summaryrefslogtreecommitdiffstats
path: root/third_party/aom/av1/common/x86
diff options
context:
space:
mode:
Diffstat (limited to '')
-rw-r--r--third_party/aom/av1/common/x86/cdef_block_sse2.c40
-rw-r--r--third_party/aom/av1/common/x86/cdef_block_ssse3.c11
-rw-r--r--third_party/aom/av1/common/x86/convolve_2d_avx2.c18
-rw-r--r--third_party/aom/av1/common/x86/convolve_2d_sse2.c17
-rw-r--r--third_party/aom/av1/common/x86/convolve_sse2.c26
-rw-r--r--third_party/aom/av1/common/x86/jnt_convolve_sse2.c229
6 files changed, 40 insertions, 301 deletions
diff --git a/third_party/aom/av1/common/x86/cdef_block_sse2.c b/third_party/aom/av1/common/x86/cdef_block_sse2.c
deleted file mode 100644
index 5ab7ffa2ff..0000000000
--- a/third_party/aom/av1/common/x86/cdef_block_sse2.c
+++ /dev/null
@@ -1,40 +0,0 @@
-/*
- * 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 "aom_dsp/aom_simd.h"
-#define SIMD_FUNC(name) name##_sse2
-#include "av1/common/cdef_block_simd.h"
-
-void cdef_find_dir_dual_sse2(const uint16_t *img1, const uint16_t *img2,
- int stride, int32_t *var_out_1st,
- int32_t *var_out_2nd, int coeff_shift,
- int *out_dir_1st_8x8, int *out_dir_2nd_8x8) {
- // Process first 8x8.
- *out_dir_1st_8x8 = cdef_find_dir(img1, stride, var_out_1st, coeff_shift);
-
- // Process second 8x8.
- *out_dir_2nd_8x8 = cdef_find_dir(img2, stride, var_out_2nd, coeff_shift);
-}
-
-void cdef_copy_rect8_8bit_to_16bit_sse2(uint16_t *dst, int dstride,
- const uint8_t *src, int sstride,
- int width, int height) {
- int j = 0;
- for (int i = 0; i < height; i++) {
- for (j = 0; j < (width & ~0x7); j += 8) {
- v64 row = v64_load_unaligned(&src[i * sstride + j]);
- v128_store_unaligned(&dst[i * dstride + j], v128_unpack_u8_s16(row));
- }
- for (; j < width; j++) {
- dst[i * dstride + j] = src[i * sstride + j];
- }
- }
-}
diff --git a/third_party/aom/av1/common/x86/cdef_block_ssse3.c b/third_party/aom/av1/common/x86/cdef_block_ssse3.c
index 0fb36eb6e0..14eb6c9e31 100644
--- a/third_party/aom/av1/common/x86/cdef_block_ssse3.c
+++ b/third_party/aom/av1/common/x86/cdef_block_ssse3.c
@@ -9,6 +9,17 @@
* PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/
+// Include SSSE3 CDEF code only for 32-bit x86, to support Valgrind.
+// For normal use, we require SSE4.1, so cdef_*_sse4_1 will be used instead of
+// these functions. However, 32-bit Valgrind does not support SSE4.1, so we
+// include a fallback to SSSE3 to improve performance
+
+#include "config/aom_config.h"
+
+#if !AOM_ARCH_X86
+#error "cdef_block_ssse3.c is included for compatibility with 32-bit x86 only"
+#endif // !AOM_ARCH_X86
+
#include "aom_dsp/aom_simd.h"
#define SIMD_FUNC(name) name##_ssse3
#include "av1/common/cdef_block_simd.h"
diff --git a/third_party/aom/av1/common/x86/convolve_2d_avx2.c b/third_party/aom/av1/common/x86/convolve_2d_avx2.c
index 1b39a0a8d5..d4c1169cc3 100644
--- a/third_party/aom/av1/common/x86/convolve_2d_avx2.c
+++ b/third_party/aom/av1/common/x86/convolve_2d_avx2.c
@@ -21,13 +21,11 @@
#include "av1/common/convolve.h"
-void av1_convolve_2d_sr_general_avx2(const uint8_t *src, int src_stride,
- uint8_t *dst, int dst_stride, int w, int h,
- const InterpFilterParams *filter_params_x,
- const InterpFilterParams *filter_params_y,
- const int subpel_x_qn,
- const int subpel_y_qn,
- ConvolveParams *conv_params) {
+static void convolve_2d_sr_general_avx2(
+ const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w,
+ int h, const InterpFilterParams *filter_params_x,
+ const InterpFilterParams *filter_params_y, const int subpel_x_qn,
+ const int subpel_y_qn, ConvolveParams *conv_params) {
if (filter_params_x->taps > 8) {
const int bd = 8;
int im_stride = 8, i;
@@ -150,9 +148,9 @@ void av1_convolve_2d_sr_avx2(
const bool use_general = (tap_x == 12 || tap_y == 12);
if (use_general) {
- av1_convolve_2d_sr_general_avx2(src, src_stride, dst, dst_stride, w, h,
- filter_params_x, filter_params_y,
- subpel_x_q4, subpel_y_q4, conv_params);
+ convolve_2d_sr_general_avx2(src, src_stride, dst, dst_stride, w, h,
+ filter_params_x, filter_params_y, subpel_x_q4,
+ subpel_y_q4, conv_params);
} else {
av1_convolve_2d_sr_specialized_avx2(src, src_stride, dst, dst_stride, w, h,
filter_params_x, filter_params_y,
diff --git a/third_party/aom/av1/common/x86/convolve_2d_sse2.c b/third_party/aom/av1/common/x86/convolve_2d_sse2.c
index 1b85f37294..68971eacc1 100644
--- a/third_party/aom/av1/common/x86/convolve_2d_sse2.c
+++ b/third_party/aom/av1/common/x86/convolve_2d_sse2.c
@@ -19,12 +19,11 @@
#include "aom_dsp/x86/convolve_common_intrin.h"
#include "av1/common/convolve.h"
-void av1_convolve_2d_sr_12tap_sse2(const uint8_t *src, int src_stride,
- uint8_t *dst, int dst_stride, int w, int h,
- const InterpFilterParams *filter_params_x,
- const InterpFilterParams *filter_params_y,
- const int subpel_x_qn, const int subpel_y_qn,
- ConvolveParams *conv_params) {
+static void convolve_2d_sr_12tap_sse2(
+ const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w,
+ int h, const InterpFilterParams *filter_params_x,
+ const InterpFilterParams *filter_params_y, const int subpel_x_qn,
+ const int subpel_y_qn, ConvolveParams *conv_params) {
const int bd = 8;
DECLARE_ALIGNED(16, int16_t,
@@ -231,9 +230,9 @@ void av1_convolve_2d_sr_sse2(const uint8_t *src, int src_stride, uint8_t *dst,
filter_params_x, filter_params_y, subpel_x_qn,
subpel_y_qn, conv_params);
} else {
- av1_convolve_2d_sr_12tap_sse2(src, src_stride, dst, dst_stride, w, h,
- filter_params_x, filter_params_y,
- subpel_x_qn, subpel_y_qn, conv_params);
+ convolve_2d_sr_12tap_sse2(src, src_stride, dst, dst_stride, w, h,
+ filter_params_x, filter_params_y, subpel_x_qn,
+ subpel_y_qn, conv_params);
}
} else {
const int bd = 8;
diff --git a/third_party/aom/av1/common/x86/convolve_sse2.c b/third_party/aom/av1/common/x86/convolve_sse2.c
index 012e75c1ae..6383567a48 100644
--- a/third_party/aom/av1/common/x86/convolve_sse2.c
+++ b/third_party/aom/av1/common/x86/convolve_sse2.c
@@ -75,10 +75,10 @@ static INLINE __m128i convolve_hi_y(const __m128i *const s,
return convolve(ss, coeffs);
}
-void av1_convolve_y_sr_12tap_sse2(const uint8_t *src, int src_stride,
- uint8_t *dst, int dst_stride, int w, int h,
- const InterpFilterParams *filter_params_y,
- int subpel_y_qn) {
+static void convolve_y_sr_12tap_sse2(const uint8_t *src, int src_stride,
+ uint8_t *dst, int dst_stride, int w, int h,
+ const InterpFilterParams *filter_params_y,
+ int subpel_y_qn) {
const int fo_vert = filter_params_y->taps / 2 - 1;
const uint8_t *src_ptr = src - fo_vert * src_stride;
const __m128i round_const = _mm_set1_epi32((1 << FILTER_BITS) >> 1);
@@ -185,8 +185,8 @@ void av1_convolve_y_sr_sse2(const uint8_t *src, int src_stride, uint8_t *dst,
av1_convolve_y_sr_c(src, src_stride, dst, dst_stride, w, h,
filter_params_y, subpel_y_qn);
} else {
- av1_convolve_y_sr_12tap_sse2(src, src_stride, dst, dst_stride, w, h,
- filter_params_y, subpel_y_qn);
+ convolve_y_sr_12tap_sse2(src, src_stride, dst, dst_stride, w, h,
+ filter_params_y, subpel_y_qn);
}
} else {
const int fo_vert = filter_params_y->taps / 2 - 1;
@@ -337,11 +337,11 @@ void av1_convolve_y_sr_sse2(const uint8_t *src, int src_stride, uint8_t *dst,
}
}
-void av1_convolve_x_sr_12tap_sse2(const uint8_t *src, int src_stride,
- uint8_t *dst, int dst_stride, int w, int h,
- const InterpFilterParams *filter_params_x,
- int subpel_x_qn,
- ConvolveParams *conv_params) {
+static void convolve_x_sr_12tap_sse2(const uint8_t *src, int src_stride,
+ uint8_t *dst, int dst_stride, int w, int h,
+ const InterpFilterParams *filter_params_x,
+ int subpel_x_qn,
+ ConvolveParams *conv_params) {
const int fo_horiz = filter_params_x->taps / 2 - 1;
const uint8_t *src_ptr = src - fo_horiz;
const int bits = FILTER_BITS - conv_params->round_0;
@@ -402,8 +402,8 @@ void av1_convolve_x_sr_sse2(const uint8_t *src, int src_stride, uint8_t *dst,
av1_convolve_x_sr_c(src, src_stride, dst, dst_stride, w, h,
filter_params_x, subpel_x_qn, conv_params);
} else {
- av1_convolve_x_sr_12tap_sse2(src, src_stride, dst, dst_stride, w, h,
- filter_params_x, subpel_x_qn, conv_params);
+ convolve_x_sr_12tap_sse2(src, src_stride, dst, dst_stride, w, h,
+ filter_params_x, subpel_x_qn, conv_params);
}
} else {
const int fo_horiz = filter_params_x->taps / 2 - 1;
diff --git a/third_party/aom/av1/common/x86/jnt_convolve_sse2.c b/third_party/aom/av1/common/x86/jnt_convolve_sse2.c
index 8c5d9918fb..d5d2db7455 100644
--- a/third_party/aom/av1/common/x86/jnt_convolve_sse2.c
+++ b/third_party/aom/av1/common/x86/jnt_convolve_sse2.c
@@ -375,232 +375,3 @@ void av1_dist_wtd_convolve_y_sse2(const uint8_t *src, int src_stride,
} while (j < w);
}
}
-
-void av1_dist_wtd_convolve_2d_sse2(const uint8_t *src, int src_stride,
- uint8_t *dst0, int dst_stride0, int w, int h,
- const InterpFilterParams *filter_params_x,
- const InterpFilterParams *filter_params_y,
- const int subpel_x_qn, const int subpel_y_qn,
- ConvolveParams *conv_params) {
- CONV_BUF_TYPE *dst = conv_params->dst;
- int dst_stride = conv_params->dst_stride;
- const int bd = 8;
-
- DECLARE_ALIGNED(16, int16_t,
- im_block[(MAX_SB_SIZE + MAX_FILTER_TAP - 1) * MAX_SB_SIZE]);
- int im_h = h + filter_params_y->taps - 1;
- int im_stride = MAX_SB_SIZE;
- int i, j;
- const int fo_vert = filter_params_y->taps / 2 - 1;
- const int fo_horiz = filter_params_x->taps / 2 - 1;
- const int do_average = conv_params->do_average;
- const int use_dist_wtd_comp_avg = conv_params->use_dist_wtd_comp_avg;
- const uint8_t *const src_ptr = src - fo_vert * src_stride - fo_horiz;
-
- const __m128i zero = _mm_setzero_si128();
-
- const int w0 = conv_params->fwd_offset;
- const int w1 = conv_params->bck_offset;
- const __m128i wt0 = _mm_set1_epi16(w0);
- const __m128i wt1 = _mm_set1_epi16(w1);
- const __m128i wt = _mm_unpacklo_epi16(wt0, wt1);
-
- const int offset_0 =
- bd + 2 * FILTER_BITS - conv_params->round_0 - conv_params->round_1;
- const int offset = (1 << offset_0) + (1 << (offset_0 - 1));
- const __m128i offset_const = _mm_set1_epi16(offset);
- const int rounding_shift =
- 2 * FILTER_BITS - conv_params->round_0 - conv_params->round_1;
- const __m128i rounding_const = _mm_set1_epi16((1 << rounding_shift) >> 1);
-
- /* Horizontal filter */
- {
- const int16_t *x_filter = av1_get_interp_filter_subpel_kernel(
- filter_params_x, subpel_x_qn & SUBPEL_MASK);
- const __m128i coeffs_x = _mm_loadu_si128((__m128i *)x_filter);
-
- // coeffs 0 1 0 1 2 3 2 3
- const __m128i tmp_0 = _mm_unpacklo_epi32(coeffs_x, coeffs_x);
- // coeffs 4 5 4 5 6 7 6 7
- const __m128i tmp_1 = _mm_unpackhi_epi32(coeffs_x, coeffs_x);
-
- // coeffs 0 1 0 1 0 1 0 1
- const __m128i coeff_01 = _mm_unpacklo_epi64(tmp_0, tmp_0);
- // coeffs 2 3 2 3 2 3 2 3
- const __m128i coeff_23 = _mm_unpackhi_epi64(tmp_0, tmp_0);
- // coeffs 4 5 4 5 4 5 4 5
- const __m128i coeff_45 = _mm_unpacklo_epi64(tmp_1, tmp_1);
- // coeffs 6 7 6 7 6 7 6 7
- const __m128i coeff_67 = _mm_unpackhi_epi64(tmp_1, tmp_1);
-
- const __m128i round_const = _mm_set1_epi32(
- ((1 << conv_params->round_0) >> 1) + (1 << (bd + FILTER_BITS - 1)));
- const __m128i round_shift = _mm_cvtsi32_si128(conv_params->round_0);
-
- for (i = 0; i < im_h; ++i) {
- for (j = 0; j < w; j += 8) {
- __m128i temp_lo, temp_hi;
- const __m128i data =
- _mm_loadu_si128((__m128i *)&src_ptr[i * src_stride + j]);
-
- const __m128i src_lo = _mm_unpacklo_epi8(data, zero);
- const __m128i src_hi = _mm_unpackhi_epi8(data, zero);
-
- // Filter even-index pixels
- const __m128i res_0 = _mm_madd_epi16(src_lo, coeff_01);
- temp_lo = _mm_srli_si128(src_lo, 4);
- temp_hi = _mm_slli_si128(src_hi, 12);
- const __m128i src_2 = _mm_or_si128(temp_hi, temp_lo);
- const __m128i res_2 = _mm_madd_epi16(src_2, coeff_23);
- temp_lo = _mm_srli_si128(src_lo, 8);
- temp_hi = _mm_slli_si128(src_hi, 8);
- const __m128i src_4 = _mm_or_si128(temp_hi, temp_lo);
- const __m128i res_4 = _mm_madd_epi16(src_4, coeff_45);
- temp_lo = _mm_srli_si128(src_lo, 12);
- temp_hi = _mm_slli_si128(src_hi, 4);
- const __m128i src_6 = _mm_or_si128(temp_hi, temp_lo);
- const __m128i res_6 = _mm_madd_epi16(src_6, coeff_67);
-
- __m128i res_even = _mm_add_epi32(_mm_add_epi32(res_0, res_4),
- _mm_add_epi32(res_2, res_6));
- res_even =
- _mm_sra_epi32(_mm_add_epi32(res_even, round_const), round_shift);
-
- // Filter odd-index pixels
- temp_lo = _mm_srli_si128(src_lo, 2);
- temp_hi = _mm_slli_si128(src_hi, 14);
- const __m128i src_1 = _mm_or_si128(temp_hi, temp_lo);
- const __m128i res_1 = _mm_madd_epi16(src_1, coeff_01);
- temp_lo = _mm_srli_si128(src_lo, 6);
- temp_hi = _mm_slli_si128(src_hi, 10);
- const __m128i src_3 = _mm_or_si128(temp_hi, temp_lo);
- const __m128i res_3 = _mm_madd_epi16(src_3, coeff_23);
- temp_lo = _mm_srli_si128(src_lo, 10);
- temp_hi = _mm_slli_si128(src_hi, 6);
- const __m128i src_5 = _mm_or_si128(temp_hi, temp_lo);
- const __m128i res_5 = _mm_madd_epi16(src_5, coeff_45);
- temp_lo = _mm_srli_si128(src_lo, 14);
- temp_hi = _mm_slli_si128(src_hi, 2);
- const __m128i src_7 = _mm_or_si128(temp_hi, temp_lo);
- const __m128i res_7 = _mm_madd_epi16(src_7, coeff_67);
-
- __m128i res_odd = _mm_add_epi32(_mm_add_epi32(res_1, res_5),
- _mm_add_epi32(res_3, res_7));
- res_odd =
- _mm_sra_epi32(_mm_add_epi32(res_odd, round_const), round_shift);
-
- // Pack in the column order 0, 2, 4, 6, 1, 3, 5, 7
- __m128i res = _mm_packs_epi32(res_even, res_odd);
- _mm_store_si128((__m128i *)&im_block[i * im_stride + j], res);
- }
- }
- }
-
- /* Vertical filter */
- {
- const int16_t *y_filter = av1_get_interp_filter_subpel_kernel(
- filter_params_y, subpel_y_qn & SUBPEL_MASK);
- const __m128i coeffs_y = _mm_loadu_si128((__m128i *)y_filter);
-
- // coeffs 0 1 0 1 2 3 2 3
- const __m128i tmp_0 = _mm_unpacklo_epi32(coeffs_y, coeffs_y);
- // coeffs 4 5 4 5 6 7 6 7
- const __m128i tmp_1 = _mm_unpackhi_epi32(coeffs_y, coeffs_y);
-
- // coeffs 0 1 0 1 0 1 0 1
- const __m128i coeff_01 = _mm_unpacklo_epi64(tmp_0, tmp_0);
- // coeffs 2 3 2 3 2 3 2 3
- const __m128i coeff_23 = _mm_unpackhi_epi64(tmp_0, tmp_0);
- // coeffs 4 5 4 5 4 5 4 5
- const __m128i coeff_45 = _mm_unpacklo_epi64(tmp_1, tmp_1);
- // coeffs 6 7 6 7 6 7 6 7
- const __m128i coeff_67 = _mm_unpackhi_epi64(tmp_1, tmp_1);
-
- const __m128i round_const = _mm_set1_epi32(
- ((1 << conv_params->round_1) >> 1) -
- (1 << (bd + 2 * FILTER_BITS - conv_params->round_0 - 1)));
- const __m128i round_shift = _mm_cvtsi32_si128(conv_params->round_1);
-
- for (i = 0; i < h; ++i) {
- for (j = 0; j < w; j += 8) {
- // Filter even-index pixels
- const int16_t *data = &im_block[i * im_stride + j];
- const __m128i src_0 =
- _mm_unpacklo_epi16(*(__m128i *)(data + 0 * im_stride),
- *(__m128i *)(data + 1 * im_stride));
- const __m128i src_2 =
- _mm_unpacklo_epi16(*(__m128i *)(data + 2 * im_stride),
- *(__m128i *)(data + 3 * im_stride));
- const __m128i src_4 =
- _mm_unpacklo_epi16(*(__m128i *)(data + 4 * im_stride),
- *(__m128i *)(data + 5 * im_stride));
- const __m128i src_6 =
- _mm_unpacklo_epi16(*(__m128i *)(data + 6 * im_stride),
- *(__m128i *)(data + 7 * im_stride));
-
- const __m128i res_0 = _mm_madd_epi16(src_0, coeff_01);
- const __m128i res_2 = _mm_madd_epi16(src_2, coeff_23);
- const __m128i res_4 = _mm_madd_epi16(src_4, coeff_45);
- const __m128i res_6 = _mm_madd_epi16(src_6, coeff_67);
-
- const __m128i res_even = _mm_add_epi32(_mm_add_epi32(res_0, res_2),
- _mm_add_epi32(res_4, res_6));
-
- // Filter odd-index pixels
- const __m128i src_1 =
- _mm_unpackhi_epi16(*(__m128i *)(data + 0 * im_stride),
- *(__m128i *)(data + 1 * im_stride));
- const __m128i src_3 =
- _mm_unpackhi_epi16(*(__m128i *)(data + 2 * im_stride),
- *(__m128i *)(data + 3 * im_stride));
- const __m128i src_5 =
- _mm_unpackhi_epi16(*(__m128i *)(data + 4 * im_stride),
- *(__m128i *)(data + 5 * im_stride));
- const __m128i src_7 =
- _mm_unpackhi_epi16(*(__m128i *)(data + 6 * im_stride),
- *(__m128i *)(data + 7 * im_stride));
-
- const __m128i res_1 = _mm_madd_epi16(src_1, coeff_01);
- const __m128i res_3 = _mm_madd_epi16(src_3, coeff_23);
- const __m128i res_5 = _mm_madd_epi16(src_5, coeff_45);
- const __m128i res_7 = _mm_madd_epi16(src_7, coeff_67);
-
- const __m128i res_odd = _mm_add_epi32(_mm_add_epi32(res_1, res_3),
- _mm_add_epi32(res_5, res_7));
-
- // Rearrange pixels back into the order 0 ... 7
- const __m128i res_lo = _mm_unpacklo_epi32(res_even, res_odd);
- const __m128i res_hi = _mm_unpackhi_epi32(res_even, res_odd);
-
- const __m128i res_lo_round =
- _mm_sra_epi32(_mm_add_epi32(res_lo, round_const), round_shift);
- const __m128i res_hi_round =
- _mm_sra_epi32(_mm_add_epi32(res_hi, round_const), round_shift);
-
- const __m128i res_16b = _mm_packs_epi32(res_lo_round, res_hi_round);
- const __m128i res_unsigned = _mm_add_epi16(res_16b, offset_const);
-
- // Accumulate values into the destination buffer
- if (do_average) {
- const __m128i data_ref_0 =
- _mm_loadu_si128((__m128i *)(&dst[i * dst_stride + j]));
-
- const __m128i comp_avg_res =
- comp_avg(&data_ref_0, &res_unsigned, &wt, use_dist_wtd_comp_avg);
-
- const __m128i round_result = convolve_rounding(
- &comp_avg_res, &offset_const, &rounding_const, rounding_shift);
-
- const __m128i res_8 = _mm_packus_epi16(round_result, round_result);
-
- if (w > 4)
- _mm_storel_epi64((__m128i *)(&dst0[i * dst_stride0 + j]), res_8);
- else
- *(int *)(&dst0[i * dst_stride0 + j]) = _mm_cvtsi128_si32(res_8);
- } else {
- _mm_store_si128((__m128i *)(&dst[i * dst_stride + j]), res_unsigned);
- }
- }
- }
- }
-}
generated by cgit v1.2.3 (git 2.39.1) at 2024-07-19 05:12:52 +0000