Skip to content
GitLab
Commit 5023627c authored by Dmitry Kovalev's avatar Dmitry Kovalev Committed by Gerrit Code Review
Browse files

Merge "Cleaning up vp9_variance_sse2.c."

parents f9501295 a789bfec
v1.14.0-linphone 1.4.X feature/update_to_v1.9.0-linphone feature/uwp_nuget frame_parallel highbitdepth indianrunnerduck javanwhistlingduck khakicampbell linphone linphone-android linphone-old longtailedduck m49-2623 m52-2743 m54-2840 m56-2924 m66-3359 m68-3440 mandarinduck nextgen nextgenv2 playground sandbox/Jingning/experimental sandbox/Jingning/transcode sandbox/Jingning/vpx sandbox/aconverse@google.com/ansbench sandbox/hkuang/frame_parallel sandbox/hkuang@google.com/decode sandbox/jimbankoski@google.com/proposed-aom sandbox/jingning@google.com/decoder_test_suite sandbox/jingning@google.com/experimental sandbox/jzern@google.com/test sandbox/wangch@google.com/vp9 sandbox/yaowu@google.com/mergeaom v1.12.0-linphone v1.6.1_linphone v1.7.0-linphone v1.9.0-linphone v1.9.0 v1.9.0-rc1 v1.8.2 v1.8.1 v1.8.0 v1.7.0 v1.6.1 v1.6.0 v1.5.0 v1.4.0
No related merge requests found
Hide whitespace changes
Inline Side-by-side
Showing
with 114 additions and 215 deletions
vp9/encoder/x86/vp9_variance_sse2.c
+ 114
− 215
View file @ 5023627c
...@@ -13,50 +13,27 @@ ...@@ -13,50 +13,27 @@
#include "vp9/encoder/vp9_variance.h" #include "vp9/encoder/vp9_variance.h"
#include "vpx_ports/mem.h" #include "vpx_ports/mem.h"
extern unsigned int vp9_get4x4var_mmx typedef unsigned int (*variance_fn_t) (const unsigned char *src, int src_stride,
( const unsigned char *ref, int ref_stride,
const unsigned char *src_ptr, unsigned int *sse, int *sum);
int source_stride,
const unsigned char *ref_ptr, unsigned int vp9_get4x4var_mmx(const unsigned char *src, int src_stride,
int recon_stride, const unsigned char *ref, int ref_stride,
unsigned int *SSE, unsigned int *sse, int *sum);
int *Sum
);
unsigned int vp9_get8x8var_sse2(const unsigned char *src, int src_stride,
unsigned int vp9_get16x16var_sse2 const unsigned char *ref, int ref_stride,
( unsigned int *sse, int *sum);
const unsigned char *src_ptr,
int source_stride, unsigned int vp9_get16x16var_sse2(const unsigned char *src, int src_stride,
const unsigned char *ref_ptr, const unsigned char *ref, int ref_stride,
int recon_stride, unsigned int *sse, int *sum);
unsigned int *SSE,
int *Sum static void variance_sse2(const unsigned char *src, int src_stride,
); const unsigned char *ref, int ref_stride,
unsigned int vp9_get8x8var_sse2 int w, int h, unsigned int *sse, int *sum,
( variance_fn_t var_fn, int block_size) {
const unsigned char *src_ptr,
int source_stride,
const unsigned char *ref_ptr,
int recon_stride,
unsigned int *SSE,
int *Sum
);
typedef unsigned int (*get_var_sse2) (
const unsigned char *src_ptr,
int source_stride,
const unsigned char *ref_ptr,
int recon_stride,
unsigned int *SSE,
int *Sum
);
static void variance_sse2(const unsigned char *src_ptr, int source_stride,
const unsigned char *ref_ptr, int recon_stride,
int w, int h, unsigned int *sse, int *sum,
get_var_sse2 var_fn, int block_size) {
unsigned int sse0;
int sum0;
int i, j; int i, j;
*sse = 0; *sse = 0;
...@@ -64,217 +41,139 @@ static void variance_sse2(const unsigned char *src_ptr, int source_stride, ...@@ -64,217 +41,139 @@ static void variance_sse2(const unsigned char *src_ptr, int source_stride,
for (i = 0; i < h; i += block_size) { for (i = 0; i < h; i += block_size) {
for (j = 0; j < w; j += block_size) { for (j = 0; j < w; j += block_size) {
var_fn(src_ptr + source_stride * i + j, source_stride, unsigned int sse0;
ref_ptr + recon_stride * i + j, recon_stride, &sse0, &sum0); int sum0;
var_fn(src + src_stride * i + j, src_stride,
ref + ref_stride * i + j, ref_stride, &sse0, &sum0);
*sse += sse0; *sse += sse0;
*sum += sum0; *sum += sum0;
} }
} }
} }
unsigned int vp9_variance4x4_sse2( unsigned int vp9_variance4x4_sse2(const unsigned char *src, int src_stride,
const unsigned char *src_ptr, const unsigned char *ref, int ref_stride,
int source_stride, unsigned int *sse) {
const unsigned char *ref_ptr, int sum;
int recon_stride, variance_sse2(src, src_stride, ref, ref_stride, 4, 4,
unsigned int *sse) { sse, &sum, vp9_get4x4var_mmx, 4);
unsigned int var; return *sse - (((unsigned int)sum * sum) >> 4);
int avg;
variance_sse2(src_ptr, source_stride, ref_ptr, recon_stride, 4, 4,
&var, &avg, vp9_get4x4var_mmx, 4);
*sse = var;
return (var - (((unsigned int)avg * avg) >> 4));
} }
unsigned int vp9_variance8x4_sse2(const uint8_t *src_ptr, unsigned int vp9_variance8x4_sse2(const uint8_t *src, int src_stride,
int source_stride, const uint8_t *ref, int ref_stride,
const uint8_t *ref_ptr,
int recon_stride,
unsigned int *sse) { unsigned int *sse) {
unsigned int var; int sum;
int avg; variance_sse2(src, src_stride, ref, ref_stride, 8, 4,
sse, &sum, vp9_get4x4var_mmx, 4);
variance_sse2(src_ptr, source_stride, ref_ptr, recon_stride, 8, 4, return *sse - (((unsigned int)sum * sum) >> 5);
&var, &avg, vp9_get4x4var_mmx, 4);
*sse = var;
return (var - (((unsigned int)avg * avg) >> 5));
} }
unsigned int vp9_variance4x8_sse2(const uint8_t *src_ptr, unsigned int vp9_variance4x8_sse2(const uint8_t *src, int src_stride,
int source_stride, const uint8_t *ref, int ref_stride,
const uint8_t *ref_ptr,
int recon_stride,
unsigned int *sse) { unsigned int *sse) {
unsigned int var; int sum;
int avg; variance_sse2(src, src_stride, ref, ref_stride, 4, 8,
sse, &sum, vp9_get4x4var_mmx, 4);
variance_sse2(src_ptr, source_stride, ref_ptr, recon_stride, 4, 8, return *sse - (((unsigned int)sum * sum) >> 5);
&var, &avg, vp9_get4x4var_mmx, 4);
*sse = var;
return (var - (((unsigned int)avg * avg) >> 5));
} }
unsigned int vp9_variance8x8_sse2 unsigned int vp9_variance8x8_sse2(const unsigned char *src, int src_stride,
( const unsigned char *ref, int ref_stride,
const unsigned char *src_ptr, unsigned int *sse) {
int source_stride, int sum;
const unsigned char *ref_ptr, variance_sse2(src, src_stride, ref, ref_stride, 8, 8,
int recon_stride, sse, &sum, vp9_get8x8var_sse2, 8);
unsigned int *sse) { return *sse - (((unsigned int)sum * sum) >> 6);
unsigned int var;
int avg;
variance_sse2(src_ptr, source_stride, ref_ptr, recon_stride, 8, 8,
&var, &avg, vp9_get8x8var_sse2, 8);
*sse = var;
return (var - (((unsigned int)avg * avg) >> 6));
} }
unsigned int vp9_variance16x8_sse2 unsigned int vp9_variance16x8_sse2(const unsigned char *src, int src_stride,
( const unsigned char *ref, int ref_stride,
const unsigned char *src_ptr, unsigned int *sse) {
int source_stride, int sum;
const unsigned char *ref_ptr, variance_sse2(src, src_stride, ref, ref_stride, 16, 8,
int recon_stride, sse, &sum, vp9_get8x8var_sse2, 8);
unsigned int *sse) { return *sse - (((unsigned int)sum * sum) >> 7);
unsigned int var;
int avg;
variance_sse2(src_ptr, source_stride, ref_ptr, recon_stride, 16, 8,
&var, &avg, vp9_get8x8var_sse2, 8);
*sse = var;
return (var - (((unsigned int)avg * avg) >> 7));
} }
unsigned int vp9_variance8x16_sse2 unsigned int vp9_variance8x16_sse2(const unsigned char *src, int src_stride,
( const unsigned char *ref, int ref_stride,
const unsigned char *src_ptr, unsigned int *sse) {
int source_stride, int sum;
const unsigned char *ref_ptr, variance_sse2(src, src_stride, ref, ref_stride, 8, 16,
int recon_stride, sse, &sum, vp9_get8x8var_sse2, 8);
unsigned int *sse) { return *sse - (((unsigned int)sum * sum) >> 7);
unsigned int var;
int avg;
variance_sse2(src_ptr, source_stride, ref_ptr, recon_stride, 8, 16,
&var, &avg, vp9_get8x8var_sse2, 8);
*sse = var;
return (var - (((unsigned int)avg * avg) >> 7));
} }
unsigned int vp9_variance16x16_sse2 unsigned int vp9_variance16x16_sse2(const unsigned char *src, int src_stride,
( const unsigned char *ref, int ref_stride,
const unsigned char *src_ptr, unsigned int *sse) {
int source_stride, int sum;
const unsigned char *ref_ptr, variance_sse2(src, src_stride, ref, ref_stride, 16, 16,
int recon_stride, sse, &sum, vp9_get16x16var_sse2, 16);
unsigned int *sse) { return *sse - (((unsigned int)sum * sum) >> 8);
unsigned int var;
int avg;
variance_sse2(src_ptr, source_stride, ref_ptr, recon_stride, 16, 16,
&var, &avg, vp9_get16x16var_sse2, 16);
*sse = var;
return (var - (((unsigned int)avg * avg) >> 8));
} }
unsigned int vp9_mse16x16_sse2( unsigned int vp9_mse16x16_sse2(const unsigned char *src, int src_stride,
const unsigned char *src_ptr, const unsigned char *ref, int ref_stride,
int source_stride, unsigned int *sse) {
const unsigned char *ref_ptr, int sum;
int recon_stride, vp9_get16x16var_sse2(src, src_stride, ref, ref_stride, sse, &sum);
unsigned int *sse) { return *sse;
unsigned int sse0;
int sum0;
vp9_get16x16var_sse2(src_ptr, source_stride, ref_ptr, recon_stride, &sse0,
&sum0);
*sse = sse0;
return sse0;
} }
unsigned int vp9_variance32x32_sse2(const uint8_t *src_ptr, unsigned int vp9_variance32x32_sse2(const uint8_t *src, int src_stride,
int source_stride, const uint8_t *ref, int ref_stride,
const uint8_t *ref_ptr,
int recon_stride,
unsigned int *sse) { unsigned int *sse) {
unsigned int var; int sum;
int avg; variance_sse2(src, src_stride, ref, ref_stride, 32, 32,
sse, &sum, vp9_get16x16var_sse2, 16);
variance_sse2(src_ptr, source_stride, ref_ptr, recon_stride, 32, 32, return *sse - (((int64_t)sum * sum) >> 10);
&var, &avg, vp9_get16x16var_sse2, 16);
*sse = var;
return (var - (((int64_t)avg * avg) >> 10));
} }
unsigned int vp9_variance32x16_sse2(const uint8_t *src_ptr, unsigned int vp9_variance32x16_sse2(const uint8_t *src, int src_stride,
int source_stride, const uint8_t *ref, int ref_stride,
const uint8_t *ref_ptr,
int recon_stride,
unsigned int *sse) { unsigned int *sse) {
unsigned int var; int sum;
int avg; variance_sse2(src, src_stride, ref, ref_stride, 32, 16,
sse, &sum, vp9_get16x16var_sse2, 16);
variance_sse2(src_ptr, source_stride, ref_ptr, recon_stride, 32, 16, return *sse - (((int64_t)sum * sum) >> 9);
&var, &avg, vp9_get16x16var_sse2, 16);
*sse = var;
return (var - (((int64_t)avg * avg) >> 9));
} }
unsigned int vp9_variance16x32_sse2(const uint8_t *src_ptr, unsigned int vp9_variance16x32_sse2(const uint8_t *src, int src_stride,
int source_stride, const uint8_t *ref, int ref_stride,
const uint8_t *ref_ptr,
int recon_stride,
unsigned int *sse) { unsigned int *sse) {
unsigned int var; int sum;
int avg; variance_sse2(src, src_stride, ref, ref_stride, 16, 32,
sse, &sum, vp9_get16x16var_sse2, 16);
variance_sse2(src_ptr, source_stride, ref_ptr, recon_stride, 16, 32, return *sse - (((int64_t)sum * sum) >> 9);
&var, &avg, vp9_get16x16var_sse2, 16);
*sse = var;
return (var - (((int64_t)avg * avg) >> 9));
} }
unsigned int vp9_variance64x64_sse2(const uint8_t *src_ptr, unsigned int vp9_variance64x64_sse2(const uint8_t *src, int src_stride,
int source_stride, const uint8_t *ref, int ref_stride,
const uint8_t *ref_ptr,
int recon_stride,
unsigned int *sse) { unsigned int *sse) {
unsigned int var; int sum;
int avg; variance_sse2(src, src_stride, ref, ref_stride, 64, 64,
sse, &sum, vp9_get16x16var_sse2, 16);
variance_sse2(src_ptr, source_stride, ref_ptr, recon_stride, 64, 64, return *sse - (((int64_t)sum * sum) >> 12);
&var, &avg, vp9_get16x16var_sse2, 16);
*sse = var;
return (var - (((int64_t)avg * avg) >> 12));
} }
unsigned int vp9_variance64x32_sse2(const uint8_t *src_ptr, unsigned int vp9_variance64x32_sse2(const uint8_t *src, int src_stride,
int source_stride, const uint8_t *ref, int ref_stride,
const uint8_t *ref_ptr,
int recon_stride,
unsigned int *sse) { unsigned int *sse) {
unsigned int var; int sum;
int avg; variance_sse2(src, src_stride, ref, ref_stride, 64, 32,
sse, &sum, vp9_get16x16var_sse2, 16);
variance_sse2(src_ptr, source_stride, ref_ptr, recon_stride, 64, 32, return *sse - (((int64_t)sum * sum) >> 11);
&var, &avg, vp9_get16x16var_sse2, 16);
*sse = var;
return (var - (((int64_t)avg * avg) >> 11));
} }
unsigned int vp9_variance32x64_sse2(const uint8_t *src_ptr, unsigned int vp9_variance32x64_sse2(const uint8_t *src, int src_stride,
int source_stride, const uint8_t *ref, int ref_stride,
const uint8_t *ref_ptr,
int recon_stride,
unsigned int *sse) { unsigned int *sse) {
unsigned int var; int sum;
int avg; variance_sse2(src, src_stride, ref, ref_stride, 32, 64,
sse, &sum, vp9_get16x16var_sse2, 16);
variance_sse2(src_ptr, source_stride, ref_ptr, recon_stride, 32, 64, return *sse - (((int64_t)sum * sum) >> 11);
&var, &avg, vp9_get16x16var_sse2, 16);
*sse = var;
return (var - (((int64_t)avg * avg) >> 11));
} }
#define DECL(w, opt) \ #define DECL(w, opt) \
......
Supports Markdown
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