generic-library/vpx
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Restore SSSE3 subpixel filters in new convolve framework

Browse Source 
This commit adds the 8 tap SSSE3 subpixel filters back into the code
underneath the convolve API. The C code is still called for 4x4
blocks, as well as compound prediction modes. This restores the
encode performance to be within about 8% of the baseline.

Change-Id: Ife0d81477075ae33c05b53c65003951efdc8b09c
This commit is contained in:
John Koleszar
2013-02-07 17:00:37 -08:00
parent 7a07eea13f
commit 29d47ac80e
5 changed files with 723 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

129
vp9/common/x86/vp9_asm_stubs.c
View File

@@ -8,8 +8,10 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#include <assert.h>
#include "./vpx_config.h"
#include "./vp9_rtcd.h"
#include "vpx_ports/mem.h"
///////////////////////////////////////////////////////////////////////////
// the mmx function that does the bilinear filtering and var calculation //
@@ -33,3 +35,130 @@ DECLARE_ALIGNED(16, const short, vp9_bilinear_filters_mmx[16][8]) = {
{ 16, 16, 16, 16, 112, 112, 112, 112 },
{ 8, 8, 8, 8, 120, 120, 120, 120 }
};
#if HAVE_SSSE3
void vp9_filter_block1d16_v8_ssse3(const unsigned char *src_ptr,
const unsigned int src_pitch,
unsigned char *output_ptr,
unsigned int out_pitch,
unsigned int output_height,
const short *filter);
void vp9_filter_block1d16_h8_ssse3(const unsigned char *src_ptr,
const unsigned int src_pitch,
unsigned char *output_ptr,
unsigned int out_pitch,
unsigned int output_height,
const short *filter);
void vp9_filter_block1d8_v8_ssse3(const unsigned char *src_ptr,
const unsigned int src_pitch,
unsigned char *output_ptr,
unsigned int out_pitch,
unsigned int output_height,
const short *filter);
void vp9_filter_block1d8_h8_ssse3(const unsigned char *src_ptr,
const unsigned int src_pitch,
unsigned char *output_ptr,
unsigned int out_pitch,
unsigned int output_height,
const short *filter);
void vp9_convolve8_horiz_ssse3(const uint8_t *src, int src_stride,
uint8_t *dst, int 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 (x_step_q4 == 16 && filter_x[3] != 128) {
while (w >= 16) {
vp9_filter_block1d16_h8_ssse3(src, src_stride,
dst, dst_stride,
h, filter_x);
src += 16;
dst += 16;
w -= 16;
}
while (w >= 8) {
vp9_filter_block1d8_h8_ssse3(src, src_stride,
dst, dst_stride,
h, filter_x);
src += 8;
dst += 8;
w -= 8;
}
}
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, int src_stride,
uint8_t *dst, int 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_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_ssse3(src - src_stride * 3, src_stride,
dst, dst_stride,
h, filter_y);
src += 8;
dst += 8;
w -= 8;
}
}
if (w) {
vp9_convolve8_vert_c(src, src_stride, dst, dst_stride,
filter_x, x_step_q4, filter_y, y_step_q4,
w, h);
}
}
void vp9_convolve8_ssse3(const uint8_t *src, int src_stride,
uint8_t *dst, int dst_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, 16*23);
// check w/h due to fixed size fdata2 array
assert(w <= 16);
assert(h <= 16);
if (x_step_q4 == 16 && y_step_q4 == 16 &&
filter_x[3] != 128 && filter_y[3] != 128) {
if (w == 16) {
vp9_filter_block1d16_h8_ssse3(src - 3 * src_stride, src_stride,
fdata2, 16,
h + 7, filter_x);
vp9_filter_block1d16_v8_ssse3(fdata2, 16,
dst, dst_stride,
h, filter_y);
return;
}
if (w == 8) {
vp9_filter_block1d8_h8_ssse3(src - 3 * src_stride, src_stride,
fdata2, 16,
h + 7, filter_x);
vp9_filter_block1d8_v8_ssse3(fdata2, 16,
dst, dst_stride,
h, filter_y);
return;
}
}
vp9_convolve8_c(src, src_stride, dst, dst_stride,
filter_x, x_step_q4, filter_y, y_step_q4,
w, h);
}
#endif