2015-05-15 20:52:03 +02:00
|
|
|
/*
|
|
|
|
|
* Copyright (c) 2014 The WebM project authors. All Rights Reserved.
|
|
|
|
|
*
|
|
|
|
|
* Use of this source code is governed by a BSD-style license
|
|
|
|
|
* that can be found in the LICENSE file in the root of the source
|
|
|
|
|
* tree. An additional intellectual property rights grant can be found
|
|
|
|
|
* in the file PATENTS. All contributing project authors may
|
|
|
|
|
* be found in the AUTHORS file in the root of the source tree.
|
|
|
|
|
*/
|
|
|
|
|
#include "./vpx_config.h"
|
|
|
|
|
|
|
|
|
|
#include "vpx_ports/mem.h"
|
|
|
|
|
|
2016-07-23 05:07:03 +02:00
|
|
|
typedef uint32_t (*high_variance_fn_t)(const uint16_t *src, int src_stride,
|
|
|
|
|
const uint16_t *ref, int ref_stride,
|
|
|
|
|
uint32_t *sse, int *sum);
|
2015-05-15 20:52:03 +02:00
|
|
|
|
|
|
|
|
uint32_t vpx_highbd_calc8x8var_sse2(const uint16_t *src, int src_stride,
|
|
|
|
|
const uint16_t *ref, int ref_stride,
|
|
|
|
|
uint32_t *sse, int *sum);
|
|
|
|
|
|
|
|
|
|
uint32_t vpx_highbd_calc16x16var_sse2(const uint16_t *src, int src_stride,
|
|
|
|
|
const uint16_t *ref, int ref_stride,
|
|
|
|
|
uint32_t *sse, int *sum);
|
|
|
|
|
|
|
|
|
|
static void highbd_8_variance_sse2(const uint16_t *src, int src_stride,
|
2016-07-23 05:07:03 +02:00
|
|
|
const uint16_t *ref, int ref_stride, int w,
|
|
|
|
|
int h, uint32_t *sse, int *sum,
|
2015-05-15 20:52:03 +02:00
|
|
|
high_variance_fn_t var_fn, int block_size) {
|
|
|
|
|
int i, j;
|
|
|
|
|
|
|
|
|
|
*sse = 0;
|
|
|
|
|
*sum = 0;
|
|
|
|
|
|
|
|
|
|
for (i = 0; i < h; i += block_size) {
|
|
|
|
|
for (j = 0; j < w; j += block_size) {
|
|
|
|
|
unsigned int sse0;
|
|
|
|
|
int sum0;
|
2016-07-23 05:07:03 +02:00
|
|
|
var_fn(src + src_stride * i + j, src_stride, ref + ref_stride * i + j,
|
|
|
|
|
ref_stride, &sse0, &sum0);
|
2015-05-15 20:52:03 +02:00
|
|
|
*sse += sse0;
|
|
|
|
|
*sum += sum0;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static void highbd_10_variance_sse2(const uint16_t *src, int src_stride,
|
2016-07-23 05:07:03 +02:00
|
|
|
const uint16_t *ref, int ref_stride, int w,
|
|
|
|
|
int h, uint32_t *sse, int *sum,
|
2015-05-15 20:52:03 +02:00
|
|
|
high_variance_fn_t var_fn, int block_size) {
|
|
|
|
|
int i, j;
|
|
|
|
|
uint64_t sse_long = 0;
|
2015-07-07 23:22:07 +02:00
|
|
|
int32_t sum_long = 0;
|
2015-05-15 20:52:03 +02:00
|
|
|
|
|
|
|
|
for (i = 0; i < h; i += block_size) {
|
|
|
|
|
for (j = 0; j < w; j += block_size) {
|
|
|
|
|
unsigned int sse0;
|
|
|
|
|
int sum0;
|
2016-07-23 05:07:03 +02:00
|
|
|
var_fn(src + src_stride * i + j, src_stride, ref + ref_stride * i + j,
|
|
|
|
|
ref_stride, &sse0, &sum0);
|
2015-05-15 20:52:03 +02:00
|
|
|
sse_long += sse0;
|
|
|
|
|
sum_long += sum0;
|
|
|
|
|
}
|
|
|
|
|
}
|
2016-01-13 20:18:19 +01:00
|
|
|
*sum = ROUND_POWER_OF_TWO(sum_long, 2);
|
|
|
|
|
*sse = (uint32_t)ROUND_POWER_OF_TWO(sse_long, 4);
|
2015-05-15 20:52:03 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static void highbd_12_variance_sse2(const uint16_t *src, int src_stride,
|
2016-07-23 05:07:03 +02:00
|
|
|
const uint16_t *ref, int ref_stride, int w,
|
|
|
|
|
int h, uint32_t *sse, int *sum,
|
2015-05-15 20:52:03 +02:00
|
|
|
high_variance_fn_t var_fn, int block_size) {
|
|
|
|
|
int i, j;
|
|
|
|
|
uint64_t sse_long = 0;
|
2015-07-07 23:22:07 +02:00
|
|
|
int32_t sum_long = 0;
|
2015-05-15 20:52:03 +02:00
|
|
|
|
|
|
|
|
for (i = 0; i < h; i += block_size) {
|
|
|
|
|
for (j = 0; j < w; j += block_size) {
|
|
|
|
|
unsigned int sse0;
|
|
|
|
|
int sum0;
|
2016-07-23 05:07:03 +02:00
|
|
|
var_fn(src + src_stride * i + j, src_stride, ref + ref_stride * i + j,
|
|
|
|
|
ref_stride, &sse0, &sum0);
|
2015-05-15 20:52:03 +02:00
|
|
|
sse_long += sse0;
|
|
|
|
|
sum_long += sum0;
|
|
|
|
|
}
|
|
|
|
|
}
|
2016-01-13 20:18:19 +01:00
|
|
|
*sum = ROUND_POWER_OF_TWO(sum_long, 4);
|
|
|
|
|
*sse = (uint32_t)ROUND_POWER_OF_TWO(sse_long, 8);
|
2015-05-15 20:52:03 +02:00
|
|
|
}
|
|
|
|
|
|
2016-07-23 05:07:03 +02:00
|
|
|
#define HIGH_GET_VAR(S) \
|
|
|
|
|
void vpx_highbd_get##S##x##S##var_sse2(const uint8_t *src8, int src_stride, \
|
|
|
|
|
const uint8_t *ref8, int ref_stride, \
|
|
|
|
|
uint32_t *sse, int *sum) { \
|
|
|
|
|
uint16_t *src = CONVERT_TO_SHORTPTR(src8); \
|
|
|
|
|
uint16_t *ref = CONVERT_TO_SHORTPTR(ref8); \
|
|
|
|
|
vpx_highbd_calc##S##x##S##var_sse2(src, src_stride, ref, ref_stride, sse, \
|
|
|
|
|
sum); \
|
|
|
|
|
} \
|
|
|
|
|
\
|
|
|
|
|
void vpx_highbd_10_get##S##x##S##var_sse2( \
|
|
|
|
|
const uint8_t *src8, int src_stride, const uint8_t *ref8, \
|
|
|
|
|
int ref_stride, uint32_t *sse, int *sum) { \
|
|
|
|
|
uint16_t *src = CONVERT_TO_SHORTPTR(src8); \
|
|
|
|
|
uint16_t *ref = CONVERT_TO_SHORTPTR(ref8); \
|
|
|
|
|
vpx_highbd_calc##S##x##S##var_sse2(src, src_stride, ref, ref_stride, sse, \
|
|
|
|
|
sum); \
|
|
|
|
|
*sum = ROUND_POWER_OF_TWO(*sum, 2); \
|
|
|
|
|
*sse = ROUND_POWER_OF_TWO(*sse, 4); \
|
|
|
|
|
} \
|
|
|
|
|
\
|
|
|
|
|
void vpx_highbd_12_get##S##x##S##var_sse2( \
|
|
|
|
|
const uint8_t *src8, int src_stride, const uint8_t *ref8, \
|
|
|
|
|
int ref_stride, uint32_t *sse, int *sum) { \
|
|
|
|
|
uint16_t *src = CONVERT_TO_SHORTPTR(src8); \
|
|
|
|
|
uint16_t *ref = CONVERT_TO_SHORTPTR(ref8); \
|
|
|
|
|
vpx_highbd_calc##S##x##S##var_sse2(src, src_stride, ref, ref_stride, sse, \
|
|
|
|
|
sum); \
|
|
|
|
|
*sum = ROUND_POWER_OF_TWO(*sum, 4); \
|
|
|
|
|
*sse = ROUND_POWER_OF_TWO(*sse, 8); \
|
|
|
|
|
}
|
2015-05-15 20:52:03 +02:00
|
|
|
|
|
|
|
|
HIGH_GET_VAR(16);
|
|
|
|
|
HIGH_GET_VAR(8);
|
|
|
|
|
|
|
|
|
|
#undef HIGH_GET_VAR
|
|
|
|
|
|
2016-07-23 05:07:03 +02:00
|
|
|
#define VAR_FN(w, h, block_size, shift) \
|
|
|
|
|
uint32_t vpx_highbd_8_variance##w##x##h##_sse2( \
|
|
|
|
|
const uint8_t *src8, int src_stride, const uint8_t *ref8, \
|
|
|
|
|
int ref_stride, uint32_t *sse) { \
|
|
|
|
|
int sum; \
|
|
|
|
|
uint16_t *src = CONVERT_TO_SHORTPTR(src8); \
|
|
|
|
|
uint16_t *ref = CONVERT_TO_SHORTPTR(ref8); \
|
|
|
|
|
highbd_8_variance_sse2( \
|
|
|
|
|
src, src_stride, ref, ref_stride, w, h, sse, &sum, \
|
|
|
|
|
vpx_highbd_calc##block_size##x##block_size##var_sse2, block_size); \
|
2017-04-05 05:37:17 +02:00
|
|
|
return *sse - (uint32_t)(((int64_t)sum * sum) >> shift); \
|
2016-07-23 05:07:03 +02:00
|
|
|
} \
|
|
|
|
|
\
|
|
|
|
|
uint32_t vpx_highbd_10_variance##w##x##h##_sse2( \
|
|
|
|
|
const uint8_t *src8, int src_stride, const uint8_t *ref8, \
|
|
|
|
|
int ref_stride, uint32_t *sse) { \
|
|
|
|
|
int sum; \
|
|
|
|
|
int64_t var; \
|
|
|
|
|
uint16_t *src = CONVERT_TO_SHORTPTR(src8); \
|
|
|
|
|
uint16_t *ref = CONVERT_TO_SHORTPTR(ref8); \
|
|
|
|
|
highbd_10_variance_sse2( \
|
|
|
|
|
src, src_stride, ref, ref_stride, w, h, sse, &sum, \
|
|
|
|
|
vpx_highbd_calc##block_size##x##block_size##var_sse2, block_size); \
|
|
|
|
|
var = (int64_t)(*sse) - (((int64_t)sum * sum) >> shift); \
|
|
|
|
|
return (var >= 0) ? (uint32_t)var : 0; \
|
|
|
|
|
} \
|
|
|
|
|
\
|
|
|
|
|
uint32_t vpx_highbd_12_variance##w##x##h##_sse2( \
|
|
|
|
|
const uint8_t *src8, int src_stride, const uint8_t *ref8, \
|
|
|
|
|
int ref_stride, uint32_t *sse) { \
|
|
|
|
|
int sum; \
|
|
|
|
|
int64_t var; \
|
|
|
|
|
uint16_t *src = CONVERT_TO_SHORTPTR(src8); \
|
|
|
|
|
uint16_t *ref = CONVERT_TO_SHORTPTR(ref8); \
|
|
|
|
|
highbd_12_variance_sse2( \
|
|
|
|
|
src, src_stride, ref, ref_stride, w, h, sse, &sum, \
|
|
|
|
|
vpx_highbd_calc##block_size##x##block_size##var_sse2, block_size); \
|
|
|
|
|
var = (int64_t)(*sse) - (((int64_t)sum * sum) >> shift); \
|
|
|
|
|
return (var >= 0) ? (uint32_t)var : 0; \
|
|
|
|
|
}
|
2015-05-15 20:52:03 +02:00
|
|
|
|
|
|
|
|
VAR_FN(64, 64, 16, 12);
|
|
|
|
|
VAR_FN(64, 32, 16, 11);
|
|
|
|
|
VAR_FN(32, 64, 16, 11);
|
|
|
|
|
VAR_FN(32, 32, 16, 10);
|
|
|
|
|
VAR_FN(32, 16, 16, 9);
|
|
|
|
|
VAR_FN(16, 32, 16, 9);
|
|
|
|
|
VAR_FN(16, 16, 16, 8);
|
|
|
|
|
VAR_FN(16, 8, 8, 7);
|
|
|
|
|
VAR_FN(8, 16, 8, 7);
|
|
|
|
|
VAR_FN(8, 8, 8, 6);
|
|
|
|
|
|
|
|
|
|
#undef VAR_FN
|
|
|
|
|
|
|
|
|
|
unsigned int vpx_highbd_8_mse16x16_sse2(const uint8_t *src8, int src_stride,
|
2016-07-23 05:07:03 +02:00
|
|
|
const uint8_t *ref8, int ref_stride,
|
|
|
|
|
unsigned int *sse) {
|
2015-05-15 20:52:03 +02:00
|
|
|
int sum;
|
|
|
|
|
uint16_t *src = CONVERT_TO_SHORTPTR(src8);
|
|
|
|
|
uint16_t *ref = CONVERT_TO_SHORTPTR(ref8);
|
2016-07-23 05:07:03 +02:00
|
|
|
highbd_8_variance_sse2(src, src_stride, ref, ref_stride, 16, 16, sse, &sum,
|
|
|
|
|
vpx_highbd_calc16x16var_sse2, 16);
|
2015-05-15 20:52:03 +02:00
|
|
|
return *sse;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
unsigned int vpx_highbd_10_mse16x16_sse2(const uint8_t *src8, int src_stride,
|
|
|
|
|
const uint8_t *ref8, int ref_stride,
|
|
|
|
|
unsigned int *sse) {
|
|
|
|
|
int sum;
|
|
|
|
|
uint16_t *src = CONVERT_TO_SHORTPTR(src8);
|
|
|
|
|
uint16_t *ref = CONVERT_TO_SHORTPTR(ref8);
|
2016-07-23 05:07:03 +02:00
|
|
|
highbd_10_variance_sse2(src, src_stride, ref, ref_stride, 16, 16, sse, &sum,
|
|
|
|
|
vpx_highbd_calc16x16var_sse2, 16);
|
2015-05-15 20:52:03 +02:00
|
|
|
return *sse;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
unsigned int vpx_highbd_12_mse16x16_sse2(const uint8_t *src8, int src_stride,
|
|
|
|
|
const uint8_t *ref8, int ref_stride,
|
|
|
|
|
unsigned int *sse) {
|
|
|
|
|
int sum;
|
|
|
|
|
uint16_t *src = CONVERT_TO_SHORTPTR(src8);
|
|
|
|
|
uint16_t *ref = CONVERT_TO_SHORTPTR(ref8);
|
2016-07-23 05:07:03 +02:00
|
|
|
highbd_12_variance_sse2(src, src_stride, ref, ref_stride, 16, 16, sse, &sum,
|
|
|
|
|
vpx_highbd_calc16x16var_sse2, 16);
|
2015-05-15 20:52:03 +02:00
|
|
|
return *sse;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
unsigned int vpx_highbd_8_mse8x8_sse2(const uint8_t *src8, int src_stride,
|
2016-07-23 05:07:03 +02:00
|
|
|
const uint8_t *ref8, int ref_stride,
|
|
|
|
|
unsigned int *sse) {
|
2015-05-15 20:52:03 +02:00
|
|
|
int sum;
|
|
|
|
|
uint16_t *src = CONVERT_TO_SHORTPTR(src8);
|
|
|
|
|
uint16_t *ref = CONVERT_TO_SHORTPTR(ref8);
|
2016-07-23 05:07:03 +02:00
|
|
|
highbd_8_variance_sse2(src, src_stride, ref, ref_stride, 8, 8, sse, &sum,
|
|
|
|
|
vpx_highbd_calc8x8var_sse2, 8);
|
2015-05-15 20:52:03 +02:00
|
|
|
return *sse;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
unsigned int vpx_highbd_10_mse8x8_sse2(const uint8_t *src8, int src_stride,
|
|
|
|
|
const uint8_t *ref8, int ref_stride,
|
|
|
|
|
unsigned int *sse) {
|
|
|
|
|
int sum;
|
|
|
|
|
uint16_t *src = CONVERT_TO_SHORTPTR(src8);
|
|
|
|
|
uint16_t *ref = CONVERT_TO_SHORTPTR(ref8);
|
2016-07-23 05:07:03 +02:00
|
|
|
highbd_10_variance_sse2(src, src_stride, ref, ref_stride, 8, 8, sse, &sum,
|
|
|
|
|
vpx_highbd_calc8x8var_sse2, 8);
|
2015-05-15 20:52:03 +02:00
|
|
|
return *sse;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
unsigned int vpx_highbd_12_mse8x8_sse2(const uint8_t *src8, int src_stride,
|
|
|
|
|
const uint8_t *ref8, int ref_stride,
|
|
|
|
|
unsigned int *sse) {
|
|
|
|
|
int sum;
|
|
|
|
|
uint16_t *src = CONVERT_TO_SHORTPTR(src8);
|
|
|
|
|
uint16_t *ref = CONVERT_TO_SHORTPTR(ref8);
|
2016-07-23 05:07:03 +02:00
|
|
|
highbd_12_variance_sse2(src, src_stride, ref, ref_stride, 8, 8, sse, &sum,
|
|
|
|
|
vpx_highbd_calc8x8var_sse2, 8);
|
2015-05-15 20:52:03 +02:00
|
|
|
return *sse;
|
|
|
|
|
}
|
2015-06-05 18:54:19 +02:00
|
|
|
|
2015-12-14 17:35:29 +01:00
|
|
|
// The 2 unused parameters are place holders for PIC enabled build.
|
|
|
|
|
// These definitions are for functions defined in
|
|
|
|
|
// highbd_subpel_variance_impl_sse2.asm
|
2016-07-23 05:07:03 +02:00
|
|
|
#define DECL(w, opt) \
|
|
|
|
|
int vpx_highbd_sub_pixel_variance##w##xh_##opt( \
|
|
|
|
|
const uint16_t *src, ptrdiff_t src_stride, int x_offset, int y_offset, \
|
|
|
|
|
const uint16_t *dst, ptrdiff_t dst_stride, int height, \
|
|
|
|
|
unsigned int *sse, void *unused0, void *unused);
|
2016-05-07 03:38:04 +02:00
|
|
|
#define DECLS(opt) \
|
2016-07-23 05:07:03 +02:00
|
|
|
DECL(8, opt); \
|
2016-05-07 03:38:04 +02:00
|
|
|
DECL(16, opt)
|
2015-06-05 18:54:19 +02:00
|
|
|
|
2016-05-07 03:38:04 +02:00
|
|
|
DECLS(sse2);
|
2016-05-07 03:57:04 +02:00
|
|
|
|
2015-06-05 18:54:19 +02:00
|
|
|
#undef DECLS
|
|
|
|
|
#undef DECL
|
|
|
|
|
|
2016-07-23 05:07:03 +02:00
|
|
|
#define FN(w, h, wf, wlog2, hlog2, opt, cast) \
|
|
|
|
|
uint32_t vpx_highbd_8_sub_pixel_variance##w##x##h##_##opt( \
|
|
|
|
|
const uint8_t *src8, int src_stride, int x_offset, int y_offset, \
|
|
|
|
|
const uint8_t *dst8, int dst_stride, uint32_t *sse_ptr) { \
|
|
|
|
|
uint32_t sse; \
|
|
|
|
|
uint16_t *src = CONVERT_TO_SHORTPTR(src8); \
|
|
|
|
|
uint16_t *dst = CONVERT_TO_SHORTPTR(dst8); \
|
|
|
|
|
int se = vpx_highbd_sub_pixel_variance##wf##xh_##opt( \
|
|
|
|
|
src, src_stride, x_offset, y_offset, dst, dst_stride, h, &sse, NULL, \
|
|
|
|
|
NULL); \
|
|
|
|
|
if (w > wf) { \
|
|
|
|
|
unsigned int sse2; \
|
|
|
|
|
int se2 = vpx_highbd_sub_pixel_variance##wf##xh_##opt( \
|
|
|
|
|
src + 16, src_stride, x_offset, y_offset, dst + 16, dst_stride, h, \
|
|
|
|
|
&sse2, NULL, NULL); \
|
|
|
|
|
se += se2; \
|
|
|
|
|
sse += sse2; \
|
|
|
|
|
if (w > wf * 2) { \
|
|
|
|
|
se2 = vpx_highbd_sub_pixel_variance##wf##xh_##opt( \
|
|
|
|
|
src + 32, src_stride, x_offset, y_offset, dst + 32, dst_stride, h, \
|
|
|
|
|
&sse2, NULL, NULL); \
|
|
|
|
|
se += se2; \
|
|
|
|
|
sse += sse2; \
|
|
|
|
|
se2 = vpx_highbd_sub_pixel_variance##wf##xh_##opt( \
|
|
|
|
|
src + 48, src_stride, x_offset, y_offset, dst + 48, dst_stride, h, \
|
|
|
|
|
&sse2, NULL, NULL); \
|
|
|
|
|
se += se2; \
|
|
|
|
|
sse += sse2; \
|
|
|
|
|
} \
|
|
|
|
|
} \
|
|
|
|
|
*sse_ptr = sse; \
|
2017-04-05 05:37:17 +02:00
|
|
|
return sse - (uint32_t)((cast se * se) >> (wlog2 + hlog2)); \
|
2016-07-23 05:07:03 +02:00
|
|
|
} \
|
|
|
|
|
\
|
|
|
|
|
uint32_t vpx_highbd_10_sub_pixel_variance##w##x##h##_##opt( \
|
|
|
|
|
const uint8_t *src8, int src_stride, int x_offset, int y_offset, \
|
|
|
|
|
const uint8_t *dst8, int dst_stride, uint32_t *sse_ptr) { \
|
2017-04-05 05:37:17 +02:00
|
|
|
int64_t var; \
|
2016-07-23 05:07:03 +02:00
|
|
|
uint32_t sse; \
|
|
|
|
|
uint16_t *src = CONVERT_TO_SHORTPTR(src8); \
|
|
|
|
|
uint16_t *dst = CONVERT_TO_SHORTPTR(dst8); \
|
|
|
|
|
int se = vpx_highbd_sub_pixel_variance##wf##xh_##opt( \
|
|
|
|
|
src, src_stride, x_offset, y_offset, dst, dst_stride, h, &sse, NULL, \
|
|
|
|
|
NULL); \
|
|
|
|
|
if (w > wf) { \
|
|
|
|
|
uint32_t sse2; \
|
|
|
|
|
int se2 = vpx_highbd_sub_pixel_variance##wf##xh_##opt( \
|
|
|
|
|
src + 16, src_stride, x_offset, y_offset, dst + 16, dst_stride, h, \
|
|
|
|
|
&sse2, NULL, NULL); \
|
|
|
|
|
se += se2; \
|
|
|
|
|
sse += sse2; \
|
|
|
|
|
if (w > wf * 2) { \
|
|
|
|
|
se2 = vpx_highbd_sub_pixel_variance##wf##xh_##opt( \
|
|
|
|
|
src + 32, src_stride, x_offset, y_offset, dst + 32, dst_stride, h, \
|
|
|
|
|
&sse2, NULL, NULL); \
|
|
|
|
|
se += se2; \
|
|
|
|
|
sse += sse2; \
|
|
|
|
|
se2 = vpx_highbd_sub_pixel_variance##wf##xh_##opt( \
|
|
|
|
|
src + 48, src_stride, x_offset, y_offset, dst + 48, dst_stride, h, \
|
|
|
|
|
&sse2, NULL, NULL); \
|
|
|
|
|
se += se2; \
|
|
|
|
|
sse += sse2; \
|
|
|
|
|
} \
|
|
|
|
|
} \
|
|
|
|
|
se = ROUND_POWER_OF_TWO(se, 2); \
|
|
|
|
|
sse = ROUND_POWER_OF_TWO(sse, 4); \
|
|
|
|
|
*sse_ptr = sse; \
|
2017-04-05 05:37:17 +02:00
|
|
|
var = (int64_t)(sse) - ((cast se * se) >> (wlog2 + hlog2)); \
|
|
|
|
|
return (var >= 0) ? (uint32_t)var : 0; \
|
2016-07-23 05:07:03 +02:00
|
|
|
} \
|
|
|
|
|
\
|
|
|
|
|
uint32_t vpx_highbd_12_sub_pixel_variance##w##x##h##_##opt( \
|
|
|
|
|
const uint8_t *src8, int src_stride, int x_offset, int y_offset, \
|
|
|
|
|
const uint8_t *dst8, int dst_stride, uint32_t *sse_ptr) { \
|
|
|
|
|
int start_row; \
|
|
|
|
|
uint32_t sse; \
|
|
|
|
|
int se = 0; \
|
2017-04-05 05:37:17 +02:00
|
|
|
int64_t var; \
|
2016-07-23 05:07:03 +02:00
|
|
|
uint64_t long_sse = 0; \
|
|
|
|
|
uint16_t *src = CONVERT_TO_SHORTPTR(src8); \
|
|
|
|
|
uint16_t *dst = CONVERT_TO_SHORTPTR(dst8); \
|
|
|
|
|
for (start_row = 0; start_row < h; start_row += 16) { \
|
|
|
|
|
uint32_t sse2; \
|
|
|
|
|
int height = h - start_row < 16 ? h - start_row : 16; \
|
|
|
|
|
int se2 = vpx_highbd_sub_pixel_variance##wf##xh_##opt( \
|
|
|
|
|
src + (start_row * src_stride), src_stride, x_offset, y_offset, \
|
|
|
|
|
dst + (start_row * dst_stride), dst_stride, height, &sse2, NULL, \
|
|
|
|
|
NULL); \
|
|
|
|
|
se += se2; \
|
|
|
|
|
long_sse += sse2; \
|
|
|
|
|
if (w > wf) { \
|
|
|
|
|
se2 = vpx_highbd_sub_pixel_variance##wf##xh_##opt( \
|
|
|
|
|
src + 16 + (start_row * src_stride), src_stride, x_offset, \
|
|
|
|
|
y_offset, dst + 16 + (start_row * dst_stride), dst_stride, height, \
|
|
|
|
|
&sse2, NULL, NULL); \
|
|
|
|
|
se += se2; \
|
|
|
|
|
long_sse += sse2; \
|
|
|
|
|
if (w > wf * 2) { \
|
|
|
|
|
se2 = vpx_highbd_sub_pixel_variance##wf##xh_##opt( \
|
|
|
|
|
src + 32 + (start_row * src_stride), src_stride, x_offset, \
|
|
|
|
|
y_offset, dst + 32 + (start_row * dst_stride), dst_stride, \
|
|
|
|
|
height, &sse2, NULL, NULL); \
|
|
|
|
|
se += se2; \
|
|
|
|
|
long_sse += sse2; \
|
|
|
|
|
se2 = vpx_highbd_sub_pixel_variance##wf##xh_##opt( \
|
|
|
|
|
src + 48 + (start_row * src_stride), src_stride, x_offset, \
|
|
|
|
|
y_offset, dst + 48 + (start_row * dst_stride), dst_stride, \
|
|
|
|
|
height, &sse2, NULL, NULL); \
|
|
|
|
|
se += se2; \
|
|
|
|
|
long_sse += sse2; \
|
|
|
|
|
} \
|
|
|
|
|
} \
|
|
|
|
|
} \
|
|
|
|
|
se = ROUND_POWER_OF_TWO(se, 4); \
|
|
|
|
|
sse = (uint32_t)ROUND_POWER_OF_TWO(long_sse, 8); \
|
|
|
|
|
*sse_ptr = sse; \
|
2017-04-05 05:37:17 +02:00
|
|
|
var = (int64_t)(sse) - ((cast se * se) >> (wlog2 + hlog2)); \
|
|
|
|
|
return (var >= 0) ? (uint32_t)var : 0; \
|
2016-07-23 05:07:03 +02:00
|
|
|
}
|
2016-05-07 03:38:04 +02:00
|
|
|
|
2016-07-23 05:07:03 +02:00
|
|
|
#define FNS(opt) \
|
|
|
|
|
FN(64, 64, 16, 6, 6, opt, (int64_t)); \
|
|
|
|
|
FN(64, 32, 16, 6, 5, opt, (int64_t)); \
|
|
|
|
|
FN(32, 64, 16, 5, 6, opt, (int64_t)); \
|
|
|
|
|
FN(32, 32, 16, 5, 5, opt, (int64_t)); \
|
|
|
|
|
FN(32, 16, 16, 5, 4, opt, (int64_t)); \
|
|
|
|
|
FN(16, 32, 16, 4, 5, opt, (int64_t)); \
|
|
|
|
|
FN(16, 16, 16, 4, 4, opt, (int64_t)); \
|
|
|
|
|
FN(16, 8, 16, 4, 3, opt, (int64_t)); \
|
|
|
|
|
FN(8, 16, 8, 3, 4, opt, (int64_t)); \
|
|
|
|
|
FN(8, 8, 8, 3, 3, opt, (int64_t)); \
|
|
|
|
|
FN(8, 4, 8, 3, 2, opt, (int64_t));
|
2016-05-07 03:38:04 +02:00
|
|
|
|
|
|
|
|
FNS(sse2);
|
2015-06-05 18:54:19 +02:00
|
|
|
|
|
|
|
|
#undef FNS
|
|
|
|
|
#undef FN
|
|
|
|
|
|
2015-12-14 17:35:29 +01:00
|
|
|
// The 2 unused parameters are place holders for PIC enabled build.
|
2016-07-23 05:07:03 +02:00
|
|
|
#define DECL(w, opt) \
|
|
|
|
|
int vpx_highbd_sub_pixel_avg_variance##w##xh_##opt( \
|
|
|
|
|
const uint16_t *src, ptrdiff_t src_stride, int x_offset, int y_offset, \
|
|
|
|
|
const uint16_t *dst, ptrdiff_t dst_stride, const uint16_t *sec, \
|
|
|
|
|
ptrdiff_t sec_stride, int height, unsigned int *sse, void *unused0, \
|
|
|
|
|
void *unused);
|
2015-06-05 18:54:19 +02:00
|
|
|
#define DECLS(opt1) \
|
2016-07-23 05:07:03 +02:00
|
|
|
DECL(16, opt1) \
|
|
|
|
|
DECL(8, opt1)
|
2015-06-05 18:54:19 +02:00
|
|
|
|
|
|
|
|
DECLS(sse2);
|
|
|
|
|
#undef DECL
|
|
|
|
|
#undef DECLS
|
|
|
|
|
|
2016-07-23 05:07:03 +02:00
|
|
|
#define FN(w, h, wf, wlog2, hlog2, opt, cast) \
|
|
|
|
|
uint32_t vpx_highbd_8_sub_pixel_avg_variance##w##x##h##_##opt( \
|
|
|
|
|
const uint8_t *src8, int src_stride, int x_offset, int y_offset, \
|
|
|
|
|
const uint8_t *dst8, int dst_stride, uint32_t *sse_ptr, \
|
|
|
|
|
const uint8_t *sec8) { \
|
|
|
|
|
uint32_t sse; \
|
|
|
|
|
uint16_t *src = CONVERT_TO_SHORTPTR(src8); \
|
|
|
|
|
uint16_t *dst = CONVERT_TO_SHORTPTR(dst8); \
|
|
|
|
|
uint16_t *sec = CONVERT_TO_SHORTPTR(sec8); \
|
|
|
|
|
int se = vpx_highbd_sub_pixel_avg_variance##wf##xh_##opt( \
|
|
|
|
|
src, src_stride, x_offset, y_offset, dst, dst_stride, sec, w, h, &sse, \
|
|
|
|
|
NULL, NULL); \
|
|
|
|
|
if (w > wf) { \
|
|
|
|
|
uint32_t sse2; \
|
|
|
|
|
int se2 = vpx_highbd_sub_pixel_avg_variance##wf##xh_##opt( \
|
|
|
|
|
src + 16, src_stride, x_offset, y_offset, dst + 16, dst_stride, \
|
|
|
|
|
sec + 16, w, h, &sse2, NULL, NULL); \
|
|
|
|
|
se += se2; \
|
|
|
|
|
sse += sse2; \
|
|
|
|
|
if (w > wf * 2) { \
|
|
|
|
|
se2 = vpx_highbd_sub_pixel_avg_variance##wf##xh_##opt( \
|
|
|
|
|
src + 32, src_stride, x_offset, y_offset, dst + 32, dst_stride, \
|
|
|
|
|
sec + 32, w, h, &sse2, NULL, NULL); \
|
|
|
|
|
se += se2; \
|
|
|
|
|
sse += sse2; \
|
|
|
|
|
se2 = vpx_highbd_sub_pixel_avg_variance##wf##xh_##opt( \
|
|
|
|
|
src + 48, src_stride, x_offset, y_offset, dst + 48, dst_stride, \
|
|
|
|
|
sec + 48, w, h, &sse2, NULL, NULL); \
|
|
|
|
|
se += se2; \
|
|
|
|
|
sse += sse2; \
|
|
|
|
|
} \
|
|
|
|
|
} \
|
|
|
|
|
*sse_ptr = sse; \
|
2017-04-05 05:37:17 +02:00
|
|
|
return sse - (uint32_t)((cast se * se) >> (wlog2 + hlog2)); \
|
2016-07-23 05:07:03 +02:00
|
|
|
} \
|
|
|
|
|
\
|
|
|
|
|
uint32_t vpx_highbd_10_sub_pixel_avg_variance##w##x##h##_##opt( \
|
|
|
|
|
const uint8_t *src8, int src_stride, int x_offset, int y_offset, \
|
|
|
|
|
const uint8_t *dst8, int dst_stride, uint32_t *sse_ptr, \
|
|
|
|
|
const uint8_t *sec8) { \
|
2017-04-05 05:37:17 +02:00
|
|
|
int64_t var; \
|
2016-07-23 05:07:03 +02:00
|
|
|
uint32_t sse; \
|
|
|
|
|
uint16_t *src = CONVERT_TO_SHORTPTR(src8); \
|
|
|
|
|
uint16_t *dst = CONVERT_TO_SHORTPTR(dst8); \
|
|
|
|
|
uint16_t *sec = CONVERT_TO_SHORTPTR(sec8); \
|
|
|
|
|
int se = vpx_highbd_sub_pixel_avg_variance##wf##xh_##opt( \
|
|
|
|
|
src, src_stride, x_offset, y_offset, dst, dst_stride, sec, w, h, &sse, \
|
|
|
|
|
NULL, NULL); \
|
|
|
|
|
if (w > wf) { \
|
|
|
|
|
uint32_t sse2; \
|
|
|
|
|
int se2 = vpx_highbd_sub_pixel_avg_variance##wf##xh_##opt( \
|
|
|
|
|
src + 16, src_stride, x_offset, y_offset, dst + 16, dst_stride, \
|
|
|
|
|
sec + 16, w, h, &sse2, NULL, NULL); \
|
|
|
|
|
se += se2; \
|
|
|
|
|
sse += sse2; \
|
|
|
|
|
if (w > wf * 2) { \
|
|
|
|
|
se2 = vpx_highbd_sub_pixel_avg_variance##wf##xh_##opt( \
|
|
|
|
|
src + 32, src_stride, x_offset, y_offset, dst + 32, dst_stride, \
|
|
|
|
|
sec + 32, w, h, &sse2, NULL, NULL); \
|
|
|
|
|
se += se2; \
|
|
|
|
|
sse += sse2; \
|
|
|
|
|
se2 = vpx_highbd_sub_pixel_avg_variance##wf##xh_##opt( \
|
|
|
|
|
src + 48, src_stride, x_offset, y_offset, dst + 48, dst_stride, \
|
|
|
|
|
sec + 48, w, h, &sse2, NULL, NULL); \
|
|
|
|
|
se += se2; \
|
|
|
|
|
sse += sse2; \
|
|
|
|
|
} \
|
|
|
|
|
} \
|
|
|
|
|
se = ROUND_POWER_OF_TWO(se, 2); \
|
|
|
|
|
sse = ROUND_POWER_OF_TWO(sse, 4); \
|
|
|
|
|
*sse_ptr = sse; \
|
2017-04-05 05:37:17 +02:00
|
|
|
var = (int64_t)(sse) - ((cast se * se) >> (wlog2 + hlog2)); \
|
|
|
|
|
return (var >= 0) ? (uint32_t)var : 0; \
|
2016-07-23 05:07:03 +02:00
|
|
|
} \
|
|
|
|
|
\
|
|
|
|
|
uint32_t vpx_highbd_12_sub_pixel_avg_variance##w##x##h##_##opt( \
|
|
|
|
|
const uint8_t *src8, int src_stride, int x_offset, int y_offset, \
|
|
|
|
|
const uint8_t *dst8, int dst_stride, uint32_t *sse_ptr, \
|
|
|
|
|
const uint8_t *sec8) { \
|
|
|
|
|
int start_row; \
|
2017-04-05 05:37:17 +02:00
|
|
|
int64_t var; \
|
2016-07-23 05:07:03 +02:00
|
|
|
uint32_t sse; \
|
|
|
|
|
int se = 0; \
|
|
|
|
|
uint64_t long_sse = 0; \
|
|
|
|
|
uint16_t *src = CONVERT_TO_SHORTPTR(src8); \
|
|
|
|
|
uint16_t *dst = CONVERT_TO_SHORTPTR(dst8); \
|
|
|
|
|
uint16_t *sec = CONVERT_TO_SHORTPTR(sec8); \
|
|
|
|
|
for (start_row = 0; start_row < h; start_row += 16) { \
|
|
|
|
|
uint32_t sse2; \
|
|
|
|
|
int height = h - start_row < 16 ? h - start_row : 16; \
|
|
|
|
|
int se2 = vpx_highbd_sub_pixel_avg_variance##wf##xh_##opt( \
|
|
|
|
|
src + (start_row * src_stride), src_stride, x_offset, y_offset, \
|
|
|
|
|
dst + (start_row * dst_stride), dst_stride, sec + (start_row * w), \
|
|
|
|
|
w, height, &sse2, NULL, NULL); \
|
|
|
|
|
se += se2; \
|
|
|
|
|
long_sse += sse2; \
|
|
|
|
|
if (w > wf) { \
|
|
|
|
|
se2 = vpx_highbd_sub_pixel_avg_variance##wf##xh_##opt( \
|
|
|
|
|
src + 16 + (start_row * src_stride), src_stride, x_offset, \
|
|
|
|
|
y_offset, dst + 16 + (start_row * dst_stride), dst_stride, \
|
|
|
|
|
sec + 16 + (start_row * w), w, height, &sse2, NULL, NULL); \
|
|
|
|
|
se += se2; \
|
|
|
|
|
long_sse += sse2; \
|
|
|
|
|
if (w > wf * 2) { \
|
|
|
|
|
se2 = vpx_highbd_sub_pixel_avg_variance##wf##xh_##opt( \
|
|
|
|
|
src + 32 + (start_row * src_stride), src_stride, x_offset, \
|
|
|
|
|
y_offset, dst + 32 + (start_row * dst_stride), dst_stride, \
|
|
|
|
|
sec + 32 + (start_row * w), w, height, &sse2, NULL, NULL); \
|
|
|
|
|
se += se2; \
|
|
|
|
|
long_sse += sse2; \
|
|
|
|
|
se2 = vpx_highbd_sub_pixel_avg_variance##wf##xh_##opt( \
|
|
|
|
|
src + 48 + (start_row * src_stride), src_stride, x_offset, \
|
|
|
|
|
y_offset, dst + 48 + (start_row * dst_stride), dst_stride, \
|
|
|
|
|
sec + 48 + (start_row * w), w, height, &sse2, NULL, NULL); \
|
|
|
|
|
se += se2; \
|
|
|
|
|
long_sse += sse2; \
|
|
|
|
|
} \
|
|
|
|
|
} \
|
|
|
|
|
} \
|
|
|
|
|
se = ROUND_POWER_OF_TWO(se, 4); \
|
|
|
|
|
sse = (uint32_t)ROUND_POWER_OF_TWO(long_sse, 8); \
|
|
|
|
|
*sse_ptr = sse; \
|
2017-04-05 05:37:17 +02:00
|
|
|
var = (int64_t)(sse) - ((cast se * se) >> (wlog2 + hlog2)); \
|
|
|
|
|
return (var >= 0) ? (uint32_t)var : 0; \
|
2016-07-23 05:07:03 +02:00
|
|
|
}
|
2015-06-05 18:54:19 +02:00
|
|
|
|
2016-07-23 05:07:03 +02:00
|
|
|
#define FNS(opt1) \
|
|
|
|
|
FN(64, 64, 16, 6, 6, opt1, (int64_t)); \
|
|
|
|
|
FN(64, 32, 16, 6, 5, opt1, (int64_t)); \
|
|
|
|
|
FN(32, 64, 16, 5, 6, opt1, (int64_t)); \
|
|
|
|
|
FN(32, 32, 16, 5, 5, opt1, (int64_t)); \
|
|
|
|
|
FN(32, 16, 16, 5, 4, opt1, (int64_t)); \
|
|
|
|
|
FN(16, 32, 16, 4, 5, opt1, (int64_t)); \
|
|
|
|
|
FN(16, 16, 16, 4, 4, opt1, (int64_t)); \
|
|
|
|
|
FN(16, 8, 16, 4, 3, opt1, (int64_t)); \
|
|
|
|
|
FN(8, 16, 8, 4, 3, opt1, (int64_t)); \
|
|
|
|
|
FN(8, 8, 8, 3, 3, opt1, (int64_t)); \
|
|
|
|
|
FN(8, 4, 8, 3, 2, opt1, (int64_t));
|
2015-06-05 18:54:19 +02:00
|
|
|
|
|
|
|
|
FNS(sse2);
|
|
|
|
|
|
|
|
|
|
#undef FNS
|
|
|
|
|
#undef FN
|
