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

Some cosmetic improvements since HBD variance 4x4 optimization

Browse Source 
Change-Id: I414c1fabd2e3a9b1d9daa8a90f85a0bace8bd3cd
This commit is contained in:
Yi Luo
2016-04-07 15:06:43 -07:00
parent 454989ff32
commit e5f4e8eab9
4 changed files with 95 additions and 77 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
vp10/encoder/mcomp.c
View File

@@ -367,8 +367,8 @@ static unsigned int setup_center_error(const MACROBLOCKD *xd,
if (second_pred != NULL) { if (second_pred != NULL) {
if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) { if (xd->cur_buf->flags & YV12_FLAG_HIGHBITDEPTH) {
DECLARE_ALIGNED(16, uint16_t, comp_pred16[MAX_SB_SQUARE]); DECLARE_ALIGNED(16, uint16_t, comp_pred16[MAX_SB_SQUARE]);
vpx_highbd_comp_avg_pred_c(comp_pred16, second_pred, w, h, y + offset, vpx_highbd_comp_avg_pred(comp_pred16, second_pred, w, h, y + offset,
y_stride); y_stride);
besterr = vfp->vf(CONVERT_TO_BYTEPTR(comp_pred16), w, src, src_stride, besterr = vfp->vf(CONVERT_TO_BYTEPTR(comp_pred16), w, src, src_stride,
sse1); sse1);
} else { } else {

92
vpx_dsp/variance.c
View File

@@ -433,7 +433,7 @@ uint32_t vpx_highbd_12_mse##W##x##H##_c(const uint8_t *src, \
return *sse; \ return *sse; \
} }
void highbd_var_filter_block2d_bil_first_pass( void vpx_highbd_var_filter_block2d_bil_first_pass(
const uint8_t *src_ptr8, const uint8_t *src_ptr8,
uint16_t *output_ptr, uint16_t *output_ptr,
unsigned int src_pixels_per_line, unsigned int src_pixels_per_line,
@@ -459,7 +459,7 @@ void highbd_var_filter_block2d_bil_first_pass(
} }
} }
void highbd_var_filter_block2d_bil_second_pass( void vpx_highbd_var_filter_block2d_bil_second_pass(
const uint16_t *src_ptr, const uint16_t *src_ptr,
uint16_t *output_ptr, uint16_t *output_ptr,
unsigned int src_pixels_per_line, unsigned int src_pixels_per_line,
@@ -492,13 +492,14 @@ uint32_t vpx_highbd_8_sub_pixel_variance##W##x##H##_c( \
uint16_t fdata3[(H + 1) * W]; \ uint16_t fdata3[(H + 1) * W]; \
uint16_t temp2[H * W]; \ uint16_t temp2[H * W]; \
\ \
highbd_var_filter_block2d_bil_first_pass(src, fdata3, src_stride, 1, H + 1, \ vpx_highbd_var_filter_block2d_bil_first_pass( \
W, bilinear_filters_2t[xoffset]); \ src, fdata3, src_stride, 1, H + 1, \
highbd_var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \ W, bilinear_filters_2t[xoffset]); \
bilinear_filters_2t[yoffset]); \ vpx_highbd_var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \
bilinear_filters_2t[yoffset]); \
\ \
return vpx_highbd_8_variance##W##x##H##_c(CONVERT_TO_BYTEPTR(temp2), W, dst, \ return vpx_highbd_8_variance##W##x##H##_c(CONVERT_TO_BYTEPTR(temp2), W, dst, \
dst_stride, sse); \ dst_stride, sse); \
} \ } \
\ \
uint32_t vpx_highbd_10_sub_pixel_variance##W##x##H##_c( \ uint32_t vpx_highbd_10_sub_pixel_variance##W##x##H##_c( \
@@ -509,10 +510,11 @@ uint32_t vpx_highbd_10_sub_pixel_variance##W##x##H##_c( \
uint16_t fdata3[(H + 1) * W]; \ uint16_t fdata3[(H + 1) * W]; \
uint16_t temp2[H * W]; \ uint16_t temp2[H * W]; \
\ \
highbd_var_filter_block2d_bil_first_pass(src, fdata3, src_stride, 1, H + 1, \ vpx_highbd_var_filter_block2d_bil_first_pass( \
W, bilinear_filters_2t[xoffset]); \ src, fdata3, src_stride, 1, H + 1, \
highbd_var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \ W, bilinear_filters_2t[xoffset]); \
bilinear_filters_2t[yoffset]); \ vpx_highbd_var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \
bilinear_filters_2t[yoffset]); \
\ \
return vpx_highbd_10_variance##W##x##H##_c(CONVERT_TO_BYTEPTR(temp2), \ return vpx_highbd_10_variance##W##x##H##_c(CONVERT_TO_BYTEPTR(temp2), \
W, dst, dst_stride, sse); \ W, dst, dst_stride, sse); \
@@ -526,10 +528,11 @@ uint32_t vpx_highbd_12_sub_pixel_variance##W##x##H##_c( \
uint16_t fdata3[(H + 1) * W]; \ uint16_t fdata3[(H + 1) * W]; \
uint16_t temp2[H * W]; \ uint16_t temp2[H * W]; \
\ \
highbd_var_filter_block2d_bil_first_pass(src, fdata3, src_stride, 1, H + 1, \ vpx_highbd_var_filter_block2d_bil_first_pass( \
W, bilinear_filters_2t[xoffset]); \ src, fdata3, src_stride, 1, H + 1, \
highbd_var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \ W, bilinear_filters_2t[xoffset]); \
bilinear_filters_2t[yoffset]); \ vpx_highbd_var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \
bilinear_filters_2t[yoffset]); \
\ \
return vpx_highbd_12_variance##W##x##H##_c(CONVERT_TO_BYTEPTR(temp2), \ return vpx_highbd_12_variance##W##x##H##_c(CONVERT_TO_BYTEPTR(temp2), \
W, dst, dst_stride, sse); \ W, dst, dst_stride, sse); \
@@ -546,16 +549,17 @@ uint32_t vpx_highbd_8_sub_pixel_avg_variance##W##x##H##_c( \
uint16_t temp2[H * W]; \ uint16_t temp2[H * W]; \
DECLARE_ALIGNED(16, uint16_t, temp3[H * W]); \ DECLARE_ALIGNED(16, uint16_t, temp3[H * W]); \
\ \
highbd_var_filter_block2d_bil_first_pass(src, fdata3, src_stride, 1, H + 1, \ vpx_highbd_var_filter_block2d_bil_first_pass( \
W, bilinear_filters_2t[xoffset]); \ src, fdata3, src_stride, 1, H + 1, \
highbd_var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \ W, bilinear_filters_2t[xoffset]); \
bilinear_filters_2t[yoffset]); \ vpx_highbd_var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \
bilinear_filters_2t[yoffset]); \
\ \
vpx_highbd_comp_avg_pred_c(temp3, second_pred, W, H, \ vpx_highbd_comp_avg_pred_c(temp3, second_pred, W, H, \
CONVERT_TO_BYTEPTR(temp2), W); \ CONVERT_TO_BYTEPTR(temp2), W); \
\ \
return vpx_highbd_8_variance##W##x##H##_c(CONVERT_TO_BYTEPTR(temp3), W, dst, \ return vpx_highbd_8_variance##W##x##H##_c(CONVERT_TO_BYTEPTR(temp3), W, dst, \
dst_stride, sse); \ dst_stride, sse); \
} \ } \
\ \
uint32_t vpx_highbd_10_sub_pixel_avg_variance##W##x##H##_c( \ uint32_t vpx_highbd_10_sub_pixel_avg_variance##W##x##H##_c( \
@@ -568,10 +572,11 @@ uint32_t vpx_highbd_10_sub_pixel_avg_variance##W##x##H##_c( \
uint16_t temp2[H * W]; \ uint16_t temp2[H * W]; \
DECLARE_ALIGNED(16, uint16_t, temp3[H * W]); \ DECLARE_ALIGNED(16, uint16_t, temp3[H * W]); \
\ \
highbd_var_filter_block2d_bil_first_pass(src, fdata3, src_stride, 1, H + 1, \ vpx_highbd_var_filter_block2d_bil_first_pass( \
W, bilinear_filters_2t[xoffset]); \ src, fdata3, src_stride, 1, H + 1, \
highbd_var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \ W, bilinear_filters_2t[xoffset]); \
bilinear_filters_2t[yoffset]); \ vpx_highbd_var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \
bilinear_filters_2t[yoffset]); \
\ \
vpx_highbd_comp_avg_pred_c(temp3, second_pred, W, H, \ vpx_highbd_comp_avg_pred_c(temp3, second_pred, W, H, \
CONVERT_TO_BYTEPTR(temp2), W); \ CONVERT_TO_BYTEPTR(temp2), W); \
@@ -590,10 +595,11 @@ uint32_t vpx_highbd_12_sub_pixel_avg_variance##W##x##H##_c( \
uint16_t temp2[H * W]; \ uint16_t temp2[H * W]; \
DECLARE_ALIGNED(16, uint16_t, temp3[H * W]); \ DECLARE_ALIGNED(16, uint16_t, temp3[H * W]); \
\ \
highbd_var_filter_block2d_bil_first_pass(src, fdata3, src_stride, 1, H + 1, \ vpx_highbd_var_filter_block2d_bil_first_pass( \
W, bilinear_filters_2t[xoffset]); \ src, fdata3, src_stride, 1, H + 1, \
highbd_var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \ W, bilinear_filters_2t[xoffset]); \
bilinear_filters_2t[yoffset]); \ vpx_highbd_var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \
bilinear_filters_2t[yoffset]); \
\ \
vpx_highbd_comp_avg_pred_c(temp3, second_pred, W, H, \ vpx_highbd_comp_avg_pred_c(temp3, second_pred, W, H, \
CONVERT_TO_BYTEPTR(temp2), W); \ CONVERT_TO_BYTEPTR(temp2), W); \
@@ -914,11 +920,11 @@ unsigned int vpx_highbd_masked_sub_pixel_variance##W##x##H##_c( \
uint16_t fdata3[(H + 1) * W]; \ uint16_t fdata3[(H + 1) * W]; \
uint16_t temp2[H * W]; \ uint16_t temp2[H * W]; \
\ \
highbd_var_filter_block2d_bil_first_pass(src, fdata3, src_stride, 1, \ vpx_highbd_var_filter_block2d_bil_first_pass(src, fdata3, src_stride, 1, \
H + 1, W, \ H + 1, W, \
bilinear_filters_2t[xoffset]); \ bilinear_filters_2t[xoffset]); \
highbd_var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \ vpx_highbd_var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \
bilinear_filters_2t[yoffset]); \ bilinear_filters_2t[yoffset]); \
\ \
return vpx_highbd_masked_variance##W##x##H##_c(CONVERT_TO_BYTEPTR(temp2), \ return vpx_highbd_masked_variance##W##x##H##_c(CONVERT_TO_BYTEPTR(temp2), \
W, dst, dst_stride, \ W, dst, dst_stride, \
@@ -934,11 +940,11 @@ unsigned int vpx_highbd_10_masked_sub_pixel_variance##W##x##H##_c( \
uint16_t fdata3[(H + 1) * W]; \ uint16_t fdata3[(H + 1) * W]; \
uint16_t temp2[H * W]; \ uint16_t temp2[H * W]; \
\ \
highbd_var_filter_block2d_bil_first_pass(src, fdata3, src_stride, 1, \ vpx_highbd_var_filter_block2d_bil_first_pass(src, fdata3, src_stride, 1, \
H + 1, W, \ H + 1, W, \
bilinear_filters_2t[xoffset]); \ bilinear_filters_2t[xoffset]); \
highbd_var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \ vpx_highbd_var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \
bilinear_filters_2t[yoffset]); \ bilinear_filters_2t[yoffset]); \
\ \
return vpx_highbd_10_masked_variance##W##x##H##_c(CONVERT_TO_BYTEPTR(temp2), \ return vpx_highbd_10_masked_variance##W##x##H##_c(CONVERT_TO_BYTEPTR(temp2), \
W, dst, dst_stride, \ W, dst, dst_stride, \
@@ -954,11 +960,11 @@ unsigned int vpx_highbd_12_masked_sub_pixel_variance##W##x##H##_c( \
uint16_t fdata3[(H + 1) * W]; \ uint16_t fdata3[(H + 1) * W]; \
uint16_t temp2[H * W]; \ uint16_t temp2[H * W]; \
\ \
highbd_var_filter_block2d_bil_first_pass(src, fdata3, src_stride, 1, \ vpx_highbd_var_filter_block2d_bil_first_pass(src, fdata3, src_stride, 1, \
H + 1, W, \ H + 1, W, \
bilinear_filters_2t[xoffset]); \ bilinear_filters_2t[xoffset]); \
highbd_var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \ vpx_highbd_var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \
bilinear_filters_2t[yoffset]); \ bilinear_filters_2t[yoffset]); \
\ \
return vpx_highbd_12_masked_variance##W##x##H##_c(CONVERT_TO_BYTEPTR(temp2), \ return vpx_highbd_12_masked_variance##W##x##H##_c(CONVERT_TO_BYTEPTR(temp2), \
W, dst, dst_stride, \ W, dst, dst_stride, \

4
vpx_dsp/variance.h
View File

@@ -130,7 +130,7 @@ typedef struct vp10_variance_vtable {
} vp10_variance_fn_ptr_t; } vp10_variance_fn_ptr_t;
#endif // CONFIG_VP10 #endif // CONFIG_VP10
void highbd_var_filter_block2d_bil_first_pass( void vpx_highbd_var_filter_block2d_bil_first_pass(
const uint8_t *src_ptr8, const uint8_t *src_ptr8,
uint16_t *output_ptr, uint16_t *output_ptr,
unsigned int src_pixels_per_line, unsigned int src_pixels_per_line,
@@ -139,7 +139,7 @@ void highbd_var_filter_block2d_bil_first_pass(
unsigned int output_width, unsigned int output_width,
const uint8_t *filter); const uint8_t *filter);
void highbd_var_filter_block2d_bil_second_pass( void vpx_highbd_var_filter_block2d_bil_second_pass(
const uint16_t *src_ptr, const uint16_t *src_ptr,
uint16_t *output_ptr, uint16_t *output_ptr,
unsigned int src_pixels_per_line, unsigned int src_pixels_per_line,

72
vpx_dsp/x86/highbd_variance_sse4.c
View File

@@ -119,10 +119,12 @@ uint32_t vpx_highbd_8_sub_pixel_variance4x4_sse4_1(
uint16_t fdata3[(4 + 1) * 4]; uint16_t fdata3[(4 + 1) * 4];
uint16_t temp2[4 * 4]; uint16_t temp2[4 * 4];
highbd_var_filter_block2d_bil_first_pass(src, fdata3, src_stride, 1, 4 + 1, vpx_highbd_var_filter_block2d_bil_first_pass(
4, bilinear_filters_2t[xoffset]); src, fdata3, src_stride, 1, 4 + 1,
highbd_var_filter_block2d_bil_second_pass(fdata3, temp2, 4, 4, 4, 4, 4, bilinear_filters_2t[xoffset]);
bilinear_filters_2t[yoffset]); vpx_highbd_var_filter_block2d_bil_second_pass(
fdata3, temp2, 4, 4, 4, 4,
bilinear_filters_2t[yoffset]);
return vpx_highbd_8_variance4x4(CONVERT_TO_BYTEPTR(temp2), return vpx_highbd_8_variance4x4(CONVERT_TO_BYTEPTR(temp2),
4, dst, dst_stride, sse); 4, dst, dst_stride, sse);
@@ -137,10 +139,12 @@ uint32_t vpx_highbd_10_sub_pixel_variance4x4_sse4_1(
uint16_t fdata3[(4 + 1) * 4]; uint16_t fdata3[(4 + 1) * 4];
uint16_t temp2[4 * 4]; uint16_t temp2[4 * 4];
highbd_var_filter_block2d_bil_first_pass(src, fdata3, src_stride, 1, 4 + 1, vpx_highbd_var_filter_block2d_bil_first_pass(
4, bilinear_filters_2t[xoffset]); src, fdata3, src_stride, 1, 4 + 1,
highbd_var_filter_block2d_bil_second_pass(fdata3, temp2, 4, 4, 4, 4, 4, bilinear_filters_2t[xoffset]);
bilinear_filters_2t[yoffset]); vpx_highbd_var_filter_block2d_bil_second_pass(
fdata3, temp2, 4, 4, 4, 4,
bilinear_filters_2t[yoffset]);
return vpx_highbd_10_variance4x4(CONVERT_TO_BYTEPTR(temp2), return vpx_highbd_10_variance4x4(CONVERT_TO_BYTEPTR(temp2),
4, dst, dst_stride, sse); 4, dst, dst_stride, sse);
@@ -155,10 +159,12 @@ uint32_t vpx_highbd_12_sub_pixel_variance4x4_sse4_1(
uint16_t fdata3[(4 + 1) * 4]; uint16_t fdata3[(4 + 1) * 4];
uint16_t temp2[4 * 4]; uint16_t temp2[4 * 4];
highbd_var_filter_block2d_bil_first_pass(src, fdata3, src_stride, 1, 4 + 1, vpx_highbd_var_filter_block2d_bil_first_pass(
4, bilinear_filters_2t[xoffset]); src, fdata3, src_stride, 1, 4 + 1,
highbd_var_filter_block2d_bil_second_pass(fdata3, temp2, 4, 4, 4, 4, 4, bilinear_filters_2t[xoffset]);
bilinear_filters_2t[yoffset]); vpx_highbd_var_filter_block2d_bil_second_pass(
fdata3, temp2, 4, 4, 4, 4,
bilinear_filters_2t[yoffset]);
return vpx_highbd_12_variance4x4(CONVERT_TO_BYTEPTR(temp2), return vpx_highbd_12_variance4x4(CONVERT_TO_BYTEPTR(temp2),
4, dst, dst_stride, sse); 4, dst, dst_stride, sse);
@@ -177,13 +183,15 @@ uint32_t vpx_highbd_8_sub_pixel_avg_variance4x4_sse4_1(
uint16_t temp2[4 * 4]; uint16_t temp2[4 * 4];
DECLARE_ALIGNED(16, uint16_t, temp3[4 * 4]); DECLARE_ALIGNED(16, uint16_t, temp3[4 * 4]);
highbd_var_filter_block2d_bil_first_pass(src, fdata3, src_stride, 1, 4 + 1, vpx_highbd_var_filter_block2d_bil_first_pass(
4, bilinear_filters_2t[xoffset]); src, fdata3, src_stride, 1, 4 + 1,
highbd_var_filter_block2d_bil_second_pass(fdata3, temp2, 4, 4, 4, 4, 4, bilinear_filters_2t[xoffset]);
bilinear_filters_2t[yoffset]); vpx_highbd_var_filter_block2d_bil_second_pass(
fdata3, temp2, 4, 4, 4, 4,
bilinear_filters_2t[yoffset]);
vpx_highbd_comp_avg_pred_c(temp3, second_pred, 4, 4, vpx_highbd_comp_avg_pred(temp3, second_pred, 4, 4,
CONVERT_TO_BYTEPTR(temp2), 4); CONVERT_TO_BYTEPTR(temp2), 4);
return vpx_highbd_8_variance4x4(CONVERT_TO_BYTEPTR(temp3), return vpx_highbd_8_variance4x4(CONVERT_TO_BYTEPTR(temp3),
4, dst, dst_stride, sse); 4, dst, dst_stride, sse);
@@ -200,13 +208,15 @@ uint32_t vpx_highbd_10_sub_pixel_avg_variance4x4_sse4_1(
uint16_t temp2[4 * 4]; uint16_t temp2[4 * 4];
DECLARE_ALIGNED(16, uint16_t, temp3[4 * 4]); DECLARE_ALIGNED(16, uint16_t, temp3[4 * 4]);
highbd_var_filter_block2d_bil_first_pass(src, fdata3, src_stride, 1, 4 + 1, vpx_highbd_var_filter_block2d_bil_first_pass(
4, bilinear_filters_2t[xoffset]); src, fdata3, src_stride, 1, 4 + 1,
highbd_var_filter_block2d_bil_second_pass(fdata3, temp2, 4, 4, 4, 4, 4, bilinear_filters_2t[xoffset]);
bilinear_filters_2t[yoffset]); vpx_highbd_var_filter_block2d_bil_second_pass(
fdata3, temp2, 4, 4, 4, 4,
bilinear_filters_2t[yoffset]);
vpx_highbd_comp_avg_pred_c(temp3, second_pred, 4, 4, vpx_highbd_comp_avg_pred(temp3, second_pred, 4, 4,
CONVERT_TO_BYTEPTR(temp2), 4); CONVERT_TO_BYTEPTR(temp2), 4);
return vpx_highbd_10_variance4x4(CONVERT_TO_BYTEPTR(temp3), return vpx_highbd_10_variance4x4(CONVERT_TO_BYTEPTR(temp3),
4, dst, dst_stride, sse); 4, dst, dst_stride, sse);
@@ -223,13 +233,15 @@ uint32_t vpx_highbd_12_sub_pixel_avg_variance4x4_sse4_1(
uint16_t temp2[4 * 4]; uint16_t temp2[4 * 4];
DECLARE_ALIGNED(16, uint16_t, temp3[4 * 4]); DECLARE_ALIGNED(16, uint16_t, temp3[4 * 4]);
highbd_var_filter_block2d_bil_first_pass(src, fdata3, src_stride, 1, 4 + 1, vpx_highbd_var_filter_block2d_bil_first_pass(
4, bilinear_filters_2t[xoffset]); src, fdata3, src_stride, 1, 4 + 1,
highbd_var_filter_block2d_bil_second_pass(fdata3, temp2, 4, 4, 4, 4, 4, bilinear_filters_2t[xoffset]);
bilinear_filters_2t[yoffset]); vpx_highbd_var_filter_block2d_bil_second_pass(
fdata3, temp2, 4, 4, 4, 4,
bilinear_filters_2t[yoffset]);
vpx_highbd_comp_avg_pred_c(temp3, second_pred, 4, 4, vpx_highbd_comp_avg_pred(temp3, second_pred, 4, 4,
CONVERT_TO_BYTEPTR(temp2), 4); CONVERT_TO_BYTEPTR(temp2), 4);
return vpx_highbd_12_variance4x4(CONVERT_TO_BYTEPTR(temp3), return vpx_highbd_12_variance4x4(CONVERT_TO_BYTEPTR(temp3),
4, dst, dst_stride, sse); 4, dst, dst_stride, sse);