Commit 1e4136d3 authored by Deb Mukherjee's avatar Deb Mukherjee
Browse files

Adds high bit depth sad and variance functions

Moves high bit depth sad/var functions from highbitdepth
branch to master.

Change-Id: If03845d8ef9c9c494e13350e7a587c289306b94d
parent ac2f2e78
This diff is collapsed.
......@@ -14,6 +14,9 @@
#include "./vpx_config.h"
#include "vpx/vpx_integer.h"
#if CONFIG_VP9_HIGHBITDEPTH
#include "vp9/common/vp9_common.h"
#endif
#include "vp9/encoder/vp9_variance.h"
static INLINE unsigned int sad(const uint8_t *a, int a_stride,
......@@ -131,3 +134,138 @@ sadMxN(4, 4)
sadMxNxK(4, 4, 3)
sadMxNxK(4, 4, 8)
sadMxNx4D(4, 4)
#if CONFIG_VP9_HIGHBITDEPTH
static INLINE unsigned int high_sad(const uint8_t *a8, int a_stride,
const uint8_t *b8, int b_stride,
int width, int height) {
int y, x;
unsigned int sad = 0;
const uint16_t *a = CONVERT_TO_SHORTPTR(a8);
const uint16_t *b = CONVERT_TO_SHORTPTR(b8);
for (y = 0; y < height; y++) {
for (x = 0; x < width; x++)
sad += abs(a[x] - b[x]);
a += a_stride;
b += b_stride;
}
return sad;
}
static INLINE unsigned int high_sadb(const uint8_t *a8, int a_stride,
const uint16_t *b, int b_stride,
int width, int height) {
int y, x;
unsigned int sad = 0;
const uint16_t *a = CONVERT_TO_SHORTPTR(a8);
for (y = 0; y < height; y++) {
for (x = 0; x < width; x++)
sad += abs(a[x] - b[x]);
a += a_stride;
b += b_stride;
}
return sad;
}
#define high_sadMxN(m, n) \
unsigned int vp9_high_sad##m##x##n##_c(const uint8_t *src, int src_stride, \
const uint8_t *ref, int ref_stride) { \
return high_sad(src, src_stride, ref, ref_stride, m, n); \
} \
unsigned int vp9_high_sad##m##x##n##_avg_c(const uint8_t *src, int src_stride, \
const uint8_t *ref, int ref_stride, \
const uint8_t *second_pred) { \
uint16_t comp_pred[m * n]; \
vp9_high_comp_avg_pred(comp_pred, second_pred, m, n, ref, ref_stride); \
return high_sadb(src, src_stride, comp_pred, m, m, n); \
}
#define high_sadMxNxK(m, n, k) \
void vp9_high_sad##m##x##n##x##k##_c(const uint8_t *src, int src_stride, \
const uint8_t *ref, int ref_stride, \
unsigned int *sads) { \
int i; \
for (i = 0; i < k; ++i) \
sads[i] = vp9_high_sad##m##x##n##_c(src, src_stride, &ref[i], ref_stride); \
}
#define high_sadMxNx4D(m, n) \
void vp9_high_sad##m##x##n##x4d_c(const uint8_t *src, int src_stride, \
const uint8_t *const refs[], \
int ref_stride, unsigned int *sads) { \
int i; \
for (i = 0; i < 4; ++i) \
sads[i] = vp9_high_sad##m##x##n##_c(src, src_stride, refs[i], ref_stride); \
}
// 64x64
high_sadMxN(64, 64)
high_sadMxNxK(64, 64, 3)
high_sadMxNxK(64, 64, 8)
high_sadMxNx4D(64, 64)
// 64x32
high_sadMxN(64, 32)
high_sadMxNx4D(64, 32)
// 32x64
high_sadMxN(32, 64)
high_sadMxNx4D(32, 64)
// 32x32
high_sadMxN(32, 32)
high_sadMxNxK(32, 32, 3)
high_sadMxNxK(32, 32, 8)
high_sadMxNx4D(32, 32)
// 32x16
high_sadMxN(32, 16)
high_sadMxNx4D(32, 16)
// 16x32
high_sadMxN(16, 32)
high_sadMxNx4D(16, 32)
// 16x16
high_sadMxN(16, 16)
high_sadMxNxK(16, 16, 3)
high_sadMxNxK(16, 16, 8)
high_sadMxNx4D(16, 16)
// 16x8
high_sadMxN(16, 8)
high_sadMxNxK(16, 8, 3)
high_sadMxNxK(16, 8, 8)
high_sadMxNx4D(16, 8)
// 8x16
high_sadMxN(8, 16)
high_sadMxNxK(8, 16, 3)
high_sadMxNxK(8, 16, 8)
high_sadMxNx4D(8, 16)
// 8x8
high_sadMxN(8, 8)
high_sadMxNxK(8, 8, 3)
high_sadMxNxK(8, 8, 8)
high_sadMxNx4D(8, 8)
// 8x4
high_sadMxN(8, 4)
high_sadMxNxK(8, 4, 8)
high_sadMxNx4D(8, 4)
// 4x8
high_sadMxN(4, 8)
high_sadMxNxK(4, 8, 8)
high_sadMxNx4D(4, 8)
// 4x4
high_sadMxN(4, 4)
high_sadMxNxK(4, 4, 3)
high_sadMxNxK(4, 4, 8)
high_sadMxNx4D(4, 4)
#endif // CONFIG_VP9_HIGHBITDEPTH
......@@ -267,3 +267,375 @@ void vp9_comp_avg_pred(uint8_t *comp_pred, const uint8_t *pred, int width,
ref += ref_stride;
}
}
#if CONFIG_VP9_HIGHBITDEPTH
void high_variance64(const uint8_t *a8, int a_stride,
const uint8_t *b8, int b_stride,
int w, int h, uint64_t *sse,
uint64_t *sum) {
int i, j;
uint16_t *a = CONVERT_TO_SHORTPTR(a8);
uint16_t *b = CONVERT_TO_SHORTPTR(b8);
*sum = 0;
*sse = 0;
for (i = 0; i < h; i++) {
for (j = 0; j < w; j++) {
const int diff = a[j] - b[j];
*sum += diff;
*sse += diff * diff;
}
a += a_stride;
b += b_stride;
}
}
void high_variance(const uint8_t *a8, int a_stride,
const uint8_t *b8, int b_stride,
int w, int h, unsigned int *sse,
int *sum) {
uint64_t sse_long = 0;
uint64_t sum_long = 0;
high_variance64(a8, a_stride, b8, b_stride, w, h, &sse_long, &sum_long);
*sse = sse_long;
*sum = sum_long;
}
void high_10_variance(const uint8_t *a8, int a_stride,
const uint8_t *b8, int b_stride,
int w, int h, unsigned int *sse,
int *sum) {
uint64_t sse_long = 0;
uint64_t sum_long = 0;
high_variance64(a8, a_stride, b8, b_stride, w, h, &sse_long, &sum_long);
*sum = ROUND_POWER_OF_TWO(sum_long, 2);
*sse = ROUND_POWER_OF_TWO(sse_long, 4);
}
void high_12_variance(const uint8_t *a8, int a_stride,
const uint8_t *b8, int b_stride,
int w, int h, unsigned int *sse,
int *sum) {
uint64_t sse_long = 0;
uint64_t sum_long = 0;
high_variance64(a8, a_stride, b8, b_stride, w, h, &sse_long, &sum_long);
*sum = ROUND_POWER_OF_TWO(sum_long, 4);
*sse = ROUND_POWER_OF_TWO(sse_long, 8);
}
static void high_var_filter_block2d_bil_first_pass(
const uint8_t *src_ptr8,
uint16_t *output_ptr,
unsigned int src_pixels_per_line,
int pixel_step,
unsigned int output_height,
unsigned int output_width,
const int16_t *vp9_filter) {
unsigned int i, j;
uint16_t *src_ptr = CONVERT_TO_SHORTPTR(src_ptr8);
for (i = 0; i < output_height; i++) {
for (j = 0; j < output_width; j++) {
output_ptr[j] =
ROUND_POWER_OF_TWO((int)src_ptr[0] * vp9_filter[0] +
(int)src_ptr[pixel_step] * vp9_filter[1],
FILTER_BITS);
src_ptr++;
}
// Next row...
src_ptr += src_pixels_per_line - output_width;
output_ptr += output_width;
}
}
static void high_var_filter_block2d_bil_second_pass(
const uint16_t *src_ptr,
uint16_t *output_ptr,
unsigned int src_pixels_per_line,
unsigned int pixel_step,
unsigned int output_height,
unsigned int output_width,
const int16_t *vp9_filter) {
unsigned int i, j;
for (i = 0; i < output_height; i++) {
for (j = 0; j < output_width; j++) {
output_ptr[j] =
ROUND_POWER_OF_TWO((int)src_ptr[0] * vp9_filter[0] +
(int)src_ptr[pixel_step] * vp9_filter[1],
FILTER_BITS);
src_ptr++;
}
src_ptr += src_pixels_per_line - output_width;
output_ptr += output_width;
}
}
#define HIGH_VAR(W, H) \
unsigned int vp9_high_variance##W##x##H##_c(const uint8_t *a, int a_stride, \
const uint8_t *b, int b_stride, \
unsigned int *sse) { \
int sum; \
high_variance(a, a_stride, b, b_stride, W, H, sse, &sum); \
return *sse - (((int64_t)sum * sum) / (W * H)); \
} \
\
unsigned int vp9_high_10_variance##W##x##H##_c(const uint8_t *a, int a_stride, \
const uint8_t *b, int b_stride, \
unsigned int *sse) { \
int sum; \
high_10_variance(a, a_stride, b, b_stride, W, H, sse, &sum); \
return *sse - (((int64_t)sum * sum) / (W * H)); \
} \
\
unsigned int vp9_high_12_variance##W##x##H##_c(const uint8_t *a, int a_stride, \
const uint8_t *b, int b_stride, \
unsigned int *sse) { \
int sum; \
high_12_variance(a, a_stride, b, b_stride, W, H, sse, &sum); \
return *sse - (((int64_t)sum * sum) / (W * H)); \
}
#define HIGH_SUBPIX_VAR(W, H) \
unsigned int vp9_high_sub_pixel_variance##W##x##H##_c( \
const uint8_t *src, int src_stride, \
int xoffset, int yoffset, \
const uint8_t *dst, int dst_stride, \
unsigned int *sse) { \
uint16_t fdata3[(H + 1) * W]; \
uint16_t temp2[H * W]; \
\
high_var_filter_block2d_bil_first_pass(src, fdata3, src_stride, 1, H + 1, \
W, BILINEAR_FILTERS_2TAP(xoffset)); \
high_var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \
BILINEAR_FILTERS_2TAP(yoffset)); \
\
return vp9_high_variance##W##x##H##_c(CONVERT_TO_BYTEPTR(temp2), W, dst, \
dst_stride, sse); \
} \
\
unsigned int vp9_high_10_sub_pixel_variance##W##x##H##_c( \
const uint8_t *src, int src_stride, \
int xoffset, int yoffset, \
const uint8_t *dst, int dst_stride, \
unsigned int *sse) { \
uint16_t fdata3[(H + 1) * W]; \
uint16_t temp2[H * W]; \
\
high_var_filter_block2d_bil_first_pass(src, fdata3, src_stride, 1, H + 1, \
W, BILINEAR_FILTERS_2TAP(xoffset)); \
high_var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \
BILINEAR_FILTERS_2TAP(yoffset)); \
\
return vp9_high_10_variance##W##x##H##_c(CONVERT_TO_BYTEPTR(temp2), W, dst, \
dst_stride, sse); \
} \
\
unsigned int vp9_high_12_sub_pixel_variance##W##x##H##_c( \
const uint8_t *src, int src_stride, \
int xoffset, int yoffset, \
const uint8_t *dst, int dst_stride, \
unsigned int *sse) { \
uint16_t fdata3[(H + 1) * W]; \
uint16_t temp2[H * W]; \
\
high_var_filter_block2d_bil_first_pass(src, fdata3, src_stride, 1, H + 1, \
W, BILINEAR_FILTERS_2TAP(xoffset)); \
high_var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \
BILINEAR_FILTERS_2TAP(yoffset)); \
\
return vp9_high_12_variance##W##x##H##_c(CONVERT_TO_BYTEPTR(temp2), W, dst, \
dst_stride, sse); \
}
#define HIGH_SUBPIX_AVG_VAR(W, H) \
unsigned int vp9_high_sub_pixel_avg_variance##W##x##H##_c( \
const uint8_t *src, int src_stride, \
int xoffset, int yoffset, \
const uint8_t *dst, int dst_stride, \
unsigned int *sse, \
const uint8_t *second_pred) { \
uint16_t fdata3[(H + 1) * W]; \
uint16_t temp2[H * W]; \
DECLARE_ALIGNED_ARRAY(16, uint16_t, temp3, H * W); \
\
high_var_filter_block2d_bil_first_pass(src, fdata3, src_stride, 1, H + 1, \
W, BILINEAR_FILTERS_2TAP(xoffset)); \
high_var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \
BILINEAR_FILTERS_2TAP(yoffset)); \
\
vp9_high_comp_avg_pred(temp3, second_pred, W, H, CONVERT_TO_BYTEPTR(temp2), \
W); \
\
return vp9_high_variance##W##x##H##_c(CONVERT_TO_BYTEPTR(temp3), W, dst, \
dst_stride, sse); \
} \
\
unsigned int vp9_high_10_sub_pixel_avg_variance##W##x##H##_c( \
const uint8_t *src, int src_stride, \
int xoffset, int yoffset, \
const uint8_t *dst, int dst_stride, \
unsigned int *sse, \
const uint8_t *second_pred) { \
uint16_t fdata3[(H + 1) * W]; \
uint16_t temp2[H * W]; \
DECLARE_ALIGNED_ARRAY(16, uint16_t, temp3, H * W); \
\
high_var_filter_block2d_bil_first_pass(src, fdata3, src_stride, 1, H + 1, \
W, BILINEAR_FILTERS_2TAP(xoffset)); \
high_var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \
BILINEAR_FILTERS_2TAP(yoffset)); \
\
vp9_high_comp_avg_pred(temp3, second_pred, W, H, CONVERT_TO_BYTEPTR(temp2), \
W); \
\
return vp9_high_10_variance##W##x##H##_c(CONVERT_TO_BYTEPTR(temp3), W, dst, \
dst_stride, sse); \
} \
\
unsigned int vp9_high_12_sub_pixel_avg_variance##W##x##H##_c( \
const uint8_t *src, int src_stride, \
int xoffset, int yoffset, \
const uint8_t *dst, int dst_stride, \
unsigned int *sse, \
const uint8_t *second_pred) { \
uint16_t fdata3[(H + 1) * W]; \
uint16_t temp2[H * W]; \
DECLARE_ALIGNED_ARRAY(16, uint16_t, temp3, H * W); \
\
high_var_filter_block2d_bil_first_pass(src, fdata3, src_stride, 1, H + 1, \
W, BILINEAR_FILTERS_2TAP(xoffset)); \
high_var_filter_block2d_bil_second_pass(fdata3, temp2, W, W, H, W, \
BILINEAR_FILTERS_2TAP(yoffset)); \
\
vp9_high_comp_avg_pred(temp3, second_pred, W, H, CONVERT_TO_BYTEPTR(temp2), \
W); \
\
return vp9_high_12_variance##W##x##H##_c(CONVERT_TO_BYTEPTR(temp3), W, dst, \
dst_stride, sse); \
}
#define HIGH_GET_VAR(S) \
void vp9_high_get##S##x##S##var_c(const uint8_t *src, int src_stride, \
const uint8_t *ref, int ref_stride, \
unsigned int *sse, int *sum) { \
high_variance(src, src_stride, ref, ref_stride, S, S, sse, sum); \
} \
\
void vp9_high_10_get##S##x##S##var_c(const uint8_t *src, int src_stride, \
const uint8_t *ref, int ref_stride, \
unsigned int *sse, int *sum) { \
high_10_variance(src, src_stride, ref, ref_stride, S, S, sse, sum); \
} \
\
void vp9_high_12_get##S##x##S##var_c(const uint8_t *src, int src_stride, \
const uint8_t *ref, int ref_stride, \
unsigned int *sse, int *sum) { \
high_12_variance(src, src_stride, ref, ref_stride, S, S, sse, sum); \
}
#define HIGH_MSE(W, H) \
unsigned int vp9_high_mse##W##x##H##_c(const uint8_t *src, int src_stride, \
const uint8_t *ref, int ref_stride, \
unsigned int *sse) { \
int sum; \
high_variance(src, src_stride, ref, ref_stride, W, H, sse, &sum); \
return *sse; \
} \
\
unsigned int vp9_high_10_mse##W##x##H##_c(const uint8_t *src, int src_stride, \
const uint8_t *ref, int ref_stride, \
unsigned int *sse) { \
int sum; \
high_10_variance(src, src_stride, ref, ref_stride, W, H, sse, &sum); \
return *sse; \
} \
\
unsigned int vp9_high_12_mse##W##x##H##_c(const uint8_t *src, int src_stride, \
const uint8_t *ref, int ref_stride, \
unsigned int *sse) { \
int sum; \
high_12_variance(src, src_stride, ref, ref_stride, W, H, sse, &sum); \
return *sse; \
}
HIGH_GET_VAR(8)
HIGH_GET_VAR(16)
HIGH_MSE(16, 16)
HIGH_MSE(16, 8)
HIGH_MSE(8, 16)
HIGH_MSE(8, 8)
HIGH_VAR(4, 4)
HIGH_SUBPIX_VAR(4, 4)
HIGH_SUBPIX_AVG_VAR(4, 4)
HIGH_VAR(4, 8)
HIGH_SUBPIX_VAR(4, 8)
HIGH_SUBPIX_AVG_VAR(4, 8)
HIGH_VAR(8, 4)
HIGH_SUBPIX_VAR(8, 4)
HIGH_SUBPIX_AVG_VAR(8, 4)
HIGH_VAR(8, 8)
HIGH_SUBPIX_VAR(8, 8)
HIGH_SUBPIX_AVG_VAR(8, 8)
HIGH_VAR(8, 16)
HIGH_SUBPIX_VAR(8, 16)
HIGH_SUBPIX_AVG_VAR(8, 16)
HIGH_VAR(16, 8)
HIGH_SUBPIX_VAR(16, 8)
HIGH_SUBPIX_AVG_VAR(16, 8)
HIGH_VAR(16, 16)
HIGH_SUBPIX_VAR(16, 16)
HIGH_SUBPIX_AVG_VAR(16, 16)
HIGH_VAR(16, 32)
HIGH_SUBPIX_VAR(16, 32)
HIGH_SUBPIX_AVG_VAR(16, 32)
HIGH_VAR(32, 16)
HIGH_SUBPIX_VAR(32, 16)
HIGH_SUBPIX_AVG_VAR(32, 16)
HIGH_VAR(32, 32)
HIGH_SUBPIX_VAR(32, 32)
HIGH_SUBPIX_AVG_VAR(32, 32)
HIGH_VAR(32, 64)
HIGH_SUBPIX_VAR(32, 64)
HIGH_SUBPIX_AVG_VAR(32, 64)
HIGH_VAR(64, 32)
HIGH_SUBPIX_VAR(64, 32)
HIGH_SUBPIX_AVG_VAR(64, 32)
HIGH_VAR(64, 64)
HIGH_SUBPIX_VAR(64, 64)
HIGH_SUBPIX_AVG_VAR(64, 64)
void vp9_high_comp_avg_pred(uint16_t *comp_pred, const uint8_t *pred8,
int width, int height, const uint8_t *ref8,
int ref_stride) {
int i, j;
uint16_t *pred = CONVERT_TO_SHORTPTR(pred8);
uint16_t *ref = CONVERT_TO_SHORTPTR(ref8);
for (i = 0; i < height; i++) {
for (j = 0; j < width; j++) {
const int tmp = pred[j] + ref[j];
comp_pred[j] = ROUND_POWER_OF_TWO(tmp, 1);
}
comp_pred += width;
pred += width;
ref += ref_stride;
}
}
#endif // CONFIG_VP9_HIGHBITDEPTH
......@@ -22,6 +22,23 @@ void variance(const uint8_t *a, int a_stride,
int w, int h,
unsigned int *sse, int *sum);
#if CONFIG_VP9_HIGHBITDEPTH
void high_variance(const uint8_t *a8, int a_stride,
const uint8_t *b8, int b_stride,
int w, int h,
unsigned int *sse, int *sum);
void high_10_variance(const uint8_t *a8, int a_stride,
const uint8_t *b8, int b_stride,
int w, int h,
unsigned int *sse, int *sum);
void high_12_variance(const uint8_t *a8, int a_stride,
const uint8_t *b8, int b_stride,
int w, int h,
unsigned int *sse, int *sum);
#endif
typedef unsigned int(*vp9_sad_fn_t)(const uint8_t *src_ptr,
int source_stride,
const uint8_t *ref_ptr,
......@@ -81,6 +98,11 @@ typedef struct vp9_variance_vtable {
void vp9_comp_avg_pred(uint8_t *comp_pred, const uint8_t *pred, int width,
int height, const uint8_t *ref, int ref_stride);
#if CONFIG_VP9_HIGHBITDEPTH
void vp9_high_comp_avg_pred(uint16_t *comp_pred, const uint8_t *pred, int width,
int height, const uint8_t *ref, int ref_stride);
#endif
#ifdef __cplusplus
} // extern "C"
#endif
......
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