Skip to content

L
libyuv
Commit 6a192487 authored by fbarchard@google.com's avatar fbarchard@google.com
Browse files
Options

Switch SSSE3 row wrappers from variable sized malloc to fixed size array with… 

Switch SSSE3 row wrappers from variable sized malloc to fixed size array with loop to process a portion of the row at a time.  This helps performance in the case where the image has been coalesced into a single large row and the allocator, although only called once, is slow to clear the pages.  Also the smaller temporary buffer fits cache, further improving performance.
BUG=403
TESTED=YUY2ToARGB unittest
R=harryjin@google.com

Review URL: https://webrtc-codereview.appspot.com/40849004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1286 16f28f9a-4ce2-e073-06de-1de4eb20be90
parent 194f740d
Hide whitespace changes
Inline Side-by-side
Showing with 102 additions and 66 deletions
README.chromium
View file @ 6a192487
Name: libyuv Name: libyuv
URL: http://code.google.com/p/libyuv/ URL: http://code.google.com/p/libyuv/
Version: 1285 Version: 1286
License: BSD License: BSD
License File: LICENSE License File: LICENSE
......
include/libyuv/version.h
View file @ 6a192487
...@@ -11,6 +11,6 @@ ...@@ -11,6 +11,6 @@
#ifndef INCLUDE_LIBYUV_VERSION_H_ // NOLINT #ifndef INCLUDE_LIBYUV_VERSION_H_ // NOLINT
#define INCLUDE_LIBYUV_VERSION_H_ #define INCLUDE_LIBYUV_VERSION_H_
#define LIBYUV_VERSION 1285 #define LIBYUV_VERSION 1286
#endif // INCLUDE_LIBYUV_VERSION_H_ NOLINT #endif // INCLUDE_LIBYUV_VERSION_H_ NOLINT
source/row_common.cc
View file @ 6a192487
...@@ -2122,6 +2122,9 @@ void I422ToUYVYRow_C(const uint8* src_y, ...@@ -2122,6 +2122,9 @@ void I422ToUYVYRow_C(const uint8* src_y,
} }
} }
// Maximum temporary width for wrappers to process at a time, in pixels.
#define MAXTWIDTH 4096
#if !defined(LIBYUV_DISABLE_X86) && defined(HAS_I422TOARGBROW_SSSE3) #if !defined(LIBYUV_DISABLE_X86) && defined(HAS_I422TOARGBROW_SSSE3)
// row_win.cc has asm version, but GCC uses 2 step wrapper. // row_win.cc has asm version, but GCC uses 2 step wrapper.
#if !defined(_MSC_VER) && (defined(__x86_64__) || defined(__i386__)) #if !defined(_MSC_VER) && (defined(__x86_64__) || defined(__i386__))
...@@ -2130,11 +2133,17 @@ void I422ToRGB565Row_SSSE3(const uint8* src_y, ...@@ -2130,11 +2133,17 @@ void I422ToRGB565Row_SSSE3(const uint8* src_y,
const uint8* src_v, const uint8* src_v,
uint8* rgb_buf, uint8* rgb_buf,
int width) { int width) {
// Allocate a row of ARGB. SIMD_ALIGNED(uint8 row[MAXTWIDTH * 4]);
align_buffer_64(row, width * 4); while (width > 0) {
I422ToARGBRow_SSSE3(src_y, src_u, src_v, row, width); int twidth = width > MAXTWIDTH ? MAXTWIDTH : width;
ARGBToRGB565Row_SSE2(row, rgb_buf, width); I422ToARGBRow_SSSE3(src_y, src_u, src_v, row, twidth);
free_aligned_buffer_64(row); ARGBToRGB565Row_SSE2(row, rgb_buf, twidth);
src_y += twidth;
src_u += twidth / 2;
src_v += twidth / 2;
dst_argb += twidth * 2;
width -= twidth;
}
} }
#endif // !defined(_MSC_VER) && (defined(__x86_64__) || defined(__i386__)) #endif // !defined(_MSC_VER) && (defined(__x86_64__) || defined(__i386__))
...@@ -2144,11 +2153,18 @@ void I422ToARGB1555Row_SSSE3(const uint8* src_y, ...@@ -2144,11 +2153,18 @@ void I422ToARGB1555Row_SSSE3(const uint8* src_y,
const uint8* src_v, const uint8* src_v,
uint8* rgb_buf, uint8* rgb_buf,
int width) { int width) {
// Allocate a row of ARGB. // Row buffer for intermediate ARGB pixels.
align_buffer_64(row, width * 4); SIMD_ALIGNED(uint8 row[MAXTWIDTH * 4]);
I422ToARGBRow_SSSE3(src_y, src_u, src_v, row, width); while (width > 0) {
ARGBToARGB1555Row_SSE2(row, rgb_buf, width); int twidth = width > MAXTWIDTH ? MAXTWIDTH : width;
free_aligned_buffer_64(row); I422ToARGBRow_SSSE3(src_y, src_u, src_v, row, twidth);
ARGBToARGB1555Row_SSE2(row, rgb_buf, twidth);
src_y += twidth;
src_u += twidth / 2;
src_v += twidth / 2;
rgb_buf += twidth * 2;
width -= twidth;
}
} }
void I422ToARGB4444Row_SSSE3(const uint8* src_y, void I422ToARGB4444Row_SSSE3(const uint8* src_y,
...@@ -2156,61 +2172,81 @@ void I422ToARGB4444Row_SSSE3(const uint8* src_y, ...@@ -2156,61 +2172,81 @@ void I422ToARGB4444Row_SSSE3(const uint8* src_y,
const uint8* src_v, const uint8* src_v,
uint8* rgb_buf, uint8* rgb_buf,
int width) { int width) {
// Allocate a row of ARGB. // Row buffer for intermediate ARGB pixels.
align_buffer_64(row, width * 4); SIMD_ALIGNED(uint8 row[MAXTWIDTH * 4]);
I422ToARGBRow_SSSE3(src_y, src_u, src_v, row, width); while (width > 0) {
ARGBToARGB4444Row_SSE2(row, rgb_buf, width); int twidth = width > MAXTWIDTH ? MAXTWIDTH : width;
free_aligned_buffer_64(row); I422ToARGBRow_SSSE3(src_y, src_u, src_v, row, twidth);
} ARGBToARGB4444Row_SSE2(row, rgb_buf, twidth);
src_y += twidth;
void NV12ToRGB565Row_SSSE3(const uint8* src_y, src_u += twidth / 2;
const uint8* src_uv, src_v += twidth / 2;
uint8* dst_rgb565, rgb_buf += twidth * 2;
int width) { width -= twidth;
// Allocate a row of ARGB. }
align_buffer_64(row, width * 4); }
NV12ToARGBRow_SSSE3(src_y, src_uv, row, width);
ARGBToRGB565Row_SSE2(row, dst_rgb565, width); void NV12ToRGB565Row_SSSE3(const uint8* src_y, const uint8* src_uv,
free_aligned_buffer_64(row); uint8* dst_rgb565, int width) {
} // Row buffer for intermediate ARGB pixels.
SIMD_ALIGNED(uint8 row[MAXTWIDTH * 4]);
void NV21ToRGB565Row_SSSE3(const uint8* src_y, while (width > 0) {
const uint8* src_vu, int twidth = width > MAXTWIDTH ? MAXTWIDTH : width;
uint8* dst_rgb565, NV12ToARGBRow_SSSE3(src_y, src_uv, row, twidth);
int width) { ARGBToRGB565Row_SSE2(row, dst_rgb565, twidth);
// Allocate a row of ARGB. src_y += twidth;
align_buffer_64(row, width * 4); src_uv += twidth;
NV21ToARGBRow_SSSE3(src_y, src_vu, row, width); dst_rgb565 += twidth * 2;
ARGBToRGB565Row_SSE2(row, dst_rgb565, width); width -= twidth;
free_aligned_buffer_64(row); }
} }
void YUY2ToARGBRow_SSSE3(const uint8* src_yuy2, void NV21ToRGB565Row_SSSE3(const uint8* src_y, const uint8* src_vu,
uint8* dst_argb, uint8* dst_rgb565, int width) {
int width) { // Row buffer for intermediate ARGB pixels.
// Allocate a rows of yuv. SIMD_ALIGNED(uint8 row[MAXTWIDTH * 4]);
align_buffer_64(row_y, ((width + 63) & ~63) * 2); while (width > 0) {
uint8* row_u = row_y + ((width + 63) & ~63); int twidth = width > MAXTWIDTH ? MAXTWIDTH : width;
uint8* row_v = row_u + ((width + 63) & ~63) / 2; NV21ToARGBRow_SSSE3(src_y, src_vu, row, twidth);
YUY2ToUV422Row_SSE2(src_yuy2, row_u, row_v, width); ARGBToRGB565Row_SSE2(row, dst_rgb565, twidth);
YUY2ToYRow_SSE2(src_yuy2, row_y, width); src_y += twidth;
I422ToARGBRow_SSSE3(row_y, row_u, row_v, dst_argb, width); src_vu += twidth;
free_aligned_buffer_64(row_y); dst_rgb565 += twidth * 2;
} width -= twidth;
}
void UYVYToARGBRow_SSSE3(const uint8* src_uyvy, }
uint8* dst_argb,
int width) { void YUY2ToARGBRow_SSSE3(const uint8* src_yuy2, uint8* dst_argb, int width) {
// Allocate a rows of yuv. // Row buffers for intermediate YUV pixels.
align_buffer_64(row_y, ((width + 63) & ~63) * 2); SIMD_ALIGNED(uint8 row_y[MAXTWIDTH]);
uint8* row_u = row_y + ((width + 63) & ~63); SIMD_ALIGNED(uint8 row_u[MAXTWIDTH / 2]);
uint8* row_v = row_u + ((width + 63) & ~63) / 2; SIMD_ALIGNED(uint8 row_v[MAXTWIDTH / 2]);
UYVYToUV422Row_SSE2(src_uyvy, row_u, row_v, width); while (width > 0) {
UYVYToYRow_SSE2(src_uyvy, row_y, width); int twidth = width > MAXTWIDTH ? MAXTWIDTH : width;
I422ToARGBRow_SSSE3(row_y, row_u, row_v, dst_argb, width); YUY2ToUV422Row_SSE2(src_yuy2, row_u, row_v, twidth);
free_aligned_buffer_64(row_y); YUY2ToYRow_SSE2(src_yuy2, row_y, twidth);
I422ToARGBRow_SSSE3(row_y, row_u, row_v, dst_argb, twidth);
src_yuy2 += twidth * 2;
dst_argb += twidth * 4;
width -= twidth;
}
}
void UYVYToARGBRow_SSSE3(const uint8* src_uyvy, uint8* dst_argb, int width) {
// Row buffers for intermediate YUV pixels.
SIMD_ALIGNED(uint8 row_y[MAXTWIDTH]);
SIMD_ALIGNED(uint8 row_u[MAXTWIDTH / 2]);
SIMD_ALIGNED(uint8 row_v[MAXTWIDTH / 2]);
while (width > 0) {
int twidth = width > MAXTWIDTH ? MAXTWIDTH : width;
UYVYToUV422Row_SSE2(src_uyvy, row_u, row_v, twidth);
UYVYToYRow_SSE2(src_uyvy, row_y, twidth);
I422ToARGBRow_SSSE3(row_y, row_u, row_v, dst_argb, twidth);
src_uyvy += twidth * 2;
dst_argb += twidth * 4;
width -= twidth;
}
} }
#endif // defined(_M_IX86) || defined(__x86_64__) || defined(__i386__) #endif // defined(_M_IX86) || defined(__x86_64__) || defined(__i386__)
#endif // !defined(LIBYUV_DISABLE_X86) #endif // !defined(LIBYUV_DISABLE_X86)
......
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