer part gives pointer to buffered input */ + buf_ptr = buf + silk_RSHIFT( index_Q16, 16 ); + + /* Fractional part gives interpolation coefficients */ + interpol_ind = silk_SMULWB( index_Q16 & 0xFFFF, FIR_Fracs ); + + /* Inner product */ + interpol_ptr = &FIR_Coefs[ RESAMPLER_DOWN_ORDER_FIR0 / 2 * interpol_ind ]; + res_Q6 = silk_SMULWB( buf_ptr[ 0 ], interpol_ptr[ 0 ] ); + res_Q6 = silk_SMLAWB( res_Q6, buf_ptr[ 1 ], interpol_ptr[ 1 ] ); + res_Q6 = silk_SMLAWB( res_Q6, buf_ptr[ 2 ], interpol_ptr[ 2 ] ); + res_Q6 = silk_SMLAWB( res_Q6, buf_ptr[ 3 ], interpol_ptr[ 3 ] ); + res_Q6 = silk_SMLAWB( res_Q6, buf_ptr[ 4 ], interpol_ptr[ 4 ] ); + res_Q6 = silk_SMLAWB( res_Q6, buf_ptr[ 5 ], interpol_ptr[ 5 ] ); + res_Q6 = silk_SMLAWB( res_Q6, buf_ptr[ 6 ], interpol_ptr[ 6 ] ); + res_Q6 = silk_SMLAWB( res_Q6, buf_ptr[ 7 ], interpol_ptr[ 7 ] ); + res_Q6 = silk_SMLAWB( res_Q6, buf_ptr[ 8 ], interpol_ptr[ 8 ] ); + interpol_ptr = &FIR_Coefs[ RESAMPLER_DOWN_ORDER_FIR0 / 2 * ( FIR_Fracs - 1 - interpol_ind ) ]; + res_Q6 = silk_SMLAWB( res_Q6, buf_ptr[ 17 ], interpol_ptr[ 0 ] ); + res_Q6 = silk_SMLAWB( res_Q6, buf_ptr[ 16 ], interpol_ptr[ 1 ] ); + res_Q6 = silk_SMLAWB( res_Q6, buf_ptr[ 15 ], interpol_ptr[ 2 ] ); + res_Q6 = silk_SMLAWB( res_Q6, buf_ptr[ 14 ], interpol_ptr[ 3 ] ); + res_Q6 = silk_SMLAWB( res_Q6, buf_ptr[ 13 ], interpol_ptr[ 4 ] ); + res_Q6 = silk_SMLAWB( res_Q6, buf_ptr[ 12 ], interpol_ptr[ 5 ] ); + res_Q6 = silk_SMLAWB( res_Q6, buf_ptr[ 11 ], interpol_ptr[ 6 ] ); + res_Q6 = silk_SMLAWB( res_Q6, buf_ptr[ 10 ], interpol_ptr[ 7 ] ); + res_Q6 = silk_SMLAWB( res_Q6, buf_ptr[ 9 ], interpol_ptr[ 8 ] ); + + /* Scale down, saturate and store in output array */ + *out++ = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( res_Q6, 6 ) ); + } + break; + case RESAMPLER_DOWN_ORDER_FIR1: + for( index_Q16 = 0; index_Q16 < max_index_Q16; index_Q16 += index_increment_Q16 ) { + /* Integer part gives pointer to buffered input */ + buf_ptr = buf + silk_RSHIFT( index_Q16, 16 ); + + /* Inner product */ + res_Q6 = silk_SMULWB( silk_ADD32( buf_ptr[ 0 ], buf_ptr[ 23 ] ), FIR_Coefs[ 0 ] ); + res_Q6 = silk_SMLAWB( res_Q6, silk_ADD32( buf_ptr[ 1 ], buf_ptr[ 22 ] ), FIR_Coefs[ 1 ] ); + res_Q6 = silk_SMLAWB( res_Q6, silk_ADD32( buf_ptr[ 2 ], buf_ptr[ 21 ] ), FIR_Coefs[ 2 ] ); + res_Q6 = silk_SMLAWB( res_Q6, silk_ADD32( buf_ptr[ 3 ], buf_ptr[ 20 ] ), FIR_Coefs[ 3 ] ); + res_Q6 = silk_SMLAWB( res_Q6, silk_ADD32( buf_ptr[ 4 ], buf_ptr[ 19 ] ), FIR_Coefs[ 4 ] ); + res_Q6 = silk_SMLAWB( res_Q6, silk_ADD32( buf_ptr[ 5 ], buf_ptr[ 18 ] ), FIR_Coefs[ 5 ] ); + res_Q6 = silk_SMLAWB( res_Q6, silk_ADD32( buf_ptr[ 6 ], buf_ptr[ 17 ] ), FIR_Coefs[ 6 ] ); + res_Q6 = silk_SMLAWB( res_Q6, silk_ADD32( buf_ptr[ 7 ], buf_ptr[ 16 ] ), FIR_Coefs[ 7 ] ); + res_Q6 = silk_SMLAWB( res_Q6, silk_ADD32( buf_ptr[ 8 ], buf_ptr[ 15 ] ), FIR_Coefs[ 8 ] ); + res_Q6 = silk_SMLAWB( res_Q6, silk_ADD32( buf_ptr[ 9 ], buf_ptr[ 14 ] ), FIR_Coefs[ 9 ] ); + res_Q6 = silk_SMLAWB( res_Q6, silk_ADD32( buf_ptr[ 10 ], buf_ptr[ 13 ] ), FIR_Coefs[ 10 ] ); + res_Q6 = silk_SMLAWB( res_Q6, silk_ADD32( buf_ptr[ 11 ], buf_ptr[ 12 ] ), FIR_Coefs[ 11 ] ); + + /* Scale down, saturate and store in output array */ + *out++ = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( res_Q6, 6 ) ); + } + break; + case RESAMPLER_DOWN_ORDER_FIR2: + for( index_Q16 = 0; index_Q16 < max_index_Q16; index_Q16 += index_increment_Q16 ) { + /* Integer part gives pointer to buffered input */ + buf_ptr = buf + silk_RSHIFT( index_Q16, 16 ); + + /* Inner product */ + res_Q6 = silk_SMULWB( silk_ADD32( buf_ptr[ 0 ], buf_ptr[ 35 ] ), FIR_Coefs[ 0 ] ); + res_Q6 = silk_SMLAWB( res_Q6, silk_ADD32( buf_ptr[ 1 ], buf_ptr[ 34 ] ), FIR_Coefs[ 1 ] ); + res_Q6 = silk_SMLAWB( res_Q6, silk_ADD32( buf_ptr[ 2 ], buf_ptr[ 33 ] ), FIR_Coefs[ 2 ] ); + res_Q6 = silk_SMLAWB( res_Q6, silk_ADD32( buf_ptr[ 3 ], buf_ptr[ 32 ] ), FIR_Coefs[ 3 ] ); + res_Q6 = silk_SMLAWB( res_Q6, silk_ADD32( buf_ptr[ 4 ], buf_ptr[ 31 ] ), FIR_Coefs[ 4 ] ); + res_Q6 = silk_SMLAWB( res_Q6, silk_ADD32( buf_ptr[ 5 ], buf_ptr[ 30 ] ), FIR_Coefs[ 5 ] ); + res_Q6 = silk_SMLAWB( res_Q6, silk_ADD32( buf_ptr[ 6 ], buf_ptr[ 29 ] ), FIR_Coefs[ 6 ] ); + res_Q6 = silk_SMLAWB( res_Q6, silk_ADD32( buf_ptr[ 7 ], buf_ptr[ 28 ] ), FIR_Coefs[ 7 ] ); + res_Q6 = silk_SMLAWB( res_Q6, silk_ADD32( buf_ptr[ 8 ], buf_ptr[ 27 ] ), FIR_Coefs[ 8 ] ); + res_Q6 = silk_SMLAWB( res_Q6, silk_ADD32( buf_ptr[ 9 ], buf_ptr[ 26 ] ), FIR_Coefs[ 9 ] ); + res_Q6 = silk_SMLAWB( res_Q6, silk_ADD32( buf_ptr[ 10 ], buf_ptr[ 25 ] ), FIR_Coefs[ 10 ] ); + res_Q6 = silk_SMLAWB( res_Q6, silk_ADD32( buf_ptr[ 11 ], buf_ptr[ 24 ] ), FIR_Coefs[ 11 ] ); + res_Q6 = silk_SMLAWB( res_Q6, silk_ADD32( buf_ptr[ 12 ], buf_ptr[ 23 ] ), FIR_Coefs[ 12 ] ); + res_Q6 = silk_SMLAWB( res_Q6, silk_ADD32( buf_ptr[ 13 ], buf_ptr[ 22 ] ), FIR_Coefs[ 13 ] ); + res_Q6 = silk_SMLAWB( res_Q6, silk_ADD32( buf_ptr[ 14 ], buf_ptr[ 21 ] ), FIR_Coefs[ 14 ] ); + res_Q6 = silk_SMLAWB( res_Q6, silk_ADD32( buf_ptr[ 15 ], buf_ptr[ 20 ] ), FIR_Coefs[ 15 ] ); + res_Q6 = silk_SMLAWB( res_Q6, silk_ADD32( buf_ptr[ 16 ], buf_ptr[ 19 ] ), FIR_Coefs[ 16 ] ); + res_Q6 = silk_SMLAWB( res_Q6, silk_ADD32( buf_ptr[ 17 ], buf_ptr[ 18 ] ), FIR_Coefs[ 17 ] ); + + /* Scale down, saturate and store in output array */ + *out++ = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( res_Q6, 6 ) ); + } + break; + default: + celt_assert( 0 ); + } + return out; +} + +/* Resample with a 2nd order AR filter followed by FIR interpolation */ +void silk_resampler_private_down_FIR( + void *SS, /* I/O Resampler state */ + opus_int16 out[], /* O Output signal */ + const opus_int16 in[], /* I Input signal */ + opus_int32 inLen /* I Number of input samples */ +) +{ + silk_resampler_state_struct *S = (silk_resampler_state_struct *)SS; + opus_int32 nSamplesIn; + opus_int32 max_index_Q16, index_increment_Q16; + VARDECL( opus_int32, buf ); + const opus_int16 *FIR_Coefs; + SAVE_STACK; + + ALLOC( buf, S->batchSize + S->FIR_Order, opus_int32 ); + + /* Copy buffered samples to start of buffer */ + silk_memcpy( buf, S->sFIR.i32, S->FIR_Order * sizeof( opus_int32 ) ); + + FIR_Coefs = &S->Coefs[ 2 ]; + + /* Iterate over blocks of frameSizeIn input samples */ + index_increment_Q16 = S->invRatio_Q16; + while( 1 ) { + nSamplesIn = silk_min( inLen, S->batchSize ); + + /* Second-order AR filter (output in Q8) */ + silk_resampler_private_AR2( S->sIIR, &buf[ S->FIR_Order ], in, S->Coefs, nSamplesIn ); + + max_index_Q16 = silk_LSHIFT32( nSamplesIn, 16 ); + + /* Interpolate filtered signal */ + out = silk_resampler_private_down_FIR_INTERPOL( out, buf, FIR_Coefs, S->FIR_Order, + S->FIR_Fracs, max_index_Q16, index_increment_Q16 ); + + in += nSamplesIn; + inLen -= nSamplesIn; + + if( inLen > 1 ) { + /* More iterations to do; copy last part of filtered signal to beginning of buffer */ + silk_memcpy( buf, &buf[ nSamplesIn ], S->FIR_Order * sizeof( opus_int32 ) ); + } else { + break; + } + } + + /* Copy last part of filtered signal to the state for the next call */ + silk_memcpy( S->sFIR.i32, &buf[ nSamplesIn ], S->FIR_Order * sizeof( opus_int32 ) ); + RESTORE_STACK; +} diff --git a/media/libopus/silk/resampler_private_up2_HQ.c b/media/libopus/silk/resampler_private_up2_HQ.c new file mode 100644 index 0000000000..c7ec8de365 --- /dev/null +++ b/media/libopus/silk/resampler_private_up2_HQ.c @@ -0,0 +1,113 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "SigProc_FIX.h" +#include "resampler_private.h" + +/* Upsample by a factor 2, high quality */ +/* Uses 2nd order allpass filters for the 2x upsampling, followed by a */ +/* notch filter just above Nyquist. */ +void silk_resampler_private_up2_HQ( + opus_int32 *S, /* I/O Resampler state [ 6 ] */ + opus_int16 *out, /* O Output signal [ 2 * len ] */ + const opus_int16 *in, /* I Input signal [ len ] */ + opus_int32 len /* I Number of input samples */ +) +{ + opus_int32 k; + opus_int32 in32, out32_1, out32_2, Y, X; + + silk_assert( silk_resampler_up2_hq_0[ 0 ] > 0 ); + silk_assert( silk_resampler_up2_hq_0[ 1 ] > 0 ); + silk_assert( silk_resampler_up2_hq_0[ 2 ] < 0 ); + silk_assert( silk_resampler_up2_hq_1[ 0 ] > 0 ); + silk_assert( silk_resampler_up2_hq_1[ 1 ] > 0 ); + silk_assert( silk_resampler_up2_hq_1[ 2 ] < 0 ); + + /* Internal variables and state are in Q10 format */ + for( k = 0; k < len; k++ ) { + /* Convert to Q10 */ + in32 = silk_LSHIFT( (opus_int32)in[ k ], 10 ); + + /* First all-pass section for even output sample */ + Y = silk_SUB32( in32, S[ 0 ] ); + X = silk_SMULWB( Y, silk_resampler_up2_hq_0[ 0 ] ); + out32_1 = silk_ADD32( S[ 0 ], X ); + S[ 0 ] = silk_ADD32( in32, X ); + + /* Second all-pass section for even output sample */ + Y = silk_SUB32( out32_1, S[ 1 ] ); + X = silk_SMULWB( Y, silk_resampler_up2_hq_0[ 1 ] ); + out32_2 = silk_ADD32( S[ 1 ], X ); + S[ 1 ] = silk_ADD32( out32_1, X ); + + /* Third all-pass section for even output sample */ + Y = silk_SUB32( out32_2, S[ 2 ] ); + X = silk_SMLAWB( Y, Y, silk_resampler_up2_hq_0[ 2 ] ); + out32_1 = silk_ADD32( S[ 2 ], X ); + S[ 2 ] = silk_ADD32( out32_2, X ); + + /* Apply gain in Q15, convert back to int16 and store to output */ + out[ 2 * k ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( out32_1, 10 ) ); + + /* First all-pass section for odd output sample */ + Y = silk_SUB32( in32, S[ 3 ] ); + X = silk_SMULWB( Y, silk_resampler_up2_hq_1[ 0 ] ); + out32_1 = silk_ADD32( S[ 3 ], X ); + S[ 3 ] = silk_ADD32( in32, X ); + + /* Second all-pass section for odd output sample */ + Y = silk_SUB32( out32_1, S[ 4 ] ); + X = silk_SMULWB( Y, silk_resampler_up2_hq_1[ 1 ] ); + out32_2 = silk_ADD32( S[ 4 ], X ); + S[ 4 ] = silk_ADD32( out32_1, X ); + + /* Third all-pass section for odd output sample */ + Y = silk_SUB32( out32_2, S[ 5 ] ); + X = silk_SMLAWB( Y, Y, silk_resampler_up2_hq_1[ 2 ] ); + out32_1 = silk_ADD32( S[ 5 ], X ); + S[ 5 ] = silk_ADD32( out32_2, X ); + + /* Apply gain in Q15, convert back to int16 and store to output */ + out[ 2 * k + 1 ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( out32_1, 10 ) ); + } +} + +void silk_resampler_private_up2_HQ_wrapper( + void *SS, /* I/O Resampler state (unused) */ + opus_int16 *out, /* O Output signal [ 2 * len ] */ + const opus_int16 *in, /* I Input signal [ len ] */ + opus_int32 len /* I Number of input samples */ +) +{ + silk_resampler_state_struct *S = (silk_resampler_state_struct *)SS; + silk_resampler_private_up2_HQ( S->sIIR, out, in, len ); +} diff --git a/media/libopus/silk/resampler_rom.c b/media/libopus/silk/resampler_rom.c new file mode 100644 index 0000000000..5e6b04476a --- /dev/null +++ b/media/libopus/silk/resampler_rom.c @@ -0,0 +1,96 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +/* Filter coefficients for IIR/FIR polyphase resampling * + * Total size: 179 Words (358 Bytes) */ + +#include "resampler_private.h" + +/* Matlab code for the notch filter coefficients: */ +/* B = [1, 0.147, 1]; A = [1, 0.107, 0.89]; G = 0.93; freqz(G * B, A, 2^14, 16e3); axis([0, 8000, -10, 1]) */ +/* fprintf('\t%6d, %6d, %6d, %6d\n', round(B(2)*2^16), round(-A(2)*2^16), round((1-A(3))*2^16), round(G*2^15)) */ +/* const opus_int16 silk_resampler_up2_hq_notch[ 4 ] = { 9634, -7012, 7209, 30474 }; */ + +/* Tables with IIR and FIR coefficients for fractional downsamplers (123 Words) */ +silk_DWORD_ALIGN const opus_int16 silk_Resampler_3_4_COEFS[ 2 + 3 * RESAMPLER_DOWN_ORDER_FIR0 / 2 ] = { + -20694, -13867, + -49, 64, 17, -157, 353, -496, 163, 11047, 22205, + -39, 6, 91, -170, 186, 23, -896, 6336, 19928, + -19, -36, 102, -89, -24, 328, -951, 2568, 15909, +}; + +silk_DWORD_ALIGN const opus_int16 silk_Resampler_2_3_COEFS[ 2 + 2 * RESAMPLER_DOWN_ORDER_FIR0 / 2 ] = { + -14457, -14019, + 64, 128, -122, 36, 310, -768, 584, 9267, 17733, + 12, 128, 18, -142, 288, -117, -865, 4123, 14459, +}; + +silk_DWORD_ALIGN const opus_int16 silk_Resampler_1_2_COEFS[ 2 + RESAMPLER_DOWN_ORDER_FIR1 / 2 ] = { + 616, -14323, + -10, 39, 58, -46, -84, 120, 184, -315, -541, 1284, 5380, 9024, +}; + +silk_DWORD_ALIGN const opus_int16 silk_Resampler_1_3_COEFS[ 2 + RESAMPLER_DOWN_ORDER_FIR2 / 2 ] = { + 16102, -15162, + -13, 0, 20, 26, 5, -31, -43, -4, 65, 90, 7, -157, -248, -44, 593, 1583, 2612, 3271, +}; + +silk_DWORD_ALIGN const opus_int16 silk_Resampler_1_4_COEFS[ 2 + RESAMPLER_DOWN_ORDER_FIR2 / 2 ] = { + 22500, -15099, + 3, -14, -20, -15, 2, 25, 37, 25, -16, -71, -107, -79, 50, 292, 623, 982, 1288, 1464, +}; + +silk_DWORD_ALIGN const opus_int16 silk_Resampler_1_6_COEFS[ 2 + RESAMPLER_DOWN_ORDER_FIR2 / 2 ] = { + 27540, -15257, + 17, 12, 8, 1, -10, -22, -30, -32, -22, 3, 44, 100, 168, 243, 317, 381, 429, 455, +}; + +silk_DWORD_ALIGN const opus_int16 silk_Resampler_2_3_COEFS_LQ[ 2 + 2 * 2 ] = { + -2797, -6507, + 4697, 10739, + 1567, 8276, +}; + +/* Table with interplation fractions of 1/24, 3/24, 5/24, ... , 23/24 : 23/24 (46 Words) */ +silk_DWORD_ALIGN const opus_int16 silk_resampler_frac_FIR_12[ 12 ][ RESAMPLER_ORDER_FIR_12 / 2 ] = { + { 189, -600, 617, 30567 }, + { 117, -159, -1070, 29704 }, + { 52, 221, -2392, 28276 }, + { -4, 529, -3350, 26341 }, + { -48, 758, -3956, 23973 }, + { -80, 905, -4235, 21254 }, + { -99, 972, -4222, 18278 }, + { -107, 967, -3957, 15143 }, + { -103, 896, -3487, 11950 }, + { -91, 773, -2865, 8798 }, + { -71, 611, -2143, 5784 }, + { -46, 425, -1375, 2996 }, +}; diff --git a/media/libopus/silk/resampler_rom.h b/media/libopus/silk/resampler_rom.h new file mode 100644 index 0000000000..490b3388dc --- /dev/null +++ b/media/libopus/silk/resampler_rom.h @@ -0,0 +1,68 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef SILK_FIX_RESAMPLER_ROM_H +#define SILK_FIX_RESAMPLER_ROM_H + +#ifdef __cplusplus +extern "C" +{ +#endif + +#include "typedef.h" +#include "resampler_structs.h" + +#define RESAMPLER_DOWN_ORDER_FIR0 18 +#define RESAMPLER_DOWN_ORDER_FIR1 24 +#define RESAMPLER_DOWN_ORDER_FIR2 36 +#define RESAMPLER_ORDER_FIR_12 8 + +/* Tables for 2x downsampler */ +static const opus_int16 silk_resampler_down2_0 = 9872; +static const opus_int16 silk_resampler_down2_1 = 39809 - 65536; + +/* Tables for 2x upsampler, high quality */ +static const opus_int16 silk_resampler_up2_hq_0[ 3 ] = { 1746, 14986, 39083 - 65536 }; +static const opus_int16 silk_resampler_up2_hq_1[ 3 ] = { 6854, 25769, 55542 - 65536 }; + +/* Tables with IIR and FIR coefficients for fractional downsamplers */ +extern const opus_int16 silk_Resampler_3_4_COEFS[ 2 + 3 * RESAMPLER_DOWN_ORDER_FIR0 / 2 ]; +extern const opus_int16 silk_Resampler_2_3_COEFS[ 2 + 2 * RESAMPLER_DOWN_ORDER_FIR0 / 2 ]; +extern const opus_int16 silk_Resampler_1_2_COEFS[ 2 + RESAMPLER_DOWN_ORDER_FIR1 / 2 ]; +extern const opus_int16 silk_Resampler_1_3_COEFS[ 2 + RESAMPLER_DOWN_ORDER_FIR2 / 2 ]; +extern const opus_int16 silk_Resampler_1_4_COEFS[ 2 + RESAMPLER_DOWN_ORDER_FIR2 / 2 ]; +extern const opus_int16 silk_Resampler_1_6_COEFS[ 2 + RESAMPLER_DOWN_ORDER_FIR2 / 2 ]; +extern const opus_int16 silk_Resampler_2_3_COEFS_LQ[ 2 + 2 * 2 ]; + +/* Table with interplation fractions of 1/24, 3/24, ..., 23/24 */ +extern const opus_int16 silk_resampler_frac_FIR_12[ 12 ][ RESAMPLER_ORDER_FIR_12 / 2 ]; + +#ifdef __cplusplus +} +#endif + +#endif /* SILK_FIX_RESAMPLER_ROM_H */ diff --git a/media/libopus/silk/resampler_structs.h b/media/libopus/silk/resampler_structs.h new file mode 100644 index 0000000000..9e9457d11c --- /dev/null +++ b/media/libopus/silk/resampler_structs.h @@ -0,0 +1,60 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef SILK_RESAMPLER_STRUCTS_H +#define SILK_RESAMPLER_STRUCTS_H + +#ifdef __cplusplus +extern "C" { +#endif + +#define SILK_RESAMPLER_MAX_FIR_ORDER 36 +#define SILK_RESAMPLER_MAX_IIR_ORDER 6 + +typedef struct _silk_resampler_state_struct{ + opus_int32 sIIR[ SILK_RESAMPLER_MAX_IIR_ORDER ]; /* this must be the first element of this struct */ + union{ + opus_int32 i32[ SILK_RESAMPLER_MAX_FIR_ORDER ]; + opus_int16 i16[ SILK_RESAMPLER_MAX_FIR_ORDER ]; + } sFIR; + opus_int16 delayBuf[ 48 ]; + opus_int resampler_function; + opus_int batchSize; + opus_int32 invRatio_Q16; + opus_int FIR_Order; + opus_int FIR_Fracs; + opus_int Fs_in_kHz; + opus_int Fs_out_kHz; + opus_int inputDelay; + const opus_int16 *Coefs; +} silk_resampler_state_struct; + +#ifdef __cplusplus +} +#endif +#endif /* SILK_RESAMPLER_STRUCTS_H */ + diff --git a/media/libopus/silk/shell_coder.c b/media/libopus/silk/shell_coder.c new file mode 100644 index 0000000000..4af341474b --- /dev/null +++ b/media/libopus/silk/shell_coder.c @@ -0,0 +1,151 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" + +/* shell coder; pulse-subframe length is hardcoded */ + +static OPUS_INLINE void combine_pulses( + opus_int *out, /* O combined pulses vector [len] */ + const opus_int *in, /* I input vector [2 * len] */ + const opus_int len /* I number of OUTPUT samples */ +) +{ + opus_int k; + for( k = 0; k < len; k++ ) { + out[ k ] = in[ 2 * k ] + in[ 2 * k + 1 ]; + } +} + +static OPUS_INLINE void encode_split( + ec_enc *psRangeEnc, /* I/O compressor data structure */ + const opus_int p_child1, /* I pulse amplitude of first child subframe */ + const opus_int p, /* I pulse amplitude of current subframe */ + const opus_uint8 *shell_table /* I table of shell cdfs */ +) +{ + if( p > 0 ) { + ec_enc_icdf( psRangeEnc, p_child1, &shell_table[ silk_shell_code_table_offsets[ p ] ], 8 ); + } +} + +static OPUS_INLINE void decode_split( + opus_int16 *p_child1, /* O pulse amplitude of first child subframe */ + opus_int16 *p_child2, /* O pulse amplitude of second child subframe */ + ec_dec *psRangeDec, /* I/O Compressor data structure */ + const opus_int p, /* I pulse amplitude of current subframe */ + const opus_uint8 *shell_table /* I table of shell cdfs */ +) +{ + if( p > 0 ) { + p_child1[ 0 ] = ec_dec_icdf( psRangeDec, &shell_table[ silk_shell_code_table_offsets[ p ] ], 8 ); + p_child2[ 0 ] = p - p_child1[ 0 ]; + } else { + p_child1[ 0 ] = 0; + p_child2[ 0 ] = 0; + } +} + +/* Shell encoder, operates on one shell code frame of 16 pulses */ +void silk_shell_encoder( + ec_enc *psRangeEnc, /* I/O compressor data structure */ + const opus_int *pulses0 /* I data: nonnegative pulse amplitudes */ +) +{ + opus_int pulses1[ 8 ], pulses2[ 4 ], pulses3[ 2 ], pulses4[ 1 ]; + + /* this function operates on one shell code frame of 16 pulses */ + silk_assert( SHELL_CODEC_FRAME_LENGTH == 16 ); + + /* tree representation per pulse-subframe */ + combine_pulses( pulses1, pulses0, 8 ); + combine_pulses( pulses2, pulses1, 4 ); + combine_pulses( pulses3, pulses2, 2 ); + combine_pulses( pulses4, pulses3, 1 ); + + encode_split( psRangeEnc, pulses3[ 0 ], pulses4[ 0 ], silk_shell_code_table3 ); + + encode_split( psRangeEnc, pulses2[ 0 ], pulses3[ 0 ], silk_shell_code_table2 ); + + encode_split( psRangeEnc, pulses1[ 0 ], pulses2[ 0 ], silk_shell_code_table1 ); + encode_split( psRangeEnc, pulses0[ 0 ], pulses1[ 0 ], silk_shell_code_table0 ); + encode_split( psRangeEnc, pulses0[ 2 ], pulses1[ 1 ], silk_shell_code_table0 ); + + encode_split( psRangeEnc, pulses1[ 2 ], pulses2[ 1 ], silk_shell_code_table1 ); + encode_split( psRangeEnc, pulses0[ 4 ], pulses1[ 2 ], silk_shell_code_table0 ); + encode_split( psRangeEnc, pulses0[ 6 ], pulses1[ 3 ], silk_shell_code_table0 ); + + encode_split( psRangeEnc, pulses2[ 2 ], pulses3[ 1 ], silk_shell_code_table2 ); + + encode_split( psRangeEnc, pulses1[ 4 ], pulses2[ 2 ], silk_shell_code_table1 ); + encode_split( psRangeEnc, pulses0[ 8 ], pulses1[ 4 ], silk_shell_code_table0 ); + encode_split( psRangeEnc, pulses0[ 10 ], pulses1[ 5 ], silk_shell_code_table0 ); + + encode_split( psRangeEnc, pulses1[ 6 ], pulses2[ 3 ], silk_shell_code_table1 ); + encode_split( psRangeEnc, pulses0[ 12 ], pulses1[ 6 ], silk_shell_code_table0 ); + encode_split( psRangeEnc, pulses0[ 14 ], pulses1[ 7 ], silk_shell_code_table0 ); +} + + +/* Shell decoder, operates on one shell code frame of 16 pulses */ +void silk_shell_decoder( + opus_int16 *pulses0, /* O data: nonnegative pulse amplitudes */ + ec_dec *psRangeDec, /* I/O Compressor data structure */ + const opus_int pulses4 /* I number of pulses per pulse-subframe */ +) +{ + opus_int16 pulses3[ 2 ], pulses2[ 4 ], pulses1[ 8 ]; + + /* this function operates on one shell code frame of 16 pulses */ + silk_assert( SHELL_CODEC_FRAME_LENGTH == 16 ); + + decode_split( &pulses3[ 0 ], &pulses3[ 1 ], psRangeDec, pulses4, silk_shell_code_table3 ); + + decode_split( &pulses2[ 0 ], &pulses2[ 1 ], psRangeDec, pulses3[ 0 ], silk_shell_code_table2 ); + + decode_split( &pulses1[ 0 ], &pulses1[ 1 ], psRangeDec, pulses2[ 0 ], silk_shell_code_table1 ); + decode_split( &pulses0[ 0 ], &pulses0[ 1 ], psRangeDec, pulses1[ 0 ], silk_shell_code_table0 ); + decode_split( &pulses0[ 2 ], &pulses0[ 3 ], psRangeDec, pulses1[ 1 ], silk_shell_code_table0 ); + + decode_split( &pulses1[ 2 ], &pulses1[ 3 ], psRangeDec, pulses2[ 1 ], silk_shell_code_table1 ); + decode_split( &pulses0[ 4 ], &pulses0[ 5 ], psRangeDec, pulses1[ 2 ], silk_shell_code_table0 ); + decode_split( &pulses0[ 6 ], &pulses0[ 7 ], psRangeDec, pulses1[ 3 ], silk_shell_code_table0 ); + + decode_split( &pulses2[ 2 ], &pulses2[ 3 ], psRangeDec, pulses3[ 1 ], silk_shell_code_table2 ); + + decode_split( &pulses1[ 4 ], &pulses1[ 5 ], psRangeDec, pulses2[ 2 ], silk_shell_code_table1 ); + decode_split( &pulses0[ 8 ], &pulses0[ 9 ], psRangeDec, pulses1[ 4 ], silk_shell_code_table0 ); + decode_split( &pulses0[ 10 ], &pulses0[ 11 ], psRangeDec, pulses1[ 5 ], silk_shell_code_table0 ); + + decode_split( &pulses1[ 6 ], &pulses1[ 7 ], psRangeDec, pulses2[ 3 ], silk_shell_code_table1 ); + decode_split( &pulses0[ 12 ], &pulses0[ 13 ], psRangeDec, pulses1[ 6 ], silk_shell_code_table0 ); + decode_split( &pulses0[ 14 ], &pulses0[ 15 ], psRangeDec, pulses1[ 7 ], silk_shell_code_table0 ); +} diff --git a/media/libopus/silk/sigm_Q15.c b/media/libopus/silk/sigm_Q15.c new file mode 100644 index 0000000000..3c507d255b --- /dev/null +++ b/media/libopus/silk/sigm_Q15.c @@ -0,0 +1,76 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +/* Approximate sigmoid function */ + +#include "SigProc_FIX.h" + +/* fprintf(1, '%d, ', round(1024 * ([1 ./ (1 + exp(-(1:5))), 1] - 1 ./ (1 + exp(-(0:5)))))); */ +static const opus_int32 sigm_LUT_slope_Q10[ 6 ] = { + 237, 153, 73, 30, 12, 7 +}; +/* fprintf(1, '%d, ', round(32767 * 1 ./ (1 + exp(-(0:5))))); */ +static const opus_int32 sigm_LUT_pos_Q15[ 6 ] = { + 16384, 23955, 28861, 31213, 32178, 32548 +}; +/* fprintf(1, '%d, ', round(32767 * 1 ./ (1 + exp((0:5))))); */ +static const opus_int32 sigm_LUT_neg_Q15[ 6 ] = { + 16384, 8812, 3906, 1554, 589, 219 +}; + +opus_int silk_sigm_Q15( + opus_int in_Q5 /* I */ +) +{ + opus_int ind; + + if( in_Q5 < 0 ) { + /* Negative input */ + in_Q5 = -in_Q5; + if( in_Q5 >= 6 * 32 ) { + return 0; /* Clip */ + } else { + /* Linear interpolation of look up table */ + ind = silk_RSHIFT( in_Q5, 5 ); + return( sigm_LUT_neg_Q15[ ind ] - silk_SMULBB( sigm_LUT_slope_Q10[ ind ], in_Q5 & 0x1F ) ); + } + } else { + /* Positive input */ + if( in_Q5 >= 6 * 32 ) { + return 32767; /* clip */ + } else { + /* Linear interpolation of look up table */ + ind = silk_RSHIFT( in_Q5, 5 ); + return( sigm_LUT_pos_Q15[ ind ] + silk_SMULBB( sigm_LUT_slope_Q10[ ind ], in_Q5 & 0x1F ) ); + } + } +} + diff --git a/media/libopus/silk/sort.c b/media/libopus/silk/sort.c new file mode 100644 index 0000000000..4fba16f831 --- /dev/null +++ b/media/libopus/silk/sort.c @@ -0,0 +1,154 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +/* Insertion sort (fast for already almost sorted arrays): */ +/* Best case: O(n) for an already sorted array */ +/* Worst case: O(n^2) for an inversely sorted array */ +/* */ +/* Shell short: https://en.wikipedia.org/wiki/Shell_sort */ + +#include "SigProc_FIX.h" + +void silk_insertion_sort_increasing( + opus_int32 *a, /* I/O Unsorted / Sorted vector */ + opus_int *idx, /* O Index vector for the sorted elements */ + const opus_int L, /* I Vector length */ + const opus_int K /* I Number of correctly sorted positions */ +) +{ + opus_int32 value; + opus_int i, j; + + /* Safety checks */ + celt_assert( K > 0 ); + celt_assert( L > 0 ); + celt_assert( L >= K ); + + /* Write start indices in index vector */ + for( i = 0; i < K; i++ ) { + idx[ i ] = i; + } + + /* Sort vector elements by value, increasing order */ + for( i = 1; i < K; i++ ) { + value = a[ i ]; + for( j = i - 1; ( j >= 0 ) && ( value < a[ j ] ); j-- ) { + a[ j + 1 ] = a[ j ]; /* Shift value */ + idx[ j + 1 ] = idx[ j ]; /* Shift index */ + } + a[ j + 1 ] = value; /* Write value */ + idx[ j + 1 ] = i; /* Write index */ + } + + /* If less than L values are asked for, check the remaining values, */ + /* but only spend CPU to ensure that the K first values are correct */ + for( i = K; i < L; i++ ) { + value = a[ i ]; + if( value < a[ K - 1 ] ) { + for( j = K - 2; ( j >= 0 ) && ( value < a[ j ] ); j-- ) { + a[ j + 1 ] = a[ j ]; /* Shift value */ + idx[ j + 1 ] = idx[ j ]; /* Shift index */ + } + a[ j + 1 ] = value; /* Write value */ + idx[ j + 1 ] = i; /* Write index */ + } + } +} + +#ifdef FIXED_POINT +/* This function is only used by the fixed-point build */ +void silk_insertion_sort_decreasing_int16( + opus_int16 *a, /* I/O Unsorted / Sorted vector */ + opus_int *idx, /* O Index vector for the sorted elements */ + const opus_int L, /* I Vector length */ + const opus_int K /* I Number of correctly sorted positions */ +) +{ + opus_int i, j; + opus_int value; + + /* Safety checks */ + celt_assert( K > 0 ); + celt_assert( L > 0 ); + celt_assert( L >= K ); + + /* Write start indices in index vector */ + for( i = 0; i < K; i++ ) { + idx[ i ] = i; + } + + /* Sort vector elements by value, decreasing order */ + for( i = 1; i < K; i++ ) { + value = a[ i ]; + for( j = i - 1; ( j >= 0 ) && ( value > a[ j ] ); j-- ) { + a[ j + 1 ] = a[ j ]; /* Shift value */ + idx[ j + 1 ] = idx[ j ]; /* Shift index */ + } + a[ j + 1 ] = value; /* Write value */ + idx[ j + 1 ] = i; /* Write index */ + } + + /* If less than L values are asked for, check the remaining values, */ + /* but only spend CPU to ensure that the K first values are correct */ + for( i = K; i < L; i++ ) { + value = a[ i ]; + if( value > a[ K - 1 ] ) { + for( j = K - 2; ( j >= 0 ) && ( value > a[ j ] ); j-- ) { + a[ j + 1 ] = a[ j ]; /* Shift value */ + idx[ j + 1 ] = idx[ j ]; /* Shift index */ + } + a[ j + 1 ] = value; /* Write value */ + idx[ j + 1 ] = i; /* Write index */ + } + } +} +#endif + +void silk_insertion_sort_increasing_all_values_int16( + opus_int16 *a, /* I/O Unsorted / Sorted vector */ + const opus_int L /* I Vector length */ +) +{ + opus_int value; + opus_int i, j; + + /* Safety checks */ + celt_assert( L > 0 ); + + /* Sort vector elements by value, increasing order */ + for( i = 1; i < L; i++ ) { + value = a[ i ]; + for( j = i - 1; ( j >= 0 ) && ( value < a[ j ] ); j-- ) { + a[ j + 1 ] = a[ j ]; /* Shift value */ + } + a[ j + 1 ] = value; /* Write value */ + } +} diff --git a/media/libopus/silk/stereo_LR_to_MS.c b/media/libopus/silk/stereo_LR_to_MS.c new file mode 100644 index 0000000000..751452cb19 --- /dev/null +++ b/media/libopus/silk/stereo_LR_to_MS.c @@ -0,0 +1,229 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" +#include "stack_alloc.h" + +/* Convert Left/Right stereo signal to adaptive Mid/Side representation */ +void silk_stereo_LR_to_MS( + stereo_enc_state *state, /* I/O State */ + opus_int16 x1[], /* I/O Left input signal, becomes mid signal */ + opus_int16 x2[], /* I/O Right input signal, becomes side signal */ + opus_int8 ix[ 2 ][ 3 ], /* O Quantization indices */ + opus_int8 *mid_only_flag, /* O Flag: only mid signal coded */ + opus_int32 mid_side_rates_bps[], /* O Bitrates for mid and side signals */ + opus_int32 total_rate_bps, /* I Total bitrate */ + opus_int prev_speech_act_Q8, /* I Speech activity level in previous frame */ + opus_int toMono, /* I Last frame before a stereo->mono transition */ + opus_int fs_kHz, /* I Sample rate (kHz) */ + opus_int frame_length /* I Number of samples */ +) +{ + opus_int n, is10msFrame, denom_Q16, delta0_Q13, delta1_Q13; + opus_int32 sum, diff, smooth_coef_Q16, pred_Q13[ 2 ], pred0_Q13, pred1_Q13; + opus_int32 LP_ratio_Q14, HP_ratio_Q14, frac_Q16, frac_3_Q16, min_mid_rate_bps, width_Q14, w_Q24, deltaw_Q24; + VARDECL( opus_int16, side ); + VARDECL( opus_int16, LP_mid ); + VARDECL( opus_int16, HP_mid ); + VARDECL( opus_int16, LP_side ); + VARDECL( opus_int16, HP_side ); + opus_int16 *mid = &x1[ -2 ]; + SAVE_STACK; + + ALLOC( side, frame_length + 2, opus_int16 ); + /* Convert to basic mid/side signals */ + for( n = 0; n < frame_length + 2; n++ ) { + sum = x1[ n - 2 ] + (opus_int32)x2[ n - 2 ]; + diff = x1[ n - 2 ] - (opus_int32)x2[ n - 2 ]; + mid[ n ] = (opus_int16)silk_RSHIFT_ROUND( sum, 1 ); + side[ n ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( diff, 1 ) ); + } + + /* Buffering */ + silk_memcpy( mid, state->sMid, 2 * sizeof( opus_int16 ) ); + silk_memcpy( side, state->sSide, 2 * sizeof( opus_int16 ) ); + silk_memcpy( state->sMid, &mid[ frame_length ], 2 * sizeof( opus_int16 ) ); + silk_memcpy( state->sSide, &side[ frame_length ], 2 * sizeof( opus_int16 ) ); + + /* LP and HP filter mid signal */ + ALLOC( LP_mid, frame_length, opus_int16 ); + ALLOC( HP_mid, frame_length, opus_int16 ); + for( n = 0; n < frame_length; n++ ) { + sum = silk_RSHIFT_ROUND( silk_ADD_LSHIFT32( mid[ n ] + (opus_int32)mid[ n + 2 ], mid[ n + 1 ], 1 ), 2 ); + LP_mid[ n ] = sum; + HP_mid[ n ] = mid[ n + 1 ] - sum; + } + + /* LP and HP filter side signal */ + ALLOC( LP_side, frame_length, opus_int16 ); + ALLOC( HP_side, frame_length, opus_int16 ); + for( n = 0; n < frame_length; n++ ) { + sum = silk_RSHIFT_ROUND( silk_ADD_LSHIFT32( side[ n ] + (opus_int32)side[ n + 2 ], side[ n + 1 ], 1 ), 2 ); + LP_side[ n ] = sum; + HP_side[ n ] = side[ n + 1 ] - sum; + } + + /* Find energies and predictors */ + is10msFrame = frame_length == 10 * fs_kHz; + smooth_coef_Q16 = is10msFrame ? + SILK_FIX_CONST( STEREO_RATIO_SMOOTH_COEF / 2, 16 ) : + SILK_FIX_CONST( STEREO_RATIO_SMOOTH_COEF, 16 ); + smooth_coef_Q16 = silk_SMULWB( silk_SMULBB( prev_speech_act_Q8, prev_speech_act_Q8 ), smooth_coef_Q16 ); + + pred_Q13[ 0 ] = silk_stereo_find_predictor( &LP_ratio_Q14, LP_mid, LP_side, &state->mid_side_amp_Q0[ 0 ], frame_length, smooth_coef_Q16 ); + pred_Q13[ 1 ] = silk_stereo_find_predictor( &HP_ratio_Q14, HP_mid, HP_side, &state->mid_side_amp_Q0[ 2 ], frame_length, smooth_coef_Q16 ); + /* Ratio of the norms of residual and mid signals */ + frac_Q16 = silk_SMLABB( HP_ratio_Q14, LP_ratio_Q14, 3 ); + frac_Q16 = silk_min( frac_Q16, SILK_FIX_CONST( 1, 16 ) ); + + /* Determine bitrate distribution between mid and side, and possibly reduce stereo width */ + total_rate_bps -= is10msFrame ? 1200 : 600; /* Subtract approximate bitrate for coding stereo parameters */ + if( total_rate_bps < 1 ) { + total_rate_bps = 1; + } + min_mid_rate_bps = silk_SMLABB( 2000, fs_kHz, 600 ); + silk_assert( min_mid_rate_bps < 32767 ); + /* Default bitrate distribution: 8 parts for Mid and (5+3*frac) parts for Side. so: mid_rate = ( 8 / ( 13 + 3 * frac ) ) * total_ rate */ + frac_3_Q16 = silk_MUL( 3, frac_Q16 ); + mid_side_rates_bps[ 0 ] = silk_DIV32_varQ( total_rate_bps, SILK_FIX_CONST( 8 + 5, 16 ) + frac_3_Q16, 16+3 ); + /* If Mid bitrate below minimum, reduce stereo width */ + if( mid_side_rates_bps[ 0 ] < min_mid_rate_bps ) { + mid_side_rates_bps[ 0 ] = min_mid_rate_bps; + mid_side_rates_bps[ 1 ] = total_rate_bps - mid_side_rates_bps[ 0 ]; + /* width = 4 * ( 2 * side_rate - min_rate ) / ( ( 1 + 3 * frac ) * min_rate ) */ + width_Q14 = silk_DIV32_varQ( silk_LSHIFT( mid_side_rates_bps[ 1 ], 1 ) - min_mid_rate_bps, + silk_SMULWB( SILK_FIX_CONST( 1, 16 ) + frac_3_Q16, min_mid_rate_bps ), 14+2 ); + width_Q14 = silk_LIMIT( width_Q14, 0, SILK_FIX_CONST( 1, 14 ) ); + } else { + mid_side_rates_bps[ 1 ] = total_rate_bps - mid_side_rates_bps[ 0 ]; + width_Q14 = SILK_FIX_CONST( 1, 14 ); + } + + /* Smoother */ + state->smth_width_Q14 = (opus_int16)silk_SMLAWB( state->smth_width_Q14, width_Q14 - state->smth_width_Q14, smooth_coef_Q16 ); + + /* At very low bitrates or for inputs that are nearly amplitude panned, switch to panned-mono coding */ + *mid_only_flag = 0; + if( toMono ) { + /* Last frame before stereo->mono transition; collapse stereo width */ + width_Q14 = 0; + pred_Q13[ 0 ] = 0; + pred_Q13[ 1 ] = 0; + silk_stereo_quant_pred( pred_Q13, ix ); + } else if( state->width_prev_Q14 == 0 && + ( 8 * total_rate_bps < 13 * min_mid_rate_bps || silk_SMULWB( frac_Q16, state->smth_width_Q14 ) < SILK_FIX_CONST( 0.05, 14 ) ) ) + { + /* Code as panned-mono; previous frame already had zero width */ + /* Scale down and quantize predictors */ + pred_Q13[ 0 ] = silk_RSHIFT( silk_SMULBB( state->smth_width_Q14, pred_Q13[ 0 ] ), 14 ); + pred_Q13[ 1 ] = silk_RSHIFT( silk_SMULBB( state->smth_width_Q14, pred_Q13[ 1 ] ), 14 ); + silk_stereo_quant_pred( pred_Q13, ix ); + /* Collapse stereo width */ + width_Q14 = 0; + pred_Q13[ 0 ] = 0; + pred_Q13[ 1 ] = 0; + mid_side_rates_bps[ 0 ] = total_rate_bps; + mid_side_rates_bps[ 1 ] = 0; + *mid_only_flag = 1; + } else if( state->width_prev_Q14 != 0 && + ( 8 * total_rate_bps < 11 * min_mid_rate_bps || silk_SMULWB( frac_Q16, state->smth_width_Q14 ) < SILK_FIX_CONST( 0.02, 14 ) ) ) + { + /* Transition to zero-width stereo */ + /* Scale down and quantize predictors */ + pred_Q13[ 0 ] = silk_RSHIFT( silk_SMULBB( state->smth_width_Q14, pred_Q13[ 0 ] ), 14 ); + pred_Q13[ 1 ] = silk_RSHIFT( silk_SMULBB( state->smth_width_Q14, pred_Q13[ 1 ] ), 14 ); + silk_stereo_quant_pred( pred_Q13, ix ); + /* Collapse stereo width */ + width_Q14 = 0; + pred_Q13[ 0 ] = 0; + pred_Q13[ 1 ] = 0; + } else if( state->smth_width_Q14 > SILK_FIX_CONST( 0.95, 14 ) ) { + /* Full-width stereo coding */ + silk_stereo_quant_pred( pred_Q13, ix ); + width_Q14 = SILK_FIX_CONST( 1, 14 ); + } else { + /* Reduced-width stereo coding; scale down and quantize predictors */ + pred_Q13[ 0 ] = silk_RSHIFT( silk_SMULBB( state->smth_width_Q14, pred_Q13[ 0 ] ), 14 ); + pred_Q13[ 1 ] = silk_RSHIFT( silk_SMULBB( state->smth_width_Q14, pred_Q13[ 1 ] ), 14 ); + silk_stereo_quant_pred( pred_Q13, ix ); + width_Q14 = state->smth_width_Q14; + } + + /* Make sure to keep on encoding until the tapered output has been transmitted */ + if( *mid_only_flag == 1 ) { + state->silent_side_len += frame_length - STEREO_INTERP_LEN_MS * fs_kHz; + if( state->silent_side_len < LA_SHAPE_MS * fs_kHz ) { + *mid_only_flag = 0; + } else { + /* Limit to avoid wrapping around */ + state->silent_side_len = 10000; + } + } else { + state->silent_side_len = 0; + } + + if( *mid_only_flag == 0 && mid_side_rates_bps[ 1 ] < 1 ) { + mid_side_rates_bps[ 1 ] = 1; + mid_side_rates_bps[ 0 ] = silk_max_int( 1, total_rate_bps - mid_side_rates_bps[ 1 ]); + } + + /* Interpolate predictors and subtract prediction from side channel */ + pred0_Q13 = -state->pred_prev_Q13[ 0 ]; + pred1_Q13 = -state->pred_prev_Q13[ 1 ]; + w_Q24 = silk_LSHIFT( state->width_prev_Q14, 10 ); + denom_Q16 = silk_DIV32_16( (opus_int32)1 << 16, STEREO_INTERP_LEN_MS * fs_kHz ); + delta0_Q13 = -silk_RSHIFT_ROUND( silk_SMULBB( pred_Q13[ 0 ] - state->pred_prev_Q13[ 0 ], denom_Q16 ), 16 ); + delta1_Q13 = -silk_RSHIFT_ROUND( silk_SMULBB( pred_Q13[ 1 ] - state->pred_prev_Q13[ 1 ], denom_Q16 ), 16 ); + deltaw_Q24 = silk_LSHIFT( silk_SMULWB( width_Q14 - state->width_prev_Q14, denom_Q16 ), 10 ); + for( n = 0; n < STEREO_INTERP_LEN_MS * fs_kHz; n++ ) { + pred0_Q13 += delta0_Q13; + pred1_Q13 += delta1_Q13; + w_Q24 += deltaw_Q24; + sum = silk_LSHIFT( silk_ADD_LSHIFT32( mid[ n ] + (opus_int32)mid[ n + 2 ], mid[ n + 1 ], 1 ), 9 ); /* Q11 */ + sum = silk_SMLAWB( silk_SMULWB( w_Q24, side[ n + 1 ] ), sum, pred0_Q13 ); /* Q8 */ + sum = silk_SMLAWB( sum, silk_LSHIFT( (opus_int32)mid[ n + 1 ], 11 ), pred1_Q13 ); /* Q8 */ + x2[ n - 1 ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( sum, 8 ) ); + } + + pred0_Q13 = -pred_Q13[ 0 ]; + pred1_Q13 = -pred_Q13[ 1 ]; + w_Q24 = silk_LSHIFT( width_Q14, 10 ); + for( n = STEREO_INTERP_LEN_MS * fs_kHz; n < frame_length; n++ ) { + sum = silk_LSHIFT( silk_ADD_LSHIFT32( mid[ n ] + (opus_int32)mid[ n + 2 ], mid[ n + 1 ], 1 ), 9 ); /* Q11 */ + sum = silk_SMLAWB( silk_SMULWB( w_Q24, side[ n + 1 ] ), sum, pred0_Q13 ); /* Q8 */ + sum = silk_SMLAWB( sum, silk_LSHIFT( (opus_int32)mid[ n + 1 ], 11 ), pred1_Q13 ); /* Q8 */ + x2[ n - 1 ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( sum, 8 ) ); + } + state->pred_prev_Q13[ 0 ] = (opus_int16)pred_Q13[ 0 ]; + state->pred_prev_Q13[ 1 ] = (opus_int16)pred_Q13[ 1 ]; + state->width_prev_Q14 = (opus_int16)width_Q14; + RESTORE_STACK; +} diff --git a/media/libopus/silk/stereo_MS_to_LR.c b/media/libopus/silk/stereo_MS_to_LR.c new file mode 100644 index 0000000000..1e01bb6eb3 --- /dev/null +++ b/media/libopus/silk/stereo_MS_to_LR.c @@ -0,0 +1,85 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" + +/* Convert adaptive Mid/Side representation to Left/Right stereo signal */ +void silk_stereo_MS_to_LR( + stereo_dec_state *state, /* I/O State */ + opus_int16 x1[], /* I/O Left input signal, becomes mid signal */ + opus_int16 x2[], /* I/O Right input signal, becomes side signal */ + const opus_int32 pred_Q13[], /* I Predictors */ + opus_int fs_kHz, /* I Samples rate (kHz) */ + opus_int frame_length /* I Number of samples */ +) +{ + opus_int n, denom_Q16, delta0_Q13, delta1_Q13; + opus_int32 sum, diff, pred0_Q13, pred1_Q13; + + /* Buffering */ + silk_memcpy( x1, state->sMid, 2 * sizeof( opus_int16 ) ); + silk_memcpy( x2, state->sSide, 2 * sizeof( opus_int16 ) ); + silk_memcpy( state->sMid, &x1[ frame_length ], 2 * sizeof( opus_int16 ) ); + silk_memcpy( state->sSide, &x2[ frame_length ], 2 * sizeof( opus_int16 ) ); + + /* Interpolate predictors and add prediction to side channel */ + pred0_Q13 = state->pred_prev_Q13[ 0 ]; + pred1_Q13 = state->pred_prev_Q13[ 1 ]; + denom_Q16 = silk_DIV32_16( (opus_int32)1 << 16, STEREO_INTERP_LEN_MS * fs_kHz ); + delta0_Q13 = silk_RSHIFT_ROUND( silk_SMULBB( pred_Q13[ 0 ] - state->pred_prev_Q13[ 0 ], denom_Q16 ), 16 ); + delta1_Q13 = silk_RSHIFT_ROUND( silk_SMULBB( pred_Q13[ 1 ] - state->pred_prev_Q13[ 1 ], denom_Q16 ), 16 ); + for( n = 0; n < STEREO_INTERP_LEN_MS * fs_kHz; n++ ) { + pred0_Q13 += delta0_Q13; + pred1_Q13 += delta1_Q13; + sum = silk_LSHIFT( silk_ADD_LSHIFT32( x1[ n ] + (opus_int32)x1[ n + 2 ], x1[ n + 1 ], 1 ), 9 ); /* Q11 */ + sum = silk_SMLAWB( silk_LSHIFT( (opus_int32)x2[ n + 1 ], 8 ), sum, pred0_Q13 ); /* Q8 */ + sum = silk_SMLAWB( sum, silk_LSHIFT( (opus_int32)x1[ n + 1 ], 11 ), pred1_Q13 ); /* Q8 */ + x2[ n + 1 ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( sum, 8 ) ); + } + pred0_Q13 = pred_Q13[ 0 ]; + pred1_Q13 = pred_Q13[ 1 ]; + for( n = STEREO_INTERP_LEN_MS * fs_kHz; n < frame_length; n++ ) { + sum = silk_LSHIFT( silk_ADD_LSHIFT32( x1[ n ] + (opus_int32)x1[ n + 2 ], x1[ n + 1 ], 1 ), 9 ); /* Q11 */ + sum = silk_SMLAWB( silk_LSHIFT( (opus_int32)x2[ n + 1 ], 8 ), sum, pred0_Q13 ); /* Q8 */ + sum = silk_SMLAWB( sum, silk_LSHIFT( (opus_int32)x1[ n + 1 ], 11 ), pred1_Q13 ); /* Q8 */ + x2[ n + 1 ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( sum, 8 ) ); + } + state->pred_prev_Q13[ 0 ] = pred_Q13[ 0 ]; + state->pred_prev_Q13[ 1 ] = pred_Q13[ 1 ]; + + /* Convert to left/right signals */ + for( n = 0; n < frame_length; n++ ) { + sum = x1[ n + 1 ] + (opus_int32)x2[ n + 1 ]; + diff = x1[ n + 1 ] - (opus_int32)x2[ n + 1 ]; + x1[ n + 1 ] = (opus_int16)silk_SAT16( sum ); + x2[ n + 1 ] = (opus_int16)silk_SAT16( diff ); + } +} diff --git a/media/libopus/silk/stereo_decode_pred.c b/media/libopus/silk/stereo_decode_pred.c new file mode 100644 index 0000000000..56ba3925e8 --- /dev/null +++ b/media/libopus/silk/stereo_decode_pred.c @@ -0,0 +1,73 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" + +/* Decode mid/side predictors */ +void silk_stereo_decode_pred( + ec_dec *psRangeDec, /* I/O Compressor data structure */ + opus_int32 pred_Q13[] /* O Predictors */ +) +{ + opus_int n, ix[ 2 ][ 3 ]; + opus_int32 low_Q13, step_Q13; + + /* Entropy decoding */ + n = ec_dec_icdf( psRangeDec, silk_stereo_pred_joint_iCDF, 8 ); + ix[ 0 ][ 2 ] = silk_DIV32_16( n, 5 ); + ix[ 1 ][ 2 ] = n - 5 * ix[ 0 ][ 2 ]; + for( n = 0; n < 2; n++ ) { + ix[ n ][ 0 ] = ec_dec_icdf( psRangeDec, silk_uniform3_iCDF, 8 ); + ix[ n ][ 1 ] = ec_dec_icdf( psRangeDec, silk_uniform5_iCDF, 8 ); + } + + /* Dequantize */ + for( n = 0; n < 2; n++ ) { + ix[ n ][ 0 ] += 3 * ix[ n ][ 2 ]; + low_Q13 = silk_stereo_pred_quant_Q13[ ix[ n ][ 0 ] ]; + step_Q13 = silk_SMULWB( silk_stereo_pred_quant_Q13[ ix[ n ][ 0 ] + 1 ] - low_Q13, + SILK_FIX_CONST( 0.5 / STEREO_QUANT_SUB_STEPS, 16 ) ); + pred_Q13[ n ] = silk_SMLABB( low_Q13, step_Q13, 2 * ix[ n ][ 1 ] + 1 ); + } + + /* Subtract second from first predictor (helps when actually applying these) */ + pred_Q13[ 0 ] -= pred_Q13[ 1 ]; +} + +/* Decode mid-only flag */ +void silk_stereo_decode_mid_only( + ec_dec *psRangeDec, /* I/O Compressor data structure */ + opus_int *decode_only_mid /* O Flag that only mid channel has been coded */ +) +{ + /* Decode flag that only mid channel is coded */ + *decode_only_mid = ec_dec_icdf( psRangeDec, silk_stereo_only_code_mid_iCDF, 8 ); +} diff --git a/media/libopus/silk/stereo_encode_pred.c b/media/libopus/silk/stereo_encode_pred.c new file mode 100644 index 0000000000..03becb6736 --- /dev/null +++ b/media/libopus/silk/stereo_encode_pred.c @@ -0,0 +1,62 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" + +/* Entropy code the mid/side quantization indices */ +void silk_stereo_encode_pred( + ec_enc *psRangeEnc, /* I/O Compressor data structure */ + opus_int8 ix[ 2 ][ 3 ] /* I Quantization indices */ +) +{ + opus_int n; + + /* Entropy coding */ + n = 5 * ix[ 0 ][ 2 ] + ix[ 1 ][ 2 ]; + celt_assert( n < 25 ); + ec_enc_icdf( psRangeEnc, n, silk_stereo_pred_joint_iCDF, 8 ); + for( n = 0; n < 2; n++ ) { + celt_assert( ix[ n ][ 0 ] < 3 ); + celt_assert( ix[ n ][ 1 ] < STEREO_QUANT_SUB_STEPS ); + ec_enc_icdf( psRangeEnc, ix[ n ][ 0 ], silk_uniform3_iCDF, 8 ); + ec_enc_icdf( psRangeEnc, ix[ n ][ 1 ], silk_uniform5_iCDF, 8 ); + } +} + +/* Entropy code the mid-only flag */ +void silk_stereo_encode_mid_only( + ec_enc *psRangeEnc, /* I/O Compressor data structure */ + opus_int8 mid_only_flag +) +{ + /* Encode flag that only mid channel is coded */ + ec_enc_icdf( psRangeEnc, mid_only_flag, silk_stereo_only_code_mid_iCDF, 8 ); +} diff --git a/media/libopus/silk/stereo_find_predictor.c b/media/libopus/silk/stereo_find_predictor.c new file mode 100644 index 0000000000..e30e90bddc --- /dev/null +++ b/media/libopus/silk/stereo_find_predictor.c @@ -0,0 +1,79 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" + +/* Find least-squares prediction gain for one signal based on another and quantize it */ +opus_int32 silk_stereo_find_predictor( /* O Returns predictor in Q13 */ + opus_int32 *ratio_Q14, /* O Ratio of residual and mid energies */ + const opus_int16 x[], /* I Basis signal */ + const opus_int16 y[], /* I Target signal */ + opus_int32 mid_res_amp_Q0[], /* I/O Smoothed mid, residual norms */ + opus_int length, /* I Number of samples */ + opus_int smooth_coef_Q16 /* I Smoothing coefficient */ +) +{ + opus_int scale, scale1, scale2; + opus_int32 nrgx, nrgy, corr, pred_Q13, pred2_Q10; + + /* Find predictor */ + silk_sum_sqr_shift( &nrgx, &scale1, x, length ); + silk_sum_sqr_shift( &nrgy, &scale2, y, length ); + scale = silk_max_int( scale1, scale2 ); + scale = scale + ( scale & 1 ); /* make even */ + nrgy = silk_RSHIFT32( nrgy, scale - scale2 ); + nrgx = silk_RSHIFT32( nrgx, scale - scale1 ); + nrgx = silk_max_int( nrgx, 1 ); + corr = silk_inner_prod_aligned_scale( x, y, scale, length ); + pred_Q13 = silk_DIV32_varQ( corr, nrgx, 13 ); + pred_Q13 = silk_LIMIT( pred_Q13, -(1 << 14), 1 << 14 ); + pred2_Q10 = silk_SMULWB( pred_Q13, pred_Q13 ); + + /* Faster update for signals with large prediction parameters */ + smooth_coef_Q16 = (opus_int)silk_max_int( smooth_coef_Q16, silk_abs( pred2_Q10 ) ); + + /* Smoothed mid and residual norms */ + silk_assert( smooth_coef_Q16 < 32768 ); + scale = silk_RSHIFT( scale, 1 ); + mid_res_amp_Q0[ 0 ] = silk_SMLAWB( mid_res_amp_Q0[ 0 ], silk_LSHIFT( silk_SQRT_APPROX( nrgx ), scale ) - mid_res_amp_Q0[ 0 ], + smooth_coef_Q16 ); + /* Residual energy = nrgy - 2 * pred * corr + pred^2 * nrgx */ + nrgy = silk_SUB_LSHIFT32( nrgy, silk_SMULWB( corr, pred_Q13 ), 3 + 1 ); + nrgy = silk_ADD_LSHIFT32( nrgy, silk_SMULWB( nrgx, pred2_Q10 ), 6 ); + mid_res_amp_Q0[ 1 ] = silk_SMLAWB( mid_res_amp_Q0[ 1 ], silk_LSHIFT( silk_SQRT_APPROX( nrgy ), scale ) - mid_res_amp_Q0[ 1 ], + smooth_coef_Q16 ); + + /* Ratio of smoothed residual and mid norms */ + *ratio_Q14 = silk_DIV32_varQ( mid_res_amp_Q0[ 1 ], silk_max( mid_res_amp_Q0[ 0 ], 1 ), 14 ); + *ratio_Q14 = silk_LIMIT( *ratio_Q14, 0, 32767 ); + + return pred_Q13; +} diff --git a/media/libopus/silk/stereo_quant_pred.c b/media/libopus/silk/stereo_quant_pred.c new file mode 100644 index 0000000000..d4ced6c3e8 --- /dev/null +++ b/media/libopus/silk/stereo_quant_pred.c @@ -0,0 +1,73 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "main.h" + +/* Quantize mid/side predictors */ +void silk_stereo_quant_pred( + opus_int32 pred_Q13[], /* I/O Predictors (out: quantized) */ + opus_int8 ix[ 2 ][ 3 ] /* O Quantization indices */ +) +{ + opus_int i, j, n; + opus_int32 low_Q13, step_Q13, lvl_Q13, err_min_Q13, err_Q13, quant_pred_Q13 = 0; + + /* Quantize */ + for( n = 0; n < 2; n++ ) { + /* Brute-force search over quantization levels */ + err_min_Q13 = silk_int32_MAX; + for( i = 0; i < STEREO_QUANT_TAB_SIZE - 1; i++ ) { + low_Q13 = silk_stereo_pred_quant_Q13[ i ]; + step_Q13 = silk_SMULWB( silk_stereo_pred_quant_Q13[ i + 1 ] - low_Q13, + SILK_FIX_CONST( 0.5 / STEREO_QUANT_SUB_STEPS, 16 ) ); + for( j = 0; j < STEREO_QUANT_SUB_STEPS; j++ ) { + lvl_Q13 = silk_SMLABB( low_Q13, step_Q13, 2 * j + 1 ); + err_Q13 = silk_abs( pred_Q13[ n ] - lvl_Q13 ); + if( err_Q13 < err_min_Q13 ) { + err_min_Q13 = err_Q13; + quant_pred_Q13 = lvl_Q13; + ix[ n ][ 0 ] = i; + ix[ n ][ 1 ] = j; + } else { + /* Error increasing, so we're past the optimum */ + goto done; + } + } + } + done: + ix[ n ][ 2 ] = silk_DIV32_16( ix[ n ][ 0 ], 3 ); + ix[ n ][ 0 ] -= ix[ n ][ 2 ] * 3; + pred_Q13[ n ] = quant_pred_Q13; + } + + /* Subtract second from first predictor (helps when actually applying these) */ + pred_Q13[ 0 ] -= pred_Q13[ 1 ]; +} diff --git a/media/libopus/silk/structs.h b/media/libopus/silk/structs.h new file mode 100644 index 0000000000..3380c757b2 --- /dev/null +++ b/media/libopus/silk/structs.h @@ -0,0 +1,329 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef SILK_STRUCTS_H +#define SILK_STRUCTS_H + +#include "typedef.h" +#include "SigProc_FIX.h" +#include "define.h" +#include "entenc.h" +#include "entdec.h" + +#ifdef __cplusplus +extern "C" +{ +#endif + +/************************************/ +/* Noise shaping quantization state */ +/************************************/ +typedef struct { + opus_int16 xq[ 2 * MAX_FRAME_LENGTH ]; /* Buffer for quantized output signal */ + opus_int32 sLTP_shp_Q14[ 2 * MAX_FRAME_LENGTH ]; + opus_int32 sLPC_Q14[ MAX_SUB_FRAME_LENGTH + NSQ_LPC_BUF_LENGTH ]; + opus_int32 sAR2_Q14[ MAX_SHAPE_LPC_ORDER ]; + opus_int32 sLF_AR_shp_Q14; + opus_int32 sDiff_shp_Q14; + opus_int lagPrev; + opus_int sLTP_buf_idx; + opus_int sLTP_shp_buf_idx; + opus_int32 rand_seed; + opus_int32 prev_gain_Q16; + opus_int rewhite_flag; +} silk_nsq_state; + +/********************************/ +/* VAD state */ +/********************************/ +typedef struct { + opus_int32 AnaState[ 2 ]; /* Analysis filterbank state: 0-8 kHz */ + opus_int32 AnaState1[ 2 ]; /* Analysis filterbank state: 0-4 kHz */ + opus_int32 AnaState2[ 2 ]; /* Analysis filterbank state: 0-2 kHz */ + opus_int32 XnrgSubfr[ VAD_N_BANDS ]; /* Subframe energies */ + opus_int32 NrgRatioSmth_Q8[ VAD_N_BANDS ]; /* Smoothed energy level in each band */ + opus_int16 HPstate; /* State of differentiator in the lowest band */ + opus_int32 NL[ VAD_N_BANDS ]; /* Noise energy level in each band */ + opus_int32 inv_NL[ VAD_N_BANDS ]; /* Inverse noise energy level in each band */ + opus_int32 NoiseLevelBias[ VAD_N_BANDS ]; /* Noise level estimator bias/offset */ + opus_int32 counter; /* Frame counter used in the initial phase */ +} silk_VAD_state; + +/* Variable cut-off low-pass filter state */ +typedef struct { + opus_int32 In_LP_State[ 2 ]; /* Low pass filter state */ + opus_int32 transition_frame_no; /* Counter which is mapped to a cut-off frequency */ + opus_int mode; /* Operating mode, <0: switch down, >0: switch up; 0: do nothing */ + opus_int32 saved_fs_kHz; /* If non-zero, holds the last sampling rate before a bandwidth switching reset. */ +} silk_LP_state; + +/* Structure containing NLSF codebook */ +typedef struct { + const opus_int16 nVectors; + const opus_int16 order; + const opus_int16 quantStepSize_Q16; + const opus_int16 invQuantStepSize_Q6; + const opus_uint8 *CB1_NLSF_Q8; + const opus_int16 *CB1_Wght_Q9; + const opus_uint8 *CB1_iCDF; + const opus_uint8 *pred_Q8; + const opus_uint8 *ec_sel; + const opus_uint8 *ec_iCDF; + const opus_uint8 *ec_Rates_Q5; + const opus_int16 *deltaMin_Q15; +} silk_NLSF_CB_struct; + +typedef struct { + opus_int16 pred_prev_Q13[ 2 ]; + opus_int16 sMid[ 2 ]; + opus_int16 sSide[ 2 ]; + opus_int32 mid_side_amp_Q0[ 4 ]; + opus_int16 smth_width_Q14; + opus_int16 width_prev_Q14; + opus_int16 silent_side_len; + opus_int8 predIx[ MAX_FRAMES_PER_PACKET ][ 2 ][ 3 ]; + opus_int8 mid_only_flags[ MAX_FRAMES_PER_PACKET ]; +} stereo_enc_state; + +typedef struct { + opus_int16 pred_prev_Q13[ 2 ]; + opus_int16 sMid[ 2 ]; + opus_int16 sSide[ 2 ]; +} stereo_dec_state; + +typedef struct { + opus_int8 GainsIndices[ MAX_NB_SUBFR ]; + opus_int8 LTPIndex[ MAX_NB_SUBFR ]; + opus_int8 NLSFIndices[ MAX_LPC_ORDER + 1 ]; + opus_int16 lagIndex; + opus_int8 contourIndex; + opus_int8 signalType; + opus_int8 quantOffsetType; + opus_int8 NLSFInterpCoef_Q2; + opus_int8 PERIndex; + opus_int8 LTP_scaleIndex; + opus_int8 Seed; +} SideInfoIndices; + +/********************************/ +/* Encoder state */ +/********************************/ +typedef struct { + opus_int32 In_HP_State[ 2 ]; /* High pass filter state */ + opus_int32 variable_HP_smth1_Q15; /* State of first smoother */ + opus_int32 variable_HP_smth2_Q15; /* State of second smoother */ + silk_LP_state sLP; /* Low pass filter state */ + silk_VAD_state sVAD; /* Voice activity detector state */ + silk_nsq_state sNSQ; /* Noise Shape Quantizer State */ + opus_int16 prev_NLSFq_Q15[ MAX_LPC_ORDER ]; /* Previously quantized NLSF vector */ + opus_int speech_activity_Q8; /* Speech activity */ + opus_int allow_bandwidth_switch; /* Flag indicating that switching of internal bandwidth is allowed */ + opus_int8 LBRRprevLastGainIndex; + opus_int8 prevSignalType; + opus_int prevLag; + opus_int pitch_LPC_win_length; + opus_int max_pitch_lag; /* Highest possible pitch lag (samples) */ + opus_int32 API_fs_Hz; /* API sampling frequency (Hz) */ + opus_int32 prev_API_fs_Hz; /* Previous API sampling frequency (Hz) */ + opus_int maxInternal_fs_Hz; /* Maximum internal sampling frequency (Hz) */ + opus_int minInternal_fs_Hz; /* Minimum internal sampling frequency (Hz) */ + opus_int desiredInternal_fs_Hz; /* Soft request for internal sampling frequency (Hz) */ + opus_int fs_kHz; /* Internal sampling frequency (kHz) */ + opus_int nb_subfr; /* Number of 5 ms subframes in a frame */ + opus_int frame_length; /* Frame length (samples) */ + opus_int subfr_length; /* Subframe length (samples) */ + opus_int ltp_mem_length; /* Length of LTP memory */ + opus_int la_pitch; /* Look-ahead for pitch analysis (samples) */ + opus_int la_shape; /* Look-ahead for noise shape analysis (samples) */ + opus_int shapeWinLength; /* Window length for noise shape analysis (samples) */ + opus_int32 TargetRate_bps; /* Target bitrate (bps) */ + opus_int PacketSize_ms; /* Number of milliseconds to put in each packet */ + opus_int PacketLoss_perc; /* Packet loss rate measured by farend */ + opus_int32 frameCounter; + opus_int Complexity; /* Complexity setting */ + opus_int nStatesDelayedDecision; /* Number of states in delayed decision quantization */ + opus_int useInterpolatedNLSFs; /* Flag for using NLSF interpolation */ + opus_int shapingLPCOrder; /* Filter order for noise shaping filters */ + opus_int predictLPCOrder; /* Filter order for prediction filters */ + opus_int pitchEstimationComplexity; /* Complexity level for pitch estimator */ + opus_int pitchEstimationLPCOrder; /* Whitening filter order for pitch estimator */ + opus_int32 pitchEstimationThreshold_Q16; /* Threshold for pitch estimator */ + opus_int32 sum_log_gain_Q7; /* Cumulative max prediction gain */ + opus_int NLSF_MSVQ_Survivors; /* Number of survivors in NLSF MSVQ */ + opus_int first_frame_after_reset; /* Flag for deactivating NLSF interpolation, pitch prediction */ + opus_int controlled_since_last_payload; /* Flag for ensuring codec_control only runs once per packet */ + opus_int warping_Q16; /* Warping parameter for warped noise shaping */ + opus_int useCBR; /* Flag to enable constant bitrate */ + opus_int prefillFlag; /* Flag to indicate that only buffers are prefilled, no coding */ + const opus_uint8 *pitch_lag_low_bits_iCDF; /* Pointer to iCDF table for low bits of pitch lag index */ + const opus_uint8 *pitch_contour_iCDF; /* Pointer to iCDF table for pitch contour index */ + const silk_NLSF_CB_struct *psNLSF_CB; /* Pointer to NLSF codebook */ + opus_int input_quality_bands_Q15[ VAD_N_BANDS ]; + opus_int input_tilt_Q15; + opus_int SNR_dB_Q7; /* Quality setting */ + + opus_int8 VAD_flags[ MAX_FRAMES_PER_PACKET ]; + opus_int8 LBRR_flag; + opus_int LBRR_flags[ MAX_FRAMES_PER_PACKET ]; + + SideInfoIndices indices; + opus_int8 pulses[ MAX_FRAME_LENGTH ]; + + int arch; + + /* Input/output buffering */ + opus_int16 inputBuf[ MAX_FRAME_LENGTH + 2 ]; /* Buffer containing input signal */ + opus_int inputBufIx; + opus_int nFramesPerPacket; + opus_int nFramesEncoded; /* Number of frames analyzed in current packet */ + + opus_int nChannelsAPI; + opus_int nChannelsInternal; + opus_int channelNb; + + /* Parameters For LTP scaling Control */ + opus_int frames_since_onset; + + /* Specifically for entropy coding */ + opus_int ec_prevSignalType; + opus_int16 ec_prevLagIndex; + + silk_resampler_state_struct resampler_state; + + /* DTX */ + opus_int useDTX; /* Flag to enable DTX */ + opus_int inDTX; /* Flag to signal DTX period */ + opus_int noSpeechCounter; /* Counts concecutive nonactive frames, used by DTX */ + + /* Inband Low Bitrate Redundancy (LBRR) data */ + opus_int useInBandFEC; /* Saves the API setting for query */ + opus_int LBRR_enabled; /* Depends on useInBandFRC, bitrate and packet loss rate */ + opus_int LBRR_GainIncreases; /* Gains increment for coding LBRR frames */ + SideInfoIndices indices_LBRR[ MAX_FRAMES_PER_PACKET ]; + opus_int8 pulses_LBRR[ MAX_FRAMES_PER_PACKET ][ MAX_FRAME_LENGTH ]; +} silk_encoder_state; + + +/* Struct for Packet Loss Concealment */ +typedef struct { + opus_int32 pitchL_Q8; /* Pitch lag to use for voiced concealment */ + opus_int16 LTPCoef_Q14[ LTP_ORDER ]; /* LTP coeficients to use for voiced concealment */ + opus_int16 prevLPC_Q12[ MAX_LPC_ORDER ]; + opus_int last_frame_lost; /* Was previous frame lost */ + opus_int32 rand_seed; /* Seed for unvoiced signal generation */ + opus_int16 randScale_Q14; /* Scaling of unvoiced random signal */ + opus_int32 conc_energy; + opus_int conc_energy_shift; + opus_int16 prevLTP_scale_Q14; + opus_int32 prevGain_Q16[ 2 ]; + opus_int fs_kHz; + opus_int nb_subfr; + opus_int subfr_length; +} silk_PLC_struct; + +/* Struct for CNG */ +typedef struct { + opus_int32 CNG_exc_buf_Q14[ MAX_FRAME_LENGTH ]; + opus_int16 CNG_smth_NLSF_Q15[ MAX_LPC_ORDER ]; + opus_int32 CNG_synth_state[ MAX_LPC_ORDER ]; + opus_int32 CNG_smth_Gain_Q16; + opus_int32 rand_seed; + opus_int fs_kHz; +} silk_CNG_struct; + +/********************************/ +/* Decoder state */ +/********************************/ +typedef struct { + opus_int32 prev_gain_Q16; + opus_int32 exc_Q14[ MAX_FRAME_LENGTH ]; + opus_int32 sLPC_Q14_buf[ MAX_LPC_ORDER ]; + opus_int16 outBuf[ MAX_FRAME_LENGTH + 2 * MAX_SUB_FRAME_LENGTH ]; /* Buffer for output signal */ + opus_int lagPrev; /* Previous Lag */ + opus_int8 LastGainIndex; /* Previous gain index */ + opus_int fs_kHz; /* Sampling frequency in kHz */ + opus_int32 fs_API_hz; /* API sample frequency (Hz) */ + opus_int nb_subfr; /* Number of 5 ms subframes in a frame */ + opus_int frame_length; /* Frame length (samples) */ + opus_int subfr_length; /* Subframe length (samples) */ + opus_int ltp_mem_length; /* Length of LTP memory */ + opus_int LPC_order; /* LPC order */ + opus_int16 prevNLSF_Q15[ MAX_LPC_ORDER ]; /* Used to interpolate LSFs */ + opus_int first_frame_after_reset; /* Flag for deactivating NLSF interpolation */ + const opus_uint8 *pitch_lag_low_bits_iCDF; /* Pointer to iCDF table for low bits of pitch lag index */ + const opus_uint8 *pitch_contour_iCDF; /* Pointer to iCDF table for pitch contour index */ + + /* For buffering payload in case of more frames per packet */ + opus_int nFramesDecoded; + opus_int nFramesPerPacket; + + /* Specifically for entropy coding */ + opus_int ec_prevSignalType; + opus_int16 ec_prevLagIndex; + + opus_int VAD_flags[ MAX_FRAMES_PER_PACKET ]; + opus_int LBRR_flag; + opus_int LBRR_flags[ MAX_FRAMES_PER_PACKET ]; + + silk_resampler_state_struct resampler_state; + + const silk_NLSF_CB_struct *psNLSF_CB; /* Pointer to NLSF codebook */ + + /* Quantization indices */ + SideInfoIndices indices; + + /* CNG state */ + silk_CNG_struct sCNG; + + /* Stuff used for PLC */ + opus_int lossCnt; + opus_int prevSignalType; + int arch; + + silk_PLC_struct sPLC; + +} silk_decoder_state; + +/************************/ +/* Decoder control */ +/************************/ +typedef struct { + /* Prediction and coding parameters */ + opus_int pitchL[ MAX_NB_SUBFR ]; + opus_int32 Gains_Q16[ MAX_NB_SUBFR ]; + /* Holds interpolated and final coefficients, 4-byte aligned */ + silk_DWORD_ALIGN opus_int16 PredCoef_Q12[ 2 ][ MAX_LPC_ORDER ]; + opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ]; + opus_int LTP_scale_Q14; +} silk_decoder_control; + + +#ifdef __cplusplus +} +#endif + +#endif diff --git a/media/libopus/silk/sum_sqr_shift.c b/media/libopus/silk/sum_sqr_shift.c new file mode 100644 index 0000000000..4fd0c3d7d5 --- /dev/null +++ b/media/libopus/silk/sum_sqr_shift.c @@ -0,0 +1,83 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "SigProc_FIX.h" + +/* Compute number of bits to right shift the sum of squares of a vector */ +/* of int16s to make it fit in an int32 */ +void silk_sum_sqr_shift( + opus_int32 *energy, /* O Energy of x, after shifting to the right */ + opus_int *shift, /* O Number of bits right shift applied to energy */ + const opus_int16 *x, /* I Input vector */ + opus_int len /* I Length of input vector */ +) +{ + opus_int i, shft; + opus_uint32 nrg_tmp; + opus_int32 nrg; + + /* Do a first run with the maximum shift we could have. */ + shft = 31-silk_CLZ32(len); + /* Let's be conservative with rounding and start with nrg=len. */ + nrg = len; + for( i = 0; i < len - 1; i += 2 ) { + nrg_tmp = silk_SMULBB( x[ i ], x[ i ] ); + nrg_tmp = silk_SMLABB_ovflw( nrg_tmp, x[ i + 1 ], x[ i + 1 ] ); + nrg = (opus_int32)silk_ADD_RSHIFT_uint( nrg, nrg_tmp, shft ); + } + if( i < len ) { + /* One sample left to process */ + nrg_tmp = silk_SMULBB( x[ i ], x[ i ] ); + nrg = (opus_int32)silk_ADD_RSHIFT_uint( nrg, nrg_tmp, shft ); + } + silk_assert( nrg >= 0 ); + /* Make sure the result will fit in a 32-bit signed integer with two bits + of headroom. */ + shft = silk_max_32(0, shft+3 - silk_CLZ32(nrg)); + nrg = 0; + for( i = 0 ; i < len - 1; i += 2 ) { + nrg_tmp = silk_SMULBB( x[ i ], x[ i ] ); + nrg_tmp = silk_SMLABB_ovflw( nrg_tmp, x[ i + 1 ], x[ i + 1 ] ); + nrg = (opus_int32)silk_ADD_RSHIFT_uint( nrg, nrg_tmp, shft ); + } + if( i < len ) { + /* One sample left to process */ + nrg_tmp = silk_SMULBB( x[ i ], x[ i ] ); + nrg = (opus_int32)silk_ADD_RSHIFT_uint( nrg, nrg_tmp, shft ); + } + + silk_assert( nrg >= 0 ); + + /* Output arguments */ + *shift = shft; + *energy = nrg; +} + diff --git a/media/libopus/silk/table_LSF_cos.c b/media/libopus/silk/table_LSF_cos.c new file mode 100644 index 0000000000..ec9dc63927 --- /dev/null +++ b/media/libopus/silk/table_LSF_cos.c @@ -0,0 +1,70 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "tables.h" + +/* Cosine approximation table for LSF conversion */ +/* Q12 values (even) */ +const opus_int16 silk_LSFCosTab_FIX_Q12[ LSF_COS_TAB_SZ_FIX + 1 ] = { + 8192, 8190, 8182, 8170, + 8152, 8130, 8104, 8072, + 8034, 7994, 7946, 7896, + 7840, 7778, 7714, 7644, + 7568, 7490, 7406, 7318, + 7226, 7128, 7026, 6922, + 6812, 6698, 6580, 6458, + 6332, 6204, 6070, 5934, + 5792, 5648, 5502, 5352, + 5198, 5040, 4880, 4718, + 4552, 4382, 4212, 4038, + 3862, 3684, 3502, 3320, + 3136, 2948, 2760, 2570, + 2378, 2186, 1990, 1794, + 1598, 1400, 1202, 1002, + 802, 602, 402, 202, + 0, -202, -402, -602, + -802, -1002, -1202, -1400, + -1598, -1794, -1990, -2186, + -2378, -2570, -2760, -2948, + -3136, -3320, -3502, -3684, + -3862, -4038, -4212, -4382, + -4552, -4718, -4880, -5040, + -5198, -5352, -5502, -5648, + -5792, -5934, -6070, -6204, + -6332, -6458, -6580, -6698, + -6812, -6922, -7026, -7128, + -7226, -7318, -7406, -7490, + -7568, -7644, -7714, -7778, + -7840, -7896, -7946, -7994, + -8034, -8072, -8104, -8130, + -8152, -8170, -8182, -8190, + -8192 +}; diff --git a/media/libopus/silk/tables.h b/media/libopus/silk/tables.h new file mode 100644 index 0000000000..95230c451a --- /dev/null +++ b/media/libopus/silk/tables.h @@ -0,0 +1,114 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef SILK_TABLES_H +#define SILK_TABLES_H + +#include "define.h" +#include "structs.h" + +#ifdef __cplusplus +extern "C" +{ +#endif + +/* Entropy coding tables (with size in bytes indicated) */ +extern const opus_uint8 silk_gain_iCDF[ 3 ][ N_LEVELS_QGAIN / 8 ]; /* 24 */ +extern const opus_uint8 silk_delta_gain_iCDF[ MAX_DELTA_GAIN_QUANT - MIN_DELTA_GAIN_QUANT + 1 ]; /* 41 */ + +extern const opus_uint8 silk_pitch_lag_iCDF[ 2 * ( PITCH_EST_MAX_LAG_MS - PITCH_EST_MIN_LAG_MS ) ];/* 32 */ +extern const opus_uint8 silk_pitch_delta_iCDF[ 21 ]; /* 21 */ +extern const opus_uint8 silk_pitch_contour_iCDF[ 34 ]; /* 34 */ +extern const opus_uint8 silk_pitch_contour_NB_iCDF[ 11 ]; /* 11 */ +extern const opus_uint8 silk_pitch_contour_10_ms_iCDF[ 12 ]; /* 12 */ +extern const opus_uint8 silk_pitch_contour_10_ms_NB_iCDF[ 3 ]; /* 3 */ + +extern const opus_uint8 silk_pulses_per_block_iCDF[ N_RATE_LEVELS ][ SILK_MAX_PULSES + 2 ]; /* 180 */ +extern const opus_uint8 silk_pulses_per_block_BITS_Q5[ N_RATE_LEVELS - 1 ][ SILK_MAX_PULSES + 2 ]; /* 162 */ + +extern const opus_uint8 silk_rate_levels_iCDF[ 2 ][ N_RATE_LEVELS - 1 ]; /* 18 */ +extern const opus_uint8 silk_rate_levels_BITS_Q5[ 2 ][ N_RATE_LEVELS - 1 ]; /* 18 */ + +extern const opus_uint8 silk_max_pulses_table[ 4 ]; /* 4 */ + +extern const opus_uint8 silk_shell_code_table0[ 152 ]; /* 152 */ +extern const opus_uint8 silk_shell_code_table1[ 152 ]; /* 152 */ +extern const opus_uint8 silk_shell_code_table2[ 152 ]; /* 152 */ +extern const opus_uint8 silk_shell_code_table3[ 152 ]; /* 152 */ +extern const opus_uint8 silk_shell_code_table_offsets[ SILK_MAX_PULSES + 1 ]; /* 17 */ + +extern const opus_uint8 silk_lsb_iCDF[ 2 ]; /* 2 */ + +extern const opus_uint8 silk_sign_iCDF[ 42 ]; /* 42 */ + +extern const opus_uint8 silk_uniform3_iCDF[ 3 ]; /* 3 */ +extern const opus_uint8 silk_uniform4_iCDF[ 4 ]; /* 4 */ +extern const opus_uint8 silk_uniform5_iCDF[ 5 ]; /* 5 */ +extern const opus_uint8 silk_uniform6_iCDF[ 6 ]; /* 6 */ +extern const opus_uint8 silk_uniform8_iCDF[ 8 ]; /* 8 */ + +extern const opus_uint8 silk_NLSF_EXT_iCDF[ 7 ]; /* 7 */ + +extern const opus_uint8 silk_LTP_per_index_iCDF[ 3 ]; /* 3 */ +extern const opus_uint8 * const silk_LTP_gain_iCDF_ptrs[ NB_LTP_CBKS ]; /* 3 */ +extern const opus_uint8 * const silk_LTP_gain_BITS_Q5_ptrs[ NB_LTP_CBKS ]; /* 3 */ +extern const opus_int8 * const silk_LTP_vq_ptrs_Q7[ NB_LTP_CBKS ]; /* 168 */ +extern const opus_uint8 * const silk_LTP_vq_gain_ptrs_Q7[NB_LTP_CBKS]; +extern const opus_int8 silk_LTP_vq_sizes[ NB_LTP_CBKS ]; /* 3 */ + +extern const opus_uint8 silk_LTPscale_iCDF[ 3 ]; /* 4 */ +extern const opus_int16 silk_LTPScales_table_Q14[ 3 ]; /* 6 */ + +extern const opus_uint8 silk_type_offset_VAD_iCDF[ 4 ]; /* 4 */ +extern const opus_uint8 silk_type_offset_no_VAD_iCDF[ 2 ]; /* 2 */ + +extern const opus_int16 silk_stereo_pred_quant_Q13[ STEREO_QUANT_TAB_SIZE ]; /* 32 */ +extern const opus_uint8 silk_stereo_pred_joint_iCDF[ 25 ]; /* 25 */ +extern const opus_uint8 silk_stereo_only_code_mid_iCDF[ 2 ]; /* 2 */ + +extern const opus_uint8 * const silk_LBRR_flags_iCDF_ptr[ 2 ]; /* 10 */ + +extern const opus_uint8 silk_NLSF_interpolation_factor_iCDF[ 5 ]; /* 5 */ + +extern const silk_NLSF_CB_struct silk_NLSF_CB_WB; /* 1040 */ +extern const silk_NLSF_CB_struct silk_NLSF_CB_NB_MB; /* 728 */ + +/* Quantization offsets */ +extern const opus_int16 silk_Quantization_Offsets_Q10[ 2 ][ 2 ]; /* 8 */ + +/* Interpolation points for filter coefficients used in the bandwidth transition smoother */ +extern const opus_int32 silk_Transition_LP_B_Q28[ TRANSITION_INT_NUM ][ TRANSITION_NB ]; /* 60 */ +extern const opus_int32 silk_Transition_LP_A_Q28[ TRANSITION_INT_NUM ][ TRANSITION_NA ]; /* 60 */ + +/* Rom table with cosine values */ +extern const opus_int16 silk_LSFCosTab_FIX_Q12[ LSF_COS_TAB_SZ_FIX + 1 ]; /* 258 */ + +#ifdef __cplusplus +} +#endif + +#endif diff --git a/media/libopus/silk/tables_LTP.c b/media/libopus/silk/tables_LTP.c new file mode 100644 index 0000000000..5e12c8643e --- /dev/null +++ b/media/libopus/silk/tables_LTP.c @@ -0,0 +1,294 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "tables.h" + +const opus_uint8 silk_LTP_per_index_iCDF[3] = { + 179, 99, 0 +}; + +static const opus_uint8 silk_LTP_gain_iCDF_0[8] = { + 71, 56, 43, 30, 21, 12, 6, 0 +}; + +static const opus_uint8 silk_LTP_gain_iCDF_1[16] = { + 199, 165, 144, 124, 109, 96, 84, 71, + 61, 51, 42, 32, 23, 15, 8, 0 +}; + +static const opus_uint8 silk_LTP_gain_iCDF_2[32] = { + 241, 225, 211, 199, 187, 175, 164, 153, + 142, 132, 123, 114, 105, 96, 88, 80, + 72, 64, 57, 50, 44, 38, 33, 29, + 24, 20, 16, 12, 9, 5, 2, 0 +}; + +static const opus_uint8 silk_LTP_gain_BITS_Q5_0[8] = { + 15, 131, 138, 138, 155, 155, 173, 173 +}; + +static const opus_uint8 silk_LTP_gain_BITS_Q5_1[16] = { + 69, 93, 115, 118, 131, 138, 141, 138, + 150, 150, 155, 150, 155, 160, 166, 160 +}; + +static const opus_uint8 silk_LTP_gain_BITS_Q5_2[32] = { + 131, 128, 134, 141, 141, 141, 145, 145, + 145, 150, 155, 155, 155, 155, 160, 160, + 160, 160, 166, 166, 173, 173, 182, 192, + 182, 192, 192, 192, 205, 192, 205, 224 +}; + +const opus_uint8 * const silk_LTP_gain_iCDF_ptrs[NB_LTP_CBKS] = { + silk_LTP_gain_iCDF_0, + silk_LTP_gain_iCDF_1, + silk_LTP_gain_iCDF_2 +}; + +const opus_uint8 * const silk_LTP_gain_BITS_Q5_ptrs[NB_LTP_CBKS] = { + silk_LTP_gain_BITS_Q5_0, + silk_LTP_gain_BITS_Q5_1, + silk_LTP_gain_BITS_Q5_2 +}; + +static const opus_int8 silk_LTP_gain_vq_0[8][5] = +{ +{ + 4, 6, 24, 7, 5 +}, +{ + 0, 0, 2, 0, 0 +}, +{ + 12, 28, 41, 13, -4 +}, +{ + -9, 15, 42, 25, 14 +}, +{ + 1, -2, 62, 41, -9 +}, +{ + -10, 37, 65, -4, 3 +}, +{ + -6, 4, 66, 7, -8 +}, +{ + 16, 14, 38, -3, 33 +} +}; + +static const opus_int8 silk_LTP_gain_vq_1[16][5] = +{ +{ + 13, 22, 39, 23, 12 +}, +{ + -1, 36, 64, 27, -6 +}, +{ + -7, 10, 55, 43, 17 +}, +{ + 1, 1, 8, 1, 1 +}, +{ + 6, -11, 74, 53, -9 +}, +{ + -12, 55, 76, -12, 8 +}, +{ + -3, 3, 93, 27, -4 +}, +{ + 26, 39, 59, 3, -8 +}, +{ + 2, 0, 77, 11, 9 +}, +{ + -8, 22, 44, -6, 7 +}, +{ + 40, 9, 26, 3, 9 +}, +{ + -7, 20, 101, -7, 4 +}, +{ + 3, -8, 42, 26, 0 +}, +{ + -15, 33, 68, 2, 23 +}, +{ + -2, 55, 46, -2, 15 +}, +{ + 3, -1, 21, 16, 41 +} +}; + +static const opus_int8 silk_LTP_gain_vq_2[32][5] = +{ +{ + -6, 27, 61, 39, 5 +}, +{ + -11, 42, 88, 4, 1 +}, +{ + -2, 60, 65, 6, -4 +}, +{ + -1, -5, 73, 56, 1 +}, +{ + -9, 19, 94, 29, -9 +}, +{ + 0, 12, 99, 6, 4 +}, +{ + 8, -19, 102, 46, -13 +}, +{ + 3, 2, 13, 3, 2 +}, +{ + 9, -21, 84, 72, -18 +}, +{ + -11, 46, 104, -22, 8 +}, +{ + 18, 38, 48, 23, 0 +}, +{ + -16, 70, 83, -21, 11 +}, +{ + 5, -11, 117, 22, -8 +}, +{ + -6, 23, 117, -12, 3 +}, +{ + 3, -8, 95, 28, 4 +}, +{ + -10, 15, 77, 60, -15 +}, +{ + -1, 4, 124, 2, -4 +}, +{ + 3, 38, 84, 24, -25 +}, +{ + 2, 13, 42, 13, 31 +}, +{ + 21, -4, 56, 46, -1 +}, +{ + -1, 35, 79, -13, 19 +}, +{ + -7, 65, 88, -9, -14 +}, +{ + 20, 4, 81, 49, -29 +}, +{ + 20, 0, 75, 3, -17 +}, +{ + 5, -9, 44, 92, -8 +}, +{ + 1, -3, 22, 69, 31 +}, +{ + -6, 95, 41, -12, 5 +}, +{ + 39, 67, 16, -4, 1 +}, +{ + 0, -6, 120, 55, -36 +}, +{ + -13, 44, 122, 4, -24 +}, +{ + 81, 5, 11, 3, 7 +}, +{ + 2, 0, 9, 10, 88 +} +}; + +const opus_int8 * const silk_LTP_vq_ptrs_Q7[NB_LTP_CBKS] = { + (opus_int8 *)&silk_LTP_gain_vq_0[0][0], + (opus_int8 *)&silk_LTP_gain_vq_1[0][0], + (opus_int8 *)&silk_LTP_gain_vq_2[0][0] +}; + +/* Maximum frequency-dependent response of the pitch taps above, + computed as max(abs(freqz(taps))) */ +static const opus_uint8 silk_LTP_gain_vq_0_gain[8] = { + 46, 2, 90, 87, 93, 91, 82, 98 +}; + +static const opus_uint8 silk_LTP_gain_vq_1_gain[16] = { + 109, 120, 118, 12, 113, 115, 117, 119, + 99, 59, 87, 111, 63, 111, 112, 80 +}; + +static const opus_uint8 silk_LTP_gain_vq_2_gain[32] = { + 126, 124, 125, 124, 129, 121, 126, 23, + 132, 127, 127, 127, 126, 127, 122, 133, + 130, 134, 101, 118, 119, 145, 126, 86, + 124, 120, 123, 119, 170, 173, 107, 109 +}; + +const opus_uint8 * const silk_LTP_vq_gain_ptrs_Q7[NB_LTP_CBKS] = { + &silk_LTP_gain_vq_0_gain[0], + &silk_LTP_gain_vq_1_gain[0], + &silk_LTP_gain_vq_2_gain[0] +}; + +const opus_int8 silk_LTP_vq_sizes[NB_LTP_CBKS] = { + 8, 16, 32 +}; diff --git a/media/libopus/silk/tables_NLSF_CB_NB_MB.c b/media/libopus/silk/tables_NLSF_CB_NB_MB.c new file mode 100644 index 0000000000..195d5b95bd --- /dev/null +++ b/media/libopus/silk/tables_NLSF_CB_NB_MB.c @@ -0,0 +1,195 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "tables.h" + +static const opus_uint8 silk_NLSF_CB1_NB_MB_Q8[ 320 ] = { + 12, 35, 60, 83, 108, 132, 157, 180, + 206, 228, 15, 32, 55, 77, 101, 125, + 151, 175, 201, 225, 19, 42, 66, 89, + 114, 137, 162, 184, 209, 230, 12, 25, + 50, 72, 97, 120, 147, 172, 200, 223, + 26, 44, 69, 90, 114, 135, 159, 180, + 205, 225, 13, 22, 53, 80, 106, 130, + 156, 180, 205, 228, 15, 25, 44, 64, + 90, 115, 142, 168, 196, 222, 19, 24, + 62, 82, 100, 120, 145, 168, 190, 214, + 22, 31, 50, 79, 103, 120, 151, 170, + 203, 227, 21, 29, 45, 65, 106, 124, + 150, 171, 196, 224, 30, 49, 75, 97, + 121, 142, 165, 186, 209, 229, 19, 25, + 52, 70, 93, 116, 143, 166, 192, 219, + 26, 34, 62, 75, 97, 118, 145, 167, + 194, 217, 25, 33, 56, 70, 91, 113, + 143, 165, 196, 223, 21, 34, 51, 72, + 97, 117, 145, 171, 196, 222, 20, 29, + 50, 67, 90, 117, 144, 168, 197, 221, + 22, 31, 48, 66, 95, 117, 146, 168, + 196, 222, 24, 33, 51, 77, 116, 134, + 158, 180, 200, 224, 21, 28, 70, 87, + 106, 124, 149, 170, 194, 217, 26, 33, + 53, 64, 83, 117, 152, 173, 204, 225, + 27, 34, 65, 95, 108, 129, 155, 174, + 210, 225, 20, 26, 72, 99, 113, 131, + 154, 176, 200, 219, 34, 43, 61, 78, + 93, 114, 155, 177, 205, 229, 23, 29, + 54, 97, 124, 138, 163, 179, 209, 229, + 30, 38, 56, 89, 118, 129, 158, 178, + 200, 231, 21, 29, 49, 63, 85, 111, + 142, 163, 193, 222, 27, 48, 77, 103, + 133, 158, 179, 196, 215, 232, 29, 47, + 74, 99, 124, 151, 176, 198, 220, 237, + 33, 42, 61, 76, 93, 121, 155, 174, + 207, 225, 29, 53, 87, 112, 136, 154, + 170, 188, 208, 227, 24, 30, 52, 84, + 131, 150, 166, 186, 203, 229, 37, 48, + 64, 84, 104, 118, 156, 177, 201, 230 +}; + +static const opus_int16 silk_NLSF_CB1_Wght_Q9[ 320 ] = { + 2897, 2314, 2314, 2314, 2287, 2287, 2314, 2300, 2327, 2287, + 2888, 2580, 2394, 2367, 2314, 2274, 2274, 2274, 2274, 2194, + 2487, 2340, 2340, 2314, 2314, 2314, 2340, 2340, 2367, 2354, + 3216, 2766, 2340, 2340, 2314, 2274, 2221, 2207, 2261, 2194, + 2460, 2474, 2367, 2394, 2394, 2394, 2394, 2367, 2407, 2314, + 3479, 3056, 2127, 2207, 2274, 2274, 2274, 2287, 2314, 2261, + 3282, 3141, 2580, 2394, 2247, 2221, 2207, 2194, 2194, 2114, + 4096, 3845, 2221, 2620, 2620, 2407, 2314, 2394, 2367, 2074, + 3178, 3244, 2367, 2221, 2553, 2434, 2340, 2314, 2167, 2221, + 3338, 3488, 2726, 2194, 2261, 2460, 2354, 2367, 2207, 2101, + 2354, 2420, 2327, 2367, 2394, 2420, 2420, 2420, 2460, 2367, + 3779, 3629, 2434, 2527, 2367, 2274, 2274, 2300, 2207, 2048, + 3254, 3225, 2713, 2846, 2447, 2327, 2300, 2300, 2274, 2127, + 3263, 3300, 2753, 2806, 2447, 2261, 2261, 2247, 2127, 2101, + 2873, 2981, 2633, 2367, 2407, 2354, 2194, 2247, 2247, 2114, + 3225, 3197, 2633, 2580, 2274, 2181, 2247, 2221, 2221, 2141, + 3178, 3310, 2740, 2407, 2274, 2274, 2274, 2287, 2194, 2114, + 3141, 3272, 2460, 2061, 2287, 2500, 2367, 2487, 2434, 2181, + 3507, 3282, 2314, 2700, 2647, 2474, 2367, 2394, 2340, 2127, + 3423, 3535, 3038, 3056, 2300, 1950, 2221, 2274, 2274, 2274, + 3404, 3366, 2087, 2687, 2873, 2354, 2420, 2274, 2474, 2540, + 3760, 3488, 1950, 2660, 2897, 2527, 2394, 2367, 2460, 2261, + 3028, 3272, 2740, 2888, 2740, 2154, 2127, 2287, 2234, 2247, + 3695, 3657, 2025, 1969, 2660, 2700, 2580, 2500, 2327, 2367, + 3207, 3413, 2354, 2074, 2888, 2888, 2340, 2487, 2247, 2167, + 3338, 3366, 2846, 2780, 2327, 2154, 2274, 2287, 2114, 2061, + 2327, 2300, 2181, 2167, 2181, 2367, 2633, 2700, 2700, 2553, + 2407, 2434, 2221, 2261, 2221, 2221, 2340, 2420, 2607, 2700, + 3038, 3244, 2806, 2888, 2474, 2074, 2300, 2314, 2354, 2380, + 2221, 2154, 2127, 2287, 2500, 2793, 2793, 2620, 2580, 2367, + 3676, 3713, 2234, 1838, 2181, 2753, 2726, 2673, 2513, 2207, + 2793, 3160, 2726, 2553, 2846, 2513, 2181, 2394, 2221, 2181 +}; + +static const opus_uint8 silk_NLSF_CB1_iCDF_NB_MB[ 64 ] = { + 212, 178, 148, 129, 108, 96, 85, 82, + 79, 77, 61, 59, 57, 56, 51, 49, + 48, 45, 42, 41, 40, 38, 36, 34, + 31, 30, 21, 12, 10, 3, 1, 0, + 255, 245, 244, 236, 233, 225, 217, 203, + 190, 176, 175, 161, 149, 136, 125, 114, + 102, 91, 81, 71, 60, 52, 43, 35, + 28, 20, 19, 18, 12, 11, 5, 0 +}; + +static const opus_uint8 silk_NLSF_CB2_SELECT_NB_MB[ 160 ] = { + 16, 0, 0, 0, 0, 99, 66, 36, + 36, 34, 36, 34, 34, 34, 34, 83, + 69, 36, 52, 34, 116, 102, 70, 68, + 68, 176, 102, 68, 68, 34, 65, 85, + 68, 84, 36, 116, 141, 152, 139, 170, + 132, 187, 184, 216, 137, 132, 249, 168, + 185, 139, 104, 102, 100, 68, 68, 178, + 218, 185, 185, 170, 244, 216, 187, 187, + 170, 244, 187, 187, 219, 138, 103, 155, + 184, 185, 137, 116, 183, 155, 152, 136, + 132, 217, 184, 184, 170, 164, 217, 171, + 155, 139, 244, 169, 184, 185, 170, 164, + 216, 223, 218, 138, 214, 143, 188, 218, + 168, 244, 141, 136, 155, 170, 168, 138, + 220, 219, 139, 164, 219, 202, 216, 137, + 168, 186, 246, 185, 139, 116, 185, 219, + 185, 138, 100, 100, 134, 100, 102, 34, + 68, 68, 100, 68, 168, 203, 221, 218, + 168, 167, 154, 136, 104, 70, 164, 246, + 171, 137, 139, 137, 155, 218, 219, 139 +}; + +static const opus_uint8 silk_NLSF_CB2_iCDF_NB_MB[ 72 ] = { + 255, 254, 253, 238, 14, 3, 2, 1, + 0, 255, 254, 252, 218, 35, 3, 2, + 1, 0, 255, 254, 250, 208, 59, 4, + 2, 1, 0, 255, 254, 246, 194, 71, + 10, 2, 1, 0, 255, 252, 236, 183, + 82, 8, 2, 1, 0, 255, 252, 235, + 180, 90, 17, 2, 1, 0, 255, 248, + 224, 171, 97, 30, 4, 1, 0, 255, + 254, 236, 173, 95, 37, 7, 1, 0 +}; + +static const opus_uint8 silk_NLSF_CB2_BITS_NB_MB_Q5[ 72 ] = { + 255, 255, 255, 131, 6, 145, 255, 255, + 255, 255, 255, 236, 93, 15, 96, 255, + 255, 255, 255, 255, 194, 83, 25, 71, + 221, 255, 255, 255, 255, 162, 73, 34, + 66, 162, 255, 255, 255, 210, 126, 73, + 43, 57, 173, 255, 255, 255, 201, 125, + 71, 48, 58, 130, 255, 255, 255, 166, + 110, 73, 57, 62, 104, 210, 255, 255, + 251, 123, 65, 55, 68, 100, 171, 255 +}; + +static const opus_uint8 silk_NLSF_PRED_NB_MB_Q8[ 18 ] = { + 179, 138, 140, 148, 151, 149, 153, 151, + 163, 116, 67, 82, 59, 92, 72, 100, + 89, 92 +}; + +static const opus_int16 silk_NLSF_DELTA_MIN_NB_MB_Q15[ 11 ] = { + 250, 3, 6, 3, 3, 3, 4, 3, + 3, 3, 461 +}; + +const silk_NLSF_CB_struct silk_NLSF_CB_NB_MB = +{ + 32, + 10, + SILK_FIX_CONST( 0.18, 16 ), + SILK_FIX_CONST( 1.0 / 0.18, 6 ), + silk_NLSF_CB1_NB_MB_Q8, + silk_NLSF_CB1_Wght_Q9, + silk_NLSF_CB1_iCDF_NB_MB, + silk_NLSF_PRED_NB_MB_Q8, + silk_NLSF_CB2_SELECT_NB_MB, + silk_NLSF_CB2_iCDF_NB_MB, + silk_NLSF_CB2_BITS_NB_MB_Q5, + silk_NLSF_DELTA_MIN_NB_MB_Q15, +}; diff --git a/media/libopus/silk/tables_NLSF_CB_WB.c b/media/libopus/silk/tables_NLSF_CB_WB.c new file mode 100644 index 0000000000..5cc9f57bff --- /dev/null +++ b/media/libopus/silk/tables_NLSF_CB_WB.c @@ -0,0 +1,234 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "tables.h" + +static const opus_uint8 silk_NLSF_CB1_WB_Q8[ 512 ] = { + 7, 23, 38, 54, 69, 85, 100, 116, + 131, 147, 162, 178, 193, 208, 223, 239, + 13, 25, 41, 55, 69, 83, 98, 112, + 127, 142, 157, 171, 187, 203, 220, 236, + 15, 21, 34, 51, 61, 78, 92, 106, + 126, 136, 152, 167, 185, 205, 225, 240, + 10, 21, 36, 50, 63, 79, 95, 110, + 126, 141, 157, 173, 189, 205, 221, 237, + 17, 20, 37, 51, 59, 78, 89, 107, + 123, 134, 150, 164, 184, 205, 224, 240, + 10, 15, 32, 51, 67, 81, 96, 112, + 129, 142, 158, 173, 189, 204, 220, 236, + 8, 21, 37, 51, 65, 79, 98, 113, + 126, 138, 155, 168, 179, 192, 209, 218, + 12, 15, 34, 55, 63, 78, 87, 108, + 118, 131, 148, 167, 185, 203, 219, 236, + 16, 19, 32, 36, 56, 79, 91, 108, + 118, 136, 154, 171, 186, 204, 220, 237, + 11, 28, 43, 58, 74, 89, 105, 120, + 135, 150, 165, 180, 196, 211, 226, 241, + 6, 16, 33, 46, 60, 75, 92, 107, + 123, 137, 156, 169, 185, 199, 214, 225, + 11, 19, 30, 44, 57, 74, 89, 105, + 121, 135, 152, 169, 186, 202, 218, 234, + 12, 19, 29, 46, 57, 71, 88, 100, + 120, 132, 148, 165, 182, 199, 216, 233, + 17, 23, 35, 46, 56, 77, 92, 106, + 123, 134, 152, 167, 185, 204, 222, 237, + 14, 17, 45, 53, 63, 75, 89, 107, + 115, 132, 151, 171, 188, 206, 221, 240, + 9, 16, 29, 40, 56, 71, 88, 103, + 119, 137, 154, 171, 189, 205, 222, 237, + 16, 19, 36, 48, 57, 76, 87, 105, + 118, 132, 150, 167, 185, 202, 218, 236, + 12, 17, 29, 54, 71, 81, 94, 104, + 126, 136, 149, 164, 182, 201, 221, 237, + 15, 28, 47, 62, 79, 97, 115, 129, + 142, 155, 168, 180, 194, 208, 223, 238, + 8, 14, 30, 45, 62, 78, 94, 111, + 127, 143, 159, 175, 192, 207, 223, 239, + 17, 30, 49, 62, 79, 92, 107, 119, + 132, 145, 160, 174, 190, 204, 220, 235, + 14, 19, 36, 45, 61, 76, 91, 108, + 121, 138, 154, 172, 189, 205, 222, 238, + 12, 18, 31, 45, 60, 76, 91, 107, + 123, 138, 154, 171, 187, 204, 221, 236, + 13, 17, 31, 43, 53, 70, 83, 103, + 114, 131, 149, 167, 185, 203, 220, 237, + 17, 22, 35, 42, 58, 78, 93, 110, + 125, 139, 155, 170, 188, 206, 224, 240, + 8, 15, 34, 50, 67, 83, 99, 115, + 131, 146, 162, 178, 193, 209, 224, 239, + 13, 16, 41, 66, 73, 86, 95, 111, + 128, 137, 150, 163, 183, 206, 225, 241, + 17, 25, 37, 52, 63, 75, 92, 102, + 119, 132, 144, 160, 175, 191, 212, 231, + 19, 31, 49, 65, 83, 100, 117, 133, + 147, 161, 174, 187, 200, 213, 227, 242, + 18, 31, 52, 68, 88, 103, 117, 126, + 138, 149, 163, 177, 192, 207, 223, 239, + 16, 29, 47, 61, 76, 90, 106, 119, + 133, 147, 161, 176, 193, 209, 224, 240, + 15, 21, 35, 50, 61, 73, 86, 97, + 110, 119, 129, 141, 175, 198, 218, 237 +}; + +static const opus_int16 silk_NLSF_CB1_WB_Wght_Q9[ 512 ] = { + 3657, 2925, 2925, 2925, 2925, 2925, 2925, 2925, 2925, 2925, 2925, 2925, 2963, 2963, 2925, 2846, + 3216, 3085, 2972, 3056, 3056, 3010, 3010, 3010, 2963, 2963, 3010, 2972, 2888, 2846, 2846, 2726, + 3920, 4014, 2981, 3207, 3207, 2934, 3056, 2846, 3122, 3244, 2925, 2846, 2620, 2553, 2780, 2925, + 3516, 3197, 3010, 3103, 3019, 2888, 2925, 2925, 2925, 2925, 2888, 2888, 2888, 2888, 2888, 2753, + 5054, 5054, 2934, 3573, 3385, 3056, 3085, 2793, 3160, 3160, 2972, 2846, 2513, 2540, 2753, 2888, + 4428, 4149, 2700, 2753, 2972, 3010, 2925, 2846, 2981, 3019, 2925, 2925, 2925, 2925, 2888, 2726, + 3620, 3019, 2972, 3056, 3056, 2873, 2806, 3056, 3216, 3047, 2981, 3291, 3291, 2981, 3310, 2991, + 5227, 5014, 2540, 3338, 3526, 3385, 3197, 3094, 3376, 2981, 2700, 2647, 2687, 2793, 2846, 2673, + 5081, 5174, 4615, 4428, 2460, 2897, 3047, 3207, 3169, 2687, 2740, 2888, 2846, 2793, 2846, 2700, + 3122, 2888, 2963, 2925, 2925, 2925, 2925, 2963, 2963, 2963, 2963, 2925, 2925, 2963, 2963, 2963, + 4202, 3207, 2981, 3103, 3010, 2888, 2888, 2925, 2972, 2873, 2916, 3019, 2972, 3010, 3197, 2873, + 3760, 3760, 3244, 3103, 2981, 2888, 2925, 2888, 2972, 2934, 2793, 2793, 2846, 2888, 2888, 2660, + 3854, 4014, 3207, 3122, 3244, 2934, 3047, 2963, 2963, 3085, 2846, 2793, 2793, 2793, 2793, 2580, + 3845, 4080, 3357, 3516, 3094, 2740, 3010, 2934, 3122, 3085, 2846, 2846, 2647, 2647, 2846, 2806, + 5147, 4894, 3225, 3845, 3441, 3169, 2897, 3413, 3451, 2700, 2580, 2673, 2740, 2846, 2806, 2753, + 4109, 3789, 3291, 3160, 2925, 2888, 2888, 2925, 2793, 2740, 2793, 2740, 2793, 2846, 2888, 2806, + 5081, 5054, 3047, 3545, 3244, 3056, 3085, 2944, 3103, 2897, 2740, 2740, 2740, 2846, 2793, 2620, + 4309, 4309, 2860, 2527, 3207, 3376, 3376, 3075, 3075, 3376, 3056, 2846, 2647, 2580, 2726, 2753, + 3056, 2916, 2806, 2888, 2740, 2687, 2897, 3103, 3150, 3150, 3216, 3169, 3056, 3010, 2963, 2846, + 4375, 3882, 2925, 2888, 2846, 2888, 2846, 2846, 2888, 2888, 2888, 2846, 2888, 2925, 2888, 2846, + 2981, 2916, 2916, 2981, 2981, 3056, 3122, 3216, 3150, 3056, 3010, 2972, 2972, 2972, 2925, 2740, + 4229, 4149, 3310, 3347, 2925, 2963, 2888, 2981, 2981, 2846, 2793, 2740, 2846, 2846, 2846, 2793, + 4080, 4014, 3103, 3010, 2925, 2925, 2925, 2888, 2925, 2925, 2846, 2846, 2846, 2793, 2888, 2780, + 4615, 4575, 3169, 3441, 3207, 2981, 2897, 3038, 3122, 2740, 2687, 2687, 2687, 2740, 2793, 2700, + 4149, 4269, 3789, 3657, 2726, 2780, 2888, 2888, 3010, 2972, 2925, 2846, 2687, 2687, 2793, 2888, + 4215, 3554, 2753, 2846, 2846, 2888, 2888, 2888, 2925, 2925, 2888, 2925, 2925, 2925, 2963, 2888, + 5174, 4921, 2261, 3432, 3789, 3479, 3347, 2846, 3310, 3479, 3150, 2897, 2460, 2487, 2753, 2925, + 3451, 3685, 3122, 3197, 3357, 3047, 3207, 3207, 2981, 3216, 3085, 2925, 2925, 2687, 2540, 2434, + 2981, 3010, 2793, 2793, 2740, 2793, 2846, 2972, 3056, 3103, 3150, 3150, 3150, 3103, 3010, 3010, + 2944, 2873, 2687, 2726, 2780, 3010, 3432, 3545, 3357, 3244, 3056, 3010, 2963, 2925, 2888, 2846, + 3019, 2944, 2897, 3010, 3010, 2972, 3019, 3103, 3056, 3056, 3010, 2888, 2846, 2925, 2925, 2888, + 3920, 3967, 3010, 3197, 3357, 3216, 3291, 3291, 3479, 3704, 3441, 2726, 2181, 2460, 2580, 2607 +}; + +static const opus_uint8 silk_NLSF_CB1_iCDF_WB[ 64 ] = { + 225, 204, 201, 184, 183, 175, 158, 154, + 153, 135, 119, 115, 113, 110, 109, 99, + 98, 95, 79, 68, 52, 50, 48, 45, + 43, 32, 31, 27, 18, 10, 3, 0, + 255, 251, 235, 230, 212, 201, 196, 182, + 167, 166, 163, 151, 138, 124, 110, 104, + 90, 78, 76, 70, 69, 57, 45, 34, + 24, 21, 11, 6, 5, 4, 3, 0 +}; + +static const opus_uint8 silk_NLSF_CB2_SELECT_WB[ 256 ] = { + 0, 0, 0, 0, 0, 0, 0, 1, + 100, 102, 102, 68, 68, 36, 34, 96, + 164, 107, 158, 185, 180, 185, 139, 102, + 64, 66, 36, 34, 34, 0, 1, 32, + 208, 139, 141, 191, 152, 185, 155, 104, + 96, 171, 104, 166, 102, 102, 102, 132, + 1, 0, 0, 0, 0, 16, 16, 0, + 80, 109, 78, 107, 185, 139, 103, 101, + 208, 212, 141, 139, 173, 153, 123, 103, + 36, 0, 0, 0, 0, 0, 0, 1, + 48, 0, 0, 0, 0, 0, 0, 32, + 68, 135, 123, 119, 119, 103, 69, 98, + 68, 103, 120, 118, 118, 102, 71, 98, + 134, 136, 157, 184, 182, 153, 139, 134, + 208, 168, 248, 75, 189, 143, 121, 107, + 32, 49, 34, 34, 34, 0, 17, 2, + 210, 235, 139, 123, 185, 137, 105, 134, + 98, 135, 104, 182, 100, 183, 171, 134, + 100, 70, 68, 70, 66, 66, 34, 131, + 64, 166, 102, 68, 36, 2, 1, 0, + 134, 166, 102, 68, 34, 34, 66, 132, + 212, 246, 158, 139, 107, 107, 87, 102, + 100, 219, 125, 122, 137, 118, 103, 132, + 114, 135, 137, 105, 171, 106, 50, 34, + 164, 214, 141, 143, 185, 151, 121, 103, + 192, 34, 0, 0, 0, 0, 0, 1, + 208, 109, 74, 187, 134, 249, 159, 137, + 102, 110, 154, 118, 87, 101, 119, 101, + 0, 2, 0, 36, 36, 66, 68, 35, + 96, 164, 102, 100, 36, 0, 2, 33, + 167, 138, 174, 102, 100, 84, 2, 2, + 100, 107, 120, 119, 36, 197, 24, 0 +}; + +static const opus_uint8 silk_NLSF_CB2_iCDF_WB[ 72 ] = { + 255, 254, 253, 244, 12, 3, 2, 1, + 0, 255, 254, 252, 224, 38, 3, 2, + 1, 0, 255, 254, 251, 209, 57, 4, + 2, 1, 0, 255, 254, 244, 195, 69, + 4, 2, 1, 0, 255, 251, 232, 184, + 84, 7, 2, 1, 0, 255, 254, 240, + 186, 86, 14, 2, 1, 0, 255, 254, + 239, 178, 91, 30, 5, 1, 0, 255, + 248, 227, 177, 100, 19, 2, 1, 0 +}; + +static const opus_uint8 silk_NLSF_CB2_BITS_WB_Q5[ 72 ] = { + 255, 255, 255, 156, 4, 154, 255, 255, + 255, 255, 255, 227, 102, 15, 92, 255, + 255, 255, 255, 255, 213, 83, 24, 72, + 236, 255, 255, 255, 255, 150, 76, 33, + 63, 214, 255, 255, 255, 190, 121, 77, + 43, 55, 185, 255, 255, 255, 245, 137, + 71, 43, 59, 139, 255, 255, 255, 255, + 131, 66, 50, 66, 107, 194, 255, 255, + 166, 116, 76, 55, 53, 125, 255, 255 +}; + +static const opus_uint8 silk_NLSF_PRED_WB_Q8[ 30 ] = { + 175, 148, 160, 176, 178, 173, 174, 164, + 177, 174, 196, 182, 198, 192, 182, 68, + 62, 66, 60, 72, 117, 85, 90, 118, + 136, 151, 142, 160, 142, 155 +}; + +static const opus_int16 silk_NLSF_DELTA_MIN_WB_Q15[ 17 ] = { + 100, 3, 40, 3, 3, 3, 5, 14, + 14, 10, 11, 3, 8, 9, 7, 3, + 347 +}; + +const silk_NLSF_CB_struct silk_NLSF_CB_WB = +{ + 32, + 16, + SILK_FIX_CONST( 0.15, 16 ), + SILK_FIX_CONST( 1.0 / 0.15, 6 ), + silk_NLSF_CB1_WB_Q8, + silk_NLSF_CB1_WB_Wght_Q9, + silk_NLSF_CB1_iCDF_WB, + silk_NLSF_PRED_WB_Q8, + silk_NLSF_CB2_SELECT_WB, + silk_NLSF_CB2_iCDF_WB, + silk_NLSF_CB2_BITS_WB_Q5, + silk_NLSF_DELTA_MIN_WB_Q15, +}; + diff --git a/media/libopus/silk/tables_gain.c b/media/libopus/silk/tables_gain.c new file mode 100644 index 0000000000..37e41d890c --- /dev/null +++ b/media/libopus/silk/tables_gain.c @@ -0,0 +1,63 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "tables.h" + +#ifdef __cplusplus +extern "C" +{ +#endif + +const opus_uint8 silk_gain_iCDF[ 3 ][ N_LEVELS_QGAIN / 8 ] = +{ +{ + 224, 112, 44, 15, 3, 2, 1, 0 +}, +{ + 254, 237, 192, 132, 70, 23, 4, 0 +}, +{ + 255, 252, 226, 155, 61, 11, 2, 0 +} +}; + +const opus_uint8 silk_delta_gain_iCDF[ MAX_DELTA_GAIN_QUANT - MIN_DELTA_GAIN_QUANT + 1 ] = { + 250, 245, 234, 203, 71, 50, 42, 38, + 35, 33, 31, 29, 28, 27, 26, 25, + 24, 23, 22, 21, 20, 19, 18, 17, + 16, 15, 14, 13, 12, 11, 10, 9, + 8, 7, 6, 5, 4, 3, 2, 1, + 0 +}; + +#ifdef __cplusplus +} +#endif diff --git a/media/libopus/silk/tables_other.c b/media/libopus/silk/tables_other.c new file mode 100644 index 0000000000..e34d90777b --- /dev/null +++ b/media/libopus/silk/tables_other.c @@ -0,0 +1,124 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "structs.h" +#include "define.h" +#include "tables.h" + +#ifdef __cplusplus +extern "C" +{ +#endif + +/* Tables for stereo predictor coding */ +const opus_int16 silk_stereo_pred_quant_Q13[ STEREO_QUANT_TAB_SIZE ] = { + -13732, -10050, -8266, -7526, -6500, -5000, -2950, -820, + 820, 2950, 5000, 6500, 7526, 8266, 10050, 13732 +}; +const opus_uint8 silk_stereo_pred_joint_iCDF[ 25 ] = { + 249, 247, 246, 245, 244, + 234, 210, 202, 201, 200, + 197, 174, 82, 59, 56, + 55, 54, 46, 22, 12, + 11, 10, 9, 7, 0 +}; +const opus_uint8 silk_stereo_only_code_mid_iCDF[ 2 ] = { 64, 0 }; + +/* Tables for LBRR flags */ +static const opus_uint8 silk_LBRR_flags_2_iCDF[ 3 ] = { 203, 150, 0 }; +static const opus_uint8 silk_LBRR_flags_3_iCDF[ 7 ] = { 215, 195, 166, 125, 110, 82, 0 }; +const opus_uint8 * const silk_LBRR_flags_iCDF_ptr[ 2 ] = { + silk_LBRR_flags_2_iCDF, + silk_LBRR_flags_3_iCDF +}; + +/* Table for LSB coding */ +const opus_uint8 silk_lsb_iCDF[ 2 ] = { 120, 0 }; + +/* Tables for LTPScale */ +const opus_uint8 silk_LTPscale_iCDF[ 3 ] = { 128, 64, 0 }; + +/* Tables for signal type and offset coding */ +const opus_uint8 silk_type_offset_VAD_iCDF[ 4 ] = { + 232, 158, 10, 0 +}; +const opus_uint8 silk_type_offset_no_VAD_iCDF[ 2 ] = { + 230, 0 +}; + +/* Tables for NLSF interpolation factor */ +const opus_uint8 silk_NLSF_interpolation_factor_iCDF[ 5 ] = { 243, 221, 192, 181, 0 }; + +/* Quantization offsets */ +const opus_int16 silk_Quantization_Offsets_Q10[ 2 ][ 2 ] = { + { OFFSET_UVL_Q10, OFFSET_UVH_Q10 }, { OFFSET_VL_Q10, OFFSET_VH_Q10 } +}; + +/* Table for LTPScale */ +const opus_int16 silk_LTPScales_table_Q14[ 3 ] = { 15565, 12288, 8192 }; + +/* Uniform entropy tables */ +const opus_uint8 silk_uniform3_iCDF[ 3 ] = { 171, 85, 0 }; +const opus_uint8 silk_uniform4_iCDF[ 4 ] = { 192, 128, 64, 0 }; +const opus_uint8 silk_uniform5_iCDF[ 5 ] = { 205, 154, 102, 51, 0 }; +const opus_uint8 silk_uniform6_iCDF[ 6 ] = { 213, 171, 128, 85, 43, 0 }; +const opus_uint8 silk_uniform8_iCDF[ 8 ] = { 224, 192, 160, 128, 96, 64, 32, 0 }; + +const opus_uint8 silk_NLSF_EXT_iCDF[ 7 ] = { 100, 40, 16, 7, 3, 1, 0 }; + +/* Elliptic/Cauer filters designed with 0.1 dB passband ripple, + 80 dB minimum stopband attenuation, and + [0.95 : 0.15 : 0.35] normalized cut off frequencies. */ + +/* Interpolation points for filter coefficients used in the bandwidth transition smoother */ +const opus_int32 silk_Transition_LP_B_Q28[ TRANSITION_INT_NUM ][ TRANSITION_NB ] = +{ +{ 250767114, 501534038, 250767114 }, +{ 209867381, 419732057, 209867381 }, +{ 170987846, 341967853, 170987846 }, +{ 131531482, 263046905, 131531482 }, +{ 89306658, 178584282, 89306658 } +}; + +/* Interpolation points for filter coefficients used in the bandwidth transition smoother */ +const opus_int32 silk_Transition_LP_A_Q28[ TRANSITION_INT_NUM ][ TRANSITION_NA ] = +{ +{ 506393414, 239854379 }, +{ 411067935, 169683996 }, +{ 306733530, 116694253 }, +{ 185807084, 77959395 }, +{ 35497197, 57401098 } +}; + +#ifdef __cplusplus +} +#endif + diff --git a/media/libopus/silk/tables_pitch_lag.c b/media/libopus/silk/tables_pitch_lag.c new file mode 100644 index 0000000000..e80cc59a27 --- /dev/null +++ b/media/libopus/silk/tables_pitch_lag.c @@ -0,0 +1,69 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "tables.h" + +const opus_uint8 silk_pitch_lag_iCDF[ 2 * ( PITCH_EST_MAX_LAG_MS - PITCH_EST_MIN_LAG_MS ) ] = { + 253, 250, 244, 233, 212, 182, 150, 131, + 120, 110, 98, 85, 72, 60, 49, 40, + 32, 25, 19, 15, 13, 11, 9, 8, + 7, 6, 5, 4, 3, 2, 1, 0 +}; + +const opus_uint8 silk_pitch_delta_iCDF[21] = { + 210, 208, 206, 203, 199, 193, 183, 168, + 142, 104, 74, 52, 37, 27, 20, 14, + 10, 6, 4, 2, 0 +}; + +const opus_uint8 silk_pitch_contour_iCDF[34] = { + 223, 201, 183, 167, 152, 138, 124, 111, + 98, 88, 79, 70, 62, 56, 50, 44, + 39, 35, 31, 27, 24, 21, 18, 16, + 14, 12, 10, 8, 6, 4, 3, 2, + 1, 0 +}; + +const opus_uint8 silk_pitch_contour_NB_iCDF[11] = { + 188, 176, 155, 138, 119, 97, 67, 43, + 26, 10, 0 +}; + +const opus_uint8 silk_pitch_contour_10_ms_iCDF[12] = { + 165, 119, 80, 61, 47, 35, 27, 20, + 14, 9, 4, 0 +}; + +const opus_uint8 silk_pitch_contour_10_ms_NB_iCDF[3] = { + 113, 63, 0 +}; + + diff --git a/media/libopus/silk/tables_pulses_per_block.c b/media/libopus/silk/tables_pulses_per_block.c new file mode 100644 index 0000000000..c7c01c8893 --- /dev/null +++ b/media/libopus/silk/tables_pulses_per_block.c @@ -0,0 +1,264 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "tables.h" + +const opus_uint8 silk_max_pulses_table[ 4 ] = { + 8, 10, 12, 16 +}; + +const opus_uint8 silk_pulses_per_block_iCDF[ 10 ][ 18 ] = { +{ + 125, 51, 26, 18, 15, 12, 11, 10, + 9, 8, 7, 6, 5, 4, 3, 2, + 1, 0 +}, +{ + 198, 105, 45, 22, 15, 12, 11, 10, + 9, 8, 7, 6, 5, 4, 3, 2, + 1, 0 +}, +{ + 213, 162, 116, 83, 59, 43, 32, 24, + 18, 15, 12, 9, 7, 6, 5, 3, + 2, 0 +}, +{ + 239, 187, 116, 59, 28, 16, 11, 10, + 9, 8, 7, 6, 5, 4, 3, 2, + 1, 0 +}, +{ + 250, 229, 188, 135, 86, 51, 30, 19, + 13, 10, 8, 6, 5, 4, 3, 2, + 1, 0 +}, +{ + 249, 235, 213, 185, 156, 128, 103, 83, + 66, 53, 42, 33, 26, 21, 17, 13, + 10, 0 +}, +{ + 254, 249, 235, 206, 164, 118, 77, 46, + 27, 16, 10, 7, 5, 4, 3, 2, + 1, 0 +}, +{ + 255, 253, 249, 239, 220, 191, 156, 119, + 85, 57, 37, 23, 15, 10, 6, 4, + 2, 0 +}, +{ + 255, 253, 251, 246, 237, 223, 203, 179, + 152, 124, 98, 75, 55, 40, 29, 21, + 15, 0 +}, +{ + 255, 254, 253, 247, 220, 162, 106, 67, + 42, 28, 18, 12, 9, 6, 4, 3, + 2, 0 +} +}; + +const opus_uint8 silk_pulses_per_block_BITS_Q5[ 9 ][ 18 ] = { +{ + 31, 57, 107, 160, 205, 205, 255, 255, + 255, 255, 255, 255, 255, 255, 255, 255, + 255, 255 +}, +{ + 69, 47, 67, 111, 166, 205, 255, 255, + 255, 255, 255, 255, 255, 255, 255, 255, + 255, 255 +}, +{ + 82, 74, 79, 95, 109, 128, 145, 160, + 173, 205, 205, 205, 224, 255, 255, 224, + 255, 224 +}, +{ + 125, 74, 59, 69, 97, 141, 182, 255, + 255, 255, 255, 255, 255, 255, 255, 255, + 255, 255 +}, +{ + 173, 115, 85, 73, 76, 92, 115, 145, + 173, 205, 224, 224, 255, 255, 255, 255, + 255, 255 +}, +{ + 166, 134, 113, 102, 101, 102, 107, 118, + 125, 138, 145, 155, 166, 182, 192, 192, + 205, 150 +}, +{ + 224, 182, 134, 101, 83, 79, 85, 97, + 120, 145, 173, 205, 224, 255, 255, 255, + 255, 255 +}, +{ + 255, 224, 192, 150, 120, 101, 92, 89, + 93, 102, 118, 134, 160, 182, 192, 224, + 224, 224 +}, +{ + 255, 224, 224, 182, 155, 134, 118, 109, + 104, 102, 106, 111, 118, 131, 145, 160, + 173, 131 +} +}; + +const opus_uint8 silk_rate_levels_iCDF[ 2 ][ 9 ] = +{ +{ + 241, 190, 178, 132, 87, 74, 41, 14, + 0 +}, +{ + 223, 193, 157, 140, 106, 57, 39, 18, + 0 +} +}; + +const opus_uint8 silk_rate_levels_BITS_Q5[ 2 ][ 9 ] = +{ +{ + 131, 74, 141, 79, 80, 138, 95, 104, + 134 +}, +{ + 95, 99, 91, 125, 93, 76, 123, 115, + 123 +} +}; + +const opus_uint8 silk_shell_code_table0[ 152 ] = { + 128, 0, 214, 42, 0, 235, 128, 21, + 0, 244, 184, 72, 11, 0, 248, 214, + 128, 42, 7, 0, 248, 225, 170, 80, + 25, 5, 0, 251, 236, 198, 126, 54, + 18, 3, 0, 250, 238, 211, 159, 82, + 35, 15, 5, 0, 250, 231, 203, 168, + 128, 88, 53, 25, 6, 0, 252, 238, + 216, 185, 148, 108, 71, 40, 18, 4, + 0, 253, 243, 225, 199, 166, 128, 90, + 57, 31, 13, 3, 0, 254, 246, 233, + 212, 183, 147, 109, 73, 44, 23, 10, + 2, 0, 255, 250, 240, 223, 198, 166, + 128, 90, 58, 33, 16, 6, 1, 0, + 255, 251, 244, 231, 210, 181, 146, 110, + 75, 46, 25, 12, 5, 1, 0, 255, + 253, 248, 238, 221, 196, 164, 128, 92, + 60, 35, 18, 8, 3, 1, 0, 255, + 253, 249, 242, 229, 208, 180, 146, 110, + 76, 48, 27, 14, 7, 3, 1, 0 +}; + +const opus_uint8 silk_shell_code_table1[ 152 ] = { + 129, 0, 207, 50, 0, 236, 129, 20, + 0, 245, 185, 72, 10, 0, 249, 213, + 129, 42, 6, 0, 250, 226, 169, 87, + 27, 4, 0, 251, 233, 194, 130, 62, + 20, 4, 0, 250, 236, 207, 160, 99, + 47, 17, 3, 0, 255, 240, 217, 182, + 131, 81, 41, 11, 1, 0, 255, 254, + 233, 201, 159, 107, 61, 20, 2, 1, + 0, 255, 249, 233, 206, 170, 128, 86, + 50, 23, 7, 1, 0, 255, 250, 238, + 217, 186, 148, 108, 70, 39, 18, 6, + 1, 0, 255, 252, 243, 226, 200, 166, + 128, 90, 56, 30, 13, 4, 1, 0, + 255, 252, 245, 231, 209, 180, 146, 110, + 76, 47, 25, 11, 4, 1, 0, 255, + 253, 248, 237, 219, 194, 163, 128, 93, + 62, 37, 19, 8, 3, 1, 0, 255, + 254, 250, 241, 226, 205, 177, 145, 111, + 79, 51, 30, 15, 6, 2, 1, 0 +}; + +const opus_uint8 silk_shell_code_table2[ 152 ] = { + 129, 0, 203, 54, 0, 234, 129, 23, + 0, 245, 184, 73, 10, 0, 250, 215, + 129, 41, 5, 0, 252, 232, 173, 86, + 24, 3, 0, 253, 240, 200, 129, 56, + 15, 2, 0, 253, 244, 217, 164, 94, + 38, 10, 1, 0, 253, 245, 226, 189, + 132, 71, 27, 7, 1, 0, 253, 246, + 231, 203, 159, 105, 56, 23, 6, 1, + 0, 255, 248, 235, 213, 179, 133, 85, + 47, 19, 5, 1, 0, 255, 254, 243, + 221, 194, 159, 117, 70, 37, 12, 2, + 1, 0, 255, 254, 248, 234, 208, 171, + 128, 85, 48, 22, 8, 2, 1, 0, + 255, 254, 250, 240, 220, 189, 149, 107, + 67, 36, 16, 6, 2, 1, 0, 255, + 254, 251, 243, 227, 201, 166, 128, 90, + 55, 29, 13, 5, 2, 1, 0, 255, + 254, 252, 246, 234, 213, 183, 147, 109, + 73, 43, 22, 10, 4, 2, 1, 0 +}; + +const opus_uint8 silk_shell_code_table3[ 152 ] = { + 130, 0, 200, 58, 0, 231, 130, 26, + 0, 244, 184, 76, 12, 0, 249, 214, + 130, 43, 6, 0, 252, 232, 173, 87, + 24, 3, 0, 253, 241, 203, 131, 56, + 14, 2, 0, 254, 246, 221, 167, 94, + 35, 8, 1, 0, 254, 249, 232, 193, + 130, 65, 23, 5, 1, 0, 255, 251, + 239, 211, 162, 99, 45, 15, 4, 1, + 0, 255, 251, 243, 223, 186, 131, 74, + 33, 11, 3, 1, 0, 255, 252, 245, + 230, 202, 158, 105, 57, 24, 8, 2, + 1, 0, 255, 253, 247, 235, 214, 179, + 132, 84, 44, 19, 7, 2, 1, 0, + 255, 254, 250, 240, 223, 196, 159, 112, + 69, 36, 15, 6, 2, 1, 0, 255, + 254, 253, 245, 231, 209, 176, 136, 93, + 55, 27, 11, 3, 2, 1, 0, 255, + 254, 253, 252, 239, 221, 194, 158, 117, + 76, 42, 18, 4, 3, 2, 1, 0 +}; + +const opus_uint8 silk_shell_code_table_offsets[ 17 ] = { + 0, 0, 2, 5, 9, 14, 20, 27, + 35, 44, 54, 65, 77, 90, 104, 119, + 135 +}; + +const opus_uint8 silk_sign_iCDF[ 42 ] = { + 254, 49, 67, 77, 82, 93, 99, + 198, 11, 18, 24, 31, 36, 45, + 255, 46, 66, 78, 87, 94, 104, + 208, 14, 21, 32, 42, 51, 66, + 255, 94, 104, 109, 112, 115, 118, + 248, 53, 69, 80, 88, 95, 102 +}; diff --git a/media/libopus/silk/tuning_parameters.h b/media/libopus/silk/tuning_parameters.h new file mode 100644 index 0000000000..d70275fd8f --- /dev/null +++ b/media/libopus/silk/tuning_parameters.h @@ -0,0 +1,155 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef SILK_TUNING_PARAMETERS_H +#define SILK_TUNING_PARAMETERS_H + +#ifdef __cplusplus +extern "C" +{ +#endif + +/* Decay time for bitreservoir */ +#define BITRESERVOIR_DECAY_TIME_MS 500 + +/*******************/ +/* Pitch estimator */ +/*******************/ + +/* Level of noise floor for whitening filter LPC analysis in pitch analysis */ +#define FIND_PITCH_WHITE_NOISE_FRACTION 1e-3f + +/* Bandwidth expansion for whitening filter in pitch analysis */ +#define FIND_PITCH_BANDWIDTH_EXPANSION 0.99f + +/*********************/ +/* Linear prediction */ +/*********************/ + +/* LPC analysis regularization */ +#define FIND_LPC_COND_FAC 1e-5f + +/* Max cumulative LTP gain */ +#define MAX_SUM_LOG_GAIN_DB 250.0f + +/* LTP analysis defines */ +#define LTP_CORR_INV_MAX 0.03f + +/***********************/ +/* High pass filtering */ +/***********************/ + +/* Smoothing parameters for low end of pitch frequency range estimation */ +#define VARIABLE_HP_SMTH_COEF1 0.1f +#define VARIABLE_HP_SMTH_COEF2 0.015f +#define VARIABLE_HP_MAX_DELTA_FREQ 0.4f + +/* Min and max cut-off frequency values (-3 dB points) */ +#define VARIABLE_HP_MIN_CUTOFF_HZ 60 +#define VARIABLE_HP_MAX_CUTOFF_HZ 100 + +/***********/ +/* Various */ +/***********/ + +/* VAD threshold */ +#define SPEECH_ACTIVITY_DTX_THRES 0.05f + +/* Speech Activity LBRR enable threshold */ +#define LBRR_SPEECH_ACTIVITY_THRES 0.3f + +/*************************/ +/* Perceptual parameters */ +/*************************/ + +/* reduction in coding SNR during low speech activity */ +#define BG_SNR_DECR_dB 2.0f + +/* factor for reducing quantization noise during voiced speech */ +#define HARM_SNR_INCR_dB 2.0f + +/* factor for reducing quantization noise for unvoiced sparse signals */ +#define SPARSE_SNR_INCR_dB 2.0f + +/* threshold for sparseness measure above which to use lower quantization offset during unvoiced */ +#define ENERGY_VARIATION_THRESHOLD_QNT_OFFSET 0.6f + +/* warping control */ +#define WARPING_MULTIPLIER 0.015f + +/* fraction added to first autocorrelation value */ +#define SHAPE_WHITE_NOISE_FRACTION 3e-5f + +/* noise shaping filter chirp factor */ +#define BANDWIDTH_EXPANSION 0.94f + +/* harmonic noise shaping */ +#define HARMONIC_SHAPING 0.3f + +/* extra harmonic noise shaping for high bitrates or noisy input */ +#define HIGH_RATE_OR_LOW_QUALITY_HARMONIC_SHAPING 0.2f + +/* parameter for shaping noise towards higher frequencies */ +#define HP_NOISE_COEF 0.25f + +/* parameter for shaping noise even more towards higher frequencies during voiced speech */ +#define HARM_HP_NOISE_COEF 0.35f + +/* parameter for applying a high-pass tilt to the input signal */ +#define INPUT_TILT 0.05f + +/* parameter for extra high-pass tilt to the input signal at high rates */ +#define HIGH_RATE_INPUT_TILT 0.1f + +/* parameter for reducing noise at the very low frequencies */ +#define LOW_FREQ_SHAPING 4.0f + +/* less reduction of noise at the very low frequencies for signals with low SNR at low frequencies */ +#define LOW_QUALITY_LOW_FREQ_SHAPING_DECR 0.5f + +/* subframe smoothing coefficient for HarmBoost, HarmShapeGain, Tilt (lower -> more smoothing) */ +#define SUBFR_SMTH_COEF 0.4f + +/* parameters defining the R/D tradeoff in the residual quantizer */ +#define LAMBDA_OFFSET 1.2f +#define LAMBDA_SPEECH_ACT -0.2f +#define LAMBDA_DELAYED_DECISIONS -0.05f +#define LAMBDA_INPUT_QUALITY -0.1f +#define LAMBDA_CODING_QUALITY -0.2f +#define LAMBDA_QUANT_OFFSET 0.8f + +/* Compensation in bitrate calculations for 10 ms modes */ +#define REDUCE_BITRATE_10_MS_BPS 2200 + +/* Maximum time before allowing a bandwidth transition */ +#define MAX_BANDWIDTH_SWITCH_DELAY_MS 5000 + +#ifdef __cplusplus +} +#endif + +#endif /* SILK_TUNING_PARAMETERS_H */ diff --git a/media/libopus/silk/typedef.h b/media/libopus/silk/typedef.h new file mode 100644 index 0000000000..793d2c0c1d --- /dev/null +++ b/media/libopus/silk/typedef.h @@ -0,0 +1,81 @@ +/*********************************************************************** +Copyright (c) 2006-2011, Skype Limited. All rights reserved. +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions +are met: +- Redistributions of source code must retain the above copyright notice, +this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright +notice, this list of conditions and the following disclaimer in the +documentation and/or other materials provided with the distribution. +- Neither the name of Internet Society, IETF or IETF Trust, nor the +names of specific contributors, may be used to endorse or promote +products derived from this software without specific prior written +permission. +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. +***********************************************************************/ + +#ifndef SILK_TYPEDEF_H +#define SILK_TYPEDEF_H + +#include "opus_types.h" +#include "opus_defines.h" + +#ifndef FIXED_POINT +# include <float.h> +# define silk_float float +# define silk_float_MAX FLT_MAX +#endif + +#define silk_int64_MAX ((opus_int64)0x7FFFFFFFFFFFFFFFLL) /* 2^63 - 1 */ +#define silk_int64_MIN ((opus_int64)0x8000000000000000LL) /* -2^63 */ +#define silk_int32_MAX 0x7FFFFFFF /* 2^31 - 1 = 2147483647 */ +#define silk_int32_MIN ((opus_int32)0x80000000) /* -2^31 = -2147483648 */ +#define silk_int16_MAX 0x7FFF /* 2^15 - 1 = 32767 */ +#define silk_int16_MIN ((opus_int16)0x8000) /* -2^15 = -32768 */ +#define silk_int8_MAX 0x7F /* 2^7 - 1 = 127 */ +#define silk_int8_MIN ((opus_int8)0x80) /* -2^7 = -128 */ +#define silk_uint8_MAX 0xFF /* 2^8 - 1 = 255 */ + +#define silk_TRUE 1 +#define silk_FALSE 0 + +/* assertions */ +#if (defined _WIN32 && !defined _WINCE && !defined(__GNUC__) && !defined(NO_ASSERTS)) +# ifndef silk_assert +# include <crtdbg.h> /* ASSERTE() */ +# define silk_assert(COND) _ASSERTE(COND) +# endif +#else +# ifdef ENABLE_ASSERTIONS +# include <stdio.h> +# include <stdlib.h> +#define silk_fatal(str) _silk_fatal(str, __FILE__, __LINE__); +#ifdef __GNUC__ +__attribute__((noreturn)) +#endif +static OPUS_INLINE void _silk_fatal(const char *str, const char *file, int line) +{ + fprintf (stderr, "Fatal (internal) error in %s, line %d: %s\n", file, line, str); +#if defined(_MSC_VER) + _set_abort_behavior( 0, _WRITE_ABORT_MSG); +#endif + abort(); +} +# define silk_assert(COND) {if (!(COND)) {silk_fatal("assertion failed: " #COND);}} +# else +# define silk_assert(COND) +# endif +#endif + +#endif /* SILK_TYPEDEF_H */ diff --git a/media/libopus/silk/x86/NSQ_del_dec_sse4_1.c b/media/libopus/silk/x86/NSQ_del_dec_sse4_1.c new file mode 100644 index 0000000000..a58a76cd8d --- /dev/null +++ b/media/libopus/silk/x86/NSQ_del_dec_sse4_1.c @@ -0,0 +1,915 @@ +/* Copyright (c) 2014-2020, Cisco Systems, INC + Written by XiangMingZhu WeiZhou MinPeng YanWang FrancisQuiers + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include <xmmintrin.h> +#include <emmintrin.h> +#include <smmintrin.h> +#include "main.h" +#include "celt/x86/x86cpu.h" + +#include "stack_alloc.h" + +typedef struct { + opus_int32 sLPC_Q14[ MAX_SUB_FRAME_LENGTH + NSQ_LPC_BUF_LENGTH ]; + opus_int32 RandState[ DECISION_DELAY ]; + opus_int32 Q_Q10[ DECISION_DELAY ]; + opus_int32 Xq_Q14[ DECISION_DELAY ]; + opus_int32 Pred_Q15[ DECISION_DELAY ]; + opus_int32 Shape_Q14[ DECISION_DELAY ]; + opus_int32 sAR2_Q14[ MAX_SHAPE_LPC_ORDER ]; + opus_int32 LF_AR_Q14; + opus_int32 Diff_Q14; + opus_int32 Seed; + opus_int32 SeedInit; + opus_int32 RD_Q10; +} NSQ_del_dec_struct; + +typedef struct { + opus_int32 Q_Q10; + opus_int32 RD_Q10; + opus_int32 xq_Q14; + opus_int32 LF_AR_Q14; + opus_int32 Diff_Q14; + opus_int32 sLTP_shp_Q14; + opus_int32 LPC_exc_Q14; +} NSQ_sample_struct; + +typedef NSQ_sample_struct NSQ_sample_pair[ 2 ]; + +static OPUS_INLINE void silk_nsq_del_dec_scale_states_sse4_1( + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + NSQ_del_dec_struct psDelDec[], /* I/O Delayed decision states */ + const opus_int16 x16[], /* I Input */ + opus_int32 x_sc_Q10[], /* O Input scaled with 1/Gain in Q10 */ + const opus_int16 sLTP[], /* I Re-whitened LTP state in Q0 */ + opus_int32 sLTP_Q15[], /* O LTP state matching scaled input */ + opus_int subfr, /* I Subframe number */ + opus_int nStatesDelayedDecision, /* I Number of del dec states */ + const opus_int LTP_scale_Q14, /* I LTP state scaling */ + const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lag */ + const opus_int signal_type, /* I Signal type */ + const opus_int decisionDelay /* I Decision delay */ +); + +/******************************************/ +/* Noise shape quantizer for one subframe */ +/******************************************/ +static OPUS_INLINE void silk_noise_shape_quantizer_del_dec_sse4_1( + silk_nsq_state *NSQ, /* I/O NSQ state */ + NSQ_del_dec_struct psDelDec[], /* I/O Delayed decision states */ + opus_int signalType, /* I Signal type */ + const opus_int32 x_Q10[], /* I */ + opus_int8 pulses[], /* O */ + opus_int16 xq[], /* O */ + opus_int32 sLTP_Q15[], /* I/O LTP filter state */ + opus_int32 delayedGain_Q10[], /* I/O Gain delay buffer */ + const opus_int16 a_Q12[], /* I Short term prediction coefs */ + const opus_int16 b_Q14[], /* I Long term prediction coefs */ + const opus_int16 AR_shp_Q13[], /* I Noise shaping coefs */ + opus_int lag, /* I Pitch lag */ + opus_int32 HarmShapeFIRPacked_Q14, /* I */ + opus_int Tilt_Q14, /* I Spectral tilt */ + opus_int32 LF_shp_Q14, /* I */ + opus_int32 Gain_Q16, /* I */ + opus_int Lambda_Q10, /* I */ + opus_int offset_Q10, /* I */ + opus_int length, /* I Input length */ + opus_int subfr, /* I Subframe number */ + opus_int shapingLPCOrder, /* I Shaping LPC filter order */ + opus_int predictLPCOrder, /* I Prediction filter order */ + opus_int warping_Q16, /* I */ + opus_int nStatesDelayedDecision, /* I Number of states in decision tree */ + opus_int *smpl_buf_idx, /* I/O Index to newest samples in buffers */ + opus_int decisionDelay /* I */ +); + +void silk_NSQ_del_dec_sse4_1( + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + SideInfoIndices *psIndices, /* I/O Quantization Indices */ + const opus_int16 x16[], /* I Input */ + opus_int8 pulses[], /* O Quantized pulse signal */ + const opus_int16 PredCoef_Q12[ 2 * MAX_LPC_ORDER ], /* I Short term prediction coefs */ + const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ + const opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ + const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ + const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ + const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ + const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ + const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ + const opus_int LTP_scale_Q14 /* I LTP state scaling */ +) +{ + opus_int i, k, lag, start_idx, LSF_interpolation_flag, Winner_ind, subfr; + opus_int last_smple_idx, smpl_buf_idx, decisionDelay; + const opus_int16 *A_Q12, *B_Q14, *AR_shp_Q13; + opus_int16 *pxq; + VARDECL( opus_int32, sLTP_Q15 ); + VARDECL( opus_int16, sLTP ); + opus_int32 HarmShapeFIRPacked_Q14; + opus_int offset_Q10; + opus_int32 RDmin_Q10, Gain_Q10; + VARDECL( opus_int32, x_sc_Q10 ); + VARDECL( opus_int32, delayedGain_Q10 ); + VARDECL( NSQ_del_dec_struct, psDelDec ); + NSQ_del_dec_struct *psDD; +#ifdef OPUS_CHECK_ASM + silk_nsq_state NSQ_c; + SideInfoIndices psIndices_c; + opus_int8 pulses_c[ MAX_FRAME_LENGTH ]; + const opus_int8 *const pulses_a = pulses; +#endif + SAVE_STACK; + +#ifdef OPUS_CHECK_ASM + ( void )pulses_a; + silk_memcpy( &NSQ_c, NSQ, sizeof( NSQ_c ) ); + silk_memcpy( &psIndices_c, psIndices, sizeof( psIndices_c ) ); + silk_assert( psEncC->nb_subfr * psEncC->subfr_length <= MAX_FRAME_LENGTH ); + silk_memcpy( pulses_c, pulses, psEncC->nb_subfr * psEncC->subfr_length * sizeof( pulses[0] ) ); + silk_NSQ_del_dec_c( + psEncC, + &NSQ_c, + &psIndices_c, + x16, + pulses_c, + PredCoef_Q12, + LTPCoef_Q14, + AR_Q13, + HarmShapeGain_Q14, + Tilt_Q14, + LF_shp_Q14, + Gains_Q16, + pitchL, + Lambda_Q10, + LTP_scale_Q14 + ); +#endif + + /* Set unvoiced lag to the previous one, overwrite later for voiced */ + lag = NSQ->lagPrev; + + silk_assert( NSQ->prev_gain_Q16 != 0 ); + + /* Initialize delayed decision states */ + ALLOC( psDelDec, psEncC->nStatesDelayedDecision, NSQ_del_dec_struct ); + silk_memset( psDelDec, 0, psEncC->nStatesDelayedDecision * sizeof( NSQ_del_dec_struct ) ); + for( k = 0; k < psEncC->nStatesDelayedDecision; k++ ) { + psDD = &psDelDec[ k ]; + psDD->Seed = ( k + psIndices->Seed ) & 3; + psDD->SeedInit = psDD->Seed; + psDD->RD_Q10 = 0; + psDD->LF_AR_Q14 = NSQ->sLF_AR_shp_Q14; + psDD->Diff_Q14 = NSQ->sDiff_shp_Q14; + psDD->Shape_Q14[ 0 ] = NSQ->sLTP_shp_Q14[ psEncC->ltp_mem_length - 1 ]; + silk_memcpy( psDD->sLPC_Q14, NSQ->sLPC_Q14, NSQ_LPC_BUF_LENGTH * sizeof( opus_int32 ) ); + silk_memcpy( psDD->sAR2_Q14, NSQ->sAR2_Q14, sizeof( NSQ->sAR2_Q14 ) ); + } + + offset_Q10 = silk_Quantization_Offsets_Q10[ psIndices->signalType >> 1 ][ psIndices->quantOffsetType ]; + smpl_buf_idx = 0; /* index of oldest samples */ + + decisionDelay = silk_min_int( DECISION_DELAY, psEncC->subfr_length ); + + /* For voiced frames limit the decision delay to lower than the pitch lag */ + if( psIndices->signalType == TYPE_VOICED ) { + for( k = 0; k < psEncC->nb_subfr; k++ ) { + decisionDelay = silk_min_int( decisionDelay, pitchL[ k ] - LTP_ORDER / 2 - 1 ); + } + } else { + if( lag > 0 ) { + decisionDelay = silk_min_int( decisionDelay, lag - LTP_ORDER / 2 - 1 ); + } + } + + if( psIndices->NLSFInterpCoef_Q2 == 4 ) { + LSF_interpolation_flag = 0; + } else { + LSF_interpolation_flag = 1; + } + + ALLOC( sLTP_Q15, psEncC->ltp_mem_length + psEncC->frame_length, opus_int32 ); + ALLOC( sLTP, psEncC->ltp_mem_length + psEncC->frame_length, opus_int16 ); + ALLOC( x_sc_Q10, psEncC->subfr_length, opus_int32 ); + ALLOC( delayedGain_Q10, DECISION_DELAY, opus_int32 ); + /* Set up pointers to start of sub frame */ + pxq = &NSQ->xq[ psEncC->ltp_mem_length ]; + NSQ->sLTP_shp_buf_idx = psEncC->ltp_mem_length; + NSQ->sLTP_buf_idx = psEncC->ltp_mem_length; + subfr = 0; + for( k = 0; k < psEncC->nb_subfr; k++ ) { + A_Q12 = &PredCoef_Q12[ ( ( k >> 1 ) | ( 1 - LSF_interpolation_flag ) ) * MAX_LPC_ORDER ]; + B_Q14 = <PCoef_Q14[ k * LTP_ORDER ]; + AR_shp_Q13 = &AR_Q13[ k * MAX_SHAPE_LPC_ORDER ]; + + /* Noise shape parameters */ + silk_assert( HarmShapeGain_Q14[ k ] >= 0 ); + HarmShapeFIRPacked_Q14 = silk_RSHIFT( HarmShapeGain_Q14[ k ], 2 ); + HarmShapeFIRPacked_Q14 |= silk_LSHIFT( (opus_int32)silk_RSHIFT( HarmShapeGain_Q14[ k ], 1 ), 16 ); + + NSQ->rewhite_flag = 0; + if( psIndices->signalType == TYPE_VOICED ) { + /* Voiced */ + lag = pitchL[ k ]; + + /* Re-whitening */ + if( ( k & ( 3 - silk_LSHIFT( LSF_interpolation_flag, 1 ) ) ) == 0 ) { + if( k == 2 ) { + /* RESET DELAYED DECISIONS */ + /* Find winner */ + RDmin_Q10 = psDelDec[ 0 ].RD_Q10; + Winner_ind = 0; + for( i = 1; i < psEncC->nStatesDelayedDecision; i++ ) { + if( psDelDec[ i ].RD_Q10 < RDmin_Q10 ) { + RDmin_Q10 = psDelDec[ i ].RD_Q10; + Winner_ind = i; + } + } + for( i = 0; i < psEncC->nStatesDelayedDecision; i++ ) { + if( i != Winner_ind ) { + psDelDec[ i ].RD_Q10 += ( silk_int32_MAX >> 4 ); + silk_assert( psDelDec[ i ].RD_Q10 >= 0 ); + } + } + + /* Copy final part of signals from winner state to output and long-term filter states */ + psDD = &psDelDec[ Winner_ind ]; + last_smple_idx = smpl_buf_idx + decisionDelay; + for( i = 0; i < decisionDelay; i++ ) { + last_smple_idx = ( last_smple_idx - 1 ) % DECISION_DELAY; + if( last_smple_idx < 0 ) last_smple_idx += DECISION_DELAY; + pulses[ i - decisionDelay ] = (opus_int8)silk_RSHIFT_ROUND( psDD->Q_Q10[ last_smple_idx ], 10 ); + pxq[ i - decisionDelay ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( + silk_SMULWW( psDD->Xq_Q14[ last_smple_idx ], Gains_Q16[ 1 ] ), 14 ) ); + NSQ->sLTP_shp_Q14[ NSQ->sLTP_shp_buf_idx - decisionDelay + i ] = psDD->Shape_Q14[ last_smple_idx ]; + } + + subfr = 0; + } + + /* Rewhiten with new A coefs */ + start_idx = psEncC->ltp_mem_length - lag - psEncC->predictLPCOrder - LTP_ORDER / 2; + celt_assert( start_idx > 0 ); + + silk_LPC_analysis_filter( &sLTP[ start_idx ], &NSQ->xq[ start_idx + k * psEncC->subfr_length ], + A_Q12, psEncC->ltp_mem_length - start_idx, psEncC->predictLPCOrder, psEncC->arch ); + + NSQ->sLTP_buf_idx = psEncC->ltp_mem_length; + NSQ->rewhite_flag = 1; + } + } + + silk_nsq_del_dec_scale_states_sse4_1( psEncC, NSQ, psDelDec, x16, x_sc_Q10, sLTP, sLTP_Q15, k, + psEncC->nStatesDelayedDecision, LTP_scale_Q14, Gains_Q16, pitchL, psIndices->signalType, decisionDelay ); + + silk_noise_shape_quantizer_del_dec_sse4_1( NSQ, psDelDec, psIndices->signalType, x_sc_Q10, pulses, pxq, sLTP_Q15, + delayedGain_Q10, A_Q12, B_Q14, AR_shp_Q13, lag, HarmShapeFIRPacked_Q14, Tilt_Q14[ k ], LF_shp_Q14[ k ], + Gains_Q16[ k ], Lambda_Q10, offset_Q10, psEncC->subfr_length, subfr++, psEncC->shapingLPCOrder, + psEncC->predictLPCOrder, psEncC->warping_Q16, psEncC->nStatesDelayedDecision, &smpl_buf_idx, decisionDelay ); + + x16 += psEncC->subfr_length; + pulses += psEncC->subfr_length; + pxq += psEncC->subfr_length; + } + + /* Find winner */ + RDmin_Q10 = psDelDec[ 0 ].RD_Q10; + Winner_ind = 0; + for( k = 1; k < psEncC->nStatesDelayedDecision; k++ ) { + if( psDelDec[ k ].RD_Q10 < RDmin_Q10 ) { + RDmin_Q10 = psDelDec[ k ].RD_Q10; + Winner_ind = k; + } + } + + /* Copy final part of signals from winner state to output and long-term filter states */ + psDD = &psDelDec[ Winner_ind ]; + psIndices->Seed = psDD->SeedInit; + last_smple_idx = smpl_buf_idx + decisionDelay; + Gain_Q10 = silk_RSHIFT32( Gains_Q16[ psEncC->nb_subfr - 1 ], 6 ); + for( i = 0; i < decisionDelay; i++ ) { + last_smple_idx = ( last_smple_idx - 1 ) % DECISION_DELAY; + if( last_smple_idx < 0 ) last_smple_idx += DECISION_DELAY; + + pulses[ i - decisionDelay ] = (opus_int8)silk_RSHIFT_ROUND( psDD->Q_Q10[ last_smple_idx ], 10 ); + pxq[ i - decisionDelay ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( + silk_SMULWW( psDD->Xq_Q14[ last_smple_idx ], Gain_Q10 ), 8 ) ); + NSQ->sLTP_shp_Q14[ NSQ->sLTP_shp_buf_idx - decisionDelay + i ] = psDD->Shape_Q14[ last_smple_idx ]; + } + silk_memcpy( NSQ->sLPC_Q14, &psDD->sLPC_Q14[ psEncC->subfr_length ], NSQ_LPC_BUF_LENGTH * sizeof( opus_int32 ) ); + silk_memcpy( NSQ->sAR2_Q14, psDD->sAR2_Q14, sizeof( psDD->sAR2_Q14 ) ); + + /* Update states */ + NSQ->sLF_AR_shp_Q14 = psDD->LF_AR_Q14; + NSQ->sDiff_shp_Q14 = psDD->Diff_Q14; + NSQ->lagPrev = pitchL[ psEncC->nb_subfr - 1 ]; + + /* Save quantized speech signal */ + silk_memmove( NSQ->xq, &NSQ->xq[ psEncC->frame_length ], psEncC->ltp_mem_length * sizeof( opus_int16 ) ); + silk_memmove( NSQ->sLTP_shp_Q14, &NSQ->sLTP_shp_Q14[ psEncC->frame_length ], psEncC->ltp_mem_length * sizeof( opus_int32 ) ); + +#ifdef OPUS_CHECK_ASM + silk_assert( !memcmp( &NSQ_c, NSQ, sizeof( NSQ_c ) ) ); + silk_assert( !memcmp( &psIndices_c, psIndices, sizeof( psIndices_c ) ) ); + silk_assert( !memcmp( pulses_c, pulses_a, psEncC->nb_subfr * psEncC->subfr_length * sizeof( pulses[0] ) ) ); +#endif + + RESTORE_STACK; +} + +/******************************************/ +/* Noise shape quantizer for one subframe */ +/******************************************/ +static OPUS_INLINE void silk_noise_shape_quantizer_del_dec_sse4_1( + silk_nsq_state *NSQ, /* I/O NSQ state */ + NSQ_del_dec_struct psDelDec[], /* I/O Delayed decision states */ + opus_int signalType, /* I Signal type */ + const opus_int32 x_Q10[], /* I */ + opus_int8 pulses[], /* O */ + opus_int16 xq[], /* O */ + opus_int32 sLTP_Q15[], /* I/O LTP filter state */ + opus_int32 delayedGain_Q10[], /* I/O Gain delay buffer */ + const opus_int16 a_Q12[], /* I Short term prediction coefs */ + const opus_int16 b_Q14[], /* I Long term prediction coefs */ + const opus_int16 AR_shp_Q13[], /* I Noise shaping coefs */ + opus_int lag, /* I Pitch lag */ + opus_int32 HarmShapeFIRPacked_Q14, /* I */ + opus_int Tilt_Q14, /* I Spectral tilt */ + opus_int32 LF_shp_Q14, /* I */ + opus_int32 Gain_Q16, /* I */ + opus_int Lambda_Q10, /* I */ + opus_int offset_Q10, /* I */ + opus_int length, /* I Input length */ + opus_int subfr, /* I Subframe number */ + opus_int shapingLPCOrder, /* I Shaping LPC filter order */ + opus_int predictLPCOrder, /* I Prediction filter order */ + opus_int warping_Q16, /* I */ + opus_int nStatesDelayedDecision, /* I Number of states in decision tree */ + opus_int *smpl_buf_idx, /* I/O Index to newest samples in buffers */ + opus_int decisionDelay /* I */ +) +{ + opus_int i, j, k, Winner_ind, RDmin_ind, RDmax_ind, last_smple_idx; + opus_int32 Winner_rand_state; + opus_int32 LTP_pred_Q14, LPC_pred_Q14, n_AR_Q14, n_LTP_Q14; + opus_int32 n_LF_Q14, r_Q10, rr_Q10, rd1_Q10, rd2_Q10, RDmin_Q10, RDmax_Q10; + opus_int32 q1_Q0, q1_Q10, q2_Q10, exc_Q14, LPC_exc_Q14, xq_Q14, Gain_Q10; + opus_int32 tmp1, tmp2, sLF_AR_shp_Q14; + opus_int32 *pred_lag_ptr, *shp_lag_ptr, *psLPC_Q14; + int rdo_offset; + + VARDECL( NSQ_sample_pair, psSampleState ); + NSQ_del_dec_struct *psDD; + NSQ_sample_struct *psSS; + + __m128i a_Q12_0123, a_Q12_4567, a_Q12_89AB, a_Q12_CDEF; + __m128i b_Q12_0123, b_sr_Q12_0123; + SAVE_STACK; + + celt_assert( nStatesDelayedDecision > 0 ); + ALLOC( psSampleState, nStatesDelayedDecision, NSQ_sample_pair ); + + rdo_offset = (Lambda_Q10 >> 1) - 512; + + shp_lag_ptr = &NSQ->sLTP_shp_Q14[ NSQ->sLTP_shp_buf_idx - lag + HARM_SHAPE_FIR_TAPS / 2 ]; + pred_lag_ptr = &sLTP_Q15[ NSQ->sLTP_buf_idx - lag + LTP_ORDER / 2 ]; + Gain_Q10 = silk_RSHIFT( Gain_Q16, 6 ); + + a_Q12_0123 = OP_CVTEPI16_EPI32_M64( a_Q12 ); + a_Q12_4567 = OP_CVTEPI16_EPI32_M64( a_Q12 + 4 ); + + if( opus_likely( predictLPCOrder == 16 ) ) { + a_Q12_89AB = OP_CVTEPI16_EPI32_M64( a_Q12 + 8 ); + a_Q12_CDEF = OP_CVTEPI16_EPI32_M64( a_Q12 + 12 ); + } + + if( signalType == TYPE_VOICED ){ + b_Q12_0123 = OP_CVTEPI16_EPI32_M64( b_Q14 ); + b_sr_Q12_0123 = _mm_shuffle_epi32( b_Q12_0123, _MM_SHUFFLE( 0, 3, 2, 1 ) ); /* equal shift right 4 bytes */ + } + for( i = 0; i < length; i++ ) { + /* Perform common calculations used in all states */ + + /* Long-term prediction */ + if( signalType == TYPE_VOICED ) { + /* Unrolled loop */ + /* Avoids introducing a bias because silk_SMLAWB() always rounds to -inf */ + LTP_pred_Q14 = 2; + { + __m128i tmpa, tmpb, pred_lag_ptr_tmp; + pred_lag_ptr_tmp = _mm_loadu_si128( (__m128i *)(&pred_lag_ptr[ -3 ] ) ); + pred_lag_ptr_tmp = _mm_shuffle_epi32( pred_lag_ptr_tmp, 0x1B ); + tmpa = _mm_mul_epi32( pred_lag_ptr_tmp, b_Q12_0123 ); + tmpa = _mm_srli_si128( tmpa, 2 ); + + pred_lag_ptr_tmp = _mm_shuffle_epi32( pred_lag_ptr_tmp, _MM_SHUFFLE( 0, 3, 2, 1 ) );/* equal shift right 4 bytes */ + pred_lag_ptr_tmp = _mm_mul_epi32( pred_lag_ptr_tmp, b_sr_Q12_0123 ); + pred_lag_ptr_tmp = _mm_srli_si128( pred_lag_ptr_tmp, 2 ); + pred_lag_ptr_tmp = _mm_add_epi32( pred_lag_ptr_tmp, tmpa ); + + tmpb = _mm_shuffle_epi32( pred_lag_ptr_tmp, _MM_SHUFFLE( 0, 0, 3, 2 ) );/* equal shift right 8 bytes */ + pred_lag_ptr_tmp = _mm_add_epi32( pred_lag_ptr_tmp, tmpb ); + LTP_pred_Q14 += _mm_cvtsi128_si32( pred_lag_ptr_tmp ); + + LTP_pred_Q14 = silk_SMLAWB( LTP_pred_Q14, pred_lag_ptr[ -4 ], b_Q14[ 4 ] ); + LTP_pred_Q14 = silk_LSHIFT( LTP_pred_Q14, 1 ); /* Q13 -> Q14 */ + pred_lag_ptr++; + } + } else { + LTP_pred_Q14 = 0; + } + + /* Long-term shaping */ + if( lag > 0 ) { + /* Symmetric, packed FIR coefficients */ + n_LTP_Q14 = silk_SMULWB( silk_ADD_SAT32( shp_lag_ptr[ 0 ], shp_lag_ptr[ -2 ] ), HarmShapeFIRPacked_Q14 ); + n_LTP_Q14 = silk_SMLAWT( n_LTP_Q14, shp_lag_ptr[ -1 ], HarmShapeFIRPacked_Q14 ); + n_LTP_Q14 = silk_SUB_LSHIFT32( LTP_pred_Q14, n_LTP_Q14, 2 ); /* Q12 -> Q14 */ + shp_lag_ptr++; + } else { + n_LTP_Q14 = 0; + } + { + __m128i tmpa, tmpb, psLPC_Q14_tmp, a_Q12_tmp; + + for( k = 0; k < nStatesDelayedDecision; k++ ) { + /* Delayed decision state */ + psDD = &psDelDec[ k ]; + + /* Sample state */ + psSS = psSampleState[ k ]; + + /* Generate dither */ + psDD->Seed = silk_RAND( psDD->Seed ); + + /* Pointer used in short term prediction and shaping */ + psLPC_Q14 = &psDD->sLPC_Q14[ NSQ_LPC_BUF_LENGTH - 1 + i ]; + /* Short-term prediction */ + silk_assert( predictLPCOrder == 10 || predictLPCOrder == 16 ); + /* Avoids introducing a bias because silk_SMLAWB() always rounds to -inf */ + LPC_pred_Q14 = silk_RSHIFT( predictLPCOrder, 1 ); + + tmpb = _mm_setzero_si128(); + + /* step 1 */ + psLPC_Q14_tmp = _mm_loadu_si128( (__m128i *)(&psLPC_Q14[ -3 ] ) ); /* -3, -2 , -1, 0 */ + psLPC_Q14_tmp = _mm_shuffle_epi32( psLPC_Q14_tmp, 0x1B ); /* 0, -1, -2, -3 */ + tmpa = _mm_mul_epi32( psLPC_Q14_tmp, a_Q12_0123 ); /* 0, -1, -2, -3 * 0123 -> 0*0, 2*-2 */ + + tmpa = _mm_srli_epi64( tmpa, 16 ); + tmpb = _mm_add_epi32( tmpb, tmpa ); + + psLPC_Q14_tmp = _mm_shuffle_epi32( psLPC_Q14_tmp, _MM_SHUFFLE( 0, 3, 2, 1 ) ); /* equal shift right 4 bytes */ + a_Q12_tmp = _mm_shuffle_epi32( a_Q12_0123, _MM_SHUFFLE(0, 3, 2, 1 ) ); /* equal shift right 4 bytes */ + psLPC_Q14_tmp = _mm_mul_epi32( psLPC_Q14_tmp, a_Q12_tmp ); /* 1*-1, 3*-3 */ + psLPC_Q14_tmp = _mm_srli_epi64( psLPC_Q14_tmp, 16 ); + tmpb = _mm_add_epi32( tmpb, psLPC_Q14_tmp ); + + /* step 2 */ + psLPC_Q14_tmp = _mm_loadu_si128( (__m128i *)(&psLPC_Q14[ -7 ] ) ); + psLPC_Q14_tmp = _mm_shuffle_epi32( psLPC_Q14_tmp, 0x1B ); + tmpa = _mm_mul_epi32( psLPC_Q14_tmp, a_Q12_4567 ); + tmpa = _mm_srli_epi64( tmpa, 16 ); + tmpb = _mm_add_epi32( tmpb, tmpa ); + + psLPC_Q14_tmp = _mm_shuffle_epi32( psLPC_Q14_tmp, _MM_SHUFFLE( 0, 3, 2, 1 ) ); /* equal shift right 4 bytes */ + a_Q12_tmp = _mm_shuffle_epi32( a_Q12_4567, _MM_SHUFFLE(0, 3, 2, 1 ) ); /* equal shift right 4 bytes */ + psLPC_Q14_tmp = _mm_mul_epi32( psLPC_Q14_tmp, a_Q12_tmp ); + psLPC_Q14_tmp = _mm_srli_epi64( psLPC_Q14_tmp, 16 ); + tmpb = _mm_add_epi32( tmpb, psLPC_Q14_tmp ); + + if ( opus_likely( predictLPCOrder == 16 ) ) + { + /* step 3 */ + psLPC_Q14_tmp = _mm_loadu_si128( (__m128i *)(&psLPC_Q14[ -11 ] ) ); + psLPC_Q14_tmp = _mm_shuffle_epi32( psLPC_Q14_tmp, 0x1B ); + tmpa = _mm_mul_epi32( psLPC_Q14_tmp, a_Q12_89AB ); + tmpa = _mm_srli_epi64( tmpa, 16 ); + tmpb = _mm_add_epi32( tmpb, tmpa ); + + psLPC_Q14_tmp = _mm_shuffle_epi32( psLPC_Q14_tmp, _MM_SHUFFLE( 0, 3, 2, 1 ) ); /* equal shift right 4 bytes */ + a_Q12_tmp = _mm_shuffle_epi32( a_Q12_89AB, _MM_SHUFFLE(0, 3, 2, 1 ) );/* equal shift right 4 bytes */ + psLPC_Q14_tmp = _mm_mul_epi32( psLPC_Q14_tmp, a_Q12_tmp ); + psLPC_Q14_tmp = _mm_srli_epi64( psLPC_Q14_tmp, 16 ); + tmpb = _mm_add_epi32( tmpb, psLPC_Q14_tmp ); + + /* step 4 */ + psLPC_Q14_tmp = _mm_loadu_si128( (__m128i *)(&psLPC_Q14[ -15 ] ) ); + psLPC_Q14_tmp = _mm_shuffle_epi32( psLPC_Q14_tmp, 0x1B ); + tmpa = _mm_mul_epi32( psLPC_Q14_tmp, a_Q12_CDEF ); + tmpa = _mm_srli_epi64( tmpa, 16 ); + tmpb = _mm_add_epi32( tmpb, tmpa ); + + psLPC_Q14_tmp = _mm_shuffle_epi32( psLPC_Q14_tmp, _MM_SHUFFLE( 0, 3, 2, 1 ) ); /* equal shift right 4 bytes */ + a_Q12_tmp = _mm_shuffle_epi32( a_Q12_CDEF, _MM_SHUFFLE(0, 3, 2, 1 ) ); /* equal shift right 4 bytes */ + psLPC_Q14_tmp = _mm_mul_epi32( psLPC_Q14_tmp, a_Q12_tmp ); + psLPC_Q14_tmp = _mm_srli_epi64( psLPC_Q14_tmp, 16 ); + tmpb = _mm_add_epi32( tmpb, psLPC_Q14_tmp ); + + /* add at last */ + /* equal shift right 8 bytes*/ + tmpa = _mm_shuffle_epi32( tmpb, _MM_SHUFFLE( 0, 0, 3, 2 ) ); + tmpb = _mm_add_epi32( tmpb, tmpa ); + LPC_pred_Q14 += _mm_cvtsi128_si32( tmpb ); + } + else + { + /* add at last */ + tmpa = _mm_shuffle_epi32( tmpb, _MM_SHUFFLE( 0, 0, 3, 2 ) ); /* equal shift right 8 bytes*/ + tmpb = _mm_add_epi32( tmpb, tmpa ); + LPC_pred_Q14 += _mm_cvtsi128_si32( tmpb ); + + LPC_pred_Q14 = silk_SMLAWB( LPC_pred_Q14, psLPC_Q14[ -8 ], a_Q12[ 8 ] ); + LPC_pred_Q14 = silk_SMLAWB( LPC_pred_Q14, psLPC_Q14[ -9 ], a_Q12[ 9 ] ); + } + + LPC_pred_Q14 = silk_LSHIFT( LPC_pred_Q14, 4 ); /* Q10 -> Q14 */ + + /* Noise shape feedback */ + celt_assert( ( shapingLPCOrder & 1 ) == 0 ); /* check that order is even */ + /* Output of lowpass section */ + tmp2 = silk_SMLAWB( psDD->Diff_Q14, psDD->sAR2_Q14[ 0 ], warping_Q16 ); + /* Output of allpass section */ + tmp1 = silk_SMLAWB( psDD->sAR2_Q14[ 0 ], psDD->sAR2_Q14[ 1 ] - tmp2, warping_Q16 ); + psDD->sAR2_Q14[ 0 ] = tmp2; + n_AR_Q14 = silk_RSHIFT( shapingLPCOrder, 1 ); + n_AR_Q14 = silk_SMLAWB( n_AR_Q14, tmp2, AR_shp_Q13[ 0 ] ); + /* Loop over allpass sections */ + for( j = 2; j < shapingLPCOrder; j += 2 ) { + /* Output of allpass section */ + tmp2 = silk_SMLAWB( psDD->sAR2_Q14[ j - 1 ], psDD->sAR2_Q14[ j + 0 ] - tmp1, warping_Q16 ); + psDD->sAR2_Q14[ j - 1 ] = tmp1; + n_AR_Q14 = silk_SMLAWB( n_AR_Q14, tmp1, AR_shp_Q13[ j - 1 ] ); + /* Output of allpass section */ + tmp1 = silk_SMLAWB( psDD->sAR2_Q14[ j + 0 ], psDD->sAR2_Q14[ j + 1 ] - tmp2, warping_Q16 ); + psDD->sAR2_Q14[ j + 0 ] = tmp2; + n_AR_Q14 = silk_SMLAWB( n_AR_Q14, tmp2, AR_shp_Q13[ j ] ); + } + psDD->sAR2_Q14[ shapingLPCOrder - 1 ] = tmp1; + n_AR_Q14 = silk_SMLAWB( n_AR_Q14, tmp1, AR_shp_Q13[ shapingLPCOrder - 1 ] ); + + n_AR_Q14 = silk_LSHIFT( n_AR_Q14, 1 ); /* Q11 -> Q12 */ + n_AR_Q14 = silk_SMLAWB( n_AR_Q14, psDD->LF_AR_Q14, Tilt_Q14 ); /* Q12 */ + n_AR_Q14 = silk_LSHIFT( n_AR_Q14, 2 ); /* Q12 -> Q14 */ + + n_LF_Q14 = silk_SMULWB( psDD->Shape_Q14[ *smpl_buf_idx ], LF_shp_Q14 ); /* Q12 */ + n_LF_Q14 = silk_SMLAWT( n_LF_Q14, psDD->LF_AR_Q14, LF_shp_Q14 ); /* Q12 */ + n_LF_Q14 = silk_LSHIFT( n_LF_Q14, 2 ); /* Q12 -> Q14 */ + + /* Input minus prediction plus noise feedback */ + /* r = x[ i ] - LTP_pred - LPC_pred + n_AR + n_Tilt + n_LF + n_LTP */ + tmp1 = silk_ADD_SAT32( n_AR_Q14, n_LF_Q14 ); /* Q14 */ + tmp2 = silk_ADD32( n_LTP_Q14, LPC_pred_Q14 ); /* Q13 */ + tmp1 = silk_SUB_SAT32( tmp2, tmp1 ); /* Q13 */ + tmp1 = silk_RSHIFT_ROUND( tmp1, 4 ); /* Q10 */ + + r_Q10 = silk_SUB32( x_Q10[ i ], tmp1 ); /* residual error Q10 */ + + /* Flip sign depending on dither */ + if ( psDD->Seed < 0 ) { + r_Q10 = -r_Q10; + } + r_Q10 = silk_LIMIT_32( r_Q10, -(31 << 10), 30 << 10 ); + + /* Find two quantization level candidates and measure their rate-distortion */ + q1_Q10 = silk_SUB32( r_Q10, offset_Q10 ); + q1_Q0 = silk_RSHIFT( q1_Q10, 10 ); + if (Lambda_Q10 > 2048) { + /* For aggressive RDO, the bias becomes more than one pulse. */ + if (q1_Q10 > rdo_offset) { + q1_Q0 = silk_RSHIFT( q1_Q10 - rdo_offset, 10 ); + } else if (q1_Q10 < -rdo_offset) { + q1_Q0 = silk_RSHIFT( q1_Q10 + rdo_offset, 10 ); + } else if (q1_Q10 < 0) { + q1_Q0 = -1; + } else { + q1_Q0 = 0; + } + } + if( q1_Q0 > 0 ) { + q1_Q10 = silk_SUB32( silk_LSHIFT( q1_Q0, 10 ), QUANT_LEVEL_ADJUST_Q10 ); + q1_Q10 = silk_ADD32( q1_Q10, offset_Q10 ); + q2_Q10 = silk_ADD32( q1_Q10, 1024 ); + rd1_Q10 = silk_SMULBB( q1_Q10, Lambda_Q10 ); + rd2_Q10 = silk_SMULBB( q2_Q10, Lambda_Q10 ); + } else if( q1_Q0 == 0 ) { + q1_Q10 = offset_Q10; + q2_Q10 = silk_ADD32( q1_Q10, 1024 - QUANT_LEVEL_ADJUST_Q10 ); + rd1_Q10 = silk_SMULBB( q1_Q10, Lambda_Q10 ); + rd2_Q10 = silk_SMULBB( q2_Q10, Lambda_Q10 ); + } else if( q1_Q0 == -1 ) { + q2_Q10 = offset_Q10; + q1_Q10 = silk_SUB32( q2_Q10, 1024 - QUANT_LEVEL_ADJUST_Q10 ); + rd1_Q10 = silk_SMULBB( -q1_Q10, Lambda_Q10 ); + rd2_Q10 = silk_SMULBB( q2_Q10, Lambda_Q10 ); + } else { /* q1_Q0 < -1 */ + q1_Q10 = silk_ADD32( silk_LSHIFT( q1_Q0, 10 ), QUANT_LEVEL_ADJUST_Q10 ); + q1_Q10 = silk_ADD32( q1_Q10, offset_Q10 ); + q2_Q10 = silk_ADD32( q1_Q10, 1024 ); + rd1_Q10 = silk_SMULBB( -q1_Q10, Lambda_Q10 ); + rd2_Q10 = silk_SMULBB( -q2_Q10, Lambda_Q10 ); + } + rr_Q10 = silk_SUB32( r_Q10, q1_Q10 ); + rd1_Q10 = silk_RSHIFT( silk_SMLABB( rd1_Q10, rr_Q10, rr_Q10 ), 10 ); + rr_Q10 = silk_SUB32( r_Q10, q2_Q10 ); + rd2_Q10 = silk_RSHIFT( silk_SMLABB( rd2_Q10, rr_Q10, rr_Q10 ), 10 ); + + if( rd1_Q10 < rd2_Q10 ) { + psSS[ 0 ].RD_Q10 = silk_ADD32( psDD->RD_Q10, rd1_Q10 ); + psSS[ 1 ].RD_Q10 = silk_ADD32( psDD->RD_Q10, rd2_Q10 ); + psSS[ 0 ].Q_Q10 = q1_Q10; + psSS[ 1 ].Q_Q10 = q2_Q10; + } else { + psSS[ 0 ].RD_Q10 = silk_ADD32( psDD->RD_Q10, rd2_Q10 ); + psSS[ 1 ].RD_Q10 = silk_ADD32( psDD->RD_Q10, rd1_Q10 ); + psSS[ 0 ].Q_Q10 = q2_Q10; + psSS[ 1 ].Q_Q10 = q1_Q10; + } + + /* Update states for best quantization */ + + /* Quantized excitation */ + exc_Q14 = silk_LSHIFT32( psSS[ 0 ].Q_Q10, 4 ); + if ( psDD->Seed < 0 ) { + exc_Q14 = -exc_Q14; + } + + /* Add predictions */ + LPC_exc_Q14 = silk_ADD32( exc_Q14, LTP_pred_Q14 ); + xq_Q14 = silk_ADD32( LPC_exc_Q14, LPC_pred_Q14 ); + + /* Update states */ + psSS[ 0 ].Diff_Q14 = silk_SUB_LSHIFT32( xq_Q14, x_Q10[ i ], 4 ); + sLF_AR_shp_Q14 = silk_SUB32( psSS[ 0 ].Diff_Q14, n_AR_Q14 ); + psSS[ 0 ].sLTP_shp_Q14 = silk_SUB_SAT32( sLF_AR_shp_Q14, n_LF_Q14 ); + psSS[ 0 ].LF_AR_Q14 = sLF_AR_shp_Q14; + psSS[ 0 ].LPC_exc_Q14 = LPC_exc_Q14; + psSS[ 0 ].xq_Q14 = xq_Q14; + + /* Update states for second best quantization */ + + /* Quantized excitation */ + exc_Q14 = silk_LSHIFT32( psSS[ 1 ].Q_Q10, 4 ); + if ( psDD->Seed < 0 ) { + exc_Q14 = -exc_Q14; + } + + /* Add predictions */ + LPC_exc_Q14 = silk_ADD32( exc_Q14, LTP_pred_Q14 ); + xq_Q14 = silk_ADD32( LPC_exc_Q14, LPC_pred_Q14 ); + + /* Update states */ + psSS[ 1 ].Diff_Q14 = silk_SUB_LSHIFT32( xq_Q14, x_Q10[ i ], 4 ); + sLF_AR_shp_Q14 = silk_SUB32( psSS[ 1 ].Diff_Q14, n_AR_Q14 ); + psSS[ 1 ].sLTP_shp_Q14 = silk_SUB_SAT32( sLF_AR_shp_Q14, n_LF_Q14 ); + psSS[ 1 ].LF_AR_Q14 = sLF_AR_shp_Q14; + psSS[ 1 ].LPC_exc_Q14 = LPC_exc_Q14; + psSS[ 1 ].xq_Q14 = xq_Q14; + } + } + *smpl_buf_idx = ( *smpl_buf_idx - 1 ) % DECISION_DELAY; + if( *smpl_buf_idx < 0 ) *smpl_buf_idx += DECISION_DELAY; + last_smple_idx = ( *smpl_buf_idx + decisionDelay ) % DECISION_DELAY; + + /* Find winner */ + RDmin_Q10 = psSampleState[ 0 ][ 0 ].RD_Q10; + Winner_ind = 0; + for( k = 1; k < nStatesDelayedDecision; k++ ) { + if( psSampleState[ k ][ 0 ].RD_Q10 < RDmin_Q10 ) { + RDmin_Q10 = psSampleState[ k ][ 0 ].RD_Q10; + Winner_ind = k; + } + } + + /* Increase RD values of expired states */ + Winner_rand_state = psDelDec[ Winner_ind ].RandState[ last_smple_idx ]; + for( k = 0; k < nStatesDelayedDecision; k++ ) { + if( psDelDec[ k ].RandState[ last_smple_idx ] != Winner_rand_state ) { + psSampleState[ k ][ 0 ].RD_Q10 = silk_ADD32( psSampleState[ k ][ 0 ].RD_Q10, silk_int32_MAX >> 4 ); + psSampleState[ k ][ 1 ].RD_Q10 = silk_ADD32( psSampleState[ k ][ 1 ].RD_Q10, silk_int32_MAX >> 4 ); + silk_assert( psSampleState[ k ][ 0 ].RD_Q10 >= 0 ); + } + } + + /* Find worst in first set and best in second set */ + RDmax_Q10 = psSampleState[ 0 ][ 0 ].RD_Q10; + RDmin_Q10 = psSampleState[ 0 ][ 1 ].RD_Q10; + RDmax_ind = 0; + RDmin_ind = 0; + for( k = 1; k < nStatesDelayedDecision; k++ ) { + /* find worst in first set */ + if( psSampleState[ k ][ 0 ].RD_Q10 > RDmax_Q10 ) { + RDmax_Q10 = psSampleState[ k ][ 0 ].RD_Q10; + RDmax_ind = k; + } + /* find best in second set */ + if( psSampleState[ k ][ 1 ].RD_Q10 < RDmin_Q10 ) { + RDmin_Q10 = psSampleState[ k ][ 1 ].RD_Q10; + RDmin_ind = k; + } + } + + /* Replace a state if best from second set outperforms worst in first set */ + if( RDmin_Q10 < RDmax_Q10 ) { + silk_memcpy( ( (opus_int32 *)&psDelDec[ RDmax_ind ] ) + i, + ( (opus_int32 *)&psDelDec[ RDmin_ind ] ) + i, sizeof( NSQ_del_dec_struct ) - i * sizeof( opus_int32) ); + silk_memcpy( &psSampleState[ RDmax_ind ][ 0 ], &psSampleState[ RDmin_ind ][ 1 ], sizeof( NSQ_sample_struct ) ); + } + + /* Write samples from winner to output and long-term filter states */ + psDD = &psDelDec[ Winner_ind ]; + if( subfr > 0 || i >= decisionDelay ) { + pulses[ i - decisionDelay ] = (opus_int8)silk_RSHIFT_ROUND( psDD->Q_Q10[ last_smple_idx ], 10 ); + xq[ i - decisionDelay ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( + silk_SMULWW( psDD->Xq_Q14[ last_smple_idx ], delayedGain_Q10[ last_smple_idx ] ), 8 ) ); + NSQ->sLTP_shp_Q14[ NSQ->sLTP_shp_buf_idx - decisionDelay ] = psDD->Shape_Q14[ last_smple_idx ]; + sLTP_Q15[ NSQ->sLTP_buf_idx - decisionDelay ] = psDD->Pred_Q15[ last_smple_idx ]; + } + NSQ->sLTP_shp_buf_idx++; + NSQ->sLTP_buf_idx++; + + /* Update states */ + for( k = 0; k < nStatesDelayedDecision; k++ ) { + psDD = &psDelDec[ k ]; + psSS = &psSampleState[ k ][ 0 ]; + psDD->LF_AR_Q14 = psSS->LF_AR_Q14; + psDD->Diff_Q14 = psSS->Diff_Q14; + psDD->sLPC_Q14[ NSQ_LPC_BUF_LENGTH + i ] = psSS->xq_Q14; + psDD->Xq_Q14[ *smpl_buf_idx ] = psSS->xq_Q14; + psDD->Q_Q10[ *smpl_buf_idx ] = psSS->Q_Q10; + psDD->Pred_Q15[ *smpl_buf_idx ] = silk_LSHIFT32( psSS->LPC_exc_Q14, 1 ); + psDD->Shape_Q14[ *smpl_buf_idx ] = psSS->sLTP_shp_Q14; + psDD->Seed = silk_ADD32_ovflw( psDD->Seed, silk_RSHIFT_ROUND( psSS->Q_Q10, 10 ) ); + psDD->RandState[ *smpl_buf_idx ] = psDD->Seed; + psDD->RD_Q10 = psSS->RD_Q10; + } + delayedGain_Q10[ *smpl_buf_idx ] = Gain_Q10; + } + /* Update LPC states */ + for( k = 0; k < nStatesDelayedDecision; k++ ) { + psDD = &psDelDec[ k ]; + silk_memcpy( psDD->sLPC_Q14, &psDD->sLPC_Q14[ length ], NSQ_LPC_BUF_LENGTH * sizeof( opus_int32 ) ); + } + RESTORE_STACK; +} + +static OPUS_INLINE void silk_nsq_del_dec_scale_states_sse4_1( + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + NSQ_del_dec_struct psDelDec[], /* I/O Delayed decision states */ + const opus_int16 x16[], /* I Input */ + opus_int32 x_sc_Q10[], /* O Input scaled with 1/Gain in Q10 */ + const opus_int16 sLTP[], /* I Re-whitened LTP state in Q0 */ + opus_int32 sLTP_Q15[], /* O LTP state matching scaled input */ + opus_int subfr, /* I Subframe number */ + opus_int nStatesDelayedDecision, /* I Number of del dec states */ + const opus_int LTP_scale_Q14, /* I LTP state scaling */ + const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lag */ + const opus_int signal_type, /* I Signal type */ + const opus_int decisionDelay /* I Decision delay */ +) +{ + opus_int i, k, lag; + opus_int32 gain_adj_Q16, inv_gain_Q31, inv_gain_Q26; + NSQ_del_dec_struct *psDD; + __m128i xmm_inv_gain_Q26, xmm_x16_x2x0, xmm_x16_x3x1; + + lag = pitchL[ subfr ]; + inv_gain_Q31 = silk_INVERSE32_varQ( silk_max( Gains_Q16[ subfr ], 1 ), 47 ); + silk_assert( inv_gain_Q31 != 0 ); + + /* Scale input */ + inv_gain_Q26 = silk_RSHIFT_ROUND( inv_gain_Q31, 5 ); + + /* prepare inv_gain_Q26 in packed 4 32-bits */ + xmm_inv_gain_Q26 = _mm_set1_epi32(inv_gain_Q26); + + for( i = 0; i < psEncC->subfr_length - 3; i += 4 ) { + xmm_x16_x2x0 = OP_CVTEPI16_EPI32_M64( &(x16[ i ] ) ); + + /* equal shift right 4 bytes*/ + xmm_x16_x3x1 = _mm_shuffle_epi32( xmm_x16_x2x0, _MM_SHUFFLE( 0, 3, 2, 1 ) ); + + xmm_x16_x2x0 = _mm_mul_epi32( xmm_x16_x2x0, xmm_inv_gain_Q26 ); + xmm_x16_x3x1 = _mm_mul_epi32( xmm_x16_x3x1, xmm_inv_gain_Q26 ); + + xmm_x16_x2x0 = _mm_srli_epi64( xmm_x16_x2x0, 16 ); + xmm_x16_x3x1 = _mm_slli_epi64( xmm_x16_x3x1, 16 ); + + xmm_x16_x2x0 = _mm_blend_epi16( xmm_x16_x2x0, xmm_x16_x3x1, 0xCC ); + + _mm_storeu_si128( (__m128i *)(&(x_sc_Q10[ i ] ) ), xmm_x16_x2x0 ); + } + + for( ; i < psEncC->subfr_length; i++ ) { + x_sc_Q10[ i ] = silk_SMULWW( x16[ i ], inv_gain_Q26 ); + } + + /* After rewhitening the LTP state is un-scaled, so scale with inv_gain_Q16 */ + if( NSQ->rewhite_flag ) { + if( subfr == 0 ) { + /* Do LTP downscaling */ + inv_gain_Q31 = silk_LSHIFT( silk_SMULWB( inv_gain_Q31, LTP_scale_Q14 ), 2 ); + } + for( i = NSQ->sLTP_buf_idx - lag - LTP_ORDER / 2; i < NSQ->sLTP_buf_idx; i++ ) { + silk_assert( i < MAX_FRAME_LENGTH ); + sLTP_Q15[ i ] = silk_SMULWB( inv_gain_Q31, sLTP[ i ] ); + } + } + + /* Adjust for changing gain */ + if( Gains_Q16[ subfr ] != NSQ->prev_gain_Q16 ) { + gain_adj_Q16 = silk_DIV32_varQ( NSQ->prev_gain_Q16, Gains_Q16[ subfr ], 16 ); + + /* Scale long-term shaping state */ + { + __m128i xmm_gain_adj_Q16, xmm_sLTP_shp_Q14_x2x0, xmm_sLTP_shp_Q14_x3x1; + + /* prepare gain_adj_Q16 in packed 4 32-bits */ + xmm_gain_adj_Q16 = _mm_set1_epi32( gain_adj_Q16 ); + + for( i = NSQ->sLTP_shp_buf_idx - psEncC->ltp_mem_length; i < NSQ->sLTP_shp_buf_idx - 3; i += 4 ) + { + xmm_sLTP_shp_Q14_x2x0 = _mm_loadu_si128( (__m128i *)(&(NSQ->sLTP_shp_Q14[ i ] ) ) ); + /* equal shift right 4 bytes*/ + xmm_sLTP_shp_Q14_x3x1 = _mm_shuffle_epi32( xmm_sLTP_shp_Q14_x2x0, _MM_SHUFFLE( 0, 3, 2, 1 ) ); + + xmm_sLTP_shp_Q14_x2x0 = _mm_mul_epi32( xmm_sLTP_shp_Q14_x2x0, xmm_gain_adj_Q16 ); + xmm_sLTP_shp_Q14_x3x1 = _mm_mul_epi32( xmm_sLTP_shp_Q14_x3x1, xmm_gain_adj_Q16 ); + + xmm_sLTP_shp_Q14_x2x0 = _mm_srli_epi64( xmm_sLTP_shp_Q14_x2x0, 16 ); + xmm_sLTP_shp_Q14_x3x1 = _mm_slli_epi64( xmm_sLTP_shp_Q14_x3x1, 16 ); + + xmm_sLTP_shp_Q14_x2x0 = _mm_blend_epi16( xmm_sLTP_shp_Q14_x2x0, xmm_sLTP_shp_Q14_x3x1, 0xCC ); + + _mm_storeu_si128( (__m128i *)(&(NSQ->sLTP_shp_Q14[ i ] ) ), xmm_sLTP_shp_Q14_x2x0 ); + } + + for( ; i < NSQ->sLTP_shp_buf_idx; i++ ) { + NSQ->sLTP_shp_Q14[ i ] = silk_SMULWW( gain_adj_Q16, NSQ->sLTP_shp_Q14[ i ] ); + } + + /* Scale long-term prediction state */ + if( signal_type == TYPE_VOICED && NSQ->rewhite_flag == 0 ) { + for( i = NSQ->sLTP_buf_idx - lag - LTP_ORDER / 2; i < NSQ->sLTP_buf_idx - decisionDelay; i++ ) { + sLTP_Q15[ i ] = silk_SMULWW( gain_adj_Q16, sLTP_Q15[ i ] ); + } + } + + for( k = 0; k < nStatesDelayedDecision; k++ ) { + psDD = &psDelDec[ k ]; + + /* Scale scalar states */ + psDD->LF_AR_Q14 = silk_SMULWW( gain_adj_Q16, psDD->LF_AR_Q14 ); + psDD->Diff_Q14 = silk_SMULWW( gain_adj_Q16, psDD->Diff_Q14 ); + + /* Scale short-term prediction and shaping states */ + for( i = 0; i < NSQ_LPC_BUF_LENGTH; i++ ) { + psDD->sLPC_Q14[ i ] = silk_SMULWW( gain_adj_Q16, psDD->sLPC_Q14[ i ] ); + } + for( i = 0; i < MAX_SHAPE_LPC_ORDER; i++ ) { + psDD->sAR2_Q14[ i ] = silk_SMULWW( gain_adj_Q16, psDD->sAR2_Q14[ i ] ); + } + for( i = 0; i < DECISION_DELAY; i++ ) { + psDD->Pred_Q15[ i ] = silk_SMULWW( gain_adj_Q16, psDD->Pred_Q15[ i ] ); + psDD->Shape_Q14[ i ] = silk_SMULWW( gain_adj_Q16, psDD->Shape_Q14[ i ] ); + } + } + } + + /* Save inverse gain */ + NSQ->prev_gain_Q16 = Gains_Q16[ subfr ]; + } +} diff --git a/media/libopus/silk/x86/NSQ_sse4_1.c b/media/libopus/silk/x86/NSQ_sse4_1.c new file mode 100644 index 0000000000..d5ae1d3b1c --- /dev/null +++ b/media/libopus/silk/x86/NSQ_sse4_1.c @@ -0,0 +1,772 @@ +/* Copyright (c) 2014-2020, Cisco Systems, INC + Written by XiangMingZhu WeiZhou MinPeng YanWang FrancisQuiers + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include <xmmintrin.h> +#include <emmintrin.h> +#include <smmintrin.h> +#include "main.h" +#include "celt/x86/x86cpu.h" +#include "stack_alloc.h" + +static OPUS_INLINE void silk_nsq_scale_states_sse4_1( + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + const opus_int16 x16[], /* I input */ + opus_int32 x_sc_Q10[], /* O input scaled with 1/Gain */ + const opus_int16 sLTP[], /* I re-whitened LTP state in Q0 */ + opus_int32 sLTP_Q15[], /* O LTP state matching scaled input */ + opus_int subfr, /* I subframe number */ + const opus_int LTP_scale_Q14, /* I */ + const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lag */ + const opus_int signal_type /* I Signal type */ +); + +static OPUS_INLINE void silk_noise_shape_quantizer_10_16_sse4_1( + silk_nsq_state *NSQ, /* I/O NSQ state */ + opus_int signalType, /* I Signal type */ + const opus_int32 x_sc_Q10[], /* I */ + opus_int8 pulses[], /* O */ + opus_int16 xq[], /* O */ + opus_int32 sLTP_Q15[], /* I/O LTP state */ + const opus_int16 a_Q12[], /* I Short term prediction coefs */ + const opus_int16 b_Q14[], /* I Long term prediction coefs */ + const opus_int16 AR_shp_Q13[], /* I Noise shaping AR coefs */ + opus_int lag, /* I Pitch lag */ + opus_int32 HarmShapeFIRPacked_Q14, /* I */ + opus_int Tilt_Q14, /* I Spectral tilt */ + opus_int32 LF_shp_Q14, /* I */ + opus_int32 Gain_Q16, /* I */ + opus_int Lambda_Q10, /* I */ + opus_int offset_Q10, /* I */ + opus_int length, /* I Input length */ + opus_int32 table[][4] /* I */ +); + +void silk_NSQ_sse4_1( + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + SideInfoIndices *psIndices, /* I/O Quantization Indices */ + const opus_int16 x16[], /* I Input */ + opus_int8 pulses[], /* O Quantized pulse signal */ + const opus_int16 PredCoef_Q12[ 2 * MAX_LPC_ORDER ], /* I Short term prediction coefs */ + const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ + const opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ + const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ + const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ + const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ + const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ + const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ + const opus_int LTP_scale_Q14 /* I LTP state scaling */ +) +{ + opus_int k, lag, start_idx, LSF_interpolation_flag; + const opus_int16 *A_Q12, *B_Q14, *AR_shp_Q13; + opus_int16 *pxq; + VARDECL( opus_int32, sLTP_Q15 ); + VARDECL( opus_int16, sLTP ); + opus_int32 HarmShapeFIRPacked_Q14; + opus_int offset_Q10; + VARDECL( opus_int32, x_sc_Q10 ); + + opus_int32 table[ 64 ][ 4 ]; + opus_int32 tmp1; + opus_int32 q1_Q10, q2_Q10, rd1_Q20, rd2_Q20; + +#ifdef OPUS_CHECK_ASM + silk_nsq_state NSQ_c; + SideInfoIndices psIndices_c; + opus_int8 pulses_c[ MAX_FRAME_LENGTH ]; + const opus_int8 *const pulses_a = pulses; +#endif + + SAVE_STACK; + +#ifdef OPUS_CHECK_ASM + ( void )pulses_a; + silk_memcpy( &NSQ_c, NSQ, sizeof( NSQ_c ) ); + silk_memcpy( &psIndices_c, psIndices, sizeof( psIndices_c ) ); + silk_assert( psEncC->nb_subfr * psEncC->subfr_length <= MAX_FRAME_LENGTH ); + silk_memcpy( pulses_c, pulses, psEncC->nb_subfr * psEncC->subfr_length * sizeof( pulses[0] ) ); + + silk_NSQ_c( + psEncC, + &NSQ_c, + &psIndices_c, + x16, + pulses_c, + PredCoef_Q12, + LTPCoef_Q14, + AR_Q13, + HarmShapeGain_Q14, + Tilt_Q14, + LF_shp_Q14, + Gains_Q16, + pitchL, + Lambda_Q10, + LTP_scale_Q14 + ); +#endif + + NSQ->rand_seed = psIndices->Seed; + + /* Set unvoiced lag to the previous one, overwrite later for voiced */ + lag = NSQ->lagPrev; + + silk_assert( NSQ->prev_gain_Q16 != 0 ); + + offset_Q10 = silk_Quantization_Offsets_Q10[ psIndices->signalType >> 1 ][ psIndices->quantOffsetType ]; + + /* 0 */ + q1_Q10 = offset_Q10; + q2_Q10 = offset_Q10 + ( 1024 - QUANT_LEVEL_ADJUST_Q10 ); + rd1_Q20 = q1_Q10 * Lambda_Q10; + rd2_Q20 = q2_Q10 * Lambda_Q10; + + table[ 32 ][ 0 ] = q1_Q10; + table[ 32 ][ 1 ] = q2_Q10; + table[ 32 ][ 2 ] = 2 * (q1_Q10 - q2_Q10); + table[ 32 ][ 3 ] = (rd1_Q20 - rd2_Q20) + (q1_Q10 * q1_Q10 - q2_Q10 * q2_Q10); + + /* -1 */ + q1_Q10 = offset_Q10 - ( 1024 - QUANT_LEVEL_ADJUST_Q10 ); + q2_Q10 = offset_Q10; + rd1_Q20 = - q1_Q10 * Lambda_Q10; + rd2_Q20 = q2_Q10 * Lambda_Q10; + + table[ 31 ][ 0 ] = q1_Q10; + table[ 31 ][ 1 ] = q2_Q10; + table[ 31 ][ 2 ] = 2 * (q1_Q10 - q2_Q10); + table[ 31 ][ 3 ] = (rd1_Q20 - rd2_Q20) + (q1_Q10 * q1_Q10 - q2_Q10 * q2_Q10); + + /* > 0 */ + for (k = 1; k <= 31; k++) + { + tmp1 = offset_Q10 + silk_LSHIFT( k, 10 ); + + q1_Q10 = tmp1 - QUANT_LEVEL_ADJUST_Q10; + q2_Q10 = tmp1 - QUANT_LEVEL_ADJUST_Q10 + 1024; + rd1_Q20 = q1_Q10 * Lambda_Q10; + rd2_Q20 = q2_Q10 * Lambda_Q10; + + table[ 32 + k ][ 0 ] = q1_Q10; + table[ 32 + k ][ 1 ] = q2_Q10; + table[ 32 + k ][ 2 ] = 2 * (q1_Q10 - q2_Q10); + table[ 32 + k ][ 3 ] = (rd1_Q20 - rd2_Q20) + (q1_Q10 * q1_Q10 - q2_Q10 * q2_Q10); + } + + /* < -1 */ + for (k = -32; k <= -2; k++) + { + tmp1 = offset_Q10 + silk_LSHIFT( k, 10 ); + + q1_Q10 = tmp1 + QUANT_LEVEL_ADJUST_Q10; + q2_Q10 = tmp1 + QUANT_LEVEL_ADJUST_Q10 + 1024; + rd1_Q20 = - q1_Q10 * Lambda_Q10; + rd2_Q20 = - q2_Q10 * Lambda_Q10; + + table[ 32 + k ][ 0 ] = q1_Q10; + table[ 32 + k ][ 1 ] = q2_Q10; + table[ 32 + k ][ 2 ] = 2 * (q1_Q10 - q2_Q10); + table[ 32 + k ][ 3 ] = (rd1_Q20 - rd2_Q20) + (q1_Q10 * q1_Q10 - q2_Q10 * q2_Q10); + } + + if( psIndices->NLSFInterpCoef_Q2 == 4 ) { + LSF_interpolation_flag = 0; + } else { + LSF_interpolation_flag = 1; + } + + ALLOC( sLTP_Q15, psEncC->ltp_mem_length + psEncC->frame_length, opus_int32 ); + ALLOC( sLTP, psEncC->ltp_mem_length + psEncC->frame_length, opus_int16 ); + ALLOC( x_sc_Q10, psEncC->subfr_length, opus_int32 ); + /* Set up pointers to start of sub frame */ + NSQ->sLTP_shp_buf_idx = psEncC->ltp_mem_length; + NSQ->sLTP_buf_idx = psEncC->ltp_mem_length; + pxq = &NSQ->xq[ psEncC->ltp_mem_length ]; + for( k = 0; k < psEncC->nb_subfr; k++ ) { + A_Q12 = &PredCoef_Q12[ (( k >> 1 ) | ( 1 - LSF_interpolation_flag )) * MAX_LPC_ORDER ]; + B_Q14 = <PCoef_Q14[ k * LTP_ORDER ]; + AR_shp_Q13 = &AR_Q13[ k * MAX_SHAPE_LPC_ORDER ]; + + /* Noise shape parameters */ + silk_assert( HarmShapeGain_Q14[ k ] >= 0 ); + HarmShapeFIRPacked_Q14 = silk_RSHIFT( HarmShapeGain_Q14[ k ], 2 ); + HarmShapeFIRPacked_Q14 |= silk_LSHIFT( (opus_int32)silk_RSHIFT( HarmShapeGain_Q14[ k ], 1 ), 16 ); + + NSQ->rewhite_flag = 0; + if( psIndices->signalType == TYPE_VOICED ) { + /* Voiced */ + lag = pitchL[ k ]; + + /* Re-whitening */ + if( ( k & ( 3 - silk_LSHIFT( LSF_interpolation_flag, 1 ) ) ) == 0 ) { + /* Rewhiten with new A coefs */ + start_idx = psEncC->ltp_mem_length - lag - psEncC->predictLPCOrder - LTP_ORDER / 2; + celt_assert( start_idx > 0 ); + + silk_LPC_analysis_filter( &sLTP[ start_idx ], &NSQ->xq[ start_idx + k * psEncC->subfr_length ], + A_Q12, psEncC->ltp_mem_length - start_idx, psEncC->predictLPCOrder, psEncC->arch ); + + NSQ->rewhite_flag = 1; + NSQ->sLTP_buf_idx = psEncC->ltp_mem_length; + } + } + + silk_nsq_scale_states_sse4_1( psEncC, NSQ, x16, x_sc_Q10, sLTP, sLTP_Q15, k, LTP_scale_Q14, Gains_Q16, pitchL, psIndices->signalType ); + + if ( opus_likely( ( 10 == psEncC->shapingLPCOrder ) && ( 16 == psEncC->predictLPCOrder) ) ) + { + silk_noise_shape_quantizer_10_16_sse4_1( NSQ, psIndices->signalType, x_sc_Q10, pulses, pxq, sLTP_Q15, A_Q12, B_Q14, + AR_shp_Q13, lag, HarmShapeFIRPacked_Q14, Tilt_Q14[ k ], LF_shp_Q14[ k ], Gains_Q16[ k ], Lambda_Q10, + offset_Q10, psEncC->subfr_length, &(table[32]) ); + } + else + { + silk_noise_shape_quantizer( NSQ, psIndices->signalType, x_sc_Q10, pulses, pxq, sLTP_Q15, A_Q12, B_Q14, + AR_shp_Q13, lag, HarmShapeFIRPacked_Q14, Tilt_Q14[ k ], LF_shp_Q14[ k ], Gains_Q16[ k ], Lambda_Q10, + offset_Q10, psEncC->subfr_length, psEncC->shapingLPCOrder, psEncC->predictLPCOrder, psEncC->arch ); + } + + x16 += psEncC->subfr_length; + pulses += psEncC->subfr_length; + pxq += psEncC->subfr_length; + } + + /* Update lagPrev for next frame */ + NSQ->lagPrev = pitchL[ psEncC->nb_subfr - 1 ]; + + /* Save quantized speech and noise shaping signals */ + silk_memmove( NSQ->xq, &NSQ->xq[ psEncC->frame_length ], psEncC->ltp_mem_length * sizeof( opus_int16 ) ); + silk_memmove( NSQ->sLTP_shp_Q14, &NSQ->sLTP_shp_Q14[ psEncC->frame_length ], psEncC->ltp_mem_length * sizeof( opus_int32 ) ); + +#ifdef OPUS_CHECK_ASM + silk_assert( !memcmp( &NSQ_c, NSQ, sizeof( NSQ_c ) ) ); + silk_assert( !memcmp( &psIndices_c, psIndices, sizeof( psIndices_c ) ) ); + silk_assert( !memcmp( pulses_c, pulses_a, psEncC->nb_subfr * psEncC->subfr_length * sizeof( pulses[0] ) ) ); +#endif + + RESTORE_STACK; +} + +/************************************/ +/* silk_noise_shape_quantizer_10_16 */ +/************************************/ +static OPUS_INLINE void silk_noise_shape_quantizer_10_16_sse4_1( + silk_nsq_state *NSQ, /* I/O NSQ state */ + opus_int signalType, /* I Signal type */ + const opus_int32 x_sc_Q10[], /* I */ + opus_int8 pulses[], /* O */ + opus_int16 xq[], /* O */ + opus_int32 sLTP_Q15[], /* I/O LTP state */ + const opus_int16 a_Q12[], /* I Short term prediction coefs */ + const opus_int16 b_Q14[], /* I Long term prediction coefs */ + const opus_int16 AR_shp_Q13[], /* I Noise shaping AR coefs */ + opus_int lag, /* I Pitch lag */ + opus_int32 HarmShapeFIRPacked_Q14, /* I */ + opus_int Tilt_Q14, /* I Spectral tilt */ + opus_int32 LF_shp_Q14, /* I */ + opus_int32 Gain_Q16, /* I */ + opus_int Lambda_Q10, /* I */ + opus_int offset_Q10, /* I */ + opus_int length, /* I Input length */ + opus_int32 table[][4] /* I */ +) +{ + opus_int i; + opus_int32 LTP_pred_Q13, LPC_pred_Q10, n_AR_Q12, n_LTP_Q13; + opus_int32 n_LF_Q12, r_Q10, q1_Q0, q1_Q10, q2_Q10; + opus_int32 exc_Q14, LPC_exc_Q14, xq_Q14, Gain_Q10, sDiff_shp_Q14; + opus_int32 tmp1, tmp2, sLF_AR_shp_Q14; + opus_int32 *psLPC_Q14, *shp_lag_ptr, *pred_lag_ptr; + + __m128i xmm_tempa, xmm_tempb; + + __m128i xmm_one; + + __m128i psLPC_Q14_hi_01234567, psLPC_Q14_hi_89ABCDEF; + __m128i psLPC_Q14_lo_01234567, psLPC_Q14_lo_89ABCDEF; + __m128i a_Q12_01234567, a_Q12_89ABCDEF; + + __m128i sAR2_Q14_hi_76543210, sAR2_Q14_lo_76543210; + __m128i AR_shp_Q13_76543210; + + int rdo_offset = (Lambda_Q10 >> 1) - 512; + + shp_lag_ptr = &NSQ->sLTP_shp_Q14[ NSQ->sLTP_shp_buf_idx - lag + HARM_SHAPE_FIR_TAPS / 2 ]; + pred_lag_ptr = &sLTP_Q15[ NSQ->sLTP_buf_idx - lag + LTP_ORDER / 2 ]; + Gain_Q10 = silk_RSHIFT( Gain_Q16, 6 ); + + /* Set up short term AR state */ + psLPC_Q14 = &NSQ->sLPC_Q14[ NSQ_LPC_BUF_LENGTH - 1 ]; + + sLF_AR_shp_Q14 = NSQ->sLF_AR_shp_Q14; + xq_Q14 = psLPC_Q14[ 0 ]; + sDiff_shp_Q14 = NSQ->sDiff_shp_Q14; + LTP_pred_Q13 = 0; + + /* load a_Q12 */ + xmm_one = _mm_set_epi8( 1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14 ); + + /* load a_Q12[0] - a_Q12[7] */ + a_Q12_01234567 = _mm_loadu_si128( (__m128i *)(&a_Q12[ 0 ] ) ); + /* load a_Q12[ 8 ] - a_Q12[ 15 ] */ + a_Q12_89ABCDEF = _mm_loadu_si128( (__m128i *)(&a_Q12[ 8 ] ) ); + + a_Q12_01234567 = _mm_shuffle_epi8( a_Q12_01234567, xmm_one ); + a_Q12_89ABCDEF = _mm_shuffle_epi8( a_Q12_89ABCDEF, xmm_one ); + + /* load AR_shp_Q13 */ + AR_shp_Q13_76543210 = _mm_loadu_si128( (__m128i *)(&AR_shp_Q13[0] ) ); + + /* load psLPC_Q14 */ + xmm_one = _mm_set_epi8(15, 14, 11, 10, 7, 6, 3, 2, 13, 12, 9, 8, 5, 4, 1, 0 ); + + xmm_tempa = _mm_loadu_si128( (__m128i *)(&psLPC_Q14[-16]) ); + xmm_tempb = _mm_loadu_si128( (__m128i *)(&psLPC_Q14[-12]) ); + + xmm_tempa = _mm_shuffle_epi8( xmm_tempa, xmm_one ); + xmm_tempb = _mm_shuffle_epi8( xmm_tempb, xmm_one ); + + psLPC_Q14_hi_89ABCDEF = _mm_unpackhi_epi64( xmm_tempa, xmm_tempb ); + psLPC_Q14_lo_89ABCDEF = _mm_unpacklo_epi64( xmm_tempa, xmm_tempb ); + + xmm_tempa = _mm_loadu_si128( (__m128i *)(&psLPC_Q14[ -8 ]) ); + xmm_tempb = _mm_loadu_si128( (__m128i *)(&psLPC_Q14[ -4 ]) ); + + xmm_tempa = _mm_shuffle_epi8( xmm_tempa, xmm_one ); + xmm_tempb = _mm_shuffle_epi8( xmm_tempb, xmm_one ); + + psLPC_Q14_hi_01234567 = _mm_unpackhi_epi64( xmm_tempa, xmm_tempb ); + psLPC_Q14_lo_01234567 = _mm_unpacklo_epi64( xmm_tempa, xmm_tempb ); + + /* load sAR2_Q14 */ + xmm_tempa = _mm_loadu_si128( (__m128i *)(&(NSQ->sAR2_Q14[ 0 ]) ) ); + xmm_tempb = _mm_loadu_si128( (__m128i *)(&(NSQ->sAR2_Q14[ 4 ]) ) ); + + xmm_tempa = _mm_shuffle_epi8( xmm_tempa, xmm_one ); + xmm_tempb = _mm_shuffle_epi8( xmm_tempb, xmm_one ); + + sAR2_Q14_hi_76543210 = _mm_unpackhi_epi64( xmm_tempa, xmm_tempb ); + sAR2_Q14_lo_76543210 = _mm_unpacklo_epi64( xmm_tempa, xmm_tempb ); + + /* prepare 1 in 8 * 16bit */ + xmm_one = _mm_set1_epi16(1); + + for( i = 0; i < length; i++ ) + { + /* Short-term prediction */ + __m128i xmm_hi_07, xmm_hi_8F, xmm_lo_07, xmm_lo_8F; + + /* Avoids introducing a bias because silk_SMLAWB() always rounds to -inf */ + LPC_pred_Q10 = 8; /* silk_RSHIFT( predictLPCOrder, 1 ); */ + + /* shift psLPC_Q14 */ + psLPC_Q14_hi_89ABCDEF = _mm_alignr_epi8( psLPC_Q14_hi_01234567, psLPC_Q14_hi_89ABCDEF, 2 ); + psLPC_Q14_lo_89ABCDEF = _mm_alignr_epi8( psLPC_Q14_lo_01234567, psLPC_Q14_lo_89ABCDEF, 2 ); + + psLPC_Q14_hi_01234567 = _mm_srli_si128( psLPC_Q14_hi_01234567, 2 ); + psLPC_Q14_lo_01234567 = _mm_srli_si128( psLPC_Q14_lo_01234567, 2 ); + + psLPC_Q14_hi_01234567 = _mm_insert_epi16( psLPC_Q14_hi_01234567, (xq_Q14 >> 16), 7 ); + psLPC_Q14_lo_01234567 = _mm_insert_epi16( psLPC_Q14_lo_01234567, (xq_Q14), 7 ); + + /* high part, use pmaddwd, results in 4 32-bit */ + xmm_hi_07 = _mm_madd_epi16( psLPC_Q14_hi_01234567, a_Q12_01234567 ); + xmm_hi_8F = _mm_madd_epi16( psLPC_Q14_hi_89ABCDEF, a_Q12_89ABCDEF ); + + /* low part, use pmulhw, results in 8 16-bit, note we need simulate unsigned * signed, _mm_srai_epi16(psLPC_Q14_lo_01234567, 15) */ + xmm_tempa = _mm_cmpgt_epi16( _mm_setzero_si128(), psLPC_Q14_lo_01234567 ); + xmm_tempb = _mm_cmpgt_epi16( _mm_setzero_si128(), psLPC_Q14_lo_89ABCDEF ); + + xmm_tempa = _mm_and_si128( xmm_tempa, a_Q12_01234567 ); + xmm_tempb = _mm_and_si128( xmm_tempb, a_Q12_89ABCDEF ); + + xmm_lo_07 = _mm_mulhi_epi16( psLPC_Q14_lo_01234567, a_Q12_01234567 ); + xmm_lo_8F = _mm_mulhi_epi16( psLPC_Q14_lo_89ABCDEF, a_Q12_89ABCDEF ); + + xmm_lo_07 = _mm_add_epi16( xmm_lo_07, xmm_tempa ); + xmm_lo_8F = _mm_add_epi16( xmm_lo_8F, xmm_tempb ); + + xmm_lo_07 = _mm_madd_epi16( xmm_lo_07, xmm_one ); + xmm_lo_8F = _mm_madd_epi16( xmm_lo_8F, xmm_one ); + + /* accumulate */ + xmm_hi_07 = _mm_add_epi32( xmm_hi_07, xmm_hi_8F ); + xmm_lo_07 = _mm_add_epi32( xmm_lo_07, xmm_lo_8F ); + + xmm_hi_07 = _mm_add_epi32( xmm_hi_07, xmm_lo_07 ); + + xmm_hi_07 = _mm_add_epi32( xmm_hi_07, _mm_unpackhi_epi64(xmm_hi_07, xmm_hi_07 ) ); + xmm_hi_07 = _mm_add_epi32( xmm_hi_07, _mm_shufflelo_epi16(xmm_hi_07, 0x0E ) ); + + LPC_pred_Q10 += _mm_cvtsi128_si32( xmm_hi_07 ); + + /* Long-term prediction */ + if ( opus_likely( signalType == TYPE_VOICED ) ) { + /* Unrolled loop */ + /* Avoids introducing a bias because silk_SMLAWB() always rounds to -inf */ + LTP_pred_Q13 = 2; + { + __m128i b_Q14_3210, b_Q14_0123, pred_lag_ptr_0123; + + b_Q14_3210 = OP_CVTEPI16_EPI32_M64( b_Q14 ); + b_Q14_0123 = _mm_shuffle_epi32( b_Q14_3210, 0x1B ); + + /* loaded: [0] [-1] [-2] [-3] */ + pred_lag_ptr_0123 = _mm_loadu_si128( (__m128i *)(&pred_lag_ptr[ -3 ] ) ); + /* shuffle to [-3] [-2] [-1] [0] and to new xmm */ + xmm_tempa = _mm_shuffle_epi32( pred_lag_ptr_0123, 0x1B ); + /*64-bit multiply, a[2] * b[-2], a[0] * b[0] */ + xmm_tempa = _mm_mul_epi32( xmm_tempa, b_Q14_3210 ); + /* right shift 2 bytes (16 bits), zero extended */ + xmm_tempa = _mm_srli_si128( xmm_tempa, 2 ); + + /* a[1] * b[-1], a[3] * b[-3] */ + pred_lag_ptr_0123 = _mm_mul_epi32( pred_lag_ptr_0123, b_Q14_0123 ); + pred_lag_ptr_0123 = _mm_srli_si128( pred_lag_ptr_0123, 2 ); + + pred_lag_ptr_0123 = _mm_add_epi32( pred_lag_ptr_0123, xmm_tempa ); + /* equal shift right 8 bytes*/ + xmm_tempa = _mm_shuffle_epi32( pred_lag_ptr_0123, _MM_SHUFFLE( 0, 0, 3, 2 ) ); + xmm_tempa = _mm_add_epi32( xmm_tempa, pred_lag_ptr_0123 ); + + LTP_pred_Q13 += _mm_cvtsi128_si32( xmm_tempa ); + + LTP_pred_Q13 = silk_SMLAWB( LTP_pred_Q13, pred_lag_ptr[ -4 ], b_Q14[ 4 ] ); + pred_lag_ptr++; + } + } + + /* Noise shape feedback */ + NSQ->sAR2_Q14[ 9 ] = NSQ->sAR2_Q14[ 8 ]; + NSQ->sAR2_Q14[ 8 ] = _mm_cvtsi128_si32( _mm_srli_si128(_mm_unpackhi_epi16( sAR2_Q14_lo_76543210, sAR2_Q14_hi_76543210 ), 12 ) ); + + sAR2_Q14_hi_76543210 = _mm_slli_si128( sAR2_Q14_hi_76543210, 2 ); + sAR2_Q14_lo_76543210 = _mm_slli_si128( sAR2_Q14_lo_76543210, 2 ); + + sAR2_Q14_hi_76543210 = _mm_insert_epi16( sAR2_Q14_hi_76543210, (sDiff_shp_Q14 >> 16), 0 ); + sAR2_Q14_lo_76543210 = _mm_insert_epi16( sAR2_Q14_lo_76543210, (sDiff_shp_Q14), 0 ); + + /* high part, use pmaddwd, results in 4 32-bit */ + xmm_hi_07 = _mm_madd_epi16( sAR2_Q14_hi_76543210, AR_shp_Q13_76543210 ); + + /* low part, use pmulhw, results in 8 16-bit, note we need simulate unsigned * signed,_mm_srai_epi16(sAR2_Q14_lo_76543210, 15) */ + xmm_tempa = _mm_cmpgt_epi16( _mm_setzero_si128(), sAR2_Q14_lo_76543210 ); + xmm_tempa = _mm_and_si128( xmm_tempa, AR_shp_Q13_76543210 ); + + xmm_lo_07 = _mm_mulhi_epi16( sAR2_Q14_lo_76543210, AR_shp_Q13_76543210 ); + xmm_lo_07 = _mm_add_epi16( xmm_lo_07, xmm_tempa ); + + xmm_lo_07 = _mm_madd_epi16( xmm_lo_07, xmm_one ); + + /* accumulate */ + xmm_hi_07 = _mm_add_epi32( xmm_hi_07, xmm_lo_07 ); + + xmm_hi_07 = _mm_add_epi32( xmm_hi_07, _mm_unpackhi_epi64(xmm_hi_07, xmm_hi_07 ) ); + xmm_hi_07 = _mm_add_epi32( xmm_hi_07, _mm_shufflelo_epi16(xmm_hi_07, 0x0E ) ); + + n_AR_Q12 = 5 + _mm_cvtsi128_si32( xmm_hi_07 ); + + n_AR_Q12 = silk_SMLAWB( n_AR_Q12, NSQ->sAR2_Q14[ 8 ], AR_shp_Q13[ 8 ] ); + n_AR_Q12 = silk_SMLAWB( n_AR_Q12, NSQ->sAR2_Q14[ 9 ], AR_shp_Q13[ 9 ] ); + + n_AR_Q12 = silk_LSHIFT32( n_AR_Q12, 1 ); /* Q11 -> Q12 */ + n_AR_Q12 = silk_SMLAWB( n_AR_Q12, sLF_AR_shp_Q14, Tilt_Q14 ); + + n_LF_Q12 = silk_SMULWB( NSQ->sLTP_shp_Q14[ NSQ->sLTP_shp_buf_idx - 1 ], LF_shp_Q14 ); + n_LF_Q12 = silk_SMLAWT( n_LF_Q12, sLF_AR_shp_Q14, LF_shp_Q14 ); + + celt_assert( lag > 0 || signalType != TYPE_VOICED ); + + /* Combine prediction and noise shaping signals */ + tmp1 = silk_SUB32( silk_LSHIFT32( LPC_pred_Q10, 2 ), n_AR_Q12 ); /* Q12 */ + tmp1 = silk_SUB32( tmp1, n_LF_Q12 ); /* Q12 */ + if( lag > 0 ) { + /* Symmetric, packed FIR coefficients */ + n_LTP_Q13 = silk_SMULWB( silk_ADD_SAT32( shp_lag_ptr[ 0 ], shp_lag_ptr[ -2 ] ), HarmShapeFIRPacked_Q14 ); + n_LTP_Q13 = silk_SMLAWT( n_LTP_Q13, shp_lag_ptr[ -1 ], HarmShapeFIRPacked_Q14 ); + n_LTP_Q13 = silk_LSHIFT( n_LTP_Q13, 1 ); + shp_lag_ptr++; + + tmp2 = silk_SUB32( LTP_pred_Q13, n_LTP_Q13 ); /* Q13 */ + tmp1 = silk_ADD_LSHIFT32( tmp2, tmp1, 1 ); /* Q13 */ + tmp1 = silk_RSHIFT_ROUND( tmp1, 3 ); /* Q10 */ + } else { + tmp1 = silk_RSHIFT_ROUND( tmp1, 2 ); /* Q10 */ + } + + r_Q10 = silk_SUB32( x_sc_Q10[ i ], tmp1 ); /* residual error Q10 */ + + /* Generate dither */ + NSQ->rand_seed = silk_RAND( NSQ->rand_seed ); + + /* Flip sign depending on dither */ + tmp2 = -r_Q10; + if ( NSQ->rand_seed < 0 ) r_Q10 = tmp2; + + r_Q10 = silk_LIMIT_32( r_Q10, -(31 << 10), 30 << 10 ); + + /* Find two quantization level candidates and measure their rate-distortion */ + q1_Q10 = silk_SUB32( r_Q10, offset_Q10 ); + q1_Q0 = silk_RSHIFT( q1_Q10, 10 ); + if (Lambda_Q10 > 2048) { + /* For aggressive RDO, the bias becomes more than one pulse. */ + if (q1_Q10 > rdo_offset) { + q1_Q0 = silk_RSHIFT( q1_Q10 - rdo_offset, 10 ); + } else if (q1_Q10 < -rdo_offset) { + q1_Q0 = silk_RSHIFT( q1_Q10 + rdo_offset, 10 ); + } else if (q1_Q10 < 0) { + q1_Q0 = -1; + } else { + q1_Q0 = 0; + } + } + + q1_Q10 = table[q1_Q0][0]; + q2_Q10 = table[q1_Q0][1]; + + if (r_Q10 * table[q1_Q0][2] - table[q1_Q0][3] < 0) + { + q1_Q10 = q2_Q10; + } + + pulses[ i ] = (opus_int8)silk_RSHIFT_ROUND( q1_Q10, 10 ); + + /* Excitation */ + exc_Q14 = silk_LSHIFT( q1_Q10, 4 ); + + tmp2 = -exc_Q14; + if ( NSQ->rand_seed < 0 ) exc_Q14 = tmp2; + + /* Add predictions */ + LPC_exc_Q14 = silk_ADD_LSHIFT32( exc_Q14, LTP_pred_Q13, 1 ); + xq_Q14 = silk_ADD_LSHIFT32( LPC_exc_Q14, LPC_pred_Q10, 4 ); + + /* Update states */ + psLPC_Q14++; + *psLPC_Q14 = xq_Q14; + NSQ->sDiff_shp_Q14 = silk_SUB_LSHIFT32( xq_Q14, x_sc_Q10[ i ], 4 ); + sLF_AR_shp_Q14 = silk_SUB_LSHIFT32( NSQ->sDiff_shp_Q14, n_AR_Q12, 2 ); + + NSQ->sLTP_shp_Q14[ NSQ->sLTP_shp_buf_idx ] = silk_SUB_LSHIFT32( sLF_AR_shp_Q14, n_LF_Q12, 2 ); + sLTP_Q15[ NSQ->sLTP_buf_idx ] = silk_LSHIFT( LPC_exc_Q14, 1 ); + NSQ->sLTP_shp_buf_idx++; + NSQ->sLTP_buf_idx++; + + /* Make dither dependent on quantized signal */ + NSQ->rand_seed = silk_ADD32_ovflw( NSQ->rand_seed, pulses[ i ] ); + } + + NSQ->sLF_AR_shp_Q14 = sLF_AR_shp_Q14; + + /* Scale XQ back to normal level before saving */ + psLPC_Q14 = &NSQ->sLPC_Q14[ NSQ_LPC_BUF_LENGTH ]; + + /* write back sAR2_Q14 */ + xmm_tempa = _mm_unpackhi_epi16( sAR2_Q14_lo_76543210, sAR2_Q14_hi_76543210 ); + xmm_tempb = _mm_unpacklo_epi16( sAR2_Q14_lo_76543210, sAR2_Q14_hi_76543210 ); + _mm_storeu_si128( (__m128i *)(&NSQ->sAR2_Q14[ 4 ]), xmm_tempa ); + _mm_storeu_si128( (__m128i *)(&NSQ->sAR2_Q14[ 0 ]), xmm_tempb ); + + /* xq[ i ] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( silk_SMULWW( psLPC_Q14[ i ], Gain_Q10 ), 8 ) ); */ + { + __m128i xmm_Gain_Q10; + __m128i xmm_xq_Q14_3210, xmm_xq_Q14_x3x1, xmm_xq_Q14_7654, xmm_xq_Q14_x7x5; + + /* prepare (1 << 7) in packed 4 32-bits */ + xmm_tempa = _mm_set1_epi32( (1 << 7) ); + + /* prepare Gain_Q10 in packed 4 32-bits */ + xmm_Gain_Q10 = _mm_set1_epi32( Gain_Q10 ); + + /* process xq */ + for (i = 0; i < length - 7; i += 8) + { + xmm_xq_Q14_3210 = _mm_loadu_si128( (__m128i *)(&(psLPC_Q14[ i + 0 ] ) ) ); + xmm_xq_Q14_7654 = _mm_loadu_si128( (__m128i *)(&(psLPC_Q14[ i + 4 ] ) ) ); + + /* equal shift right 4 bytes*/ + xmm_xq_Q14_x3x1 = _mm_shuffle_epi32( xmm_xq_Q14_3210, _MM_SHUFFLE( 0, 3, 2, 1 ) ); + /* equal shift right 4 bytes*/ + xmm_xq_Q14_x7x5 = _mm_shuffle_epi32( xmm_xq_Q14_7654, _MM_SHUFFLE( 0, 3, 2, 1 ) ); + + xmm_xq_Q14_3210 = _mm_mul_epi32( xmm_xq_Q14_3210, xmm_Gain_Q10 ); + xmm_xq_Q14_x3x1 = _mm_mul_epi32( xmm_xq_Q14_x3x1, xmm_Gain_Q10 ); + xmm_xq_Q14_7654 = _mm_mul_epi32( xmm_xq_Q14_7654, xmm_Gain_Q10 ); + xmm_xq_Q14_x7x5 = _mm_mul_epi32( xmm_xq_Q14_x7x5, xmm_Gain_Q10 ); + + xmm_xq_Q14_3210 = _mm_srli_epi64( xmm_xq_Q14_3210, 16 ); + xmm_xq_Q14_x3x1 = _mm_slli_epi64( xmm_xq_Q14_x3x1, 16 ); + xmm_xq_Q14_7654 = _mm_srli_epi64( xmm_xq_Q14_7654, 16 ); + xmm_xq_Q14_x7x5 = _mm_slli_epi64( xmm_xq_Q14_x7x5, 16 ); + + xmm_xq_Q14_3210 = _mm_blend_epi16( xmm_xq_Q14_3210, xmm_xq_Q14_x3x1, 0xCC ); + xmm_xq_Q14_7654 = _mm_blend_epi16( xmm_xq_Q14_7654, xmm_xq_Q14_x7x5, 0xCC ); + + /* silk_RSHIFT_ROUND(xq, 8) */ + xmm_xq_Q14_3210 = _mm_add_epi32( xmm_xq_Q14_3210, xmm_tempa ); + xmm_xq_Q14_7654 = _mm_add_epi32( xmm_xq_Q14_7654, xmm_tempa ); + + xmm_xq_Q14_3210 = _mm_srai_epi32( xmm_xq_Q14_3210, 8 ); + xmm_xq_Q14_7654 = _mm_srai_epi32( xmm_xq_Q14_7654, 8 ); + + /* silk_SAT16 */ + xmm_xq_Q14_3210 = _mm_packs_epi32( xmm_xq_Q14_3210, xmm_xq_Q14_7654 ); + + /* save to xq */ + _mm_storeu_si128( (__m128i *)(&xq[ i ] ), xmm_xq_Q14_3210 ); + } + } + for ( ; i < length; i++) + { + xq[i] = (opus_int16)silk_SAT16( silk_RSHIFT_ROUND( silk_SMULWW( psLPC_Q14[ i ], Gain_Q10 ), 8 ) ); + } + + /* Update LPC synth buffer */ + silk_memcpy( NSQ->sLPC_Q14, &NSQ->sLPC_Q14[ length ], NSQ_LPC_BUF_LENGTH * sizeof( opus_int32 ) ); +} + +static OPUS_INLINE void silk_nsq_scale_states_sse4_1( + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + const opus_int16 x16[], /* I input */ + opus_int32 x_sc_Q10[], /* O input scaled with 1/Gain */ + const opus_int16 sLTP[], /* I re-whitened LTP state in Q0 */ + opus_int32 sLTP_Q15[], /* O LTP state matching scaled input */ + opus_int subfr, /* I subframe number */ + const opus_int LTP_scale_Q14, /* I */ + const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lag */ + const opus_int signal_type /* I Signal type */ +) +{ + opus_int i, lag; + opus_int32 gain_adj_Q16, inv_gain_Q31, inv_gain_Q26; + __m128i xmm_inv_gain_Q26, xmm_x16_x2x0, xmm_x16_x3x1; + + lag = pitchL[ subfr ]; + inv_gain_Q31 = silk_INVERSE32_varQ( silk_max( Gains_Q16[ subfr ], 1 ), 47 ); + silk_assert( inv_gain_Q31 != 0 ); + + /* Scale input */ + inv_gain_Q26 = silk_RSHIFT_ROUND( inv_gain_Q31, 5 ); + + /* prepare inv_gain_Q26 in packed 4 32-bits */ + xmm_inv_gain_Q26 = _mm_set1_epi32(inv_gain_Q26); + + for( i = 0; i < psEncC->subfr_length - 3; i += 4 ) { + xmm_x16_x2x0 = OP_CVTEPI16_EPI32_M64( &(x16[ i ] ) ); + + /* equal shift right 4 bytes*/ + xmm_x16_x3x1 = _mm_shuffle_epi32( xmm_x16_x2x0, _MM_SHUFFLE( 0, 3, 2, 1 ) ); + + xmm_x16_x2x0 = _mm_mul_epi32( xmm_x16_x2x0, xmm_inv_gain_Q26 ); + xmm_x16_x3x1 = _mm_mul_epi32( xmm_x16_x3x1, xmm_inv_gain_Q26 ); + + xmm_x16_x2x0 = _mm_srli_epi64( xmm_x16_x2x0, 16 ); + xmm_x16_x3x1 = _mm_slli_epi64( xmm_x16_x3x1, 16 ); + + xmm_x16_x2x0 = _mm_blend_epi16( xmm_x16_x2x0, xmm_x16_x3x1, 0xCC ); + + _mm_storeu_si128( (__m128i *)(&(x_sc_Q10[ i ] ) ), xmm_x16_x2x0 ); + } + + for( ; i < psEncC->subfr_length; i++ ) { + x_sc_Q10[ i ] = silk_SMULWW( x16[ i ], inv_gain_Q26 ); + } + + /* After rewhitening the LTP state is un-scaled, so scale with inv_gain_Q16 */ + if( NSQ->rewhite_flag ) { + if( subfr == 0 ) { + /* Do LTP downscaling */ + inv_gain_Q31 = silk_LSHIFT( silk_SMULWB( inv_gain_Q31, LTP_scale_Q14 ), 2 ); + } + for( i = NSQ->sLTP_buf_idx - lag - LTP_ORDER / 2; i < NSQ->sLTP_buf_idx; i++ ) { + silk_assert( i < MAX_FRAME_LENGTH ); + sLTP_Q15[ i ] = silk_SMULWB( inv_gain_Q31, sLTP[ i ] ); + } + } + + /* Adjust for changing gain */ + if( Gains_Q16[ subfr ] != NSQ->prev_gain_Q16 ) { + __m128i xmm_gain_adj_Q16, xmm_sLTP_shp_Q14_x2x0, xmm_sLTP_shp_Q14_x3x1; + gain_adj_Q16 = silk_DIV32_varQ( NSQ->prev_gain_Q16, Gains_Q16[ subfr ], 16 ); + + /* Scale long-term shaping state */ + + /* prepare gain_adj_Q16 in packed 4 32-bits */ + xmm_gain_adj_Q16 = _mm_set1_epi32(gain_adj_Q16); + + for( i = NSQ->sLTP_shp_buf_idx - psEncC->ltp_mem_length; i < NSQ->sLTP_shp_buf_idx - 3; i += 4 ) + { + xmm_sLTP_shp_Q14_x2x0 = _mm_loadu_si128( (__m128i *)(&(NSQ->sLTP_shp_Q14[ i ] ) ) ); + /* equal shift right 4 bytes*/ + xmm_sLTP_shp_Q14_x3x1 = _mm_shuffle_epi32( xmm_sLTP_shp_Q14_x2x0, _MM_SHUFFLE( 0, 3, 2, 1 ) ); + + xmm_sLTP_shp_Q14_x2x0 = _mm_mul_epi32( xmm_sLTP_shp_Q14_x2x0, xmm_gain_adj_Q16 ); + xmm_sLTP_shp_Q14_x3x1 = _mm_mul_epi32( xmm_sLTP_shp_Q14_x3x1, xmm_gain_adj_Q16 ); + + xmm_sLTP_shp_Q14_x2x0 = _mm_srli_epi64( xmm_sLTP_shp_Q14_x2x0, 16 ); + xmm_sLTP_shp_Q14_x3x1 = _mm_slli_epi64( xmm_sLTP_shp_Q14_x3x1, 16 ); + + xmm_sLTP_shp_Q14_x2x0 = _mm_blend_epi16( xmm_sLTP_shp_Q14_x2x0, xmm_sLTP_shp_Q14_x3x1, 0xCC ); + + _mm_storeu_si128( (__m128i *)(&(NSQ->sLTP_shp_Q14[ i ] ) ), xmm_sLTP_shp_Q14_x2x0 ); + } + + for( ; i < NSQ->sLTP_shp_buf_idx; i++ ) { + NSQ->sLTP_shp_Q14[ i ] = silk_SMULWW( gain_adj_Q16, NSQ->sLTP_shp_Q14[ i ] ); + } + + /* Scale long-term prediction state */ + if( signal_type == TYPE_VOICED && NSQ->rewhite_flag == 0 ) { + for( i = NSQ->sLTP_buf_idx - lag - LTP_ORDER / 2; i < NSQ->sLTP_buf_idx; i++ ) { + sLTP_Q15[ i ] = silk_SMULWW( gain_adj_Q16, sLTP_Q15[ i ] ); + } + } + + NSQ->sLF_AR_shp_Q14 = silk_SMULWW( gain_adj_Q16, NSQ->sLF_AR_shp_Q14 ); + NSQ->sDiff_shp_Q14 = silk_SMULWW( gain_adj_Q16, NSQ->sDiff_shp_Q14 ); + + /* Scale short-term prediction and shaping states */ + for( i = 0; i < NSQ_LPC_BUF_LENGTH; i++ ) { + NSQ->sLPC_Q14[ i ] = silk_SMULWW( gain_adj_Q16, NSQ->sLPC_Q14[ i ] ); + } + for( i = 0; i < MAX_SHAPE_LPC_ORDER; i++ ) { + NSQ->sAR2_Q14[ i ] = silk_SMULWW( gain_adj_Q16, NSQ->sAR2_Q14[ i ] ); + } + + /* Save inverse gain */ + NSQ->prev_gain_Q16 = Gains_Q16[ subfr ]; + } +} diff --git a/media/libopus/silk/x86/SigProc_FIX_sse.h b/media/libopus/silk/x86/SigProc_FIX_sse.h new file mode 100644 index 0000000000..89a5ec888a --- /dev/null +++ b/media/libopus/silk/x86/SigProc_FIX_sse.h @@ -0,0 +1,99 @@ +/* Copyright (c) 2014, Cisco Systems, INC + Written by XiangMingZhu WeiZhou MinPeng YanWang + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef SIGPROC_FIX_SSE_H +# define SIGPROC_FIX_SSE_H + +# ifdef HAVE_CONFIG_H +# include "config.h" +# endif + +# if defined(OPUS_X86_MAY_HAVE_SSE4_1) +void silk_burg_modified_sse4_1( + opus_int32 *res_nrg, /* O Residual energy */ + opus_int *res_nrg_Q, /* O Residual energy Q value */ + opus_int32 A_Q16[], /* O Prediction coefficients (length order) */ + const opus_int16 x[], /* I Input signal, length: nb_subfr * ( D + subfr_length ) */ + const opus_int32 minInvGain_Q30, /* I Inverse of max prediction gain */ + const opus_int subfr_length, /* I Input signal subframe length (incl. D preceding samples) */ + const opus_int nb_subfr, /* I Number of subframes stacked in x */ + const opus_int D, /* I Order */ + int arch /* I Run-time architecture */ +); + +# if defined(OPUS_X86_PRESUME_SSE4_1) + +# define OVERRIDE_silk_burg_modified +# define silk_burg_modified(res_nrg, res_nrg_Q, A_Q16, x, minInvGain_Q30, subfr_length, nb_subfr, D, arch) \ + ((void)(arch), silk_burg_modified_sse4_1(res_nrg, res_nrg_Q, A_Q16, x, minInvGain_Q30, subfr_length, nb_subfr, D, arch)) + +# elif defined(OPUS_HAVE_RTCD) + +extern void (*const SILK_BURG_MODIFIED_IMPL[OPUS_ARCHMASK + 1])( + opus_int32 *res_nrg, /* O Residual energy */ + opus_int *res_nrg_Q, /* O Residual energy Q value */ + opus_int32 A_Q16[], /* O Prediction coefficients (length order) */ + const opus_int16 x[], /* I Input signal, length: nb_subfr * ( D + subfr_length ) */ + const opus_int32 minInvGain_Q30, /* I Inverse of max prediction gain */ + const opus_int subfr_length, /* I Input signal subframe length (incl. D preceding samples) */ + const opus_int nb_subfr, /* I Number of subframes stacked in x */ + const opus_int D, /* I Order */ + int arch /* I Run-time architecture */); + +# define OVERRIDE_silk_burg_modified +# define silk_burg_modified(res_nrg, res_nrg_Q, A_Q16, x, minInvGain_Q30, subfr_length, nb_subfr, D, arch) \ + ((*SILK_BURG_MODIFIED_IMPL[(arch) & OPUS_ARCHMASK])(res_nrg, res_nrg_Q, A_Q16, x, minInvGain_Q30, subfr_length, nb_subfr, D, arch)) + +# endif + +opus_int64 silk_inner_prod16_sse4_1( + const opus_int16 *inVec1, + const opus_int16 *inVec2, + const opus_int len +); + + +# if defined(OPUS_X86_PRESUME_SSE4_1) + +# define OVERRIDE_silk_inner_prod16 +# define silk_inner_prod16(inVec1, inVec2, len, arch) \ + ((void)(arch),silk_inner_prod16_sse4_1(inVec1, inVec2, len)) + +# elif defined(OPUS_HAVE_RTCD) + +extern opus_int64 (*const SILK_INNER_PROD16_IMPL[OPUS_ARCHMASK + 1])( + const opus_int16 *inVec1, + const opus_int16 *inVec2, + const opus_int len); + +# define OVERRIDE_silk_inner_prod16 +# define silk_inner_prod16(inVec1, inVec2, len, arch) \ + ((*SILK_INNER_PROD16_IMPL[(arch) & OPUS_ARCHMASK])(inVec1, inVec2, len)) + +# endif +# endif +#endif diff --git a/media/libopus/silk/x86/VAD_sse4_1.c b/media/libopus/silk/x86/VAD_sse4_1.c new file mode 100644 index 0000000000..e7eaf9714a --- /dev/null +++ b/media/libopus/silk/x86/VAD_sse4_1.c @@ -0,0 +1,289 @@ +/* Copyright (c) 2014-2020, Cisco Systems, INC + Written by XiangMingZhu WeiZhou MinPeng YanWang FrancisQuiers + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include <xmmintrin.h> +#include <emmintrin.h> +#include <smmintrin.h> + +#include "main.h" +#include "stack_alloc.h" + +/* Weighting factors for tilt measure */ +static const opus_int32 tiltWeights[ VAD_N_BANDS ] = { 30000, 6000, -12000, -12000 }; + +/***************************************/ +/* Get the speech activity level in Q8 */ +/***************************************/ +opus_int silk_VAD_GetSA_Q8_sse4_1( /* O Return value, 0 if success */ + silk_encoder_state *psEncC, /* I/O Encoder state */ + const opus_int16 pIn[] /* I PCM input */ +) +{ + opus_int SA_Q15, pSNR_dB_Q7, input_tilt; + opus_int decimated_framelength1, decimated_framelength2; + opus_int decimated_framelength; + opus_int dec_subframe_length, dec_subframe_offset, SNR_Q7, i, b, s; + opus_int32 sumSquared, smooth_coef_Q16; + opus_int16 HPstateTmp; + VARDECL( opus_int16, X ); + opus_int32 Xnrg[ VAD_N_BANDS ]; + opus_int32 NrgToNoiseRatio_Q8[ VAD_N_BANDS ]; + opus_int32 speech_nrg, x_tmp; + opus_int X_offset[ VAD_N_BANDS ]; + opus_int ret = 0; + silk_VAD_state *psSilk_VAD = &psEncC->sVAD; + + SAVE_STACK; + +#ifdef OPUS_CHECK_ASM + silk_encoder_state psEncC_c; + opus_int ret_c; + + silk_memcpy( &psEncC_c, psEncC, sizeof( psEncC_c ) ); + ret_c = silk_VAD_GetSA_Q8_c( &psEncC_c, pIn ); +#endif + + /* Safety checks */ + silk_assert( VAD_N_BANDS == 4 ); + celt_assert( MAX_FRAME_LENGTH >= psEncC->frame_length ); + celt_assert( psEncC->frame_length <= 512 ); + celt_assert( psEncC->frame_length == 8 * silk_RSHIFT( psEncC->frame_length, 3 ) ); + + /***********************/ + /* Filter and Decimate */ + /***********************/ + decimated_framelength1 = silk_RSHIFT( psEncC->frame_length, 1 ); + decimated_framelength2 = silk_RSHIFT( psEncC->frame_length, 2 ); + decimated_framelength = silk_RSHIFT( psEncC->frame_length, 3 ); + /* Decimate into 4 bands: + 0 L 3L L 3L 5L + - -- - -- -- + 8 8 2 4 4 + + [0-1 kHz| temp. |1-2 kHz| 2-4 kHz | 4-8 kHz | + + They're arranged to allow the minimal ( frame_length / 4 ) extra + scratch space during the downsampling process */ + X_offset[ 0 ] = 0; + X_offset[ 1 ] = decimated_framelength + decimated_framelength2; + X_offset[ 2 ] = X_offset[ 1 ] + decimated_framelength; + X_offset[ 3 ] = X_offset[ 2 ] + decimated_framelength2; + ALLOC( X, X_offset[ 3 ] + decimated_framelength1, opus_int16 ); + + /* 0-8 kHz to 0-4 kHz and 4-8 kHz */ + silk_ana_filt_bank_1( pIn, &psSilk_VAD->AnaState[ 0 ], + X, &X[ X_offset[ 3 ] ], psEncC->frame_length ); + + /* 0-4 kHz to 0-2 kHz and 2-4 kHz */ + silk_ana_filt_bank_1( X, &psSilk_VAD->AnaState1[ 0 ], + X, &X[ X_offset[ 2 ] ], decimated_framelength1 ); + + /* 0-2 kHz to 0-1 kHz and 1-2 kHz */ + silk_ana_filt_bank_1( X, &psSilk_VAD->AnaState2[ 0 ], + X, &X[ X_offset[ 1 ] ], decimated_framelength2 ); + + /*********************************************/ + /* HP filter on lowest band (differentiator) */ + /*********************************************/ + X[ decimated_framelength - 1 ] = silk_RSHIFT( X[ decimated_framelength - 1 ], 1 ); + HPstateTmp = X[ decimated_framelength - 1 ]; + for( i = decimated_framelength - 1; i > 0; i-- ) { + X[ i - 1 ] = silk_RSHIFT( X[ i - 1 ], 1 ); + X[ i ] -= X[ i - 1 ]; + } + X[ 0 ] -= psSilk_VAD->HPstate; + psSilk_VAD->HPstate = HPstateTmp; + + /*************************************/ + /* Calculate the energy in each band */ + /*************************************/ + for( b = 0; b < VAD_N_BANDS; b++ ) { + /* Find the decimated framelength in the non-uniformly divided bands */ + decimated_framelength = silk_RSHIFT( psEncC->frame_length, silk_min_int( VAD_N_BANDS - b, VAD_N_BANDS - 1 ) ); + + /* Split length into subframe lengths */ + dec_subframe_length = silk_RSHIFT( decimated_framelength, VAD_INTERNAL_SUBFRAMES_LOG2 ); + dec_subframe_offset = 0; + + /* Compute energy per sub-frame */ + /* initialize with summed energy of last subframe */ + Xnrg[ b ] = psSilk_VAD->XnrgSubfr[ b ]; + for( s = 0; s < VAD_INTERNAL_SUBFRAMES; s++ ) { + __m128i xmm_X, xmm_acc; + sumSquared = 0; + + xmm_acc = _mm_setzero_si128(); + + for( i = 0; i < dec_subframe_length - 7; i += 8 ) + { + xmm_X = _mm_loadu_si128( (__m128i *)&(X[ X_offset[ b ] + i + dec_subframe_offset ] ) ); + xmm_X = _mm_srai_epi16( xmm_X, 3 ); + xmm_X = _mm_madd_epi16( xmm_X, xmm_X ); + xmm_acc = _mm_add_epi32( xmm_acc, xmm_X ); + } + + xmm_acc = _mm_add_epi32( xmm_acc, _mm_unpackhi_epi64( xmm_acc, xmm_acc ) ); + xmm_acc = _mm_add_epi32( xmm_acc, _mm_shufflelo_epi16( xmm_acc, 0x0E ) ); + + sumSquared += _mm_cvtsi128_si32( xmm_acc ); + + for( ; i < dec_subframe_length; i++ ) { + /* The energy will be less than dec_subframe_length * ( silk_int16_MIN / 8 ) ^ 2. */ + /* Therefore we can accumulate with no risk of overflow (unless dec_subframe_length > 128) */ + x_tmp = silk_RSHIFT( + X[ X_offset[ b ] + i + dec_subframe_offset ], 3 ); + sumSquared = silk_SMLABB( sumSquared, x_tmp, x_tmp ); + + /* Safety check */ + silk_assert( sumSquared >= 0 ); + } + + /* Add/saturate summed energy of current subframe */ + if( s < VAD_INTERNAL_SUBFRAMES - 1 ) { + Xnrg[ b ] = silk_ADD_POS_SAT32( Xnrg[ b ], sumSquared ); + } else { + /* Look-ahead subframe */ + Xnrg[ b ] = silk_ADD_POS_SAT32( Xnrg[ b ], silk_RSHIFT( sumSquared, 1 ) ); + } + + dec_subframe_offset += dec_subframe_length; + } + psSilk_VAD->XnrgSubfr[ b ] = sumSquared; + } + + /********************/ + /* Noise estimation */ + /********************/ + silk_VAD_GetNoiseLevels( &Xnrg[ 0 ], psSilk_VAD ); + + /***********************************************/ + /* Signal-plus-noise to noise ratio estimation */ + /***********************************************/ + sumSquared = 0; + input_tilt = 0; + for( b = 0; b < VAD_N_BANDS; b++ ) { + speech_nrg = Xnrg[ b ] - psSilk_VAD->NL[ b ]; + if( speech_nrg > 0 ) { + /* Divide, with sufficient resolution */ + if( ( Xnrg[ b ] & 0xFF800000 ) == 0 ) { + NrgToNoiseRatio_Q8[ b ] = silk_DIV32( silk_LSHIFT( Xnrg[ b ], 8 ), psSilk_VAD->NL[ b ] + 1 ); + } else { + NrgToNoiseRatio_Q8[ b ] = silk_DIV32( Xnrg[ b ], silk_RSHIFT( psSilk_VAD->NL[ b ], 8 ) + 1 ); + } + + /* Convert to log domain */ + SNR_Q7 = silk_lin2log( NrgToNoiseRatio_Q8[ b ] ) - 8 * 128; + + /* Sum-of-squares */ + sumSquared = silk_SMLABB( sumSquared, SNR_Q7, SNR_Q7 ); /* Q14 */ + + /* Tilt measure */ + if( speech_nrg < ( (opus_int32)1 << 20 ) ) { + /* Scale down SNR value for small subband speech energies */ + SNR_Q7 = silk_SMULWB( silk_LSHIFT( silk_SQRT_APPROX( speech_nrg ), 6 ), SNR_Q7 ); + } + input_tilt = silk_SMLAWB( input_tilt, tiltWeights[ b ], SNR_Q7 ); + } else { + NrgToNoiseRatio_Q8[ b ] = 256; + } + } + + /* Mean-of-squares */ + sumSquared = silk_DIV32_16( sumSquared, VAD_N_BANDS ); /* Q14 */ + + /* Root-mean-square approximation, scale to dBs, and write to output pointer */ + pSNR_dB_Q7 = (opus_int16)( 3 * silk_SQRT_APPROX( sumSquared ) ); /* Q7 */ + + /*********************************/ + /* Speech Probability Estimation */ + /*********************************/ + SA_Q15 = silk_sigm_Q15( silk_SMULWB( VAD_SNR_FACTOR_Q16, pSNR_dB_Q7 ) - VAD_NEGATIVE_OFFSET_Q5 ); + + /**************************/ + /* Frequency Tilt Measure */ + /**************************/ + psEncC->input_tilt_Q15 = silk_LSHIFT( silk_sigm_Q15( input_tilt ) - 16384, 1 ); + + /**************************************************/ + /* Scale the sigmoid output based on power levels */ + /**************************************************/ + speech_nrg = 0; + for( b = 0; b < VAD_N_BANDS; b++ ) { + /* Accumulate signal-without-noise energies, higher frequency bands have more weight */ + speech_nrg += ( b + 1 ) * silk_RSHIFT( Xnrg[ b ] - psSilk_VAD->NL[ b ], 4 ); + } + + if( psEncC->frame_length == 20 * psEncC->fs_kHz ) { + speech_nrg = silk_RSHIFT32( speech_nrg, 1 ); + } + /* Power scaling */ + if( speech_nrg <= 0 ) { + SA_Q15 = silk_RSHIFT( SA_Q15, 1 ); + } else if( speech_nrg < 16384 ) { + speech_nrg = silk_LSHIFT32( speech_nrg, 16 ); + + /* square-root */ + speech_nrg = silk_SQRT_APPROX( speech_nrg ); + SA_Q15 = silk_SMULWB( 32768 + speech_nrg, SA_Q15 ); + } + + /* Copy the resulting speech activity in Q8 */ + psEncC->speech_activity_Q8 = silk_min_int( silk_RSHIFT( SA_Q15, 7 ), silk_uint8_MAX ); + + /***********************************/ + /* Energy Level and SNR estimation */ + /***********************************/ + /* Smoothing coefficient */ + smooth_coef_Q16 = silk_SMULWB( VAD_SNR_SMOOTH_COEF_Q18, silk_SMULWB( (opus_int32)SA_Q15, SA_Q15 ) ); + + if( psEncC->frame_length == 10 * psEncC->fs_kHz ) { + smooth_coef_Q16 >>= 1; + } + + for( b = 0; b < VAD_N_BANDS; b++ ) { + /* compute smoothed energy-to-noise ratio per band */ + psSilk_VAD->NrgRatioSmth_Q8[ b ] = silk_SMLAWB( psSilk_VAD->NrgRatioSmth_Q8[ b ], + NrgToNoiseRatio_Q8[ b ] - psSilk_VAD->NrgRatioSmth_Q8[ b ], smooth_coef_Q16 ); + + /* signal to noise ratio in dB per band */ + SNR_Q7 = 3 * ( silk_lin2log( psSilk_VAD->NrgRatioSmth_Q8[b] ) - 8 * 128 ); + /* quality = sigmoid( 0.25 * ( SNR_dB - 16 ) ); */ + psEncC->input_quality_bands_Q15[ b ] = silk_sigm_Q15( silk_RSHIFT( SNR_Q7 - 16 * 128, 4 ) ); + } + +#ifdef OPUS_CHECK_ASM + silk_assert( ret == ret_c ); + silk_assert( !memcmp( &psEncC_c, psEncC, sizeof( psEncC_c ) ) ); +#endif + + RESTORE_STACK; + return( ret ); +} diff --git a/media/libopus/silk/x86/VQ_WMat_EC_sse4_1.c b/media/libopus/silk/x86/VQ_WMat_EC_sse4_1.c new file mode 100644 index 0000000000..2c7d18d05e --- /dev/null +++ b/media/libopus/silk/x86/VQ_WMat_EC_sse4_1.c @@ -0,0 +1,173 @@ +/* Copyright (c) 2014-2020, Cisco Systems, INC + Written by XiangMingZhu WeiZhou MinPeng YanWang FrancisQuiers + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include <xmmintrin.h> +#include <emmintrin.h> +#include <smmintrin.h> +#include "main.h" +#include "celt/x86/x86cpu.h" + +/* Entropy constrained matrix-weighted VQ, hard-coded to 5-element vectors, for a single input data vector */ +void silk_VQ_WMat_EC_sse4_1( + opus_int8 *ind, /* O index of best codebook vector */ + opus_int32 *res_nrg_Q15, /* O best residual energy */ + opus_int32 *rate_dist_Q8, /* O best total bitrate */ + opus_int *gain_Q7, /* O sum of absolute LTP coefficients */ + const opus_int32 *XX_Q17, /* I correlation matrix */ + const opus_int32 *xX_Q17, /* I correlation vector */ + const opus_int8 *cb_Q7, /* I codebook */ + const opus_uint8 *cb_gain_Q7, /* I codebook effective gain */ + const opus_uint8 *cl_Q5, /* I code length for each codebook vector */ + const opus_int subfr_len, /* I number of samples per subframe */ + const opus_int32 max_gain_Q7, /* I maximum sum of absolute LTP coefficients */ + const opus_int L /* I number of vectors in codebook */ +) +{ + opus_int k, gain_tmp_Q7; + const opus_int8 *cb_row_Q7; + opus_int32 neg_xX_Q24[ 5 ]; + opus_int32 sum1_Q15, sum2_Q24; + opus_int32 bits_res_Q8, bits_tot_Q8; + __m128i v_XX_31_Q17, v_XX_42_Q17, v_cb_row_31_Q7, v_cb_row_42_Q7, v_acc1_Q24, v_acc2_Q24; + + /* Negate and convert to new Q domain */ + neg_xX_Q24[ 0 ] = -silk_LSHIFT32( xX_Q17[ 0 ], 7 ); + neg_xX_Q24[ 1 ] = -silk_LSHIFT32( xX_Q17[ 1 ], 7 ); + neg_xX_Q24[ 2 ] = -silk_LSHIFT32( xX_Q17[ 2 ], 7 ); + neg_xX_Q24[ 3 ] = -silk_LSHIFT32( xX_Q17[ 3 ], 7 ); + neg_xX_Q24[ 4 ] = -silk_LSHIFT32( xX_Q17[ 4 ], 7 ); + + v_XX_31_Q17 = _mm_loadu_si128( (__m128i *)(&XX_Q17[ 1 ] ) ); + v_XX_42_Q17 = _mm_shuffle_epi32( v_XX_31_Q17, _MM_SHUFFLE( 0, 3, 2, 1 ) ); + + /* Loop over codebook */ + *rate_dist_Q8 = silk_int32_MAX; + *res_nrg_Q15 = silk_int32_MAX; + cb_row_Q7 = cb_Q7; + /* If things go really bad, at least *ind is set to something safe. */ + *ind = 0; + for( k = 0; k < L; k++ ) { + opus_int32 penalty; + gain_tmp_Q7 = cb_gain_Q7[k]; + /* Weighted rate */ + /* Quantization error: 1 - 2 * xX * cb + cb' * XX * cb */ + sum1_Q15 = SILK_FIX_CONST( 1.001, 15 ); + + /* Penalty for too large gain */ + penalty = silk_LSHIFT32( silk_max( silk_SUB32( gain_tmp_Q7, max_gain_Q7 ), 0 ), 11 ); + + /* first row of XX_Q17 */ + v_cb_row_31_Q7 = OP_CVTEPI8_EPI32_M32( &cb_row_Q7[ 1 ] ); + v_cb_row_42_Q7 = _mm_shuffle_epi32( v_cb_row_31_Q7, _MM_SHUFFLE( 0, 3, 2, 1 ) ); + v_cb_row_31_Q7 = _mm_mul_epi32( v_XX_31_Q17, v_cb_row_31_Q7 ); + v_cb_row_42_Q7 = _mm_mul_epi32( v_XX_42_Q17, v_cb_row_42_Q7 ); + v_acc1_Q24 = _mm_add_epi64( v_cb_row_31_Q7, v_cb_row_42_Q7); + v_acc2_Q24 = _mm_shuffle_epi32( v_acc1_Q24, _MM_SHUFFLE( 1, 0, 3, 2 ) ); + v_acc1_Q24 = _mm_add_epi64( v_acc1_Q24, v_acc2_Q24); + sum2_Q24 = _mm_cvtsi128_si32( v_acc1_Q24 ); + sum2_Q24 = silk_ADD32( neg_xX_Q24[ 0 ], sum2_Q24 ); + sum2_Q24 = silk_LSHIFT32( sum2_Q24, 1 ); + sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 0 ], cb_row_Q7[ 0 ] ); + sum1_Q15 = silk_SMLAWB( sum1_Q15, sum2_Q24, cb_row_Q7[ 0 ] ); + + /* second row of XX_Q17 */ + sum2_Q24 = silk_MLA( neg_xX_Q24[ 1 ], XX_Q17[ 7 ], cb_row_Q7[ 2 ] ); + sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 8 ], cb_row_Q7[ 3 ] ); + sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 9 ], cb_row_Q7[ 4 ] ); + sum2_Q24 = silk_LSHIFT32( sum2_Q24, 1 ); + sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 6 ], cb_row_Q7[ 1 ] ); + sum1_Q15 = silk_SMLAWB( sum1_Q15, sum2_Q24, cb_row_Q7[ 1 ] ); + + /* third row of XX_Q17 */ + sum2_Q24 = silk_MLA( neg_xX_Q24[ 2 ], XX_Q17[ 13 ], cb_row_Q7[ 3 ] ); + sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 14 ], cb_row_Q7[ 4 ] ); + sum2_Q24 = silk_LSHIFT32( sum2_Q24, 1 ); + sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 12 ], cb_row_Q7[ 2 ] ); + sum1_Q15 = silk_SMLAWB( sum1_Q15, sum2_Q24, cb_row_Q7[ 2 ] ); + + /* fourth row of XX_Q17 */ + sum2_Q24 = silk_MLA( neg_xX_Q24[ 3 ], XX_Q17[ 19 ], cb_row_Q7[ 4 ] ); + sum2_Q24 = silk_LSHIFT32( sum2_Q24, 1 ); + sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 18 ], cb_row_Q7[ 3 ] ); + sum1_Q15 = silk_SMLAWB( sum1_Q15, sum2_Q24, cb_row_Q7[ 3 ] ); + + /* last row of XX_Q17 */ + sum2_Q24 = silk_LSHIFT32( neg_xX_Q24[ 4 ], 1 ); + sum2_Q24 = silk_MLA( sum2_Q24, XX_Q17[ 24 ], cb_row_Q7[ 4 ] ); + sum1_Q15 = silk_SMLAWB( sum1_Q15, sum2_Q24, cb_row_Q7[ 4 ] ); + + /* find best */ + if( sum1_Q15 >= 0 ) { + /* Translate residual energy to bits using high-rate assumption (6 dB ==> 1 bit/sample) */ + bits_res_Q8 = silk_SMULBB( subfr_len, silk_lin2log( sum1_Q15 + penalty) - (15 << 7) ); + /* In the following line we reduce the codelength component by half ("-1"); seems to slightly improve quality */ + bits_tot_Q8 = silk_ADD_LSHIFT32( bits_res_Q8, cl_Q5[ k ], 3-1 ); + if( bits_tot_Q8 <= *rate_dist_Q8 ) { + *rate_dist_Q8 = bits_tot_Q8; + *res_nrg_Q15 = sum1_Q15 + penalty; + *ind = (opus_int8)k; + *gain_Q7 = gain_tmp_Q7; + } + } + + /* Go to next cbk vector */ + cb_row_Q7 += LTP_ORDER; + } + +#ifdef OPUS_CHECK_ASM + { + opus_int8 ind_c = 0; + opus_int32 res_nrg_Q15_c = 0; + opus_int32 rate_dist_Q8_c = 0; + opus_int gain_Q7_c = 0; + + silk_VQ_WMat_EC_c( + &ind_c, + &res_nrg_Q15_c, + &rate_dist_Q8_c, + &gain_Q7_c, + XX_Q17, + xX_Q17, + cb_Q7, + cb_gain_Q7, + cl_Q5, + subfr_len, + max_gain_Q7, + L + ); + + silk_assert( *ind == ind_c ); + silk_assert( *res_nrg_Q15 == res_nrg_Q15_c ); + silk_assert( *rate_dist_Q8 == rate_dist_Q8_c ); + silk_assert( *gain_Q7 == gain_Q7_c ); + } +#endif +} diff --git a/media/libopus/silk/x86/main_sse.h b/media/libopus/silk/x86/main_sse.h new file mode 100644 index 0000000000..a01d7f6c75 --- /dev/null +++ b/media/libopus/silk/x86/main_sse.h @@ -0,0 +1,247 @@ +/* Copyright (c) 2014, Cisco Systems, INC + Written by XiangMingZhu WeiZhou MinPeng YanWang + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef MAIN_SSE_H +# define MAIN_SSE_H + +# ifdef HAVE_CONFIG_H +# include "config.h" +# endif + +# if defined(OPUS_X86_MAY_HAVE_SSE4_1) + +void silk_VQ_WMat_EC_sse4_1( + opus_int8 *ind, /* O index of best codebook vector */ + opus_int32 *res_nrg_Q15, /* O best residual energy */ + opus_int32 *rate_dist_Q8, /* O best total bitrate */ + opus_int *gain_Q7, /* O sum of absolute LTP coefficients */ + const opus_int32 *XX_Q17, /* I correlation matrix */ + const opus_int32 *xX_Q17, /* I correlation vector */ + const opus_int8 *cb_Q7, /* I codebook */ + const opus_uint8 *cb_gain_Q7, /* I codebook effective gain */ + const opus_uint8 *cl_Q5, /* I code length for each codebook vector */ + const opus_int subfr_len, /* I number of samples per subframe */ + const opus_int32 max_gain_Q7, /* I maximum sum of absolute LTP coefficients */ + const opus_int L /* I number of vectors in codebook */ +); + +# if defined OPUS_X86_PRESUME_SSE4_1 + +# define OVERRIDE_silk_VQ_WMat_EC +# define silk_VQ_WMat_EC(ind, res_nrg_Q15, rate_dist_Q8, gain_Q7, XX_Q17, xX_Q17, cb_Q7, cb_gain_Q7, cl_Q5, \ + subfr_len, max_gain_Q7, L, arch) \ + ((void)(arch),silk_VQ_WMat_EC_sse4_1(ind, res_nrg_Q15, rate_dist_Q8, gain_Q7, XX_Q17, xX_Q17, cb_Q7, cb_gain_Q7, cl_Q5, \ + subfr_len, max_gain_Q7, L)) + +# elif defined(OPUS_HAVE_RTCD) + +extern void (*const SILK_VQ_WMAT_EC_IMPL[OPUS_ARCHMASK + 1])( + opus_int8 *ind, /* O index of best codebook vector */ + opus_int32 *res_nrg_Q15, /* O best residual energy */ + opus_int32 *rate_dist_Q8, /* O best total bitrate */ + opus_int *gain_Q7, /* O sum of absolute LTP coefficients */ + const opus_int32 *XX_Q17, /* I correlation matrix */ + const opus_int32 *xX_Q17, /* I correlation vector */ + const opus_int8 *cb_Q7, /* I codebook */ + const opus_uint8 *cb_gain_Q7, /* I codebook effective gain */ + const opus_uint8 *cl_Q5, /* I code length for each codebook vector */ + const opus_int subfr_len, /* I number of samples per subframe */ + const opus_int32 max_gain_Q7, /* I maximum sum of absolute LTP coefficients */ + const opus_int L /* I number of vectors in codebook */ +); + +# define OVERRIDE_silk_VQ_WMat_EC +# define silk_VQ_WMat_EC(ind, res_nrg_Q15, rate_dist_Q8, gain_Q7, XX_Q17, xX_Q17, cb_Q7, cb_gain_Q7, cl_Q5, \ + subfr_len, max_gain_Q7, L, arch) \ + ((*SILK_VQ_WMAT_EC_IMPL[(arch) & OPUS_ARCHMASK])(ind, res_nrg_Q15, rate_dist_Q8, gain_Q7, XX_Q17, xX_Q17, cb_Q7, cb_gain_Q7, cl_Q5, \ + subfr_len, max_gain_Q7, L)) + +# endif + +void silk_NSQ_sse4_1( + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + SideInfoIndices *psIndices, /* I/O Quantization Indices */ + const opus_int16 x16[], /* I Input */ + opus_int8 pulses[], /* O Quantized pulse signal */ + const opus_int16 PredCoef_Q12[ 2 * MAX_LPC_ORDER ], /* I Short term prediction coefs */ + const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ + const opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ + const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ + const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ + const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ + const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ + const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ + const opus_int LTP_scale_Q14 /* I LTP state scaling */ +); + +# if defined OPUS_X86_PRESUME_SSE4_1 + +# define OVERRIDE_silk_NSQ +# define silk_NSQ(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ + HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14, arch) \ + ((void)(arch),silk_NSQ_sse4_1(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ + HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14)) + +# elif defined(OPUS_HAVE_RTCD) + +extern void (*const SILK_NSQ_IMPL[OPUS_ARCHMASK + 1])( + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + SideInfoIndices *psIndices, /* I/O Quantization Indices */ + const opus_int16 x16[], /* I Input */ + opus_int8 pulses[], /* O Quantized pulse signal */ + const opus_int16 PredCoef_Q12[ 2 * MAX_LPC_ORDER ], /* I Short term prediction coefs */ + const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ + const opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ + const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ + const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ + const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ + const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ + const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ + const opus_int LTP_scale_Q14 /* I LTP state scaling */ +); + +# define OVERRIDE_silk_NSQ +# define silk_NSQ(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ + HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14, arch) \ + ((*SILK_NSQ_IMPL[(arch) & OPUS_ARCHMASK])(psEncC, NSQ, psIndices, x_Q3, pulses, PredCoef_Q12, LTPCoef_Q14, AR2_Q13, \ + HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14)) + +# endif + +void silk_NSQ_del_dec_sse4_1( + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + SideInfoIndices *psIndices, /* I/O Quantization Indices */ + const opus_int16 x16[], /* I Input */ + opus_int8 pulses[], /* O Quantized pulse signal */ + const opus_int16 PredCoef_Q12[ 2 * MAX_LPC_ORDER ], /* I Short term prediction coefs */ + const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ + const opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ + const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ + const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ + const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ + const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ + const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ + const opus_int LTP_scale_Q14 /* I LTP state scaling */ +); + +# if defined OPUS_X86_PRESUME_SSE4_1 + +# define OVERRIDE_silk_NSQ_del_dec +# define silk_NSQ_del_dec(psEncC, NSQ, psIndices, x16, pulses, PredCoef_Q12, LTPCoef_Q14, AR_Q13, \ + HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14, arch) \ + ((void)(arch),silk_NSQ_del_dec_sse4_1(psEncC, NSQ, psIndices, x16, pulses, PredCoef_Q12, LTPCoef_Q14, AR_Q13, \ + HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14)) + +# elif defined(OPUS_HAVE_RTCD) + +extern void (*const SILK_NSQ_DEL_DEC_IMPL[OPUS_ARCHMASK + 1])( + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + SideInfoIndices *psIndices, /* I/O Quantization Indices */ + const opus_int16 x16[], /* I Input */ + opus_int8 pulses[], /* O Quantized pulse signal */ + const opus_int16 PredCoef_Q12[ 2 * MAX_LPC_ORDER ], /* I Short term prediction coefs */ + const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ + const opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ + const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ + const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ + const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ + const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ + const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ + const opus_int LTP_scale_Q14 /* I LTP state scaling */ +); + +# define OVERRIDE_silk_NSQ_del_dec +# define silk_NSQ_del_dec(psEncC, NSQ, psIndices, x16, pulses, PredCoef_Q12, LTPCoef_Q14, AR_Q13, \ + HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14, arch) \ + ((*SILK_NSQ_DEL_DEC_IMPL[(arch) & OPUS_ARCHMASK])(psEncC, NSQ, psIndices, x16, pulses, PredCoef_Q12, LTPCoef_Q14, AR_Q13, \ + HarmShapeGain_Q14, Tilt_Q14, LF_shp_Q14, Gains_Q16, pitchL, Lambda_Q10, LTP_scale_Q14)) + +# endif + +void silk_noise_shape_quantizer( + silk_nsq_state *NSQ, /* I/O NSQ state */ + opus_int signalType, /* I Signal type */ + const opus_int32 x_sc_Q10[], /* I */ + opus_int8 pulses[], /* O */ + opus_int16 xq[], /* O */ + opus_int32 sLTP_Q15[], /* I/O LTP state */ + const opus_int16 a_Q12[], /* I Short term prediction coefs */ + const opus_int16 b_Q14[], /* I Long term prediction coefs */ + const opus_int16 AR_shp_Q13[], /* I Noise shaping AR coefs */ + opus_int lag, /* I Pitch lag */ + opus_int32 HarmShapeFIRPacked_Q14, /* I */ + opus_int Tilt_Q14, /* I Spectral tilt */ + opus_int32 LF_shp_Q14, /* I */ + opus_int32 Gain_Q16, /* I */ + opus_int Lambda_Q10, /* I */ + opus_int offset_Q10, /* I */ + opus_int length, /* I Input length */ + opus_int shapingLPCOrder, /* I Noise shaping AR filter order */ + opus_int predictLPCOrder, /* I Prediction filter order */ + int arch /* I Architecture */ +); + +/**************************/ +/* Noise level estimation */ +/**************************/ +void silk_VAD_GetNoiseLevels( + const opus_int32 pX[ VAD_N_BANDS ], /* I subband energies */ + silk_VAD_state *psSilk_VAD /* I/O Pointer to Silk VAD state */ +); + +opus_int silk_VAD_GetSA_Q8_sse4_1( + silk_encoder_state *psEnC, + const opus_int16 pIn[] +); + +# if defined(OPUS_X86_PRESUME_SSE4_1) + +# define OVERRIDE_silk_VAD_GetSA_Q8 +# define silk_VAD_GetSA_Q8(psEnC, pIn, arch) ((void)(arch),silk_VAD_GetSA_Q8_sse4_1(psEnC, pIn)) + +# elif defined(OPUS_HAVE_RTCD) + +extern opus_int (*const SILK_VAD_GETSA_Q8_IMPL[OPUS_ARCHMASK + 1])( + silk_encoder_state *psEnC, + const opus_int16 pIn[]); + +# define OVERRIDE_silk_VAD_GetSA_Q8 +# define silk_VAD_GetSA_Q8(psEnC, pIn, arch) \ + ((*SILK_VAD_GETSA_Q8_IMPL[(arch) & OPUS_ARCHMASK])(psEnC, pIn)) + +# endif + +# endif +#endif diff --git a/media/libopus/silk/x86/x86_silk_map.c b/media/libopus/silk/x86/x86_silk_map.c new file mode 100644 index 0000000000..70f60078cf --- /dev/null +++ b/media/libopus/silk/x86/x86_silk_map.c @@ -0,0 +1,159 @@ +/* Copyright (c) 2014, Cisco Systems, INC + Written by XiangMingZhu WeiZhou MinPeng YanWang + + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#if defined(HAVE_CONFIG_H) +#include "config.h" +#endif + +#include "celt/x86/x86cpu.h" +#include "structs.h" +#include "SigProc_FIX.h" +#include "pitch.h" +#include "main.h" + +#if defined(OPUS_HAVE_RTCD) && !defined(OPUS_X86_PRESUME_SSE4_1) + +#if defined(FIXED_POINT) + +#include "fixed/main_FIX.h" + +opus_int64 (*const SILK_INNER_PROD16_IMPL[ OPUS_ARCHMASK + 1 ] )( + const opus_int16 *inVec1, + const opus_int16 *inVec2, + const opus_int len +) = { + silk_inner_prod16_c, /* non-sse */ + silk_inner_prod16_c, + silk_inner_prod16_c, + MAY_HAVE_SSE4_1( silk_inner_prod16 ), /* sse4.1 */ + MAY_HAVE_SSE4_1( silk_inner_prod16 ) /* avx */ +}; + +#endif + +opus_int (*const SILK_VAD_GETSA_Q8_IMPL[ OPUS_ARCHMASK + 1 ] )( + silk_encoder_state *psEncC, + const opus_int16 pIn[] +) = { + silk_VAD_GetSA_Q8_c, /* non-sse */ + silk_VAD_GetSA_Q8_c, + silk_VAD_GetSA_Q8_c, + MAY_HAVE_SSE4_1( silk_VAD_GetSA_Q8 ), /* sse4.1 */ + MAY_HAVE_SSE4_1( silk_VAD_GetSA_Q8 ) /* avx */ +}; + +void (*const SILK_NSQ_IMPL[ OPUS_ARCHMASK + 1 ] )( + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + SideInfoIndices *psIndices, /* I/O Quantization Indices */ + const opus_int16 x16[], /* I Input */ + opus_int8 pulses[], /* O Quantized pulse signal */ + const opus_int16 PredCoef_Q12[ 2 * MAX_LPC_ORDER ], /* I Short term prediction coefs */ + const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ + const opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ + const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ + const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ + const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ + const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ + const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ + const opus_int LTP_scale_Q14 /* I LTP state scaling */ +) = { + silk_NSQ_c, /* non-sse */ + silk_NSQ_c, + silk_NSQ_c, + MAY_HAVE_SSE4_1( silk_NSQ ), /* sse4.1 */ + MAY_HAVE_SSE4_1( silk_NSQ ) /* avx */ +}; + +void (*const SILK_VQ_WMAT_EC_IMPL[ OPUS_ARCHMASK + 1 ] )( + opus_int8 *ind, /* O index of best codebook vector */ + opus_int32 *res_nrg_Q15, /* O best residual energy */ + opus_int32 *rate_dist_Q8, /* O best total bitrate */ + opus_int *gain_Q7, /* O sum of absolute LTP coefficients */ + const opus_int32 *XX_Q17, /* I correlation matrix */ + const opus_int32 *xX_Q17, /* I correlation vector */ + const opus_int8 *cb_Q7, /* I codebook */ + const opus_uint8 *cb_gain_Q7, /* I codebook effective gain */ + const opus_uint8 *cl_Q5, /* I code length for each codebook vector */ + const opus_int subfr_len, /* I number of samples per subframe */ + const opus_int32 max_gain_Q7, /* I maximum sum of absolute LTP coefficients */ + const opus_int L /* I number of vectors in codebook */ +) = { + silk_VQ_WMat_EC_c, /* non-sse */ + silk_VQ_WMat_EC_c, + silk_VQ_WMat_EC_c, + MAY_HAVE_SSE4_1( silk_VQ_WMat_EC ), /* sse4.1 */ + MAY_HAVE_SSE4_1( silk_VQ_WMat_EC ) /* avx */ +}; + +void (*const SILK_NSQ_DEL_DEC_IMPL[ OPUS_ARCHMASK + 1 ] )( + const silk_encoder_state *psEncC, /* I Encoder State */ + silk_nsq_state *NSQ, /* I/O NSQ state */ + SideInfoIndices *psIndices, /* I/O Quantization Indices */ + const opus_int16 x16[], /* I Input */ + opus_int8 pulses[], /* O Quantized pulse signal */ + const opus_int16 PredCoef_Q12[ 2 * MAX_LPC_ORDER ], /* I Short term prediction coefs */ + const opus_int16 LTPCoef_Q14[ LTP_ORDER * MAX_NB_SUBFR ], /* I Long term prediction coefs */ + const opus_int16 AR_Q13[ MAX_NB_SUBFR * MAX_SHAPE_LPC_ORDER ], /* I Noise shaping coefs */ + const opus_int HarmShapeGain_Q14[ MAX_NB_SUBFR ], /* I Long term shaping coefs */ + const opus_int Tilt_Q14[ MAX_NB_SUBFR ], /* I Spectral tilt */ + const opus_int32 LF_shp_Q14[ MAX_NB_SUBFR ], /* I Low frequency shaping coefs */ + const opus_int32 Gains_Q16[ MAX_NB_SUBFR ], /* I Quantization step sizes */ + const opus_int pitchL[ MAX_NB_SUBFR ], /* I Pitch lags */ + const opus_int Lambda_Q10, /* I Rate/distortion tradeoff */ + const opus_int LTP_scale_Q14 /* I LTP state scaling */ +) = { + silk_NSQ_del_dec_c, /* non-sse */ + silk_NSQ_del_dec_c, + silk_NSQ_del_dec_c, + MAY_HAVE_SSE4_1( silk_NSQ_del_dec ), /* sse4.1 */ + MAY_HAVE_SSE4_1( silk_NSQ_del_dec ) /* avx */ +}; + +#if defined(FIXED_POINT) + +void (*const SILK_BURG_MODIFIED_IMPL[ OPUS_ARCHMASK + 1 ] )( + opus_int32 *res_nrg, /* O Residual energy */ + opus_int *res_nrg_Q, /* O Residual energy Q value */ + opus_int32 A_Q16[], /* O Prediction coefficients (length order) */ + const opus_int16 x[], /* I Input signal, length: nb_subfr * ( D + subfr_length ) */ + const opus_int32 minInvGain_Q30, /* I Inverse of max prediction gain */ + const opus_int subfr_length, /* I Input signal subframe length (incl. D preceding samples) */ + const opus_int nb_subfr, /* I Number of subframes stacked in x */ + const opus_int D, /* I Order */ + int arch /* I Run-time architecture */ +) = { + silk_burg_modified_c, /* non-sse */ + silk_burg_modified_c, + silk_burg_modified_c, + MAY_HAVE_SSE4_1( silk_burg_modified ), /* sse4.1 */ + MAY_HAVE_SSE4_1( silk_burg_modified ) /* avx */ +}; + +#endif +#endif diff --git a/media/libopus/sources.mozbuild b/media/libopus/sources.mozbuild new file mode 100644 index 0000000000..891f5b32ae --- /dev/null +++ b/media/libopus/sources.mozbuild @@ -0,0 +1,251 @@ +# THIS FILE WAS AUTOMATICALLY GENERATED BY gen-sources.py. DO NOT EDIT. +celt_sources = [ + 'celt/bands.c', + 'celt/celt_lpc.c', + 'celt/cwrs.c', + 'celt/entcode.c', + 'celt/entdec.c', + 'celt/entenc.c', + 'celt/kiss_fft.c', + 'celt/laplace.c', + 'celt/mathops.c', + 'celt/mdct.c', + 'celt/modes.c', + 'celt/pitch.c', + 'celt/quant_bands.c', + 'celt/rate.c', + 'celt/vq.c', +] + +celt_sources_x86_rtcd = [ + 'celt/x86/x86_celt_map.c', + 'celt/x86/x86cpu.c', +] + +opus_nonunified_sources = [ + # Disabled because of undefined reference to celt_fatal at link time + 'celt/celt.c', + # Disabled because of name clash of opus_custom_encoder_get_size. + 'celt/celt_decoder.c', + 'celt/celt_encoder.c', + # Disabled for (safe) warning about QA redefinition. + 'silk/LPC_inv_pred_gain.c', + 'silk/NLSF2A.c', +] + +celt_sources_sse = [ + 'celt/x86/pitch_sse.c', +] + +celt_sources_sse2 = [ + 'celt/x86/pitch_sse2.c', + 'celt/x86/vq_sse2.c', +] + +celt_sources_sse4_1 = [ + 'celt/x86/celt_lpc_sse4_1.c', + 'celt/x86/pitch_sse4_1.c', +] + +celt_sources_arm_rtcd = [ + 'celt/arm/arm_celt_map.c', + 'celt/arm/armcpu.c', +] + +celt_sources_arm_asm = [ + 'celt/arm/celt_pitch_xcorr_arm.s', +] + +celt_am_sources_arm_asm = [ + 'celt/arm/armopts.s.in', +] + +celt_sources_arm_neon_intr = [ + 'celt/arm/celt_neon_intr.c', + 'celt/arm/pitch_neon_intr.c', +] + +celt_sources_arm_ne10 = [ + 'celt/arm/celt_fft_ne10.c', + 'celt/arm/celt_mdct_ne10.c', +] + +opus_sources = [ + 'src/opus.c', + 'src/opus_decoder.c', + 'src/opus_encoder.c', + 'src/opus_multistream.c', + 'src/opus_multistream_decoder.c', + 'src/opus_multistream_encoder.c', + 'src/repacketizer.c', +] + +opus_sources_float = [ + 'src/analysis.c', + 'src/mlp.c', + 'src/mlp_data.c', +] + +silk_sources = [ + 'silk/A2NLSF.c', + 'silk/ana_filt_bank_1.c', + 'silk/biquad_alt.c', + 'silk/bwexpander.c', + 'silk/bwexpander_32.c', + 'silk/check_control_input.c', + 'silk/CNG.c', + 'silk/code_signs.c', + 'silk/control_audio_bandwidth.c', + 'silk/control_codec.c', + 'silk/control_SNR.c', + 'silk/dec_API.c', + 'silk/decode_core.c', + 'silk/decode_frame.c', + 'silk/decode_indices.c', + 'silk/decode_parameters.c', + 'silk/decode_pitch.c', + 'silk/decode_pulses.c', + 'silk/decoder_set_fs.c', + 'silk/enc_API.c', + 'silk/encode_indices.c', + 'silk/encode_pulses.c', + 'silk/gain_quant.c', + 'silk/HP_variable_cutoff.c', + 'silk/init_decoder.c', + 'silk/init_encoder.c', + 'silk/inner_prod_aligned.c', + 'silk/interpolate.c', + 'silk/lin2log.c', + 'silk/log2lin.c', + 'silk/LP_variable_cutoff.c', + 'silk/LPC_analysis_filter.c', + 'silk/LPC_fit.c', + 'silk/NLSF_decode.c', + 'silk/NLSF_del_dec_quant.c', + 'silk/NLSF_encode.c', + 'silk/NLSF_stabilize.c', + 'silk/NLSF_unpack.c', + 'silk/NLSF_VQ.c', + 'silk/NLSF_VQ_weights_laroia.c', + 'silk/NSQ.c', + 'silk/NSQ_del_dec.c', + 'silk/pitch_est_tables.c', + 'silk/PLC.c', + 'silk/process_NLSFs.c', + 'silk/quant_LTP_gains.c', + 'silk/resampler.c', + 'silk/resampler_down2.c', + 'silk/resampler_down2_3.c', + 'silk/resampler_private_AR2.c', + 'silk/resampler_private_down_FIR.c', + 'silk/resampler_private_IIR_FIR.c', + 'silk/resampler_private_up2_HQ.c', + 'silk/resampler_rom.c', + 'silk/shell_coder.c', + 'silk/sigm_Q15.c', + 'silk/sort.c', + 'silk/stereo_decode_pred.c', + 'silk/stereo_encode_pred.c', + 'silk/stereo_find_predictor.c', + 'silk/stereo_LR_to_MS.c', + 'silk/stereo_MS_to_LR.c', + 'silk/stereo_quant_pred.c', + 'silk/sum_sqr_shift.c', + 'silk/table_LSF_cos.c', + 'silk/tables_gain.c', + 'silk/tables_LTP.c', + 'silk/tables_NLSF_CB_NB_MB.c', + 'silk/tables_NLSF_CB_WB.c', + 'silk/tables_other.c', + 'silk/tables_pitch_lag.c', + 'silk/tables_pulses_per_block.c', + 'silk/VAD.c', + 'silk/VQ_WMat_EC.c', +] + +silk_sources_x86_rtcd = [ + 'silk/x86/x86_silk_map.c', +] + +silk_sources_sse4_1 = [ + 'silk/x86/NSQ_del_dec_sse4_1.c', + 'silk/x86/NSQ_sse4_1.c', + 'silk/x86/VAD_sse4_1.c', + 'silk/x86/VQ_WMat_EC_sse4_1.c', +] + +silk_sources_arm_rtcd = [ + 'silk/arm/arm_silk_map.c', +] + +silk_sources_arm_neon_intr = [ + 'silk/arm/biquad_alt_neon_intr.c', + 'silk/arm/LPC_inv_pred_gain_neon_intr.c', + 'silk/arm/NSQ_del_dec_neon_intr.c', + 'silk/arm/NSQ_neon.c', +] + +silk_sources_fixed = [ + 'silk/fixed/apply_sine_window_FIX.c', + 'silk/fixed/autocorr_FIX.c', + 'silk/fixed/burg_modified_FIX.c', + 'silk/fixed/corrMatrix_FIX.c', + 'silk/fixed/encode_frame_FIX.c', + 'silk/fixed/find_LPC_FIX.c', + 'silk/fixed/find_LTP_FIX.c', + 'silk/fixed/find_pitch_lags_FIX.c', + 'silk/fixed/find_pred_coefs_FIX.c', + 'silk/fixed/k2a_FIX.c', + 'silk/fixed/k2a_Q16_FIX.c', + 'silk/fixed/LTP_analysis_filter_FIX.c', + 'silk/fixed/LTP_scale_ctrl_FIX.c', + 'silk/fixed/noise_shape_analysis_FIX.c', + 'silk/fixed/pitch_analysis_core_FIX.c', + 'silk/fixed/process_gains_FIX.c', + 'silk/fixed/regularize_correlations_FIX.c', + 'silk/fixed/residual_energy16_FIX.c', + 'silk/fixed/residual_energy_FIX.c', + 'silk/fixed/schur64_FIX.c', + 'silk/fixed/schur_FIX.c', + 'silk/fixed/vector_ops_FIX.c', + 'silk/fixed/warped_autocorrelation_FIX.c', +] + +silk_sources_fixed_sse4_1 = [ + 'silk/fixed/x86/burg_modified_FIX_sse4_1.c', + 'silk/fixed/x86/vector_ops_FIX_sse4_1.c', +] + +silk_sources_fixed_arm_neon_intr = [ + 'silk/fixed/arm/warped_autocorrelation_FIX_neon_intr.c', +] + +silk_sources_float = [ + 'silk/float/apply_sine_window_FLP.c', + 'silk/float/autocorrelation_FLP.c', + 'silk/float/burg_modified_FLP.c', + 'silk/float/bwexpander_FLP.c', + 'silk/float/corrMatrix_FLP.c', + 'silk/float/encode_frame_FLP.c', + 'silk/float/energy_FLP.c', + 'silk/float/find_LPC_FLP.c', + 'silk/float/find_LTP_FLP.c', + 'silk/float/find_pitch_lags_FLP.c', + 'silk/float/find_pred_coefs_FLP.c', + 'silk/float/inner_product_FLP.c', + 'silk/float/k2a_FLP.c', + 'silk/float/LPC_analysis_filter_FLP.c', + 'silk/float/LTP_analysis_filter_FLP.c', + 'silk/float/LTP_scale_ctrl_FLP.c', + 'silk/float/noise_shape_analysis_FLP.c', + 'silk/float/pitch_analysis_core_FLP.c', + 'silk/float/process_gains_FLP.c', + 'silk/float/residual_energy_FLP.c', + 'silk/float/scale_copy_vector_FLP.c', + 'silk/float/scale_vector_FLP.c', + 'silk/float/schur_FLP.c', + 'silk/float/sort_FLP.c', + 'silk/float/warped_autocorrelation_FLP.c', + 'silk/float/wrappers_FLP.c', +] + diff --git a/media/libopus/src/analysis.c b/media/libopus/src/analysis.c new file mode 100644 index 0000000000..058328f0fd --- /dev/null +++ b/media/libopus/src/analysis.c @@ -0,0 +1,983 @@ +/* Copyright (c) 2011 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR + CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#define ANALYSIS_C + +#ifdef MLP_TRAINING +#include <stdio.h> +#endif + +#include "mathops.h" +#include "kiss_fft.h" +#include "celt.h" +#include "modes.h" +#include "arch.h" +#include "quant_bands.h" +#include "analysis.h" +#include "mlp.h" +#include "stack_alloc.h" +#include "float_cast.h" + +#ifndef M_PI +#define M_PI 3.141592653 +#endif + +#ifndef DISABLE_FLOAT_API + +#define TRANSITION_PENALTY 10 + +static const float dct_table[128] = { + 0.250000f, 0.250000f, 0.250000f, 0.250000f, 0.250000f, 0.250000f, 0.250000f, 0.250000f, + 0.250000f, 0.250000f, 0.250000f, 0.250000f, 0.250000f, 0.250000f, 0.250000f, 0.250000f, + 0.351851f, 0.338330f, 0.311806f, 0.273300f, 0.224292f, 0.166664f, 0.102631f, 0.034654f, + -0.034654f,-0.102631f,-0.166664f,-0.224292f,-0.273300f,-0.311806f,-0.338330f,-0.351851f, + 0.346760f, 0.293969f, 0.196424f, 0.068975f,-0.068975f,-0.196424f,-0.293969f,-0.346760f, + -0.346760f,-0.293969f,-0.196424f,-0.068975f, 0.068975f, 0.196424f, 0.293969f, 0.346760f, + 0.338330f, 0.224292f, 0.034654f,-0.166664f,-0.311806f,-0.351851f,-0.273300f,-0.102631f, + 0.102631f, 0.273300f, 0.351851f, 0.311806f, 0.166664f,-0.034654f,-0.224292f,-0.338330f, + 0.326641f, 0.135299f,-0.135299f,-0.326641f,-0.326641f,-0.135299f, 0.135299f, 0.326641f, + 0.326641f, 0.135299f,-0.135299f,-0.326641f,-0.326641f,-0.135299f, 0.135299f, 0.326641f, + 0.311806f, 0.034654f,-0.273300f,-0.338330f,-0.102631f, 0.224292f, 0.351851f, 0.166664f, + -0.166664f,-0.351851f,-0.224292f, 0.102631f, 0.338330f, 0.273300f,-0.034654f,-0.311806f, + 0.293969f,-0.068975f,-0.346760f,-0.196424f, 0.196424f, 0.346760f, 0.068975f,-0.293969f, + -0.293969f, 0.068975f, 0.346760f, 0.196424f,-0.196424f,-0.346760f,-0.068975f, 0.293969f, + 0.273300f,-0.166664f,-0.338330f, 0.034654f, 0.351851f, 0.102631f,-0.311806f,-0.224292f, + 0.224292f, 0.311806f,-0.102631f,-0.351851f,-0.034654f, 0.338330f, 0.166664f,-0.273300f, +}; + +static const float analysis_window[240] = { + 0.000043f, 0.000171f, 0.000385f, 0.000685f, 0.001071f, 0.001541f, 0.002098f, 0.002739f, + 0.003466f, 0.004278f, 0.005174f, 0.006156f, 0.007222f, 0.008373f, 0.009607f, 0.010926f, + 0.012329f, 0.013815f, 0.015385f, 0.017037f, 0.018772f, 0.020590f, 0.022490f, 0.024472f, + 0.026535f, 0.028679f, 0.030904f, 0.033210f, 0.035595f, 0.038060f, 0.040604f, 0.043227f, + 0.045928f, 0.048707f, 0.051564f, 0.054497f, 0.057506f, 0.060591f, 0.063752f, 0.066987f, + 0.070297f, 0.073680f, 0.077136f, 0.080665f, 0.084265f, 0.087937f, 0.091679f, 0.095492f, + 0.099373f, 0.103323f, 0.107342f, 0.111427f, 0.115579f, 0.119797f, 0.124080f, 0.128428f, + 0.132839f, 0.137313f, 0.141849f, 0.146447f, 0.151105f, 0.155823f, 0.160600f, 0.165435f, + 0.170327f, 0.175276f, 0.180280f, 0.185340f, 0.190453f, 0.195619f, 0.200838f, 0.206107f, + 0.211427f, 0.216797f, 0.222215f, 0.227680f, 0.233193f, 0.238751f, 0.244353f, 0.250000f, + 0.255689f, 0.261421f, 0.267193f, 0.273005f, 0.278856f, 0.284744f, 0.290670f, 0.296632f, + 0.302628f, 0.308658f, 0.314721f, 0.320816f, 0.326941f, 0.333097f, 0.339280f, 0.345492f, + 0.351729f, 0.357992f, 0.364280f, 0.370590f, 0.376923f, 0.383277f, 0.389651f, 0.396044f, + 0.402455f, 0.408882f, 0.415325f, 0.421783f, 0.428254f, 0.434737f, 0.441231f, 0.447736f, + 0.454249f, 0.460770f, 0.467298f, 0.473832f, 0.480370f, 0.486912f, 0.493455f, 0.500000f, + 0.506545f, 0.513088f, 0.519630f, 0.526168f, 0.532702f, 0.539230f, 0.545751f, 0.552264f, + 0.558769f, 0.565263f, 0.571746f, 0.578217f, 0.584675f, 0.591118f, 0.597545f, 0.603956f, + 0.610349f, 0.616723f, 0.623077f, 0.629410f, 0.635720f, 0.642008f, 0.648271f, 0.654508f, + 0.660720f, 0.666903f, 0.673059f, 0.679184f, 0.685279f, 0.691342f, 0.697372f, 0.703368f, + 0.709330f, 0.715256f, 0.721144f, 0.726995f, 0.732807f, 0.738579f, 0.744311f, 0.750000f, + 0.755647f, 0.761249f, 0.766807f, 0.772320f, 0.777785f, 0.783203f, 0.788573f, 0.793893f, + 0.799162f, 0.804381f, 0.809547f, 0.814660f, 0.819720f, 0.824724f, 0.829673f, 0.834565f, + 0.839400f, 0.844177f, 0.848895f, 0.853553f, 0.858151f, 0.862687f, 0.867161f, 0.871572f, + 0.875920f, 0.880203f, 0.884421f, 0.888573f, 0.892658f, 0.896677f, 0.900627f, 0.904508f, + 0.908321f, 0.912063f, 0.915735f, 0.919335f, 0.922864f, 0.926320f, 0.929703f, 0.933013f, + 0.936248f, 0.939409f, 0.942494f, 0.945503f, 0.948436f, 0.951293f, 0.954072f, 0.956773f, + 0.959396f, 0.961940f, 0.964405f, 0.966790f, 0.969096f, 0.971321f, 0.973465f, 0.975528f, + 0.977510f, 0.979410f, 0.981228f, 0.982963f, 0.984615f, 0.986185f, 0.987671f, 0.989074f, + 0.990393f, 0.991627f, 0.992778f, 0.993844f, 0.994826f, 0.995722f, 0.996534f, 0.997261f, + 0.997902f, 0.998459f, 0.998929f, 0.999315f, 0.999615f, 0.999829f, 0.999957f, 1.000000f, +}; + +static const int tbands[NB_TBANDS+1] = { + 4, 8, 12, 16, 20, 24, 28, 32, 40, 48, 56, 64, 80, 96, 112, 136, 160, 192, 240 +}; + +#define NB_TONAL_SKIP_BANDS 9 + +static opus_val32 silk_resampler_down2_hp( + opus_val32 *S, /* I/O State vector [ 2 ] */ + opus_val32 *out, /* O Output signal [ floor(len/2) ] */ + const opus_val32 *in, /* I Input signal [ len ] */ + int inLen /* I Number of input samples */ +) +{ + int k, len2 = inLen/2; + opus_val32 in32, out32, out32_hp, Y, X; + opus_val64 hp_ener = 0; + /* Internal variables and state are in Q10 format */ + for( k = 0; k < len2; k++ ) { + /* Convert to Q10 */ + in32 = in[ 2 * k ]; + + /* All-pass section for even input sample */ + Y = SUB32( in32, S[ 0 ] ); + X = MULT16_32_Q15(QCONST16(0.6074371f, 15), Y); + out32 = ADD32( S[ 0 ], X ); + S[ 0 ] = ADD32( in32, X ); + out32_hp = out32; + /* Convert to Q10 */ + in32 = in[ 2 * k + 1 ]; + + /* All-pass section for odd input sample, and add to output of previous section */ + Y = SUB32( in32, S[ 1 ] ); + X = MULT16_32_Q15(QCONST16(0.15063f, 15), Y); + out32 = ADD32( out32, S[ 1 ] ); + out32 = ADD32( out32, X ); + S[ 1 ] = ADD32( in32, X ); + + Y = SUB32( -in32, S[ 2 ] ); + X = MULT16_32_Q15(QCONST16(0.15063f, 15), Y); + out32_hp = ADD32( out32_hp, S[ 2 ] ); + out32_hp = ADD32( out32_hp, X ); + S[ 2 ] = ADD32( -in32, X ); + + hp_ener += out32_hp*(opus_val64)out32_hp; + /* Add, convert back to int16 and store to output */ + out[ k ] = HALF32(out32); + } +#ifdef FIXED_POINT + /* len2 can be up to 480, so we shift by 8 more to make it fit. */ + hp_ener = hp_ener >> (2*SIG_SHIFT + 8); +#endif + return (opus_val32)hp_ener; +} + +static opus_val32 downmix_and_resample(downmix_func downmix, const void *_x, opus_val32 *y, opus_val32 S[3], int subframe, int offset, int c1, int c2, int C, int Fs) +{ + VARDECL(opus_val32, tmp); + opus_val32 scale; + int j; + opus_val32 ret = 0; + SAVE_STACK; + + if (subframe==0) return 0; + if (Fs == 48000) + { + subframe *= 2; + offset *= 2; + } else if (Fs == 16000) { + subframe = subframe*2/3; + offset = offset*2/3; + } + ALLOC(tmp, subframe, opus_val32); + + downmix(_x, tmp, subframe, offset, c1, c2, C); +#ifdef FIXED_POINT + scale = (1<<SIG_SHIFT); +#else + scale = 1.f/32768; +#endif + if (c2==-2) + scale /= C; + else if (c2>-1) + scale /= 2; + for (j=0;j<subframe;j++) + tmp[j] *= scale; + if (Fs == 48000) + { + ret = silk_resampler_down2_hp(S, y, tmp, subframe); + } else if (Fs == 24000) { + OPUS_COPY(y, tmp, subframe); + } else if (Fs == 16000) { + VARDECL(opus_val32, tmp3x); + ALLOC(tmp3x, 3*subframe, opus_val32); + /* Don't do this at home! This resampler is horrible and it's only (barely) + usable for the purpose of the analysis because we don't care about all + the aliasing between 8 kHz and 12 kHz. */ + for (j=0;j<subframe;j++) + { + tmp3x[3*j] = tmp[j]; + tmp3x[3*j+1] = tmp[j]; + tmp3x[3*j+2] = tmp[j]; + } + silk_resampler_down2_hp(S, y, tmp3x, 3*subframe); + } + RESTORE_STACK; + return ret; +} + +void tonality_analysis_init(TonalityAnalysisState *tonal, opus_int32 Fs) +{ + /* Initialize reusable fields. */ + tonal->arch = opus_select_arch(); + tonal->Fs = Fs; + /* Clear remaining fields. */ + tonality_analysis_reset(tonal); +} + +void tonality_analysis_reset(TonalityAnalysisState *tonal) +{ + /* Clear non-reusable fields. */ + char *start = (char*)&tonal->TONALITY_ANALYSIS_RESET_START; + OPUS_CLEAR(start, sizeof(TonalityAnalysisState) - (start - (char*)tonal)); +} + +void tonality_get_info(TonalityAnalysisState *tonal, AnalysisInfo *info_out, int len) +{ + int pos; + int curr_lookahead; + float tonality_max; + float tonality_avg; + int tonality_count; + int i; + int pos0; + float prob_avg; + float prob_count; + float prob_min, prob_max; + float vad_prob; + int mpos, vpos; + int bandwidth_span; + + pos = tonal->read_pos; + curr_lookahead = tonal->write_pos-tonal->read_pos; + if (curr_lookahead<0) + curr_lookahead += DETECT_SIZE; + + tonal->read_subframe += len/(tonal->Fs/400); + while (tonal->read_subframe>=8) + { + tonal->read_subframe -= 8; + tonal->read_pos++; + } + if (tonal->read_pos>=DETECT_SIZE) + tonal->read_pos-=DETECT_SIZE; + + /* On long frames, look at the second analysis window rather than the first. */ + if (len > tonal->Fs/50 && pos != tonal->write_pos) + { + pos++; + if (pos==DETECT_SIZE) + pos=0; + } + if (pos == tonal->write_pos) + pos--; + if (pos<0) + pos = DETECT_SIZE-1; + pos0 = pos; + OPUS_COPY(info_out, &tonal->info[pos], 1); + if (!info_out->valid) + return; + tonality_max = tonality_avg = info_out->tonality; + tonality_count = 1; + /* Look at the neighbouring frames and pick largest bandwidth found (to be safe). */ + bandwidth_span = 6; + /* If possible, look ahead for a tone to compensate for the delay in the tone detector. */ + for (i=0;i<3;i++) + { + pos++; + if (pos==DETECT_SIZE) + pos = 0; + if (pos == tonal->write_pos) + break; + tonality_max = MAX32(tonality_max, tonal->info[pos].tonality); + tonality_avg += tonal->info[pos].tonality; + tonality_count++; + info_out->bandwidth = IMAX(info_out->bandwidth, tonal->info[pos].bandwidth); + bandwidth_span--; + } + pos = pos0; + /* Look back in time to see if any has a wider bandwidth than the current frame. */ + for (i=0;i<bandwidth_span;i++) + { + pos--; + if (pos < 0) + pos = DETECT_SIZE-1; + if (pos == tonal->write_pos) + break; + info_out->bandwidth = IMAX(info_out->bandwidth, tonal->info[pos].bandwidth); + } + info_out->tonality = MAX32(tonality_avg/tonality_count, tonality_max-.2f); + + mpos = vpos = pos0; + /* If we have enough look-ahead, compensate for the ~5-frame delay in the music prob and + ~1 frame delay in the VAD prob. */ + if (curr_lookahead > 15) + { + mpos += 5; + if (mpos>=DETECT_SIZE) + mpos -= DETECT_SIZE; + vpos += 1; + if (vpos>=DETECT_SIZE) + vpos -= DETECT_SIZE; + } + + /* The following calculations attempt to minimize a "badness function" + for the transition. When switching from speech to music, the badness + of switching at frame k is + b_k = S*v_k + \sum_{i=0}^{k-1} v_i*(p_i - T) + where + v_i is the activity probability (VAD) at frame i, + p_i is the music probability at frame i + T is the probability threshold for switching + S is the penalty for switching during active audio rather than silence + the current frame has index i=0 + + Rather than apply badness to directly decide when to switch, what we compute + instead is the threshold for which the optimal switching point is now. When + considering whether to switch now (frame 0) or at frame k, we have: + S*v_0 = S*v_k + \sum_{i=0}^{k-1} v_i*(p_i - T) + which gives us: + T = ( \sum_{i=0}^{k-1} v_i*p_i + S*(v_k-v_0) ) / ( \sum_{i=0}^{k-1} v_i ) + We take the min threshold across all positive values of k (up to the maximum + amount of lookahead we have) to give us the threshold for which the current + frame is the optimal switch point. + + The last step is that we need to consider whether we want to switch at all. + For that we use the average of the music probability over the entire window. + If the threshold is higher than that average we're not going to + switch, so we compute a min with the average as well. The result of all these + min operations is music_prob_min, which gives the threshold for switching to music + if we're currently encoding for speech. + + We do the exact opposite to compute music_prob_max which is used for switching + from music to speech. + */ + prob_min = 1.f; + prob_max = 0.f; + vad_prob = tonal->info[vpos].activity_probability; + prob_count = MAX16(.1f, vad_prob); + prob_avg = MAX16(.1f, vad_prob)*tonal->info[mpos].music_prob; + while (1) + { + float pos_vad; + mpos++; + if (mpos==DETECT_SIZE) + mpos = 0; + if (mpos == tonal->write_pos) + break; + vpos++; + if (vpos==DETECT_SIZE) + vpos = 0; + if (vpos == tonal->write_pos) + break; + pos_vad = tonal->info[vpos].activity_probability; + prob_min = MIN16((prob_avg - TRANSITION_PENALTY*(vad_prob - pos_vad))/prob_count, prob_min); + prob_max = MAX16((prob_avg + TRANSITION_PENALTY*(vad_prob - pos_vad))/prob_count, prob_max); + prob_count += MAX16(.1f, pos_vad); + prob_avg += MAX16(.1f, pos_vad)*tonal->info[mpos].music_prob; + } + info_out->music_prob = prob_avg/prob_count; + prob_min = MIN16(prob_avg/prob_count, prob_min); + prob_max = MAX16(prob_avg/prob_count, prob_max); + prob_min = MAX16(prob_min, 0.f); + prob_max = MIN16(prob_max, 1.f); + + /* If we don't have enough look-ahead, do our best to make a decent decision. */ + if (curr_lookahead < 10) + { + float pmin, pmax; + pmin = prob_min; + pmax = prob_max; + pos = pos0; + /* Look for min/max in the past. */ + for (i=0;i<IMIN(tonal->count-1, 15);i++) + { + pos--; + if (pos < 0) + pos = DETECT_SIZE-1; + pmin = MIN16(pmin, tonal->info[pos].music_prob); + pmax = MAX16(pmax, tonal->info[pos].music_prob); + } + /* Bias against switching on active audio. */ + pmin = MAX16(0.f, pmin - .1f*vad_prob); + pmax = MIN16(1.f, pmax + .1f*vad_prob); + prob_min += (1.f-.1f*curr_lookahead)*(pmin - prob_min); + prob_max += (1.f-.1f*curr_lookahead)*(pmax - prob_max); + } + info_out->music_prob_min = prob_min; + info_out->music_prob_max = prob_max; + + /* printf("%f %f %f %f %f\n", prob_min, prob_max, prob_avg/prob_count, vad_prob, info_out->music_prob); */ +} + +static const float std_feature_bias[9] = { + 5.684947f, 3.475288f, 1.770634f, 1.599784f, 3.773215f, + 2.163313f, 1.260756f, 1.116868f, 1.918795f +}; + +#define LEAKAGE_OFFSET 2.5f +#define LEAKAGE_SLOPE 2.f + +#ifdef FIXED_POINT +/* For fixed-point, the input is +/-2^15 shifted up by SIG_SHIFT, so we need to + compensate for that in the energy. */ +#define SCALE_COMPENS (1.f/((opus_int32)1<<(15+SIG_SHIFT))) +#define SCALE_ENER(e) ((SCALE_COMPENS*SCALE_COMPENS)*(e)) +#else +#define SCALE_ENER(e) (e) +#endif + +#ifdef FIXED_POINT +static int is_digital_silence32(const opus_val32* pcm, int frame_size, int channels, int lsb_depth) +{ + int silence = 0; + opus_val32 sample_max = 0; +#ifdef MLP_TRAINING + return 0; +#endif + sample_max = celt_maxabs32(pcm, frame_size*channels); + + silence = (sample_max == 0); + (void)lsb_depth; + return silence; +} +#else +#define is_digital_silence32(pcm, frame_size, channels, lsb_depth) is_digital_silence(pcm, frame_size, channels, lsb_depth) +#endif + +static void tonality_analysis(TonalityAnalysisState *tonal, const CELTMode *celt_mode, const void *x, int len, int offset, int c1, int c2, int C, int lsb_depth, downmix_func downmix) +{ + int i, b; + const kiss_fft_state *kfft; + VARDECL(kiss_fft_cpx, in); + VARDECL(kiss_fft_cpx, out); + int N = 480, N2=240; + float * OPUS_RESTRICT A = tonal->angle; + float * OPUS_RESTRICT dA = tonal->d_angle; + float * OPUS_RESTRICT d2A = tonal->d2_angle; + VARDECL(float, tonality); + VARDECL(float, noisiness); + float band_tonality[NB_TBANDS]; + float logE[NB_TBANDS]; + float BFCC[8]; + float features[25]; + float frame_tonality; + float max_frame_tonality; + /*float tw_sum=0;*/ + float frame_noisiness; + const float pi4 = (float)(M_PI*M_PI*M_PI*M_PI); + float slope=0; + float frame_stationarity; + float relativeE; + float frame_probs[2]; + float alpha, alphaE, alphaE2; + float frame_loudness; + float bandwidth_mask; + int is_masked[NB_TBANDS+1]; + int bandwidth=0; + float maxE = 0; + float noise_floor; + int remaining; + AnalysisInfo *info; + float hp_ener; + float tonality2[240]; + float midE[8]; + float spec_variability=0; + float band_log2[NB_TBANDS+1]; + float leakage_from[NB_TBANDS+1]; + float leakage_to[NB_TBANDS+1]; + float layer_out[MAX_NEURONS]; + float below_max_pitch; + float above_max_pitch; + int is_silence; + SAVE_STACK; + + if (!tonal->initialized) + { + tonal->mem_fill = 240; + tonal->initialized = 1; + } + alpha = 1.f/IMIN(10, 1+tonal->count); + alphaE = 1.f/IMIN(25, 1+tonal->count); + /* Noise floor related decay for bandwidth detection: -2.2 dB/second */ + alphaE2 = 1.f/IMIN(100, 1+tonal->count); + if (tonal->count <= 1) alphaE2 = 1; + + if (tonal->Fs == 48000) + { + /* len and offset are now at 24 kHz. */ + len/= 2; + offset /= 2; + } else if (tonal->Fs == 16000) { + len = 3*len/2; + offset = 3*offset/2; + } + + kfft = celt_mode->mdct.kfft[0]; + tonal->hp_ener_accum += (float)downmix_and_resample(downmix, x, + &tonal->inmem[tonal->mem_fill], tonal->downmix_state, + IMIN(len, ANALYSIS_BUF_SIZE-tonal->mem_fill), offset, c1, c2, C, tonal->Fs); + if (tonal->mem_fill+len < ANALYSIS_BUF_SIZE) + { + tonal->mem_fill += len; + /* Don't have enough to update the analysis */ + RESTORE_STACK; + return; + } + hp_ener = tonal->hp_ener_accum; + info = &tonal->info[tonal->write_pos++]; + if (tonal->write_pos>=DETECT_SIZE) + tonal->write_pos-=DETECT_SIZE; + + is_silence = is_digital_silence32(tonal->inmem, ANALYSIS_BUF_SIZE, 1, lsb_depth); + + ALLOC(in, 480, kiss_fft_cpx); + ALLOC(out, 480, kiss_fft_cpx); + ALLOC(tonality, 240, float); + ALLOC(noisiness, 240, float); + for (i=0;i<N2;i++) + { + float w = analysis_window[i]; + in[i].r = (kiss_fft_scalar)(w*tonal->inmem[i]); + in[i].i = (kiss_fft_scalar)(w*tonal->inmem[N2+i]); + in[N-i-1].r = (kiss_fft_scalar)(w*tonal->inmem[N-i-1]); + in[N-i-1].i = (kiss_fft_scalar)(w*tonal->inmem[N+N2-i-1]); + } + OPUS_MOVE(tonal->inmem, tonal->inmem+ANALYSIS_BUF_SIZE-240, 240); + remaining = len - (ANALYSIS_BUF_SIZE-tonal->mem_fill); + tonal->hp_ener_accum = (float)downmix_and_resample(downmix, x, + &tonal->inmem[240], tonal->downmix_state, remaining, + offset+ANALYSIS_BUF_SIZE-tonal->mem_fill, c1, c2, C, tonal->Fs); + tonal->mem_fill = 240 + remaining; + if (is_silence) + { + /* On silence, copy the previous analysis. */ + int prev_pos = tonal->write_pos-2; + if (prev_pos < 0) + prev_pos += DETECT_SIZE; + OPUS_COPY(info, &tonal->info[prev_pos], 1); + RESTORE_STACK; + return; + } + opus_fft(kfft, in, out, tonal->arch); +#ifndef FIXED_POINT + /* If there's any NaN on the input, the entire output will be NaN, so we only need to check one value. */ + if (celt_isnan(out[0].r)) + { + info->valid = 0; + RESTORE_STACK; + return; + } +#endif + + for (i=1;i<N2;i++) + { + float X1r, X2r, X1i, X2i; + float angle, d_angle, d2_angle; + float angle2, d_angle2, d2_angle2; + float mod1, mod2, avg_mod; + X1r = (float)out[i].r+out[N-i].r; + X1i = (float)out[i].i-out[N-i].i; + X2r = (float)out[i].i+out[N-i].i; + X2i = (float)out[N-i].r-out[i].r; + + angle = (float)(.5f/M_PI)*fast_atan2f(X1i, X1r); + d_angle = angle - A[i]; + d2_angle = d_angle - dA[i]; + + angle2 = (float)(.5f/M_PI)*fast_atan2f(X2i, X2r); + d_angle2 = angle2 - angle; + d2_angle2 = d_angle2 - d_angle; + + mod1 = d2_angle - (float)float2int(d2_angle); + noisiness[i] = ABS16(mod1); + mod1 *= mod1; + mod1 *= mod1; + + mod2 = d2_angle2 - (float)float2int(d2_angle2); + noisiness[i] += ABS16(mod2); + mod2 *= mod2; + mod2 *= mod2; + + avg_mod = .25f*(d2A[i]+mod1+2*mod2); + /* This introduces an extra delay of 2 frames in the detection. */ + tonality[i] = 1.f/(1.f+40.f*16.f*pi4*avg_mod)-.015f; + /* No delay on this detection, but it's less reliable. */ + tonality2[i] = 1.f/(1.f+40.f*16.f*pi4*mod2)-.015f; + + A[i] = angle2; + dA[i] = d_angle2; + d2A[i] = mod2; + } + for (i=2;i<N2-1;i++) + { + float tt = MIN32(tonality2[i], MAX32(tonality2[i-1], tonality2[i+1])); + tonality[i] = .9f*MAX32(tonality[i], tt-.1f); + } + frame_tonality = 0; + max_frame_tonality = 0; + /*tw_sum = 0;*/ + info->activity = 0; + frame_noisiness = 0; + frame_stationarity = 0; + if (!tonal->count) + { + for (b=0;b<NB_TBANDS;b++) + { + tonal->lowE[b] = 1e10; + tonal->highE[b] = -1e10; + } + } + relativeE = 0; + frame_loudness = 0; + /* The energy of the very first band is special because of DC. */ + { + float E = 0; + float X1r, X2r; + X1r = 2*(float)out[0].r; + X2r = 2*(float)out[0].i; + E = X1r*X1r + X2r*X2r; + for (i=1;i<4;i++) + { + float binE = out[i].r*(float)out[i].r + out[N-i].r*(float)out[N-i].r + + out[i].i*(float)out[i].i + out[N-i].i*(float)out[N-i].i; + E += binE; + } + E = SCALE_ENER(E); + band_log2[0] = .5f*1.442695f*(float)log(E+1e-10f); + } + for (b=0;b<NB_TBANDS;b++) + { + float E=0, tE=0, nE=0; + float L1, L2; + float stationarity; + for (i=tbands[b];i<tbands[b+1];i++) + { + float binE = out[i].r*(float)out[i].r + out[N-i].r*(float)out[N-i].r + + out[i].i*(float)out[i].i + out[N-i].i*(float)out[N-i].i; + binE = SCALE_ENER(binE); + E += binE; + tE += binE*MAX32(0, tonality[i]); + nE += binE*2.f*(.5f-noisiness[i]); + } +#ifndef FIXED_POINT + /* Check for extreme band energies that could cause NaNs later. */ + if (!(E<1e9f) || celt_isnan(E)) + { + info->valid = 0; + RESTORE_STACK; + return; + } +#endif + + tonal->E[tonal->E_count][b] = E; + frame_noisiness += nE/(1e-15f+E); + + frame_loudness += (float)sqrt(E+1e-10f); + logE[b] = (float)log(E+1e-10f); + band_log2[b+1] = .5f*1.442695f*(float)log(E+1e-10f); + tonal->logE[tonal->E_count][b] = logE[b]; + if (tonal->count==0) + tonal->highE[b] = tonal->lowE[b] = logE[b]; + if (tonal->highE[b] > tonal->lowE[b] + 7.5) + { + if (tonal->highE[b] - logE[b] > logE[b] - tonal->lowE[b]) + tonal->highE[b] -= .01f; + else + tonal->lowE[b] += .01f; + } + if (logE[b] > tonal->highE[b]) + { + tonal->highE[b] = logE[b]; + tonal->lowE[b] = MAX32(tonal->highE[b]-15, tonal->lowE[b]); + } else if (logE[b] < tonal->lowE[b]) + { + tonal->lowE[b] = logE[b]; + tonal->highE[b] = MIN32(tonal->lowE[b]+15, tonal->highE[b]); + } + relativeE += (logE[b]-tonal->lowE[b])/(1e-5f + (tonal->highE[b]-tonal->lowE[b])); + + L1=L2=0; + for (i=0;i<NB_FRAMES;i++) + { + L1 += (float)sqrt(tonal->E[i][b]); + L2 += tonal->E[i][b]; + } + + stationarity = MIN16(0.99f,L1/(float)sqrt(1e-15+NB_FRAMES*L2)); + stationarity *= stationarity; + stationarity *= stationarity; + frame_stationarity += stationarity; + /*band_tonality[b] = tE/(1e-15+E)*/; + band_tonality[b] = MAX16(tE/(1e-15f+E), stationarity*tonal->prev_band_tonality[b]); +#if 0 + if (b>=NB_TONAL_SKIP_BANDS) + { + frame_tonality += tweight[b]*band_tonality[b]; + tw_sum += tweight[b]; + } +#else + frame_tonality += band_tonality[b]; + if (b>=NB_TBANDS-NB_TONAL_SKIP_BANDS) + frame_tonality -= band_tonality[b-NB_TBANDS+NB_TONAL_SKIP_BANDS]; +#endif + max_frame_tonality = MAX16(max_frame_tonality, (1.f+.03f*(b-NB_TBANDS))*frame_tonality); + slope += band_tonality[b]*(b-8); + /*printf("%f %f ", band_tonality[b], stationarity);*/ + tonal->prev_band_tonality[b] = band_tonality[b]; + } + + leakage_from[0] = band_log2[0]; + leakage_to[0] = band_log2[0] - LEAKAGE_OFFSET; + for (b=1;b<NB_TBANDS+1;b++) + { + float leak_slope = LEAKAGE_SLOPE*(tbands[b]-tbands[b-1])/4; + leakage_from[b] = MIN16(leakage_from[b-1]+leak_slope, band_log2[b]); + leakage_to[b] = MAX16(leakage_to[b-1]-leak_slope, band_log2[b]-LEAKAGE_OFFSET); + } + for (b=NB_TBANDS-2;b>=0;b--) + { + float leak_slope = LEAKAGE_SLOPE*(tbands[b+1]-tbands[b])/4; + leakage_from[b] = MIN16(leakage_from[b+1]+leak_slope, leakage_from[b]); + leakage_to[b] = MAX16(leakage_to[b+1]-leak_slope, leakage_to[b]); + } + celt_assert(NB_TBANDS+1 <= LEAK_BANDS); + for (b=0;b<NB_TBANDS+1;b++) + { + /* leak_boost[] is made up of two terms. The first, based on leakage_to[], + represents the boost needed to overcome the amount of analysis leakage + cause in a weaker band b by louder neighbouring bands. + The second, based on leakage_from[], applies to a loud band b for + which the quantization noise causes synthesis leakage to the weaker + neighbouring bands. */ + float boost = MAX16(0, leakage_to[b] - band_log2[b]) + + MAX16(0, band_log2[b] - (leakage_from[b]+LEAKAGE_OFFSET)); + info->leak_boost[b] = IMIN(255, (int)floor(.5 + 64.f*boost)); + } + for (;b<LEAK_BANDS;b++) info->leak_boost[b] = 0; + + for (i=0;i<NB_FRAMES;i++) + { + int j; + float mindist = 1e15f; + for (j=0;j<NB_FRAMES;j++) + { + int k; + float dist=0; + for (k=0;k<NB_TBANDS;k++) + { + float tmp; + tmp = tonal->logE[i][k] - tonal->logE[j][k]; + dist += tmp*tmp; + } + if (j!=i) + mindist = MIN32(mindist, dist); + } + spec_variability += mindist; + } + spec_variability = (float)sqrt(spec_variability/NB_FRAMES/NB_TBANDS); + bandwidth_mask = 0; + bandwidth = 0; + maxE = 0; + noise_floor = 5.7e-4f/(1<<(IMAX(0,lsb_depth-8))); + noise_floor *= noise_floor; + below_max_pitch=0; + above_max_pitch=0; + for (b=0;b<NB_TBANDS;b++) + { + float E=0; + float Em; + int band_start, band_end; + /* Keep a margin of 300 Hz for aliasing */ + band_start = tbands[b]; + band_end = tbands[b+1]; + for (i=band_start;i<band_end;i++) + { + float binE = out[i].r*(float)out[i].r + out[N-i].r*(float)out[N-i].r + + out[i].i*(float)out[i].i + out[N-i].i*(float)out[N-i].i; + E += binE; + } + E = SCALE_ENER(E); + maxE = MAX32(maxE, E); + if (band_start < 64) + { + below_max_pitch += E; + } else { + above_max_pitch += E; + } + tonal->meanE[b] = MAX32((1-alphaE2)*tonal->meanE[b], E); + Em = MAX32(E, tonal->meanE[b]); + /* Consider the band "active" only if all these conditions are met: + 1) less than 90 dB below the peak band (maximal masking possible considering + both the ATH and the loudness-dependent slope of the spreading function) + 2) above the PCM quantization noise floor + We use b+1 because the first CELT band isn't included in tbands[] + */ + if (E*1e9f > maxE && (Em > 3*noise_floor*(band_end-band_start) || E > noise_floor*(band_end-band_start))) + bandwidth = b+1; + /* Check if the band is masked (see below). */ + is_masked[b] = E < (tonal->prev_bandwidth >= b+1 ? .01f : .05f)*bandwidth_mask; + /* Use a simple follower with 13 dB/Bark slope for spreading function. */ + bandwidth_mask = MAX32(.05f*bandwidth_mask, E); + } + /* Special case for the last two bands, for which we don't have spectrum but only + the energy above 12 kHz. The difficulty here is that the high-pass we use + leaks some LF energy, so we need to increase the threshold without accidentally cutting + off the band. */ + if (tonal->Fs == 48000) { + float noise_ratio; + float Em; + float E = hp_ener*(1.f/(60*60)); + noise_ratio = tonal->prev_bandwidth==20 ? 10.f : 30.f; + +#ifdef FIXED_POINT + /* silk_resampler_down2_hp() shifted right by an extra 8 bits. */ + E *= 256.f*(1.f/Q15ONE)*(1.f/Q15ONE); +#endif + above_max_pitch += E; + tonal->meanE[b] = MAX32((1-alphaE2)*tonal->meanE[b], E); + Em = MAX32(E, tonal->meanE[b]); + if (Em > 3*noise_ratio*noise_floor*160 || E > noise_ratio*noise_floor*160) + bandwidth = 20; + /* Check if the band is masked (see below). */ + is_masked[b] = E < (tonal->prev_bandwidth == 20 ? .01f : .05f)*bandwidth_mask; + } + if (above_max_pitch > below_max_pitch) + info->max_pitch_ratio = below_max_pitch/above_max_pitch; + else + info->max_pitch_ratio = 1; + /* In some cases, resampling aliasing can create a small amount of energy in the first band + being cut. So if the last band is masked, we don't include it. */ + if (bandwidth == 20 && is_masked[NB_TBANDS]) + bandwidth-=2; + else if (bandwidth > 0 && bandwidth <= NB_TBANDS && is_masked[bandwidth-1]) + bandwidth--; + if (tonal->count<=2) + bandwidth = 20; + frame_loudness = 20*(float)log10(frame_loudness); + tonal->Etracker = MAX32(tonal->Etracker-.003f, frame_loudness); + tonal->lowECount *= (1-alphaE); + if (frame_loudness < tonal->Etracker-30) + tonal->lowECount += alphaE; + + for (i=0;i<8;i++) + { + float sum=0; + for (b=0;b<16;b++) + sum += dct_table[i*16+b]*logE[b]; + BFCC[i] = sum; + } + for (i=0;i<8;i++) + { + float sum=0; + for (b=0;b<16;b++) + sum += dct_table[i*16+b]*.5f*(tonal->highE[b]+tonal->lowE[b]); + midE[i] = sum; + } + + frame_stationarity /= NB_TBANDS; + relativeE /= NB_TBANDS; + if (tonal->count<10) + relativeE = .5f; + frame_noisiness /= NB_TBANDS; +#if 1 + info->activity = frame_noisiness + (1-frame_noisiness)*relativeE; +#else + info->activity = .5*(1+frame_noisiness-frame_stationarity); +#endif + frame_tonality = (max_frame_tonality/(NB_TBANDS-NB_TONAL_SKIP_BANDS)); + frame_tonality = MAX16(frame_tonality, tonal->prev_tonality*.8f); + tonal->prev_tonality = frame_tonality; + + slope /= 8*8; + info->tonality_slope = slope; + + tonal->E_count = (tonal->E_count+1)%NB_FRAMES; + tonal->count = IMIN(tonal->count+1, ANALYSIS_COUNT_MAX); + info->tonality = frame_tonality; + + for (i=0;i<4;i++) + features[i] = -0.12299f*(BFCC[i]+tonal->mem[i+24]) + 0.49195f*(tonal->mem[i]+tonal->mem[i+16]) + 0.69693f*tonal->mem[i+8] - 1.4349f*tonal->cmean[i]; + + for (i=0;i<4;i++) + tonal->cmean[i] = (1-alpha)*tonal->cmean[i] + alpha*BFCC[i]; + + for (i=0;i<4;i++) + features[4+i] = 0.63246f*(BFCC[i]-tonal->mem[i+24]) + 0.31623f*(tonal->mem[i]-tonal->mem[i+16]); + for (i=0;i<3;i++) + features[8+i] = 0.53452f*(BFCC[i]+tonal->mem[i+24]) - 0.26726f*(tonal->mem[i]+tonal->mem[i+16]) -0.53452f*tonal->mem[i+8]; + + if (tonal->count > 5) + { + for (i=0;i<9;i++) + tonal->std[i] = (1-alpha)*tonal->std[i] + alpha*features[i]*features[i]; + } + for (i=0;i<4;i++) + features[i] = BFCC[i]-midE[i]; + + for (i=0;i<8;i++) + { + tonal->mem[i+24] = tonal->mem[i+16]; + tonal->mem[i+16] = tonal->mem[i+8]; + tonal->mem[i+8] = tonal->mem[i]; + tonal->mem[i] = BFCC[i]; + } + for (i=0;i<9;i++) + features[11+i] = (float)sqrt(tonal->std[i]) - std_feature_bias[i]; + features[18] = spec_variability - 0.78f; + features[20] = info->tonality - 0.154723f; + features[21] = info->activity - 0.724643f; + features[22] = frame_stationarity - 0.743717f; + features[23] = info->tonality_slope + 0.069216f; + features[24] = tonal->lowECount - 0.067930f; + + compute_dense(&layer0, layer_out, features); + compute_gru(&layer1, tonal->rnn_state, layer_out); + compute_dense(&layer2, frame_probs, tonal->rnn_state); + + /* Probability of speech or music vs noise */ + info->activity_probability = frame_probs[1]; + info->music_prob = frame_probs[0]; + + /*printf("%f %f %f\n", frame_probs[0], frame_probs[1], info->music_prob);*/ +#ifdef MLP_TRAINING + for (i=0;i<25;i++) + printf("%f ", features[i]); + printf("\n"); +#endif + + info->bandwidth = bandwidth; + tonal->prev_bandwidth = bandwidth; + /*printf("%d %d\n", info->bandwidth, info->opus_bandwidth);*/ + info->noisiness = frame_noisiness; + info->valid = 1; + RESTORE_STACK; +} + +void run_analysis(TonalityAnalysisState *analysis, const CELTMode *celt_mode, const void *analysis_pcm, + int analysis_frame_size, int frame_size, int c1, int c2, int C, opus_int32 Fs, + int lsb_depth, downmix_func downmix, AnalysisInfo *analysis_info) +{ + int offset; + int pcm_len; + + analysis_frame_size -= analysis_frame_size&1; + if (analysis_pcm != NULL) + { + /* Avoid overflow/wrap-around of the analysis buffer */ + analysis_frame_size = IMIN((DETECT_SIZE-5)*Fs/50, analysis_frame_size); + + pcm_len = analysis_frame_size - analysis->analysis_offset; + offset = analysis->analysis_offset; + while (pcm_len>0) { + tonality_analysis(analysis, celt_mode, analysis_pcm, IMIN(Fs/50, pcm_len), offset, c1, c2, C, lsb_depth, downmix); + offset += Fs/50; + pcm_len -= Fs/50; + } + analysis->analysis_offset = analysis_frame_size; + + analysis->analysis_offset -= frame_size; + } + + tonality_get_info(analysis, analysis_info, frame_size); +} + +#endif /* DISABLE_FLOAT_API */ diff --git a/media/libopus/src/analysis.h b/media/libopus/src/analysis.h new file mode 100644 index 0000000000..0b66555f21 --- /dev/null +++ b/media/libopus/src/analysis.h @@ -0,0 +1,103 @@ +/* Copyright (c) 2011 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR + CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef ANALYSIS_H +#define ANALYSIS_H + +#include "celt.h" +#include "opus_private.h" +#include "mlp.h" + +#define NB_FRAMES 8 +#define NB_TBANDS 18 +#define ANALYSIS_BUF_SIZE 720 /* 30 ms at 24 kHz */ + +/* At that point we can stop counting frames because it no longer matters. */ +#define ANALYSIS_COUNT_MAX 10000 + +#define DETECT_SIZE 100 + +/* Uncomment this to print the MLP features on stdout. */ +/*#define MLP_TRAINING*/ + +typedef struct { + int arch; + int application; + opus_int32 Fs; +#define TONALITY_ANALYSIS_RESET_START angle + float angle[240]; + float d_angle[240]; + float d2_angle[240]; + opus_val32 inmem[ANALYSIS_BUF_SIZE]; + int mem_fill; /* number of usable samples in the buffer */ + float prev_band_tonality[NB_TBANDS]; + float prev_tonality; + int prev_bandwidth; + float E[NB_FRAMES][NB_TBANDS]; + float logE[NB_FRAMES][NB_TBANDS]; + float lowE[NB_TBANDS]; + float highE[NB_TBANDS]; + float meanE[NB_TBANDS+1]; + float mem[32]; + float cmean[8]; + float std[9]; + float Etracker; + float lowECount; + int E_count; + int count; + int analysis_offset; + int write_pos; + int read_pos; + int read_subframe; + float hp_ener_accum; + int initialized; + float rnn_state[MAX_NEURONS]; + opus_val32 downmix_state[3]; + AnalysisInfo info[DETECT_SIZE]; +} TonalityAnalysisState; + +/** Initialize a TonalityAnalysisState struct. + * + * This performs some possibly slow initialization steps which should + * not be repeated every analysis step. No allocated memory is retained + * by the state struct, so no cleanup call is required. + */ +void tonality_analysis_init(TonalityAnalysisState *analysis, opus_int32 Fs); + +/** Reset a TonalityAnalysisState stuct. + * + * Call this when there's a discontinuity in the data. + */ +void tonality_analysis_reset(TonalityAnalysisState *analysis); + +void tonality_get_info(TonalityAnalysisState *tonal, AnalysisInfo *info_out, int len); + +void run_analysis(TonalityAnalysisState *analysis, const CELTMode *celt_mode, const void *analysis_pcm, + int analysis_frame_size, int frame_size, int c1, int c2, int C, opus_int32 Fs, + int lsb_depth, downmix_func downmix, AnalysisInfo *analysis_info); + +#endif diff --git a/media/libopus/src/mapping_matrix.c b/media/libopus/src/mapping_matrix.c new file mode 100644 index 0000000000..31298af057 --- /dev/null +++ b/media/libopus/src/mapping_matrix.c @@ -0,0 +1,378 @@ +/* Copyright (c) 2017 Google Inc. + Written by Andrew Allen */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "arch.h" +#include "float_cast.h" +#include "opus_private.h" +#include "opus_defines.h" +#include "mapping_matrix.h" + +#define MATRIX_INDEX(nb_rows, row, col) (nb_rows * col + row) + +opus_int32 mapping_matrix_get_size(int rows, int cols) +{ + opus_int32 size; + + /* Mapping Matrix must only support up to 255 channels in or out. + * Additionally, the total cell count must be <= 65004 octets in order + * for the matrix to be stored in an OGG header. + */ + if (rows > 255 || cols > 255) + return 0; + size = rows * (opus_int32)cols * sizeof(opus_int16); + if (size > 65004) + return 0; + + return align(sizeof(MappingMatrix)) + align(size); +} + +opus_int16 *mapping_matrix_get_data(const MappingMatrix *matrix) +{ + /* void* cast avoids clang -Wcast-align warning */ + return (opus_int16*)(void*)((char*)matrix + align(sizeof(MappingMatrix))); +} + +void mapping_matrix_init(MappingMatrix * const matrix, + int rows, int cols, int gain, const opus_int16 *data, opus_int32 data_size) +{ + int i; + opus_int16 *ptr; + +#if !defined(ENABLE_ASSERTIONS) + (void)data_size; +#endif + celt_assert(align(data_size) == align(rows * cols * sizeof(opus_int16))); + + matrix->rows = rows; + matrix->cols = cols; + matrix->gain = gain; + ptr = mapping_matrix_get_data(matrix); + for (i = 0; i < rows * cols; i++) + { + ptr[i] = data[i]; + } +} + +#ifndef DISABLE_FLOAT_API +void mapping_matrix_multiply_channel_in_float( + const MappingMatrix *matrix, + const float *input, + int input_rows, + opus_val16 *output, + int output_row, + int output_rows, + int frame_size) +{ + /* Matrix data is ordered col-wise. */ + opus_int16* matrix_data; + int i, col; + + celt_assert(input_rows <= matrix->cols && output_rows <= matrix->rows); + + matrix_data = mapping_matrix_get_data(matrix); + + for (i = 0; i < frame_size; i++) + { + float tmp = 0; + for (col = 0; col < input_rows; col++) + { + tmp += + matrix_data[MATRIX_INDEX(matrix->rows, output_row, col)] * + input[MATRIX_INDEX(input_rows, col, i)]; + } +#if defined(FIXED_POINT) + output[output_rows * i] = FLOAT2INT16((1/32768.f)*tmp); +#else + output[output_rows * i] = (1/32768.f)*tmp; +#endif + } +} + +void mapping_matrix_multiply_channel_out_float( + const MappingMatrix *matrix, + const opus_val16 *input, + int input_row, + int input_rows, + float *output, + int output_rows, + int frame_size +) +{ + /* Matrix data is ordered col-wise. */ + opus_int16* matrix_data; + int i, row; + float input_sample; + + celt_assert(input_rows <= matrix->cols && output_rows <= matrix->rows); + + matrix_data = mapping_matrix_get_data(matrix); + + for (i = 0; i < frame_size; i++) + { +#if defined(FIXED_POINT) + input_sample = (1/32768.f)*input[input_rows * i]; +#else + input_sample = input[input_rows * i]; +#endif + for (row = 0; row < output_rows; row++) + { + float tmp = + (1/32768.f)*matrix_data[MATRIX_INDEX(matrix->rows, row, input_row)] * + input_sample; + output[MATRIX_INDEX(output_rows, row, i)] += tmp; + } + } +} +#endif /* DISABLE_FLOAT_API */ + +void mapping_matrix_multiply_channel_in_short( + const MappingMatrix *matrix, + const opus_int16 *input, + int input_rows, + opus_val16 *output, + int output_row, + int output_rows, + int frame_size) +{ + /* Matrix data is ordered col-wise. */ + opus_int16* matrix_data; + int i, col; + + celt_assert(input_rows <= matrix->cols && output_rows <= matrix->rows); + + matrix_data = mapping_matrix_get_data(matrix); + + for (i = 0; i < frame_size; i++) + { + opus_val32 tmp = 0; + for (col = 0; col < input_rows; col++) + { +#if defined(FIXED_POINT) + tmp += + ((opus_int32)matrix_data[MATRIX_INDEX(matrix->rows, output_row, col)] * + (opus_int32)input[MATRIX_INDEX(input_rows, col, i)]) >> 8; +#else + tmp += + matrix_data[MATRIX_INDEX(matrix->rows, output_row, col)] * + input[MATRIX_INDEX(input_rows, col, i)]; +#endif + } +#if defined(FIXED_POINT) + output[output_rows * i] = (opus_int16)((tmp + 64) >> 7); +#else + output[output_rows * i] = (1/(32768.f*32768.f))*tmp; +#endif + } +} + +void mapping_matrix_multiply_channel_out_short( + const MappingMatrix *matrix, + const opus_val16 *input, + int input_row, + int input_rows, + opus_int16 *output, + int output_rows, + int frame_size) +{ + /* Matrix data is ordered col-wise. */ + opus_int16* matrix_data; + int i, row; + opus_int32 input_sample; + + celt_assert(input_rows <= matrix->cols && output_rows <= matrix->rows); + + matrix_data = mapping_matrix_get_data(matrix); + + for (i = 0; i < frame_size; i++) + { +#if defined(FIXED_POINT) + input_sample = (opus_int32)input[input_rows * i]; +#else + input_sample = (opus_int32)FLOAT2INT16(input[input_rows * i]); +#endif + for (row = 0; row < output_rows; row++) + { + opus_int32 tmp = + (opus_int32)matrix_data[MATRIX_INDEX(matrix->rows, row, input_row)] * + input_sample; + output[MATRIX_INDEX(output_rows, row, i)] += (tmp + 16384) >> 15; + } + } +} + +const MappingMatrix mapping_matrix_foa_mixing = { 6, 6, 0 }; +const opus_int16 mapping_matrix_foa_mixing_data[36] = { + 16384, 0, -16384, 23170, 0, 0, 16384, 23170, + 16384, 0, 0, 0, 16384, 0, -16384, -23170, + 0, 0, 16384, -23170, 16384, 0, 0, 0, + 0, 0, 0, 0, 32767, 0, 0, 0, + 0, 0, 0, 32767 +}; + +const MappingMatrix mapping_matrix_soa_mixing = { 11, 11, 0 }; +const opus_int16 mapping_matrix_soa_mixing_data[121] = { + 10923, 7723, 13377, -13377, 11585, 9459, 7723, -16384, + -6689, 0, 0, 10923, 7723, 13377, 13377, -11585, + 9459, 7723, 16384, -6689, 0, 0, 10923, -15447, + 13377, 0, 0, -18919, 7723, 0, 13377, 0, + 0, 10923, 7723, -13377, -13377, 11585, -9459, 7723, + 16384, -6689, 0, 0, 10923, -7723, 0, 13377, + -16384, 0, -15447, 0, 9459, 0, 0, 10923, + -7723, 0, -13377, 16384, 0, -15447, 0, 9459, + 0, 0, 10923, 15447, 0, 0, 0, 0, + -15447, 0, -18919, 0, 0, 10923, 7723, -13377, + 13377, -11585, -9459, 7723, -16384, -6689, 0, 0, + 10923, -15447, -13377, 0, 0, 18919, 7723, 0, + 13377, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 32767, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 32767 +}; + +const MappingMatrix mapping_matrix_toa_mixing = { 18, 18, 0 }; +const opus_int16 mapping_matrix_toa_mixing_data[324] = { + 8208, 0, -881, 14369, 0, 0, -8192, -4163, + 13218, 0, 0, 0, 11095, -8836, -6218, 14833, + 0, 0, 8208, -10161, 881, 10161, -13218, -2944, + -8192, 2944, 0, -10488, -6218, 6248, -11095, -6248, + 0, -10488, 0, 0, 8208, 10161, 881, -10161, + -13218, 2944, -8192, -2944, 0, 10488, -6218, -6248, + -11095, 6248, 0, 10488, 0, 0, 8176, 5566, + -11552, 5566, 9681, -11205, 8192, -11205, 0, 4920, + -15158, 9756, -3334, 9756, 0, -4920, 0, 0, + 8176, 7871, 11552, 0, 0, 15846, 8192, 0, + -9681, -6958, 0, 13797, 3334, 0, -15158, 0, + 0, 0, 8176, 0, 11552, 7871, 0, 0, + 8192, 15846, 9681, 0, 0, 0, 3334, 13797, + 15158, 6958, 0, 0, 8176, 5566, -11552, -5566, + -9681, -11205, 8192, 11205, 0, 4920, 15158, 9756, + -3334, -9756, 0, 4920, 0, 0, 8208, 14369, + -881, 0, 0, -4163, -8192, 0, -13218, -14833, + 0, -8836, 11095, 0, 6218, 0, 0, 0, + 8208, 10161, 881, 10161, 13218, 2944, -8192, 2944, + 0, 10488, 6218, -6248, -11095, -6248, 0, -10488, + 0, 0, 8208, -14369, -881, 0, 0, 4163, + -8192, 0, -13218, 14833, 0, 8836, 11095, 0, + 6218, 0, 0, 0, 8208, 0, -881, -14369, + 0, 0, -8192, 4163, 13218, 0, 0, 0, + 11095, 8836, -6218, -14833, 0, 0, 8176, -5566, + -11552, 5566, -9681, 11205, 8192, -11205, 0, -4920, + 15158, -9756, -3334, 9756, 0, -4920, 0, 0, + 8176, 0, 11552, -7871, 0, 0, 8192, -15846, + 9681, 0, 0, 0, 3334, -13797, 15158, -6958, + 0, 0, 8176, -7871, 11552, 0, 0, -15846, + 8192, 0, -9681, 6958, 0, -13797, 3334, 0, + -15158, 0, 0, 0, 8176, -5566, -11552, -5566, + 9681, 11205, 8192, 11205, 0, -4920, -15158, -9756, + -3334, -9756, 0, 4920, 0, 0, 8208, -10161, + 881, -10161, 13218, -2944, -8192, -2944, 0, -10488, + 6218, 6248, -11095, 6248, 0, 10488, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 32767, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 32767 +}; + +const MappingMatrix mapping_matrix_foa_demixing = { 6, 6, 0 }; +const opus_int16 mapping_matrix_foa_demixing_data[36] = { + 16384, 16384, 16384, 16384, 0, 0, 0, 23170, + 0, -23170, 0, 0, -16384, 16384, -16384, 16384, + 0, 0, 23170, 0, -23170, 0, 0, 0, + 0, 0, 0, 0, 32767, 0, 0, 0, + 0, 0, 0, 32767 +}; + +const MappingMatrix mapping_matrix_soa_demixing = { 11, 11, 3050 }; +const opus_int16 mapping_matrix_soa_demixing_data[121] = { + 2771, 2771, 2771, 2771, 2771, 2771, 2771, 2771, + 2771, 0, 0, 10033, 10033, -20066, 10033, 14189, + 14189, -28378, 10033, -20066, 0, 0, 3393, 3393, + 3393, -3393, 0, 0, 0, -3393, -3393, 0, + 0, -17378, 17378, 0, -17378, -24576, 24576, 0, + 17378, 0, 0, 0, -14189, 14189, 0, -14189, + -28378, 28378, 0, 14189, 0, 0, 0, 2399, + 2399, -4799, -2399, 0, 0, 0, -2399, 4799, + 0, 0, 1959, 1959, 1959, 1959, -3918, -3918, + -3918, 1959, 1959, 0, 0, -4156, 4156, 0, + 4156, 0, 0, 0, -4156, 0, 0, 0, + 8192, 8192, -16384, 8192, 16384, 16384, -32768, 8192, + -16384, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 8312, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 8312 +}; + +const MappingMatrix mapping_matrix_toa_demixing = { 18, 18, 0 }; +const opus_int16 mapping_matrix_toa_demixing_data[324] = { + 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, + 8192, 8192, 8192, 8192, 8192, 8192, 8192, 8192, + 0, 0, 0, -9779, 9779, 6263, 8857, 0, + 6263, 13829, 9779, -13829, 0, -6263, 0, -8857, + -6263, -9779, 0, 0, -3413, 3413, 3413, -11359, + 11359, 11359, -11359, -3413, 3413, -3413, -3413, -11359, + 11359, 11359, -11359, 3413, 0, 0, 13829, 9779, + -9779, 6263, 0, 8857, -6263, 0, 9779, 0, + -13829, 6263, -8857, 0, -6263, -9779, 0, 0, + 0, -15617, -15617, 6406, 0, 0, -6406, 0, + 15617, 0, 0, -6406, 0, 0, 6406, 15617, + 0, 0, 0, -5003, 5003, -10664, 15081, 0, + -10664, -7075, 5003, 7075, 0, 10664, 0, -15081, + 10664, -5003, 0, 0, -8176, -8176, -8176, 8208, + 8208, 8208, 8208, -8176, -8176, -8176, -8176, 8208, + 8208, 8208, 8208, -8176, 0, 0, -7075, 5003, + -5003, -10664, 0, 15081, 10664, 0, 5003, 0, + 7075, -10664, -15081, 0, 10664, -5003, 0, 0, + 15617, 0, 0, 0, -6406, 6406, 0, -15617, + 0, -15617, 15617, 0, 6406, -6406, 0, 0, + 0, 0, 0, -11393, 11393, 2993, -4233, 0, + 2993, -16112, 11393, 16112, 0, -2993, 0, 4233, + -2993, -11393, 0, 0, 0, -9974, -9974, -13617, + 0, 0, 13617, 0, 9974, 0, 0, 13617, + 0, 0, -13617, 9974, 0, 0, 0, 5579, + -5579, 10185, 14403, 0, 10185, -7890, -5579, 7890, + 0, -10185, 0, -14403, -10185, 5579, 0, 0, + 11826, -11826, -11826, -901, 901, 901, -901, 11826, + -11826, 11826, 11826, -901, 901, 901, -901, -11826, + 0, 0, -7890, -5579, 5579, 10185, 0, 14403, + -10185, 0, -5579, 0, 7890, 10185, -14403, 0, + -10185, 5579, 0, 0, -9974, 0, 0, 0, + -13617, 13617, 0, 9974, 0, 9974, -9974, 0, + 13617, -13617, 0, 0, 0, 0, 16112, -11393, + 11393, -2993, 0, 4233, 2993, 0, -11393, 0, + -16112, -2993, -4233, 0, 2993, 11393, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 32767, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 32767 +}; + diff --git a/media/libopus/src/mapping_matrix.h b/media/libopus/src/mapping_matrix.h new file mode 100644 index 0000000000..98bc82df3e --- /dev/null +++ b/media/libopus/src/mapping_matrix.h @@ -0,0 +1,133 @@ +/* Copyright (c) 2017 Google Inc. + Written by Andrew Allen */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +/** + * @file mapping_matrix.h + * @brief Opus reference implementation mapping matrix API + */ + +#ifndef MAPPING_MATRIX_H +#define MAPPING_MATRIX_H + +#include "opus_types.h" +#include "opus_projection.h" + +#ifdef __cplusplus +extern "C" { +#endif + +typedef struct MappingMatrix +{ + int rows; /* number of channels outputted from matrix. */ + int cols; /* number of channels inputted to matrix. */ + int gain; /* in dB. S7.8-format. */ + /* Matrix cell data goes here using col-wise ordering. */ +} MappingMatrix; + +opus_int32 mapping_matrix_get_size(int rows, int cols); + +opus_int16 *mapping_matrix_get_data(const MappingMatrix *matrix); + +void mapping_matrix_init( + MappingMatrix * const matrix, + int rows, + int cols, + int gain, + const opus_int16 *data, + opus_int32 data_size +); + +#ifndef DISABLE_FLOAT_API +void mapping_matrix_multiply_channel_in_float( + const MappingMatrix *matrix, + const float *input, + int input_rows, + opus_val16 *output, + int output_row, + int output_rows, + int frame_size +); + +void mapping_matrix_multiply_channel_out_float( + const MappingMatrix *matrix, + const opus_val16 *input, + int input_row, + int input_rows, + float *output, + int output_rows, + int frame_size +); +#endif /* DISABLE_FLOAT_API */ + +void mapping_matrix_multiply_channel_in_short( + const MappingMatrix *matrix, + const opus_int16 *input, + int input_rows, + opus_val16 *output, + int output_row, + int output_rows, + int frame_size +); + +void mapping_matrix_multiply_channel_out_short( + const MappingMatrix *matrix, + const opus_val16 *input, + int input_row, + int input_rows, + opus_int16 *output, + int output_rows, + int frame_size +); + +/* Pre-computed mixing and demixing matrices for 1st to 3rd-order ambisonics. + * foa: first-order ambisonics + * soa: second-order ambisonics + * toa: third-order ambisonics + */ +extern const MappingMatrix mapping_matrix_foa_mixing; +extern const opus_int16 mapping_matrix_foa_mixing_data[36]; + +extern const MappingMatrix mapping_matrix_soa_mixing; +extern const opus_int16 mapping_matrix_soa_mixing_data[121]; + +extern const MappingMatrix mapping_matrix_toa_mixing; +extern const opus_int16 mapping_matrix_toa_mixing_data[324]; + +extern const MappingMatrix mapping_matrix_foa_demixing; +extern const opus_int16 mapping_matrix_foa_demixing_data[36]; + +extern const MappingMatrix mapping_matrix_soa_demixing; +extern const opus_int16 mapping_matrix_soa_demixing_data[121]; + +extern const MappingMatrix mapping_matrix_toa_demixing; +extern const opus_int16 mapping_matrix_toa_demixing_data[324]; + +#ifdef __cplusplus +} +#endif + +#endif /* MAPPING_MATRIX_H */ diff --git a/media/libopus/src/mlp.c b/media/libopus/src/mlp.c new file mode 100644 index 0000000000..964c6a98f6 --- /dev/null +++ b/media/libopus/src/mlp.c @@ -0,0 +1,144 @@ +/* Copyright (c) 2008-2011 Octasic Inc. + 2012-2017 Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR + CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include <math.h> +#include "opus_types.h" +#include "opus_defines.h" +#include "arch.h" +#include "tansig_table.h" +#include "mlp.h" + +static OPUS_INLINE float tansig_approx(float x) +{ + int i; + float y, dy; + float sign=1; + /* Tests are reversed to catch NaNs */ + if (!(x<8)) + return 1; + if (!(x>-8)) + return -1; +#ifndef FIXED_POINT + /* Another check in case of -ffast-math */ + if (celt_isnan(x)) + return 0; +#endif + if (x<0) + { + x=-x; + sign=-1; + } + i = (int)floor(.5f+25*x); + x -= .04f*i; + y = tansig_table[i]; + dy = 1-y*y; + y = y + x*dy*(1 - y*x); + return sign*y; +} + +static OPUS_INLINE float sigmoid_approx(float x) +{ + return .5f + .5f*tansig_approx(.5f*x); +} + +static void gemm_accum(float *out, const opus_int8 *weights, int rows, int cols, int col_stride, const float *x) +{ + int i, j; + for (i=0;i<rows;i++) + { + for (j=0;j<cols;j++) + out[i] += weights[j*col_stride + i]*x[j]; + } +} + +void compute_dense(const DenseLayer *layer, float *output, const float *input) +{ + int i; + int N, M; + int stride; + M = layer->nb_inputs; + N = layer->nb_neurons; + stride = N; + for (i=0;i<N;i++) + output[i] = layer->bias[i]; + gemm_accum(output, layer->input_weights, N, M, stride, input); + for (i=0;i<N;i++) + output[i] *= WEIGHTS_SCALE; + if (layer->sigmoid) { + for (i=0;i<N;i++) + output[i] = sigmoid_approx(output[i]); + } else { + for (i=0;i<N;i++) + output[i] = tansig_approx(output[i]); + } +} + +void compute_gru(const GRULayer *gru, float *state, const float *input) +{ + int i; + int N, M; + int stride; + float tmp[MAX_NEURONS]; + float z[MAX_NEURONS]; + float r[MAX_NEURONS]; + float h[MAX_NEURONS]; + M = gru->nb_inputs; + N = gru->nb_neurons; + stride = 3*N; + /* Compute update gate. */ + for (i=0;i<N;i++) + z[i] = gru->bias[i]; + gemm_accum(z, gru->input_weights, N, M, stride, input); + gemm_accum(z, gru->recurrent_weights, N, N, stride, state); + for (i=0;i<N;i++) + z[i] = sigmoid_approx(WEIGHTS_SCALE*z[i]); + + /* Compute reset gate. */ + for (i=0;i<N;i++) + r[i] = gru->bias[N + i]; + gemm_accum(r, &gru->input_weights[N], N, M, stride, input); + gemm_accum(r, &gru->recurrent_weights[N], N, N, stride, state); + for (i=0;i<N;i++) + r[i] = sigmoid_approx(WEIGHTS_SCALE*r[i]); + + /* Compute output. */ + for (i=0;i<N;i++) + h[i] = gru->bias[2*N + i]; + for (i=0;i<N;i++) + tmp[i] = state[i] * r[i]; + gemm_accum(h, &gru->input_weights[2*N], N, M, stride, input); + gemm_accum(h, &gru->recurrent_weights[2*N], N, N, stride, tmp); + for (i=0;i<N;i++) + h[i] = z[i]*state[i] + (1-z[i])*tansig_approx(WEIGHTS_SCALE*h[i]); + for (i=0;i<N;i++) + state[i] = h[i]; +} + diff --git a/media/libopus/src/mlp.h b/media/libopus/src/mlp.h new file mode 100644 index 0000000000..d7670550fd --- /dev/null +++ b/media/libopus/src/mlp.h @@ -0,0 +1,60 @@ +/* Copyright (c) 2017 Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR + CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifndef _MLP_H_ +#define _MLP_H_ + +#include "opus_types.h" + +#define WEIGHTS_SCALE (1.f/128) + +#define MAX_NEURONS 32 + +typedef struct { + const opus_int8 *bias; + const opus_int8 *input_weights; + int nb_inputs; + int nb_neurons; + int sigmoid; +} DenseLayer; + +typedef struct { + const opus_int8 *bias; + const opus_int8 *input_weights; + const opus_int8 *recurrent_weights; + int nb_inputs; + int nb_neurons; +} GRULayer; + +extern const DenseLayer layer0; +extern const GRULayer layer1; +extern const DenseLayer layer2; + +void compute_dense(const DenseLayer *layer, float *output, const float *input); + +void compute_gru(const GRULayer *gru, float *state, const float *input); + +#endif /* _MLP_H_ */ diff --git a/media/libopus/src/mlp_data.c b/media/libopus/src/mlp_data.c new file mode 100644 index 0000000000..ae4178df76 --- /dev/null +++ b/media/libopus/src/mlp_data.c @@ -0,0 +1,672 @@ +/*This file is automatically generated from a Keras model*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "mlp.h" + +static const opus_int8 layer0_weights[800] = { + -30, -9, 2, -12, 5, -1, 8, 9, + 9, 8, -13, 18, -17, -34, -5, 17, + -11, 0, -4, 10, 2, 10, 15, -8, + 2, -1, 0, 5, 13, -3, -16, 1, + -5, 3, 7, -28, -13, 6, 36, -3, + 19, -60, -17, -28, 7, -11, -30, -7, + 2, -42, -21, -3, 6, -22, 33, -9, + 7, -30, 21, -14, 24, -11, -20, -18, + -5, -12, 12, -49, -50, -49, 16, 9, + -37, -1, 9, 34, -13, -31, -31, 12, + 16, 44, -42, 2, -9, 8, -18, -6, + 9, 36, 19, 11, 13, 12, -21, 3, + -28, -12, 3, 33, 25, -14, 11, 1, + -94, -39, 18, -12, -11, -15, -7, 49, + 52, 10, -43, 9, 57, 8, 21, -6, + 14, -15, 44, -8, 7, -30, -13, -2, + -9, 25, -2, -127, 18, -11, -52, 26, + -27, 27, 10, -10, 7, 43, 6, -24, + 41, 10, -18, -27, 10, 17, 9, 10, + -17, -10, 20, -6, 22, 55, 35, -80, + 36, 25, -24, -36, 15, 9, -19, 88, + 19, 64, -51, -35, 17, 0, -7, 41, + -16, 27, 4, 15, -1, 18, -16, 47, + -39, -54, -8, 13, -25, -20, 102, -18, + -5, 44, 11, -28, 71, 2, -51, -5, + 5, 2, -83, -9, -29, 8, 21, -53, + 58, -37, -7, 13, 38, 9, 34, -1, + -41, 21, 4, -24, -36, -33, -21, 32, + 75, -2, 1, -68, -1, 47, -29, 32, + 20, 12, -65, -87, 5, 16, -12, 24, + 40, 15, 7, 19, -26, -17, 17, 6, + -2, -37, -30, -9, 32, -127, -39, 0, + -31, -27, 4, -22, 23, -6, -77, 35, + -61, 32, -37, -24, 13, -11, -1, -40, + -3, 17, -7, 13, 11, 59, -19, 10, + 6, -18, 0, 13, 3, -6, -23, 19, + 11, -17, 13, -1, -80, 40, -53, 69, + -29, -54, 0, -4, 33, -25, -2, 38, + 35, 36, -15, 46, 2, -13, -16, -8, + -8, 12, -24, -9, -55, -5, -9, 32, + 11, 7, 12, -18, -10, -86, -38, 54, + 37, -25, 18, -43, 7, -27, -27, -54, + 13, 9, 22, 70, 6, 35, -7, 23, + -15, -44, -6, 7, -66, -85, 32, 40, + -19, -9, -7, 12, -15, 7, 2, 6, + -35, 11, 28, 0, 26, 14, 1, 1, + 4, 12, 18, 35, 22, -18, -3, 14, + -1, 7, 14, -8, -14, -3, 4, -3, + -19, -7, -1, -25, -27, 25, -26, -2, + 33, -22, -27, -25, 4, -9, 7, 21, + 26, -30, 10, -9, -20, 11, 27, 10, + 5, -18, 14, -4, 2, -17, -5, -7, + -9, -13, 15, 29, 1, -10, -16, -10, + 35, 36, -7, -22, -44, 17, 30, 22, + 21, -1, 22, -11, 32, -8, -7, 5, + -10, 5, 30, -20, 29, -20, -34, 12, + -4, -6, 6, -13, 10, -5, -68, -1, + 24, 9, 19, -24, -64, 31, 19, 27, + -26, 75, -45, 41, 39, -42, 8, 6, + 23, -30, 16, -25, 30, 34, 8, -38, + -3, 18, 16, -31, 22, -4, -9, 1, + 20, 9, 38, -32, 0, -45, 0, -6, + -13, 11, -25, -32, -22, 31, -24, -11, + -11, -4, -4, 20, -34, 22, 20, 9, + -25, 27, -5, 28, -29, 29, 6, 21, + -6, -18, 54, 4, -46, 23, 21, -14, + -31, 36, -41, -24, 4, 22, 10, 11, + 7, 36, -32, -13, -52, -17, 24, 28, + -37, -36, -1, 24, 9, -38, 35, 48, + 18, 2, -1, 45, 10, 39, 24, -38, + 13, 8, -16, 8, 25, 11, 7, -29, + -11, 7, 20, -30, -38, -45, 14, -18, + -28, -9, 65, 61, 22, -53, -38, -16, + 36, 46, 20, -39, 32, -61, -6, -6, + -36, -33, -18, -28, 56, 101, 45, 11, + -28, -23, -29, -61, 20, -47, 2, 48, + 27, -17, 1, 40, 1, 3, -51, 15, + 35, 28, 22, 35, 53, -61, -29, 12, + -6, -21, 10, 3, -20, 2, -25, 1, + -6, 31, 11, -3, 1, -10, -52, 6, + 126, -105, 122, 127, -128, 127, 127, -128, + 127, 108, 12, 127, 48, -128, -36, -128, + 127, 127, -128, -128, 127, 89, -128, 127, + -128, -128, -128, 127, 127, -128, -128, -93, + -82, 20, 125, 65, -82, 127, 38, -74, + 81, 88, -88, 79, 51, -47, -111, -26, + 14, 83, -88, -112, 24, 35, -101, 98, + -99, -48, -45, 46, 83, -60, -79, 45, + -20, -41, 9, 4, 52, 54, 93, -10, + 4, 13, 3, 123, 6, 94, -111, -69, + -14, -31, 10, 12, 53, -79, -11, -21, + -2, -44, -72, 92, 65, -57, 56, -38, + 127, -56, -128, 127, 127, -128, 86, 117, + -75, -128, 127, -19, -99, -112, 127, -128, + 127, -48, 114, 118, -128, -128, 117, -17, + -6, 121, -128, 127, -128, 82, 54, -106, + 127, 127, -33, 100, -39, -23, 18, -78, + -34, -29, -1, -30, 127, -26, 127, -128, + 126, -128, 27, -23, -79, -120, -127, 127, + 72, 66, 29, 7, -66, -56, -117, -128 +}; + +static const opus_int8 layer0_bias[32] = { + 51, -16, 1, 13, -5, -6, -16, -7, + 11, -6, 106, 26, 28, -14, 21, -29, + 7, 18, -18, -17, 21, -17, -9, 20, + -25, -3, -34, 48, 11, -13, -31, -20 +}; + +static const opus_int8 layer1_weights[2304] = { + 22, -1, -7, 7, 29, -27, -31, -17, + -13, 33, 44, -8, 11, 33, 24, 78, + 15, 19, 30, -2, -24, 5, 49, 5, + 36, 29, -14, -11, -48, -33, 21, -42, + -38, -12, 55, -37, 54, -8, 1, 36, + 17, 0, 51, 31, 59, 7, -12, 53, + 4, 32, -14, 48, 5, -10, -16, -8, + 1, -16, -56, -24, -6, 18, -2, 23, + 6, 46, -6, -10, 20, 35, -44, -15, + -49, 36, 16, 5, -7, -79, -67, 12, + 70, -3, -79, -54, -85, -24, 47, -22, + 33, 21, 69, -1, 11, 22, 14, -16, + -16, -22, -28, -11, 11, -41, 31, -26, + -33, -19, -4, 27, 32, -50, 5, -10, + -38, -22, -8, 35, -31, 1, -41, -15, + -11, 44, 28, -17, -41, -23, 17, 2, + -23, -26, -13, -13, -17, 6, 14, -31, + -25, 9, -19, 39, -8, 4, 31, -1, + -45, -11, -28, -92, -46, -15, 21, 118, + -22, 45, -51, 11, -20, -20, -15, 13, + -21, -97, -29, -32, -23, -42, 94, 1, + 23, -8, 63, -3, -46, 19, -26, 32, + -40, -74, -26, 26, -4, -13, 30, -20, + -30, -25, -14, -31, -45, -43, 4, -60, + -48, -12, -34, 2, 2, 3, 13, 15, + 11, 16, 5, 46, -9, -55, -16, -57, + 29, 14, 38, -50, -2, -44, -11, -8, + 52, -27, -38, -7, 20, 47, 17, -59, + 0, 47, 46, -63, 35, -17, 19, 33, + 68, -19, 2, 15, -16, 28, -16, -103, + 26, -35, 47, -39, -60, 30, 31, -23, + -52, -13, 116, 47, -25, 30, 40, 30, + -22, 2, 12, -27, -18, 31, -10, 27, + -8, -66, 12, 14, 4, -26, -28, -13, + 3, 13, -26, -51, 37, 5, 2, -21, + 47, 3, 13, 25, -41, -27, -8, -4, + 5, -76, -33, 28, 10, 9, -46, -74, + 19, 28, 25, 31, 54, -55, 68, 38, + -24, -32, 2, 4, 68, 11, -1, 99, + 5, 16, -2, -74, 40, 26, -26, 33, + 31, -1, -68, 14, -6, 25, 9, 29, + 60, 61, 7, -7, 0, -24, 7, 77, + 4, -1, 16, -7, 13, -15, -19, 28, + -31, -24, -16, 37, 24, 13, 30, 10, + -30, 11, 11, -10, 22, 60, 28, 45, + -3, -40, -62, -5, -102, 9, -32, -27, + -54, 21, 15, -5, 37, -43, -11, 37, + -19, 47, -64, -128, -27, -114, 21, -66, + 59, 46, -3, -12, -87, -9, 4, 19, + -113, -36, 78, 57, -26, -38, -77, -10, + 6, 6, -75, 25, -97, -11, 33, -46, + 1, 13, -21, -33, -20, 16, -6, -3, + -11, -4, -27, 38, 8, -41, -2, -33, + 18, 19, -26, 1, -29, -22, -4, -14, + -55, -11, -80, -3, 11, 34, 90, 51, + 11, 17, 43, 36, 127, -32, 29, 103, + 9, 27, 13, 64, 56, 70, -14, 3, + -12, 10, 37, 3, 12, -22, -10, 46, + 28, 10, 20, 26, -24, 18, 9, 7, + 14, 34, -5, -7, 31, -14, -56, 11, + -18, -8, -17, -7, -10, -40, 10, -33, + -32, -43, 5, 9, 11, -4, 10, 50, + -12, -5, 46, 9, 7, 1, 11, 15, + 91, -17, 7, -50, 23, 6, -30, -99, + 0, -17, 14, 8, -10, -25, -30, -69, + -62, 31, 127, 114, -23, 101, -5, -54, + -6, -22, 7, -56, 39, 18, -29, 0, + 46, 8, -79, 4, -21, 18, -32, 62, + -12, -8, -12, -58, 31, -32, 17, 6, + -24, 25, 24, 9, -4, -19, 45, 6, + 17, -14, 5, -27, 16, -4, -41, 25, + -36, 5, 15, 12, 50, 27, 25, 23, + -44, -69, -9, -19, -48, -8, 4, 12, + -6, 13, -19, -30, -36, 26, 37, -1, + -3, -30, -42, -14, -10, -20, 26, -54, + -27, -44, 4, 73, -26, 90, 32, -69, + -29, -16, 3, 103, 15, -17, 37, 24, + -23, -31, 33, -37, -64, 25, 13, -81, + -28, -32, 27, 5, -35, -23, 15, -22, + 19, -7, 9, 30, 19, -23, 27, -13, + 43, 29, -29, -6, 9, -40, -33, -33, + -32, 9, 11, -48, -8, -23, -52, 46, + 17, -22, -42, 35, -15, -41, 16, 34, + 31, -42, -19, -11, 55, 7, -39, 89, + -11, -33, 20, -14, 22, 32, 3, -17, + -6, 14, 34, 1, 55, -21, -90, -8, + 18, 27, 13, -29, 21, 15, -33, -51, + -9, -11, 4, -16, -18, 23, -4, -4, + 48, 1, 7, 29, -14, -12, -16, 17, + 35, 8, 0, -7, -2, 9, 8, 17, + -6, 53, -32, -21, -50, 5, 99, -60, + -5, -53, 10, -31, 12, -5, 7, 80, + 36, 18, -31, 9, 98, 36, -63, -35, + 4, -13, -28, -24, 28, -13, 18, 16, + -1, -18, -34, 10, 20, 7, 4, 29, + 11, 25, -7, 36, 14, 45, 24, 1, + -16, 30, 6, 35, -6, -11, -24, 13, + -1, 27, 39, 20, 48, -11, -4, -13, + 28, 11, -31, -18, 31, -29, 22, -2, + -20, -16, 5, 30, -12, -28, -3, 93, + -16, 23, 18, -29, 6, -54, -37, 28, + -3, -3, -47, -3, -36, -55, -3, 41, + -10, 47, -2, 23, 42, -7, -71, -27, + 83, -64, 7, -24, 8, 26, -17, 15, + 12, 31, -30, -38, -13, -33, -56, 4, + -17, 20, 18, 1, -30, -5, -6, -31, + -14, -37, 0, 22, 10, -30, 37, -17, + 18, 6, 5, 23, -36, -32, 14, 18, + -13, -61, -52, -69, 44, -30, 16, 18, + -4, -25, 14, 81, 26, -8, -23, -59, + 52, -104, 17, 119, -32, 26, 17, 1, + 23, 45, 29, -64, -57, -14, 73, 21, + -13, -13, 9, -68, -7, -52, 3, 24, + -39, 44, -15, 27, 14, 19, -9, -28, + -11, 5, 3, -34, -2, 2, 22, -6, + -23, 4, 3, 13, -22, -13, -10, -18, + 29, 6, 44, -13, -24, -8, 2, 30, + 14, 43, 6, 17, -73, -6, -7, 20, + -80, -7, -7, -28, 15, -69, -38, -5, + -100, -35, 15, -79, 23, 29, -18, -27, + 21, -66, -37, 8, -22, -39, 48, 4, + -13, 1, -9, 11, -29, 22, 6, -49, + 32, -14, 47, -18, -4, 44, -52, -74, + 43, 30, 23, -14, 5, 0, -27, 4, + -7, 10, -4, 10, 1, -16, 11, -18, + -2, -5, 2, -11, 0, -20, -4, 38, + 74, 59, 39, 64, -10, 26, -3, -40, + -68, 3, -30, -51, 8, -19, -27, -46, + 51, 52, 54, 36, 90, 92, 14, 13, + -5, 0, 16, -62, 16, 11, -47, -37, + -6, -5, 21, 54, -57, 32, 42, -6, + 62, -9, 16, 21, 24, 9, -10, -4, + 33, 50, 13, -15, 1, -35, -48, 18, + -11, -17, -67, -13, 21, 38, -44, 36, + -16, 29, 17, 5, -10, 18, 17, -32, + 2, 8, 22, -56, -15, -32, 40, 43, + 19, 46, -7, -100, -96, 19, 53, 24, + 21, -26, -48, -101, -82, 61, 38, -85, + -28, -34, -1, 63, -5, -5, 39, 39, + -38, 32, -12, -28, 20, 40, -8, 2, + 31, 12, -35, -13, 20, -25, 30, 8, + 3, -13, -9, -20, 2, -13, 24, 37, + -10, 33, 6, 20, -16, -24, -6, -6, + -19, -5, 22, 21, 10, 11, -4, -39, + -1, 6, 49, 41, -15, -57, 21, -62, + 77, -69, -13, 0, -74, 1, -7, -38, + -8, 6, 63, 28, 4, 26, -52, 82, + 63, 13, 45, -33, 44, -52, -65, -21, + -46, -49, 64, -17, 32, 24, 68, -39, + -16, -5, -26, 28, 5, -61, -28, 2, + 24, 11, -12, -33, 9, -37, -3, -28, + 22, -37, -12, 19, 0, -18, -2, 14, + 1, 4, 8, -9, -2, 43, -17, -2, + -66, -31, 56, -40, -87, -36, -2, -4, + -42, -45, -1, 31, -43, -15, 27, 63, + -11, 32, -10, -33, 27, -19, 4, 15, + -26, -34, 29, -4, -39, -65, 14, -20, + -21, -17, -36, 13, 59, 47, -38, -33, + 13, -37, -8, -37, -7, -6, -76, -31, + -12, -46, 7, 24, -21, -30, -14, 9, + 15, -12, -13, 47, -27, -25, -1, -39, + 0, 20, -9, 6, 7, 4, 3, 7, + 39, 50, 22, -7, 14, -20, 1, 70, + -28, 29, -41, 10, -16, -5, -28, -2, + -37, 32, -18, 17, 62, -11, -20, -50, + 36, 21, -62, -12, -56, 52, 50, 17, + 3, 48, 44, -41, -25, 3, 16, -3, + 0, 33, -6, 15, 27, 34, -25, 22, + 9, 17, -11, 36, 16, -2, 12, 21, + -52, 45, -2, -10, 46, 21, -18, 67, + -28, -13, 30, 37, 42, 16, -9, 11, + 75, 7, -64, -40, -10, 29, 57, -23, + 5, 53, -77, 3, -17, -5, 47, -55, + -35, -36, -13, 52, -53, -71, 52, -111, + -23, -26, -28, 29, -43, 55, -19, 43, + -19, 54, -12, -33, -44, -39, -19, -10, + -31, -10, 21, 38, -57, -20, 2, -25, + 8, -6, 50, 12, 15, 25, -25, 15, + -30, -6, 9, 25, 37, 19, -4, 31, + -22, 2, 4, 2, 36, 7, 3, -34, + -80, 36, -10, -2, -5, 31, -36, 49, + -70, 20, -36, 21, 24, 25, -46, -51, + 36, -58, -48, -40, -10, 55, 71, 47, + 10, -1, 1, 2, -46, -68, 16, 13, + 0, -74, -29, 73, -52, -18, -11, 7, + -44, -82, -32, -70, -28, -1, -39, -68, + -6, -41, 12, -22, -16, 40, -11, -25, + 51, -9, 21, 4, 4, -34, 7, -78, + 16, 6, -38, -30, -2, -44, 32, 0, + 22, 64, 5, -72, -2, -14, -10, -16, + -8, -25, 12, 102, -58, 37, -10, -23, + 15, 49, 7, -7, 2, -20, -32, 45, + -6, 48, 28, 30, 33, -1, 22, -6, + 30, 65, -17, 29, 74, 37, -26, -10, + 15, -24, 19, -66, 22, -10, -31, -1, + -18, -9, 11, 37, -4, 45, 5, 41, + 17, 1, 1, 24, -58, 41, 5, -51, + 14, 8, 43, 16, -10, -1, 45, 32, + -64, 3, -33, -25, -3, -27, -68, 12, + 23, -11, -13, -37, -40, 4, -21, -12, + 32, -23, -19, 76, 41, -23, -24, -44, + -65, -1, -15, 1, 71, 63, 5, 20, + -3, 21, -23, 31, -32, 18, -2, 27, + 31, 46, -5, -39, -5, -35, 18, -18, + -40, -10, 3, 12, 2, -2, -22, 40, + 5, -6, 60, 36, 3, 29, -27, 10, + 25, -54, 5, 26, 39, 35, -24, -37, + 30, -91, 28, -4, -21, -27, -39, -6, + 5, 12, -128, 38, -16, 29, -95, -29, + 82, -2, 35, 2, 12, 8, -22, 10, + 80, -47, 2, -25, -73, -79, 16, -30, + -32, -66, 48, 21, -45, -11, -47, 14, + -27, -17, -7, 15, -44, -14, -44, -26, + -32, 26, -23, 17, -7, -28, 26, -6, + 28, 6, -26, 2, 13, -14, -23, -14, + 19, 46, 16, 2, -33, -21, 28, -17, + -42, 44, -37, 1, -39, 28, 84, -46, + 15, 10, 13, -44, 72, -26, 26, 32, + -28, -12, -83, 2, 10, -30, -44, -10, + -28, 53, 45, 65, 0, -25, 57, 36, + -33, 6, 29, 44, -53, 11, 19, -2, + -27, 35, 32, 49, 4, 23, 38, 36, + 24, 10, 51, -39, 4, -7, 26, 37, + -35, 11, -47, -18, 28, 16, -35, 42, + 17, -21, -41, 28, 14, -12, 11, -45, + 7, -43, -15, 18, -5, 38, -40, -50, + -30, -21, 9, -98, 13, 12, 23, 75, + -56, -7, -3, -4, -1, -34, 12, -49, + 11, 26, -18, -28, -17, 33, 13, -14, + 40, 24, -72, -37, 10, 17, -6, 22, + 16, 16, -6, -12, -30, -14, 10, 40, + -23, 12, 15, -3, -15, 13, -56, -4, + -30, 1, -3, -17, 27, 50, -5, 64, + -36, -19, 7, 29, 22, 25, 9, -16, + -58, -69, -40, -61, -71, -14, 42, 93, + 26, 11, -6, -58, -11, 70, -52, 19, + 9, -30, -33, 11, -37, -47, -21, -22, + -40, 10, 47, 4, -23, 17, 48, 41, + -48, 14, 10, 15, 34, -23, -2, -47, + 23, -32, -13, -10, -26, -26, -4, 16, + 38, -14, 0, -12, -7, -7, 20, 44, + -1, -32, -27, -16, 4, -6, -18, 14, + 5, 4, -29, 28, 7, -7, 15, -11, + -20, -45, -36, 16, 84, 34, -59, -30, + 22, 126, 8, 68, 79, -17, 21, -68, + 37, 5, 15, 63, 49, 127, -90, 85, + 43, 7, 16, 9, 6, -45, -57, -43, + 57, 11, -23, -11, -29, 60, -26, 0, + 7, 42, -24, 10, 23, -25, 8, -7, + -40, 19, -17, 35, 4, 27, -39, -91, + 27, -36, 34, 2, 16, -24, 25, 7, + -21, 5, 17, 10, -22, -30, 9, -17, + -61, -26, 33, 21, 58, -51, -14, 69, + -38, 20, 7, 80, -4, -65, -6, -27, + 53, -12, 47, -1, -15, 1, 60, 102, + -79, -4, 12, 9, 22, 37, -8, -4, + 37, 2, -3, -15, -16, -11, -5, 19, + -6, -43, 20, -25, -18, 10, -27, 0, + -28, -27, -11, 10, -18, -2, -4, -16, + 26, 14, -6, 7, -6, 1, 53, -2, + -29, 23, 9, -30, -6, -4, -6, 56, + 70, 0, -33, -20, -17, -9, -24, 46, + -5, -105, 47, -46, -51, 20, 20, -53, + -81, -1, -7, 75, -5, -21, -65, 12, + -52, 22, -50, -12, 49, 54, 76, -81, + 10, 45, -41, -59, 18, -19, 25, 14, + -31, -53, -5, 12, 31, 84, -23, 2, + 7, 2, 10, -32, 39, -2, -12, 1, + -9, 0, -10, -11, 9, 15, -8, -2, + 2, -1, 10, 14, -5, -40, 19, -7, + -7, 26, -4, 2, 1, -27, 35, 32, + 21, -31, 26, 43, -9, 4, -32, 40, + -62, -52, 36, 22, 38, 22, 36, -96, + 6, -10, -23, -49, 15, -33, -18, -3, + 0, 41, 21, -19, 21, 23, -39, -23, + -6, 6, 47, 56, 4, 74, 0, -98, + 29, -47, -14, -36, 21, -22, 22, 16, + 13, 12, 16, -5, 13, 17, -13, -15, + 1, -34, -26, 26, 12, 32, 27, 13, + -67, 27, 2, 8, 10, 18, 16, 20, + -17, -17, 57, -64, 5, 14, 19, 31, + -18, -44, -46, -16, 4, -25, 17, -126, + -24, 39, 4, 8, 55, -25, -34, 39, + -16, 3, 9, 71, 72, -31, -55, 6, + 10, -25, 32, -85, -21, 18, -8, 15, + 12, -27, -7, 1, -21, -2, -5, 48, + -16, 18, 1, -22, -26, 16, 14, -31, + 27, -6, -15, -21, 4, -14, 18, -36 +}; + +static const opus_int8 layer1_recur_weights[1728] = { + 20, 67, -99, 12, 41, -25, 49, -44, + 35, 81, 110, 47, 34, -66, -14, 14, + -60, 34, 29, -73, 10, 41, 35, 89, + 7, -35, 22, 7, 27, -20, -6, 56, + 26, 66, 6, 33, -55, 53, 1, -21, + 14, 17, 68, 55, 59, 0, 18, -9, + 5, -41, 6, -5, -114, -12, 29, 42, + -23, 10, 81, -27, 20, -53, -30, -62, + 40, 95, 25, -4, 3, 18, -8, -15, + -29, -82, 2, -57, -3, -61, -29, -29, + 49, 2, -55, 5, -69, -99, -49, -51, + 6, -25, 12, 89, 44, -33, 5, 41, + 1, 23, -37, -37, -28, -48, 3, 4, + -41, -30, -57, -35, -39, -1, -13, -56, + -5, 50, 49, 41, -4, -4, 33, -22, + -1, 33, 34, 18, 40, -42, 12, 1, + -6, -2, 18, 17, 39, 44, 11, 65, + -60, -45, 10, 91, 21, 9, -62, -11, + 8, 69, 37, 24, -30, 21, 26, -27, + 1, -28, 24, 66, -8, 6, -71, 34, + 24, 44, 58, -78, -19, 57, 17, -60, + 1, 12, -3, -1, -40, 22, 11, -5, + 25, 12, 1, 72, 79, 7, -50, 23, + 18, 13, 21, -11, -20, 5, 77, -94, + 24, 15, 57, -51, 3, 36, 53, -1, + 4, 14, 30, -31, 22, 40, 32, -11, + -34, -36, -59, 58, 25, 21, -54, -23, + 40, 46, 18, 0, 12, 54, -96, -99, + -59, 5, 119, -38, 50, 55, 12, -16, + 67, 0, 34, 35, 39, 35, -1, 69, + 24, 27, -30, -35, -4, -70, 2, -44, + -7, -6, 19, -9, 60, 44, -21, -10, + 37, 43, -16, -3, 30, -15, -65, 31, + -55, 18, -98, 76, 64, 25, 24, -18, + -7, -68, -10, 38, 27, -60, 36, 33, + 16, 30, 34, -39, -37, 31, 12, 53, + -54, 14, -26, -49, -128, -13, -5, -22, + -11, -85, 55, -8, -51, -11, -33, -10, + -31, -76, -41, 23, 44, -40, -54, -127, + -101, 19, -23, -15, 15, 27, 58, -60, + 8, 14, -33, 1, 48, -9, -11, -123, + 3, 53, 23, 4, -28, 22, 2, -29, + -67, 36, 12, 7, 55, -21, 88, 20, + -1, -21, -17, 3, 41, 32, -10, -14, + -5, -57, 67, 57, 21, 23, -2, -27, + -73, -24, 120, 21, 18, -35, 42, -7, + 3, -45, -25, 76, -34, 50, 11, -54, + -91, 3, -113, -20, -5, 47, 15, -47, + 17, 27, -3, -26, -7, 10, 7, 74, + -40, 64, -7, -5, -24, -49, -24, -3, + -10, 27, -17, -8, -3, 14, -27, 33, + 13, 39, 28, -7, -38, 29, 16, 44, + 19, 55, -3, 9, -13, -57, 43, 43, + 31, 0, -93, -17, 19, -56, 4, -12, + -25, 37, -85, -13, -118, 33, -17, 56, + 71, -80, -4, 6, -11, -18, 47, -52, + 25, 9, 48, -107, 1, 21, 20, -3, + 10, -16, -4, 24, 17, 31, -61, -18, + -50, 24, -10, 12, 71, 26, 11, -3, + 4, 1, 0, -7, -40, 18, 38, -34, + 38, 17, 8, -34, 2, 21, 123, -32, + -26, 43, 14, -34, -1, -9, 37, -16, + 6, -17, -62, 68, 22, 17, 11, -75, + 33, -80, 62, -9, -75, 76, 36, -41, + -8, -40, -11, -71, 40, -39, 62, -49, + -81, 16, -9, -52, 52, 61, 17, -103, + -27, -10, -8, -54, -57, 21, 23, -16, + -52, 36, 18, 10, -5, 8, 15, -29, + 5, -19, -37, 8, -53, 6, 19, -37, + 38, -17, 48, 10, 0, 81, 46, 70, + -29, 101, 11, 44, -44, -3, 24, 11, + 3, 14, -9, 11, 14, -45, 13, 46, + -3, -57, 68, 44, 63, 98, 25, -28, + -23, 15, 32, -10, 53, -6, -2, -9, + -6, 16, -107, -11, -11, -28, 59, 57, + -22, 38, 42, 83, 27, 5, 29, -30, + 12, -21, -13, 31, 38, -21, 58, -10, + -10, -15, -2, -5, 11, 12, -73, -28, + -38, 22, 2, -25, 73, -52, -12, -55, + 32, -63, 21, 51, 33, 52, -26, 55, + -26, -26, 57, -32, -4, -52, -61, 21, + -33, -91, -51, 69, -90, -53, -38, -44, + 12, -76, -20, 77, -45, -7, 86, 43, + -109, -33, -105, -40, -121, -10, 0, -72, + 45, -51, -75, -49, -38, -1, -62, 18, + -1, 30, -44, -14, -10, -67, 40, -10, + -34, 46, -64, -32, 29, -13, 33, 3, + -32, -5, 28, -27, -25, 93, 24, 68, + -40, 57, 23, -3, -21, -58, 17, -39, + -17, -22, -89, 11, 18, -46, 27, 24, + 46, 127, 61, 87, 31, 127, -36, 47, + -23, 47, 127, -24, 110, 122, 30, 100, + 0, 96, -12, 6, 50, 44, -13, 73, + 4, 55, -11, -15, 49, 42, -6, 20, + -35, 58, 18, 38, 42, 72, 19, -21, + 11, 9, -37, 7, 29, 31, 16, -17, + 13, -50, 19, 5, -23, 51, -16, -5, + 4, -24, 76, 10, -53, -28, -7, -65, + 74, 40, -16, -29, 32, -16, -49, -35, + -3, 59, -96, -50, -43, -43, -61, -15, + -8, -36, -34, -33, -14, 11, -3, -39, + 4, -114, -123, -11, -49, -21, 14, -56, + 1, 43, -63, 26, 40, 18, -10, -26, + -14, -15, -35, -35, -11, 32, -44, -67, + 2, 22, 7, 3, -9, -30, -51, -28, + 28, 6, -22, 16, 34, -25, -52, -54, + -8, -6, 5, 8, 20, -16, -17, -44, + 27, 3, 31, -5, -48, -1, -3, 116, + 11, 71, -31, -47, 109, 50, -22, -12, + -57, 32, 66, 8, -25, -93, -54, -10, + 19, -76, -34, 97, 48, -36, -18, -30, + -39, -26, -12, 28, 14, 12, -12, -31, + 38, 2, 10, 4, -40, 20, 16, -61, + 2, 64, 39, 5, 15, 33, 40, -61, + -49, 93, -10, 33, 28, -11, -27, -18, + 39, -62, -6, -6, 62, 11, -8, 38, + -67, 12, 27, 39, -27, 123, -18, -6, + -65, 83, -64, 20, 19, -11, 33, 24, + 17, 56, 78, 7, -15, 54, -101, -9, + 115, -96, 50, 51, 35, 34, 27, 37, + -40, -11, 8, -36, 42, -45, 2, -23, + 0, 67, -8, -9, -13, 50, -14, -27, + 4, 0, -8, -14, 30, -9, 29, 15, + 9, -38, 37, -8, 50, -46, 54, 41, + -11, -8, -11, -26, 39, 45, 14, -26, + -17, -27, 69, 38, 39, 98, 66, 0, + 42, 123, -101, -19, -83, 117, -32, 56, + 10, 12, -88, 79, -53, 56, 63, 95, + -62, 9, 36, -13, -79, -16, 37, -46, + 35, -34, 14, 17, -54, 5, 21, -7, + 7, 63, 56, 15, 27, -76, -25, 4, + -26, -63, 28, -67, -52, 43, -47, -70, + 40, -12, 40, -66, -37, 0, 35, 37, + -53, 4, -17, -51, 11, 21, 14, -34, + -4, 24, -42, 29, 22, 7, 28, 12, + 37, 39, -39, -19, 65, -60, -50, -2, + 1, 82, 39, 19, -23, -43, -22, -67, + -35, -34, 32, 102, 81, 127, 36, 67, + -45, 1, -67, -52, -4, 35, 20, 28, + 71, 86, -35, -9, -83, -34, 12, 9, + -23, 2, 14, 28, -23, 7, -25, 45, + 7, 17, -37, 0, -19, 31, 26, 40, + -27, -16, 17, 5, -21, 23, 24, 96, + -55, 52, -19, -14, -6, 1, 50, -34, + 86, -53, 38, 2, -52, -36, -13, 60, + -85, -120, 32, 7, -12, 22, 70, -7, + -94, 38, -76, -31, -20, 15, -28, 7, + 6, 40, 53, 88, 3, 38, 18, -8, + -22, -23, 51, 37, -9, 13, -32, 25, + -21, 27, 31, 20, 18, -9, -13, 1, + 21, -24, -13, 39, 15, -11, -29, -36, + 18, 15, 8, 27, 21, -94, -1, -22, + 49, 66, -1, 6, -3, -40, -18, 6, + 28, 12, 33, -59, 62, 60, -48, 90, + -1, 108, 9, 18, -2, 27, 77, -65, + 82, -48, -38, -19, -11, 127, 50, 66, + 18, -13, -22, 60, -38, 40, -14, -26, + -13, 38, 67, 57, 30, 33, 26, 36, + 38, -17, 27, -28, 20, 12, -64, 18, + 5, -33, -27, 13, -26, 32, 35, -5, + -48, -14, 92, 43, -47, -14, 40, 11, + 51, 66, 22, -63, -16, -61, 4, -28, + 27, 20, -33, -30, -21, -29, -53, 31, + -40, 24, 43, -4, -19, 21, 67, 20, + 100, -16, -93, 78, -6, -18, -52, -37, + -9, 66, -31, -8, 26, 18, 4, 24, + -22, 17, -2, -13, 27, 0, 8, -18, + -25, 5, -21, -24, -7, 18, -93, 21, + 7, 2, -75, 69, 50, -5, -15, -17, + 60, -42, 55, 1, -4, 3, 10, 46, + 16, -13, 45, -7, -10, -44, -108, 49, + 2, -15, -64, -12, -72, 32, -38, -45, + 10, -54, 13, -13, -27, -36, -64, 58, + -62, -101, 88, -86, -71, -39, -9, -128, + 32, 15, -4, 54, -16, -39, -26, -36, + 46, 48, -64, -10, 19, 30, -13, 34, + -8, 50, 60, -22, -6, -11, -30, 5, + 50, 32, 56, 0, 25, 6, 68, 11, + -29, 45, -9, -12, 4, 1, 18, -49, + 0, -38, -19, 90, 29, 35, 51, 8, + -48, 96, -1, -12, -9, -32, -63, -65, + -7, 38, 89, 28, -85, -28, -23, -25, + -128, 56, 79, -36, 99, -6, -37, 7, + -13, -69, -46, -29, 25, 64, -21, 17, + 1, 42, -66, 1, 80, 26, -32, 21, + 15, 15, 6, 6, -10, 15, 127, 5, + 38, 27, 87, -57, -25, 11, 72, -21, + -5, 11, -13, -66, 78, 36, -3, 41, + -21, 8, -33, 23, 73, 28, 57, -25, + -5, 4, -22, -47, 15, 4, -57, -72, + 33, 1, 18, 2, 53, -71, -99, -21, + -3, -111, 108, 71, -14, 82, 25, 61, + -48, 5, 9, -51, -20, -25, -3, 14, + -33, 14, -3, -34, 22, 12, -19, -38, + -16, 2, 21, 16, 26, -31, 75, 44, + -31, 16, 26, 66, 17, -9, -22, -22, + 22, -44, 22, 27, 2, 58, -14, 10, + -73, -42, 55, -25, -61, 72, -1, 30, + -58, -25, 63, 26, -48, -40, 26, -30, + 60, 8, -17, -1, -18, -20, 43, -20, + -4, -28, 127, -106, 29, 70, 64, -27, + 39, -33, -5, -88, -40, -52, 26, 44, + -17, 23, 2, -49, 22, -9, -8, 86, + 49, -43, -60, 1, 10, 45, 36, -53, + -4, 33, 38, 48, -72, 1, 19, 21, + -65, 4, -5, -62, 27, -25, 17, -6, + 6, -45, -39, -46, 4, 26, 127, -9, + 18, -33, -18, -3, 33, 2, -5, 15, + -26, -22, -117, -63, -17, -59, 61, -74, + 7, -47, -58, -128, -67, 15, -16, -128, + 12, 2, 20, 9, -48, -40, 43, 3, + -40, -16, -38, -6, -22, -28, -16, -59, + -22, 6, -5, 11, -12, -66, -40, 27, + -62, -44, -19, 38, -3, 39, -8, 40, + -24, 13, 21, 50, -60, -22, 53, -29, + -6, 1, 22, -59, 0, 17, -39, 115 +}; + +static const opus_int8 layer1_bias[72] = { + -42, 20, 16, 0, 105, 60, 1, -97, + 24, 60, 18, 13, 62, 25, 127, 34, + 79, 55, 118, 127, 95, 31, -4, 87, + 21, 12, 2, -14, 18, 23, 8, 17, + -1, -8, 5, 4, 24, 37, 21, 13, + 36, 13, 17, 18, 37, 30, 33, 1, + 8, -16, -11, -5, -31, -3, -5, 0, + 6, 3, 58, -7, -1, -16, 5, -13, + 16, 10, -2, -14, 11, -4, 3, -11 +}; + +static const opus_int8 layer2_weights[48] = { + -113, -88, 31, -128, -126, -61, 85, -35, + 118, -128, -61, 127, -128, -17, -128, 127, + 104, -9, -128, 33, 45, 127, 5, 83, + 84, -128, -85, -128, -45, 48, -53, -128, + 46, 127, -17, 125, 117, -41, -117, -91, + -127, -68, -1, -89, -80, 32, 106, 7 +}; + +static const opus_int8 layer2_bias[2] = { + 14, 117 +}; + +const DenseLayer layer0 = { + layer0_bias, + layer0_weights, + 25, 32, 0 +}; + +const GRULayer layer1 = { + layer1_bias, + layer1_weights, + layer1_recur_weights, + 32, 24 +}; + +const DenseLayer layer2 = { + layer2_bias, + layer2_weights, + 24, 2, 1 +}; + diff --git a/media/libopus/src/opus.c b/media/libopus/src/opus.c new file mode 100644 index 0000000000..538b5ea74e --- /dev/null +++ b/media/libopus/src/opus.c @@ -0,0 +1,356 @@ +/* Copyright (c) 2011 Xiph.Org Foundation, Skype Limited + Written by Jean-Marc Valin and Koen Vos */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "opus.h" +#include "opus_private.h" + +#ifndef DISABLE_FLOAT_API +OPUS_EXPORT void opus_pcm_soft_clip(float *_x, int N, int C, float *declip_mem) +{ + int c; + int i; + float *x; + + if (C<1 || N<1 || !_x || !declip_mem) return; + + /* First thing: saturate everything to +/- 2 which is the highest level our + non-linearity can handle. At the point where the signal reaches +/-2, + the derivative will be zero anyway, so this doesn't introduce any + discontinuity in the derivative. */ + for (i=0;i<N*C;i++) + _x[i] = MAX16(-2.f, MIN16(2.f, _x[i])); + for (c=0;c<C;c++) + { + float a; + float x0; + int curr; + + x = _x+c; + a = declip_mem[c]; + /* Continue applying the non-linearity from the previous frame to avoid + any discontinuity. */ + for (i=0;i<N;i++) + { + if (x[i*C]*a>=0) + break; + x[i*C] = x[i*C]+a*x[i*C]*x[i*C]; + } + + curr=0; + x0 = x[0]; + while(1) + { + int start, end; + float maxval; + int special=0; + int peak_pos; + for (i=curr;i<N;i++) + { + if (x[i*C]>1 || x[i*C]<-1) + break; + } + if (i==N) + { + a=0; + break; + } + peak_pos = i; + start=end=i; + maxval=ABS16(x[i*C]); + /* Look for first zero crossing before clipping */ + while (start>0 && x[i*C]*x[(start-1)*C]>=0) + start--; + /* Look for first zero crossing after clipping */ + while (end<N && x[i*C]*x[end*C]>=0) + { + /* Look for other peaks until the next zero-crossing. */ + if (ABS16(x[end*C])>maxval) + { + maxval = ABS16(x[end*C]); + peak_pos = end; + } + end++; + } + /* Detect the special case where we clip before the first zero crossing */ + special = (start==0 && x[i*C]*x[0]>=0); + + /* Compute a such that maxval + a*maxval^2 = 1 */ + a=(maxval-1)/(maxval*maxval); + /* Slightly boost "a" by 2^-22. This is just enough to ensure -ffast-math + does not cause output values larger than +/-1, but small enough not + to matter even for 24-bit output. */ + a += a*2.4e-7f; + if (x[i*C]>0) + a = -a; + /* Apply soft clipping */ + for (i=start;i<end;i++) + x[i*C] = x[i*C]+a*x[i*C]*x[i*C]; + + if (special && peak_pos>=2) + { + /* Add a linear ramp from the first sample to the signal peak. + This avoids a discontinuity at the beginning of the frame. */ + float delta; + float offset = x0-x[0]; + delta = offset / peak_pos; + for (i=curr;i<peak_pos;i++) + { + offset -= delta; + x[i*C] += offset; + x[i*C] = MAX16(-1.f, MIN16(1.f, x[i*C])); + } + } + curr = end; + if (curr==N) + break; + } + declip_mem[c] = a; + } +} +#endif + +int encode_size(int size, unsigned char *data) +{ + if (size < 252) + { + data[0] = size; + return 1; + } else { + data[0] = 252+(size&0x3); + data[1] = (size-(int)data[0])>>2; + return 2; + } +} + +static int parse_size(const unsigned char *data, opus_int32 len, opus_int16 *size) +{ + if (len<1) + { + *size = -1; + return -1; + } else if (data[0]<252) + { + *size = data[0]; + return 1; + } else if (len<2) + { + *size = -1; + return -1; + } else { + *size = 4*data[1] + data[0]; + return 2; + } +} + +int opus_packet_get_samples_per_frame(const unsigned char *data, + opus_int32 Fs) +{ + int audiosize; + if (data[0]&0x80) + { + audiosize = ((data[0]>>3)&0x3); + audiosize = (Fs<<audiosize)/400; + } else if ((data[0]&0x60) == 0x60) + { + audiosize = (data[0]&0x08) ? Fs/50 : Fs/100; + } else { + audiosize = ((data[0]>>3)&0x3); + if (audiosize == 3) + audiosize = Fs*60/1000; + else + audiosize = (Fs<<audiosize)/100; + } + return audiosize; +} + +int opus_packet_parse_impl(const unsigned char *data, opus_int32 len, + int self_delimited, unsigned char *out_toc, + const unsigned char *frames[48], opus_int16 size[48], + int *payload_offset, opus_int32 *packet_offset) +{ + int i, bytes; + int count; + int cbr; + unsigned char ch, toc; + int framesize; + opus_int32 last_size; + opus_int32 pad = 0; + const unsigned char *data0 = data; + + if (size==NULL || len<0) + return OPUS_BAD_ARG; + if (len==0) + return OPUS_INVALID_PACKET; + + framesize = opus_packet_get_samples_per_frame(data, 48000); + + cbr = 0; + toc = *data++; + len--; + last_size = len; + switch (toc&0x3) + { + /* One frame */ + case 0: + count=1; + break; + /* Two CBR frames */ + case 1: + count=2; + cbr = 1; + if (!self_delimited) + { + if (len&0x1) + return OPUS_INVALID_PACKET; + last_size = len/2; + /* If last_size doesn't fit in size[0], we'll catch it later */ + size[0] = (opus_int16)last_size; + } + break; + /* Two VBR frames */ + case 2: + count = 2; + bytes = parse_size(data, len, size); + len -= bytes; + if (size[0]<0 || size[0] > len) + return OPUS_INVALID_PACKET; + data += bytes; + last_size = len-size[0]; + break; + /* Multiple CBR/VBR frames (from 0 to 120 ms) */ + default: /*case 3:*/ + if (len<1) + return OPUS_INVALID_PACKET; + /* Number of frames encoded in bits 0 to 5 */ + ch = *data++; + count = ch&0x3F; + if (count <= 0 || framesize*(opus_int32)count > 5760) + return OPUS_INVALID_PACKET; + len--; + /* Padding flag is bit 6 */ + if (ch&0x40) + { + int p; + do { + int tmp; + if (len<=0) + return OPUS_INVALID_PACKET; + p = *data++; + len--; + tmp = p==255 ? 254: p; + len -= tmp; + pad += tmp; + } while (p==255); + } + if (len<0) + return OPUS_INVALID_PACKET; + /* VBR flag is bit 7 */ + cbr = !(ch&0x80); + if (!cbr) + { + /* VBR case */ + last_size = len; + for (i=0;i<count-1;i++) + { + bytes = parse_size(data, len, size+i); + len -= bytes; + if (size[i]<0 || size[i] > len) + return OPUS_INVALID_PACKET; + data += bytes; + last_size -= bytes+size[i]; + } + if (last_size<0) + return OPUS_INVALID_PACKET; + } else if (!self_delimited) + { + /* CBR case */ + last_size = len/count; + if (last_size*count!=len) + return OPUS_INVALID_PACKET; + for (i=0;i<count-1;i++) + size[i] = (opus_int16)last_size; + } + break; + } + /* Self-delimited framing has an extra size for the last frame. */ + if (self_delimited) + { + bytes = parse_size(data, len, size+count-1); + len -= bytes; + if (size[count-1]<0 || size[count-1] > len) + return OPUS_INVALID_PACKET; + data += bytes; + /* For CBR packets, apply the size to all the frames. */ + if (cbr) + { + if (size[count-1]*count > len) + return OPUS_INVALID_PACKET; + for (i=0;i<count-1;i++) + size[i] = size[count-1]; + } else if (bytes+size[count-1] > last_size) + return OPUS_INVALID_PACKET; + } else + { + /* Because it's not encoded explicitly, it's possible the size of the + last packet (or all the packets, for the CBR case) is larger than + 1275. Reject them here.*/ + if (last_size > 1275) + return OPUS_INVALID_PACKET; + size[count-1] = (opus_int16)last_size; + } + + if (payload_offset) + *payload_offset = (int)(data-data0); + + for (i=0;i<count;i++) + { + if (frames) + frames[i] = data; + data += size[i]; + } + + if (packet_offset) + *packet_offset = pad+(opus_int32)(data-data0); + + if (out_toc) + *out_toc = toc; + + return count; +} + +int opus_packet_parse(const unsigned char *data, opus_int32 len, + unsigned char *out_toc, const unsigned char *frames[48], + opus_int16 size[48], int *payload_offset) +{ + return opus_packet_parse_impl(data, len, 0, out_toc, + frames, size, payload_offset, NULL); +} + diff --git a/media/libopus/src/opus_decoder.c b/media/libopus/src/opus_decoder.c new file mode 100644 index 0000000000..6520e748ea --- /dev/null +++ b/media/libopus/src/opus_decoder.c @@ -0,0 +1,1041 @@ +/* Copyright (c) 2010 Xiph.Org Foundation, Skype Limited + Written by Jean-Marc Valin and Koen Vos */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +# include "config.h" +#endif + +#ifndef OPUS_BUILD +# error "OPUS_BUILD _MUST_ be defined to build Opus. This probably means you need other defines as well, as in a config.h. See the included build files for details." +#endif + +#if defined(__GNUC__) && (__GNUC__ >= 2) && !defined(__OPTIMIZE__) && !defined(OPUS_WILL_BE_SLOW) +# pragma message "You appear to be compiling without optimization, if so opus will be very slow." +#endif + +#include <stdarg.h> +#include "celt.h" +#include "opus.h" +#include "entdec.h" +#include "modes.h" +#include "API.h" +#include "stack_alloc.h" +#include "float_cast.h" +#include "opus_private.h" +#include "os_support.h" +#include "structs.h" +#include "define.h" +#include "mathops.h" +#include "cpu_support.h" + +struct OpusDecoder { + int celt_dec_offset; + int silk_dec_offset; + int channels; + opus_int32 Fs; /** Sampling rate (at the API level) */ + silk_DecControlStruct DecControl; + int decode_gain; + int arch; + + /* Everything beyond this point gets cleared on a reset */ +#define OPUS_DECODER_RESET_START stream_channels + int stream_channels; + + int bandwidth; + int mode; + int prev_mode; + int frame_size; + int prev_redundancy; + int last_packet_duration; +#ifndef FIXED_POINT + opus_val16 softclip_mem[2]; +#endif + + opus_uint32 rangeFinal; +}; + +#if defined(ENABLE_HARDENING) || defined(ENABLE_ASSERTIONS) +static void validate_opus_decoder(OpusDecoder *st) +{ + celt_assert(st->channels == 1 || st->channels == 2); + celt_assert(st->Fs == 48000 || st->Fs == 24000 || st->Fs == 16000 || st->Fs == 12000 || st->Fs == 8000); + celt_assert(st->DecControl.API_sampleRate == st->Fs); + celt_assert(st->DecControl.internalSampleRate == 0 || st->DecControl.internalSampleRate == 16000 || st->DecControl.internalSampleRate == 12000 || st->DecControl.internalSampleRate == 8000); + celt_assert(st->DecControl.nChannelsAPI == st->channels); + celt_assert(st->DecControl.nChannelsInternal == 0 || st->DecControl.nChannelsInternal == 1 || st->DecControl.nChannelsInternal == 2); + celt_assert(st->DecControl.payloadSize_ms == 0 || st->DecControl.payloadSize_ms == 10 || st->DecControl.payloadSize_ms == 20 || st->DecControl.payloadSize_ms == 40 || st->DecControl.payloadSize_ms == 60); +#ifdef OPUS_ARCHMASK + celt_assert(st->arch >= 0); + celt_assert(st->arch <= OPUS_ARCHMASK); +#endif + celt_assert(st->stream_channels == 1 || st->stream_channels == 2); +} +#define VALIDATE_OPUS_DECODER(st) validate_opus_decoder(st) +#else +#define VALIDATE_OPUS_DECODER(st) +#endif + +int opus_decoder_get_size(int channels) +{ + int silkDecSizeBytes, celtDecSizeBytes; + int ret; + if (channels<1 || channels > 2) + return 0; + ret = silk_Get_Decoder_Size( &silkDecSizeBytes ); + if(ret) + return 0; + silkDecSizeBytes = align(silkDecSizeBytes); + celtDecSizeBytes = celt_decoder_get_size(channels); + return align(sizeof(OpusDecoder))+silkDecSizeBytes+celtDecSizeBytes; +} + +int opus_decoder_init(OpusDecoder *st, opus_int32 Fs, int channels) +{ + void *silk_dec; + CELTDecoder *celt_dec; + int ret, silkDecSizeBytes; + + if ((Fs!=48000&&Fs!=24000&&Fs!=16000&&Fs!=12000&&Fs!=8000) + || (channels!=1&&channels!=2)) + return OPUS_BAD_ARG; + + OPUS_CLEAR((char*)st, opus_decoder_get_size(channels)); + /* Initialize SILK decoder */ + ret = silk_Get_Decoder_Size(&silkDecSizeBytes); + if (ret) + return OPUS_INTERNAL_ERROR; + + silkDecSizeBytes = align(silkDecSizeBytes); + st->silk_dec_offset = align(sizeof(OpusDecoder)); + st->celt_dec_offset = st->silk_dec_offset+silkDecSizeBytes; + silk_dec = (char*)st+st->silk_dec_offset; + celt_dec = (CELTDecoder*)((char*)st+st->celt_dec_offset); + st->stream_channels = st->channels = channels; + + st->Fs = Fs; + st->DecControl.API_sampleRate = st->Fs; + st->DecControl.nChannelsAPI = st->channels; + + /* Reset decoder */ + ret = silk_InitDecoder( silk_dec ); + if(ret)return OPUS_INTERNAL_ERROR; + + /* Initialize CELT decoder */ + ret = celt_decoder_init(celt_dec, Fs, channels); + if(ret!=OPUS_OK)return OPUS_INTERNAL_ERROR; + + celt_decoder_ctl(celt_dec, CELT_SET_SIGNALLING(0)); + + st->prev_mode = 0; + st->frame_size = Fs/400; + st->arch = opus_select_arch(); + return OPUS_OK; +} + +OpusDecoder *opus_decoder_create(opus_int32 Fs, int channels, int *error) +{ + int ret; + OpusDecoder *st; + if ((Fs!=48000&&Fs!=24000&&Fs!=16000&&Fs!=12000&&Fs!=8000) + || (channels!=1&&channels!=2)) + { + if (error) + *error = OPUS_BAD_ARG; + return NULL; + } + st = (OpusDecoder *)opus_alloc(opus_decoder_get_size(channels)); + if (st == NULL) + { + if (error) + *error = OPUS_ALLOC_FAIL; + return NULL; + } + ret = opus_decoder_init(st, Fs, channels); + if (error) + *error = ret; + if (ret != OPUS_OK) + { + opus_free(st); + st = NULL; + } + return st; +} + +static void smooth_fade(const opus_val16 *in1, const opus_val16 *in2, + opus_val16 *out, int overlap, int channels, + const opus_val16 *window, opus_int32 Fs) +{ + int i, c; + int inc = 48000/Fs; + for (c=0;c<channels;c++) + { + for (i=0;i<overlap;i++) + { + opus_val16 w = MULT16_16_Q15(window[i*inc], window[i*inc]); + out[i*channels+c] = SHR32(MAC16_16(MULT16_16(w,in2[i*channels+c]), + Q15ONE-w, in1[i*channels+c]), 15); + } + } +} + +static int opus_packet_get_mode(const unsigned char *data) +{ + int mode; + if (data[0]&0x80) + { + mode = MODE_CELT_ONLY; + } else if ((data[0]&0x60) == 0x60) + { + mode = MODE_HYBRID; + } else { + mode = MODE_SILK_ONLY; + } + return mode; +} + +static int opus_decode_frame(OpusDecoder *st, const unsigned char *data, + opus_int32 len, opus_val16 *pcm, int frame_size, int decode_fec) +{ + void *silk_dec; + CELTDecoder *celt_dec; + int i, silk_ret=0, celt_ret=0; + ec_dec dec; + opus_int32 silk_frame_size; + int pcm_silk_size; + VARDECL(opus_int16, pcm_silk); + int pcm_transition_silk_size; + VARDECL(opus_val16, pcm_transition_silk); + int pcm_transition_celt_size; + VARDECL(opus_val16, pcm_transition_celt); + opus_val16 *pcm_transition=NULL; + int redundant_audio_size; + VARDECL(opus_val16, redundant_audio); + + int audiosize; + int mode; + int bandwidth; + int transition=0; + int start_band; + int redundancy=0; + int redundancy_bytes = 0; + int celt_to_silk=0; + int c; + int F2_5, F5, F10, F20; + const opus_val16 *window; + opus_uint32 redundant_rng = 0; + int celt_accum; + ALLOC_STACK; + + silk_dec = (char*)st+st->silk_dec_offset; + celt_dec = (CELTDecoder*)((char*)st+st->celt_dec_offset); + F20 = st->Fs/50; + F10 = F20>>1; + F5 = F10>>1; + F2_5 = F5>>1; + if (frame_size < F2_5) + { + RESTORE_STACK; + return OPUS_BUFFER_TOO_SMALL; + } + /* Limit frame_size to avoid excessive stack allocations. */ + frame_size = IMIN(frame_size, st->Fs/25*3); + /* Payloads of 1 (2 including ToC) or 0 trigger the PLC/DTX */ + if (len<=1) + { + data = NULL; + /* In that case, don't conceal more than what the ToC says */ + frame_size = IMIN(frame_size, st->frame_size); + } + if (data != NULL) + { + audiosize = st->frame_size; + mode = st->mode; + bandwidth = st->bandwidth; + ec_dec_init(&dec,(unsigned char*)data,len); + } else { + audiosize = frame_size; + /* Run PLC using last used mode (CELT if we ended with CELT redundancy) */ + mode = st->prev_redundancy ? MODE_CELT_ONLY : st->prev_mode; + bandwidth = 0; + + if (mode == 0) + { + /* If we haven't got any packet yet, all we can do is return zeros */ + for (i=0;i<audiosize*st->channels;i++) + pcm[i] = 0; + RESTORE_STACK; + return audiosize; + } + + /* Avoids trying to run the PLC on sizes other than 2.5 (CELT), 5 (CELT), + 10, or 20 (e.g. 12.5 or 30 ms). */ + if (audiosize > F20) + { + do { + int ret = opus_decode_frame(st, NULL, 0, pcm, IMIN(audiosize, F20), 0); + if (ret<0) + { + RESTORE_STACK; + return ret; + } + pcm += ret*st->channels; + audiosize -= ret; + } while (audiosize > 0); + RESTORE_STACK; + return frame_size; + } else if (audiosize < F20) + { + if (audiosize > F10) + audiosize = F10; + else if (mode != MODE_SILK_ONLY && audiosize > F5 && audiosize < F10) + audiosize = F5; + } + } + + /* In fixed-point, we can tell CELT to do the accumulation on top of the + SILK PCM buffer. This saves some stack space. */ +#ifdef FIXED_POINT + celt_accum = (mode != MODE_CELT_ONLY) && (frame_size >= F10); +#else + celt_accum = 0; +#endif + + pcm_transition_silk_size = ALLOC_NONE; + pcm_transition_celt_size = ALLOC_NONE; + if (data!=NULL && st->prev_mode > 0 && ( + (mode == MODE_CELT_ONLY && st->prev_mode != MODE_CELT_ONLY && !st->prev_redundancy) + || (mode != MODE_CELT_ONLY && st->prev_mode == MODE_CELT_ONLY) ) + ) + { + transition = 1; + /* Decide where to allocate the stack memory for pcm_transition */ + if (mode == MODE_CELT_ONLY) + pcm_transition_celt_size = F5*st->channels; + else + pcm_transition_silk_size = F5*st->channels; + } + ALLOC(pcm_transition_celt, pcm_transition_celt_size, opus_val16); + if (transition && mode == MODE_CELT_ONLY) + { + pcm_transition = pcm_transition_celt; + opus_decode_frame(st, NULL, 0, pcm_transition, IMIN(F5, audiosize), 0); + } + if (audiosize > frame_size) + { + /*fprintf(stderr, "PCM buffer too small: %d vs %d (mode = %d)\n", audiosize, frame_size, mode);*/ + RESTORE_STACK; + return OPUS_BAD_ARG; + } else { + frame_size = audiosize; + } + + /* Don't allocate any memory when in CELT-only mode */ + pcm_silk_size = (mode != MODE_CELT_ONLY && !celt_accum) ? IMAX(F10, frame_size)*st->channels : ALLOC_NONE; + ALLOC(pcm_silk, pcm_silk_size, opus_int16); + + /* SILK processing */ + if (mode != MODE_CELT_ONLY) + { + int lost_flag, decoded_samples; + opus_int16 *pcm_ptr; +#ifdef FIXED_POINT + if (celt_accum) + pcm_ptr = pcm; + else +#endif + pcm_ptr = pcm_silk; + + if (st->prev_mode==MODE_CELT_ONLY) + silk_InitDecoder( silk_dec ); + + /* The SILK PLC cannot produce frames of less than 10 ms */ + st->DecControl.payloadSize_ms = IMAX(10, 1000 * audiosize / st->Fs); + + if (data != NULL) + { + st->DecControl.nChannelsInternal = st->stream_channels; + if( mode == MODE_SILK_ONLY ) { + if( bandwidth == OPUS_BANDWIDTH_NARROWBAND ) { + st->DecControl.internalSampleRate = 8000; + } else if( bandwidth == OPUS_BANDWIDTH_MEDIUMBAND ) { + st->DecControl.internalSampleRate = 12000; + } else if( bandwidth == OPUS_BANDWIDTH_WIDEBAND ) { + st->DecControl.internalSampleRate = 16000; + } else { + st->DecControl.internalSampleRate = 16000; + celt_assert( 0 ); + } + } else { + /* Hybrid mode */ + st->DecControl.internalSampleRate = 16000; + } + } + + lost_flag = data == NULL ? 1 : 2 * decode_fec; + decoded_samples = 0; + do { + /* Call SILK decoder */ + int first_frame = decoded_samples == 0; + silk_ret = silk_Decode( silk_dec, &st->DecControl, + lost_flag, first_frame, &dec, pcm_ptr, &silk_frame_size, st->arch ); + if( silk_ret ) { + if (lost_flag) { + /* PLC failure should not be fatal */ + silk_frame_size = frame_size; + for (i=0;i<frame_size*st->channels;i++) + pcm_ptr[i] = 0; + } else { + RESTORE_STACK; + return OPUS_INTERNAL_ERROR; + } + } + pcm_ptr += silk_frame_size * st->channels; + decoded_samples += silk_frame_size; + } while( decoded_samples < frame_size ); + } + + start_band = 0; + if (!decode_fec && mode != MODE_CELT_ONLY && data != NULL + && ec_tell(&dec)+17+20*(mode == MODE_HYBRID) <= 8*len) + { + /* Check if we have a redundant 0-8 kHz band */ + if (mode == MODE_HYBRID) + redundancy = ec_dec_bit_logp(&dec, 12); + else + redundancy = 1; + if (redundancy) + { + celt_to_silk = ec_dec_bit_logp(&dec, 1); + /* redundancy_bytes will be at least two, in the non-hybrid + case due to the ec_tell() check above */ + redundancy_bytes = mode==MODE_HYBRID ? + (opus_int32)ec_dec_uint(&dec, 256)+2 : + len-((ec_tell(&dec)+7)>>3); + len -= redundancy_bytes; + /* This is a sanity check. It should never happen for a valid + packet, so the exact behaviour is not normative. */ + if (len*8 < ec_tell(&dec)) + { + len = 0; + redundancy_bytes = 0; + redundancy = 0; + } + /* Shrink decoder because of raw bits */ + dec.storage -= redundancy_bytes; + } + } + if (mode != MODE_CELT_ONLY) + start_band = 17; + + if (redundancy) + { + transition = 0; + pcm_transition_silk_size=ALLOC_NONE; + } + + ALLOC(pcm_transition_silk, pcm_transition_silk_size, opus_val16); + + if (transition && mode != MODE_CELT_ONLY) + { + pcm_transition = pcm_transition_silk; + opus_decode_frame(st, NULL, 0, pcm_transition, IMIN(F5, audiosize), 0); + } + + + if (bandwidth) + { + int endband=21; + + switch(bandwidth) + { + case OPUS_BANDWIDTH_NARROWBAND: + endband = 13; + break; + case OPUS_BANDWIDTH_MEDIUMBAND: + case OPUS_BANDWIDTH_WIDEBAND: + endband = 17; + break; + case OPUS_BANDWIDTH_SUPERWIDEBAND: + endband = 19; + break; + case OPUS_BANDWIDTH_FULLBAND: + endband = 21; + break; + default: + celt_assert(0); + break; + } + MUST_SUCCEED(celt_decoder_ctl(celt_dec, CELT_SET_END_BAND(endband))); + } + MUST_SUCCEED(celt_decoder_ctl(celt_dec, CELT_SET_CHANNELS(st->stream_channels))); + + /* Only allocation memory for redundancy if/when needed */ + redundant_audio_size = redundancy ? F5*st->channels : ALLOC_NONE; + ALLOC(redundant_audio, redundant_audio_size, opus_val16); + + /* 5 ms redundant frame for CELT->SILK*/ + if (redundancy && celt_to_silk) + { + /* If the previous frame did not use CELT (the first redundancy frame in + a transition from SILK may have been lost) then the CELT decoder is + stale at this point and the redundancy audio is not useful, however + the final range is still needed (for testing), so the redundancy is + always decoded but the decoded audio may not be used */ + MUST_SUCCEED(celt_decoder_ctl(celt_dec, CELT_SET_START_BAND(0))); + celt_decode_with_ec(celt_dec, data+len, redundancy_bytes, + redundant_audio, F5, NULL, 0); + MUST_SUCCEED(celt_decoder_ctl(celt_dec, OPUS_GET_FINAL_RANGE(&redundant_rng))); + } + + /* MUST be after PLC */ + MUST_SUCCEED(celt_decoder_ctl(celt_dec, CELT_SET_START_BAND(start_band))); + + if (mode != MODE_SILK_ONLY) + { + int celt_frame_size = IMIN(F20, frame_size); + /* Make sure to discard any previous CELT state */ + if (mode != st->prev_mode && st->prev_mode > 0 && !st->prev_redundancy) + MUST_SUCCEED(celt_decoder_ctl(celt_dec, OPUS_RESET_STATE)); + /* Decode CELT */ + celt_ret = celt_decode_with_ec(celt_dec, decode_fec ? NULL : data, + len, pcm, celt_frame_size, &dec, celt_accum); + } else { + unsigned char silence[2] = {0xFF, 0xFF}; + if (!celt_accum) + { + for (i=0;i<frame_size*st->channels;i++) + pcm[i] = 0; + } + /* For hybrid -> SILK transitions, we let the CELT MDCT + do a fade-out by decoding a silence frame */ + if (st->prev_mode == MODE_HYBRID && !(redundancy && celt_to_silk && st->prev_redundancy) ) + { + MUST_SUCCEED(celt_decoder_ctl(celt_dec, CELT_SET_START_BAND(0))); + celt_decode_with_ec(celt_dec, silence, 2, pcm, F2_5, NULL, celt_accum); + } + } + + if (mode != MODE_CELT_ONLY && !celt_accum) + { +#ifdef FIXED_POINT + for (i=0;i<frame_size*st->channels;i++) + pcm[i] = SAT16(ADD32(pcm[i], pcm_silk[i])); +#else + for (i=0;i<frame_size*st->channels;i++) + pcm[i] = pcm[i] + (opus_val16)((1.f/32768.f)*pcm_silk[i]); +#endif + } + + { + const CELTMode *celt_mode; + MUST_SUCCEED(celt_decoder_ctl(celt_dec, CELT_GET_MODE(&celt_mode))); + window = celt_mode->window; + } + + /* 5 ms redundant frame for SILK->CELT */ + if (redundancy && !celt_to_silk) + { + MUST_SUCCEED(celt_decoder_ctl(celt_dec, OPUS_RESET_STATE)); + MUST_SUCCEED(celt_decoder_ctl(celt_dec, CELT_SET_START_BAND(0))); + + celt_decode_with_ec(celt_dec, data+len, redundancy_bytes, redundant_audio, F5, NULL, 0); + MUST_SUCCEED(celt_decoder_ctl(celt_dec, OPUS_GET_FINAL_RANGE(&redundant_rng))); + smooth_fade(pcm+st->channels*(frame_size-F2_5), redundant_audio+st->channels*F2_5, + pcm+st->channels*(frame_size-F2_5), F2_5, st->channels, window, st->Fs); + } + /* 5ms redundant frame for CELT->SILK; ignore if the previous frame did not + use CELT (the first redundancy frame in a transition from SILK may have + been lost) */ + if (redundancy && celt_to_silk && (st->prev_mode != MODE_SILK_ONLY || st->prev_redundancy)) + { + for (c=0;c<st->channels;c++) + { + for (i=0;i<F2_5;i++) + pcm[st->channels*i+c] = redundant_audio[st->channels*i+c]; + } + smooth_fade(redundant_audio+st->channels*F2_5, pcm+st->channels*F2_5, + pcm+st->channels*F2_5, F2_5, st->channels, window, st->Fs); + } + if (transition) + { + if (audiosize >= F5) + { + for (i=0;i<st->channels*F2_5;i++) + pcm[i] = pcm_transition[i]; + smooth_fade(pcm_transition+st->channels*F2_5, pcm+st->channels*F2_5, + pcm+st->channels*F2_5, F2_5, + st->channels, window, st->Fs); + } else { + /* Not enough time to do a clean transition, but we do it anyway + This will not preserve amplitude perfectly and may introduce + a bit of temporal aliasing, but it shouldn't be too bad and + that's pretty much the best we can do. In any case, generating this + transition it pretty silly in the first place */ + smooth_fade(pcm_transition, pcm, + pcm, F2_5, + st->channels, window, st->Fs); + } + } + + if(st->decode_gain) + { + opus_val32 gain; + gain = celt_exp2(MULT16_16_P15(QCONST16(6.48814081e-4f, 25), st->decode_gain)); + for (i=0;i<frame_size*st->channels;i++) + { + opus_val32 x; + x = MULT16_32_P16(pcm[i],gain); + pcm[i] = SATURATE(x, 32767); + } + } + + if (len <= 1) + st->rangeFinal = 0; + else + st->rangeFinal = dec.rng ^ redundant_rng; + + st->prev_mode = mode; + st->prev_redundancy = redundancy && !celt_to_silk; + + if (celt_ret>=0) + { + if (OPUS_CHECK_ARRAY(pcm, audiosize*st->channels)) + OPUS_PRINT_INT(audiosize); + } + + RESTORE_STACK; + return celt_ret < 0 ? celt_ret : audiosize; + +} + +int opus_decode_native(OpusDecoder *st, const unsigned char *data, + opus_int32 len, opus_val16 *pcm, int frame_size, int decode_fec, + int self_delimited, opus_int32 *packet_offset, int soft_clip) +{ + int i, nb_samples; + int count, offset; + unsigned char toc; + int packet_frame_size, packet_bandwidth, packet_mode, packet_stream_channels; + /* 48 x 2.5 ms = 120 ms */ + opus_int16 size[48]; + VALIDATE_OPUS_DECODER(st); + if (decode_fec<0 || decode_fec>1) + return OPUS_BAD_ARG; + /* For FEC/PLC, frame_size has to be to have a multiple of 2.5 ms */ + if ((decode_fec || len==0 || data==NULL) && frame_size%(st->Fs/400)!=0) + return OPUS_BAD_ARG; + if (len==0 || data==NULL) + { + int pcm_count=0; + do { + int ret; + ret = opus_decode_frame(st, NULL, 0, pcm+pcm_count*st->channels, frame_size-pcm_count, 0); + if (ret<0) + return ret; + pcm_count += ret; + } while (pcm_count < frame_size); + celt_assert(pcm_count == frame_size); + if (OPUS_CHECK_ARRAY(pcm, pcm_count*st->channels)) + OPUS_PRINT_INT(pcm_count); + st->last_packet_duration = pcm_count; + return pcm_count; + } else if (len<0) + return OPUS_BAD_ARG; + + packet_mode = opus_packet_get_mode(data); + packet_bandwidth = opus_packet_get_bandwidth(data); + packet_frame_size = opus_packet_get_samples_per_frame(data, st->Fs); + packet_stream_channels = opus_packet_get_nb_channels(data); + + count = opus_packet_parse_impl(data, len, self_delimited, &toc, NULL, + size, &offset, packet_offset); + if (count<0) + return count; + + data += offset; + + if (decode_fec) + { + int duration_copy; + int ret; + /* If no FEC can be present, run the PLC (recursive call) */ + if (frame_size < packet_frame_size || packet_mode == MODE_CELT_ONLY || st->mode == MODE_CELT_ONLY) + return opus_decode_native(st, NULL, 0, pcm, frame_size, 0, 0, NULL, soft_clip); + /* Otherwise, run the PLC on everything except the size for which we might have FEC */ + duration_copy = st->last_packet_duration; + if (frame_size-packet_frame_size!=0) + { + ret = opus_decode_native(st, NULL, 0, pcm, frame_size-packet_frame_size, 0, 0, NULL, soft_clip); + if (ret<0) + { + st->last_packet_duration = duration_copy; + return ret; + } + celt_assert(ret==frame_size-packet_frame_size); + } + /* Complete with FEC */ + st->mode = packet_mode; + st->bandwidth = packet_bandwidth; + st->frame_size = packet_frame_size; + st->stream_channels = packet_stream_channels; + ret = opus_decode_frame(st, data, size[0], pcm+st->channels*(frame_size-packet_frame_size), + packet_frame_size, 1); + if (ret<0) + return ret; + else { + if (OPUS_CHECK_ARRAY(pcm, frame_size*st->channels)) + OPUS_PRINT_INT(frame_size); + st->last_packet_duration = frame_size; + return frame_size; + } + } + + if (count*packet_frame_size > frame_size) + return OPUS_BUFFER_TOO_SMALL; + + /* Update the state as the last step to avoid updating it on an invalid packet */ + st->mode = packet_mode; + st->bandwidth = packet_bandwidth; + st->frame_size = packet_frame_size; + st->stream_channels = packet_stream_channels; + + nb_samples=0; + for (i=0;i<count;i++) + { + int ret; + ret = opus_decode_frame(st, data, size[i], pcm+nb_samples*st->channels, frame_size-nb_samples, 0); + if (ret<0) + return ret; + celt_assert(ret==packet_frame_size); + data += size[i]; + nb_samples += ret; + } + st->last_packet_duration = nb_samples; + if (OPUS_CHECK_ARRAY(pcm, nb_samples*st->channels)) + OPUS_PRINT_INT(nb_samples); +#ifndef FIXED_POINT + if (soft_clip) + opus_pcm_soft_clip(pcm, nb_samples, st->channels, st->softclip_mem); + else + st->softclip_mem[0]=st->softclip_mem[1]=0; +#endif + return nb_samples; +} + +#ifdef FIXED_POINT + +int opus_decode(OpusDecoder *st, const unsigned char *data, + opus_int32 len, opus_val16 *pcm, int frame_size, int decode_fec) +{ + if(frame_size<=0) + return OPUS_BAD_ARG; + return opus_decode_native(st, data, len, pcm, frame_size, decode_fec, 0, NULL, 0); +} + +#ifndef DISABLE_FLOAT_API +int opus_decode_float(OpusDecoder *st, const unsigned char *data, + opus_int32 len, float *pcm, int frame_size, int decode_fec) +{ + VARDECL(opus_int16, out); + int ret, i; + int nb_samples; + ALLOC_STACK; + + if(frame_size<=0) + { + RESTORE_STACK; + return OPUS_BAD_ARG; + } + if (data != NULL && len > 0 && !decode_fec) + { + nb_samples = opus_decoder_get_nb_samples(st, data, len); + if (nb_samples>0) + frame_size = IMIN(frame_size, nb_samples); + else + return OPUS_INVALID_PACKET; + } + celt_assert(st->channels == 1 || st->channels == 2); + ALLOC(out, frame_size*st->channels, opus_int16); + + ret = opus_decode_native(st, data, len, out, frame_size, decode_fec, 0, NULL, 0); + if (ret > 0) + { + for (i=0;i<ret*st->channels;i++) + pcm[i] = (1.f/32768.f)*(out[i]); + } + RESTORE_STACK; + return ret; +} +#endif + + +#else +int opus_decode(OpusDecoder *st, const unsigned char *data, + opus_int32 len, opus_int16 *pcm, int frame_size, int decode_fec) +{ + VARDECL(float, out); + int ret, i; + int nb_samples; + ALLOC_STACK; + + if(frame_size<=0) + { + RESTORE_STACK; + return OPUS_BAD_ARG; + } + + if (data != NULL && len > 0 && !decode_fec) + { + nb_samples = opus_decoder_get_nb_samples(st, data, len); + if (nb_samples>0) + frame_size = IMIN(frame_size, nb_samples); + else + return OPUS_INVALID_PACKET; + } + celt_assert(st->channels == 1 || st->channels == 2); + ALLOC(out, frame_size*st->channels, float); + + ret = opus_decode_native(st, data, len, out, frame_size, decode_fec, 0, NULL, 1); + if (ret > 0) + { + for (i=0;i<ret*st->channels;i++) + pcm[i] = FLOAT2INT16(out[i]); + } + RESTORE_STACK; + return ret; +} + +int opus_decode_float(OpusDecoder *st, const unsigned char *data, + opus_int32 len, opus_val16 *pcm, int frame_size, int decode_fec) +{ + if(frame_size<=0) + return OPUS_BAD_ARG; + return opus_decode_native(st, data, len, pcm, frame_size, decode_fec, 0, NULL, 0); +} + +#endif + +int opus_decoder_ctl(OpusDecoder *st, int request, ...) +{ + int ret = OPUS_OK; + va_list ap; + void *silk_dec; + CELTDecoder *celt_dec; + + silk_dec = (char*)st+st->silk_dec_offset; + celt_dec = (CELTDecoder*)((char*)st+st->celt_dec_offset); + + + va_start(ap, request); + + switch (request) + { + case OPUS_GET_BANDWIDTH_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->bandwidth; + } + break; + case OPUS_GET_FINAL_RANGE_REQUEST: + { + opus_uint32 *value = va_arg(ap, opus_uint32*); + if (!value) + { + goto bad_arg; + } + *value = st->rangeFinal; + } + break; + case OPUS_RESET_STATE: + { + OPUS_CLEAR((char*)&st->OPUS_DECODER_RESET_START, + sizeof(OpusDecoder)- + ((char*)&st->OPUS_DECODER_RESET_START - (char*)st)); + + celt_decoder_ctl(celt_dec, OPUS_RESET_STATE); + silk_InitDecoder( silk_dec ); + st->stream_channels = st->channels; + st->frame_size = st->Fs/400; + } + break; + case OPUS_GET_SAMPLE_RATE_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->Fs; + } + break; + case OPUS_GET_PITCH_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + if (st->prev_mode == MODE_CELT_ONLY) + ret = celt_decoder_ctl(celt_dec, OPUS_GET_PITCH(value)); + else + *value = st->DecControl.prevPitchLag; + } + break; + case OPUS_GET_GAIN_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->decode_gain; + } + break; + case OPUS_SET_GAIN_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value<-32768 || value>32767) + { + goto bad_arg; + } + st->decode_gain = value; + } + break; + case OPUS_GET_LAST_PACKET_DURATION_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->last_packet_duration; + } + break; + case OPUS_SET_PHASE_INVERSION_DISABLED_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if(value<0 || value>1) + { + goto bad_arg; + } + ret = celt_decoder_ctl(celt_dec, OPUS_SET_PHASE_INVERSION_DISABLED(value)); + } + break; + case OPUS_GET_PHASE_INVERSION_DISABLED_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + ret = celt_decoder_ctl(celt_dec, OPUS_GET_PHASE_INVERSION_DISABLED(value)); + } + break; + default: + /*fprintf(stderr, "unknown opus_decoder_ctl() request: %d", request);*/ + ret = OPUS_UNIMPLEMENTED; + break; + } + + va_end(ap); + return ret; +bad_arg: + va_end(ap); + return OPUS_BAD_ARG; +} + +void opus_decoder_destroy(OpusDecoder *st) +{ + opus_free(st); +} + + +int opus_packet_get_bandwidth(const unsigned char *data) +{ + int bandwidth; + if (data[0]&0x80) + { + bandwidth = OPUS_BANDWIDTH_MEDIUMBAND + ((data[0]>>5)&0x3); + if (bandwidth == OPUS_BANDWIDTH_MEDIUMBAND) + bandwidth = OPUS_BANDWIDTH_NARROWBAND; + } else if ((data[0]&0x60) == 0x60) + { + bandwidth = (data[0]&0x10) ? OPUS_BANDWIDTH_FULLBAND : + OPUS_BANDWIDTH_SUPERWIDEBAND; + } else { + bandwidth = OPUS_BANDWIDTH_NARROWBAND + ((data[0]>>5)&0x3); + } + return bandwidth; +} + +int opus_packet_get_nb_channels(const unsigned char *data) +{ + return (data[0]&0x4) ? 2 : 1; +} + +int opus_packet_get_nb_frames(const unsigned char packet[], opus_int32 len) +{ + int count; + if (len<1) + return OPUS_BAD_ARG; + count = packet[0]&0x3; + if (count==0) + return 1; + else if (count!=3) + return 2; + else if (len<2) + return OPUS_INVALID_PACKET; + else + return packet[1]&0x3F; +} + +int opus_packet_get_nb_samples(const unsigned char packet[], opus_int32 len, + opus_int32 Fs) +{ + int samples; + int count = opus_packet_get_nb_frames(packet, len); + + if (count<0) + return count; + + samples = count*opus_packet_get_samples_per_frame(packet, Fs); + /* Can't have more than 120 ms */ + if (samples*25 > Fs*3) + return OPUS_INVALID_PACKET; + else + return samples; +} + +int opus_decoder_get_nb_samples(const OpusDecoder *dec, + const unsigned char packet[], opus_int32 len) +{ + return opus_packet_get_nb_samples(packet, len, dec->Fs); +} diff --git a/media/libopus/src/opus_encoder.c b/media/libopus/src/opus_encoder.c new file mode 100644 index 0000000000..8c8db5a546 --- /dev/null +++ b/media/libopus/src/opus_encoder.c @@ -0,0 +1,2780 @@ +/* Copyright (c) 2010-2011 Xiph.Org Foundation, Skype Limited + Written by Jean-Marc Valin and Koen Vos */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include <stdarg.h> +#include "celt.h" +#include "entenc.h" +#include "modes.h" +#include "API.h" +#include "stack_alloc.h" +#include "float_cast.h" +#include "opus.h" +#include "arch.h" +#include "pitch.h" +#include "opus_private.h" +#include "os_support.h" +#include "cpu_support.h" +#include "analysis.h" +#include "mathops.h" +#include "tuning_parameters.h" +#ifdef FIXED_POINT +#include "fixed/structs_FIX.h" +#else +#include "float/structs_FLP.h" +#endif + +#define MAX_ENCODER_BUFFER 480 + +#ifndef DISABLE_FLOAT_API +#define PSEUDO_SNR_THRESHOLD 316.23f /* 10^(25/10) */ +#endif + +typedef struct { + opus_val32 XX, XY, YY; + opus_val16 smoothed_width; + opus_val16 max_follower; +} StereoWidthState; + +struct OpusEncoder { + int celt_enc_offset; + int silk_enc_offset; + silk_EncControlStruct silk_mode; + int application; + int channels; + int delay_compensation; + int force_channels; + int signal_type; + int user_bandwidth; + int max_bandwidth; + int user_forced_mode; + int voice_ratio; + opus_int32 Fs; + int use_vbr; + int vbr_constraint; + int variable_duration; + opus_int32 bitrate_bps; + opus_int32 user_bitrate_bps; + int lsb_depth; + int encoder_buffer; + int lfe; + int arch; + int use_dtx; /* general DTX for both SILK and CELT */ + int fec_config; +#ifndef DISABLE_FLOAT_API + TonalityAnalysisState analysis; +#endif + +#define OPUS_ENCODER_RESET_START stream_channels + int stream_channels; + opus_int16 hybrid_stereo_width_Q14; + opus_int32 variable_HP_smth2_Q15; + opus_val16 prev_HB_gain; + opus_val32 hp_mem[4]; + int mode; + int prev_mode; + int prev_channels; + int prev_framesize; + int bandwidth; + /* Bandwidth determined automatically from the rate (before any other adjustment) */ + int auto_bandwidth; + int silk_bw_switch; + /* Sampling rate (at the API level) */ + int first; + opus_val16 * energy_masking; + StereoWidthState width_mem; + opus_val16 delay_buffer[MAX_ENCODER_BUFFER*2]; +#ifndef DISABLE_FLOAT_API + int detected_bandwidth; + int nb_no_activity_ms_Q1; + opus_val32 peak_signal_energy; +#endif + int nonfinal_frame; /* current frame is not the final in a packet */ + opus_uint32 rangeFinal; +}; + +/* Transition tables for the voice and music. First column is the + middle (memoriless) threshold. The second column is the hysteresis + (difference with the middle) */ +static const opus_int32 mono_voice_bandwidth_thresholds[8] = { + 9000, 700, /* NB<->MB */ + 9000, 700, /* MB<->WB */ + 13500, 1000, /* WB<->SWB */ + 14000, 2000, /* SWB<->FB */ +}; +static const opus_int32 mono_music_bandwidth_thresholds[8] = { + 9000, 700, /* NB<->MB */ + 9000, 700, /* MB<->WB */ + 11000, 1000, /* WB<->SWB */ + 12000, 2000, /* SWB<->FB */ +}; +static const opus_int32 stereo_voice_bandwidth_thresholds[8] = { + 9000, 700, /* NB<->MB */ + 9000, 700, /* MB<->WB */ + 13500, 1000, /* WB<->SWB */ + 14000, 2000, /* SWB<->FB */ +}; +static const opus_int32 stereo_music_bandwidth_thresholds[8] = { + 9000, 700, /* NB<->MB */ + 9000, 700, /* MB<->WB */ + 11000, 1000, /* WB<->SWB */ + 12000, 2000, /* SWB<->FB */ +}; +/* Threshold bit-rates for switching between mono and stereo */ +static const opus_int32 stereo_voice_threshold = 19000; +static const opus_int32 stereo_music_threshold = 17000; + +/* Threshold bit-rate for switching between SILK/hybrid and CELT-only */ +static const opus_int32 mode_thresholds[2][2] = { + /* voice */ /* music */ + { 64000, 10000}, /* mono */ + { 44000, 10000}, /* stereo */ +}; + +static const opus_int32 fec_thresholds[] = { + 12000, 1000, /* NB */ + 14000, 1000, /* MB */ + 16000, 1000, /* WB */ + 20000, 1000, /* SWB */ + 22000, 1000, /* FB */ +}; + +int opus_encoder_get_size(int channels) +{ + int silkEncSizeBytes, celtEncSizeBytes; + int ret; + if (channels<1 || channels > 2) + return 0; + ret = silk_Get_Encoder_Size( &silkEncSizeBytes ); + if (ret) + return 0; + silkEncSizeBytes = align(silkEncSizeBytes); + celtEncSizeBytes = celt_encoder_get_size(channels); + return align(sizeof(OpusEncoder))+silkEncSizeBytes+celtEncSizeBytes; +} + +int opus_encoder_init(OpusEncoder* st, opus_int32 Fs, int channels, int application) +{ + void *silk_enc; + CELTEncoder *celt_enc; + int err; + int ret, silkEncSizeBytes; + + if((Fs!=48000&&Fs!=24000&&Fs!=16000&&Fs!=12000&&Fs!=8000)||(channels!=1&&channels!=2)|| + (application != OPUS_APPLICATION_VOIP && application != OPUS_APPLICATION_AUDIO + && application != OPUS_APPLICATION_RESTRICTED_LOWDELAY)) + return OPUS_BAD_ARG; + + OPUS_CLEAR((char*)st, opus_encoder_get_size(channels)); + /* Create SILK encoder */ + ret = silk_Get_Encoder_Size( &silkEncSizeBytes ); + if (ret) + return OPUS_BAD_ARG; + silkEncSizeBytes = align(silkEncSizeBytes); + st->silk_enc_offset = align(sizeof(OpusEncoder)); + st->celt_enc_offset = st->silk_enc_offset+silkEncSizeBytes; + silk_enc = (char*)st+st->silk_enc_offset; + celt_enc = (CELTEncoder*)((char*)st+st->celt_enc_offset); + + st->stream_channels = st->channels = channels; + + st->Fs = Fs; + + st->arch = opus_select_arch(); + + ret = silk_InitEncoder( silk_enc, st->arch, &st->silk_mode ); + if(ret)return OPUS_INTERNAL_ERROR; + + /* default SILK parameters */ + st->silk_mode.nChannelsAPI = channels; + st->silk_mode.nChannelsInternal = channels; + st->silk_mode.API_sampleRate = st->Fs; + st->silk_mode.maxInternalSampleRate = 16000; + st->silk_mode.minInternalSampleRate = 8000; + st->silk_mode.desiredInternalSampleRate = 16000; + st->silk_mode.payloadSize_ms = 20; + st->silk_mode.bitRate = 25000; + st->silk_mode.packetLossPercentage = 0; + st->silk_mode.complexity = 9; + st->silk_mode.useInBandFEC = 0; + st->silk_mode.useDTX = 0; + st->silk_mode.useCBR = 0; + st->silk_mode.reducedDependency = 0; + + /* Create CELT encoder */ + /* Initialize CELT encoder */ + err = celt_encoder_init(celt_enc, Fs, channels, st->arch); + if(err!=OPUS_OK)return OPUS_INTERNAL_ERROR; + + celt_encoder_ctl(celt_enc, CELT_SET_SIGNALLING(0)); + celt_encoder_ctl(celt_enc, OPUS_SET_COMPLEXITY(st->silk_mode.complexity)); + + st->use_vbr = 1; + /* Makes constrained VBR the default (safer for real-time use) */ + st->vbr_constraint = 1; + st->user_bitrate_bps = OPUS_AUTO; + st->bitrate_bps = 3000+Fs*channels; + st->application = application; + st->signal_type = OPUS_AUTO; + st->user_bandwidth = OPUS_AUTO; + st->max_bandwidth = OPUS_BANDWIDTH_FULLBAND; + st->force_channels = OPUS_AUTO; + st->user_forced_mode = OPUS_AUTO; + st->voice_ratio = -1; + st->encoder_buffer = st->Fs/100; + st->lsb_depth = 24; + st->variable_duration = OPUS_FRAMESIZE_ARG; + + /* Delay compensation of 4 ms (2.5 ms for SILK's extra look-ahead + + 1.5 ms for SILK resamplers and stereo prediction) */ + st->delay_compensation = st->Fs/250; + + st->hybrid_stereo_width_Q14 = 1 << 14; + st->prev_HB_gain = Q15ONE; + st->variable_HP_smth2_Q15 = silk_LSHIFT( silk_lin2log( VARIABLE_HP_MIN_CUTOFF_HZ ), 8 ); + st->first = 1; + st->mode = MODE_HYBRID; + st->bandwidth = OPUS_BANDWIDTH_FULLBAND; + +#ifndef DISABLE_FLOAT_API + tonality_analysis_init(&st->analysis, st->Fs); + st->analysis.application = st->application; +#endif + + return OPUS_OK; +} + +static unsigned char gen_toc(int mode, int framerate, int bandwidth, int channels) +{ + int period; + unsigned char toc; + period = 0; + while (framerate < 400) + { + framerate <<= 1; + period++; + } + if (mode == MODE_SILK_ONLY) + { + toc = (bandwidth-OPUS_BANDWIDTH_NARROWBAND)<<5; + toc |= (period-2)<<3; + } else if (mode == MODE_CELT_ONLY) + { + int tmp = bandwidth-OPUS_BANDWIDTH_MEDIUMBAND; + if (tmp < 0) + tmp = 0; + toc = 0x80; + toc |= tmp << 5; + toc |= period<<3; + } else /* Hybrid */ + { + toc = 0x60; + toc |= (bandwidth-OPUS_BANDWIDTH_SUPERWIDEBAND)<<4; + toc |= (period-2)<<3; + } + toc |= (channels==2)<<2; + return toc; +} + +#ifndef FIXED_POINT +static void silk_biquad_float( + const opus_val16 *in, /* I: Input signal */ + const opus_int32 *B_Q28, /* I: MA coefficients [3] */ + const opus_int32 *A_Q28, /* I: AR coefficients [2] */ + opus_val32 *S, /* I/O: State vector [2] */ + opus_val16 *out, /* O: Output signal */ + const opus_int32 len, /* I: Signal length (must be even) */ + int stride +) +{ + /* DIRECT FORM II TRANSPOSED (uses 2 element state vector) */ + opus_int k; + opus_val32 vout; + opus_val32 inval; + opus_val32 A[2], B[3]; + + A[0] = (opus_val32)(A_Q28[0] * (1.f/((opus_int32)1<<28))); + A[1] = (opus_val32)(A_Q28[1] * (1.f/((opus_int32)1<<28))); + B[0] = (opus_val32)(B_Q28[0] * (1.f/((opus_int32)1<<28))); + B[1] = (opus_val32)(B_Q28[1] * (1.f/((opus_int32)1<<28))); + B[2] = (opus_val32)(B_Q28[2] * (1.f/((opus_int32)1<<28))); + + /* Negate A_Q28 values and split in two parts */ + + for( k = 0; k < len; k++ ) { + /* S[ 0 ], S[ 1 ]: Q12 */ + inval = in[ k*stride ]; + vout = S[ 0 ] + B[0]*inval; + + S[ 0 ] = S[1] - vout*A[0] + B[1]*inval; + + S[ 1 ] = - vout*A[1] + B[2]*inval + VERY_SMALL; + + /* Scale back to Q0 and saturate */ + out[ k*stride ] = vout; + } +} +#endif + +static void hp_cutoff(const opus_val16 *in, opus_int32 cutoff_Hz, opus_val16 *out, opus_val32 *hp_mem, int len, int channels, opus_int32 Fs, int arch) +{ + opus_int32 B_Q28[ 3 ], A_Q28[ 2 ]; + opus_int32 Fc_Q19, r_Q28, r_Q22; + (void)arch; + + silk_assert( cutoff_Hz <= silk_int32_MAX / SILK_FIX_CONST( 1.5 * 3.14159 / 1000, 19 ) ); + Fc_Q19 = silk_DIV32_16( silk_SMULBB( SILK_FIX_CONST( 1.5 * 3.14159 / 1000, 19 ), cutoff_Hz ), Fs/1000 ); + silk_assert( Fc_Q19 > 0 && Fc_Q19 < 32768 ); + + r_Q28 = SILK_FIX_CONST( 1.0, 28 ) - silk_MUL( SILK_FIX_CONST( 0.92, 9 ), Fc_Q19 ); + + /* b = r * [ 1; -2; 1 ]; */ + /* a = [ 1; -2 * r * ( 1 - 0.5 * Fc^2 ); r^2 ]; */ + B_Q28[ 0 ] = r_Q28; + B_Q28[ 1 ] = silk_LSHIFT( -r_Q28, 1 ); + B_Q28[ 2 ] = r_Q28; + + /* -r * ( 2 - Fc * Fc ); */ + r_Q22 = silk_RSHIFT( r_Q28, 6 ); + A_Q28[ 0 ] = silk_SMULWW( r_Q22, silk_SMULWW( Fc_Q19, Fc_Q19 ) - SILK_FIX_CONST( 2.0, 22 ) ); + A_Q28[ 1 ] = silk_SMULWW( r_Q22, r_Q22 ); + +#ifdef FIXED_POINT + if( channels == 1 ) { + silk_biquad_alt_stride1( in, B_Q28, A_Q28, hp_mem, out, len ); + } else { + silk_biquad_alt_stride2( in, B_Q28, A_Q28, hp_mem, out, len, arch ); + } +#else + silk_biquad_float( in, B_Q28, A_Q28, hp_mem, out, len, channels ); + if( channels == 2 ) { + silk_biquad_float( in+1, B_Q28, A_Q28, hp_mem+2, out+1, len, channels ); + } +#endif +} + +#ifdef FIXED_POINT +static void dc_reject(const opus_val16 *in, opus_int32 cutoff_Hz, opus_val16 *out, opus_val32 *hp_mem, int len, int channels, opus_int32 Fs) +{ + int c, i; + int shift; + + /* Approximates -round(log2(6.3*cutoff_Hz/Fs)) */ + shift=celt_ilog2(Fs/(cutoff_Hz*4)); + for (c=0;c<channels;c++) + { + for (i=0;i<len;i++) + { + opus_val32 x, y; + x = SHL32(EXTEND32(in[channels*i+c]), 14); + y = x-hp_mem[2*c]; + hp_mem[2*c] = hp_mem[2*c] + PSHR32(x - hp_mem[2*c], shift); + out[channels*i+c] = EXTRACT16(SATURATE(PSHR32(y, 14), 32767)); + } + } +} + +#else +static void dc_reject(const opus_val16 *in, opus_int32 cutoff_Hz, opus_val16 *out, opus_val32 *hp_mem, int len, int channels, opus_int32 Fs) +{ + int i; + float coef, coef2; + coef = 6.3f*cutoff_Hz/Fs; + coef2 = 1-coef; + if (channels==2) + { + float m0, m2; + m0 = hp_mem[0]; + m2 = hp_mem[2]; + for (i=0;i<len;i++) + { + opus_val32 x0, x1, out0, out1; + x0 = in[2*i+0]; + x1 = in[2*i+1]; + out0 = x0-m0; + out1 = x1-m2; + m0 = coef*x0 + VERY_SMALL + coef2*m0; + m2 = coef*x1 + VERY_SMALL + coef2*m2; + out[2*i+0] = out0; + out[2*i+1] = out1; + } + hp_mem[0] = m0; + hp_mem[2] = m2; + } else { + float m0; + m0 = hp_mem[0]; + for (i=0;i<len;i++) + { + opus_val32 x, y; + x = in[i]; + y = x-m0; + m0 = coef*x + VERY_SMALL + coef2*m0; + out[i] = y; + } + hp_mem[0] = m0; + } +} +#endif + +static void stereo_fade(const opus_val16 *in, opus_val16 *out, opus_val16 g1, opus_val16 g2, + int overlap48, int frame_size, int channels, const opus_val16 *window, opus_int32 Fs) +{ + int i; + int overlap; + int inc; + inc = 48000/Fs; + overlap=overlap48/inc; + g1 = Q15ONE-g1; + g2 = Q15ONE-g2; + for (i=0;i<overlap;i++) + { + opus_val32 diff; + opus_val16 g, w; + w = MULT16_16_Q15(window[i*inc], window[i*inc]); + g = SHR32(MAC16_16(MULT16_16(w,g2), + Q15ONE-w, g1), 15); + diff = EXTRACT16(HALF32((opus_val32)in[i*channels] - (opus_val32)in[i*channels+1])); + diff = MULT16_16_Q15(g, diff); + out[i*channels] = out[i*channels] - diff; + out[i*channels+1] = out[i*channels+1] + diff; + } + for (;i<frame_size;i++) + { + opus_val32 diff; + diff = EXTRACT16(HALF32((opus_val32)in[i*channels] - (opus_val32)in[i*channels+1])); + diff = MULT16_16_Q15(g2, diff); + out[i*channels] = out[i*channels] - diff; + out[i*channels+1] = out[i*channels+1] + diff; + } +} + +static void gain_fade(const opus_val16 *in, opus_val16 *out, opus_val16 g1, opus_val16 g2, + int overlap48, int frame_size, int channels, const opus_val16 *window, opus_int32 Fs) +{ + int i; + int inc; + int overlap; + int c; + inc = 48000/Fs; + overlap=overlap48/inc; + if (channels==1) + { + for (i=0;i<overlap;i++) + { + opus_val16 g, w; + w = MULT16_16_Q15(window[i*inc], window[i*inc]); + g = SHR32(MAC16_16(MULT16_16(w,g2), + Q15ONE-w, g1), 15); + out[i] = MULT16_16_Q15(g, in[i]); + } + } else { + for (i=0;i<overlap;i++) + { + opus_val16 g, w; + w = MULT16_16_Q15(window[i*inc], window[i*inc]); + g = SHR32(MAC16_16(MULT16_16(w,g2), + Q15ONE-w, g1), 15); + out[i*2] = MULT16_16_Q15(g, in[i*2]); + out[i*2+1] = MULT16_16_Q15(g, in[i*2+1]); + } + } + c=0;do { + for (i=overlap;i<frame_size;i++) + { + out[i*channels+c] = MULT16_16_Q15(g2, in[i*channels+c]); + } + } + while (++c<channels); +} + +OpusEncoder *opus_encoder_create(opus_int32 Fs, int channels, int application, int *error) +{ + int ret; + OpusEncoder *st; + if((Fs!=48000&&Fs!=24000&&Fs!=16000&&Fs!=12000&&Fs!=8000)||(channels!=1&&channels!=2)|| + (application != OPUS_APPLICATION_VOIP && application != OPUS_APPLICATION_AUDIO + && application != OPUS_APPLICATION_RESTRICTED_LOWDELAY)) + { + if (error) + *error = OPUS_BAD_ARG; + return NULL; + } + st = (OpusEncoder *)opus_alloc(opus_encoder_get_size(channels)); + if (st == NULL) + { + if (error) + *error = OPUS_ALLOC_FAIL; + return NULL; + } + ret = opus_encoder_init(st, Fs, channels, application); + if (error) + *error = ret; + if (ret != OPUS_OK) + { + opus_free(st); + st = NULL; + } + return st; +} + +static opus_int32 user_bitrate_to_bitrate(OpusEncoder *st, int frame_size, int max_data_bytes) +{ + if(!frame_size)frame_size=st->Fs/400; + if (st->user_bitrate_bps==OPUS_AUTO) + return 60*st->Fs/frame_size + st->Fs*st->channels; + else if (st->user_bitrate_bps==OPUS_BITRATE_MAX) + return max_data_bytes*8*st->Fs/frame_size; + else + return st->user_bitrate_bps; +} + +#ifndef DISABLE_FLOAT_API +#ifdef FIXED_POINT +#define PCM2VAL(x) FLOAT2INT16(x) +#else +#define PCM2VAL(x) SCALEIN(x) +#endif + +void downmix_float(const void *_x, opus_val32 *y, int subframe, int offset, int c1, int c2, int C) +{ + const float *x; + int j; + + x = (const float *)_x; + for (j=0;j<subframe;j++) + y[j] = PCM2VAL(x[(j+offset)*C+c1]); + if (c2>-1) + { + for (j=0;j<subframe;j++) + y[j] += PCM2VAL(x[(j+offset)*C+c2]); + } else if (c2==-2) + { + int c; + for (c=1;c<C;c++) + { + for (j=0;j<subframe;j++) + y[j] += PCM2VAL(x[(j+offset)*C+c]); + } + } +} +#endif + +void downmix_int(const void *_x, opus_val32 *y, int subframe, int offset, int c1, int c2, int C) +{ + const opus_int16 *x; + int j; + + x = (const opus_int16 *)_x; + for (j=0;j<subframe;j++) + y[j] = x[(j+offset)*C+c1]; + if (c2>-1) + { + for (j=0;j<subframe;j++) + y[j] += x[(j+offset)*C+c2]; + } else if (c2==-2) + { + int c; + for (c=1;c<C;c++) + { + for (j=0;j<subframe;j++) + y[j] += x[(j+offset)*C+c]; + } + } +} + +opus_int32 frame_size_select(opus_int32 frame_size, int variable_duration, opus_int32 Fs) +{ + int new_size; + if (frame_size<Fs/400) + return -1; + if (variable_duration == OPUS_FRAMESIZE_ARG) + new_size = frame_size; + else if (variable_duration >= OPUS_FRAMESIZE_2_5_MS && variable_duration <= OPUS_FRAMESIZE_120_MS) + { + if (variable_duration <= OPUS_FRAMESIZE_40_MS) + new_size = (Fs/400)<<(variable_duration-OPUS_FRAMESIZE_2_5_MS); + else + new_size = (variable_duration-OPUS_FRAMESIZE_2_5_MS-2)*Fs/50; + } + else + return -1; + if (new_size>frame_size) + return -1; + if (400*new_size!=Fs && 200*new_size!=Fs && 100*new_size!=Fs && + 50*new_size!=Fs && 25*new_size!=Fs && 50*new_size!=3*Fs && + 50*new_size!=4*Fs && 50*new_size!=5*Fs && 50*new_size!=6*Fs) + return -1; + return new_size; +} + +opus_val16 compute_stereo_width(const opus_val16 *pcm, int frame_size, opus_int32 Fs, StereoWidthState *mem) +{ + opus_val32 xx, xy, yy; + opus_val16 sqrt_xx, sqrt_yy; + opus_val16 qrrt_xx, qrrt_yy; + int frame_rate; + int i; + opus_val16 short_alpha; + + frame_rate = Fs/frame_size; + short_alpha = Q15ONE - MULT16_16(25, Q15ONE)/IMAX(50,frame_rate); + xx=xy=yy=0; + /* Unroll by 4. The frame size is always a multiple of 4 *except* for + 2.5 ms frames at 12 kHz. Since this setting is very rare (and very + stupid), we just discard the last two samples. */ + for (i=0;i<frame_size-3;i+=4) + { + opus_val32 pxx=0; + opus_val32 pxy=0; + opus_val32 pyy=0; + opus_val16 x, y; + x = pcm[2*i]; + y = pcm[2*i+1]; + pxx = SHR32(MULT16_16(x,x),2); + pxy = SHR32(MULT16_16(x,y),2); + pyy = SHR32(MULT16_16(y,y),2); + x = pcm[2*i+2]; + y = pcm[2*i+3]; + pxx += SHR32(MULT16_16(x,x),2); + pxy += SHR32(MULT16_16(x,y),2); + pyy += SHR32(MULT16_16(y,y),2); + x = pcm[2*i+4]; + y = pcm[2*i+5]; + pxx += SHR32(MULT16_16(x,x),2); + pxy += SHR32(MULT16_16(x,y),2); + pyy += SHR32(MULT16_16(y,y),2); + x = pcm[2*i+6]; + y = pcm[2*i+7]; + pxx += SHR32(MULT16_16(x,x),2); + pxy += SHR32(MULT16_16(x,y),2); + pyy += SHR32(MULT16_16(y,y),2); + + xx += SHR32(pxx, 10); + xy += SHR32(pxy, 10); + yy += SHR32(pyy, 10); + } +#ifndef FIXED_POINT + if (!(xx < 1e9f) || celt_isnan(xx) || !(yy < 1e9f) || celt_isnan(yy)) + { + xy = xx = yy = 0; + } +#endif + mem->XX += MULT16_32_Q15(short_alpha, xx-mem->XX); + mem->XY += MULT16_32_Q15(short_alpha, xy-mem->XY); + mem->YY += MULT16_32_Q15(short_alpha, yy-mem->YY); + mem->XX = MAX32(0, mem->XX); + mem->XY = MAX32(0, mem->XY); + mem->YY = MAX32(0, mem->YY); + if (MAX32(mem->XX, mem->YY)>QCONST16(8e-4f, 18)) + { + opus_val16 corr; + opus_val16 ldiff; + opus_val16 width; + sqrt_xx = celt_sqrt(mem->XX); + sqrt_yy = celt_sqrt(mem->YY); + qrrt_xx = celt_sqrt(sqrt_xx); + qrrt_yy = celt_sqrt(sqrt_yy); + /* Inter-channel correlation */ + mem->XY = MIN32(mem->XY, sqrt_xx*sqrt_yy); + corr = SHR32(frac_div32(mem->XY,EPSILON+MULT16_16(sqrt_xx,sqrt_yy)),16); + /* Approximate loudness difference */ + ldiff = MULT16_16(Q15ONE, ABS16(qrrt_xx-qrrt_yy))/(EPSILON+qrrt_xx+qrrt_yy); + width = MULT16_16_Q15(celt_sqrt(QCONST32(1.f,30)-MULT16_16(corr,corr)), ldiff); + /* Smoothing over one second */ + mem->smoothed_width += (width-mem->smoothed_width)/frame_rate; + /* Peak follower */ + mem->max_follower = MAX16(mem->max_follower-QCONST16(.02f,15)/frame_rate, mem->smoothed_width); + } + /*printf("%f %f %f %f %f ", corr/(float)Q15ONE, ldiff/(float)Q15ONE, width/(float)Q15ONE, mem->smoothed_width/(float)Q15ONE, mem->max_follower/(float)Q15ONE);*/ + return EXTRACT16(MIN32(Q15ONE, MULT16_16(20, mem->max_follower))); +} + +static int decide_fec(int useInBandFEC, int PacketLoss_perc, int last_fec, int mode, int *bandwidth, opus_int32 rate) +{ + int orig_bandwidth; + if (!useInBandFEC || PacketLoss_perc == 0 || mode == MODE_CELT_ONLY) + return 0; + orig_bandwidth = *bandwidth; + for (;;) + { + opus_int32 hysteresis; + opus_int32 LBRR_rate_thres_bps; + /* Compute threshold for using FEC at the current bandwidth setting */ + LBRR_rate_thres_bps = fec_thresholds[2*(*bandwidth - OPUS_BANDWIDTH_NARROWBAND)]; + hysteresis = fec_thresholds[2*(*bandwidth - OPUS_BANDWIDTH_NARROWBAND) + 1]; + if (last_fec == 1) LBRR_rate_thres_bps -= hysteresis; + if (last_fec == 0) LBRR_rate_thres_bps += hysteresis; + LBRR_rate_thres_bps = silk_SMULWB( silk_MUL( LBRR_rate_thres_bps, + 125 - silk_min( PacketLoss_perc, 25 ) ), SILK_FIX_CONST( 0.01, 16 ) ); + /* If loss <= 5%, we look at whether we have enough rate to enable FEC. + If loss > 5%, we decrease the bandwidth until we can enable FEC. */ + if (rate > LBRR_rate_thres_bps) + return 1; + else if (PacketLoss_perc <= 5) + return 0; + else if (*bandwidth > OPUS_BANDWIDTH_NARROWBAND) + (*bandwidth)--; + else + break; + } + /* Couldn't find any bandwidth to enable FEC, keep original bandwidth. */ + *bandwidth = orig_bandwidth; + return 0; +} + +static int compute_silk_rate_for_hybrid(int rate, int bandwidth, int frame20ms, int vbr, int fec, int channels) { + int entry; + int i; + int N; + int silk_rate; + static int rate_table[][5] = { + /* |total| |-------- SILK------------| + |-- No FEC -| |--- FEC ---| + 10ms 20ms 10ms 20ms */ + { 0, 0, 0, 0, 0}, + {12000, 10000, 10000, 11000, 11000}, + {16000, 13500, 13500, 15000, 15000}, + {20000, 16000, 16000, 18000, 18000}, + {24000, 18000, 18000, 21000, 21000}, + {32000, 22000, 22000, 28000, 28000}, + {64000, 38000, 38000, 50000, 50000} + }; + /* Do the allocation per-channel. */ + rate /= channels; + entry = 1 + frame20ms + 2*fec; + N = sizeof(rate_table)/sizeof(rate_table[0]); + for (i=1;i<N;i++) + { + if (rate_table[i][0] > rate) break; + } + if (i == N) + { + silk_rate = rate_table[i-1][entry]; + /* For now, just give 50% of the extra bits to SILK. */ + silk_rate += (rate-rate_table[i-1][0])/2; + } else { + opus_int32 lo, hi, x0, x1; + lo = rate_table[i-1][entry]; + hi = rate_table[i][entry]; + x0 = rate_table[i-1][0]; + x1 = rate_table[i][0]; + silk_rate = (lo*(x1-rate) + hi*(rate-x0))/(x1-x0); + } + if (!vbr) + { + /* Tiny boost to SILK for CBR. We should probably tune this better. */ + silk_rate += 100; + } + if (bandwidth==OPUS_BANDWIDTH_SUPERWIDEBAND) + silk_rate += 300; + silk_rate *= channels; + /* Small adjustment for stereo (calibrated for 32 kb/s, haven't tried other bitrates). */ + if (channels == 2 && rate >= 12000) + silk_rate -= 1000; + return silk_rate; +} + +/* Returns the equivalent bitrate corresponding to 20 ms frames, + complexity 10 VBR operation. */ +static opus_int32 compute_equiv_rate(opus_int32 bitrate, int channels, + int frame_rate, int vbr, int mode, int complexity, int loss) +{ + opus_int32 equiv; + equiv = bitrate; + /* Take into account overhead from smaller frames. */ + if (frame_rate > 50) + equiv -= (40*channels+20)*(frame_rate - 50); + /* CBR is about a 8% penalty for both SILK and CELT. */ + if (!vbr) + equiv -= equiv/12; + /* Complexity makes about 10% difference (from 0 to 10) in general. */ + equiv = equiv * (90+complexity)/100; + if (mode == MODE_SILK_ONLY || mode == MODE_HYBRID) + { + /* SILK complexity 0-1 uses the non-delayed-decision NSQ, which + costs about 20%. */ + if (complexity<2) + equiv = equiv*4/5; + equiv -= equiv*loss/(6*loss + 10); + } else if (mode == MODE_CELT_ONLY) { + /* CELT complexity 0-4 doesn't have the pitch filter, which costs + about 10%. */ + if (complexity<5) + equiv = equiv*9/10; + } else { + /* Mode not known yet */ + /* Half the SILK loss*/ + equiv -= equiv*loss/(12*loss + 20); + } + return equiv; +} + +#ifndef DISABLE_FLOAT_API + +int is_digital_silence(const opus_val16* pcm, int frame_size, int channels, int lsb_depth) +{ + int silence = 0; + opus_val32 sample_max = 0; +#ifdef MLP_TRAINING + return 0; +#endif + sample_max = celt_maxabs16(pcm, frame_size*channels); + +#ifdef FIXED_POINT + silence = (sample_max == 0); + (void)lsb_depth; +#else + silence = (sample_max <= (opus_val16) 1 / (1 << lsb_depth)); +#endif + + return silence; +} + +#ifdef FIXED_POINT +static opus_val32 compute_frame_energy(const opus_val16 *pcm, int frame_size, int channels, int arch) +{ + int i; + opus_val32 sample_max; + int max_shift; + int shift; + opus_val32 energy = 0; + int len = frame_size*channels; + (void)arch; + /* Max amplitude in the signal */ + sample_max = celt_maxabs16(pcm, len); + + /* Compute the right shift required in the MAC to avoid an overflow */ + max_shift = celt_ilog2(len); + shift = IMAX(0, (celt_ilog2(sample_max) << 1) + max_shift - 28); + + /* Compute the energy */ + for (i=0; i<len; i++) + energy += SHR32(MULT16_16(pcm[i], pcm[i]), shift); + + /* Normalize energy by the frame size and left-shift back to the original position */ + energy /= len; + energy = SHL32(energy, shift); + + return energy; +} +#else +static opus_val32 compute_frame_energy(const opus_val16 *pcm, int frame_size, int channels, int arch) +{ + int len = frame_size*channels; + return celt_inner_prod(pcm, pcm, len, arch)/len; +} +#endif + +/* Decides if DTX should be turned on (=1) or off (=0) */ +static int decide_dtx_mode(opus_int activity, /* indicates if this frame contains speech/music */ + int *nb_no_activity_ms_Q1, /* number of consecutive milliseconds with no activity, in Q1 */ + int frame_size_ms_Q1 /* number of miliseconds in this update, in Q1 */ + ) + +{ + if (!activity) + { + /* The number of consecutive DTX frames should be within the allowed bounds. + Note that the allowed bound is defined in the SILK headers and assumes 20 ms + frames. As this function can be called with any frame length, a conversion to + milliseconds is done before the comparisons. */ + (*nb_no_activity_ms_Q1) += frame_size_ms_Q1; + if (*nb_no_activity_ms_Q1 > NB_SPEECH_FRAMES_BEFORE_DTX*20*2) + { + if (*nb_no_activity_ms_Q1 <= (NB_SPEECH_FRAMES_BEFORE_DTX + MAX_CONSECUTIVE_DTX)*20*2) + /* Valid frame for DTX! */ + return 1; + else + (*nb_no_activity_ms_Q1) = NB_SPEECH_FRAMES_BEFORE_DTX*20*2; + } + } else + (*nb_no_activity_ms_Q1) = 0; + + return 0; +} + +#endif + +static opus_int32 encode_multiframe_packet(OpusEncoder *st, + const opus_val16 *pcm, + int nb_frames, + int frame_size, + unsigned char *data, + opus_int32 out_data_bytes, + int to_celt, + int lsb_depth, + int float_api) +{ + int i; + int ret = 0; + VARDECL(unsigned char, tmp_data); + int bak_mode, bak_bandwidth, bak_channels, bak_to_mono; + VARDECL(OpusRepacketizer, rp); + int max_header_bytes; + opus_int32 bytes_per_frame; + opus_int32 cbr_bytes; + opus_int32 repacketize_len; + int tmp_len; + ALLOC_STACK; + + /* Worst cases: + * 2 frames: Code 2 with different compressed sizes + * >2 frames: Code 3 VBR */ + max_header_bytes = nb_frames == 2 ? 3 : (2+(nb_frames-1)*2); + + if (st->use_vbr || st->user_bitrate_bps==OPUS_BITRATE_MAX) + repacketize_len = out_data_bytes; + else { + cbr_bytes = 3*st->bitrate_bps/(3*8*st->Fs/(frame_size*nb_frames)); + repacketize_len = IMIN(cbr_bytes, out_data_bytes); + } + bytes_per_frame = IMIN(1276, 1+(repacketize_len-max_header_bytes)/nb_frames); + + ALLOC(tmp_data, nb_frames*bytes_per_frame, unsigned char); + ALLOC(rp, 1, OpusRepacketizer); + opus_repacketizer_init(rp); + + bak_mode = st->user_forced_mode; + bak_bandwidth = st->user_bandwidth; + bak_channels = st->force_channels; + + st->user_forced_mode = st->mode; + st->user_bandwidth = st->bandwidth; + st->force_channels = st->stream_channels; + + bak_to_mono = st->silk_mode.toMono; + if (bak_to_mono) + st->force_channels = 1; + else + st->prev_channels = st->stream_channels; + + for (i=0;i<nb_frames;i++) + { + st->silk_mode.toMono = 0; + st->nonfinal_frame = i<(nb_frames-1); + + /* When switching from SILK/Hybrid to CELT, only ask for a switch at the last frame */ + if (to_celt && i==nb_frames-1) + st->user_forced_mode = MODE_CELT_ONLY; + + tmp_len = opus_encode_native(st, pcm+i*(st->channels*frame_size), frame_size, + tmp_data+i*bytes_per_frame, bytes_per_frame, lsb_depth, NULL, 0, 0, 0, 0, + NULL, float_api); + + if (tmp_len<0) + { + RESTORE_STACK; + return OPUS_INTERNAL_ERROR; + } + + ret = opus_repacketizer_cat(rp, tmp_data+i*bytes_per_frame, tmp_len); + + if (ret<0) + { + RESTORE_STACK; + return OPUS_INTERNAL_ERROR; + } + } + + ret = opus_repacketizer_out_range_impl(rp, 0, nb_frames, data, repacketize_len, 0, !st->use_vbr); + + if (ret<0) + { + RESTORE_STACK; + return OPUS_INTERNAL_ERROR; + } + + /* Discard configs that were forced locally for the purpose of repacketization */ + st->user_forced_mode = bak_mode; + st->user_bandwidth = bak_bandwidth; + st->force_channels = bak_channels; + st->silk_mode.toMono = bak_to_mono; + + RESTORE_STACK; + return ret; +} + +static int compute_redundancy_bytes(opus_int32 max_data_bytes, opus_int32 bitrate_bps, int frame_rate, int channels) +{ + int redundancy_bytes_cap; + int redundancy_bytes; + opus_int32 redundancy_rate; + int base_bits; + opus_int32 available_bits; + base_bits = (40*channels+20); + + /* Equivalent rate for 5 ms frames. */ + redundancy_rate = bitrate_bps + base_bits*(200 - frame_rate); + /* For VBR, further increase the bitrate if we can afford it. It's pretty short + and we'll avoid artefacts. */ + redundancy_rate = 3*redundancy_rate/2; + redundancy_bytes = redundancy_rate/1600; + + /* Compute the max rate we can use given CBR or VBR with cap. */ + available_bits = max_data_bytes*8 - 2*base_bits; + redundancy_bytes_cap = (available_bits*240/(240+48000/frame_rate) + base_bits)/8; + redundancy_bytes = IMIN(redundancy_bytes, redundancy_bytes_cap); + /* It we can't get enough bits for redundancy to be worth it, rely on the decoder PLC. */ + if (redundancy_bytes > 4 + 8*channels) + redundancy_bytes = IMIN(257, redundancy_bytes); + else + redundancy_bytes = 0; + return redundancy_bytes; +} + +opus_int32 opus_encode_native(OpusEncoder *st, const opus_val16 *pcm, int frame_size, + unsigned char *data, opus_int32 out_data_bytes, int lsb_depth, + const void *analysis_pcm, opus_int32 analysis_size, int c1, int c2, + int analysis_channels, downmix_func downmix, int float_api) +{ + void *silk_enc; + CELTEncoder *celt_enc; + int i; + int ret=0; + opus_int32 nBytes; + ec_enc enc; + int bytes_target; + int prefill=0; + int start_band = 0; + int redundancy = 0; + int redundancy_bytes = 0; /* Number of bytes to use for redundancy frame */ + int celt_to_silk = 0; + VARDECL(opus_val16, pcm_buf); + int nb_compr_bytes; + int to_celt = 0; + opus_uint32 redundant_rng = 0; + int cutoff_Hz, hp_freq_smth1; + int voice_est; /* Probability of voice in Q7 */ + opus_int32 equiv_rate; + int delay_compensation; + int frame_rate; + opus_int32 max_rate; /* Max bitrate we're allowed to use */ + int curr_bandwidth; + opus_val16 HB_gain; + opus_int32 max_data_bytes; /* Max number of bytes we're allowed to use */ + int total_buffer; + opus_val16 stereo_width; + const CELTMode *celt_mode; +#ifndef DISABLE_FLOAT_API + AnalysisInfo analysis_info; + int analysis_read_pos_bak=-1; + int analysis_read_subframe_bak=-1; + int is_silence = 0; +#endif + opus_int activity = VAD_NO_DECISION; + + VARDECL(opus_val16, tmp_prefill); + + ALLOC_STACK; + + max_data_bytes = IMIN(1276, out_data_bytes); + + st->rangeFinal = 0; + if (frame_size <= 0 || max_data_bytes <= 0) + { + RESTORE_STACK; + return OPUS_BAD_ARG; + } + + /* Cannot encode 100 ms in 1 byte */ + if (max_data_bytes==1 && st->Fs==(frame_size*10)) + { + RESTORE_STACK; + return OPUS_BUFFER_TOO_SMALL; + } + + silk_enc = (char*)st+st->silk_enc_offset; + celt_enc = (CELTEncoder*)((char*)st+st->celt_enc_offset); + if (st->application == OPUS_APPLICATION_RESTRICTED_LOWDELAY) + delay_compensation = 0; + else + delay_compensation = st->delay_compensation; + + lsb_depth = IMIN(lsb_depth, st->lsb_depth); + + celt_encoder_ctl(celt_enc, CELT_GET_MODE(&celt_mode)); +#ifndef DISABLE_FLOAT_API + analysis_info.valid = 0; +#ifdef FIXED_POINT + if (st->silk_mode.complexity >= 10 && st->Fs>=16000) +#else + if (st->silk_mode.complexity >= 7 && st->Fs>=16000) +#endif + { + is_silence = is_digital_silence(pcm, frame_size, st->channels, lsb_depth); + analysis_read_pos_bak = st->analysis.read_pos; + analysis_read_subframe_bak = st->analysis.read_subframe; + run_analysis(&st->analysis, celt_mode, analysis_pcm, analysis_size, frame_size, + c1, c2, analysis_channels, st->Fs, + lsb_depth, downmix, &analysis_info); + + /* Track the peak signal energy */ + if (!is_silence && analysis_info.activity_probability > DTX_ACTIVITY_THRESHOLD) + st->peak_signal_energy = MAX32(MULT16_32_Q15(QCONST16(0.999f, 15), st->peak_signal_energy), + compute_frame_energy(pcm, frame_size, st->channels, st->arch)); + } else if (st->analysis.initialized) { + tonality_analysis_reset(&st->analysis); + } +#else + (void)analysis_pcm; + (void)analysis_size; + (void)c1; + (void)c2; + (void)analysis_channels; + (void)downmix; +#endif + +#ifndef DISABLE_FLOAT_API + /* Reset voice_ratio if this frame is not silent or if analysis is disabled. + * Otherwise, preserve voice_ratio from the last non-silent frame */ + if (!is_silence) + st->voice_ratio = -1; + + if (is_silence) + { + activity = !is_silence; + } else if (analysis_info.valid) + { + activity = analysis_info.activity_probability >= DTX_ACTIVITY_THRESHOLD; + if (!activity) + { + /* Mark as active if this noise frame is sufficiently loud */ + opus_val32 noise_energy = compute_frame_energy(pcm, frame_size, st->channels, st->arch); + activity = st->peak_signal_energy < (PSEUDO_SNR_THRESHOLD * noise_energy); + } + } + + st->detected_bandwidth = 0; + if (analysis_info.valid) + { + int analysis_bandwidth; + if (st->signal_type == OPUS_AUTO) + { + float prob; + if (st->prev_mode == 0) + prob = analysis_info.music_prob; + else if (st->prev_mode == MODE_CELT_ONLY) + prob = analysis_info.music_prob_max; + else + prob = analysis_info.music_prob_min; + st->voice_ratio = (int)floor(.5+100*(1-prob)); + } + + analysis_bandwidth = analysis_info.bandwidth; + if (analysis_bandwidth<=12) + st->detected_bandwidth = OPUS_BANDWIDTH_NARROWBAND; + else if (analysis_bandwidth<=14) + st->detected_bandwidth = OPUS_BANDWIDTH_MEDIUMBAND; + else if (analysis_bandwidth<=16) + st->detected_bandwidth = OPUS_BANDWIDTH_WIDEBAND; + else if (analysis_bandwidth<=18) + st->detected_bandwidth = OPUS_BANDWIDTH_SUPERWIDEBAND; + else + st->detected_bandwidth = OPUS_BANDWIDTH_FULLBAND; + } +#else + st->voice_ratio = -1; +#endif + + if (st->channels==2 && st->force_channels!=1) + stereo_width = compute_stereo_width(pcm, frame_size, st->Fs, &st->width_mem); + else + stereo_width = 0; + total_buffer = delay_compensation; + st->bitrate_bps = user_bitrate_to_bitrate(st, frame_size, max_data_bytes); + + frame_rate = st->Fs/frame_size; + if (!st->use_vbr) + { + int cbrBytes; + /* Multiply by 12 to make sure the division is exact. */ + int frame_rate12 = 12*st->Fs/frame_size; + /* We need to make sure that "int" values always fit in 16 bits. */ + cbrBytes = IMIN( (12*st->bitrate_bps/8 + frame_rate12/2)/frame_rate12, max_data_bytes); + st->bitrate_bps = cbrBytes*(opus_int32)frame_rate12*8/12; + /* Make sure we provide at least one byte to avoid failing. */ + max_data_bytes = IMAX(1, cbrBytes); + } + if (max_data_bytes<3 || st->bitrate_bps < 3*frame_rate*8 + || (frame_rate<50 && (max_data_bytes*frame_rate<300 || st->bitrate_bps < 2400))) + { + /*If the space is too low to do something useful, emit 'PLC' frames.*/ + int tocmode = st->mode; + int bw = st->bandwidth == 0 ? OPUS_BANDWIDTH_NARROWBAND : st->bandwidth; + int packet_code = 0; + int num_multiframes = 0; + + if (tocmode==0) + tocmode = MODE_SILK_ONLY; + if (frame_rate>100) + tocmode = MODE_CELT_ONLY; + /* 40 ms -> 2 x 20 ms if in CELT_ONLY or HYBRID mode */ + if (frame_rate==25 && tocmode!=MODE_SILK_ONLY) + { + frame_rate = 50; + packet_code = 1; + } + + /* >= 60 ms frames */ + if (frame_rate<=16) + { + /* 1 x 60 ms, 2 x 40 ms, 2 x 60 ms */ + if (out_data_bytes==1 || (tocmode==MODE_SILK_ONLY && frame_rate!=10)) + { + tocmode = MODE_SILK_ONLY; + + packet_code = frame_rate <= 12; + frame_rate = frame_rate == 12 ? 25 : 16; + } + else + { + num_multiframes = 50/frame_rate; + frame_rate = 50; + packet_code = 3; + } + } + + if(tocmode==MODE_SILK_ONLY&&bw>OPUS_BANDWIDTH_WIDEBAND) + bw=OPUS_BANDWIDTH_WIDEBAND; + else if (tocmode==MODE_CELT_ONLY&&bw==OPUS_BANDWIDTH_MEDIUMBAND) + bw=OPUS_BANDWIDTH_NARROWBAND; + else if (tocmode==MODE_HYBRID&&bw<=OPUS_BANDWIDTH_SUPERWIDEBAND) + bw=OPUS_BANDWIDTH_SUPERWIDEBAND; + + data[0] = gen_toc(tocmode, frame_rate, bw, st->stream_channels); + data[0] |= packet_code; + + ret = packet_code <= 1 ? 1 : 2; + + max_data_bytes = IMAX(max_data_bytes, ret); + + if (packet_code==3) + data[1] = num_multiframes; + + if (!st->use_vbr) + { + ret = opus_packet_pad(data, ret, max_data_bytes); + if (ret == OPUS_OK) + ret = max_data_bytes; + else + ret = OPUS_INTERNAL_ERROR; + } + RESTORE_STACK; + return ret; + } + max_rate = frame_rate*max_data_bytes*8; + + /* Equivalent 20-ms rate for mode/channel/bandwidth decisions */ + equiv_rate = compute_equiv_rate(st->bitrate_bps, st->channels, st->Fs/frame_size, + st->use_vbr, 0, st->silk_mode.complexity, st->silk_mode.packetLossPercentage); + + if (st->signal_type == OPUS_SIGNAL_VOICE) + voice_est = 127; + else if (st->signal_type == OPUS_SIGNAL_MUSIC) + voice_est = 0; + else if (st->voice_ratio >= 0) + { + voice_est = st->voice_ratio*327>>8; + /* For AUDIO, never be more than 90% confident of having speech */ + if (st->application == OPUS_APPLICATION_AUDIO) + voice_est = IMIN(voice_est, 115); + } else if (st->application == OPUS_APPLICATION_VOIP) + voice_est = 115; + else + voice_est = 48; + + if (st->force_channels!=OPUS_AUTO && st->channels == 2) + { + st->stream_channels = st->force_channels; + } else { +#ifdef FUZZING + (void)stereo_music_threshold; + (void)stereo_voice_threshold; + /* Random mono/stereo decision */ + if (st->channels == 2 && (rand()&0x1F)==0) + st->stream_channels = 3-st->stream_channels; +#else + /* Rate-dependent mono-stereo decision */ + if (st->channels == 2) + { + opus_int32 stereo_threshold; + stereo_threshold = stereo_music_threshold + ((voice_est*voice_est*(stereo_voice_threshold-stereo_music_threshold))>>14); + if (st->stream_channels == 2) + stereo_threshold -= 1000; + else + stereo_threshold += 1000; + st->stream_channels = (equiv_rate > stereo_threshold) ? 2 : 1; + } else { + st->stream_channels = st->channels; + } +#endif + } + /* Update equivalent rate for channels decision. */ + equiv_rate = compute_equiv_rate(st->bitrate_bps, st->stream_channels, st->Fs/frame_size, + st->use_vbr, 0, st->silk_mode.complexity, st->silk_mode.packetLossPercentage); + + /* Allow SILK DTX if DTX is enabled but the generalized DTX cannot be used, + e.g. because of the complexity setting or sample rate. */ +#ifndef DISABLE_FLOAT_API + st->silk_mode.useDTX = st->use_dtx && !(analysis_info.valid || is_silence); +#else + st->silk_mode.useDTX = st->use_dtx; +#endif + + /* Mode selection depending on application and signal type */ + if (st->application == OPUS_APPLICATION_RESTRICTED_LOWDELAY) + { + st->mode = MODE_CELT_ONLY; + } else if (st->user_forced_mode == OPUS_AUTO) + { +#ifdef FUZZING + (void)stereo_width; + (void)mode_thresholds; + /* Random mode switching */ + if ((rand()&0xF)==0) + { + if ((rand()&0x1)==0) + st->mode = MODE_CELT_ONLY; + else + st->mode = MODE_SILK_ONLY; + } else { + if (st->prev_mode==MODE_CELT_ONLY) + st->mode = MODE_CELT_ONLY; + else + st->mode = MODE_SILK_ONLY; + } +#else + opus_int32 mode_voice, mode_music; + opus_int32 threshold; + + /* Interpolate based on stereo width */ + mode_voice = (opus_int32)(MULT16_32_Q15(Q15ONE-stereo_width,mode_thresholds[0][0]) + + MULT16_32_Q15(stereo_width,mode_thresholds[1][0])); + mode_music = (opus_int32)(MULT16_32_Q15(Q15ONE-stereo_width,mode_thresholds[1][1]) + + MULT16_32_Q15(stereo_width,mode_thresholds[1][1])); + /* Interpolate based on speech/music probability */ + threshold = mode_music + ((voice_est*voice_est*(mode_voice-mode_music))>>14); + /* Bias towards SILK for VoIP because of some useful features */ + if (st->application == OPUS_APPLICATION_VOIP) + threshold += 8000; + + /*printf("%f %d\n", stereo_width/(float)Q15ONE, threshold);*/ + /* Hysteresis */ + if (st->prev_mode == MODE_CELT_ONLY) + threshold -= 4000; + else if (st->prev_mode>0) + threshold += 4000; + + st->mode = (equiv_rate >= threshold) ? MODE_CELT_ONLY: MODE_SILK_ONLY; + + /* When FEC is enabled and there's enough packet loss, use SILK. + Unless the FEC is set to 2, in which case we don't switch to SILK if we're confident we have music. */ + if (st->silk_mode.useInBandFEC && st->silk_mode.packetLossPercentage > (128-voice_est)>>4 && (st->fec_config != 2 || voice_est > 25)) + st->mode = MODE_SILK_ONLY; + /* When encoding voice and DTX is enabled but the generalized DTX cannot be used, + use SILK in order to make use of its DTX. */ + if (st->silk_mode.useDTX && voice_est > 100) + st->mode = MODE_SILK_ONLY; +#endif + + /* If max_data_bytes represents less than 6 kb/s, switch to CELT-only mode */ + if (max_data_bytes < (frame_rate > 50 ? 9000 : 6000)*frame_size / (st->Fs * 8)) + st->mode = MODE_CELT_ONLY; + } else { + st->mode = st->user_forced_mode; + } + + /* Override the chosen mode to make sure we meet the requested frame size */ + if (st->mode != MODE_CELT_ONLY && frame_size < st->Fs/100) + st->mode = MODE_CELT_ONLY; + if (st->lfe) + st->mode = MODE_CELT_ONLY; + + if (st->prev_mode > 0 && + ((st->mode != MODE_CELT_ONLY && st->prev_mode == MODE_CELT_ONLY) || + (st->mode == MODE_CELT_ONLY && st->prev_mode != MODE_CELT_ONLY))) + { + redundancy = 1; + celt_to_silk = (st->mode != MODE_CELT_ONLY); + if (!celt_to_silk) + { + /* Switch to SILK/hybrid if frame size is 10 ms or more*/ + if (frame_size >= st->Fs/100) + { + st->mode = st->prev_mode; + to_celt = 1; + } else { + redundancy=0; + } + } + } + + /* When encoding multiframes, we can ask for a switch to CELT only in the last frame. This switch + * is processed above as the requested mode shouldn't interrupt stereo->mono transition. */ + if (st->stream_channels == 1 && st->prev_channels ==2 && st->silk_mode.toMono==0 + && st->mode != MODE_CELT_ONLY && st->prev_mode != MODE_CELT_ONLY) + { + /* Delay stereo->mono transition by two frames so that SILK can do a smooth downmix */ + st->silk_mode.toMono = 1; + st->stream_channels = 2; + } else { + st->silk_mode.toMono = 0; + } + + /* Update equivalent rate with mode decision. */ + equiv_rate = compute_equiv_rate(st->bitrate_bps, st->stream_channels, st->Fs/frame_size, + st->use_vbr, st->mode, st->silk_mode.complexity, st->silk_mode.packetLossPercentage); + + if (st->mode != MODE_CELT_ONLY && st->prev_mode == MODE_CELT_ONLY) + { + silk_EncControlStruct dummy; + silk_InitEncoder( silk_enc, st->arch, &dummy); + prefill=1; + } + + /* Automatic (rate-dependent) bandwidth selection */ + if (st->mode == MODE_CELT_ONLY || st->first || st->silk_mode.allowBandwidthSwitch) + { + const opus_int32 *voice_bandwidth_thresholds, *music_bandwidth_thresholds; + opus_int32 bandwidth_thresholds[8]; + int bandwidth = OPUS_BANDWIDTH_FULLBAND; + + if (st->channels==2 && st->force_channels!=1) + { + voice_bandwidth_thresholds = stereo_voice_bandwidth_thresholds; + music_bandwidth_thresholds = stereo_music_bandwidth_thresholds; + } else { + voice_bandwidth_thresholds = mono_voice_bandwidth_thresholds; + music_bandwidth_thresholds = mono_music_bandwidth_thresholds; + } + /* Interpolate bandwidth thresholds depending on voice estimation */ + for (i=0;i<8;i++) + { + bandwidth_thresholds[i] = music_bandwidth_thresholds[i] + + ((voice_est*voice_est*(voice_bandwidth_thresholds[i]-music_bandwidth_thresholds[i]))>>14); + } + do { + int threshold, hysteresis; + threshold = bandwidth_thresholds[2*(bandwidth-OPUS_BANDWIDTH_MEDIUMBAND)]; + hysteresis = bandwidth_thresholds[2*(bandwidth-OPUS_BANDWIDTH_MEDIUMBAND)+1]; + if (!st->first) + { + if (st->auto_bandwidth >= bandwidth) + threshold -= hysteresis; + else + threshold += hysteresis; + } + if (equiv_rate >= threshold) + break; + } while (--bandwidth>OPUS_BANDWIDTH_NARROWBAND); + /* We don't use mediumband anymore, except when explicitly requested or during + mode transitions. */ + if (bandwidth == OPUS_BANDWIDTH_MEDIUMBAND) + bandwidth = OPUS_BANDWIDTH_WIDEBAND; + st->bandwidth = st->auto_bandwidth = bandwidth; + /* Prevents any transition to SWB/FB until the SILK layer has fully + switched to WB mode and turned the variable LP filter off */ + if (!st->first && st->mode != MODE_CELT_ONLY && !st->silk_mode.inWBmodeWithoutVariableLP && st->bandwidth > OPUS_BANDWIDTH_WIDEBAND) + st->bandwidth = OPUS_BANDWIDTH_WIDEBAND; + } + + if (st->bandwidth>st->max_bandwidth) + st->bandwidth = st->max_bandwidth; + + if (st->user_bandwidth != OPUS_AUTO) + st->bandwidth = st->user_bandwidth; + + /* This prevents us from using hybrid at unsafe CBR/max rates */ + if (st->mode != MODE_CELT_ONLY && max_rate < 15000) + { + st->bandwidth = IMIN(st->bandwidth, OPUS_BANDWIDTH_WIDEBAND); + } + + /* Prevents Opus from wasting bits on frequencies that are above + the Nyquist rate of the input signal */ + if (st->Fs <= 24000 && st->bandwidth > OPUS_BANDWIDTH_SUPERWIDEBAND) + st->bandwidth = OPUS_BANDWIDTH_SUPERWIDEBAND; + if (st->Fs <= 16000 && st->bandwidth > OPUS_BANDWIDTH_WIDEBAND) + st->bandwidth = OPUS_BANDWIDTH_WIDEBAND; + if (st->Fs <= 12000 && st->bandwidth > OPUS_BANDWIDTH_MEDIUMBAND) + st->bandwidth = OPUS_BANDWIDTH_MEDIUMBAND; + if (st->Fs <= 8000 && st->bandwidth > OPUS_BANDWIDTH_NARROWBAND) + st->bandwidth = OPUS_BANDWIDTH_NARROWBAND; +#ifndef DISABLE_FLOAT_API + /* Use detected bandwidth to reduce the encoded bandwidth. */ + if (st->detected_bandwidth && st->user_bandwidth == OPUS_AUTO) + { + int min_detected_bandwidth; + /* Makes bandwidth detection more conservative just in case the detector + gets it wrong when we could have coded a high bandwidth transparently. + When operating in SILK/hybrid mode, we don't go below wideband to avoid + more complicated switches that require redundancy. */ + if (equiv_rate <= 18000*st->stream_channels && st->mode == MODE_CELT_ONLY) + min_detected_bandwidth = OPUS_BANDWIDTH_NARROWBAND; + else if (equiv_rate <= 24000*st->stream_channels && st->mode == MODE_CELT_ONLY) + min_detected_bandwidth = OPUS_BANDWIDTH_MEDIUMBAND; + else if (equiv_rate <= 30000*st->stream_channels) + min_detected_bandwidth = OPUS_BANDWIDTH_WIDEBAND; + else if (equiv_rate <= 44000*st->stream_channels) + min_detected_bandwidth = OPUS_BANDWIDTH_SUPERWIDEBAND; + else + min_detected_bandwidth = OPUS_BANDWIDTH_FULLBAND; + + st->detected_bandwidth = IMAX(st->detected_bandwidth, min_detected_bandwidth); + st->bandwidth = IMIN(st->bandwidth, st->detected_bandwidth); + } +#endif + st->silk_mode.LBRR_coded = decide_fec(st->silk_mode.useInBandFEC, st->silk_mode.packetLossPercentage, + st->silk_mode.LBRR_coded, st->mode, &st->bandwidth, equiv_rate); + celt_encoder_ctl(celt_enc, OPUS_SET_LSB_DEPTH(lsb_depth)); + + /* CELT mode doesn't support mediumband, use wideband instead */ + if (st->mode == MODE_CELT_ONLY && st->bandwidth == OPUS_BANDWIDTH_MEDIUMBAND) + st->bandwidth = OPUS_BANDWIDTH_WIDEBAND; + if (st->lfe) + st->bandwidth = OPUS_BANDWIDTH_NARROWBAND; + + curr_bandwidth = st->bandwidth; + + /* Chooses the appropriate mode for speech + *NEVER* switch to/from CELT-only mode here as this will invalidate some assumptions */ + if (st->mode == MODE_SILK_ONLY && curr_bandwidth > OPUS_BANDWIDTH_WIDEBAND) + st->mode = MODE_HYBRID; + if (st->mode == MODE_HYBRID && curr_bandwidth <= OPUS_BANDWIDTH_WIDEBAND) + st->mode = MODE_SILK_ONLY; + + /* Can't support higher than >60 ms frames, and >20 ms when in Hybrid or CELT-only modes */ + if ((frame_size > st->Fs/50 && (st->mode != MODE_SILK_ONLY)) || frame_size > 3*st->Fs/50) + { + int enc_frame_size; + int nb_frames; + + if (st->mode == MODE_SILK_ONLY) + { + if (frame_size == 2*st->Fs/25) /* 80 ms -> 2x 40 ms */ + enc_frame_size = st->Fs/25; + else if (frame_size == 3*st->Fs/25) /* 120 ms -> 2x 60 ms */ + enc_frame_size = 3*st->Fs/50; + else /* 100 ms -> 5x 20 ms */ + enc_frame_size = st->Fs/50; + } + else + enc_frame_size = st->Fs/50; + + nb_frames = frame_size/enc_frame_size; + +#ifndef DISABLE_FLOAT_API + if (analysis_read_pos_bak!= -1) + { + st->analysis.read_pos = analysis_read_pos_bak; + st->analysis.read_subframe = analysis_read_subframe_bak; + } +#endif + + ret = encode_multiframe_packet(st, pcm, nb_frames, enc_frame_size, data, + out_data_bytes, to_celt, lsb_depth, float_api); + + RESTORE_STACK; + return ret; + } + + /* For the first frame at a new SILK bandwidth */ + if (st->silk_bw_switch) + { + redundancy = 1; + celt_to_silk = 1; + st->silk_bw_switch = 0; + /* Do a prefill without reseting the sampling rate control. */ + prefill=2; + } + + /* If we decided to go with CELT, make sure redundancy is off, no matter what + we decided earlier. */ + if (st->mode == MODE_CELT_ONLY) + redundancy = 0; + + if (redundancy) + { + redundancy_bytes = compute_redundancy_bytes(max_data_bytes, st->bitrate_bps, frame_rate, st->stream_channels); + if (redundancy_bytes == 0) + redundancy = 0; + } + + /* printf("%d %d %d %d\n", st->bitrate_bps, st->stream_channels, st->mode, curr_bandwidth); */ + bytes_target = IMIN(max_data_bytes-redundancy_bytes, st->bitrate_bps * frame_size / (st->Fs * 8)) - 1; + + data += 1; + + ec_enc_init(&enc, data, max_data_bytes-1); + + ALLOC(pcm_buf, (total_buffer+frame_size)*st->channels, opus_val16); + OPUS_COPY(pcm_buf, &st->delay_buffer[(st->encoder_buffer-total_buffer)*st->channels], total_buffer*st->channels); + + if (st->mode == MODE_CELT_ONLY) + hp_freq_smth1 = silk_LSHIFT( silk_lin2log( VARIABLE_HP_MIN_CUTOFF_HZ ), 8 ); + else + hp_freq_smth1 = ((silk_encoder*)silk_enc)->state_Fxx[0].sCmn.variable_HP_smth1_Q15; + + st->variable_HP_smth2_Q15 = silk_SMLAWB( st->variable_HP_smth2_Q15, + hp_freq_smth1 - st->variable_HP_smth2_Q15, SILK_FIX_CONST( VARIABLE_HP_SMTH_COEF2, 16 ) ); + + /* convert from log scale to Hertz */ + cutoff_Hz = silk_log2lin( silk_RSHIFT( st->variable_HP_smth2_Q15, 8 ) ); + + if (st->application == OPUS_APPLICATION_VOIP) + { + hp_cutoff(pcm, cutoff_Hz, &pcm_buf[total_buffer*st->channels], st->hp_mem, frame_size, st->channels, st->Fs, st->arch); + } else { + dc_reject(pcm, 3, &pcm_buf[total_buffer*st->channels], st->hp_mem, frame_size, st->channels, st->Fs); + } +#ifndef FIXED_POINT + if (float_api) + { + opus_val32 sum; + sum = celt_inner_prod(&pcm_buf[total_buffer*st->channels], &pcm_buf[total_buffer*st->channels], frame_size*st->channels, st->arch); + /* This should filter out both NaNs and ridiculous signals that could + cause NaNs further down. */ + if (!(sum < 1e9f) || celt_isnan(sum)) + { + OPUS_CLEAR(&pcm_buf[total_buffer*st->channels], frame_size*st->channels); + st->hp_mem[0] = st->hp_mem[1] = st->hp_mem[2] = st->hp_mem[3] = 0; + } + } +#endif + + + /* SILK processing */ + HB_gain = Q15ONE; + if (st->mode != MODE_CELT_ONLY) + { + opus_int32 total_bitRate, celt_rate; +#ifdef FIXED_POINT + const opus_int16 *pcm_silk; +#else + VARDECL(opus_int16, pcm_silk); + ALLOC(pcm_silk, st->channels*frame_size, opus_int16); +#endif + + /* Distribute bits between SILK and CELT */ + total_bitRate = 8 * bytes_target * frame_rate; + if( st->mode == MODE_HYBRID ) { + /* Base rate for SILK */ + st->silk_mode.bitRate = compute_silk_rate_for_hybrid(total_bitRate, + curr_bandwidth, st->Fs == 50 * frame_size, st->use_vbr, st->silk_mode.LBRR_coded, + st->stream_channels); + if (!st->energy_masking) + { + /* Increasingly attenuate high band when it gets allocated fewer bits */ + celt_rate = total_bitRate - st->silk_mode.bitRate; + HB_gain = Q15ONE - SHR32(celt_exp2(-celt_rate * QCONST16(1.f/1024, 10)), 1); + } + } else { + /* SILK gets all bits */ + st->silk_mode.bitRate = total_bitRate; + } + + /* Surround masking for SILK */ + if (st->energy_masking && st->use_vbr && !st->lfe) + { + opus_val32 mask_sum=0; + opus_val16 masking_depth; + opus_int32 rate_offset; + int c; + int end = 17; + opus_int16 srate = 16000; + if (st->bandwidth == OPUS_BANDWIDTH_NARROWBAND) + { + end = 13; + srate = 8000; + } else if (st->bandwidth == OPUS_BANDWIDTH_MEDIUMBAND) + { + end = 15; + srate = 12000; + } + for (c=0;c<st->channels;c++) + { + for(i=0;i<end;i++) + { + opus_val16 mask; + mask = MAX16(MIN16(st->energy_masking[21*c+i], + QCONST16(.5f, DB_SHIFT)), -QCONST16(2.0f, DB_SHIFT)); + if (mask > 0) + mask = HALF16(mask); + mask_sum += mask; + } + } + /* Conservative rate reduction, we cut the masking in half */ + masking_depth = mask_sum / end*st->channels; + masking_depth += QCONST16(.2f, DB_SHIFT); + rate_offset = (opus_int32)PSHR32(MULT16_16(srate, masking_depth), DB_SHIFT); + rate_offset = MAX32(rate_offset, -2*st->silk_mode.bitRate/3); + /* Split the rate change between the SILK and CELT part for hybrid. */ + if (st->bandwidth==OPUS_BANDWIDTH_SUPERWIDEBAND || st->bandwidth==OPUS_BANDWIDTH_FULLBAND) + st->silk_mode.bitRate += 3*rate_offset/5; + else + st->silk_mode.bitRate += rate_offset; + } + + st->silk_mode.payloadSize_ms = 1000 * frame_size / st->Fs; + st->silk_mode.nChannelsAPI = st->channels; + st->silk_mode.nChannelsInternal = st->stream_channels; + if (curr_bandwidth == OPUS_BANDWIDTH_NARROWBAND) { + st->silk_mode.desiredInternalSampleRate = 8000; + } else if (curr_bandwidth == OPUS_BANDWIDTH_MEDIUMBAND) { + st->silk_mode.desiredInternalSampleRate = 12000; + } else { + celt_assert( st->mode == MODE_HYBRID || curr_bandwidth == OPUS_BANDWIDTH_WIDEBAND ); + st->silk_mode.desiredInternalSampleRate = 16000; + } + if( st->mode == MODE_HYBRID ) { + /* Don't allow bandwidth reduction at lowest bitrates in hybrid mode */ + st->silk_mode.minInternalSampleRate = 16000; + } else { + st->silk_mode.minInternalSampleRate = 8000; + } + + st->silk_mode.maxInternalSampleRate = 16000; + if (st->mode == MODE_SILK_ONLY) + { + opus_int32 effective_max_rate = max_rate; + if (frame_rate > 50) + effective_max_rate = effective_max_rate*2/3; + if (effective_max_rate < 8000) + { + st->silk_mode.maxInternalSampleRate = 12000; + st->silk_mode.desiredInternalSampleRate = IMIN(12000, st->silk_mode.desiredInternalSampleRate); + } + if (effective_max_rate < 7000) + { + st->silk_mode.maxInternalSampleRate = 8000; + st->silk_mode.desiredInternalSampleRate = IMIN(8000, st->silk_mode.desiredInternalSampleRate); + } + } + + st->silk_mode.useCBR = !st->use_vbr; + + /* Call SILK encoder for the low band */ + + /* Max bits for SILK, counting ToC, redundancy bytes, and optionally redundancy. */ + st->silk_mode.maxBits = (max_data_bytes-1)*8; + if (redundancy && redundancy_bytes >= 2) + { + /* Counting 1 bit for redundancy position and 20 bits for flag+size (only for hybrid). */ + st->silk_mode.maxBits -= redundancy_bytes*8 + 1; + if (st->mode == MODE_HYBRID) + st->silk_mode.maxBits -= 20; + } + if (st->silk_mode.useCBR) + { + if (st->mode == MODE_HYBRID) + { + st->silk_mode.maxBits = IMIN(st->silk_mode.maxBits, st->silk_mode.bitRate * frame_size / st->Fs); + } + } else { + /* Constrained VBR. */ + if (st->mode == MODE_HYBRID) + { + /* Compute SILK bitrate corresponding to the max total bits available */ + opus_int32 maxBitRate = compute_silk_rate_for_hybrid(st->silk_mode.maxBits*st->Fs / frame_size, + curr_bandwidth, st->Fs == 50 * frame_size, st->use_vbr, st->silk_mode.LBRR_coded, + st->stream_channels); + st->silk_mode.maxBits = maxBitRate * frame_size / st->Fs; + } + } + + if (prefill) + { + opus_int32 zero=0; + int prefill_offset; + /* Use a smooth onset for the SILK prefill to avoid the encoder trying to encode + a discontinuity. The exact location is what we need to avoid leaving any "gap" + in the audio when mixing with the redundant CELT frame. Here we can afford to + overwrite st->delay_buffer because the only thing that uses it before it gets + rewritten is tmp_prefill[] and even then only the part after the ramp really + gets used (rather than sent to the encoder and discarded) */ + prefill_offset = st->channels*(st->encoder_buffer-st->delay_compensation-st->Fs/400); + gain_fade(st->delay_buffer+prefill_offset, st->delay_buffer+prefill_offset, + 0, Q15ONE, celt_mode->overlap, st->Fs/400, st->channels, celt_mode->window, st->Fs); + OPUS_CLEAR(st->delay_buffer, prefill_offset); +#ifdef FIXED_POINT + pcm_silk = st->delay_buffer; +#else + for (i=0;i<st->encoder_buffer*st->channels;i++) + pcm_silk[i] = FLOAT2INT16(st->delay_buffer[i]); +#endif + silk_Encode( silk_enc, &st->silk_mode, pcm_silk, st->encoder_buffer, NULL, &zero, prefill, activity ); + /* Prevent a second switch in the real encode call. */ + st->silk_mode.opusCanSwitch = 0; + } + +#ifdef FIXED_POINT + pcm_silk = pcm_buf+total_buffer*st->channels; +#else + for (i=0;i<frame_size*st->channels;i++) + pcm_silk[i] = FLOAT2INT16(pcm_buf[total_buffer*st->channels + i]); +#endif + ret = silk_Encode( silk_enc, &st->silk_mode, pcm_silk, frame_size, &enc, &nBytes, 0, activity ); + if( ret ) { + /*fprintf (stderr, "SILK encode error: %d\n", ret);*/ + /* Handle error */ + RESTORE_STACK; + return OPUS_INTERNAL_ERROR; + } + + /* Extract SILK internal bandwidth for signaling in first byte */ + if( st->mode == MODE_SILK_ONLY ) { + if( st->silk_mode.internalSampleRate == 8000 ) { + curr_bandwidth = OPUS_BANDWIDTH_NARROWBAND; + } else if( st->silk_mode.internalSampleRate == 12000 ) { + curr_bandwidth = OPUS_BANDWIDTH_MEDIUMBAND; + } else if( st->silk_mode.internalSampleRate == 16000 ) { + curr_bandwidth = OPUS_BANDWIDTH_WIDEBAND; + } + } else { + celt_assert( st->silk_mode.internalSampleRate == 16000 ); + } + + st->silk_mode.opusCanSwitch = st->silk_mode.switchReady && !st->nonfinal_frame; + + if (nBytes==0) + { + st->rangeFinal = 0; + data[-1] = gen_toc(st->mode, st->Fs/frame_size, curr_bandwidth, st->stream_channels); + RESTORE_STACK; + return 1; + } + + /* FIXME: How do we allocate the redundancy for CBR? */ + if (st->silk_mode.opusCanSwitch) + { + redundancy_bytes = compute_redundancy_bytes(max_data_bytes, st->bitrate_bps, frame_rate, st->stream_channels); + redundancy = (redundancy_bytes != 0); + celt_to_silk = 0; + st->silk_bw_switch = 1; + } + } + + /* CELT processing */ + { + int endband=21; + + switch(curr_bandwidth) + { + case OPUS_BANDWIDTH_NARROWBAND: + endband = 13; + break; + case OPUS_BANDWIDTH_MEDIUMBAND: + case OPUS_BANDWIDTH_WIDEBAND: + endband = 17; + break; + case OPUS_BANDWIDTH_SUPERWIDEBAND: + endband = 19; + break; + case OPUS_BANDWIDTH_FULLBAND: + endband = 21; + break; + } + celt_encoder_ctl(celt_enc, CELT_SET_END_BAND(endband)); + celt_encoder_ctl(celt_enc, CELT_SET_CHANNELS(st->stream_channels)); + } + celt_encoder_ctl(celt_enc, OPUS_SET_BITRATE(OPUS_BITRATE_MAX)); + if (st->mode != MODE_SILK_ONLY) + { + opus_val32 celt_pred=2; + celt_encoder_ctl(celt_enc, OPUS_SET_VBR(0)); + /* We may still decide to disable prediction later */ + if (st->silk_mode.reducedDependency) + celt_pred = 0; + celt_encoder_ctl(celt_enc, CELT_SET_PREDICTION(celt_pred)); + + if (st->mode == MODE_HYBRID) + { + if( st->use_vbr ) { + celt_encoder_ctl(celt_enc, OPUS_SET_BITRATE(st->bitrate_bps-st->silk_mode.bitRate)); + celt_encoder_ctl(celt_enc, OPUS_SET_VBR_CONSTRAINT(0)); + } + } else { + if (st->use_vbr) + { + celt_encoder_ctl(celt_enc, OPUS_SET_VBR(1)); + celt_encoder_ctl(celt_enc, OPUS_SET_VBR_CONSTRAINT(st->vbr_constraint)); + celt_encoder_ctl(celt_enc, OPUS_SET_BITRATE(st->bitrate_bps)); + } + } + } + + ALLOC(tmp_prefill, st->channels*st->Fs/400, opus_val16); + if (st->mode != MODE_SILK_ONLY && st->mode != st->prev_mode && st->prev_mode > 0) + { + OPUS_COPY(tmp_prefill, &st->delay_buffer[(st->encoder_buffer-total_buffer-st->Fs/400)*st->channels], st->channels*st->Fs/400); + } + + if (st->channels*(st->encoder_buffer-(frame_size+total_buffer)) > 0) + { + OPUS_MOVE(st->delay_buffer, &st->delay_buffer[st->channels*frame_size], st->channels*(st->encoder_buffer-frame_size-total_buffer)); + OPUS_COPY(&st->delay_buffer[st->channels*(st->encoder_buffer-frame_size-total_buffer)], + &pcm_buf[0], + (frame_size+total_buffer)*st->channels); + } else { + OPUS_COPY(st->delay_buffer, &pcm_buf[(frame_size+total_buffer-st->encoder_buffer)*st->channels], st->encoder_buffer*st->channels); + } + /* gain_fade() and stereo_fade() need to be after the buffer copying + because we don't want any of this to affect the SILK part */ + if( st->prev_HB_gain < Q15ONE || HB_gain < Q15ONE ) { + gain_fade(pcm_buf, pcm_buf, + st->prev_HB_gain, HB_gain, celt_mode->overlap, frame_size, st->channels, celt_mode->window, st->Fs); + } + st->prev_HB_gain = HB_gain; + if (st->mode != MODE_HYBRID || st->stream_channels==1) + { + if (equiv_rate > 32000) + st->silk_mode.stereoWidth_Q14 = 16384; + else if (equiv_rate < 16000) + st->silk_mode.stereoWidth_Q14 = 0; + else + st->silk_mode.stereoWidth_Q14 = 16384 - 2048*(opus_int32)(32000-equiv_rate)/(equiv_rate-14000); + } + if( !st->energy_masking && st->channels == 2 ) { + /* Apply stereo width reduction (at low bitrates) */ + if( st->hybrid_stereo_width_Q14 < (1 << 14) || st->silk_mode.stereoWidth_Q14 < (1 << 14) ) { + opus_val16 g1, g2; + g1 = st->hybrid_stereo_width_Q14; + g2 = (opus_val16)(st->silk_mode.stereoWidth_Q14); +#ifdef FIXED_POINT + g1 = g1==16384 ? Q15ONE : SHL16(g1,1); + g2 = g2==16384 ? Q15ONE : SHL16(g2,1); +#else + g1 *= (1.f/16384); + g2 *= (1.f/16384); +#endif + stereo_fade(pcm_buf, pcm_buf, g1, g2, celt_mode->overlap, + frame_size, st->channels, celt_mode->window, st->Fs); + st->hybrid_stereo_width_Q14 = st->silk_mode.stereoWidth_Q14; + } + } + + if ( st->mode != MODE_CELT_ONLY && ec_tell(&enc)+17+20*(st->mode == MODE_HYBRID) <= 8*(max_data_bytes-1)) + { + /* For SILK mode, the redundancy is inferred from the length */ + if (st->mode == MODE_HYBRID) + ec_enc_bit_logp(&enc, redundancy, 12); + if (redundancy) + { + int max_redundancy; + ec_enc_bit_logp(&enc, celt_to_silk, 1); + if (st->mode == MODE_HYBRID) + { + /* Reserve the 8 bits needed for the redundancy length, + and at least a few bits for CELT if possible */ + max_redundancy = (max_data_bytes-1)-((ec_tell(&enc)+8+3+7)>>3); + } + else + max_redundancy = (max_data_bytes-1)-((ec_tell(&enc)+7)>>3); + /* Target the same bit-rate for redundancy as for the rest, + up to a max of 257 bytes */ + redundancy_bytes = IMIN(max_redundancy, redundancy_bytes); + redundancy_bytes = IMIN(257, IMAX(2, redundancy_bytes)); + if (st->mode == MODE_HYBRID) + ec_enc_uint(&enc, redundancy_bytes-2, 256); + } + } else { + redundancy = 0; + } + + if (!redundancy) + { + st->silk_bw_switch = 0; + redundancy_bytes = 0; + } + if (st->mode != MODE_CELT_ONLY)start_band=17; + + if (st->mode == MODE_SILK_ONLY) + { + ret = (ec_tell(&enc)+7)>>3; + ec_enc_done(&enc); + nb_compr_bytes = ret; + } else { + nb_compr_bytes = (max_data_bytes-1)-redundancy_bytes; + ec_enc_shrink(&enc, nb_compr_bytes); + } + +#ifndef DISABLE_FLOAT_API + if (redundancy || st->mode != MODE_SILK_ONLY) + celt_encoder_ctl(celt_enc, CELT_SET_ANALYSIS(&analysis_info)); +#endif + if (st->mode == MODE_HYBRID) { + SILKInfo info; + info.signalType = st->silk_mode.signalType; + info.offset = st->silk_mode.offset; + celt_encoder_ctl(celt_enc, CELT_SET_SILK_INFO(&info)); + } + + /* 5 ms redundant frame for CELT->SILK */ + if (redundancy && celt_to_silk) + { + int err; + celt_encoder_ctl(celt_enc, CELT_SET_START_BAND(0)); + celt_encoder_ctl(celt_enc, OPUS_SET_VBR(0)); + celt_encoder_ctl(celt_enc, OPUS_SET_BITRATE(OPUS_BITRATE_MAX)); + err = celt_encode_with_ec(celt_enc, pcm_buf, st->Fs/200, data+nb_compr_bytes, redundancy_bytes, NULL); + if (err < 0) + { + RESTORE_STACK; + return OPUS_INTERNAL_ERROR; + } + celt_encoder_ctl(celt_enc, OPUS_GET_FINAL_RANGE(&redundant_rng)); + celt_encoder_ctl(celt_enc, OPUS_RESET_STATE); + } + + celt_encoder_ctl(celt_enc, CELT_SET_START_BAND(start_band)); + + if (st->mode != MODE_SILK_ONLY) + { + if (st->mode != st->prev_mode && st->prev_mode > 0) + { + unsigned char dummy[2]; + celt_encoder_ctl(celt_enc, OPUS_RESET_STATE); + + /* Prefilling */ + celt_encode_with_ec(celt_enc, tmp_prefill, st->Fs/400, dummy, 2, NULL); + celt_encoder_ctl(celt_enc, CELT_SET_PREDICTION(0)); + } + /* If false, we already busted the budget and we'll end up with a "PLC frame" */ + if (ec_tell(&enc) <= 8*nb_compr_bytes) + { + /* Set the bitrate again if it was overridden in the redundancy code above*/ + if (redundancy && celt_to_silk && st->mode==MODE_HYBRID && st->use_vbr) + celt_encoder_ctl(celt_enc, OPUS_SET_BITRATE(st->bitrate_bps-st->silk_mode.bitRate)); + celt_encoder_ctl(celt_enc, OPUS_SET_VBR(st->use_vbr)); + ret = celt_encode_with_ec(celt_enc, pcm_buf, frame_size, NULL, nb_compr_bytes, &enc); + if (ret < 0) + { + RESTORE_STACK; + return OPUS_INTERNAL_ERROR; + } + /* Put CELT->SILK redundancy data in the right place. */ + if (redundancy && celt_to_silk && st->mode==MODE_HYBRID && st->use_vbr) + { + OPUS_MOVE(data+ret, data+nb_compr_bytes, redundancy_bytes); + nb_compr_bytes = nb_compr_bytes+redundancy_bytes; + } + } + } + + /* 5 ms redundant frame for SILK->CELT */ + if (redundancy && !celt_to_silk) + { + int err; + unsigned char dummy[2]; + int N2, N4; + N2 = st->Fs/200; + N4 = st->Fs/400; + + celt_encoder_ctl(celt_enc, OPUS_RESET_STATE); + celt_encoder_ctl(celt_enc, CELT_SET_START_BAND(0)); + celt_encoder_ctl(celt_enc, CELT_SET_PREDICTION(0)); + celt_encoder_ctl(celt_enc, OPUS_SET_VBR(0)); + celt_encoder_ctl(celt_enc, OPUS_SET_BITRATE(OPUS_BITRATE_MAX)); + + if (st->mode == MODE_HYBRID) + { + /* Shrink packet to what the encoder actually used. */ + nb_compr_bytes = ret; + ec_enc_shrink(&enc, nb_compr_bytes); + } + /* NOTE: We could speed this up slightly (at the expense of code size) by just adding a function that prefills the buffer */ + celt_encode_with_ec(celt_enc, pcm_buf+st->channels*(frame_size-N2-N4), N4, dummy, 2, NULL); + + err = celt_encode_with_ec(celt_enc, pcm_buf+st->channels*(frame_size-N2), N2, data+nb_compr_bytes, redundancy_bytes, NULL); + if (err < 0) + { + RESTORE_STACK; + return OPUS_INTERNAL_ERROR; + } + celt_encoder_ctl(celt_enc, OPUS_GET_FINAL_RANGE(&redundant_rng)); + } + + + + /* Signalling the mode in the first byte */ + data--; + data[0] = gen_toc(st->mode, st->Fs/frame_size, curr_bandwidth, st->stream_channels); + + st->rangeFinal = enc.rng ^ redundant_rng; + + if (to_celt) + st->prev_mode = MODE_CELT_ONLY; + else + st->prev_mode = st->mode; + st->prev_channels = st->stream_channels; + st->prev_framesize = frame_size; + + st->first = 0; + + /* DTX decision */ +#ifndef DISABLE_FLOAT_API + if (st->use_dtx && (analysis_info.valid || is_silence)) + { + if (decide_dtx_mode(activity, &st->nb_no_activity_ms_Q1, 2*1000*frame_size/st->Fs)) + { + st->rangeFinal = 0; + data[0] = gen_toc(st->mode, st->Fs/frame_size, curr_bandwidth, st->stream_channels); + RESTORE_STACK; + return 1; + } + } else { + st->nb_no_activity_ms_Q1 = 0; + } +#endif + + /* In the unlikely case that the SILK encoder busted its target, tell + the decoder to call the PLC */ + if (ec_tell(&enc) > (max_data_bytes-1)*8) + { + if (max_data_bytes < 2) + { + RESTORE_STACK; + return OPUS_BUFFER_TOO_SMALL; + } + data[1] = 0; + ret = 1; + st->rangeFinal = 0; + } else if (st->mode==MODE_SILK_ONLY&&!redundancy) + { + /*When in LPC only mode it's perfectly + reasonable to strip off trailing zero bytes as + the required range decoder behavior is to + fill these in. This can't be done when the MDCT + modes are used because the decoder needs to know + the actual length for allocation purposes.*/ + while(ret>2&&data[ret]==0)ret--; + } + /* Count ToC and redundancy */ + ret += 1+redundancy_bytes; + if (!st->use_vbr) + { + if (opus_packet_pad(data, ret, max_data_bytes) != OPUS_OK) + { + RESTORE_STACK; + return OPUS_INTERNAL_ERROR; + } + ret = max_data_bytes; + } + RESTORE_STACK; + return ret; +} + +#ifdef FIXED_POINT + +#ifndef DISABLE_FLOAT_API +opus_int32 opus_encode_float(OpusEncoder *st, const float *pcm, int analysis_frame_size, + unsigned char *data, opus_int32 max_data_bytes) +{ + int i, ret; + int frame_size; + VARDECL(opus_int16, in); + ALLOC_STACK; + + frame_size = frame_size_select(analysis_frame_size, st->variable_duration, st->Fs); + if (frame_size <= 0) + { + RESTORE_STACK; + return OPUS_BAD_ARG; + } + ALLOC(in, frame_size*st->channels, opus_int16); + + for (i=0;i<frame_size*st->channels;i++) + in[i] = FLOAT2INT16(pcm[i]); + ret = opus_encode_native(st, in, frame_size, data, max_data_bytes, 16, + pcm, analysis_frame_size, 0, -2, st->channels, downmix_float, 1); + RESTORE_STACK; + return ret; +} +#endif + +opus_int32 opus_encode(OpusEncoder *st, const opus_int16 *pcm, int analysis_frame_size, + unsigned char *data, opus_int32 out_data_bytes) +{ + int frame_size; + frame_size = frame_size_select(analysis_frame_size, st->variable_duration, st->Fs); + return opus_encode_native(st, pcm, frame_size, data, out_data_bytes, 16, + pcm, analysis_frame_size, 0, -2, st->channels, downmix_int, 0); +} + +#else +opus_int32 opus_encode(OpusEncoder *st, const opus_int16 *pcm, int analysis_frame_size, + unsigned char *data, opus_int32 max_data_bytes) +{ + int i, ret; + int frame_size; + VARDECL(float, in); + ALLOC_STACK; + + frame_size = frame_size_select(analysis_frame_size, st->variable_duration, st->Fs); + if (frame_size <= 0) + { + RESTORE_STACK; + return OPUS_BAD_ARG; + } + ALLOC(in, frame_size*st->channels, float); + + for (i=0;i<frame_size*st->channels;i++) + in[i] = (1.0f/32768)*pcm[i]; + ret = opus_encode_native(st, in, frame_size, data, max_data_bytes, 16, + pcm, analysis_frame_size, 0, -2, st->channels, downmix_int, 0); + RESTORE_STACK; + return ret; +} +opus_int32 opus_encode_float(OpusEncoder *st, const float *pcm, int analysis_frame_size, + unsigned char *data, opus_int32 out_data_bytes) +{ + int frame_size; + frame_size = frame_size_select(analysis_frame_size, st->variable_duration, st->Fs); + return opus_encode_native(st, pcm, frame_size, data, out_data_bytes, 24, + pcm, analysis_frame_size, 0, -2, st->channels, downmix_float, 1); +} +#endif + + +int opus_encoder_ctl(OpusEncoder *st, int request, ...) +{ + int ret; + CELTEncoder *celt_enc; + va_list ap; + + ret = OPUS_OK; + va_start(ap, request); + + celt_enc = (CELTEncoder*)((char*)st+st->celt_enc_offset); + + switch (request) + { + case OPUS_SET_APPLICATION_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if ( (value != OPUS_APPLICATION_VOIP && value != OPUS_APPLICATION_AUDIO + && value != OPUS_APPLICATION_RESTRICTED_LOWDELAY) + || (!st->first && st->application != value)) + { + ret = OPUS_BAD_ARG; + break; + } + st->application = value; +#ifndef DISABLE_FLOAT_API + st->analysis.application = value; +#endif + } + break; + case OPUS_GET_APPLICATION_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->application; + } + break; + case OPUS_SET_BITRATE_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value != OPUS_AUTO && value != OPUS_BITRATE_MAX) + { + if (value <= 0) + goto bad_arg; + else if (value <= 500) + value = 500; + else if (value > (opus_int32)300000*st->channels) + value = (opus_int32)300000*st->channels; + } + st->user_bitrate_bps = value; + } + break; + case OPUS_GET_BITRATE_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = user_bitrate_to_bitrate(st, st->prev_framesize, 1276); + } + break; + case OPUS_SET_FORCE_CHANNELS_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if((value<1 || value>st->channels) && value != OPUS_AUTO) + { + goto bad_arg; + } + st->force_channels = value; + } + break; + case OPUS_GET_FORCE_CHANNELS_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->force_channels; + } + break; + case OPUS_SET_MAX_BANDWIDTH_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value < OPUS_BANDWIDTH_NARROWBAND || value > OPUS_BANDWIDTH_FULLBAND) + { + goto bad_arg; + } + st->max_bandwidth = value; + if (st->max_bandwidth == OPUS_BANDWIDTH_NARROWBAND) { + st->silk_mode.maxInternalSampleRate = 8000; + } else if (st->max_bandwidth == OPUS_BANDWIDTH_MEDIUMBAND) { + st->silk_mode.maxInternalSampleRate = 12000; + } else { + st->silk_mode.maxInternalSampleRate = 16000; + } + } + break; + case OPUS_GET_MAX_BANDWIDTH_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->max_bandwidth; + } + break; + case OPUS_SET_BANDWIDTH_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if ((value < OPUS_BANDWIDTH_NARROWBAND || value > OPUS_BANDWIDTH_FULLBAND) && value != OPUS_AUTO) + { + goto bad_arg; + } + st->user_bandwidth = value; + if (st->user_bandwidth == OPUS_BANDWIDTH_NARROWBAND) { + st->silk_mode.maxInternalSampleRate = 8000; + } else if (st->user_bandwidth == OPUS_BANDWIDTH_MEDIUMBAND) { + st->silk_mode.maxInternalSampleRate = 12000; + } else { + st->silk_mode.maxInternalSampleRate = 16000; + } + } + break; + case OPUS_GET_BANDWIDTH_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->bandwidth; + } + break; + case OPUS_SET_DTX_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if(value<0 || value>1) + { + goto bad_arg; + } + st->use_dtx = value; + } + break; + case OPUS_GET_DTX_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->use_dtx; + } + break; + case OPUS_SET_COMPLEXITY_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if(value<0 || value>10) + { + goto bad_arg; + } + st->silk_mode.complexity = value; + celt_encoder_ctl(celt_enc, OPUS_SET_COMPLEXITY(value)); + } + break; + case OPUS_GET_COMPLEXITY_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->silk_mode.complexity; + } + break; + case OPUS_SET_INBAND_FEC_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if(value<0 || value>2) + { + goto bad_arg; + } + st->fec_config = value; + st->silk_mode.useInBandFEC = (value != 0); + } + break; + case OPUS_GET_INBAND_FEC_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->fec_config; + } + break; + case OPUS_SET_PACKET_LOSS_PERC_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value < 0 || value > 100) + { + goto bad_arg; + } + st->silk_mode.packetLossPercentage = value; + celt_encoder_ctl(celt_enc, OPUS_SET_PACKET_LOSS_PERC(value)); + } + break; + case OPUS_GET_PACKET_LOSS_PERC_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->silk_mode.packetLossPercentage; + } + break; + case OPUS_SET_VBR_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if(value<0 || value>1) + { + goto bad_arg; + } + st->use_vbr = value; + st->silk_mode.useCBR = 1-value; + } + break; + case OPUS_GET_VBR_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->use_vbr; + } + break; + case OPUS_SET_VOICE_RATIO_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value<-1 || value>100) + { + goto bad_arg; + } + st->voice_ratio = value; + } + break; + case OPUS_GET_VOICE_RATIO_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->voice_ratio; + } + break; + case OPUS_SET_VBR_CONSTRAINT_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if(value<0 || value>1) + { + goto bad_arg; + } + st->vbr_constraint = value; + } + break; + case OPUS_GET_VBR_CONSTRAINT_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->vbr_constraint; + } + break; + case OPUS_SET_SIGNAL_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if(value!=OPUS_AUTO && value!=OPUS_SIGNAL_VOICE && value!=OPUS_SIGNAL_MUSIC) + { + goto bad_arg; + } + st->signal_type = value; + } + break; + case OPUS_GET_SIGNAL_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->signal_type; + } + break; + case OPUS_GET_LOOKAHEAD_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->Fs/400; + if (st->application != OPUS_APPLICATION_RESTRICTED_LOWDELAY) + *value += st->delay_compensation; + } + break; + case OPUS_GET_SAMPLE_RATE_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->Fs; + } + break; + case OPUS_GET_FINAL_RANGE_REQUEST: + { + opus_uint32 *value = va_arg(ap, opus_uint32*); + if (!value) + { + goto bad_arg; + } + *value = st->rangeFinal; + } + break; + case OPUS_SET_LSB_DEPTH_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value<8 || value>24) + { + goto bad_arg; + } + st->lsb_depth=value; + } + break; + case OPUS_GET_LSB_DEPTH_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->lsb_depth; + } + break; + case OPUS_SET_EXPERT_FRAME_DURATION_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value != OPUS_FRAMESIZE_ARG && value != OPUS_FRAMESIZE_2_5_MS && + value != OPUS_FRAMESIZE_5_MS && value != OPUS_FRAMESIZE_10_MS && + value != OPUS_FRAMESIZE_20_MS && value != OPUS_FRAMESIZE_40_MS && + value != OPUS_FRAMESIZE_60_MS && value != OPUS_FRAMESIZE_80_MS && + value != OPUS_FRAMESIZE_100_MS && value != OPUS_FRAMESIZE_120_MS) + { + goto bad_arg; + } + st->variable_duration = value; + } + break; + case OPUS_GET_EXPERT_FRAME_DURATION_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->variable_duration; + } + break; + case OPUS_SET_PREDICTION_DISABLED_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value > 1 || value < 0) + goto bad_arg; + st->silk_mode.reducedDependency = value; + } + break; + case OPUS_GET_PREDICTION_DISABLED_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + goto bad_arg; + *value = st->silk_mode.reducedDependency; + } + break; + case OPUS_SET_PHASE_INVERSION_DISABLED_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if(value<0 || value>1) + { + goto bad_arg; + } + celt_encoder_ctl(celt_enc, OPUS_SET_PHASE_INVERSION_DISABLED(value)); + } + break; + case OPUS_GET_PHASE_INVERSION_DISABLED_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + celt_encoder_ctl(celt_enc, OPUS_GET_PHASE_INVERSION_DISABLED(value)); + } + break; + case OPUS_RESET_STATE: + { + void *silk_enc; + silk_EncControlStruct dummy; + char *start; + silk_enc = (char*)st+st->silk_enc_offset; +#ifndef DISABLE_FLOAT_API + tonality_analysis_reset(&st->analysis); +#endif + + start = (char*)&st->OPUS_ENCODER_RESET_START; + OPUS_CLEAR(start, sizeof(OpusEncoder) - (start - (char*)st)); + + celt_encoder_ctl(celt_enc, OPUS_RESET_STATE); + silk_InitEncoder( silk_enc, st->arch, &dummy ); + st->stream_channels = st->channels; + st->hybrid_stereo_width_Q14 = 1 << 14; + st->prev_HB_gain = Q15ONE; + st->first = 1; + st->mode = MODE_HYBRID; + st->bandwidth = OPUS_BANDWIDTH_FULLBAND; + st->variable_HP_smth2_Q15 = silk_LSHIFT( silk_lin2log( VARIABLE_HP_MIN_CUTOFF_HZ ), 8 ); + } + break; + case OPUS_SET_FORCE_MODE_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if ((value < MODE_SILK_ONLY || value > MODE_CELT_ONLY) && value != OPUS_AUTO) + { + goto bad_arg; + } + st->user_forced_mode = value; + } + break; + case OPUS_SET_LFE_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + st->lfe = value; + ret = celt_encoder_ctl(celt_enc, OPUS_SET_LFE(value)); + } + break; + case OPUS_SET_ENERGY_MASK_REQUEST: + { + opus_val16 *value = va_arg(ap, opus_val16*); + st->energy_masking = value; + ret = celt_encoder_ctl(celt_enc, OPUS_SET_ENERGY_MASK(value)); + } + break; + case OPUS_GET_IN_DTX_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + if (st->silk_mode.useDTX && (st->prev_mode == MODE_SILK_ONLY || st->prev_mode == MODE_HYBRID)) { + /* DTX determined by Silk. */ + silk_encoder *silk_enc = (silk_encoder*)(void *)((char*)st+st->silk_enc_offset); + *value = silk_enc->state_Fxx[0].sCmn.noSpeechCounter >= NB_SPEECH_FRAMES_BEFORE_DTX; + /* Stereo: check second channel unless only the middle channel was encoded. */ + if(*value == 1 && st->silk_mode.nChannelsInternal == 2 && silk_enc->prev_decode_only_middle == 0) { + *value = silk_enc->state_Fxx[1].sCmn.noSpeechCounter >= NB_SPEECH_FRAMES_BEFORE_DTX; + } + } +#ifndef DISABLE_FLOAT_API + else if (st->use_dtx) { + /* DTX determined by Opus. */ + *value = st->nb_no_activity_ms_Q1 >= NB_SPEECH_FRAMES_BEFORE_DTX*20*2; + } +#endif + else { + *value = 0; + } + } + break; + case CELT_GET_MODE_REQUEST: + { + const CELTMode ** value = va_arg(ap, const CELTMode**); + if (!value) + { + goto bad_arg; + } + ret = celt_encoder_ctl(celt_enc, CELT_GET_MODE(value)); + } + break; + default: + /* fprintf(stderr, "unknown opus_encoder_ctl() request: %d", request);*/ + ret = OPUS_UNIMPLEMENTED; + break; + } + va_end(ap); + return ret; +bad_arg: + va_end(ap); + return OPUS_BAD_ARG; +} + +void opus_encoder_destroy(OpusEncoder *st) +{ + opus_free(st); +} diff --git a/media/libopus/src/opus_multistream.c b/media/libopus/src/opus_multistream.c new file mode 100644 index 0000000000..09c3639b7f --- /dev/null +++ b/media/libopus/src/opus_multistream.c @@ -0,0 +1,92 @@ +/* Copyright (c) 2011 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "opus_multistream.h" +#include "opus.h" +#include "opus_private.h" +#include "stack_alloc.h" +#include <stdarg.h> +#include "float_cast.h" +#include "os_support.h" + + +int validate_layout(const ChannelLayout *layout) +{ + int i, max_channel; + + max_channel = layout->nb_streams+layout->nb_coupled_streams; + if (max_channel>255) + return 0; + for (i=0;i<layout->nb_channels;i++) + { + if (layout->mapping[i] >= max_channel && layout->mapping[i] != 255) + return 0; + } + return 1; +} + + +int get_left_channel(const ChannelLayout *layout, int stream_id, int prev) +{ + int i; + i = (prev<0) ? 0 : prev+1; + for (;i<layout->nb_channels;i++) + { + if (layout->mapping[i]==stream_id*2) + return i; + } + return -1; +} + +int get_right_channel(const ChannelLayout *layout, int stream_id, int prev) +{ + int i; + i = (prev<0) ? 0 : prev+1; + for (;i<layout->nb_channels;i++) + { + if (layout->mapping[i]==stream_id*2+1) + return i; + } + return -1; +} + +int get_mono_channel(const ChannelLayout *layout, int stream_id, int prev) +{ + int i; + i = (prev<0) ? 0 : prev+1; + for (;i<layout->nb_channels;i++) + { + if (layout->mapping[i]==stream_id+layout->nb_coupled_streams) + return i; + } + return -1; +} + diff --git a/media/libopus/src/opus_multistream_decoder.c b/media/libopus/src/opus_multistream_decoder.c new file mode 100644 index 0000000000..a2837c3549 --- /dev/null +++ b/media/libopus/src/opus_multistream_decoder.c @@ -0,0 +1,552 @@ +/* Copyright (c) 2011 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "opus_multistream.h" +#include "opus.h" +#include "opus_private.h" +#include "stack_alloc.h" +#include <stdarg.h> +#include "float_cast.h" +#include "os_support.h" + +/* DECODER */ + +#if defined(ENABLE_HARDENING) || defined(ENABLE_ASSERTIONS) +static void validate_ms_decoder(OpusMSDecoder *st) +{ + validate_layout(&st->layout); +} +#define VALIDATE_MS_DECODER(st) validate_ms_decoder(st) +#else +#define VALIDATE_MS_DECODER(st) +#endif + + +opus_int32 opus_multistream_decoder_get_size(int nb_streams, int nb_coupled_streams) +{ + int coupled_size; + int mono_size; + + if(nb_streams<1||nb_coupled_streams>nb_streams||nb_coupled_streams<0)return 0; + coupled_size = opus_decoder_get_size(2); + mono_size = opus_decoder_get_size(1); + return align(sizeof(OpusMSDecoder)) + + nb_coupled_streams * align(coupled_size) + + (nb_streams-nb_coupled_streams) * align(mono_size); +} + +int opus_multistream_decoder_init( + OpusMSDecoder *st, + opus_int32 Fs, + int channels, + int streams, + int coupled_streams, + const unsigned char *mapping +) +{ + int coupled_size; + int mono_size; + int i, ret; + char *ptr; + + if ((channels>255) || (channels<1) || (coupled_streams>streams) || + (streams<1) || (coupled_streams<0) || (streams>255-coupled_streams)) + return OPUS_BAD_ARG; + + st->layout.nb_channels = channels; + st->layout.nb_streams = streams; + st->layout.nb_coupled_streams = coupled_streams; + + for (i=0;i<st->layout.nb_channels;i++) + st->layout.mapping[i] = mapping[i]; + if (!validate_layout(&st->layout)) + return OPUS_BAD_ARG; + + ptr = (char*)st + align(sizeof(OpusMSDecoder)); + coupled_size = opus_decoder_get_size(2); + mono_size = opus_decoder_get_size(1); + + for (i=0;i<st->layout.nb_coupled_streams;i++) + { + ret=opus_decoder_init((OpusDecoder*)ptr, Fs, 2); + if(ret!=OPUS_OK)return ret; + ptr += align(coupled_size); + } + for (;i<st->layout.nb_streams;i++) + { + ret=opus_decoder_init((OpusDecoder*)ptr, Fs, 1); + if(ret!=OPUS_OK)return ret; + ptr += align(mono_size); + } + return OPUS_OK; +} + + +OpusMSDecoder *opus_multistream_decoder_create( + opus_int32 Fs, + int channels, + int streams, + int coupled_streams, + const unsigned char *mapping, + int *error +) +{ + int ret; + OpusMSDecoder *st; + if ((channels>255) || (channels<1) || (coupled_streams>streams) || + (streams<1) || (coupled_streams<0) || (streams>255-coupled_streams)) + { + if (error) + *error = OPUS_BAD_ARG; + return NULL; + } + st = (OpusMSDecoder *)opus_alloc(opus_multistream_decoder_get_size(streams, coupled_streams)); + if (st==NULL) + { + if (error) + *error = OPUS_ALLOC_FAIL; + return NULL; + } + ret = opus_multistream_decoder_init(st, Fs, channels, streams, coupled_streams, mapping); + if (error) + *error = ret; + if (ret != OPUS_OK) + { + opus_free(st); + st = NULL; + } + return st; +} + +static int opus_multistream_packet_validate(const unsigned char *data, + opus_int32 len, int nb_streams, opus_int32 Fs) +{ + int s; + int count; + unsigned char toc; + opus_int16 size[48]; + int samples=0; + opus_int32 packet_offset; + + for (s=0;s<nb_streams;s++) + { + int tmp_samples; + if (len<=0) + return OPUS_INVALID_PACKET; + count = opus_packet_parse_impl(data, len, s!=nb_streams-1, &toc, NULL, + size, NULL, &packet_offset); + if (count<0) + return count; + tmp_samples = opus_packet_get_nb_samples(data, packet_offset, Fs); + if (s!=0 && samples != tmp_samples) + return OPUS_INVALID_PACKET; + samples = tmp_samples; + data += packet_offset; + len -= packet_offset; + } + return samples; +} + +int opus_multistream_decode_native( + OpusMSDecoder *st, + const unsigned char *data, + opus_int32 len, + void *pcm, + opus_copy_channel_out_func copy_channel_out, + int frame_size, + int decode_fec, + int soft_clip, + void *user_data +) +{ + opus_int32 Fs; + int coupled_size; + int mono_size; + int s, c; + char *ptr; + int do_plc=0; + VARDECL(opus_val16, buf); + ALLOC_STACK; + + VALIDATE_MS_DECODER(st); + if (frame_size <= 0) + { + RESTORE_STACK; + return OPUS_BAD_ARG; + } + /* Limit frame_size to avoid excessive stack allocations. */ + MUST_SUCCEED(opus_multistream_decoder_ctl(st, OPUS_GET_SAMPLE_RATE(&Fs))); + frame_size = IMIN(frame_size, Fs/25*3); + ALLOC(buf, 2*frame_size, opus_val16); + ptr = (char*)st + align(sizeof(OpusMSDecoder)); + coupled_size = opus_decoder_get_size(2); + mono_size = opus_decoder_get_size(1); + + if (len==0) + do_plc = 1; + if (len < 0) + { + RESTORE_STACK; + return OPUS_BAD_ARG; + } + if (!do_plc && len < 2*st->layout.nb_streams-1) + { + RESTORE_STACK; + return OPUS_INVALID_PACKET; + } + if (!do_plc) + { + int ret = opus_multistream_packet_validate(data, len, st->layout.nb_streams, Fs); + if (ret < 0) + { + RESTORE_STACK; + return ret; + } else if (ret > frame_size) + { + RESTORE_STACK; + return OPUS_BUFFER_TOO_SMALL; + } + } + for (s=0;s<st->layout.nb_streams;s++) + { + OpusDecoder *dec; + opus_int32 packet_offset; + int ret; + + dec = (OpusDecoder*)ptr; + ptr += (s < st->layout.nb_coupled_streams) ? align(coupled_size) : align(mono_size); + + if (!do_plc && len<=0) + { + RESTORE_STACK; + return OPUS_INTERNAL_ERROR; + } + packet_offset = 0; + ret = opus_decode_native(dec, data, len, buf, frame_size, decode_fec, s!=st->layout.nb_streams-1, &packet_offset, soft_clip); + if (!do_plc) + { + data += packet_offset; + len -= packet_offset; + } + if (ret <= 0) + { + RESTORE_STACK; + return ret; + } + frame_size = ret; + if (s < st->layout.nb_coupled_streams) + { + int chan, prev; + prev = -1; + /* Copy "left" audio to the channel(s) where it belongs */ + while ( (chan = get_left_channel(&st->layout, s, prev)) != -1) + { + (*copy_channel_out)(pcm, st->layout.nb_channels, chan, + buf, 2, frame_size, user_data); + prev = chan; + } + prev = -1; + /* Copy "right" audio to the channel(s) where it belongs */ + while ( (chan = get_right_channel(&st->layout, s, prev)) != -1) + { + (*copy_channel_out)(pcm, st->layout.nb_channels, chan, + buf+1, 2, frame_size, user_data); + prev = chan; + } + } else { + int chan, prev; + prev = -1; + /* Copy audio to the channel(s) where it belongs */ + while ( (chan = get_mono_channel(&st->layout, s, prev)) != -1) + { + (*copy_channel_out)(pcm, st->layout.nb_channels, chan, + buf, 1, frame_size, user_data); + prev = chan; + } + } + } + /* Handle muted channels */ + for (c=0;c<st->layout.nb_channels;c++) + { + if (st->layout.mapping[c] == 255) + { + (*copy_channel_out)(pcm, st->layout.nb_channels, c, + NULL, 0, frame_size, user_data); + } + } + RESTORE_STACK; + return frame_size; +} + +#if !defined(DISABLE_FLOAT_API) +static void opus_copy_channel_out_float( + void *dst, + int dst_stride, + int dst_channel, + const opus_val16 *src, + int src_stride, + int frame_size, + void *user_data +) +{ + float *float_dst; + opus_int32 i; + (void)user_data; + float_dst = (float*)dst; + if (src != NULL) + { + for (i=0;i<frame_size;i++) +#if defined(FIXED_POINT) + float_dst[i*dst_stride+dst_channel] = (1/32768.f)*src[i*src_stride]; +#else + float_dst[i*dst_stride+dst_channel] = src[i*src_stride]; +#endif + } + else + { + for (i=0;i<frame_size;i++) + float_dst[i*dst_stride+dst_channel] = 0; + } +} +#endif + +static void opus_copy_channel_out_short( + void *dst, + int dst_stride, + int dst_channel, + const opus_val16 *src, + int src_stride, + int frame_size, + void *user_data +) +{ + opus_int16 *short_dst; + opus_int32 i; + (void)user_data; + short_dst = (opus_int16*)dst; + if (src != NULL) + { + for (i=0;i<frame_size;i++) +#if defined(FIXED_POINT) + short_dst[i*dst_stride+dst_channel] = src[i*src_stride]; +#else + short_dst[i*dst_stride+dst_channel] = FLOAT2INT16(src[i*src_stride]); +#endif + } + else + { + for (i=0;i<frame_size;i++) + short_dst[i*dst_stride+dst_channel] = 0; + } +} + + + +#ifdef FIXED_POINT +int opus_multistream_decode( + OpusMSDecoder *st, + const unsigned char *data, + opus_int32 len, + opus_int16 *pcm, + int frame_size, + int decode_fec +) +{ + return opus_multistream_decode_native(st, data, len, + pcm, opus_copy_channel_out_short, frame_size, decode_fec, 0, NULL); +} + +#ifndef DISABLE_FLOAT_API +int opus_multistream_decode_float(OpusMSDecoder *st, const unsigned char *data, + opus_int32 len, float *pcm, int frame_size, int decode_fec) +{ + return opus_multistream_decode_native(st, data, len, + pcm, opus_copy_channel_out_float, frame_size, decode_fec, 0, NULL); +} +#endif + +#else + +int opus_multistream_decode(OpusMSDecoder *st, const unsigned char *data, + opus_int32 len, opus_int16 *pcm, int frame_size, int decode_fec) +{ + return opus_multistream_decode_native(st, data, len, + pcm, opus_copy_channel_out_short, frame_size, decode_fec, 1, NULL); +} + +int opus_multistream_decode_float( + OpusMSDecoder *st, + const unsigned char *data, + opus_int32 len, + opus_val16 *pcm, + int frame_size, + int decode_fec +) +{ + return opus_multistream_decode_native(st, data, len, + pcm, opus_copy_channel_out_float, frame_size, decode_fec, 0, NULL); +} +#endif + +int opus_multistream_decoder_ctl_va_list(OpusMSDecoder *st, int request, + va_list ap) +{ + int coupled_size, mono_size; + char *ptr; + int ret = OPUS_OK; + + coupled_size = opus_decoder_get_size(2); + mono_size = opus_decoder_get_size(1); + ptr = (char*)st + align(sizeof(OpusMSDecoder)); + switch (request) + { + case OPUS_GET_BANDWIDTH_REQUEST: + case OPUS_GET_SAMPLE_RATE_REQUEST: + case OPUS_GET_GAIN_REQUEST: + case OPUS_GET_LAST_PACKET_DURATION_REQUEST: + case OPUS_GET_PHASE_INVERSION_DISABLED_REQUEST: + { + OpusDecoder *dec; + /* For int32* GET params, just query the first stream */ + opus_int32 *value = va_arg(ap, opus_int32*); + dec = (OpusDecoder*)ptr; + ret = opus_decoder_ctl(dec, request, value); + } + break; + case OPUS_GET_FINAL_RANGE_REQUEST: + { + int s; + opus_uint32 *value = va_arg(ap, opus_uint32*); + opus_uint32 tmp; + if (!value) + { + goto bad_arg; + } + *value = 0; + for (s=0;s<st->layout.nb_streams;s++) + { + OpusDecoder *dec; + dec = (OpusDecoder*)ptr; + if (s < st->layout.nb_coupled_streams) + ptr += align(coupled_size); + else + ptr += align(mono_size); + ret = opus_decoder_ctl(dec, request, &tmp); + if (ret != OPUS_OK) break; + *value ^= tmp; + } + } + break; + case OPUS_RESET_STATE: + { + int s; + for (s=0;s<st->layout.nb_streams;s++) + { + OpusDecoder *dec; + + dec = (OpusDecoder*)ptr; + if (s < st->layout.nb_coupled_streams) + ptr += align(coupled_size); + else + ptr += align(mono_size); + ret = opus_decoder_ctl(dec, OPUS_RESET_STATE); + if (ret != OPUS_OK) + break; + } + } + break; + case OPUS_MULTISTREAM_GET_DECODER_STATE_REQUEST: + { + int s; + opus_int32 stream_id; + OpusDecoder **value; + stream_id = va_arg(ap, opus_int32); + if (stream_id<0 || stream_id >= st->layout.nb_streams) + goto bad_arg; + value = va_arg(ap, OpusDecoder**); + if (!value) + { + goto bad_arg; + } + for (s=0;s<stream_id;s++) + { + if (s < st->layout.nb_coupled_streams) + ptr += align(coupled_size); + else + ptr += align(mono_size); + } + *value = (OpusDecoder*)ptr; + } + break; + case OPUS_SET_GAIN_REQUEST: + case OPUS_SET_PHASE_INVERSION_DISABLED_REQUEST: + { + int s; + /* This works for int32 params */ + opus_int32 value = va_arg(ap, opus_int32); + for (s=0;s<st->layout.nb_streams;s++) + { + OpusDecoder *dec; + + dec = (OpusDecoder*)ptr; + if (s < st->layout.nb_coupled_streams) + ptr += align(coupled_size); + else + ptr += align(mono_size); + ret = opus_decoder_ctl(dec, request, value); + if (ret != OPUS_OK) + break; + } + } + break; + default: + ret = OPUS_UNIMPLEMENTED; + break; + } + return ret; +bad_arg: + return OPUS_BAD_ARG; +} + +int opus_multistream_decoder_ctl(OpusMSDecoder *st, int request, ...) +{ + int ret; + va_list ap; + va_start(ap, request); + ret = opus_multistream_decoder_ctl_va_list(st, request, ap); + va_end(ap); + return ret; +} + +void opus_multistream_decoder_destroy(OpusMSDecoder *st) +{ + opus_free(st); +} diff --git a/media/libopus/src/opus_multistream_encoder.c b/media/libopus/src/opus_multistream_encoder.c new file mode 100644 index 0000000000..213e3eb2c2 --- /dev/null +++ b/media/libopus/src/opus_multistream_encoder.c @@ -0,0 +1,1329 @@ +/* Copyright (c) 2011 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "opus_multistream.h" +#include "opus.h" +#include "opus_private.h" +#include "stack_alloc.h" +#include <stdarg.h> +#include "float_cast.h" +#include "os_support.h" +#include "mathops.h" +#include "mdct.h" +#include "modes.h" +#include "bands.h" +#include "quant_bands.h" +#include "pitch.h" + +typedef struct { + int nb_streams; + int nb_coupled_streams; + unsigned char mapping[8]; +} VorbisLayout; + +/* Index is nb_channel-1*/ +static const VorbisLayout vorbis_mappings[8] = { + {1, 0, {0}}, /* 1: mono */ + {1, 1, {0, 1}}, /* 2: stereo */ + {2, 1, {0, 2, 1}}, /* 3: 1-d surround */ + {2, 2, {0, 1, 2, 3}}, /* 4: quadraphonic surround */ + {3, 2, {0, 4, 1, 2, 3}}, /* 5: 5-channel surround */ + {4, 2, {0, 4, 1, 2, 3, 5}}, /* 6: 5.1 surround */ + {4, 3, {0, 4, 1, 2, 3, 5, 6}}, /* 7: 6.1 surround */ + {5, 3, {0, 6, 1, 2, 3, 4, 5, 7}}, /* 8: 7.1 surround */ +}; + +static opus_val32 *ms_get_preemph_mem(OpusMSEncoder *st) +{ + int s; + char *ptr; + int coupled_size, mono_size; + + coupled_size = opus_encoder_get_size(2); + mono_size = opus_encoder_get_size(1); + ptr = (char*)st + align(sizeof(OpusMSEncoder)); + for (s=0;s<st->layout.nb_streams;s++) + { + if (s < st->layout.nb_coupled_streams) + ptr += align(coupled_size); + else + ptr += align(mono_size); + } + /* void* cast avoids clang -Wcast-align warning */ + return (opus_val32*)(void*)(ptr+st->layout.nb_channels*120*sizeof(opus_val32)); +} + +static opus_val32 *ms_get_window_mem(OpusMSEncoder *st) +{ + int s; + char *ptr; + int coupled_size, mono_size; + + coupled_size = opus_encoder_get_size(2); + mono_size = opus_encoder_get_size(1); + ptr = (char*)st + align(sizeof(OpusMSEncoder)); + for (s=0;s<st->layout.nb_streams;s++) + { + if (s < st->layout.nb_coupled_streams) + ptr += align(coupled_size); + else + ptr += align(mono_size); + } + /* void* cast avoids clang -Wcast-align warning */ + return (opus_val32*)(void*)ptr; +} + +static int validate_ambisonics(int nb_channels, int *nb_streams, int *nb_coupled_streams) +{ + int order_plus_one; + int acn_channels; + int nondiegetic_channels; + + if (nb_channels < 1 || nb_channels > 227) + return 0; + + order_plus_one = isqrt32(nb_channels); + acn_channels = order_plus_one * order_plus_one; + nondiegetic_channels = nb_channels - acn_channels; + + if (nondiegetic_channels != 0 && nondiegetic_channels != 2) + return 0; + + if (nb_streams) + *nb_streams = acn_channels + (nondiegetic_channels != 0); + if (nb_coupled_streams) + *nb_coupled_streams = nondiegetic_channels != 0; + return 1; +} + +static int validate_encoder_layout(const ChannelLayout *layout) +{ + int s; + for (s=0;s<layout->nb_streams;s++) + { + if (s < layout->nb_coupled_streams) + { + if (get_left_channel(layout, s, -1)==-1) + return 0; + if (get_right_channel(layout, s, -1)==-1) + return 0; + } else { + if (get_mono_channel(layout, s, -1)==-1) + return 0; + } + } + return 1; +} + +static void channel_pos(int channels, int pos[8]) +{ + /* Position in the mix: 0 don't mix, 1: left, 2: center, 3:right */ + if (channels==4) + { + pos[0]=1; + pos[1]=3; + pos[2]=1; + pos[3]=3; + } else if (channels==3||channels==5||channels==6) + { + pos[0]=1; + pos[1]=2; + pos[2]=3; + pos[3]=1; + pos[4]=3; + pos[5]=0; + } else if (channels==7) + { + pos[0]=1; + pos[1]=2; + pos[2]=3; + pos[3]=1; + pos[4]=3; + pos[5]=2; + pos[6]=0; + } else if (channels==8) + { + pos[0]=1; + pos[1]=2; + pos[2]=3; + pos[3]=1; + pos[4]=3; + pos[5]=1; + pos[6]=3; + pos[7]=0; + } +} + +#if 1 +/* Computes a rough approximation of log2(2^a + 2^b) */ +static opus_val16 logSum(opus_val16 a, opus_val16 b) +{ + opus_val16 max; + opus_val32 diff; + opus_val16 frac; + static const opus_val16 diff_table[17] = { + QCONST16(0.5000000f, DB_SHIFT), QCONST16(0.2924813f, DB_SHIFT), QCONST16(0.1609640f, DB_SHIFT), QCONST16(0.0849625f, DB_SHIFT), + QCONST16(0.0437314f, DB_SHIFT), QCONST16(0.0221971f, DB_SHIFT), QCONST16(0.0111839f, DB_SHIFT), QCONST16(0.0056136f, DB_SHIFT), + QCONST16(0.0028123f, DB_SHIFT) + }; + int low; + if (a>b) + { + max = a; + diff = SUB32(EXTEND32(a),EXTEND32(b)); + } else { + max = b; + diff = SUB32(EXTEND32(b),EXTEND32(a)); + } + if (!(diff < QCONST16(8.f, DB_SHIFT))) /* inverted to catch NaNs */ + return max; +#ifdef FIXED_POINT + low = SHR32(diff, DB_SHIFT-1); + frac = SHL16(diff - SHL16(low, DB_SHIFT-1), 16-DB_SHIFT); +#else + low = (int)floor(2*diff); + frac = 2*diff - low; +#endif + return max + diff_table[low] + MULT16_16_Q15(frac, SUB16(diff_table[low+1], diff_table[low])); +} +#else +opus_val16 logSum(opus_val16 a, opus_val16 b) +{ + return log2(pow(4, a)+ pow(4, b))/2; +} +#endif + +void surround_analysis(const CELTMode *celt_mode, const void *pcm, opus_val16 *bandLogE, opus_val32 *mem, opus_val32 *preemph_mem, + int len, int overlap, int channels, int rate, opus_copy_channel_in_func copy_channel_in, int arch +) +{ + int c; + int i; + int LM; + int pos[8] = {0}; + int upsample; + int frame_size; + int freq_size; + opus_val16 channel_offset; + opus_val32 bandE[21]; + opus_val16 maskLogE[3][21]; + VARDECL(opus_val32, in); + VARDECL(opus_val16, x); + VARDECL(opus_val32, freq); + SAVE_STACK; + + upsample = resampling_factor(rate); + frame_size = len*upsample; + freq_size = IMIN(960, frame_size); + + /* LM = log2(frame_size / 120) */ + for (LM=0;LM<celt_mode->maxLM;LM++) + if (celt_mode->shortMdctSize<<LM==frame_size) + break; + + ALLOC(in, frame_size+overlap, opus_val32); + ALLOC(x, len, opus_val16); + ALLOC(freq, freq_size, opus_val32); + + channel_pos(channels, pos); + + for (c=0;c<3;c++) + for (i=0;i<21;i++) + maskLogE[c][i] = -QCONST16(28.f, DB_SHIFT); + + for (c=0;c<channels;c++) + { + int frame; + int nb_frames = frame_size/freq_size; + celt_assert(nb_frames*freq_size == frame_size); + OPUS_COPY(in, mem+c*overlap, overlap); + (*copy_channel_in)(x, 1, pcm, channels, c, len, NULL); + celt_preemphasis(x, in+overlap, frame_size, 1, upsample, celt_mode->preemph, preemph_mem+c, 0); +#ifndef FIXED_POINT + { + opus_val32 sum; + sum = celt_inner_prod(in, in, frame_size+overlap, 0); + /* This should filter out both NaNs and ridiculous signals that could + cause NaNs further down. */ + if (!(sum < 1e18f) || celt_isnan(sum)) + { + OPUS_CLEAR(in, frame_size+overlap); + preemph_mem[c] = 0; + } + } +#endif + OPUS_CLEAR(bandE, 21); + for (frame=0;frame<nb_frames;frame++) + { + opus_val32 tmpE[21]; + clt_mdct_forward(&celt_mode->mdct, in+960*frame, freq, celt_mode->window, + overlap, celt_mode->maxLM-LM, 1, arch); + if (upsample != 1) + { + int bound = freq_size/upsample; + for (i=0;i<bound;i++) + freq[i] *= upsample; + for (;i<freq_size;i++) + freq[i] = 0; + } + + compute_band_energies(celt_mode, freq, tmpE, 21, 1, LM, arch); + /* If we have multiple frames, take the max energy. */ + for (i=0;i<21;i++) + bandE[i] = MAX32(bandE[i], tmpE[i]); + } + amp2Log2(celt_mode, 21, 21, bandE, bandLogE+21*c, 1); + /* Apply spreading function with -6 dB/band going up and -12 dB/band going down. */ + for (i=1;i<21;i++) + bandLogE[21*c+i] = MAX16(bandLogE[21*c+i], bandLogE[21*c+i-1]-QCONST16(1.f, DB_SHIFT)); + for (i=19;i>=0;i--) + bandLogE[21*c+i] = MAX16(bandLogE[21*c+i], bandLogE[21*c+i+1]-QCONST16(2.f, DB_SHIFT)); + if (pos[c]==1) + { + for (i=0;i<21;i++) + maskLogE[0][i] = logSum(maskLogE[0][i], bandLogE[21*c+i]); + } else if (pos[c]==3) + { + for (i=0;i<21;i++) + maskLogE[2][i] = logSum(maskLogE[2][i], bandLogE[21*c+i]); + } else if (pos[c]==2) + { + for (i=0;i<21;i++) + { + maskLogE[0][i] = logSum(maskLogE[0][i], bandLogE[21*c+i]-QCONST16(.5f, DB_SHIFT)); + maskLogE[2][i] = logSum(maskLogE[2][i], bandLogE[21*c+i]-QCONST16(.5f, DB_SHIFT)); + } + } +#if 0 + for (i=0;i<21;i++) + printf("%f ", bandLogE[21*c+i]); + float sum=0; + for (i=0;i<21;i++) + sum += bandLogE[21*c+i]; + printf("%f ", sum/21); +#endif + OPUS_COPY(mem+c*overlap, in+frame_size, overlap); + } + for (i=0;i<21;i++) + maskLogE[1][i] = MIN32(maskLogE[0][i],maskLogE[2][i]); + channel_offset = HALF16(celt_log2(QCONST32(2.f,14)/(channels-1))); + for (c=0;c<3;c++) + for (i=0;i<21;i++) + maskLogE[c][i] += channel_offset; +#if 0 + for (c=0;c<3;c++) + { + for (i=0;i<21;i++) + printf("%f ", maskLogE[c][i]); + } +#endif + for (c=0;c<channels;c++) + { + opus_val16 *mask; + if (pos[c]!=0) + { + mask = &maskLogE[pos[c]-1][0]; + for (i=0;i<21;i++) + bandLogE[21*c+i] = bandLogE[21*c+i] - mask[i]; + } else { + for (i=0;i<21;i++) + bandLogE[21*c+i] = 0; + } +#if 0 + for (i=0;i<21;i++) + printf("%f ", bandLogE[21*c+i]); + printf("\n"); +#endif +#if 0 + float sum=0; + for (i=0;i<21;i++) + sum += bandLogE[21*c+i]; + printf("%f ", sum/(float)QCONST32(21.f, DB_SHIFT)); + printf("\n"); +#endif + } + RESTORE_STACK; +} + +opus_int32 opus_multistream_encoder_get_size(int nb_streams, int nb_coupled_streams) +{ + int coupled_size; + int mono_size; + + if(nb_streams<1||nb_coupled_streams>nb_streams||nb_coupled_streams<0)return 0; + coupled_size = opus_encoder_get_size(2); + mono_size = opus_encoder_get_size(1); + return align(sizeof(OpusMSEncoder)) + + nb_coupled_streams * align(coupled_size) + + (nb_streams-nb_coupled_streams) * align(mono_size); +} + +opus_int32 opus_multistream_surround_encoder_get_size(int channels, int mapping_family) +{ + int nb_streams; + int nb_coupled_streams; + opus_int32 size; + + if (mapping_family==0) + { + if (channels==1) + { + nb_streams=1; + nb_coupled_streams=0; + } else if (channels==2) + { + nb_streams=1; + nb_coupled_streams=1; + } else + return 0; + } else if (mapping_family==1 && channels<=8 && channels>=1) + { + nb_streams=vorbis_mappings[channels-1].nb_streams; + nb_coupled_streams=vorbis_mappings[channels-1].nb_coupled_streams; + } else if (mapping_family==255) + { + nb_streams=channels; + nb_coupled_streams=0; + } else if (mapping_family==2) + { + if (!validate_ambisonics(channels, &nb_streams, &nb_coupled_streams)) + return 0; + } else + return 0; + size = opus_multistream_encoder_get_size(nb_streams, nb_coupled_streams); + if (channels>2) + { + size += channels*(120*sizeof(opus_val32) + sizeof(opus_val32)); + } + return size; +} + +static int opus_multistream_encoder_init_impl( + OpusMSEncoder *st, + opus_int32 Fs, + int channels, + int streams, + int coupled_streams, + const unsigned char *mapping, + int application, + MappingType mapping_type +) +{ + int coupled_size; + int mono_size; + int i, ret; + char *ptr; + + if ((channels>255) || (channels<1) || (coupled_streams>streams) || + (streams<1) || (coupled_streams<0) || (streams>255-coupled_streams) || + (streams+coupled_streams>channels)) + return OPUS_BAD_ARG; + + st->arch = opus_select_arch(); + st->layout.nb_channels = channels; + st->layout.nb_streams = streams; + st->layout.nb_coupled_streams = coupled_streams; + if (mapping_type != MAPPING_TYPE_SURROUND) + st->lfe_stream = -1; + st->bitrate_bps = OPUS_AUTO; + st->application = application; + st->variable_duration = OPUS_FRAMESIZE_ARG; + for (i=0;i<st->layout.nb_channels;i++) + st->layout.mapping[i] = mapping[i]; + if (!validate_layout(&st->layout)) + return OPUS_BAD_ARG; + if (!validate_encoder_layout(&st->layout)) + return OPUS_BAD_ARG; + if (mapping_type == MAPPING_TYPE_AMBISONICS && + !validate_ambisonics(st->layout.nb_channels, NULL, NULL)) + return OPUS_BAD_ARG; + ptr = (char*)st + align(sizeof(OpusMSEncoder)); + coupled_size = opus_encoder_get_size(2); + mono_size = opus_encoder_get_size(1); + + for (i=0;i<st->layout.nb_coupled_streams;i++) + { + ret = opus_encoder_init((OpusEncoder*)ptr, Fs, 2, application); + if(ret!=OPUS_OK)return ret; + if (i==st->lfe_stream) + opus_encoder_ctl((OpusEncoder*)ptr, OPUS_SET_LFE(1)); + ptr += align(coupled_size); + } + for (;i<st->layout.nb_streams;i++) + { + ret = opus_encoder_init((OpusEncoder*)ptr, Fs, 1, application); + if (i==st->lfe_stream) + opus_encoder_ctl((OpusEncoder*)ptr, OPUS_SET_LFE(1)); + if(ret!=OPUS_OK)return ret; + ptr += align(mono_size); + } + if (mapping_type == MAPPING_TYPE_SURROUND) + { + OPUS_CLEAR(ms_get_preemph_mem(st), channels); + OPUS_CLEAR(ms_get_window_mem(st), channels*120); + } + st->mapping_type = mapping_type; + return OPUS_OK; +} + +int opus_multistream_encoder_init( + OpusMSEncoder *st, + opus_int32 Fs, + int channels, + int streams, + int coupled_streams, + const unsigned char *mapping, + int application +) +{ + return opus_multistream_encoder_init_impl(st, Fs, channels, streams, + coupled_streams, mapping, + application, MAPPING_TYPE_NONE); +} + +int opus_multistream_surround_encoder_init( + OpusMSEncoder *st, + opus_int32 Fs, + int channels, + int mapping_family, + int *streams, + int *coupled_streams, + unsigned char *mapping, + int application +) +{ + MappingType mapping_type; + + if ((channels>255) || (channels<1)) + return OPUS_BAD_ARG; + st->lfe_stream = -1; + if (mapping_family==0) + { + if (channels==1) + { + *streams=1; + *coupled_streams=0; + mapping[0]=0; + } else if (channels==2) + { + *streams=1; + *coupled_streams=1; + mapping[0]=0; + mapping[1]=1; + } else + return OPUS_UNIMPLEMENTED; + } else if (mapping_family==1 && channels<=8 && channels>=1) + { + int i; + *streams=vorbis_mappings[channels-1].nb_streams; + *coupled_streams=vorbis_mappings[channels-1].nb_coupled_streams; + for (i=0;i<channels;i++) + mapping[i] = vorbis_mappings[channels-1].mapping[i]; + if (channels>=6) + st->lfe_stream = *streams-1; + } else if (mapping_family==255) + { + int i; + *streams=channels; + *coupled_streams=0; + for(i=0;i<channels;i++) + mapping[i] = i; + } else if (mapping_family==2) + { + int i; + if (!validate_ambisonics(channels, streams, coupled_streams)) + return OPUS_BAD_ARG; + for(i = 0; i < (*streams - *coupled_streams); i++) + mapping[i] = i + (*coupled_streams * 2); + for(i = 0; i < *coupled_streams * 2; i++) + mapping[i + (*streams - *coupled_streams)] = i; + } else + return OPUS_UNIMPLEMENTED; + + if (channels>2 && mapping_family==1) { + mapping_type = MAPPING_TYPE_SURROUND; + } else if (mapping_family==2) + { + mapping_type = MAPPING_TYPE_AMBISONICS; + } else + { + mapping_type = MAPPING_TYPE_NONE; + } + return opus_multistream_encoder_init_impl(st, Fs, channels, *streams, + *coupled_streams, mapping, + application, mapping_type); +} + +OpusMSEncoder *opus_multistream_encoder_create( + opus_int32 Fs, + int channels, + int streams, + int coupled_streams, + const unsigned char *mapping, + int application, + int *error +) +{ + int ret; + OpusMSEncoder *st; + if ((channels>255) || (channels<1) || (coupled_streams>streams) || + (streams<1) || (coupled_streams<0) || (streams>255-coupled_streams) || + (streams+coupled_streams>channels)) + { + if (error) + *error = OPUS_BAD_ARG; + return NULL; + } + st = (OpusMSEncoder *)opus_alloc(opus_multistream_encoder_get_size(streams, coupled_streams)); + if (st==NULL) + { + if (error) + *error = OPUS_ALLOC_FAIL; + return NULL; + } + ret = opus_multistream_encoder_init(st, Fs, channels, streams, coupled_streams, mapping, application); + if (ret != OPUS_OK) + { + opus_free(st); + st = NULL; + } + if (error) + *error = ret; + return st; +} + +OpusMSEncoder *opus_multistream_surround_encoder_create( + opus_int32 Fs, + int channels, + int mapping_family, + int *streams, + int *coupled_streams, + unsigned char *mapping, + int application, + int *error +) +{ + int ret; + opus_int32 size; + OpusMSEncoder *st; + if ((channels>255) || (channels<1)) + { + if (error) + *error = OPUS_BAD_ARG; + return NULL; + } + size = opus_multistream_surround_encoder_get_size(channels, mapping_family); + if (!size) + { + if (error) + *error = OPUS_UNIMPLEMENTED; + return NULL; + } + st = (OpusMSEncoder *)opus_alloc(size); + if (st==NULL) + { + if (error) + *error = OPUS_ALLOC_FAIL; + return NULL; + } + ret = opus_multistream_surround_encoder_init(st, Fs, channels, mapping_family, streams, coupled_streams, mapping, application); + if (ret != OPUS_OK) + { + opus_free(st); + st = NULL; + } + if (error) + *error = ret; + return st; +} + +static void surround_rate_allocation( + OpusMSEncoder *st, + opus_int32 *rate, + int frame_size, + opus_int32 Fs + ) +{ + int i; + opus_int32 channel_rate; + int stream_offset; + int lfe_offset; + int coupled_ratio; /* Q8 */ + int lfe_ratio; /* Q8 */ + int nb_lfe; + int nb_uncoupled; + int nb_coupled; + int nb_normal; + opus_int32 channel_offset; + opus_int32 bitrate; + int total; + + nb_lfe = (st->lfe_stream!=-1); + nb_coupled = st->layout.nb_coupled_streams; + nb_uncoupled = st->layout.nb_streams-nb_coupled-nb_lfe; + nb_normal = 2*nb_coupled + nb_uncoupled; + + /* Give each non-LFE channel enough bits per channel for coding band energy. */ + channel_offset = 40*IMAX(50, Fs/frame_size); + + if (st->bitrate_bps==OPUS_AUTO) + { + bitrate = nb_normal*(channel_offset + Fs + 10000) + 8000*nb_lfe; + } else if (st->bitrate_bps==OPUS_BITRATE_MAX) + { + bitrate = nb_normal*300000 + nb_lfe*128000; + } else { + bitrate = st->bitrate_bps; + } + + /* Give LFE some basic stream_channel allocation but never exceed 1/20 of the + total rate for the non-energy part to avoid problems at really low rate. */ + lfe_offset = IMIN(bitrate/20, 3000) + 15*IMAX(50, Fs/frame_size); + + /* We give each stream (coupled or uncoupled) a starting bitrate. + This models the main saving of coupled channels over uncoupled. */ + stream_offset = (bitrate - channel_offset*nb_normal - lfe_offset*nb_lfe)/nb_normal/2; + stream_offset = IMAX(0, IMIN(20000, stream_offset)); + + /* Coupled streams get twice the mono rate after the offset is allocated. */ + coupled_ratio = 512; + /* Should depend on the bitrate, for now we assume LFE gets 1/8 the bits of mono */ + lfe_ratio = 32; + + total = (nb_uncoupled<<8) /* mono */ + + coupled_ratio*nb_coupled /* stereo */ + + nb_lfe*lfe_ratio; + channel_rate = 256*(opus_int64)(bitrate - lfe_offset*nb_lfe - stream_offset*(nb_coupled+nb_uncoupled) - channel_offset*nb_normal)/total; + + for (i=0;i<st->layout.nb_streams;i++) + { + if (i<st->layout.nb_coupled_streams) + rate[i] = 2*channel_offset + IMAX(0, stream_offset+(channel_rate*coupled_ratio>>8)); + else if (i!=st->lfe_stream) + rate[i] = channel_offset + IMAX(0, stream_offset + channel_rate); + else + rate[i] = IMAX(0, lfe_offset+(channel_rate*lfe_ratio>>8)); + } +} + +static void ambisonics_rate_allocation( + OpusMSEncoder *st, + opus_int32 *rate, + int frame_size, + opus_int32 Fs + ) +{ + int i; + opus_int32 total_rate; + opus_int32 per_stream_rate; + + const int nb_channels = st->layout.nb_streams + st->layout.nb_coupled_streams; + + if (st->bitrate_bps==OPUS_AUTO) + { + total_rate = (st->layout.nb_coupled_streams + st->layout.nb_streams) * + (Fs+60*Fs/frame_size) + st->layout.nb_streams * (opus_int32)15000; + } else if (st->bitrate_bps==OPUS_BITRATE_MAX) + { + total_rate = nb_channels * 320000; + } else + { + total_rate = st->bitrate_bps; + } + + /* Allocate equal number of bits to Ambisonic (uncoupled) and non-diegetic + * (coupled) streams */ + per_stream_rate = total_rate / st->layout.nb_streams; + for (i = 0; i < st->layout.nb_streams; i++) + { + rate[i] = per_stream_rate; + } +} + +static opus_int32 rate_allocation( + OpusMSEncoder *st, + opus_int32 *rate, + int frame_size + ) +{ + int i; + opus_int32 rate_sum=0; + opus_int32 Fs; + char *ptr; + + ptr = (char*)st + align(sizeof(OpusMSEncoder)); + opus_encoder_ctl((OpusEncoder*)ptr, OPUS_GET_SAMPLE_RATE(&Fs)); + + if (st->mapping_type == MAPPING_TYPE_AMBISONICS) { + ambisonics_rate_allocation(st, rate, frame_size, Fs); + } else + { + surround_rate_allocation(st, rate, frame_size, Fs); + } + + for (i=0;i<st->layout.nb_streams;i++) + { + rate[i] = IMAX(rate[i], 500); + rate_sum += rate[i]; + } + return rate_sum; +} + +/* Max size in case the encoder decides to return six frames (6 x 20 ms = 120 ms) */ +#define MS_FRAME_TMP (6*1275+12) +int opus_multistream_encode_native +( + OpusMSEncoder *st, + opus_copy_channel_in_func copy_channel_in, + const void *pcm, + int analysis_frame_size, + unsigned char *data, + opus_int32 max_data_bytes, + int lsb_depth, + downmix_func downmix, + int float_api, + void *user_data +) +{ + opus_int32 Fs; + int coupled_size; + int mono_size; + int s; + char *ptr; + int tot_size; + VARDECL(opus_val16, buf); + VARDECL(opus_val16, bandSMR); + unsigned char tmp_data[MS_FRAME_TMP]; + OpusRepacketizer rp; + opus_int32 vbr; + const CELTMode *celt_mode; + opus_int32 bitrates[256]; + opus_val16 bandLogE[42]; + opus_val32 *mem = NULL; + opus_val32 *preemph_mem=NULL; + int frame_size; + opus_int32 rate_sum; + opus_int32 smallest_packet; + ALLOC_STACK; + + if (st->mapping_type == MAPPING_TYPE_SURROUND) + { + preemph_mem = ms_get_preemph_mem(st); + mem = ms_get_window_mem(st); + } + + ptr = (char*)st + align(sizeof(OpusMSEncoder)); + opus_encoder_ctl((OpusEncoder*)ptr, OPUS_GET_SAMPLE_RATE(&Fs)); + opus_encoder_ctl((OpusEncoder*)ptr, OPUS_GET_VBR(&vbr)); + opus_encoder_ctl((OpusEncoder*)ptr, CELT_GET_MODE(&celt_mode)); + + frame_size = frame_size_select(analysis_frame_size, st->variable_duration, Fs); + if (frame_size <= 0) + { + RESTORE_STACK; + return OPUS_BAD_ARG; + } + + /* Smallest packet the encoder can produce. */ + smallest_packet = st->layout.nb_streams*2-1; + /* 100 ms needs an extra byte per stream for the ToC. */ + if (Fs/frame_size == 10) + smallest_packet += st->layout.nb_streams; + if (max_data_bytes < smallest_packet) + { + RESTORE_STACK; + return OPUS_BUFFER_TOO_SMALL; + } + ALLOC(buf, 2*frame_size, opus_val16); + coupled_size = opus_encoder_get_size(2); + mono_size = opus_encoder_get_size(1); + + ALLOC(bandSMR, 21*st->layout.nb_channels, opus_val16); + if (st->mapping_type == MAPPING_TYPE_SURROUND) + { + surround_analysis(celt_mode, pcm, bandSMR, mem, preemph_mem, frame_size, 120, st->layout.nb_channels, Fs, copy_channel_in, st->arch); + } + + /* Compute bitrate allocation between streams (this could be a lot better) */ + rate_sum = rate_allocation(st, bitrates, frame_size); + + if (!vbr) + { + if (st->bitrate_bps == OPUS_AUTO) + { + max_data_bytes = IMIN(max_data_bytes, 3*rate_sum/(3*8*Fs/frame_size)); + } else if (st->bitrate_bps != OPUS_BITRATE_MAX) + { + max_data_bytes = IMIN(max_data_bytes, IMAX(smallest_packet, + 3*st->bitrate_bps/(3*8*Fs/frame_size))); + } + } + ptr = (char*)st + align(sizeof(OpusMSEncoder)); + for (s=0;s<st->layout.nb_streams;s++) + { + OpusEncoder *enc; + enc = (OpusEncoder*)ptr; + if (s < st->layout.nb_coupled_streams) + ptr += align(coupled_size); + else + ptr += align(mono_size); + opus_encoder_ctl(enc, OPUS_SET_BITRATE(bitrates[s])); + if (st->mapping_type == MAPPING_TYPE_SURROUND) + { + opus_int32 equiv_rate; + equiv_rate = st->bitrate_bps; + if (frame_size*50 < Fs) + equiv_rate -= 60*(Fs/frame_size - 50)*st->layout.nb_channels; + if (equiv_rate > 10000*st->layout.nb_channels) + opus_encoder_ctl(enc, OPUS_SET_BANDWIDTH(OPUS_BANDWIDTH_FULLBAND)); + else if (equiv_rate > 7000*st->layout.nb_channels) + opus_encoder_ctl(enc, OPUS_SET_BANDWIDTH(OPUS_BANDWIDTH_SUPERWIDEBAND)); + else if (equiv_rate > 5000*st->layout.nb_channels) + opus_encoder_ctl(enc, OPUS_SET_BANDWIDTH(OPUS_BANDWIDTH_WIDEBAND)); + else + opus_encoder_ctl(enc, OPUS_SET_BANDWIDTH(OPUS_BANDWIDTH_NARROWBAND)); + if (s < st->layout.nb_coupled_streams) + { + /* To preserve the spatial image, force stereo CELT on coupled streams */ + opus_encoder_ctl(enc, OPUS_SET_FORCE_MODE(MODE_CELT_ONLY)); + opus_encoder_ctl(enc, OPUS_SET_FORCE_CHANNELS(2)); + } + } + else if (st->mapping_type == MAPPING_TYPE_AMBISONICS) { + opus_encoder_ctl(enc, OPUS_SET_FORCE_MODE(MODE_CELT_ONLY)); + } + } + + ptr = (char*)st + align(sizeof(OpusMSEncoder)); + /* Counting ToC */ + tot_size = 0; + for (s=0;s<st->layout.nb_streams;s++) + { + OpusEncoder *enc; + int len; + int curr_max; + int c1, c2; + int ret; + + opus_repacketizer_init(&rp); + enc = (OpusEncoder*)ptr; + if (s < st->layout.nb_coupled_streams) + { + int i; + int left, right; + left = get_left_channel(&st->layout, s, -1); + right = get_right_channel(&st->layout, s, -1); + (*copy_channel_in)(buf, 2, + pcm, st->layout.nb_channels, left, frame_size, user_data); + (*copy_channel_in)(buf+1, 2, + pcm, st->layout.nb_channels, right, frame_size, user_data); + ptr += align(coupled_size); + if (st->mapping_type == MAPPING_TYPE_SURROUND) + { + for (i=0;i<21;i++) + { + bandLogE[i] = bandSMR[21*left+i]; + bandLogE[21+i] = bandSMR[21*right+i]; + } + } + c1 = left; + c2 = right; + } else { + int i; + int chan = get_mono_channel(&st->layout, s, -1); + (*copy_channel_in)(buf, 1, + pcm, st->layout.nb_channels, chan, frame_size, user_data); + ptr += align(mono_size); + if (st->mapping_type == MAPPING_TYPE_SURROUND) + { + for (i=0;i<21;i++) + bandLogE[i] = bandSMR[21*chan+i]; + } + c1 = chan; + c2 = -1; + } + if (st->mapping_type == MAPPING_TYPE_SURROUND) + opus_encoder_ctl(enc, OPUS_SET_ENERGY_MASK(bandLogE)); + /* number of bytes left (+Toc) */ + curr_max = max_data_bytes - tot_size; + /* Reserve one byte for the last stream and two for the others */ + curr_max -= IMAX(0,2*(st->layout.nb_streams-s-1)-1); + /* For 100 ms, reserve an extra byte per stream for the ToC */ + if (Fs/frame_size == 10) + curr_max -= st->layout.nb_streams-s-1; + curr_max = IMIN(curr_max,MS_FRAME_TMP); + /* Repacketizer will add one or two bytes for self-delimited frames */ + if (s != st->layout.nb_streams-1) curr_max -= curr_max>253 ? 2 : 1; + if (!vbr && s == st->layout.nb_streams-1) + opus_encoder_ctl(enc, OPUS_SET_BITRATE(curr_max*(8*Fs/frame_size))); + len = opus_encode_native(enc, buf, frame_size, tmp_data, curr_max, lsb_depth, + pcm, analysis_frame_size, c1, c2, st->layout.nb_channels, downmix, float_api); + if (len<0) + { + RESTORE_STACK; + return len; + } + /* We need to use the repacketizer to add the self-delimiting lengths + while taking into account the fact that the encoder can now return + more than one frame at a time (e.g. 60 ms CELT-only) */ + ret = opus_repacketizer_cat(&rp, tmp_data, len); + /* If the opus_repacketizer_cat() fails, then something's seriously wrong + with the encoder. */ + if (ret != OPUS_OK) + { + RESTORE_STACK; + return OPUS_INTERNAL_ERROR; + } + len = opus_repacketizer_out_range_impl(&rp, 0, opus_repacketizer_get_nb_frames(&rp), + data, max_data_bytes-tot_size, s != st->layout.nb_streams-1, !vbr && s == st->layout.nb_streams-1); + data += len; + tot_size += len; + } + /*printf("\n");*/ + RESTORE_STACK; + return tot_size; +} + +#if !defined(DISABLE_FLOAT_API) +static void opus_copy_channel_in_float( + opus_val16 *dst, + int dst_stride, + const void *src, + int src_stride, + int src_channel, + int frame_size, + void *user_data +) +{ + const float *float_src; + opus_int32 i; + (void)user_data; + float_src = (const float *)src; + for (i=0;i<frame_size;i++) +#if defined(FIXED_POINT) + dst[i*dst_stride] = FLOAT2INT16(float_src[i*src_stride+src_channel]); +#else + dst[i*dst_stride] = float_src[i*src_stride+src_channel]; +#endif +} +#endif + +static void opus_copy_channel_in_short( + opus_val16 *dst, + int dst_stride, + const void *src, + int src_stride, + int src_channel, + int frame_size, + void *user_data +) +{ + const opus_int16 *short_src; + opus_int32 i; + (void)user_data; + short_src = (const opus_int16 *)src; + for (i=0;i<frame_size;i++) +#if defined(FIXED_POINT) + dst[i*dst_stride] = short_src[i*src_stride+src_channel]; +#else + dst[i*dst_stride] = (1/32768.f)*short_src[i*src_stride+src_channel]; +#endif +} + + +#ifdef FIXED_POINT +int opus_multistream_encode( + OpusMSEncoder *st, + const opus_val16 *pcm, + int frame_size, + unsigned char *data, + opus_int32 max_data_bytes +) +{ + return opus_multistream_encode_native(st, opus_copy_channel_in_short, + pcm, frame_size, data, max_data_bytes, 16, downmix_int, 0, NULL); +} + +#ifndef DISABLE_FLOAT_API +int opus_multistream_encode_float( + OpusMSEncoder *st, + const float *pcm, + int frame_size, + unsigned char *data, + opus_int32 max_data_bytes +) +{ + return opus_multistream_encode_native(st, opus_copy_channel_in_float, + pcm, frame_size, data, max_data_bytes, 16, downmix_float, 1, NULL); +} +#endif + +#else + +int opus_multistream_encode_float +( + OpusMSEncoder *st, + const opus_val16 *pcm, + int frame_size, + unsigned char *data, + opus_int32 max_data_bytes +) +{ + return opus_multistream_encode_native(st, opus_copy_channel_in_float, + pcm, frame_size, data, max_data_bytes, 24, downmix_float, 1, NULL); +} + +int opus_multistream_encode( + OpusMSEncoder *st, + const opus_int16 *pcm, + int frame_size, + unsigned char *data, + opus_int32 max_data_bytes +) +{ + return opus_multistream_encode_native(st, opus_copy_channel_in_short, + pcm, frame_size, data, max_data_bytes, 16, downmix_int, 0, NULL); +} +#endif + +int opus_multistream_encoder_ctl_va_list(OpusMSEncoder *st, int request, + va_list ap) +{ + int coupled_size, mono_size; + char *ptr; + int ret = OPUS_OK; + + coupled_size = opus_encoder_get_size(2); + mono_size = opus_encoder_get_size(1); + ptr = (char*)st + align(sizeof(OpusMSEncoder)); + switch (request) + { + case OPUS_SET_BITRATE_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + if (value != OPUS_AUTO && value != OPUS_BITRATE_MAX) + { + if (value <= 0) + goto bad_arg; + value = IMIN(300000*st->layout.nb_channels, IMAX(500*st->layout.nb_channels, value)); + } + st->bitrate_bps = value; + } + break; + case OPUS_GET_BITRATE_REQUEST: + { + int s; + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = 0; + for (s=0;s<st->layout.nb_streams;s++) + { + opus_int32 rate; + OpusEncoder *enc; + enc = (OpusEncoder*)ptr; + if (s < st->layout.nb_coupled_streams) + ptr += align(coupled_size); + else + ptr += align(mono_size); + opus_encoder_ctl(enc, request, &rate); + *value += rate; + } + } + break; + case OPUS_GET_LSB_DEPTH_REQUEST: + case OPUS_GET_VBR_REQUEST: + case OPUS_GET_APPLICATION_REQUEST: + case OPUS_GET_BANDWIDTH_REQUEST: + case OPUS_GET_COMPLEXITY_REQUEST: + case OPUS_GET_PACKET_LOSS_PERC_REQUEST: + case OPUS_GET_DTX_REQUEST: + case OPUS_GET_VOICE_RATIO_REQUEST: + case OPUS_GET_VBR_CONSTRAINT_REQUEST: + case OPUS_GET_SIGNAL_REQUEST: + case OPUS_GET_LOOKAHEAD_REQUEST: + case OPUS_GET_SAMPLE_RATE_REQUEST: + case OPUS_GET_INBAND_FEC_REQUEST: + case OPUS_GET_FORCE_CHANNELS_REQUEST: + case OPUS_GET_PREDICTION_DISABLED_REQUEST: + case OPUS_GET_PHASE_INVERSION_DISABLED_REQUEST: + { + OpusEncoder *enc; + /* For int32* GET params, just query the first stream */ + opus_int32 *value = va_arg(ap, opus_int32*); + enc = (OpusEncoder*)ptr; + ret = opus_encoder_ctl(enc, request, value); + } + break; + case OPUS_GET_FINAL_RANGE_REQUEST: + { + int s; + opus_uint32 *value = va_arg(ap, opus_uint32*); + opus_uint32 tmp; + if (!value) + { + goto bad_arg; + } + *value=0; + for (s=0;s<st->layout.nb_streams;s++) + { + OpusEncoder *enc; + enc = (OpusEncoder*)ptr; + if (s < st->layout.nb_coupled_streams) + ptr += align(coupled_size); + else + ptr += align(mono_size); + ret = opus_encoder_ctl(enc, request, &tmp); + if (ret != OPUS_OK) break; + *value ^= tmp; + } + } + break; + case OPUS_SET_LSB_DEPTH_REQUEST: + case OPUS_SET_COMPLEXITY_REQUEST: + case OPUS_SET_VBR_REQUEST: + case OPUS_SET_VBR_CONSTRAINT_REQUEST: + case OPUS_SET_MAX_BANDWIDTH_REQUEST: + case OPUS_SET_BANDWIDTH_REQUEST: + case OPUS_SET_SIGNAL_REQUEST: + case OPUS_SET_APPLICATION_REQUEST: + case OPUS_SET_INBAND_FEC_REQUEST: + case OPUS_SET_PACKET_LOSS_PERC_REQUEST: + case OPUS_SET_DTX_REQUEST: + case OPUS_SET_FORCE_MODE_REQUEST: + case OPUS_SET_FORCE_CHANNELS_REQUEST: + case OPUS_SET_PREDICTION_DISABLED_REQUEST: + case OPUS_SET_PHASE_INVERSION_DISABLED_REQUEST: + { + int s; + /* This works for int32 params */ + opus_int32 value = va_arg(ap, opus_int32); + for (s=0;s<st->layout.nb_streams;s++) + { + OpusEncoder *enc; + + enc = (OpusEncoder*)ptr; + if (s < st->layout.nb_coupled_streams) + ptr += align(coupled_size); + else + ptr += align(mono_size); + ret = opus_encoder_ctl(enc, request, value); + if (ret != OPUS_OK) + break; + } + } + break; + case OPUS_MULTISTREAM_GET_ENCODER_STATE_REQUEST: + { + int s; + opus_int32 stream_id; + OpusEncoder **value; + stream_id = va_arg(ap, opus_int32); + if (stream_id<0 || stream_id >= st->layout.nb_streams) + goto bad_arg; + value = va_arg(ap, OpusEncoder**); + if (!value) + { + goto bad_arg; + } + for (s=0;s<stream_id;s++) + { + if (s < st->layout.nb_coupled_streams) + ptr += align(coupled_size); + else + ptr += align(mono_size); + } + *value = (OpusEncoder*)ptr; + } + break; + case OPUS_SET_EXPERT_FRAME_DURATION_REQUEST: + { + opus_int32 value = va_arg(ap, opus_int32); + st->variable_duration = value; + } + break; + case OPUS_GET_EXPERT_FRAME_DURATION_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = st->variable_duration; + } + break; + case OPUS_RESET_STATE: + { + int s; + if (st->mapping_type == MAPPING_TYPE_SURROUND) + { + OPUS_CLEAR(ms_get_preemph_mem(st), st->layout.nb_channels); + OPUS_CLEAR(ms_get_window_mem(st), st->layout.nb_channels*120); + } + for (s=0;s<st->layout.nb_streams;s++) + { + OpusEncoder *enc; + enc = (OpusEncoder*)ptr; + if (s < st->layout.nb_coupled_streams) + ptr += align(coupled_size); + else + ptr += align(mono_size); + ret = opus_encoder_ctl(enc, OPUS_RESET_STATE); + if (ret != OPUS_OK) + break; + } + } + break; + default: + ret = OPUS_UNIMPLEMENTED; + break; + } + return ret; +bad_arg: + return OPUS_BAD_ARG; +} + +int opus_multistream_encoder_ctl(OpusMSEncoder *st, int request, ...) +{ + int ret; + va_list ap; + va_start(ap, request); + ret = opus_multistream_encoder_ctl_va_list(st, request, ap); + va_end(ap); + return ret; +} + +void opus_multistream_encoder_destroy(OpusMSEncoder *st) +{ + opus_free(st); +} diff --git a/media/libopus/src/opus_private.h b/media/libopus/src/opus_private.h new file mode 100644 index 0000000000..5e2463f546 --- /dev/null +++ b/media/libopus/src/opus_private.h @@ -0,0 +1,201 @@ +/* Copyright (c) 2012 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + + +#ifndef OPUS_PRIVATE_H +#define OPUS_PRIVATE_H + +#include "arch.h" +#include "opus.h" +#include "celt.h" + +#include <stdarg.h> /* va_list */ +#include <stddef.h> /* offsetof */ + +struct OpusRepacketizer { + unsigned char toc; + int nb_frames; + const unsigned char *frames[48]; + opus_int16 len[48]; + int framesize; +}; + +typedef struct ChannelLayout { + int nb_channels; + int nb_streams; + int nb_coupled_streams; + unsigned char mapping[256]; +} ChannelLayout; + +typedef enum { + MAPPING_TYPE_NONE, + MAPPING_TYPE_SURROUND, + MAPPING_TYPE_AMBISONICS +} MappingType; + +struct OpusMSEncoder { + ChannelLayout layout; + int arch; + int lfe_stream; + int application; + int variable_duration; + MappingType mapping_type; + opus_int32 bitrate_bps; + /* Encoder states go here */ + /* then opus_val32 window_mem[channels*120]; */ + /* then opus_val32 preemph_mem[channels]; */ +}; + +struct OpusMSDecoder { + ChannelLayout layout; + /* Decoder states go here */ +}; + +int opus_multistream_encoder_ctl_va_list(struct OpusMSEncoder *st, int request, + va_list ap); +int opus_multistream_decoder_ctl_va_list(struct OpusMSDecoder *st, int request, + va_list ap); + +int validate_layout(const ChannelLayout *layout); +int get_left_channel(const ChannelLayout *layout, int stream_id, int prev); +int get_right_channel(const ChannelLayout *layout, int stream_id, int prev); +int get_mono_channel(const ChannelLayout *layout, int stream_id, int prev); + +typedef void (*opus_copy_channel_in_func)( + opus_val16 *dst, + int dst_stride, + const void *src, + int src_stride, + int src_channel, + int frame_size, + void *user_data +); + +typedef void (*opus_copy_channel_out_func)( + void *dst, + int dst_stride, + int dst_channel, + const opus_val16 *src, + int src_stride, + int frame_size, + void *user_data +); + +#define MODE_SILK_ONLY 1000 +#define MODE_HYBRID 1001 +#define MODE_CELT_ONLY 1002 + +#define OPUS_SET_VOICE_RATIO_REQUEST 11018 +#define OPUS_GET_VOICE_RATIO_REQUEST 11019 + +/** Configures the encoder's expected percentage of voice + * opposed to music or other signals. + * + * @note This interface is currently more aspiration than actuality. It's + * ultimately expected to bias an automatic signal classifier, but it currently + * just shifts the static bitrate to mode mapping around a little bit. + * + * @param[in] x <tt>int</tt>: Voice percentage in the range 0-100, inclusive. + * @hideinitializer */ +#define OPUS_SET_VOICE_RATIO(x) OPUS_SET_VOICE_RATIO_REQUEST, __opus_check_int(x) +/** Gets the encoder's configured voice ratio value, @see OPUS_SET_VOICE_RATIO + * + * @param[out] x <tt>int*</tt>: Voice percentage in the range 0-100, inclusive. + * @hideinitializer */ +#define OPUS_GET_VOICE_RATIO(x) OPUS_GET_VOICE_RATIO_REQUEST, __opus_check_int_ptr(x) + + +#define OPUS_SET_FORCE_MODE_REQUEST 11002 +#define OPUS_SET_FORCE_MODE(x) OPUS_SET_FORCE_MODE_REQUEST, __opus_check_int(x) + +typedef void (*downmix_func)(const void *, opus_val32 *, int, int, int, int, int); +void downmix_float(const void *_x, opus_val32 *sub, int subframe, int offset, int c1, int c2, int C); +void downmix_int(const void *_x, opus_val32 *sub, int subframe, int offset, int c1, int c2, int C); +int is_digital_silence(const opus_val16* pcm, int frame_size, int channels, int lsb_depth); + +int encode_size(int size, unsigned char *data); + +opus_int32 frame_size_select(opus_int32 frame_size, int variable_duration, opus_int32 Fs); + +opus_int32 opus_encode_native(OpusEncoder *st, const opus_val16 *pcm, int frame_size, + unsigned char *data, opus_int32 out_data_bytes, int lsb_depth, + const void *analysis_pcm, opus_int32 analysis_size, int c1, int c2, + int analysis_channels, downmix_func downmix, int float_api); + +int opus_decode_native(OpusDecoder *st, const unsigned char *data, opus_int32 len, + opus_val16 *pcm, int frame_size, int decode_fec, int self_delimited, + opus_int32 *packet_offset, int soft_clip); + +/* Make sure everything is properly aligned. */ +static OPUS_INLINE int align(int i) +{ + struct foo {char c; union { void* p; opus_int32 i; opus_val32 v; } u;}; + + unsigned int alignment = offsetof(struct foo, u); + + /* Optimizing compilers should optimize div and multiply into and + for all sensible alignment values. */ + return ((i + alignment - 1) / alignment) * alignment; +} + +int opus_packet_parse_impl(const unsigned char *data, opus_int32 len, + int self_delimited, unsigned char *out_toc, + const unsigned char *frames[48], opus_int16 size[48], + int *payload_offset, opus_int32 *packet_offset); + +opus_int32 opus_repacketizer_out_range_impl(OpusRepacketizer *rp, int begin, int end, + unsigned char *data, opus_int32 maxlen, int self_delimited, int pad); + +int pad_frame(unsigned char *data, opus_int32 len, opus_int32 new_len); + +int opus_multistream_encode_native +( + struct OpusMSEncoder *st, + opus_copy_channel_in_func copy_channel_in, + const void *pcm, + int analysis_frame_size, + unsigned char *data, + opus_int32 max_data_bytes, + int lsb_depth, + downmix_func downmix, + int float_api, + void *user_data +); + +int opus_multistream_decode_native( + struct OpusMSDecoder *st, + const unsigned char *data, + opus_int32 len, + void *pcm, + opus_copy_channel_out_func copy_channel_out, + int frame_size, + int decode_fec, + int soft_clip, + void *user_data +); + +#endif /* OPUS_PRIVATE_H */ diff --git a/media/libopus/src/opus_projection_decoder.c b/media/libopus/src/opus_projection_decoder.c new file mode 100644 index 0000000000..c2e07d5bcf --- /dev/null +++ b/media/libopus/src/opus_projection_decoder.c @@ -0,0 +1,258 @@ +/* Copyright (c) 2017 Google Inc. + Written by Andrew Allen */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "mathops.h" +#include "os_support.h" +#include "opus_private.h" +#include "opus_defines.h" +#include "opus_projection.h" +#include "opus_multistream.h" +#include "mapping_matrix.h" +#include "stack_alloc.h" + +struct OpusProjectionDecoder +{ + opus_int32 demixing_matrix_size_in_bytes; + /* Encoder states go here */ +}; + +#if !defined(DISABLE_FLOAT_API) +static void opus_projection_copy_channel_out_float( + void *dst, + int dst_stride, + int dst_channel, + const opus_val16 *src, + int src_stride, + int frame_size, + void *user_data) +{ + float *float_dst; + const MappingMatrix *matrix; + float_dst = (float *)dst; + matrix = (const MappingMatrix *)user_data; + + if (dst_channel == 0) + OPUS_CLEAR(float_dst, frame_size * dst_stride); + + if (src != NULL) + mapping_matrix_multiply_channel_out_float(matrix, src, dst_channel, + src_stride, float_dst, dst_stride, frame_size); +} +#endif + +static void opus_projection_copy_channel_out_short( + void *dst, + int dst_stride, + int dst_channel, + const opus_val16 *src, + int src_stride, + int frame_size, + void *user_data) +{ + opus_int16 *short_dst; + const MappingMatrix *matrix; + short_dst = (opus_int16 *)dst; + matrix = (const MappingMatrix *)user_data; + if (dst_channel == 0) + OPUS_CLEAR(short_dst, frame_size * dst_stride); + + if (src != NULL) + mapping_matrix_multiply_channel_out_short(matrix, src, dst_channel, + src_stride, short_dst, dst_stride, frame_size); +} + +static MappingMatrix *get_dec_demixing_matrix(OpusProjectionDecoder *st) +{ + /* void* cast avoids clang -Wcast-align warning */ + return (MappingMatrix*)(void*)((char*)st + + align(sizeof(OpusProjectionDecoder))); +} + +static OpusMSDecoder *get_multistream_decoder(OpusProjectionDecoder *st) +{ + /* void* cast avoids clang -Wcast-align warning */ + return (OpusMSDecoder*)(void*)((char*)st + + align(sizeof(OpusProjectionDecoder) + + st->demixing_matrix_size_in_bytes)); +} + +opus_int32 opus_projection_decoder_get_size(int channels, int streams, + int coupled_streams) +{ + opus_int32 matrix_size; + opus_int32 decoder_size; + + matrix_size = + mapping_matrix_get_size(streams + coupled_streams, channels); + if (!matrix_size) + return 0; + + decoder_size = opus_multistream_decoder_get_size(streams, coupled_streams); + if (!decoder_size) + return 0; + + return align(sizeof(OpusProjectionDecoder)) + matrix_size + decoder_size; +} + +int opus_projection_decoder_init(OpusProjectionDecoder *st, opus_int32 Fs, + int channels, int streams, int coupled_streams, + unsigned char *demixing_matrix, opus_int32 demixing_matrix_size) +{ + int nb_input_streams; + opus_int32 expected_matrix_size; + int i, ret; + unsigned char mapping[255]; + VARDECL(opus_int16, buf); + ALLOC_STACK; + + /* Verify supplied matrix size. */ + nb_input_streams = streams + coupled_streams; + expected_matrix_size = nb_input_streams * channels * sizeof(opus_int16); + if (expected_matrix_size != demixing_matrix_size) + { + RESTORE_STACK; + return OPUS_BAD_ARG; + } + + /* Convert demixing matrix input into internal format. */ + ALLOC(buf, nb_input_streams * channels, opus_int16); + for (i = 0; i < nb_input_streams * channels; i++) + { + int s = demixing_matrix[2*i + 1] << 8 | demixing_matrix[2*i]; + s = ((s & 0xFFFF) ^ 0x8000) - 0x8000; + buf[i] = (opus_int16)s; + } + + /* Assign demixing matrix. */ + st->demixing_matrix_size_in_bytes = + mapping_matrix_get_size(channels, nb_input_streams); + if (!st->demixing_matrix_size_in_bytes) + { + RESTORE_STACK; + return OPUS_BAD_ARG; + } + + mapping_matrix_init(get_dec_demixing_matrix(st), channels, nb_input_streams, 0, + buf, demixing_matrix_size); + + /* Set trivial mapping so each input channel pairs with a matrix column. */ + for (i = 0; i < channels; i++) + mapping[i] = i; + + ret = opus_multistream_decoder_init( + get_multistream_decoder(st), Fs, channels, streams, coupled_streams, mapping); + RESTORE_STACK; + return ret; +} + +OpusProjectionDecoder *opus_projection_decoder_create( + opus_int32 Fs, int channels, int streams, int coupled_streams, + unsigned char *demixing_matrix, opus_int32 demixing_matrix_size, int *error) +{ + int size; + int ret; + OpusProjectionDecoder *st; + + /* Allocate space for the projection decoder. */ + size = opus_projection_decoder_get_size(channels, streams, coupled_streams); + if (!size) { + if (error) + *error = OPUS_ALLOC_FAIL; + return NULL; + } + st = (OpusProjectionDecoder *)opus_alloc(size); + if (!st) + { + if (error) + *error = OPUS_ALLOC_FAIL; + return NULL; + } + + /* Initialize projection decoder with provided settings. */ + ret = opus_projection_decoder_init(st, Fs, channels, streams, coupled_streams, + demixing_matrix, demixing_matrix_size); + if (ret != OPUS_OK) + { + opus_free(st); + st = NULL; + } + if (error) + *error = ret; + return st; +} + +#ifdef FIXED_POINT +int opus_projection_decode(OpusProjectionDecoder *st, const unsigned char *data, + opus_int32 len, opus_int16 *pcm, int frame_size, + int decode_fec) +{ + return opus_multistream_decode_native(get_multistream_decoder(st), data, len, + pcm, opus_projection_copy_channel_out_short, frame_size, decode_fec, 0, + get_dec_demixing_matrix(st)); +} +#else +int opus_projection_decode(OpusProjectionDecoder *st, const unsigned char *data, + opus_int32 len, opus_int16 *pcm, int frame_size, + int decode_fec) +{ + return opus_multistream_decode_native(get_multistream_decoder(st), data, len, + pcm, opus_projection_copy_channel_out_short, frame_size, decode_fec, 1, + get_dec_demixing_matrix(st)); +} +#endif + +#ifndef DISABLE_FLOAT_API +int opus_projection_decode_float(OpusProjectionDecoder *st, const unsigned char *data, + opus_int32 len, float *pcm, int frame_size, int decode_fec) +{ + return opus_multistream_decode_native(get_multistream_decoder(st), data, len, + pcm, opus_projection_copy_channel_out_float, frame_size, decode_fec, 0, + get_dec_demixing_matrix(st)); +} +#endif + +int opus_projection_decoder_ctl(OpusProjectionDecoder *st, int request, ...) +{ + va_list ap; + int ret = OPUS_OK; + + va_start(ap, request); + ret = opus_multistream_decoder_ctl_va_list(get_multistream_decoder(st), + request, ap); + va_end(ap); + return ret; +} + +void opus_projection_decoder_destroy(OpusProjectionDecoder *st) +{ + opus_free(st); +} + diff --git a/media/libopus/src/opus_projection_encoder.c b/media/libopus/src/opus_projection_encoder.c new file mode 100644 index 0000000000..06fb2d2526 --- /dev/null +++ b/media/libopus/src/opus_projection_encoder.c @@ -0,0 +1,468 @@ +/* Copyright (c) 2017 Google Inc. + Written by Andrew Allen */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "mathops.h" +#include "os_support.h" +#include "opus_private.h" +#include "opus_defines.h" +#include "opus_projection.h" +#include "opus_multistream.h" +#include "stack_alloc.h" +#include "mapping_matrix.h" + +struct OpusProjectionEncoder +{ + opus_int32 mixing_matrix_size_in_bytes; + opus_int32 demixing_matrix_size_in_bytes; + /* Encoder states go here */ +}; + +#if !defined(DISABLE_FLOAT_API) +static void opus_projection_copy_channel_in_float( + opus_val16 *dst, + int dst_stride, + const void *src, + int src_stride, + int src_channel, + int frame_size, + void *user_data +) +{ + mapping_matrix_multiply_channel_in_float((const MappingMatrix*)user_data, + (const float*)src, src_stride, dst, src_channel, dst_stride, frame_size); +} +#endif + +static void opus_projection_copy_channel_in_short( + opus_val16 *dst, + int dst_stride, + const void *src, + int src_stride, + int src_channel, + int frame_size, + void *user_data +) +{ + mapping_matrix_multiply_channel_in_short((const MappingMatrix*)user_data, + (const opus_int16*)src, src_stride, dst, src_channel, dst_stride, frame_size); +} + +static int get_order_plus_one_from_channels(int channels, int *order_plus_one) +{ + int order_plus_one_; + int acn_channels; + int nondiegetic_channels; + + /* Allowed numbers of channels: + * (1 + n)^2 + 2j, for n = 0...14 and j = 0 or 1. + */ + if (channels < 1 || channels > 227) + return OPUS_BAD_ARG; + + order_plus_one_ = isqrt32(channels); + acn_channels = order_plus_one_ * order_plus_one_; + nondiegetic_channels = channels - acn_channels; + if (nondiegetic_channels != 0 && nondiegetic_channels != 2) + return OPUS_BAD_ARG; + + if (order_plus_one) + *order_plus_one = order_plus_one_; + return OPUS_OK; +} + +static int get_streams_from_channels(int channels, int mapping_family, + int *streams, int *coupled_streams, + int *order_plus_one) +{ + if (mapping_family == 3) + { + if (get_order_plus_one_from_channels(channels, order_plus_one) != OPUS_OK) + return OPUS_BAD_ARG; + if (streams) + *streams = (channels + 1) / 2; + if (coupled_streams) + *coupled_streams = channels / 2; + return OPUS_OK; + } + return OPUS_BAD_ARG; +} + +static MappingMatrix *get_mixing_matrix(OpusProjectionEncoder *st) +{ + /* void* cast avoids clang -Wcast-align warning */ + return (MappingMatrix *)(void*)((char*)st + + align(sizeof(OpusProjectionEncoder))); +} + +static MappingMatrix *get_enc_demixing_matrix(OpusProjectionEncoder *st) +{ + /* void* cast avoids clang -Wcast-align warning */ + return (MappingMatrix *)(void*)((char*)st + + align(sizeof(OpusProjectionEncoder) + + st->mixing_matrix_size_in_bytes)); +} + +static OpusMSEncoder *get_multistream_encoder(OpusProjectionEncoder *st) +{ + /* void* cast avoids clang -Wcast-align warning */ + return (OpusMSEncoder *)(void*)((char*)st + + align(sizeof(OpusProjectionEncoder) + + st->mixing_matrix_size_in_bytes + + st->demixing_matrix_size_in_bytes)); +} + +opus_int32 opus_projection_ambisonics_encoder_get_size(int channels, + int mapping_family) +{ + int nb_streams; + int nb_coupled_streams; + int order_plus_one; + int mixing_matrix_rows, mixing_matrix_cols; + int demixing_matrix_rows, demixing_matrix_cols; + opus_int32 mixing_matrix_size, demixing_matrix_size; + opus_int32 encoder_size; + int ret; + + ret = get_streams_from_channels(channels, mapping_family, &nb_streams, + &nb_coupled_streams, &order_plus_one); + if (ret != OPUS_OK) + return 0; + + if (order_plus_one == 2) + { + mixing_matrix_rows = mapping_matrix_foa_mixing.rows; + mixing_matrix_cols = mapping_matrix_foa_mixing.cols; + demixing_matrix_rows = mapping_matrix_foa_demixing.rows; + demixing_matrix_cols = mapping_matrix_foa_demixing.cols; + } + else if (order_plus_one == 3) + { + mixing_matrix_rows = mapping_matrix_soa_mixing.rows; + mixing_matrix_cols = mapping_matrix_soa_mixing.cols; + demixing_matrix_rows = mapping_matrix_soa_demixing.rows; + demixing_matrix_cols = mapping_matrix_soa_demixing.cols; + } + else if (order_plus_one == 4) + { + mixing_matrix_rows = mapping_matrix_toa_mixing.rows; + mixing_matrix_cols = mapping_matrix_toa_mixing.cols; + demixing_matrix_rows = mapping_matrix_toa_demixing.rows; + demixing_matrix_cols = mapping_matrix_toa_demixing.cols; + } + else + return 0; + + mixing_matrix_size = + mapping_matrix_get_size(mixing_matrix_rows, mixing_matrix_cols); + if (!mixing_matrix_size) + return 0; + + demixing_matrix_size = + mapping_matrix_get_size(demixing_matrix_rows, demixing_matrix_cols); + if (!demixing_matrix_size) + return 0; + + encoder_size = + opus_multistream_encoder_get_size(nb_streams, nb_coupled_streams); + if (!encoder_size) + return 0; + + return align(sizeof(OpusProjectionEncoder)) + + mixing_matrix_size + demixing_matrix_size + encoder_size; +} + +int opus_projection_ambisonics_encoder_init(OpusProjectionEncoder *st, opus_int32 Fs, + int channels, int mapping_family, + int *streams, int *coupled_streams, + int application) +{ + MappingMatrix *mixing_matrix; + MappingMatrix *demixing_matrix; + OpusMSEncoder *ms_encoder; + int i; + int ret; + int order_plus_one; + unsigned char mapping[255]; + + if (streams == NULL || coupled_streams == NULL) { + return OPUS_BAD_ARG; + } + + if (get_streams_from_channels(channels, mapping_family, streams, + coupled_streams, &order_plus_one) != OPUS_OK) + return OPUS_BAD_ARG; + + if (mapping_family == 3) + { + /* Assign mixing matrix based on available pre-computed matrices. */ + mixing_matrix = get_mixing_matrix(st); + if (order_plus_one == 2) + { + mapping_matrix_init(mixing_matrix, mapping_matrix_foa_mixing.rows, + mapping_matrix_foa_mixing.cols, mapping_matrix_foa_mixing.gain, + mapping_matrix_foa_mixing_data, + sizeof(mapping_matrix_foa_mixing_data)); + } + else if (order_plus_one == 3) + { + mapping_matrix_init(mixing_matrix, mapping_matrix_soa_mixing.rows, + mapping_matrix_soa_mixing.cols, mapping_matrix_soa_mixing.gain, + mapping_matrix_soa_mixing_data, + sizeof(mapping_matrix_soa_mixing_data)); + } + else if (order_plus_one == 4) + { + mapping_matrix_init(mixing_matrix, mapping_matrix_toa_mixing.rows, + mapping_matrix_toa_mixing.cols, mapping_matrix_toa_mixing.gain, + mapping_matrix_toa_mixing_data, + sizeof(mapping_matrix_toa_mixing_data)); + } + else + return OPUS_BAD_ARG; + + st->mixing_matrix_size_in_bytes = mapping_matrix_get_size( + mixing_matrix->rows, mixing_matrix->cols); + if (!st->mixing_matrix_size_in_bytes) + return OPUS_BAD_ARG; + + /* Assign demixing matrix based on available pre-computed matrices. */ + demixing_matrix = get_enc_demixing_matrix(st); + if (order_plus_one == 2) + { + mapping_matrix_init(demixing_matrix, mapping_matrix_foa_demixing.rows, + mapping_matrix_foa_demixing.cols, mapping_matrix_foa_demixing.gain, + mapping_matrix_foa_demixing_data, + sizeof(mapping_matrix_foa_demixing_data)); + } + else if (order_plus_one == 3) + { + mapping_matrix_init(demixing_matrix, mapping_matrix_soa_demixing.rows, + mapping_matrix_soa_demixing.cols, mapping_matrix_soa_demixing.gain, + mapping_matrix_soa_demixing_data, + sizeof(mapping_matrix_soa_demixing_data)); + } + else if (order_plus_one == 4) + { + mapping_matrix_init(demixing_matrix, mapping_matrix_toa_demixing.rows, + mapping_matrix_toa_demixing.cols, mapping_matrix_toa_demixing.gain, + mapping_matrix_toa_demixing_data, + sizeof(mapping_matrix_toa_demixing_data)); + } + else + return OPUS_BAD_ARG; + + st->demixing_matrix_size_in_bytes = mapping_matrix_get_size( + demixing_matrix->rows, demixing_matrix->cols); + if (!st->demixing_matrix_size_in_bytes) + return OPUS_BAD_ARG; + } + else + return OPUS_UNIMPLEMENTED; + + /* Ensure matrices are large enough for desired coding scheme. */ + if (*streams + *coupled_streams > mixing_matrix->rows || + channels > mixing_matrix->cols || + channels > demixing_matrix->rows || + *streams + *coupled_streams > demixing_matrix->cols) + return OPUS_BAD_ARG; + + /* Set trivial mapping so each input channel pairs with a matrix column. */ + for (i = 0; i < channels; i++) + mapping[i] = i; + + /* Initialize multistream encoder with provided settings. */ + ms_encoder = get_multistream_encoder(st); + ret = opus_multistream_encoder_init(ms_encoder, Fs, channels, *streams, + *coupled_streams, mapping, application); + return ret; +} + +OpusProjectionEncoder *opus_projection_ambisonics_encoder_create( + opus_int32 Fs, int channels, int mapping_family, int *streams, + int *coupled_streams, int application, int *error) +{ + int size; + int ret; + OpusProjectionEncoder *st; + + /* Allocate space for the projection encoder. */ + size = opus_projection_ambisonics_encoder_get_size(channels, mapping_family); + if (!size) { + if (error) + *error = OPUS_ALLOC_FAIL; + return NULL; + } + st = (OpusProjectionEncoder *)opus_alloc(size); + if (!st) + { + if (error) + *error = OPUS_ALLOC_FAIL; + return NULL; + } + + /* Initialize projection encoder with provided settings. */ + ret = opus_projection_ambisonics_encoder_init(st, Fs, channels, + mapping_family, streams, coupled_streams, application); + if (ret != OPUS_OK) + { + opus_free(st); + st = NULL; + } + if (error) + *error = ret; + return st; +} + +int opus_projection_encode(OpusProjectionEncoder *st, const opus_int16 *pcm, + int frame_size, unsigned char *data, + opus_int32 max_data_bytes) +{ + return opus_multistream_encode_native(get_multistream_encoder(st), + opus_projection_copy_channel_in_short, pcm, frame_size, data, + max_data_bytes, 16, downmix_int, 0, get_mixing_matrix(st)); +} + +#ifndef DISABLE_FLOAT_API +#ifdef FIXED_POINT +int opus_projection_encode_float(OpusProjectionEncoder *st, const float *pcm, + int frame_size, unsigned char *data, + opus_int32 max_data_bytes) +{ + return opus_multistream_encode_native(get_multistream_encoder(st), + opus_projection_copy_channel_in_float, pcm, frame_size, data, + max_data_bytes, 16, downmix_float, 1, get_mixing_matrix(st)); +} +#else +int opus_projection_encode_float(OpusProjectionEncoder *st, const float *pcm, + int frame_size, unsigned char *data, + opus_int32 max_data_bytes) +{ + return opus_multistream_encode_native(get_multistream_encoder(st), + opus_projection_copy_channel_in_float, pcm, frame_size, data, + max_data_bytes, 24, downmix_float, 1, get_mixing_matrix(st)); +} +#endif +#endif + +void opus_projection_encoder_destroy(OpusProjectionEncoder *st) +{ + opus_free(st); +} + +int opus_projection_encoder_ctl(OpusProjectionEncoder *st, int request, ...) +{ + va_list ap; + MappingMatrix *demixing_matrix; + OpusMSEncoder *ms_encoder; + int ret = OPUS_OK; + + ms_encoder = get_multistream_encoder(st); + demixing_matrix = get_enc_demixing_matrix(st); + + va_start(ap, request); + switch(request) + { + case OPUS_PROJECTION_GET_DEMIXING_MATRIX_SIZE_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = + ms_encoder->layout.nb_channels * (ms_encoder->layout.nb_streams + + ms_encoder->layout.nb_coupled_streams) * sizeof(opus_int16); + } + break; + case OPUS_PROJECTION_GET_DEMIXING_MATRIX_GAIN_REQUEST: + { + opus_int32 *value = va_arg(ap, opus_int32*); + if (!value) + { + goto bad_arg; + } + *value = demixing_matrix->gain; + } + break; + case OPUS_PROJECTION_GET_DEMIXING_MATRIX_REQUEST: + { + int i, j, k, l; + int nb_input_streams; + int nb_output_streams; + unsigned char *external_char; + opus_int16 *internal_short; + opus_int32 external_size; + opus_int32 internal_size; + + /* (I/O is in relation to the decoder's perspective). */ + nb_input_streams = ms_encoder->layout.nb_streams + + ms_encoder->layout.nb_coupled_streams; + nb_output_streams = ms_encoder->layout.nb_channels; + + external_char = va_arg(ap, unsigned char *); + external_size = va_arg(ap, opus_int32); + if (!external_char) + { + goto bad_arg; + } + internal_short = mapping_matrix_get_data(demixing_matrix); + internal_size = nb_input_streams * nb_output_streams * sizeof(opus_int16); + if (external_size != internal_size) + { + goto bad_arg; + } + + /* Copy demixing matrix subset to output destination. */ + l = 0; + for (i = 0; i < nb_input_streams; i++) { + for (j = 0; j < nb_output_streams; j++) { + k = demixing_matrix->rows * i + j; + external_char[2*l] = (unsigned char)internal_short[k]; + external_char[2*l+1] = (unsigned char)(internal_short[k] >> 8); + l++; + } + } + } + break; + default: + { + ret = opus_multistream_encoder_ctl_va_list(ms_encoder, request, ap); + } + break; + } + va_end(ap); + return ret; + +bad_arg: + va_end(ap); + return OPUS_BAD_ARG; +} + diff --git a/media/libopus/src/repacketizer.c b/media/libopus/src/repacketizer.c new file mode 100644 index 0000000000..bda44a148a --- /dev/null +++ b/media/libopus/src/repacketizer.c @@ -0,0 +1,349 @@ +/* Copyright (c) 2011 Xiph.Org Foundation + Written by Jean-Marc Valin */ +/* + Redistribution and use in source and binary forms, with or without + modification, are permitted provided that the following conditions + are met: + + - Redistributions of source code must retain the above copyright + notice, this list of conditions and the following disclaimer. + + - Redistributions in binary form must reproduce the above copyright + notice, this list of conditions and the following disclaimer in the + documentation and/or other materials provided with the distribution. + + THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR + A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER + OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, + EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, + PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR + PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF + LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING + NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. +*/ + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include "opus.h" +#include "opus_private.h" +#include "os_support.h" + + +int opus_repacketizer_get_size(void) +{ + return sizeof(OpusRepacketizer); +} + +OpusRepacketizer *opus_repacketizer_init(OpusRepacketizer *rp) +{ + rp->nb_frames = 0; + return rp; +} + +OpusRepacketizer *opus_repacketizer_create(void) +{ + OpusRepacketizer *rp; + rp=(OpusRepacketizer *)opus_alloc(opus_repacketizer_get_size()); + if(rp==NULL)return NULL; + return opus_repacketizer_init(rp); +} + +void opus_repacketizer_destroy(OpusRepacketizer *rp) +{ + opus_free(rp); +} + +static int opus_repacketizer_cat_impl(OpusRepacketizer *rp, const unsigned char *data, opus_int32 len, int self_delimited) +{ + unsigned char tmp_toc; + int curr_nb_frames,ret; + /* Set of check ToC */ + if (len<1) return OPUS_INVALID_PACKET; + if (rp->nb_frames == 0) + { + rp->toc = data[0]; + rp->framesize = opus_packet_get_samples_per_frame(data, 8000); + } else if ((rp->toc&0xFC) != (data[0]&0xFC)) + { + /*fprintf(stderr, "toc mismatch: 0x%x vs 0x%x\n", rp->toc, data[0]);*/ + return OPUS_INVALID_PACKET; + } + curr_nb_frames = opus_packet_get_nb_frames(data, len); + if(curr_nb_frames<1) return OPUS_INVALID_PACKET; + + /* Check the 120 ms maximum packet size */ + if ((curr_nb_frames+rp->nb_frames)*rp->framesize > 960) + { + return OPUS_INVALID_PACKET; + } + + ret=opus_packet_parse_impl(data, len, self_delimited, &tmp_toc, &rp->frames[rp->nb_frames], &rp->len[rp->nb_frames], NULL, NULL); + if(ret<1)return ret; + + rp->nb_frames += curr_nb_frames; + return OPUS_OK; +} + +int opus_repacketizer_cat(OpusRepacketizer *rp, const unsigned char *data, opus_int32 len) +{ + return opus_repacketizer_cat_impl(rp, data, len, 0); +} + +int opus_repacketizer_get_nb_frames(OpusRepacketizer *rp) +{ + return rp->nb_frames; +} + +opus_int32 opus_repacketizer_out_range_impl(OpusRepacketizer *rp, int begin, int end, + unsigned char *data, opus_int32 maxlen, int self_delimited, int pad) +{ + int i, count; + opus_int32 tot_size; + opus_int16 *len; + const unsigned char **frames; + unsigned char * ptr; + + if (begin<0 || begin>=end || end>rp->nb_frames) + { + /*fprintf(stderr, "%d %d %d\n", begin, end, rp->nb_frames);*/ + return OPUS_BAD_ARG; + } + count = end-begin; + + len = rp->len+begin; + frames = rp->frames+begin; + if (self_delimited) + tot_size = 1 + (len[count-1]>=252); + else + tot_size = 0; + + ptr = data; + if (count==1) + { + /* Code 0 */ + tot_size += len[0]+1; + if (tot_size > maxlen) + return OPUS_BUFFER_TOO_SMALL; + *ptr++ = rp->toc&0xFC; + } else if (count==2) + { + if (len[1] == len[0]) + { + /* Code 1 */ + tot_size += 2*len[0]+1; + if (tot_size > maxlen) + return OPUS_BUFFER_TOO_SMALL; + *ptr++ = (rp->toc&0xFC) | 0x1; + } else { + /* Code 2 */ + tot_size += len[0]+len[1]+2+(len[0]>=252); + if (tot_size > maxlen) + return OPUS_BUFFER_TOO_SMALL; + *ptr++ = (rp->toc&0xFC) | 0x2; + ptr += encode_size(len[0], ptr); + } + } + if (count > 2 || (pad && tot_size < maxlen)) + { + /* Code 3 */ + int vbr; + int pad_amount=0; + + /* Restart the process for the padding case */ + ptr = data; + if (self_delimited) + tot_size = 1 + (len[count-1]>=252); + else + tot_size = 0; + vbr = 0; + for (i=1;i<count;i++) + { + if (len[i] != len[0]) + { + vbr=1; + break; + } + } + if (vbr) + { + tot_size += 2; + for (i=0;i<count-1;i++) + tot_size += 1 + (len[i]>=252) + len[i]; + tot_size += len[count-1]; + + if (tot_size > maxlen) + return OPUS_BUFFER_TOO_SMALL; + *ptr++ = (rp->toc&0xFC) | 0x3; + *ptr++ = count | 0x80; + } else { + tot_size += count*len[0]+2; + if (tot_size > maxlen) + return OPUS_BUFFER_TOO_SMALL; + *ptr++ = (rp->toc&0xFC) | 0x3; + *ptr++ = count; + } + pad_amount = pad ? (maxlen-tot_size) : 0; + if (pad_amount != 0) + { + int nb_255s; + data[1] |= 0x40; + nb_255s = (pad_amount-1)/255; + for (i=0;i<nb_255s;i++) + *ptr++ = 255; + *ptr++ = pad_amount-255*nb_255s-1; + tot_size += pad_amount; + } + if (vbr) + { + for (i=0;i<count-1;i++) + ptr += encode_size(len[i], ptr); + } + } + if (self_delimited) { + int sdlen = encode_size(len[count-1], ptr); + ptr += sdlen; + } + /* Copy the actual data */ + for (i=0;i<count;i++) + { + /* Using OPUS_MOVE() instead of OPUS_COPY() in case we're doing in-place + padding from opus_packet_pad or opus_packet_unpad(). */ + /* assert disabled because it's not valid in C. */ + /* celt_assert(frames[i] + len[i] <= data || ptr <= frames[i]); */ + OPUS_MOVE(ptr, frames[i], len[i]); + ptr += len[i]; + } + if (pad) + { + /* Fill padding with zeros. */ + while (ptr<data+maxlen) + *ptr++=0; + } + return tot_size; +} + +opus_int32 opus_repacketizer_out_range(OpusRepacketizer *rp, int begin, int end, unsigned char *data, opus_int32 maxlen) +{ + return opus_repacketizer_out_range_impl(rp, begin, end, data, maxlen, 0, 0); +} + +opus_int32 opus_repacketizer_out(OpusRepacketizer *rp, unsigned char *data, opus_int32 maxlen) +{ + return opus_repacketizer_out_range_impl(rp, 0, rp->nb_frames, data, maxlen, 0, 0); +} + +int opus_packet_pad(unsigned char *data, opus_int32 len, opus_int32 new_len) +{ + OpusRepacketizer rp; + opus_int32 ret; + if (len < 1) + return OPUS_BAD_ARG; + if (len==new_len) + return OPUS_OK; + else if (len > new_len) + return OPUS_BAD_ARG; + opus_repacketizer_init(&rp); + /* Moving payload to the end of the packet so we can do in-place padding */ + OPUS_MOVE(data+new_len-len, data, len); + ret = opus_repacketizer_cat(&rp, data+new_len-len, len); + if (ret != OPUS_OK) + return ret; + ret = opus_repacketizer_out_range_impl(&rp, 0, rp.nb_frames, data, new_len, 0, 1); + if (ret > 0) + return OPUS_OK; + else + return ret; +} + +opus_int32 opus_packet_unpad(unsigned char *data, opus_int32 len) +{ + OpusRepacketizer rp; + opus_int32 ret; + if (len < 1) + return OPUS_BAD_ARG; + opus_repacketizer_init(&rp); + ret = opus_repacketizer_cat(&rp, data, len); + if (ret < 0) + return ret; + ret = opus_repacketizer_out_range_impl(&rp, 0, rp.nb_frames, data, len, 0, 0); + celt_assert(ret > 0 && ret <= len); + return ret; +} + +int opus_multistream_packet_pad(unsigned char *data, opus_int32 len, opus_int32 new_len, int nb_streams) +{ + int s; + int count; + unsigned char toc; + opus_int16 size[48]; + opus_int32 packet_offset; + opus_int32 amount; + + if (len < 1) + return OPUS_BAD_ARG; + if (len==new_len) + return OPUS_OK; + else if (len > new_len) + return OPUS_BAD_ARG; + amount = new_len - len; + /* Seek to last stream */ + for (s=0;s<nb_streams-1;s++) + { + if (len<=0) + return OPUS_INVALID_PACKET; + count = opus_packet_parse_impl(data, len, 1, &toc, NULL, + size, NULL, &packet_offset); + if (count<0) + return count; + data += packet_offset; + len -= packet_offset; + } + return opus_packet_pad(data, len, len+amount); +} + +opus_int32 opus_multistream_packet_unpad(unsigned char *data, opus_int32 len, int nb_streams) +{ + int s; + unsigned char toc; + opus_int16 size[48]; + opus_int32 packet_offset; + OpusRepacketizer rp; + unsigned char *dst; + opus_int32 dst_len; + + if (len < 1) + return OPUS_BAD_ARG; + dst = data; + dst_len = 0; + /* Unpad all frames */ + for (s=0;s<nb_streams;s++) + { + opus_int32 ret; + int self_delimited = s!=nb_streams-1; + if (len<=0) + return OPUS_INVALID_PACKET; + opus_repacketizer_init(&rp); + ret = opus_packet_parse_impl(data, len, self_delimited, &toc, NULL, + size, NULL, &packet_offset); + if (ret<0) + return ret; + ret = opus_repacketizer_cat_impl(&rp, data, packet_offset, self_delimited); + if (ret < 0) + return ret; + ret = opus_repacketizer_out_range_impl(&rp, 0, rp.nb_frames, dst, len, self_delimited, 0); + if (ret < 0) + return ret; + else + dst_len += ret; + dst += ret; + data += packet_offset; + len -= packet_offset; + } + return dst_len; +} + diff --git a/media/libopus/src/tansig_table.h b/media/libopus/src/tansig_table.h new file mode 100644 index 0000000000..c76f844a72 --- /dev/null +++ b/media/libopus/src/tansig_table.h @@ -0,0 +1,45 @@ +/* This file is auto-generated by gen_tables */ + +static const float tansig_table[201] = { +0.000000f, 0.039979f, 0.079830f, 0.119427f, 0.158649f, +0.197375f, 0.235496f, 0.272905f, 0.309507f, 0.345214f, +0.379949f, 0.413644f, 0.446244f, 0.477700f, 0.507977f, +0.537050f, 0.564900f, 0.591519f, 0.616909f, 0.641077f, +0.664037f, 0.685809f, 0.706419f, 0.725897f, 0.744277f, +0.761594f, 0.777888f, 0.793199f, 0.807569f, 0.821040f, +0.833655f, 0.845456f, 0.856485f, 0.866784f, 0.876393f, +0.885352f, 0.893698f, 0.901468f, 0.908698f, 0.915420f, +0.921669f, 0.927473f, 0.932862f, 0.937863f, 0.942503f, +0.946806f, 0.950795f, 0.954492f, 0.957917f, 0.961090f, +0.964028f, 0.966747f, 0.969265f, 0.971594f, 0.973749f, +0.975743f, 0.977587f, 0.979293f, 0.980869f, 0.982327f, +0.983675f, 0.984921f, 0.986072f, 0.987136f, 0.988119f, +0.989027f, 0.989867f, 0.990642f, 0.991359f, 0.992020f, +0.992631f, 0.993196f, 0.993718f, 0.994199f, 0.994644f, +0.995055f, 0.995434f, 0.995784f, 0.996108f, 0.996407f, +0.996682f, 0.996937f, 0.997172f, 0.997389f, 0.997590f, +0.997775f, 0.997946f, 0.998104f, 0.998249f, 0.998384f, +0.998508f, 0.998623f, 0.998728f, 0.998826f, 0.998916f, +0.999000f, 0.999076f, 0.999147f, 0.999213f, 0.999273f, +0.999329f, 0.999381f, 0.999428f, 0.999472f, 0.999513f, +0.999550f, 0.999585f, 0.999617f, 0.999646f, 0.999673f, +0.999699f, 0.999722f, 0.999743f, 0.999763f, 0.999781f, +0.999798f, 0.999813f, 0.999828f, 0.999841f, 0.999853f, +0.999865f, 0.999875f, 0.999885f, 0.999893f, 0.999902f, +0.999909f, 0.999916f, 0.999923f, 0.999929f, 0.999934f, +0.999939f, 0.999944f, 0.999948f, 0.999952f, 0.999956f, +0.999959f, 0.999962f, 0.999965f, 0.999968f, 0.999970f, +0.999973f, 0.999975f, 0.999977f, 0.999978f, 0.999980f, +0.999982f, 0.999983f, 0.999984f, 0.999986f, 0.999987f, +0.999988f, 0.999989f, 0.999990f, 0.999990f, 0.999991f, +0.999992f, 0.999992f, 0.999993f, 0.999994f, 0.999994f, +0.999994f, 0.999995f, 0.999995f, 0.999996f, 0.999996f, +0.999996f, 0.999997f, 0.999997f, 0.999997f, 0.999997f, +0.999997f, 0.999998f, 0.999998f, 0.999998f, 0.999998f, +0.999998f, 0.999998f, 0.999999f, 0.999999f, 0.999999f, +0.999999f, 0.999999f, 0.999999f, 0.999999f, 0.999999f, +0.999999f, 0.999999f, 0.999999f, 0.999999f, 0.999999f, +1.000000f, 1.000000f, 1.000000f, 1.000000f, 1.000000f, +1.000000f, 1.000000f, 1.000000f, 1.000000f, 1.000000f, +1.000000f, +}; diff --git a/media/libopus/update.sh b/media/libopus/update.sh new file mode 100755 index 0000000000..a311a3fd86 --- /dev/null +++ b/media/libopus/update.sh @@ -0,0 +1,68 @@ +#!/bin/sh +# This Source Code Form is subject to the terms of the Mozilla Public +# License, v. 2.0. If a copy of the MPL was not distributed with this +# file, You can obtain one at http://mozilla.org/MPL/2.0/. + +# +# Usage: ./update.sh <opus_src_directory> +# +# Copies the needed files from a directory containing the original +# libopus source, and applies any local patches we're carrying. + +TARGET='.' + +STATIC_FILES="COPYING celt/arm/arm2gnu.pl" +MK_FILES="opus_sources.mk celt_sources.mk silk_sources.mk \ + opus_headers.mk celt_headers.mk silk_headers.mk" + +# Make sure we have a source directory +if test -z $1 || ! test -r $1/include/opus.h; then + echo "Update the current directory from a source checkout" + echo "usage: $0 ../opus" + exit 1 +fi + +# "parse" the makefile fragments to get the list of source files +# requires GNU sed extensions +SRC_FILES=$(sed -e ':a;N;$!ba;s/#[^\n]*\(\n\)/\1/g;s/\\\n//g;s/[A-Z0-9_]*[ \t]*=[ \t]*//g' \ + $(for file in ${MK_FILES}; do echo "$1/${file}"; done)) + +# pre-release versions of the code don't list opus_custom.h +# in celt_headers.mk, so we must include it manually +HDR_FILES="include/opus_custom.h" + +# make sure the necessary subdirectories exist +for file in ${SRC_FILES}; do + base=${file##*/} + dir="${file%"${base}"}" + if test ! -d "${TARGET}/${dir}"; then + cmd="mkdir -p ${TARGET}/${dir}" + echo ${cmd} + ${cmd} + fi +done + +# copy files into the target directory +for file in ${STATIC_FILES} ${SRC_FILES} ${HDR_FILES}; do + cmd="cp $1/${file} ${TARGET}/${file}" + echo ${cmd} + ${cmd} +done + +sed \ + -e s/@OPUS_ARM_MAY_HAVE_EDSP@/1/g \ + -e s/@OPUS_ARM_MAY_HAVE_MEDIA@/1/g \ + -e s/@OPUS_ARM_MAY_HAVE_NEON@/1/g \ + $1/celt/arm/armopts.s.in > ${TARGET}/celt/arm/armopts.s + +# query git for the revision we're copying from +if test -d $1/.git; then + version=$(cd $1 && git describe --tags --match 'v*' --dirty) +else + version="UNKNOWN" +fi + +python3 gen-sources.py $1 + +# apply outstanding local patches +patch -p3 --no-backup-if-mismatch < nonunified.patch |