Skip to content

L
libyuv
Commit a1ab1945 authored by fbarchard@google.com's avatar fbarchard@google.com
Browse files
Options

Color Table x86 reoptimized and ported to gcc. 

BUG=266
TESTED=color table unittests
R=changjun.yang@intel.com

Review URL: https://webrtc-codereview.appspot.com/2216004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@791 16f28f9a-4ce2-e073-06de-1de4eb20be90
parent 2bbb64df
Show whitespace changes
Inline Side-by-side
Showing with 98 additions and 65 deletions
README.chromium
View file @ a1ab1945
Name: libyuv
URL: http://code.google.com/p/libyuv/
Version: 790
Version: 791
License: BSD
License File: LICENSE
......
include/libyuv/row.h
View file @ a1ab1945
......@@ -136,15 +136,16 @@ extern "C" {
#define HAS_YTOARGBROW_SSE2
#define HAS_YUY2TOARGBROW_SSSE3
#define HAS_YUY2TOYROW_SSE2
// Effects:
#define HAS_ARGBCOLORTABLEROW_X86
#define HAS_RGBCOLORTABLEROW_X86
#endif
// The following are Windows only:
// TODO(fbarchard): Port to gcc.
#if !defined(LIBYUV_DISABLE_X86) && defined(_M_IX86) && defined(_MSC_VER)
// Effects:
// Disabled due to C versions being faster on Sandy Bridge with VC2012.
// #define HAS_ARGBCOLORTABLEROW_X86
// #define HAS_RGBCOLORTABLEROW_X86
// SSSE3 version incomplete:
// #define HAS_ARGBLUMACOLORTABLEROW_SSSE3
......
include/libyuv/version.h
View file @ a1ab1945
......@@ -11,6 +11,6 @@
#ifndef INCLUDE_LIBYUV_VERSION_H_ // NOLINT
#define INCLUDE_LIBYUV_VERSION_H_
#define LIBYUV_VERSION 790
#define LIBYUV_VERSION 791
#endif // INCLUDE_LIBYUV_VERSION_H_ NOLINT
source/row_posix.cc
View file @ a1ab1945
......@@ -5875,6 +5875,66 @@ void ARGBPolynomialRow_SSE2(const uint8* src_argb,
}
#endif // HAS_ARGBPOLYNOMIALROW_SSE2
#ifdef HAS_ARGBCOLORTABLEROW_X86
// Tranform ARGB pixels with color table.
void ARGBColorTableRow_X86(uint8* dst_argb, const uint8* table_argb,
int width) {
uintptr_t pixel_temp = 0u;
asm volatile (
// 1 pixel loop.
".p2align 4 \n"
"1: \n"
"movzb (%0),%1 \n"
"lea 0x4(%0),%0 \n"
"movzb (%3,%1,4),%1 \n"
"mov %b1,-0x4(%0) \n"
"movzb -0x3(%0),%1 \n"
"movzb 0x1(%3,%1,4),%1 \n"
"mov %b1,-0x3(%0) \n"
"movzb -0x2(%0),%1 \n"
"movzb 0x2(%3,%1,4),%1 \n"
"mov %b1,-0x2(%0) \n"
"movzb -0x1(%0),%1 \n"
"movzb 0x3(%3,%1,4),%1 \n"
"mov %b1,-0x1(%0) \n"
"dec %2 \n"
"jg 1b \n"
: "+r"(dst_argb), // %0
"+d"(pixel_temp), // %1
"+r"(width) // %2
: "r"(table_argb) // %3
: "memory", "cc");
}
#endif // HAS_ARGBCOLORTABLEROW_X86
#ifdef HAS_RGBCOLORTABLEROW_X86
// Tranform RGB pixels with color table.
void RGBColorTableRow_X86(uint8* dst_argb, const uint8* table_argb, int width) {
uintptr_t pixel_temp = 0u;
asm volatile (
// 1 pixel loop.
".p2align 4 \n"
"1: \n"
"movzb (%0),%1 \n"
"lea 0x4(%0),%0 \n"
"movzb (%3,%1,4),%1 \n"
"mov %b1,-0x4(%0) \n"
"movzb -0x3(%0),%1 \n"
"movzb 0x1(%3,%1,4),%1 \n"
"mov %b1,-0x3(%0) \n"
"movzb -0x2(%0),%1 \n"
"movzb 0x2(%3,%1,4),%1 \n"
"mov %b1,-0x2(%0) \n"
"dec %2 \n"
"jg 1b \n"
: "+r"(dst_argb), // %0
"+d"(pixel_temp), // %1
"+r"(width) // %2
: "r"(table_argb) // %3
: "memory", "cc");
}
#endif // HAS_RGBCOLORTABLEROW_X86
#endif // defined(__x86_64__) || defined(__i386__)
#ifdef __cplusplus
......
source/row_win.cc
View file @ a1ab1945
......@@ -5097,43 +5097,28 @@ __declspec(naked) __declspec(align(16))
void ARGBColorTableRow_X86(uint8* dst_argb, const uint8* table_argb,
int width) {
__asm {
push ebx
push esi
push edi
push ebp
mov eax, [esp + 16 + 4] /* dst_argb */
mov edi, [esp + 16 + 8] /* table_argb */
mov ecx, [esp + 16 + 12] /* width */
xor ebx, ebx
xor edx, edx
mov eax, [esp + 4 + 4] /* dst_argb */
mov esi, [esp + 4 + 8] /* table_argb */
mov ecx, [esp + 4 + 12] /* width */
align 16
convertloop:
mov ebp, dword ptr [eax] // BGRA
mov esi, ebp
and ebp, 255
shr esi, 8
and esi, 255
mov bl, [edi + ebp * 4 + 0] // B
mov dl, [edi + esi * 4 + 1] // G
mov ebp, dword ptr [eax] // BGRA
mov esi, ebp
shr ebp, 16
shr esi, 24
and ebp, 255
mov [eax], bl
mov [eax + 1], dl
mov bl, [edi + ebp * 4 + 2] // R
mov dl, [edi + esi * 4 + 3] // A
mov [eax + 2], bl
mov [eax + 3], dl
movzx edx, byte ptr [eax]
lea eax, [eax + 4]
sub ecx, 1
movzx edx, byte ptr [esi + edx * 4]
mov byte ptr [eax - 4], dl
movzx edx, byte ptr [eax - 4 + 1]
movzx edx, byte ptr [esi + edx * 4 + 1]
mov byte ptr [eax - 4 + 1], dl
movzx edx, byte ptr [eax - 4 + 2]
movzx edx, byte ptr [esi + edx * 4 + 2]
mov byte ptr [eax - 4 + 2], dl
movzx edx, byte ptr [eax - 4 + 3]
movzx edx, byte ptr [esi + edx * 4 + 3]
mov byte ptr [eax - 4 + 3], dl
dec ecx
jg convertloop
pop ebp
pop edi
pop esi
pop ebx
ret
}
}
......@@ -5144,39 +5129,26 @@ void ARGBColorTableRow_X86(uint8* dst_argb, const uint8* table_argb,
__declspec(naked) __declspec(align(16))
void RGBColorTableRow_X86(uint8* dst_argb, const uint8* table_argb, int width) {
__asm {
push ebx
push esi
push edi
push ebp
mov eax, [esp + 16 + 4] /* dst_argb */
mov edi, [esp + 16 + 8] /* table_argb */
mov ecx, [esp + 16 + 12] /* width */
xor ebx, ebx
xor edx, edx
mov eax, [esp + 4 + 4] /* dst_argb */
mov esi, [esp + 4 + 8] /* table_argb */
mov ecx, [esp + 4 + 12] /* width */
align 16
convertloop:
mov ebp, dword ptr [eax] // BGRA
mov esi, ebp
and ebp, 255
shr esi, 8
and esi, 255
mov bl, [edi + ebp * 4 + 0] // B
mov dl, [edi + esi * 4 + 1] // G
mov ebp, dword ptr [eax] // BGRA
shr ebp, 16
and ebp, 255
mov [eax], bl
mov [eax + 1], dl
mov bl, [edi + ebp * 4 + 2] // R
mov [eax + 2], bl
movzx edx, byte ptr [eax]
lea eax, [eax + 4]
sub ecx, 1
movzx edx, byte ptr [esi + edx * 4]
mov byte ptr [eax - 4], dl
movzx edx, byte ptr [eax - 4 + 1]
movzx edx, byte ptr [esi + edx * 4 + 1]
mov byte ptr [eax - 4 + 1], dl
movzx edx, byte ptr [eax - 4 + 2]
movzx edx, byte ptr [esi + edx * 4 + 2]
mov byte ptr [eax - 4 + 2], dl
dec ecx
jg convertloop
pop ebp
pop edi
pop esi
pop ebx
ret
}
}
......
unit_test/planar_test.cc
View file @ a1ab1945
......@@ -543,7 +543,7 @@ TEST_F(libyuvTest, TestARGBColorTable) {
memset(orig_pixels, 0, sizeof(orig_pixels));
// Matrix for Sepia.
static const uint8 kARGBTable[1280 * 4] = {
static const uint8 kARGBTable[256 * 4] = {
1u, 2u, 3u, 4u,
5u, 6u, 7u, 8u,
9u, 10u, 11u, 12u,
......@@ -602,7 +602,7 @@ TEST_F(libyuvTest, TestRGBColorTable) {
memset(orig_pixels, 0, sizeof(orig_pixels));
// Matrix for Sepia.
static const uint8 kARGBTable[1280 * 4] = {
static const uint8 kARGBTable[256 * 4] = {
1u, 2u, 3u, 4u,
5u, 6u, 7u, 8u,
9u, 10u, 11u, 12u,
......
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