From 43a97878ce14b72f0981164f87f2e35e14151312 Mon Sep 17 00:00:00 2001 From: Daniel Baumann Date: Sun, 7 Apr 2024 11:22:09 +0200 Subject: Adding upstream version 110.0.1. Signed-off-by: Daniel Baumann --- third_party/aom/aom_scale/aom_scale.cmake | 38 ++ third_party/aom/aom_scale/aom_scale.h | 23 + third_party/aom/aom_scale/aom_scale_rtcd.c | 18 + third_party/aom/aom_scale/aom_scale_rtcd.pl | 52 +++ third_party/aom/aom_scale/generic/aom_scale.c | 506 +++++++++++++++++++++ third_party/aom/aom_scale/generic/gen_scalers.c | 201 ++++++++ third_party/aom/aom_scale/generic/yv12config.c | 203 +++++++++ third_party/aom/aom_scale/generic/yv12extend.c | 411 +++++++++++++++++ .../aom/aom_scale/mips/dspr2/yv12extend_dspr2.c | 142 ++++++ third_party/aom/aom_scale/yv12config.h | 143 ++++++ 10 files changed, 1737 insertions(+) create mode 100644 third_party/aom/aom_scale/aom_scale.cmake create mode 100644 third_party/aom/aom_scale/aom_scale.h create mode 100644 third_party/aom/aom_scale/aom_scale_rtcd.c create mode 100644 third_party/aom/aom_scale/aom_scale_rtcd.pl create mode 100644 third_party/aom/aom_scale/generic/aom_scale.c create mode 100644 third_party/aom/aom_scale/generic/gen_scalers.c create mode 100644 third_party/aom/aom_scale/generic/yv12config.c create mode 100644 third_party/aom/aom_scale/generic/yv12extend.c create mode 100644 third_party/aom/aom_scale/mips/dspr2/yv12extend_dspr2.c create mode 100644 third_party/aom/aom_scale/yv12config.h (limited to 'third_party/aom/aom_scale') diff --git a/third_party/aom/aom_scale/aom_scale.cmake b/third_party/aom/aom_scale/aom_scale.cmake new file mode 100644 index 0000000000..197dea6bd2 --- /dev/null +++ b/third_party/aom/aom_scale/aom_scale.cmake @@ -0,0 +1,38 @@ +# +# Copyright (c) 2017, Alliance for Open Media. All rights reserved +# +# This source code is subject to the terms of the BSD 2 Clause License and the +# Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License was +# not distributed with this source code in the LICENSE file, you can obtain it +# at www.aomedia.org/license/software. If the Alliance for Open Media Patent +# License 1.0 was not distributed with this source code in the PATENTS file, you +# can obtain it at www.aomedia.org/license/patent. +# +if(AOM_AOM_SCALE_AOM_SCALE_CMAKE_) + return() +endif() # AOM_AOM_SCALE_AOM_SCALE_CMAKE_ +set(AOM_AOM_SCALE_AOM_SCALE_CMAKE_ 1) + +list(APPEND AOM_SCALE_SOURCES "${AOM_ROOT}/aom_scale/aom_scale.h" + "${AOM_ROOT}/aom_scale/generic/aom_scale.c" + "${AOM_ROOT}/aom_scale/generic/gen_scalers.c" + "${AOM_ROOT}/aom_scale/generic/yv12config.c" + "${AOM_ROOT}/aom_scale/generic/yv12extend.c" + "${AOM_ROOT}/aom_scale/yv12config.h") + +list(APPEND AOM_SCALE_INTRIN_DSPR2 + "${AOM_ROOT}/aom_scale/mips/dspr2/yv12extend_dspr2.c") + +# Creates the aom_scale build target and makes libaom depend on it. The libaom +# target must exist before this function is called. +function(setup_aom_scale_targets) + add_library(aom_scale OBJECT ${AOM_SCALE_SOURCES}) + target_sources(aom PRIVATE $) + + if(HAVE_DSPR2) + add_intrinsics_object_library("" "dspr2" "aom_scale" + "AOM_SCALE_INTRIN_DSPR2" "aom") + endif() + + set(AOM_LIB_TARGETS ${AOM_LIB_TARGETS} aom_scale PARENT_SCOPE) +endfunction() diff --git a/third_party/aom/aom_scale/aom_scale.h b/third_party/aom/aom_scale/aom_scale.h new file mode 100644 index 0000000000..11812a1453 --- /dev/null +++ b/third_party/aom/aom_scale/aom_scale.h @@ -0,0 +1,23 @@ +/* + * Copyright (c) 2016, Alliance for Open Media. All rights reserved + * + * This source code is subject to the terms of the BSD 2 Clause License and + * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License + * was not distributed with this source code in the LICENSE file, you can + * obtain it at www.aomedia.org/license/software. If the Alliance for Open + * Media Patent License 1.0 was not distributed with this source code in the + * PATENTS file, you can obtain it at www.aomedia.org/license/patent. + */ + +#ifndef AOM_AOM_SCALE_AOM_SCALE_H_ +#define AOM_AOM_SCALE_AOM_SCALE_H_ + +#include "aom_scale/yv12config.h" + +extern void aom_scale_frame(YV12_BUFFER_CONFIG *src, YV12_BUFFER_CONFIG *dst, + unsigned char *temp_area, unsigned char temp_height, + unsigned int hscale, unsigned int hratio, + unsigned int vscale, unsigned int vratio, + unsigned int interlaced, const int num_planes); + +#endif // AOM_AOM_SCALE_AOM_SCALE_H_ diff --git a/third_party/aom/aom_scale/aom_scale_rtcd.c b/third_party/aom/aom_scale/aom_scale_rtcd.c new file mode 100644 index 0000000000..a04e053b0a --- /dev/null +++ b/third_party/aom/aom_scale/aom_scale_rtcd.c @@ -0,0 +1,18 @@ +/* + * Copyright (c) 2016, Alliance for Open Media. All rights reserved + * + * This source code is subject to the terms of the BSD 2 Clause License and + * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License + * was not distributed with this source code in the LICENSE file, you can + * obtain it at www.aomedia.org/license/software. If the Alliance for Open + * Media Patent License 1.0 was not distributed with this source code in the + * PATENTS file, you can obtain it at www.aomedia.org/license/patent. + */ +#include "config/aom_config.h" + +#define RTCD_C +#include "config/aom_scale_rtcd.h" + +#include "aom_ports/aom_once.h" + +void aom_scale_rtcd() { aom_once(setup_rtcd_internal); } diff --git a/third_party/aom/aom_scale/aom_scale_rtcd.pl b/third_party/aom/aom_scale/aom_scale_rtcd.pl new file mode 100644 index 0000000000..c5990b1bb3 --- /dev/null +++ b/third_party/aom/aom_scale/aom_scale_rtcd.pl @@ -0,0 +1,52 @@ +## +## Copyright (c) 2017, Alliance for Open Media. All rights reserved +## +## This source code is subject to the terms of the BSD 2 Clause License and +## the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License +## was not distributed with this source code in the LICENSE file, you can +## obtain it at www.aomedia.org/license/software. If the Alliance for Open +## Media Patent License 1.0 was not distributed with this source code in the +## PATENTS file, you can obtain it at www.aomedia.org/license/patent. +## +sub aom_scale_forward_decls() { +print <> 4); + source += source_step; + dest += dest_step; + } +} + +/**************************************************************************** + * + * ROUTINE : scale1d_2t1_ps + * + * INPUTS : const unsigned char *source : Pointer to data to be scaled. + * int source_step : Number of pixels to step on + * in source. + * unsigned int source_scale : Scale for source (UNUSED). + * unsigned int source_length : Length of source (UNUSED). + * unsigned char *dest : Pointer to output data array. + * int dest_step : Number of pixels to step on + * in destination. + * unsigned int dest_scale : Scale for destination + * (UNUSED). + * unsigned int dest_length : Length of destination. + * + * OUTPUTS : None. + * + * RETURNS : void + * + * FUNCTION : Performs 2-to-1 point subsampled scaling. + * + * SPECIAL NOTES : None. + * + ****************************************************************************/ +static void scale1d_2t1_ps(const unsigned char *source, int source_step, + unsigned int source_scale, + unsigned int source_length, unsigned char *dest, + int dest_step, unsigned int dest_scale, + unsigned int dest_length) { + const unsigned char *const dest_end = dest + dest_length * dest_step; + (void)source_length; + (void)source_scale; + (void)dest_scale; + + source_step *= 2; // Every other row. + + while (dest < dest_end) { + *dest = *source; + source += source_step; + dest += dest_step; + } +} +/**************************************************************************** + * + * ROUTINE : scale1d_c + * + * INPUTS : const unsigned char *source : Pointer to data to be scaled. + * int source_step : Number of pixels to step on + * in source. + * unsigned int source_scale : Scale for source. + * unsigned int source_length : Length of source (UNUSED). + * unsigned char *dest : Pointer to output data array. + * int dest_step : Number of pixels to step on + * in destination. + * unsigned int dest_scale : Scale for destination. + * unsigned int dest_length : Length of destination. + * + * OUTPUTS : None. + * + * RETURNS : void + * + * FUNCTION : Performs linear interpolation in one dimension. + * + * SPECIAL NOTES : None. + * + ****************************************************************************/ +static void scale1d_c(const unsigned char *source, int source_step, + unsigned int source_scale, unsigned int source_length, + unsigned char *dest, int dest_step, + unsigned int dest_scale, unsigned int dest_length) { + const unsigned char *const dest_end = dest + dest_length * dest_step; + const unsigned int round_value = dest_scale / 2; + unsigned int left_modifier = dest_scale; + unsigned int right_modifier = 0; + unsigned char left_pixel = source[0]; + unsigned char right_pixel = source[source_step]; + + (void)source_length; + + /* These asserts are needed if there are boundary issues... */ + /* assert ( dest_scale > source_scale );*/ + /* assert ( (source_length - 1) * dest_scale >= (dest_length - 1) * + * source_scale);*/ + + while (dest < dest_end) { + *dest = (unsigned char)((left_modifier * left_pixel + + right_modifier * right_pixel + round_value) / + dest_scale); + + right_modifier += source_scale; + + while (right_modifier > dest_scale) { + right_modifier -= dest_scale; + source += source_step; + left_pixel = source[0]; + right_pixel = source[source_step]; + } + + left_modifier = dest_scale - right_modifier; + } +} + +/**************************************************************************** + * + * ROUTINE : Scale2D + * + * INPUTS : const unsigned char *source : Pointer to data to be + * scaled. + * int source_pitch : Stride of source image. + * unsigned int source_width : Width of input image. + * unsigned int source_height : Height of input image. + * unsigned char *dest : Pointer to output data + * array. + * int dest_pitch : Stride of destination + * image. + * unsigned int dest_width : Width of destination image. + * unsigned int dest_height : Height of destination + * image. + * unsigned char *temp_area : Pointer to temp work area. + * unsigned char temp_area_height : Height of temp work area. + * unsigned int hscale : Horizontal scale factor + * numerator. + * unsigned int hratio : Horizontal scale factor + * denominator. + * unsigned int vscale : Vertical scale factor + * numerator. + * unsigned int vratio : Vertical scale factor + * denominator. + * unsigned int interlaced : Interlace flag. + * + * OUTPUTS : None. + * + * RETURNS : void + * + * FUNCTION : Performs 2-tap linear interpolation in two dimensions. + * + * SPECIAL NOTES : Expansion is performed one band at a time to help with + * caching. + * + ****************************************************************************/ +static void Scale2D( + /*const*/ + unsigned char *source, int source_pitch, unsigned int source_width, + unsigned int source_height, unsigned char *dest, int dest_pitch, + unsigned int dest_width, unsigned int dest_height, unsigned char *temp_area, + unsigned char temp_area_height, unsigned int hscale, unsigned int hratio, + unsigned int vscale, unsigned int vratio, unsigned int interlaced) { + unsigned int i, j, k; + unsigned int bands; + unsigned int dest_band_height; + unsigned int source_band_height; + + typedef void (*Scale1D)(const unsigned char *source, int source_step, + unsigned int source_scale, unsigned int source_length, + unsigned char *dest, int dest_step, + unsigned int dest_scale, unsigned int dest_length); + + Scale1D Scale1Dv = scale1d_c; + Scale1D Scale1Dh = scale1d_c; + + void (*horiz_line_scale)(const unsigned char *, unsigned int, unsigned char *, + unsigned int) = NULL; + void (*vert_band_scale)(unsigned char *, int, unsigned char *, int, + unsigned int) = NULL; + + int ratio_scalable = 1; + int interpolation = 0; + + unsigned char *source_base; + unsigned char *line_src; + + source_base = (unsigned char *)source; + + if (source_pitch < 0) { + int offset; + + offset = (source_height - 1); + offset *= source_pitch; + + source_base += offset; + } + + /* find out the ratio for each direction */ + switch (hratio * 10 / hscale) { + case 8: + /* 4-5 Scale in Width direction */ + horiz_line_scale = aom_horizontal_line_5_4_scale; + break; + case 6: + /* 3-5 Scale in Width direction */ + horiz_line_scale = aom_horizontal_line_5_3_scale; + break; + case 5: + /* 1-2 Scale in Width direction */ + horiz_line_scale = aom_horizontal_line_2_1_scale; + break; + default: + /* The ratio is not acceptable now */ + /* throw("The ratio is not acceptable for now!"); */ + ratio_scalable = 0; + break; + } + + switch (vratio * 10 / vscale) { + case 8: + /* 4-5 Scale in vertical direction */ + vert_band_scale = aom_vertical_band_5_4_scale; + source_band_height = 5; + dest_band_height = 4; + break; + case 6: + /* 3-5 Scale in vertical direction */ + vert_band_scale = aom_vertical_band_5_3_scale; + source_band_height = 5; + dest_band_height = 3; + break; + case 5: + /* 1-2 Scale in vertical direction */ + + if (interlaced) { + /* if the content is interlaced, point sampling is used */ + vert_band_scale = aom_vertical_band_2_1_scale; + } else { + interpolation = 1; + /* if the content is progressive, interplo */ + vert_band_scale = aom_vertical_band_2_1_scale_i; + } + + source_band_height = 2; + dest_band_height = 1; + break; + default: + /* The ratio is not acceptable now */ + /* throw("The ratio is not acceptable for now!"); */ + ratio_scalable = 0; + break; + } + + if (ratio_scalable) { + if (source_height == dest_height) { + /* for each band of the image */ + for (k = 0; k < dest_height; ++k) { + horiz_line_scale(source, source_width, dest, dest_width); + source += source_pitch; + dest += dest_pitch; + } + + return; + } + + if (interpolation) { + if (source < source_base) source = source_base; + + horiz_line_scale(source, source_width, temp_area, dest_width); + } + + for (k = 0; k < (dest_height + dest_band_height - 1) / dest_band_height; + ++k) { + /* scale one band horizontally */ + for (i = 0; i < source_band_height; ++i) { + /* Trap case where we could read off the base of the source buffer */ + + line_src = source + i * source_pitch; + + if (line_src < source_base) line_src = source_base; + + horiz_line_scale(line_src, source_width, + temp_area + (i + 1) * dest_pitch, dest_width); + } + + /* Vertical scaling is in place */ + vert_band_scale(temp_area + dest_pitch, dest_pitch, dest, dest_pitch, + dest_width); + + if (interpolation) + memcpy(temp_area, temp_area + source_band_height * dest_pitch, + dest_width); + + /* Next band... */ + source += (unsigned long)source_band_height * source_pitch; + dest += (unsigned long)dest_band_height * dest_pitch; + } + + return; + } + + if (hscale == 2 && hratio == 1) Scale1Dh = scale1d_2t1_ps; + + if (vscale == 2 && vratio == 1) { + if (interlaced) + Scale1Dv = scale1d_2t1_ps; + else + Scale1Dv = scale1d_2t1_i; + } + + if (source_height == dest_height) { + /* for each band of the image */ + for (k = 0; k < dest_height; ++k) { + Scale1Dh(source, 1, hscale, source_width + 1, dest, 1, hratio, + dest_width); + source += source_pitch; + dest += dest_pitch; + } + + return; + } + + if (dest_height > source_height) { + dest_band_height = temp_area_height - 1; + source_band_height = dest_band_height * source_height / dest_height; + } else { + source_band_height = temp_area_height - 1; + dest_band_height = source_band_height * vratio / vscale; + } + + /* first row needs to be done so that we can stay one row ahead for vertical + * zoom */ + Scale1Dh(source, 1, hscale, source_width + 1, temp_area, 1, hratio, + dest_width); + + /* for each band of the image */ + bands = (dest_height + dest_band_height - 1) / dest_band_height; + + for (k = 0; k < bands; ++k) { + /* scale one band horizontally */ + for (i = 1; i < source_band_height + 1; ++i) { + if (k * source_band_height + i < source_height) { + Scale1Dh(source + i * source_pitch, 1, hscale, source_width + 1, + temp_area + i * dest_pitch, 1, hratio, dest_width); + } else { /* Duplicate the last row */ + /* copy temp_area row 0 over from last row in the past */ + memcpy(temp_area + i * dest_pitch, temp_area + (i - 1) * dest_pitch, + dest_pitch); + } + } + + /* scale one band vertically */ + for (j = 0; j < dest_width; ++j) { + Scale1Dv(&temp_area[j], dest_pitch, vscale, source_band_height + 1, + &dest[j], dest_pitch, vratio, dest_band_height); + } + + /* copy temp_area row 0 over from last row in the past */ + memcpy(temp_area, temp_area + source_band_height * dest_pitch, dest_pitch); + + /* move to the next band */ + source += source_band_height * source_pitch; + dest += dest_band_height * dest_pitch; + } +} + +/**************************************************************************** + * + * ROUTINE : aom_scale_frame + * + * INPUTS : YV12_BUFFER_CONFIG *src : Pointer to frame to be + * scaled. + * YV12_BUFFER_CONFIG *dst : Pointer to buffer to hold + * scaled frame. + * unsigned char *temp_area : Pointer to temp work area. + * unsigned char temp_area_height : Height of temp work area. + * unsigned int hscale : Horizontal scale factor + * numerator. + * unsigned int hratio : Horizontal scale factor + * denominator. + * unsigned int vscale : Vertical scale factor + * numerator. + * unsigned int vratio : Vertical scale factor + * denominator. + * unsigned int interlaced : Interlace flag. + * + * OUTPUTS : None. + * + * RETURNS : void + * + * FUNCTION : Performs 2-tap linear interpolation in two dimensions. + * + * SPECIAL NOTES : Expansion is performed one band at a time to help with + * caching. + * + ****************************************************************************/ +void aom_scale_frame(YV12_BUFFER_CONFIG *src, YV12_BUFFER_CONFIG *dst, + unsigned char *temp_area, unsigned char temp_height, + unsigned int hscale, unsigned int hratio, + unsigned int vscale, unsigned int vratio, + unsigned int interlaced, const int num_planes) { + const int dw = (hscale - 1 + src->y_width * hratio) / hscale; + const int dh = (vscale - 1 + src->y_height * vratio) / vscale; + + for (int plane = 0; plane < num_planes; ++plane) { + const int is_uv = plane > 0; + const int plane_dw = dw >> is_uv; + const int plane_dh = dh >> is_uv; + + Scale2D((unsigned char *)src->buffers[plane], src->strides[is_uv], + src->widths[is_uv], src->heights[is_uv], + (unsigned char *)dst->buffers[plane], dst->strides[is_uv], plane_dw, + plane_dh, temp_area, temp_height, hscale, hratio, vscale, vratio, + interlaced); + + if (plane_dw < dst->widths[is_uv]) + for (int i = 0; i < plane_dh; ++i) + memset(dst->buffers[plane] + i * dst->strides[is_uv] + plane_dw - 1, + dst->buffers[plane][i * dst->strides[is_uv] + plane_dw - 2], + dst->widths[is_uv] - plane_dw + 1); + + if (plane_dh < dst->heights[is_uv]) + for (int i = plane_dh - 1; i < dst->heights[is_uv]; ++i) + memcpy(dst->buffers[plane] + i * dst->strides[is_uv], + dst->buffers[plane] + (plane_dh - 2) * dst->strides[is_uv], + dst->widths[is_uv] + 1); + } +} diff --git a/third_party/aom/aom_scale/generic/gen_scalers.c b/third_party/aom/aom_scale/generic/gen_scalers.c new file mode 100644 index 0000000000..549e2aa690 --- /dev/null +++ b/third_party/aom/aom_scale/generic/gen_scalers.c @@ -0,0 +1,201 @@ +/* + * Copyright (c) 2016, Alliance for Open Media. All rights reserved + * + * This source code is subject to the terms of the BSD 2 Clause License and + * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License + * was not distributed with this source code in the LICENSE file, you can + * obtain it at www.aomedia.org/license/software. If the Alliance for Open + * Media Patent License 1.0 was not distributed with this source code in the + * PATENTS file, you can obtain it at www.aomedia.org/license/patent. + */ + +#include "config/aom_scale_rtcd.h" + +#include "aom_scale/aom_scale.h" +#include "aom_mem/aom_mem.h" +/**************************************************************************** + * Imports + ****************************************************************************/ + +/**************************************************************************** + * + * + * INPUTS : const unsigned char *source : Pointer to source data. + * unsigned int source_width : Stride of source. + * unsigned char *dest : Pointer to destination data. + * unsigned int dest_width : Stride of destination + * (NOT USED). + * + * OUTPUTS : None. + * + * RETURNS : void + * + * FUNCTION : Copies horizontal line of pixels from source to + * destination scaling up by 4 to 5. + * + * SPECIAL NOTES : None. + * + ****************************************************************************/ +void aom_horizontal_line_5_4_scale_c(const unsigned char *source, + unsigned int source_width, + unsigned char *dest, + unsigned int dest_width) { + const unsigned char *const source_end = source + source_width; + (void)dest_width; + + while (source < source_end) { + const unsigned int a = source[0]; + const unsigned int b = source[1]; + const unsigned int c = source[2]; + const unsigned int d = source[3]; + const unsigned int e = source[4]; + + dest[0] = (unsigned char)a; + dest[1] = (unsigned char)((b * 192 + c * 64 + 128) >> 8); + dest[2] = (unsigned char)((c * 128 + d * 128 + 128) >> 8); + dest[3] = (unsigned char)((d * 64 + e * 192 + 128) >> 8); + + source += 5; + dest += 4; + } +} + +void aom_vertical_band_5_4_scale_c(unsigned char *source, int src_pitch, + unsigned char *dest, int dest_pitch, + unsigned int dest_width) { + const unsigned char *const dest_end = dest + dest_width; + while (dest < dest_end) { + const unsigned int a = source[0 * src_pitch]; + const unsigned int b = source[1 * src_pitch]; + const unsigned int c = source[2 * src_pitch]; + const unsigned int d = source[3 * src_pitch]; + const unsigned int e = source[4 * src_pitch]; + + dest[0 * dest_pitch] = (unsigned char)a; + dest[1 * dest_pitch] = (unsigned char)((b * 192 + c * 64 + 128) >> 8); + dest[2 * dest_pitch] = (unsigned char)((c * 128 + d * 128 + 128) >> 8); + dest[3 * dest_pitch] = (unsigned char)((d * 64 + e * 192 + 128) >> 8); + + ++source; + ++dest; + } +} + +/*7*************************************************************************** + * + * ROUTINE : aom_horizontal_line_3_5_scale_c + * + * INPUTS : const unsigned char *source : Pointer to source data. + * unsigned int source_width : Stride of source. + * unsigned char *dest : Pointer to destination data. + * unsigned int dest_width : Stride of destination + * (NOT USED). + * + * OUTPUTS : None. + * + * RETURNS : void + * + * FUNCTION : Copies horizontal line of pixels from source to + * destination scaling up by 3 to 5. + * + * SPECIAL NOTES : None. + * + * + ****************************************************************************/ +void aom_horizontal_line_5_3_scale_c(const unsigned char *source, + unsigned int source_width, + unsigned char *dest, + unsigned int dest_width) { + const unsigned char *const source_end = source + source_width; + (void)dest_width; + while (source < source_end) { + const unsigned int a = source[0]; + const unsigned int b = source[1]; + const unsigned int c = source[2]; + const unsigned int d = source[3]; + const unsigned int e = source[4]; + + dest[0] = (unsigned char)a; + dest[1] = (unsigned char)((b * 85 + c * 171 + 128) >> 8); + dest[2] = (unsigned char)((d * 171 + e * 85 + 128) >> 8); + + source += 5; + dest += 3; + } +} + +void aom_vertical_band_5_3_scale_c(unsigned char *source, int src_pitch, + unsigned char *dest, int dest_pitch, + unsigned int dest_width) { + const unsigned char *const dest_end = dest + dest_width; + while (dest < dest_end) { + const unsigned int a = source[0 * src_pitch]; + const unsigned int b = source[1 * src_pitch]; + const unsigned int c = source[2 * src_pitch]; + const unsigned int d = source[3 * src_pitch]; + const unsigned int e = source[4 * src_pitch]; + + dest[0 * dest_pitch] = (unsigned char)a; + dest[1 * dest_pitch] = (unsigned char)((b * 85 + c * 171 + 128) >> 8); + dest[2 * dest_pitch] = (unsigned char)((d * 171 + e * 85 + 128) >> 8); + + ++source; + ++dest; + } +} + +/**************************************************************************** + * + * ROUTINE : aom_horizontal_line_1_2_scale_c + * + * INPUTS : const unsigned char *source : Pointer to source data. + * unsigned int source_width : Stride of source. + * unsigned char *dest : Pointer to destination data. + * unsigned int dest_width : Stride of destination + * (NOT USED). + * + * OUTPUTS : None. + * + * RETURNS : void + * + * FUNCTION : Copies horizontal line of pixels from source to + * destination scaling up by 1 to 2. + * + * SPECIAL NOTES : None. + * + ****************************************************************************/ +void aom_horizontal_line_2_1_scale_c(const unsigned char *source, + unsigned int source_width, + unsigned char *dest, + unsigned int dest_width) { + const unsigned char *const source_end = source + source_width; + (void)dest_width; + while (source < source_end) { + dest[0] = source[0]; + source += 2; + ++dest; + } +} + +void aom_vertical_band_2_1_scale_c(unsigned char *source, int src_pitch, + unsigned char *dest, int dest_pitch, + unsigned int dest_width) { + (void)dest_pitch; + (void)src_pitch; + memcpy(dest, source, dest_width); +} + +void aom_vertical_band_2_1_scale_i_c(unsigned char *source, int src_pitch, + unsigned char *dest, int dest_pitch, + unsigned int dest_width) { + const unsigned char *const dest_end = dest + dest_width; + (void)dest_pitch; + while (dest < dest_end) { + const unsigned int a = source[-src_pitch] * 3; + const unsigned int b = source[0] * 10; + const unsigned int c = source[src_pitch] * 3; + dest[0] = (unsigned char)((8 + a + b + c) >> 4); + ++source; + ++dest; + } +} diff --git a/third_party/aom/aom_scale/generic/yv12config.c b/third_party/aom/aom_scale/generic/yv12config.c new file mode 100644 index 0000000000..84705e2d8b --- /dev/null +++ b/third_party/aom/aom_scale/generic/yv12config.c @@ -0,0 +1,203 @@ +/* + * Copyright (c) 2016, Alliance for Open Media. All rights reserved + * + * This source code is subject to the terms of the BSD 2 Clause License and + * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License + * was not distributed with this source code in the LICENSE file, you can + * obtain it at www.aomedia.org/license/software. If the Alliance for Open + * Media Patent License 1.0 was not distributed with this source code in the + * PATENTS file, you can obtain it at www.aomedia.org/license/patent. + */ + +#include + +#include "aom_mem/aom_mem.h" +#include "aom_ports/mem.h" +#include "aom_scale/yv12config.h" +#include "av1/common/enums.h" + +/**************************************************************************** + * Exports + ****************************************************************************/ + +/**************************************************************************** + * + ****************************************************************************/ +#define yv12_align_addr(addr, align) \ + (void *)(((size_t)(addr) + ((align)-1)) & (size_t) - (align)) + +// TODO(jkoleszar): Maybe replace this with struct aom_image + +int aom_free_frame_buffer(YV12_BUFFER_CONFIG *ybf) { + if (ybf) { + if (ybf->buffer_alloc_sz > 0) { + aom_free(ybf->buffer_alloc); + } + if (ybf->y_buffer_8bit) aom_free(ybf->y_buffer_8bit); + + /* buffer_alloc isn't accessed by most functions. Rather y_buffer, + u_buffer and v_buffer point to buffer_alloc and are used. Clear out + all of this so that a freed pointer isn't inadvertently used */ + memset(ybf, 0, sizeof(YV12_BUFFER_CONFIG)); + } else { + return -1; + } + + return 0; +} + +int aom_realloc_frame_buffer(YV12_BUFFER_CONFIG *ybf, int width, int height, + int ss_x, int ss_y, int use_highbitdepth, + int border, int byte_alignment, + aom_codec_frame_buffer_t *fb, + aom_get_frame_buffer_cb_fn_t cb, void *cb_priv) { +#if CONFIG_SIZE_LIMIT + if (width > DECODE_WIDTH_LIMIT || height > DECODE_HEIGHT_LIMIT) return -1; +#endif + + if (ybf) { + const int aom_byte_align = (byte_alignment == 0) ? 1 : byte_alignment; + const int aligned_width = (width + 7) & ~7; + const int aligned_height = (height + 7) & ~7; + const int y_stride = ((aligned_width + 2 * border) + 31) & ~31; + const uint64_t yplane_size = + (aligned_height + 2 * border) * (uint64_t)y_stride + byte_alignment; + const int uv_width = aligned_width >> ss_x; + const int uv_height = aligned_height >> ss_y; + const int uv_stride = y_stride >> ss_x; + const int uv_border_w = border >> ss_x; + const int uv_border_h = border >> ss_y; + const uint64_t uvplane_size = + (uv_height + 2 * uv_border_h) * (uint64_t)uv_stride + byte_alignment; + + const uint64_t frame_size = + (1 + use_highbitdepth) * (yplane_size + 2 * uvplane_size); + + uint8_t *buf = NULL; + +#if defined AOM_MAX_ALLOCABLE_MEMORY + // The size of ybf->buffer_alloc. + uint64_t alloc_size = frame_size; + // The size of ybf->y_buffer_8bit. + if (use_highbitdepth) alloc_size += yplane_size; + // The decoder may allocate REF_FRAMES frame buffers in the frame buffer + // pool. Bound the total amount of allocated memory as if these REF_FRAMES + // frame buffers were allocated in a single allocation. + if (alloc_size > AOM_MAX_ALLOCABLE_MEMORY / REF_FRAMES) return -1; +#endif + + if (cb != NULL) { + const int align_addr_extra_size = 31; + const uint64_t external_frame_size = frame_size + align_addr_extra_size; + + assert(fb != NULL); + + if (external_frame_size != (size_t)external_frame_size) return -1; + + // Allocation to hold larger frame, or first allocation. + if (cb(cb_priv, (size_t)external_frame_size, fb) < 0) return -1; + + if (fb->data == NULL || fb->size < external_frame_size) return -1; + + ybf->buffer_alloc = (uint8_t *)yv12_align_addr(fb->data, 32); + +#if defined(__has_feature) +#if __has_feature(memory_sanitizer) + // This memset is needed for fixing the issue of using uninitialized + // value in msan test. It will cause a perf loss, so only do this for + // msan test. + memset(ybf->buffer_alloc, 0, (size_t)frame_size); +#endif +#endif + } else if (frame_size > (size_t)ybf->buffer_alloc_sz) { + // Allocation to hold larger frame, or first allocation. + aom_free(ybf->buffer_alloc); + ybf->buffer_alloc = NULL; + + if (frame_size != (size_t)frame_size) return -1; + + ybf->buffer_alloc = (uint8_t *)aom_memalign(32, (size_t)frame_size); + if (!ybf->buffer_alloc) return -1; + + ybf->buffer_alloc_sz = (size_t)frame_size; + + // This memset is needed for fixing valgrind error from C loop filter + // due to access uninitialized memory in frame border. It could be + // removed if border is totally removed. + memset(ybf->buffer_alloc, 0, ybf->buffer_alloc_sz); + } + + /* Only support allocating buffers that have a border that's a multiple + * of 32. The border restriction is required to get 16-byte alignment of + * the start of the chroma rows without introducing an arbitrary gap + * between planes, which would break the semantics of things like + * aom_img_set_rect(). */ + if (border & 0x1f) return -3; + + ybf->y_crop_width = width; + ybf->y_crop_height = height; + ybf->y_width = aligned_width; + ybf->y_height = aligned_height; + ybf->y_stride = y_stride; + + ybf->uv_crop_width = (width + ss_x) >> ss_x; + ybf->uv_crop_height = (height + ss_y) >> ss_y; + ybf->uv_width = uv_width; + ybf->uv_height = uv_height; + ybf->uv_stride = uv_stride; + + ybf->border = border; + ybf->frame_size = (size_t)frame_size; + ybf->subsampling_x = ss_x; + ybf->subsampling_y = ss_y; + + buf = ybf->buffer_alloc; + if (use_highbitdepth) { + // Store uint16 addresses when using 16bit framebuffers + buf = CONVERT_TO_BYTEPTR(ybf->buffer_alloc); + ybf->flags = YV12_FLAG_HIGHBITDEPTH; + } else { + ybf->flags = 0; + } + + ybf->y_buffer = (uint8_t *)yv12_align_addr( + buf + (border * y_stride) + border, aom_byte_align); + ybf->u_buffer = (uint8_t *)yv12_align_addr( + buf + yplane_size + (uv_border_h * uv_stride) + uv_border_w, + aom_byte_align); + ybf->v_buffer = + (uint8_t *)yv12_align_addr(buf + yplane_size + uvplane_size + + (uv_border_h * uv_stride) + uv_border_w, + aom_byte_align); + + ybf->use_external_reference_buffers = 0; + + if (use_highbitdepth) { + if (ybf->y_buffer_8bit) aom_free(ybf->y_buffer_8bit); + ybf->y_buffer_8bit = (uint8_t *)aom_memalign(32, (size_t)yplane_size); + if (!ybf->y_buffer_8bit) return -1; + } else { + if (ybf->y_buffer_8bit) { + aom_free(ybf->y_buffer_8bit); + ybf->y_buffer_8bit = NULL; + ybf->buf_8bit_valid = 0; + } + } + + ybf->corrupted = 0; /* assume not corrupted by errors */ + return 0; + } + return -2; +} + +int aom_alloc_frame_buffer(YV12_BUFFER_CONFIG *ybf, int width, int height, + int ss_x, int ss_y, int use_highbitdepth, int border, + int byte_alignment) { + if (ybf) { + aom_free_frame_buffer(ybf); + return aom_realloc_frame_buffer(ybf, width, height, ss_x, ss_y, + use_highbitdepth, border, byte_alignment, + NULL, NULL, NULL); + } + return -2; +} diff --git a/third_party/aom/aom_scale/generic/yv12extend.c b/third_party/aom/aom_scale/generic/yv12extend.c new file mode 100644 index 0000000000..ba183520a0 --- /dev/null +++ b/third_party/aom/aom_scale/generic/yv12extend.c @@ -0,0 +1,411 @@ +/* + * Copyright (c) 2016, Alliance for Open Media. All rights reserved + * + * This source code is subject to the terms of the BSD 2 Clause License and + * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License + * was not distributed with this source code in the LICENSE file, you can + * obtain it at www.aomedia.org/license/software. If the Alliance for Open + * Media Patent License 1.0 was not distributed with this source code in the + * PATENTS file, you can obtain it at www.aomedia.org/license/patent. + */ + +#include + +#include "config/aom_config.h" +#include "config/aom_scale_rtcd.h" + +#include "aom/aom_integer.h" +#include "aom_mem/aom_mem.h" +#include "aom_ports/mem.h" +#include "aom_scale/yv12config.h" + +static void extend_plane(uint8_t *const src, int src_stride, int width, + int height, int extend_top, int extend_left, + int extend_bottom, int extend_right) { + int i; + const int linesize = extend_left + extend_right + width; + + /* copy the left and right most columns out */ + uint8_t *src_ptr1 = src; + uint8_t *src_ptr2 = src + width - 1; + uint8_t *dst_ptr1 = src - extend_left; + uint8_t *dst_ptr2 = src + width; + + for (i = 0; i < height; ++i) { + memset(dst_ptr1, src_ptr1[0], extend_left); + memset(dst_ptr2, src_ptr2[0], extend_right); + src_ptr1 += src_stride; + src_ptr2 += src_stride; + dst_ptr1 += src_stride; + dst_ptr2 += src_stride; + } + + /* Now copy the top and bottom lines into each line of the respective + * borders + */ + src_ptr1 = src - extend_left; + src_ptr2 = src + src_stride * (height - 1) - extend_left; + dst_ptr1 = src + src_stride * -extend_top - extend_left; + dst_ptr2 = src + src_stride * height - extend_left; + + for (i = 0; i < extend_top; ++i) { + memcpy(dst_ptr1, src_ptr1, linesize); + dst_ptr1 += src_stride; + } + + for (i = 0; i < extend_bottom; ++i) { + memcpy(dst_ptr2, src_ptr2, linesize); + dst_ptr2 += src_stride; + } +} + +static void extend_plane_high(uint8_t *const src8, int src_stride, int width, + int height, int extend_top, int extend_left, + int extend_bottom, int extend_right) { + int i; + const int linesize = extend_left + extend_right + width; + uint16_t *src = CONVERT_TO_SHORTPTR(src8); + + /* copy the left and right most columns out */ + uint16_t *src_ptr1 = src; + uint16_t *src_ptr2 = src + width - 1; + uint16_t *dst_ptr1 = src - extend_left; + uint16_t *dst_ptr2 = src + width; + + for (i = 0; i < height; ++i) { + aom_memset16(dst_ptr1, src_ptr1[0], extend_left); + aom_memset16(dst_ptr2, src_ptr2[0], extend_right); + src_ptr1 += src_stride; + src_ptr2 += src_stride; + dst_ptr1 += src_stride; + dst_ptr2 += src_stride; + } + + /* Now copy the top and bottom lines into each line of the respective + * borders + */ + src_ptr1 = src - extend_left; + src_ptr2 = src + src_stride * (height - 1) - extend_left; + dst_ptr1 = src + src_stride * -extend_top - extend_left; + dst_ptr2 = src + src_stride * height - extend_left; + + for (i = 0; i < extend_top; ++i) { + memcpy(dst_ptr1, src_ptr1, linesize * sizeof(uint16_t)); + dst_ptr1 += src_stride; + } + + for (i = 0; i < extend_bottom; ++i) { + memcpy(dst_ptr2, src_ptr2, linesize * sizeof(uint16_t)); + dst_ptr2 += src_stride; + } +} + +void aom_yv12_extend_frame_borders_c(YV12_BUFFER_CONFIG *ybf, + const int num_planes) { + assert(ybf->border % 2 == 0); + assert(ybf->y_height - ybf->y_crop_height < 16); + assert(ybf->y_width - ybf->y_crop_width < 16); + assert(ybf->y_height - ybf->y_crop_height >= 0); + assert(ybf->y_width - ybf->y_crop_width >= 0); + + if (ybf->flags & YV12_FLAG_HIGHBITDEPTH) { + for (int plane = 0; plane < num_planes; ++plane) { + const int is_uv = plane > 0; + const int plane_border = ybf->border >> is_uv; + extend_plane_high( + ybf->buffers[plane], ybf->strides[is_uv], ybf->crop_widths[is_uv], + ybf->crop_heights[is_uv], plane_border, plane_border, + plane_border + ybf->heights[is_uv] - ybf->crop_heights[is_uv], + plane_border + ybf->widths[is_uv] - ybf->crop_widths[is_uv]); + } + return; + } + for (int plane = 0; plane < num_planes; ++plane) { + const int is_uv = plane > 0; + const int plane_border = ybf->border >> is_uv; + extend_plane(ybf->buffers[plane], ybf->strides[is_uv], + ybf->crop_widths[is_uv], ybf->crop_heights[is_uv], + plane_border, plane_border, + plane_border + ybf->heights[is_uv] - ybf->crop_heights[is_uv], + plane_border + ybf->widths[is_uv] - ybf->crop_widths[is_uv]); + } +} + +static void extend_frame(YV12_BUFFER_CONFIG *const ybf, int ext_size, + const int num_planes) { + const int ss_x = ybf->uv_width < ybf->y_width; + const int ss_y = ybf->uv_height < ybf->y_height; + + assert(ybf->y_height - ybf->y_crop_height < 16); + assert(ybf->y_width - ybf->y_crop_width < 16); + assert(ybf->y_height - ybf->y_crop_height >= 0); + assert(ybf->y_width - ybf->y_crop_width >= 0); + + if (ybf->flags & YV12_FLAG_HIGHBITDEPTH) { + for (int plane = 0; plane < num_planes; ++plane) { + const int is_uv = plane > 0; + const int top = ext_size >> (is_uv ? ss_y : 0); + const int left = ext_size >> (is_uv ? ss_x : 0); + const int bottom = top + ybf->heights[is_uv] - ybf->crop_heights[is_uv]; + const int right = left + ybf->widths[is_uv] - ybf->crop_widths[is_uv]; + extend_plane_high(ybf->buffers[plane], ybf->strides[is_uv], + ybf->crop_widths[is_uv], ybf->crop_heights[is_uv], top, + left, bottom, right); + } + return; + } + for (int plane = 0; plane < num_planes; ++plane) { + const int is_uv = plane > 0; + const int top = ext_size >> (is_uv ? ss_y : 0); + const int left = ext_size >> (is_uv ? ss_x : 0); + const int bottom = top + ybf->heights[is_uv] - ybf->crop_heights[is_uv]; + const int right = left + ybf->widths[is_uv] - ybf->crop_widths[is_uv]; + extend_plane(ybf->buffers[plane], ybf->strides[is_uv], + ybf->crop_widths[is_uv], ybf->crop_heights[is_uv], top, left, + bottom, right); + } +} + +void aom_extend_frame_borders_c(YV12_BUFFER_CONFIG *ybf, const int num_planes) { + extend_frame(ybf, ybf->border, num_planes); +} + +void aom_extend_frame_inner_borders_c(YV12_BUFFER_CONFIG *ybf, + const int num_planes) { + const int inner_bw = (ybf->border > AOMINNERBORDERINPIXELS) + ? AOMINNERBORDERINPIXELS + : ybf->border; + extend_frame(ybf, inner_bw, num_planes); +} + +void aom_extend_frame_borders_y_c(YV12_BUFFER_CONFIG *ybf) { + int ext_size = ybf->border; + assert(ybf->y_height - ybf->y_crop_height < 16); + assert(ybf->y_width - ybf->y_crop_width < 16); + assert(ybf->y_height - ybf->y_crop_height >= 0); + assert(ybf->y_width - ybf->y_crop_width >= 0); + + if (ybf->flags & YV12_FLAG_HIGHBITDEPTH) { + extend_plane_high(ybf->y_buffer, ybf->y_stride, ybf->y_crop_width, + ybf->y_crop_height, ext_size, ext_size, + ext_size + ybf->y_height - ybf->y_crop_height, + ext_size + ybf->y_width - ybf->y_crop_width); + return; + } + extend_plane(ybf->y_buffer, ybf->y_stride, ybf->y_crop_width, + ybf->y_crop_height, ext_size, ext_size, + ext_size + ybf->y_height - ybf->y_crop_height, + ext_size + ybf->y_width - ybf->y_crop_width); +} + +static void memcpy_short_addr(uint8_t *dst8, const uint8_t *src8, int num) { + uint16_t *dst = CONVERT_TO_SHORTPTR(dst8); + uint16_t *src = CONVERT_TO_SHORTPTR(src8); + memcpy(dst, src, num * sizeof(uint16_t)); +} + +// Copies the source image into the destination image and updates the +// destination's UMV borders. +// Note: The frames are assumed to be identical in size. +void aom_yv12_copy_frame_c(const YV12_BUFFER_CONFIG *src_bc, + YV12_BUFFER_CONFIG *dst_bc, const int num_planes) { +#if 0 + /* These assertions are valid in the codec, but the libaom-tester uses + * this code slightly differently. + */ + assert(src_bc->y_width == dst_bc->y_width); + assert(src_bc->y_height == dst_bc->y_height); +#endif + + assert((src_bc->flags & YV12_FLAG_HIGHBITDEPTH) == + (dst_bc->flags & YV12_FLAG_HIGHBITDEPTH)); + + if (src_bc->flags & YV12_FLAG_HIGHBITDEPTH) { + for (int plane = 0; plane < num_planes; ++plane) { + const uint8_t *plane_src = src_bc->buffers[plane]; + uint8_t *plane_dst = dst_bc->buffers[plane]; + const int is_uv = plane > 0; + + for (int row = 0; row < src_bc->heights[is_uv]; ++row) { + memcpy_short_addr(plane_dst, plane_src, src_bc->widths[is_uv]); + plane_src += src_bc->strides[is_uv]; + plane_dst += dst_bc->strides[is_uv]; + } + } + aom_yv12_extend_frame_borders_c(dst_bc, num_planes); + return; + } + for (int plane = 0; plane < num_planes; ++plane) { + const uint8_t *plane_src = src_bc->buffers[plane]; + uint8_t *plane_dst = dst_bc->buffers[plane]; + const int is_uv = plane > 0; + + for (int row = 0; row < src_bc->heights[is_uv]; ++row) { + memcpy(plane_dst, plane_src, src_bc->widths[is_uv]); + plane_src += src_bc->strides[is_uv]; + plane_dst += dst_bc->strides[is_uv]; + } + } + aom_yv12_extend_frame_borders_c(dst_bc, num_planes); +} + +void aom_yv12_copy_y_c(const YV12_BUFFER_CONFIG *src_ybc, + YV12_BUFFER_CONFIG *dst_ybc) { + int row; + const uint8_t *src = src_ybc->y_buffer; + uint8_t *dst = dst_ybc->y_buffer; + + if (src_ybc->flags & YV12_FLAG_HIGHBITDEPTH) { + const uint16_t *src16 = CONVERT_TO_SHORTPTR(src); + uint16_t *dst16 = CONVERT_TO_SHORTPTR(dst); + for (row = 0; row < src_ybc->y_height; ++row) { + memcpy(dst16, src16, src_ybc->y_width * sizeof(uint16_t)); + src16 += src_ybc->y_stride; + dst16 += dst_ybc->y_stride; + } + return; + } + + for (row = 0; row < src_ybc->y_height; ++row) { + memcpy(dst, src, src_ybc->y_width); + src += src_ybc->y_stride; + dst += dst_ybc->y_stride; + } +} + +void aom_yv12_copy_u_c(const YV12_BUFFER_CONFIG *src_bc, + YV12_BUFFER_CONFIG *dst_bc) { + int row; + const uint8_t *src = src_bc->u_buffer; + uint8_t *dst = dst_bc->u_buffer; + + if (src_bc->flags & YV12_FLAG_HIGHBITDEPTH) { + const uint16_t *src16 = CONVERT_TO_SHORTPTR(src); + uint16_t *dst16 = CONVERT_TO_SHORTPTR(dst); + for (row = 0; row < src_bc->uv_height; ++row) { + memcpy(dst16, src16, src_bc->uv_width * sizeof(uint16_t)); + src16 += src_bc->uv_stride; + dst16 += dst_bc->uv_stride; + } + return; + } + + for (row = 0; row < src_bc->uv_height; ++row) { + memcpy(dst, src, src_bc->uv_width); + src += src_bc->uv_stride; + dst += dst_bc->uv_stride; + } +} + +void aom_yv12_copy_v_c(const YV12_BUFFER_CONFIG *src_bc, + YV12_BUFFER_CONFIG *dst_bc) { + int row; + const uint8_t *src = src_bc->v_buffer; + uint8_t *dst = dst_bc->v_buffer; + + if (src_bc->flags & YV12_FLAG_HIGHBITDEPTH) { + const uint16_t *src16 = CONVERT_TO_SHORTPTR(src); + uint16_t *dst16 = CONVERT_TO_SHORTPTR(dst); + for (row = 0; row < src_bc->uv_height; ++row) { + memcpy(dst16, src16, src_bc->uv_width * sizeof(uint16_t)); + src16 += src_bc->uv_stride; + dst16 += dst_bc->uv_stride; + } + return; + } + + for (row = 0; row < src_bc->uv_height; ++row) { + memcpy(dst, src, src_bc->uv_width); + src += src_bc->uv_stride; + dst += dst_bc->uv_stride; + } +} + +void aom_yv12_partial_copy_y_c(const YV12_BUFFER_CONFIG *src_ybc, + YV12_BUFFER_CONFIG *dst_ybc, int hstart, + int hend, int vstart, int vend) { + int row; + const uint8_t *src = src_ybc->y_buffer; + uint8_t *dst = dst_ybc->y_buffer; + + if (src_ybc->flags & YV12_FLAG_HIGHBITDEPTH) { + const uint16_t *src16 = + CONVERT_TO_SHORTPTR(src + vstart * src_ybc->y_stride + hstart); + uint16_t *dst16 = + CONVERT_TO_SHORTPTR(dst + vstart * dst_ybc->y_stride + hstart); + for (row = vstart; row < vend; ++row) { + memcpy(dst16, src16, (hend - hstart) * sizeof(uint16_t)); + src16 += src_ybc->y_stride; + dst16 += dst_ybc->y_stride; + } + return; + } + src = (src + vstart * src_ybc->y_stride + hstart); + dst = (dst + vstart * dst_ybc->y_stride + hstart); + + for (row = vstart; row < vend; ++row) { + memcpy(dst, src, (hend - hstart)); + src += src_ybc->y_stride; + dst += dst_ybc->y_stride; + } +} + +void aom_yv12_partial_copy_u_c(const YV12_BUFFER_CONFIG *src_bc, + YV12_BUFFER_CONFIG *dst_bc, int hstart, int hend, + int vstart, int vend) { + int row; + const uint8_t *src = src_bc->u_buffer; + uint8_t *dst = dst_bc->u_buffer; + + if (src_bc->flags & YV12_FLAG_HIGHBITDEPTH) { + const uint16_t *src16 = + CONVERT_TO_SHORTPTR(src + vstart * src_bc->uv_stride + hstart); + uint16_t *dst16 = + CONVERT_TO_SHORTPTR(dst + vstart * dst_bc->uv_stride + hstart); + for (row = vstart; row < vend; ++row) { + memcpy(dst16, src16, (hend - hstart) * sizeof(uint16_t)); + src16 += src_bc->uv_stride; + dst16 += dst_bc->uv_stride; + } + return; + } + + src = (src + vstart * src_bc->uv_stride + hstart); + dst = (dst + vstart * dst_bc->uv_stride + hstart); + + for (row = vstart; row < vend; ++row) { + memcpy(dst, src, (hend - hstart)); + src += src_bc->uv_stride; + dst += dst_bc->uv_stride; + } +} + +void aom_yv12_partial_copy_v_c(const YV12_BUFFER_CONFIG *src_bc, + YV12_BUFFER_CONFIG *dst_bc, int hstart, int hend, + int vstart, int vend) { + int row; + const uint8_t *src = src_bc->v_buffer; + uint8_t *dst = dst_bc->v_buffer; + + if (src_bc->flags & YV12_FLAG_HIGHBITDEPTH) { + const uint16_t *src16 = + CONVERT_TO_SHORTPTR(src + vstart * src_bc->uv_stride + hstart); + uint16_t *dst16 = + CONVERT_TO_SHORTPTR(dst + vstart * dst_bc->uv_stride + hstart); + for (row = vstart; row < vend; ++row) { + memcpy(dst16, src16, (hend - hstart) * sizeof(uint16_t)); + src16 += src_bc->uv_stride; + dst16 += dst_bc->uv_stride; + } + return; + } + + src = (src + vstart * src_bc->uv_stride + hstart); + dst = (dst + vstart * dst_bc->uv_stride + hstart); + + for (row = vstart; row < vend; ++row) { + memcpy(dst, src, (hend - hstart)); + src += src_bc->uv_stride; + dst += dst_bc->uv_stride; + } +} diff --git a/third_party/aom/aom_scale/mips/dspr2/yv12extend_dspr2.c b/third_party/aom/aom_scale/mips/dspr2/yv12extend_dspr2.c new file mode 100644 index 0000000000..869e594d79 --- /dev/null +++ b/third_party/aom/aom_scale/mips/dspr2/yv12extend_dspr2.c @@ -0,0 +1,142 @@ +/* + * Copyright (c) 2016, Alliance for Open Media. All rights reserved + * + * This source code is subject to the terms of the BSD 2 Clause License and + * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License + * was not distributed with this source code in the LICENSE file, you can + * obtain it at www.aomedia.org/license/software. If the Alliance for Open + * Media Patent License 1.0 was not distributed with this source code in the + * PATENTS file, you can obtain it at www.aomedia.org/license/patent. + */ + +#include + +#include "config/aom_config.h" + +#include "aom_scale/yv12config.h" +#include "aom_mem/aom_mem.h" +#include "aom_scale/aom_scale.h" + +#if HAVE_DSPR2 +static void extend_plane(uint8_t *const src, int src_stride, int width, + int height, int extend_top, int extend_left, + int extend_bottom, int extend_right) { + int i, j; + uint8_t *left_src, *right_src; + uint8_t *left_dst_start, *right_dst_start; + uint8_t *left_dst, *right_dst; + uint8_t *top_src, *bot_src; + uint8_t *top_dst, *bot_dst; + uint32_t left_pix; + uint32_t right_pix; + uint32_t linesize; + + /* copy the left and right most columns out */ + left_src = src; + right_src = src + width - 1; + left_dst_start = src - extend_left; + right_dst_start = src + width; + + for (i = height; i--;) { + left_dst = left_dst_start; + right_dst = right_dst_start; + + __asm__ __volatile__( + "lb %[left_pix], 0(%[left_src]) \n\t" + "lb %[right_pix], 0(%[right_src]) \n\t" + "replv.qb %[left_pix], %[left_pix] \n\t" + "replv.qb %[right_pix], %[right_pix] \n\t" + + : [left_pix] "=&r"(left_pix), [right_pix] "=&r"(right_pix) + : [left_src] "r"(left_src), [right_src] "r"(right_src)); + + for (j = extend_left / 4; j--;) { + __asm__ __volatile__( + "sw %[left_pix], 0(%[left_dst]) \n\t" + "sw %[right_pix], 0(%[right_dst]) \n\t" + + : + : [left_dst] "r"(left_dst), [left_pix] "r"(left_pix), + [right_dst] "r"(right_dst), [right_pix] "r"(right_pix)); + + left_dst += 4; + right_dst += 4; + } + + for (j = extend_left % 4; j--;) { + __asm__ __volatile__( + "sb %[left_pix], 0(%[left_dst]) \n\t" + "sb %[right_pix], 0(%[right_dst]) \n\t" + + : + : [left_dst] "r"(left_dst), [left_pix] "r"(left_pix), + [right_dst] "r"(right_dst), [right_pix] "r"(right_pix)); + + left_dst += 1; + right_dst += 1; + } + + left_src += src_stride; + right_src += src_stride; + left_dst_start += src_stride; + right_dst_start += src_stride; + } + + /* Now copy the top and bottom lines into each line of the respective + * borders + */ + top_src = src - extend_left; + bot_src = src + src_stride * (height - 1) - extend_left; + top_dst = src + src_stride * (-extend_top) - extend_left; + bot_dst = src + src_stride * (height)-extend_left; + linesize = extend_left + extend_right + width; + + for (i = 0; i < extend_top; i++) { + memcpy(top_dst, top_src, linesize); + top_dst += src_stride; + } + + for (i = 0; i < extend_bottom; i++) { + memcpy(bot_dst, bot_src, linesize); + bot_dst += src_stride; + } +} + +static void extend_frame(YV12_BUFFER_CONFIG *const ybf, int ext_size) { + const int c_w = ybf->uv_crop_width; + const int c_h = ybf->uv_crop_height; + const int ss_x = ybf->uv_width < ybf->y_width; + const int ss_y = ybf->uv_height < ybf->y_height; + const int c_et = ext_size >> ss_y; + const int c_el = ext_size >> ss_x; + const int c_eb = c_et + ybf->uv_height - ybf->uv_crop_height; + const int c_er = c_el + ybf->uv_width - ybf->uv_crop_width; + + assert(ybf->y_height - ybf->y_crop_height < 16); + assert(ybf->y_width - ybf->y_crop_width < 16); + assert(ybf->y_height - ybf->y_crop_height >= 0); + assert(ybf->y_width - ybf->y_crop_width >= 0); + + extend_plane(ybf->y_buffer, ybf->y_stride, ybf->y_crop_width, + ybf->y_crop_height, ext_size, ext_size, + ext_size + ybf->y_height - ybf->y_crop_height, + ext_size + ybf->y_width - ybf->y_crop_width); + + extend_plane(ybf->u_buffer, ybf->uv_stride, c_w, c_h, c_et, c_el, c_eb, c_er); + + extend_plane(ybf->v_buffer, ybf->uv_stride, c_w, c_h, c_et, c_el, c_eb, c_er); +} + +void aom_extend_frame_borders_dspr2(YV12_BUFFER_CONFIG *ybf, + const int num_planes) { + extend_frame(ybf, ybf->border, num_planes); +} + +void aom_extend_frame_inner_borders_dspr2(YV12_BUFFER_CONFIG *ybf, + const int num_planes) { + const int inner_bw = (ybf->border > AOMINNERBORDERINPIXELS) + ? AOMINNERBORDERINPIXELS + : ybf->border; + extend_frame(ybf, inner_bw, num_planes); +} +#endif diff --git a/third_party/aom/aom_scale/yv12config.h b/third_party/aom/aom_scale/yv12config.h new file mode 100644 index 0000000000..2fb81acd7c --- /dev/null +++ b/third_party/aom/aom_scale/yv12config.h @@ -0,0 +1,143 @@ +/* + * Copyright (c) 2016, Alliance for Open Media. All rights reserved + * + * This source code is subject to the terms of the BSD 2 Clause License and + * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License + * was not distributed with this source code in the LICENSE file, you can + * obtain it at www.aomedia.org/license/software. If the Alliance for Open + * Media Patent License 1.0 was not distributed with this source code in the + * PATENTS file, you can obtain it at www.aomedia.org/license/patent. + */ + +#ifndef AOM_AOM_SCALE_YV12CONFIG_H_ +#define AOM_AOM_SCALE_YV12CONFIG_H_ + +#ifdef __cplusplus +extern "C" { +#endif + +#include "config/aom_config.h" + +#include "aom/aom_codec.h" +#include "aom/aom_frame_buffer.h" +#include "aom/aom_integer.h" + +#define AOMINNERBORDERINPIXELS 160 +#define AOM_INTERP_EXTEND 4 + +// TODO(jingning): Use unified inter predictor for encoder and +// decoder during the development process. Revisit the frame border +// to improve the decoder performance. +#if CONFIG_REDUCED_ENCODER_BORDER +#define AOM_BORDER_IN_PIXELS 160 +#else +#define AOM_BORDER_IN_PIXELS 288 +#endif // CONFIG_REDUCED_ENCODER_BORDER + +typedef struct yv12_buffer_config { + union { + struct { + int y_width; + int uv_width; + int alpha_width; + }; + int widths[3]; + }; + union { + struct { + int y_height; + int uv_height; + int alpha_height; + }; + int heights[3]; + }; + union { + struct { + int y_crop_width; + int uv_crop_width; + }; + int crop_widths[2]; + }; + union { + struct { + int y_crop_height; + int uv_crop_height; + }; + int crop_heights[2]; + }; + union { + struct { + int y_stride; + int uv_stride; + int alpha_stride; + }; + int strides[3]; + }; + union { + struct { + uint8_t *y_buffer; + uint8_t *u_buffer; + uint8_t *v_buffer; + uint8_t *alpha_buffer; + }; + uint8_t *buffers[4]; + }; + + // Indicate whether y_buffer, u_buffer, and v_buffer points to the internally + // allocated memory or external buffers. + int use_external_reference_buffers; + // This is needed to store y_buffer, u_buffer, and v_buffer when set reference + // uses an external refernece, and restore those buffer pointers after the + // external reference frame is no longer used. + uint8_t *store_buf_adr[3]; + + // If the frame is stored in a 16-bit buffer, this stores an 8-bit version + // for use in global motion detection. It is allocated on-demand. + uint8_t *y_buffer_8bit; + int buf_8bit_valid; + + uint8_t *buffer_alloc; + size_t buffer_alloc_sz; + int border; + size_t frame_size; + int subsampling_x; + int subsampling_y; + unsigned int bit_depth; + aom_color_primaries_t color_primaries; + aom_transfer_characteristics_t transfer_characteristics; + aom_matrix_coefficients_t matrix_coefficients; + int monochrome; + aom_chroma_sample_position_t chroma_sample_position; + aom_color_range_t color_range; + int render_width; + int render_height; + + int corrupted; + int flags; +} YV12_BUFFER_CONFIG; + +#define YV12_FLAG_HIGHBITDEPTH 8 + +int aom_alloc_frame_buffer(YV12_BUFFER_CONFIG *ybf, int width, int height, + int ss_x, int ss_y, int use_highbitdepth, int border, + int byte_alignment); + +// Updates the yv12 buffer config with the frame buffer. |byte_alignment| must +// be a power of 2, from 32 to 1024. 0 sets legacy alignment. If cb is not +// NULL, then libaom is using the frame buffer callbacks to handle memory. +// If cb is not NULL, libaom will call cb with minimum size in bytes needed +// to decode the current frame. If cb is NULL, libaom will allocate memory +// internally to decode the current frame. Returns 0 on success. Returns < 0 +// on failure. +int aom_realloc_frame_buffer(YV12_BUFFER_CONFIG *ybf, int width, int height, + int ss_x, int ss_y, int use_highbitdepth, + int border, int byte_alignment, + aom_codec_frame_buffer_t *fb, + aom_get_frame_buffer_cb_fn_t cb, void *cb_priv); +int aom_free_frame_buffer(YV12_BUFFER_CONFIG *ybf); + +#ifdef __cplusplus +} +#endif + +#endif // AOM_AOM_SCALE_YV12CONFIG_H_ -- cgit v1.2.3