Skip to content

GitLab

T
tangled
Commit 985d66b3 authored by Johann's avatar Johann Committed by Commit Bot
Browse files
Options

libaom: require sse2 

Cherry pick rtcd-related changes and re-run cmake_update.sh

This allows the linker to remove unused C code when an sse2 version
is available. See discussion on libvpx:
https://chromium-review.googlesource.com/c/chromium/src/+/1195800

R=tomfinegan@chromium.org

Change-Id: I419a87e6d0b4f9166754aaaedf6b9c8cd1834dae
Reviewed-on: https://chromium-review.googlesource.com/1197466Reviewed-by: default avatarTom Finegan <tomfinegan@chromium.org>
Commit-Queue: Johann Koenig <johannkoenig@google.com>
Cr-Commit-Position: refs/heads/master@{#587780}
parent d228e1a8
Hide whitespace changes
Inline Side-by-side
Showing with 460 additions and 2979 deletions
third_party/libaom/cmake_update.sh
View file @ 985d66b3
...@@ -126,7 +126,8 @@ cp libaom_srcs.gni "${BASE}" ...@@ -126,7 +126,8 @@ cp libaom_srcs.gni "${BASE}"
cp config/aom_version.h "${CFG}/config/" cp config/aom_version.h "${CFG}/config/"
reset_dirs linux/ia32 reset_dirs linux/ia32
gen_config_files linux/ia32 "${toolchain}/x86-linux.cmake ${all_platforms}" gen_config_files linux/ia32 "${toolchain}/x86-linux.cmake ${all_platforms} \
-DAOM_RTCD_FLAGS=--require-mmx;--require-sse;--require-sse2"
reset_dirs linux/x64 reset_dirs linux/x64
gen_config_files linux/x64 "${all_platforms}" gen_config_files linux/x64 "${all_platforms}"
......
third_party/libaom/source/config/linux/ia32/config/aom_config.c
View file @ 985d66b3
...@@ -9,5 +9,5 @@ ...@@ -9,5 +9,5 @@
* PATENTS file, you can obtain it at www.aomedia.org/license/patent. * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/ */
#include "aom/aom_codec.h" #include "aom/aom_codec.h"
static const char* const cfg = "cmake ../source/libaom -G \"Unix Makefiles\" -DCMAKE_TOOLCHAIN_FILE=\"../source/libaom/build/cmake/toolchains/x86-linux.cmake\" -DCONFIG_AV1_ENCODER=0 -DCONFIG_LOWBITDEPTH=1 -DCONFIG_MAX_DECODE_PROFILE=0 -DCONFIG_NORMAL_TILE_MODE=1 -DCONFIG_SIZE_LIMIT=1 -DDECODE_HEIGHT_LIMIT=16384 -DDECODE_WIDTH_LIMIT=16384"; static const char* const cfg = "cmake ../source/libaom -G \"Unix Makefiles\" -DCMAKE_TOOLCHAIN_FILE=\"../source/libaom/build/cmake/toolchains/x86-linux.cmake\" -DCONFIG_AV1_ENCODER=0 -DCONFIG_LOWBITDEPTH=1 -DCONFIG_MAX_DECODE_PROFILE=0 -DCONFIG_NORMAL_TILE_MODE=1 -DCONFIG_SIZE_LIMIT=1 -DDECODE_HEIGHT_LIMIT=16384 -DDECODE_WIDTH_LIMIT=16384 -DAOM_RTCD_FLAGS=--require-mmx;--require-sse;--require-sse2";
const char *aom_codec_build_config(void) {return cfg;} const char *aom_codec_build_config(void) {return cfg;}
third_party/libaom/source/config/linux/ia32/config/aom_dsp_rtcd.h
View file @ 985d66b3
This source diff could not be displayed because it is too large. You can view the blob instead.
third_party/libaom/source/config/linux/ia32/config/av1_rtcd.h
View file @ 985d66b3
...@@ -2071,30 +2071,22 @@ static void setup_rtcd_internal(void) { ...@@ -2071,30 +2071,22 @@ static void setup_rtcd_internal(void) {
if (flags & HAS_AVX2) if (flags & HAS_AVX2)
av1_build_compound_diffwtd_mask_highbd = av1_build_compound_diffwtd_mask_highbd =
av1_build_compound_diffwtd_mask_highbd_avx2; av1_build_compound_diffwtd_mask_highbd_avx2;
av1_convolve_2d_copy_sr = av1_convolve_2d_copy_sr_c; av1_convolve_2d_copy_sr = av1_convolve_2d_copy_sr_sse2;
if (flags & HAS_SSE2)
av1_convolve_2d_copy_sr = av1_convolve_2d_copy_sr_sse2;
if (flags & HAS_AVX2) if (flags & HAS_AVX2)
av1_convolve_2d_copy_sr = av1_convolve_2d_copy_sr_avx2; av1_convolve_2d_copy_sr = av1_convolve_2d_copy_sr_avx2;
av1_convolve_2d_scale = av1_convolve_2d_scale_c; av1_convolve_2d_scale = av1_convolve_2d_scale_c;
if (flags & HAS_SSE4_1) if (flags & HAS_SSE4_1)
av1_convolve_2d_scale = av1_convolve_2d_scale_sse4_1; av1_convolve_2d_scale = av1_convolve_2d_scale_sse4_1;
av1_convolve_2d_sr = av1_convolve_2d_sr_c; av1_convolve_2d_sr = av1_convolve_2d_sr_sse2;
if (flags & HAS_SSE2)
av1_convolve_2d_sr = av1_convolve_2d_sr_sse2;
if (flags & HAS_AVX2) if (flags & HAS_AVX2)
av1_convolve_2d_sr = av1_convolve_2d_sr_avx2; av1_convolve_2d_sr = av1_convolve_2d_sr_avx2;
av1_convolve_horiz_rs = av1_convolve_horiz_rs_c; av1_convolve_horiz_rs = av1_convolve_horiz_rs_c;
if (flags & HAS_SSE4_1) if (flags & HAS_SSE4_1)
av1_convolve_horiz_rs = av1_convolve_horiz_rs_sse4_1; av1_convolve_horiz_rs = av1_convolve_horiz_rs_sse4_1;
av1_convolve_x_sr = av1_convolve_x_sr_c; av1_convolve_x_sr = av1_convolve_x_sr_sse2;
if (flags & HAS_SSE2)
av1_convolve_x_sr = av1_convolve_x_sr_sse2;
if (flags & HAS_AVX2) if (flags & HAS_AVX2)
av1_convolve_x_sr = av1_convolve_x_sr_avx2; av1_convolve_x_sr = av1_convolve_x_sr_avx2;
av1_convolve_y_sr = av1_convolve_y_sr_c; av1_convolve_y_sr = av1_convolve_y_sr_sse2;
if (flags & HAS_SSE2)
av1_convolve_y_sr = av1_convolve_y_sr_sse2;
if (flags & HAS_AVX2) if (flags & HAS_AVX2)
av1_convolve_y_sr = av1_convolve_y_sr_avx2; av1_convolve_y_sr = av1_convolve_y_sr_avx2;
av1_filter_intra_edge = av1_filter_intra_edge_c; av1_filter_intra_edge = av1_filter_intra_edge_c;
...@@ -2106,9 +2098,7 @@ static void setup_rtcd_internal(void) { ...@@ -2106,9 +2098,7 @@ static void setup_rtcd_internal(void) {
av1_filter_intra_predictor = av1_filter_intra_predictor_c; av1_filter_intra_predictor = av1_filter_intra_predictor_c;
if (flags & HAS_SSE4_1) if (flags & HAS_SSE4_1)
av1_filter_intra_predictor = av1_filter_intra_predictor_sse4_1; av1_filter_intra_predictor = av1_filter_intra_predictor_sse4_1;
av1_highbd_convolve_2d_copy_sr = av1_highbd_convolve_2d_copy_sr_c; av1_highbd_convolve_2d_copy_sr = av1_highbd_convolve_2d_copy_sr_sse2;
if (flags & HAS_SSE2)
av1_highbd_convolve_2d_copy_sr = av1_highbd_convolve_2d_copy_sr_sse2;
if (flags & HAS_AVX2) if (flags & HAS_AVX2)
av1_highbd_convolve_2d_copy_sr = av1_highbd_convolve_2d_copy_sr_avx2; av1_highbd_convolve_2d_copy_sr = av1_highbd_convolve_2d_copy_sr_avx2;
av1_highbd_convolve_2d_scale = av1_highbd_convolve_2d_scale_c; av1_highbd_convolve_2d_scale = av1_highbd_convolve_2d_scale_c;
...@@ -2198,19 +2188,13 @@ static void setup_rtcd_internal(void) { ...@@ -2198,19 +2188,13 @@ static void setup_rtcd_internal(void) {
av1_jnt_convolve_2d = av1_jnt_convolve_2d_ssse3; av1_jnt_convolve_2d = av1_jnt_convolve_2d_ssse3;
if (flags & HAS_AVX2) if (flags & HAS_AVX2)
av1_jnt_convolve_2d = av1_jnt_convolve_2d_avx2; av1_jnt_convolve_2d = av1_jnt_convolve_2d_avx2;
av1_jnt_convolve_2d_copy = av1_jnt_convolve_2d_copy_c; av1_jnt_convolve_2d_copy = av1_jnt_convolve_2d_copy_sse2;
if (flags & HAS_SSE2)
av1_jnt_convolve_2d_copy = av1_jnt_convolve_2d_copy_sse2;
if (flags & HAS_AVX2) if (flags & HAS_AVX2)
av1_jnt_convolve_2d_copy = av1_jnt_convolve_2d_copy_avx2; av1_jnt_convolve_2d_copy = av1_jnt_convolve_2d_copy_avx2;
av1_jnt_convolve_x = av1_jnt_convolve_x_c; av1_jnt_convolve_x = av1_jnt_convolve_x_sse2;
if (flags & HAS_SSE2)
av1_jnt_convolve_x = av1_jnt_convolve_x_sse2;
if (flags & HAS_AVX2) if (flags & HAS_AVX2)
av1_jnt_convolve_x = av1_jnt_convolve_x_avx2; av1_jnt_convolve_x = av1_jnt_convolve_x_avx2;
av1_jnt_convolve_y = av1_jnt_convolve_y_c; av1_jnt_convolve_y = av1_jnt_convolve_y_sse2;
if (flags & HAS_SSE2)
av1_jnt_convolve_y = av1_jnt_convolve_y_sse2;
if (flags & HAS_AVX2) if (flags & HAS_AVX2)
av1_jnt_convolve_y = av1_jnt_convolve_y_avx2; av1_jnt_convolve_y = av1_jnt_convolve_y_avx2;
av1_selfguided_restoration = av1_selfguided_restoration_c; av1_selfguided_restoration = av1_selfguided_restoration_c;
...@@ -2227,23 +2211,17 @@ static void setup_rtcd_internal(void) { ...@@ -2227,23 +2211,17 @@ static void setup_rtcd_internal(void) {
av1_warp_affine = av1_warp_affine_c; av1_warp_affine = av1_warp_affine_c;
if (flags & HAS_SSE4_1) if (flags & HAS_SSE4_1)
av1_warp_affine = av1_warp_affine_sse4_1; av1_warp_affine = av1_warp_affine_sse4_1;
av1_wiener_convolve_add_src = av1_wiener_convolve_add_src_c; av1_wiener_convolve_add_src = av1_wiener_convolve_add_src_sse2;
if (flags & HAS_SSE2)
av1_wiener_convolve_add_src = av1_wiener_convolve_add_src_sse2;
if (flags & HAS_AVX2) if (flags & HAS_AVX2)
av1_wiener_convolve_add_src = av1_wiener_convolve_add_src_avx2; av1_wiener_convolve_add_src = av1_wiener_convolve_add_src_avx2;
cdef_filter_block = cdef_filter_block_c; cdef_filter_block = cdef_filter_block_sse2;
if (flags & HAS_SSE2)
cdef_filter_block = cdef_filter_block_sse2;
if (flags & HAS_SSSE3) if (flags & HAS_SSSE3)
cdef_filter_block = cdef_filter_block_ssse3; cdef_filter_block = cdef_filter_block_ssse3;
if (flags & HAS_SSE4_1) if (flags & HAS_SSE4_1)
cdef_filter_block = cdef_filter_block_sse4_1; cdef_filter_block = cdef_filter_block_sse4_1;
if (flags & HAS_AVX2) if (flags & HAS_AVX2)
cdef_filter_block = cdef_filter_block_avx2; cdef_filter_block = cdef_filter_block_avx2;
cdef_find_dir = cdef_find_dir_c; cdef_find_dir = cdef_find_dir_sse2;
if (flags & HAS_SSE2)
cdef_find_dir = cdef_find_dir_sse2;
if (flags & HAS_SSSE3) if (flags & HAS_SSSE3)
cdef_find_dir = cdef_find_dir_ssse3; cdef_find_dir = cdef_find_dir_ssse3;
if (flags & HAS_SSE4_1) if (flags & HAS_SSE4_1)
...@@ -2280,18 +2258,14 @@ static void setup_rtcd_internal(void) { ...@@ -2280,18 +2258,14 @@ static void setup_rtcd_internal(void) {
cfl_get_luma_subsampling_444_lbd = cfl_get_luma_subsampling_444_lbd_ssse3; cfl_get_luma_subsampling_444_lbd = cfl_get_luma_subsampling_444_lbd_ssse3;
if (flags & HAS_AVX2) if (flags & HAS_AVX2)
cfl_get_luma_subsampling_444_lbd = cfl_get_luma_subsampling_444_lbd_avx2; cfl_get_luma_subsampling_444_lbd = cfl_get_luma_subsampling_444_lbd_avx2;
copy_rect8_16bit_to_16bit = copy_rect8_16bit_to_16bit_c; copy_rect8_16bit_to_16bit = copy_rect8_16bit_to_16bit_sse2;
if (flags & HAS_SSE2)
copy_rect8_16bit_to_16bit = copy_rect8_16bit_to_16bit_sse2;
if (flags & HAS_SSSE3) if (flags & HAS_SSSE3)
copy_rect8_16bit_to_16bit = copy_rect8_16bit_to_16bit_ssse3; copy_rect8_16bit_to_16bit = copy_rect8_16bit_to_16bit_ssse3;
if (flags & HAS_SSE4_1) if (flags & HAS_SSE4_1)
copy_rect8_16bit_to_16bit = copy_rect8_16bit_to_16bit_sse4_1; copy_rect8_16bit_to_16bit = copy_rect8_16bit_to_16bit_sse4_1;
if (flags & HAS_AVX2) if (flags & HAS_AVX2)
copy_rect8_16bit_to_16bit = copy_rect8_16bit_to_16bit_avx2; copy_rect8_16bit_to_16bit = copy_rect8_16bit_to_16bit_avx2;
copy_rect8_8bit_to_16bit = copy_rect8_8bit_to_16bit_c; copy_rect8_8bit_to_16bit = copy_rect8_8bit_to_16bit_sse2;
if (flags & HAS_SSE2)
copy_rect8_8bit_to_16bit = copy_rect8_8bit_to_16bit_sse2;
if (flags & HAS_SSSE3) if (flags & HAS_SSSE3)
copy_rect8_8bit_to_16bit = copy_rect8_8bit_to_16bit_ssse3; copy_rect8_8bit_to_16bit = copy_rect8_8bit_to_16bit_ssse3;
if (flags & HAS_SSE4_1) if (flags & HAS_SSE4_1)
...@@ -2308,9 +2282,7 @@ static void setup_rtcd_internal(void) { ...@@ -2308,9 +2282,7 @@ static void setup_rtcd_internal(void) {
get_predict_lbd_fn = get_predict_lbd_fn_ssse3; get_predict_lbd_fn = get_predict_lbd_fn_ssse3;
if (flags & HAS_AVX2) if (flags & HAS_AVX2)
get_predict_lbd_fn = get_predict_lbd_fn_avx2; get_predict_lbd_fn = get_predict_lbd_fn_avx2;
get_subtract_average_fn = get_subtract_average_fn_c; get_subtract_average_fn = get_subtract_average_fn_sse2;
if (flags & HAS_SSE2)
get_subtract_average_fn = get_subtract_average_fn_sse2;
if (flags & HAS_AVX2) if (flags & HAS_AVX2)
get_subtract_average_fn = get_subtract_average_fn_avx2; get_subtract_average_fn = get_subtract_average_fn_avx2;
} }
......
third_party/libaom/source/config/win/ia32/config/aom_config.c
View file @ 985d66b3
...@@ -9,5 +9,5 @@ ...@@ -9,5 +9,5 @@
* PATENTS file, you can obtain it at www.aomedia.org/license/patent. * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
*/ */
#include "aom/aom_codec.h" #include "aom/aom_codec.h"
static const char* const cfg = "cmake ../source/libaom -G \"Unix Makefiles\" -DCMAKE_TOOLCHAIN_FILE=\"../source/libaom/build/cmake/toolchains/x86-linux.cmake\" -DCONFIG_AV1_ENCODER=0 -DCONFIG_LOWBITDEPTH=1 -DCONFIG_MAX_DECODE_PROFILE=0 -DCONFIG_NORMAL_TILE_MODE=1 -DCONFIG_SIZE_LIMIT=1 -DDECODE_HEIGHT_LIMIT=16384 -DDECODE_WIDTH_LIMIT=16384"; static const char* const cfg = "cmake ../source/libaom -G \"Unix Makefiles\" -DCMAKE_TOOLCHAIN_FILE=\"../source/libaom/build/cmake/toolchains/x86-linux.cmake\" -DCONFIG_AV1_ENCODER=0 -DCONFIG_LOWBITDEPTH=1 -DCONFIG_MAX_DECODE_PROFILE=0 -DCONFIG_NORMAL_TILE_MODE=1 -DCONFIG_SIZE_LIMIT=1 -DDECODE_HEIGHT_LIMIT=16384 -DDECODE_WIDTH_LIMIT=16384 -DAOM_RTCD_FLAGS=--require-mmx;--require-sse;--require-sse2";
const char *aom_codec_build_config(void) {return cfg;} const char *aom_codec_build_config(void) {return cfg;}
third_party/libaom/source/config/win/ia32/config/aom_dsp_rtcd.h
View file @ 985d66b3
This source diff could not be displayed because it is too large. You can view the blob instead.
third_party/libaom/source/config/win/ia32/config/av1_rtcd.h
View file @ 985d66b3
...@@ -2071,30 +2071,22 @@ static void setup_rtcd_internal(void) { ...@@ -2071,30 +2071,22 @@ static void setup_rtcd_internal(void) {
if (flags & HAS_AVX2) if (flags & HAS_AVX2)
av1_build_compound_diffwtd_mask_highbd = av1_build_compound_diffwtd_mask_highbd =
av1_build_compound_diffwtd_mask_highbd_avx2; av1_build_compound_diffwtd_mask_highbd_avx2;
av1_convolve_2d_copy_sr = av1_convolve_2d_copy_sr_c; av1_convolve_2d_copy_sr = av1_convolve_2d_copy_sr_sse2;
if (flags & HAS_SSE2)
av1_convolve_2d_copy_sr = av1_convolve_2d_copy_sr_sse2;
if (flags & HAS_AVX2) if (flags & HAS_AVX2)
av1_convolve_2d_copy_sr = av1_convolve_2d_copy_sr_avx2; av1_convolve_2d_copy_sr = av1_convolve_2d_copy_sr_avx2;
av1_convolve_2d_scale = av1_convolve_2d_scale_c; av1_convolve_2d_scale = av1_convolve_2d_scale_c;
if (flags & HAS_SSE4_1) if (flags & HAS_SSE4_1)
av1_convolve_2d_scale = av1_convolve_2d_scale_sse4_1; av1_convolve_2d_scale = av1_convolve_2d_scale_sse4_1;
av1_convolve_2d_sr = av1_convolve_2d_sr_c; av1_convolve_2d_sr = av1_convolve_2d_sr_sse2;
if (flags & HAS_SSE2)
av1_convolve_2d_sr = av1_convolve_2d_sr_sse2;
if (flags & HAS_AVX2) if (flags & HAS_AVX2)
av1_convolve_2d_sr = av1_convolve_2d_sr_avx2; av1_convolve_2d_sr = av1_convolve_2d_sr_avx2;
av1_convolve_horiz_rs = av1_convolve_horiz_rs_c; av1_convolve_horiz_rs = av1_convolve_horiz_rs_c;
if (flags & HAS_SSE4_1) if (flags & HAS_SSE4_1)
av1_convolve_horiz_rs = av1_convolve_horiz_rs_sse4_1; av1_convolve_horiz_rs = av1_convolve_horiz_rs_sse4_1;
av1_convolve_x_sr = av1_convolve_x_sr_c; av1_convolve_x_sr = av1_convolve_x_sr_sse2;
if (flags & HAS_SSE2)
av1_convolve_x_sr = av1_convolve_x_sr_sse2;
if (flags & HAS_AVX2) if (flags & HAS_AVX2)
av1_convolve_x_sr = av1_convolve_x_sr_avx2; av1_convolve_x_sr = av1_convolve_x_sr_avx2;
av1_convolve_y_sr = av1_convolve_y_sr_c; av1_convolve_y_sr = av1_convolve_y_sr_sse2;
if (flags & HAS_SSE2)
av1_convolve_y_sr = av1_convolve_y_sr_sse2;
if (flags & HAS_AVX2) if (flags & HAS_AVX2)
av1_convolve_y_sr = av1_convolve_y_sr_avx2; av1_convolve_y_sr = av1_convolve_y_sr_avx2;
av1_filter_intra_edge = av1_filter_intra_edge_c; av1_filter_intra_edge = av1_filter_intra_edge_c;
...@@ -2106,9 +2098,7 @@ static void setup_rtcd_internal(void) { ...@@ -2106,9 +2098,7 @@ static void setup_rtcd_internal(void) {
av1_filter_intra_predictor = av1_filter_intra_predictor_c; av1_filter_intra_predictor = av1_filter_intra_predictor_c;
if (flags & HAS_SSE4_1) if (flags & HAS_SSE4_1)
av1_filter_intra_predictor = av1_filter_intra_predictor_sse4_1; av1_filter_intra_predictor = av1_filter_intra_predictor_sse4_1;
av1_highbd_convolve_2d_copy_sr = av1_highbd_convolve_2d_copy_sr_c; av1_highbd_convolve_2d_copy_sr = av1_highbd_convolve_2d_copy_sr_sse2;
if (flags & HAS_SSE2)
av1_highbd_convolve_2d_copy_sr = av1_highbd_convolve_2d_copy_sr_sse2;
if (flags & HAS_AVX2) if (flags & HAS_AVX2)
av1_highbd_convolve_2d_copy_sr = av1_highbd_convolve_2d_copy_sr_avx2; av1_highbd_convolve_2d_copy_sr = av1_highbd_convolve_2d_copy_sr_avx2;
av1_highbd_convolve_2d_scale = av1_highbd_convolve_2d_scale_c; av1_highbd_convolve_2d_scale = av1_highbd_convolve_2d_scale_c;
...@@ -2198,19 +2188,13 @@ static void setup_rtcd_internal(void) { ...@@ -2198,19 +2188,13 @@ static void setup_rtcd_internal(void) {
av1_jnt_convolve_2d = av1_jnt_convolve_2d_ssse3; av1_jnt_convolve_2d = av1_jnt_convolve_2d_ssse3;
if (flags & HAS_AVX2) if (flags & HAS_AVX2)
av1_jnt_convolve_2d = av1_jnt_convolve_2d_avx2; av1_jnt_convolve_2d = av1_jnt_convolve_2d_avx2;
av1_jnt_convolve_2d_copy = av1_jnt_convolve_2d_copy_c; av1_jnt_convolve_2d_copy = av1_jnt_convolve_2d_copy_sse2;
if (flags & HAS_SSE2)
av1_jnt_convolve_2d_copy = av1_jnt_convolve_2d_copy_sse2;
if (flags & HAS_AVX2) if (flags & HAS_AVX2)
av1_jnt_convolve_2d_copy = av1_jnt_convolve_2d_copy_avx2; av1_jnt_convolve_2d_copy = av1_jnt_convolve_2d_copy_avx2;
av1_jnt_convolve_x = av1_jnt_convolve_x_c; av1_jnt_convolve_x = av1_jnt_convolve_x_sse2;
if (flags & HAS_SSE2)
av1_jnt_convolve_x = av1_jnt_convolve_x_sse2;
if (flags & HAS_AVX2) if (flags & HAS_AVX2)
av1_jnt_convolve_x = av1_jnt_convolve_x_avx2; av1_jnt_convolve_x = av1_jnt_convolve_x_avx2;
av1_jnt_convolve_y = av1_jnt_convolve_y_c; av1_jnt_convolve_y = av1_jnt_convolve_y_sse2;
if (flags & HAS_SSE2)
av1_jnt_convolve_y = av1_jnt_convolve_y_sse2;
if (flags & HAS_AVX2) if (flags & HAS_AVX2)
av1_jnt_convolve_y = av1_jnt_convolve_y_avx2; av1_jnt_convolve_y = av1_jnt_convolve_y_avx2;
av1_selfguided_restoration = av1_selfguided_restoration_c; av1_selfguided_restoration = av1_selfguided_restoration_c;
...@@ -2227,23 +2211,17 @@ static void setup_rtcd_internal(void) { ...@@ -2227,23 +2211,17 @@ static void setup_rtcd_internal(void) {
av1_warp_affine = av1_warp_affine_c; av1_warp_affine = av1_warp_affine_c;
if (flags & HAS_SSE4_1) if (flags & HAS_SSE4_1)
av1_warp_affine = av1_warp_affine_sse4_1; av1_warp_affine = av1_warp_affine_sse4_1;
av1_wiener_convolve_add_src = av1_wiener_convolve_add_src_c; av1_wiener_convolve_add_src = av1_wiener_convolve_add_src_sse2;
if (flags & HAS_SSE2)
av1_wiener_convolve_add_src = av1_wiener_convolve_add_src_sse2;
if (flags & HAS_AVX2) if (flags & HAS_AVX2)
av1_wiener_convolve_add_src = av1_wiener_convolve_add_src_avx2; av1_wiener_convolve_add_src = av1_wiener_convolve_add_src_avx2;
cdef_filter_block = cdef_filter_block_c; cdef_filter_block = cdef_filter_block_sse2;
if (flags & HAS_SSE2)
cdef_filter_block = cdef_filter_block_sse2;
if (flags & HAS_SSSE3) if (flags & HAS_SSSE3)
cdef_filter_block = cdef_filter_block_ssse3; cdef_filter_block = cdef_filter_block_ssse3;
if (flags & HAS_SSE4_1) if (flags & HAS_SSE4_1)
cdef_filter_block = cdef_filter_block_sse4_1; cdef_filter_block = cdef_filter_block_sse4_1;
if (flags & HAS_AVX2) if (flags & HAS_AVX2)
cdef_filter_block = cdef_filter_block_avx2; cdef_filter_block = cdef_filter_block_avx2;
cdef_find_dir = cdef_find_dir_c; cdef_find_dir = cdef_find_dir_sse2;
if (flags & HAS_SSE2)
cdef_find_dir = cdef_find_dir_sse2;
if (flags & HAS_SSSE3) if (flags & HAS_SSSE3)
cdef_find_dir = cdef_find_dir_ssse3; cdef_find_dir = cdef_find_dir_ssse3;
if (flags & HAS_SSE4_1) if (flags & HAS_SSE4_1)
...@@ -2280,18 +2258,14 @@ static void setup_rtcd_internal(void) { ...@@ -2280,18 +2258,14 @@ static void setup_rtcd_internal(void) {
cfl_get_luma_subsampling_444_lbd = cfl_get_luma_subsampling_444_lbd_ssse3; cfl_get_luma_subsampling_444_lbd = cfl_get_luma_subsampling_444_lbd_ssse3;
if (flags & HAS_AVX2) if (flags & HAS_AVX2)
cfl_get_luma_subsampling_444_lbd = cfl_get_luma_subsampling_444_lbd_avx2; cfl_get_luma_subsampling_444_lbd = cfl_get_luma_subsampling_444_lbd_avx2;
copy_rect8_16bit_to_16bit = copy_rect8_16bit_to_16bit_c; copy_rect8_16bit_to_16bit = copy_rect8_16bit_to_16bit_sse2;
if (flags & HAS_SSE2)
copy_rect8_16bit_to_16bit = copy_rect8_16bit_to_16bit_sse2;
if (flags & HAS_SSSE3) if (flags & HAS_SSSE3)
copy_rect8_16bit_to_16bit = copy_rect8_16bit_to_16bit_ssse3; copy_rect8_16bit_to_16bit = copy_rect8_16bit_to_16bit_ssse3;
if (flags & HAS_SSE4_1) if (flags & HAS_SSE4_1)
copy_rect8_16bit_to_16bit = copy_rect8_16bit_to_16bit_sse4_1; copy_rect8_16bit_to_16bit = copy_rect8_16bit_to_16bit_sse4_1;
if (flags & HAS_AVX2) if (flags & HAS_AVX2)
copy_rect8_16bit_to_16bit = copy_rect8_16bit_to_16bit_avx2; copy_rect8_16bit_to_16bit = copy_rect8_16bit_to_16bit_avx2;
copy_rect8_8bit_to_16bit = copy_rect8_8bit_to_16bit_c; copy_rect8_8bit_to_16bit = copy_rect8_8bit_to_16bit_sse2;
if (flags & HAS_SSE2)
copy_rect8_8bit_to_16bit = copy_rect8_8bit_to_16bit_sse2;
if (flags & HAS_SSSE3) if (flags & HAS_SSSE3)
copy_rect8_8bit_to_16bit = copy_rect8_8bit_to_16bit_ssse3; copy_rect8_8bit_to_16bit = copy_rect8_8bit_to_16bit_ssse3;
if (flags & HAS_SSE4_1) if (flags & HAS_SSE4_1)
...@@ -2308,9 +2282,7 @@ static void setup_rtcd_internal(void) { ...@@ -2308,9 +2282,7 @@ static void setup_rtcd_internal(void) {
get_predict_lbd_fn = get_predict_lbd_fn_ssse3; get_predict_lbd_fn = get_predict_lbd_fn_ssse3;
if (flags & HAS_AVX2) if (flags & HAS_AVX2)
get_predict_lbd_fn = get_predict_lbd_fn_avx2; get_predict_lbd_fn = get_predict_lbd_fn_avx2;
get_subtract_average_fn = get_subtract_average_fn_c; get_subtract_average_fn = get_subtract_average_fn_sse2;
if (flags & HAS_SSE2)
get_subtract_average_fn = get_subtract_average_fn_sse2;
if (flags & HAS_AVX2) if (flags & HAS_AVX2)
get_subtract_average_fn = get_subtract_average_fn_avx2; get_subtract_average_fn = get_subtract_average_fn_avx2;
} }
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment