Commit f9404f24 authored by Yunqing Wang's avatar Yunqing Wang

Revert "Revert "SSSE3 convolution optimization""

This reverts commit b6452571.

Change-Id: I60d1bf57ae8e9eb6127f42f2d5a780124ac51b45
parent 9c9fdf30
......@@ -23,20 +23,105 @@ typedef void filter8_1dfunction (
const short *filter
);
#if HAVE_SSSE3
filter8_1dfunction vp9_filter_block1d16_v8_ssse3;
filter8_1dfunction vp9_filter_block1d16_h8_ssse3;
filter8_1dfunction vp9_filter_block1d8_v8_ssse3;
filter8_1dfunction vp9_filter_block1d8_h8_ssse3;
filter8_1dfunction vp9_filter_block1d4_v8_ssse3;
filter8_1dfunction vp9_filter_block1d4_h8_ssse3;
#if (HAVE_SSSE3)
filter8_1dfunction vp9_filter_block1d16_v8_avg_ssse3;
filter8_1dfunction vp9_filter_block1d16_h8_avg_ssse3;
filter8_1dfunction vp9_filter_block1d8_v8_avg_ssse3;
filter8_1dfunction vp9_filter_block1d8_h8_avg_ssse3;
filter8_1dfunction vp9_filter_block1d4_v8_avg_ssse3;
filter8_1dfunction vp9_filter_block1d4_h8_avg_ssse3;
#if (ARCH_X86_64)
filter8_1dfunction vp9_filter_block1d16_v8_intrin_ssse3;
filter8_1dfunction vp9_filter_block1d16_h8_intrin_ssse3;
filter8_1dfunction vp9_filter_block1d8_v8_intrin_ssse3;
filter8_1dfunction vp9_filter_block1d8_h8_intrin_ssse3;
filter8_1dfunction vp9_filter_block1d4_v8_intrin_ssse3;
filter8_1dfunction vp9_filter_block1d4_h8_intrin_ssse3;
void vp9_convolve8_horiz_ssse3(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4,
int w, int h) {
/* Ensure the filter can be compressed to int16_t. */
if (x_step_q4 == 16 && filter_x[3] != 128) {
while (w >= 16) {
vp9_filter_block1d16_h8_intrin_ssse3(src, src_stride,
dst, dst_stride,
h, filter_x);
src += 16;
dst += 16;
w -= 16;
}
while (w >= 8) {
vp9_filter_block1d8_h8_intrin_ssse3(src, src_stride,
dst, dst_stride,
h, filter_x);
src += 8;
dst += 8;
w -= 8;
}
while (w >= 4) {
vp9_filter_block1d4_h8_intrin_ssse3(src, src_stride,
dst, dst_stride,
h, filter_x);
src += 4;
dst += 4;
w -= 4;
}
}
if (w) {
vp9_convolve8_horiz_c(src, src_stride, dst, dst_stride,
filter_x, x_step_q4, filter_y, y_step_q4,
w, h);
}
}
void vp9_convolve8_vert_ssse3(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
const int16_t *filter_y, int y_step_q4,
int w, int h) {
if (y_step_q4 == 16 && filter_y[3] != 128) {
while (w >= 16) {
vp9_filter_block1d16_v8_intrin_ssse3(src - src_stride * 3, src_stride,
dst, dst_stride,
h, filter_y);
src += 16;
dst += 16;
w -= 16;
}
while (w >= 8) {
vp9_filter_block1d8_v8_intrin_ssse3(src - src_stride * 3, src_stride,
dst, dst_stride,
h, filter_y);
src += 8;
dst += 8;
w -= 8;
}
while (w >= 4) {
vp9_filter_block1d4_v8_intrin_ssse3(src - src_stride * 3, src_stride,
dst, dst_stride,
h, filter_y);
src += 4;
dst += 4;
w -= 4;
}
}
if (w) {
vp9_convolve8_vert_c(src, src_stride, dst, dst_stride,
filter_x, x_step_q4, filter_y, y_step_q4,
w, h);
}
}
#else
filter8_1dfunction vp9_filter_block1d16_v8_ssse3;
filter8_1dfunction vp9_filter_block1d16_h8_ssse3;
filter8_1dfunction vp9_filter_block1d8_v8_ssse3;
filter8_1dfunction vp9_filter_block1d8_h8_ssse3;
filter8_1dfunction vp9_filter_block1d4_v8_ssse3;
filter8_1dfunction vp9_filter_block1d4_h8_ssse3;
void vp9_convolve8_horiz_ssse3(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
const int16_t *filter_x, int x_step_q4,
......@@ -113,6 +198,7 @@ void vp9_convolve8_vert_ssse3(const uint8_t *src, ptrdiff_t src_stride,
w, h);
}
}
#endif
void vp9_convolve8_avg_horiz_ssse3(const uint8_t *src, ptrdiff_t src_stride,
uint8_t *dst, ptrdiff_t dst_stride,
......
This diff is collapsed.
......@@ -74,6 +74,9 @@ VP9_COMMON_SRCS-$(CONFIG_VP9_POSTPROC) += common/vp9_postproc.c
VP9_COMMON_SRCS-$(HAVE_MMX) += common/x86/vp9_loopfilter_mmx.asm
VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_subpixel_8t_sse2.asm
VP9_COMMON_SRCS-$(HAVE_SSSE3) += common/x86/vp9_subpixel_8t_ssse3.asm
ifeq ($(ARCH_X86_64),yes)
VP9_COMMON_SRCS-$(HAVE_SSSE3) += common/x86/vp9_subpixel_8t_intrin_ssse3.c
endif
ifeq ($(CONFIG_VP9_POSTPROC),yes)
VP9_COMMON_SRCS-$(HAVE_MMX) += common/x86/vp9_postproc_mmx.asm
VP9_COMMON_SRCS-$(HAVE_SSE2) += common/x86/vp9_postproc_sse2.asm
......
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