Commit 2488cb34 authored by Yunqing Wang's avatar Yunqing Wang

Optimize bilinear sub-pixel filters in sse2

Using bilinear filters could speed up the codec in real-time mode.
This patch added sse2 optimizations of bilinear filters that
operate on different-sized blocks.

Tests showed that the real-time encoder was speeded up by 3%.

Change-Id: If99a7ee4385fcc225c3ee7445d962d5752e57c3f
parent 37e5325b
......@@ -23,36 +23,63 @@ typedef void filter8_1dfunction (
const short *filter
);
#define FUN_CONV_1D(name, step_q4, filter, dir, src_start, avg, opt) \
void vp9_convolve8_##name##_##opt(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) { \
#define FUN_CONV_1D(name, step_q4, filter, dir, src_start, avg, opt1, opt2) \
void vp9_convolve8_##name##_##opt1(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 (step_q4 == 16 && filter[3] != 128) { \
while (w >= 16) { \
vp9_filter_block1d16_##dir##8_##avg##opt(src_start, src_stride, \
dst, dst_stride, \
h, filter); \
src += 16; \
dst += 16; \
w -= 16; \
} \
while (w >= 8) { \
vp9_filter_block1d8_##dir##8_##avg##opt(src_start, src_stride, \
dst, dst_stride, \
h, filter); \
src += 8; \
dst += 8; \
w -= 8; \
} \
while (w >= 4) { \
vp9_filter_block1d4_##dir##8_##avg##opt(src_start, src_stride, \
dst, dst_stride, \
h, filter); \
src += 4; \
dst += 4; \
w -= 4; \
if (filter[0] || filter[1] || filter[2]) { \
while (w >= 16) { \
vp9_filter_block1d16_##dir##8_##avg##opt1(src_start, src_stride, \
dst, dst_stride, \
h, filter); \
src += 16; \
dst += 16; \
w -= 16; \
} \
while (w >= 8) { \
vp9_filter_block1d8_##dir##8_##avg##opt1(src_start, src_stride, \
dst, dst_stride, \
h, filter); \
src += 8; \
dst += 8; \
w -= 8; \
} \
while (w >= 4) { \
vp9_filter_block1d4_##dir##8_##avg##opt1(src_start, src_stride, \
dst, dst_stride, \
h, filter); \
src += 4; \
dst += 4; \
w -= 4; \
} \
} else { \
while (w >= 16) { \
vp9_filter_block1d16_##dir##2_##avg##opt2(src, src_stride, \
dst, dst_stride, \
h, filter); \
src += 16; \
dst += 16; \
w -= 16; \
} \
while (w >= 8) { \
vp9_filter_block1d8_##dir##2_##avg##opt2(src, src_stride, \
dst, dst_stride, \
h, filter); \
src += 8; \
dst += 8; \
w -= 8; \
} \
while (w >= 4) { \
vp9_filter_block1d4_##dir##2_##avg##opt2(src, src_stride, \
dst, dst_stride, \
h, filter); \
src += 4; \
dst += 4; \
w -= 4; \
} \
} \
} \
if (w) { \
......@@ -68,17 +95,27 @@ void vp9_convolve8_##avg##opt(const uint8_t *src, ptrdiff_t src_stride, \
const int16_t *filter_x, int x_step_q4, \
const int16_t *filter_y, int y_step_q4, \
int w, int h) { \
DECLARE_ALIGNED_ARRAY(16, unsigned char, fdata2, 64 * 71); \
\
assert(w <= 64); \
assert(h <= 64); \
if (x_step_q4 == 16 && y_step_q4 == 16) { \
vp9_convolve8_horiz_##opt(src - 3 * src_stride, src_stride, fdata2, 64, \
filter_x, x_step_q4, filter_y, y_step_q4, \
w, h + 7); \
vp9_convolve8_##avg##vert_##opt(fdata2 + 3 * 64, 64, dst, dst_stride, \
filter_x, x_step_q4, filter_y, y_step_q4, \
w, h); \
if (filter_x[0] || filter_x[1] || filter_x[2] || filter_x[3] == 128 || \
filter_y[0] || filter_y[1] || filter_y[2] || filter_y[3] == 128) { \
DECLARE_ALIGNED_ARRAY(16, unsigned char, fdata2, 64 * 71); \
vp9_convolve8_horiz_##opt(src - 3 * src_stride, src_stride, fdata2, 64, \
filter_x, x_step_q4, filter_y, y_step_q4, \
w, h + 7); \
vp9_convolve8_##avg##vert_##opt(fdata2 + 3 * 64, 64, dst, dst_stride, \
filter_x, x_step_q4, filter_y, \
y_step_q4, w, h); \
} else { \
DECLARE_ALIGNED_ARRAY(16, unsigned char, fdata2, 64 * 65); \
vp9_convolve8_horiz_##opt(src, src_stride, fdata2, 64, \
filter_x, x_step_q4, filter_y, y_step_q4, \
w, h + 1); \
vp9_convolve8_##avg##vert_##opt(fdata2, 64, dst, dst_stride, \
filter_x, x_step_q4, filter_y, \
y_step_q4, w, h); \
} \
} else { \
vp9_convolve8_##avg##c(src, src_stride, dst, dst_stride, \
filter_x, x_step_q4, filter_y, y_step_q4, w, h); \
......@@ -99,6 +136,19 @@ filter8_1dfunction vp9_filter_block1d8_h8_avg_ssse3;
filter8_1dfunction vp9_filter_block1d4_v8_avg_ssse3;
filter8_1dfunction vp9_filter_block1d4_h8_avg_ssse3;
filter8_1dfunction vp9_filter_block1d16_v2_sse2;
filter8_1dfunction vp9_filter_block1d16_h2_sse2;
filter8_1dfunction vp9_filter_block1d8_v2_sse2;
filter8_1dfunction vp9_filter_block1d8_h2_sse2;
filter8_1dfunction vp9_filter_block1d4_v2_sse2;
filter8_1dfunction vp9_filter_block1d4_h2_sse2;
filter8_1dfunction vp9_filter_block1d16_v2_avg_sse2;
filter8_1dfunction vp9_filter_block1d16_h2_avg_sse2;
filter8_1dfunction vp9_filter_block1d8_v2_avg_sse2;
filter8_1dfunction vp9_filter_block1d8_h2_avg_sse2;
filter8_1dfunction vp9_filter_block1d4_v2_avg_sse2;
filter8_1dfunction vp9_filter_block1d4_h2_avg_sse2;
// 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,
......@@ -119,11 +169,11 @@ filter8_1dfunction vp9_filter_block1d4_h8_avg_ssse3;
// const int16_t *filter_x, int x_step_q4,
// const int16_t *filter_y, int y_step_q4,
// int w, int h);
FUN_CONV_1D(horiz, x_step_q4, filter_x, h, src, , ssse3);
FUN_CONV_1D(vert, y_step_q4, filter_y, v, src - src_stride * 3, , ssse3);
FUN_CONV_1D(avg_horiz, x_step_q4, filter_x, h, src, avg_, ssse3);
FUN_CONV_1D(horiz, x_step_q4, filter_x, h, src, , ssse3, sse2);
FUN_CONV_1D(vert, y_step_q4, filter_y, v, src - src_stride * 3, , ssse3, sse2);
FUN_CONV_1D(avg_horiz, x_step_q4, filter_x, h, src, avg_, ssse3, sse2);
FUN_CONV_1D(avg_vert, y_step_q4, filter_y, v, src - src_stride * 3, avg_,
ssse3);
ssse3, sse2);
// void vp9_convolve8_ssse3(const uint8_t *src, ptrdiff_t src_stride,
// uint8_t *dst, ptrdiff_t dst_stride,
......@@ -153,6 +203,19 @@ filter8_1dfunction vp9_filter_block1d8_h8_avg_sse2;
filter8_1dfunction vp9_filter_block1d4_v8_avg_sse2;
filter8_1dfunction vp9_filter_block1d4_h8_avg_sse2;
filter8_1dfunction vp9_filter_block1d16_v2_sse2;
filter8_1dfunction vp9_filter_block1d16_h2_sse2;
filter8_1dfunction vp9_filter_block1d8_v2_sse2;
filter8_1dfunction vp9_filter_block1d8_h2_sse2;
filter8_1dfunction vp9_filter_block1d4_v2_sse2;
filter8_1dfunction vp9_filter_block1d4_h2_sse2;
filter8_1dfunction vp9_filter_block1d16_v2_avg_sse2;
filter8_1dfunction vp9_filter_block1d16_h2_avg_sse2;
filter8_1dfunction vp9_filter_block1d8_v2_avg_sse2;
filter8_1dfunction vp9_filter_block1d8_h2_avg_sse2;
filter8_1dfunction vp9_filter_block1d4_v2_avg_sse2;
filter8_1dfunction vp9_filter_block1d4_h2_avg_sse2;
// void vp9_convolve8_horiz_sse2(const uint8_t *src, ptrdiff_t src_stride,
// uint8_t *dst, ptrdiff_t dst_stride,
// const int16_t *filter_x, int x_step_q4,
......@@ -173,10 +236,11 @@ filter8_1dfunction vp9_filter_block1d4_h8_avg_sse2;
// const int16_t *filter_x, int x_step_q4,
// const int16_t *filter_y, int y_step_q4,
// int w, int h);
FUN_CONV_1D(horiz, x_step_q4, filter_x, h, src, , sse2);
FUN_CONV_1D(vert, y_step_q4, filter_y, v, src - src_stride * 3, , sse2);
FUN_CONV_1D(avg_horiz, x_step_q4, filter_x, h, src, avg_, sse2);
FUN_CONV_1D(avg_vert, y_step_q4, filter_y, v, src - src_stride * 3, avg_, sse2);
FUN_CONV_1D(horiz, x_step_q4, filter_x, h, src, , sse2, sse2);
FUN_CONV_1D(vert, y_step_q4, filter_y, v, src - src_stride * 3, , sse2, sse2);
FUN_CONV_1D(avg_horiz, x_step_q4, filter_x, h, src, avg_, sse2, sse2);
FUN_CONV_1D(avg_vert, y_step_q4, filter_y, v, src - src_stride * 3, avg_, sse2,
sse2);
// void vp9_convolve8_sse2(const uint8_t *src, ptrdiff_t src_stride,
// uint8_t *dst, ptrdiff_t dst_stride,
......
This diff is collapsed.
......@@ -74,6 +74,7 @@ VP9_COMMON_SRCS-$(CONFIG_VP9_POSTPROC) += common/vp9_postproc.h
VP9_COMMON_SRCS-$(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-$(HAVE_SSE2) += common/x86/vp9_subpixel_bilinear_sse2.asm
VP9_COMMON_SRCS-$(HAVE_SSSE3) += common/x86/vp9_subpixel_8t_ssse3.asm
ifeq ($(CONFIG_VP9_POSTPROC),yes)
VP9_COMMON_SRCS-$(HAVE_MMX) += common/x86/vp9_postproc_mmx.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