Skip to content

L
libyuv
Commit ca26f860 authored by fbarchard@google.com's avatar fbarchard@google.com
Browse files
Options

ARGBRect use SetRows32 to do full image. 744 ms -> 288 ms on Linux, 688 ms ->… 

ARGBRect use SetRows32 to do full image. 744 ms -> 288 ms on Linux, 688 ms -> 277 ms on Mac.  Started ARGBCopy function.
BUG=none
TEST=none
Review URL: http://webrtc-codereview.appspot.com/290006

git-svn-id: http://libyuv.googlecode.com/svn/trunk@95 16f28f9a-4ce2-e073-06de-1de4eb20be90
parent 120b8d7e
Hide whitespace changes
Inline Side-by-side
Showing with 138 additions and 85 deletions
README.chromium
View file @ ca26f860
Name: libyuv Name: libyuv
URL: http://code.google.com/p/libyuv/ URL: http://code.google.com/p/libyuv/
Version: 94 Version: 95
License: BSD License: BSD
License File: LICENSE License File: LICENSE
......
include/libyuv/planar_functions.h
View file @ ca26f860
...@@ -189,6 +189,11 @@ int ARGBRect(uint8* dst_argb, int dst_stride_argb, ...@@ -189,6 +189,11 @@ int ARGBRect(uint8* dst_argb, int dst_stride_argb,
int width, int height, int width, int height,
uint32 value); uint32 value);
// Copy ARGB to ARGB.
int ARGBCopy(const uint8* src_argb, int src_stride_argb,
uint8* dst_argb, int dst_stride_argb,
int width, int height);
} // namespace libyuv } // namespace libyuv
#endif // INCLUDE_LIBYUV_PLANAR_FUNCTIONS_H_ #endif // INCLUDE_LIBYUV_PLANAR_FUNCTIONS_H_
source/planar_functions.cc
View file @ ca26f860
...@@ -129,6 +129,7 @@ static void SplitUV_C(const uint8* src_uv, ...@@ -129,6 +129,7 @@ static void SplitUV_C(const uint8* src_uv,
// CopyRows copys 'count' bytes using a 16 byte load/store, 64 bytes at time // CopyRows copys 'count' bytes using a 16 byte load/store, 64 bytes at time
#if defined(_M_IX86) && !defined(YUV_DISABLE_ASM) #if defined(_M_IX86) && !defined(YUV_DISABLE_ASM)
#define HAS_COPYROW_SSE2 #define HAS_COPYROW_SSE2
#define HAS_COPYROW_X86
__declspec(naked) __declspec(naked)
void CopyRow_SSE2(const uint8* src, uint8* dst, int count) { void CopyRow_SSE2(const uint8* src, uint8* dst, int count) {
__asm { __asm {
...@@ -148,6 +149,21 @@ void CopyRow_SSE2(const uint8* src, uint8* dst, int count) { ...@@ -148,6 +149,21 @@ void CopyRow_SSE2(const uint8* src, uint8* dst, int count) {
ret ret
} }
} }
void CopyRow_X86(const uint8* src, uint8* dst, int count) {
__asm {
push esi
push edi
mov esi, [esp + 4 + 4] // src
mov edi, [esp + 4 + 8] // dst
mov ecx, [esp + 4 + 12] // count
shr ecx, 2
rep movsd
pop edi
pop esi
ret
}
}
#elif (defined(__x86_64__) || defined(__i386__)) && !defined(YUV_DISABLE_ASM) #elif (defined(__x86_64__) || defined(__i386__)) && !defined(YUV_DISABLE_ASM)
#define HAS_COPYROW_SSE2 #define HAS_COPYROW_SSE2
void CopyRow_SSE2(const uint8* src, uint8* dst, int count) { void CopyRow_SSE2(const uint8* src, uint8* dst, int count) {
...@@ -184,9 +200,17 @@ static void CopyPlane(const uint8* src_y, int src_stride_y, ...@@ -184,9 +200,17 @@ static void CopyPlane(const uint8* src_y, int src_stride_y,
#if defined(HAS_COPYROW_SSE2) #if defined(HAS_COPYROW_SSE2)
if (TestCpuFlag(kCpuHasSSE2) && if (TestCpuFlag(kCpuHasSSE2) &&
IS_ALIGNED(width, 32) && IS_ALIGNED(width, 32) &&
IS_ALIGNED(src_y, 16) && IS_ALIGNED(src_stride_y, 16)) { IS_ALIGNED(src_y, 16) && IS_ALIGNED(src_stride_y, 16) &&
IS_ALIGNED(dst_y, 16) && IS_ALIGNED(dst_stride_y, 16)) {
CopyRow = CopyRow_SSE2; CopyRow = CopyRow_SSE2;
} else } else
#endif
#if defined(HAS_COPYROW_X86)
if (IS_ALIGNED(width, 4) &&
IS_ALIGNED(src_y, 4) && IS_ALIGNED(src_stride_y, 4) &&
IS_ALIGNED(dst_y, 4) && IS_ALIGNED(dst_stride_y, 4)) {
CopyRow = CopyRow_X86;
} else
#endif #endif
{ {
CopyRow = CopyRow_C; CopyRow = CopyRow_C;
...@@ -233,6 +257,28 @@ int I420Copy(const uint8* src_y, int src_stride_y, ...@@ -233,6 +257,28 @@ int I420Copy(const uint8* src_y, int src_stride_y,
return 0; return 0;
} }
// Copy ARGB with optional flipping
int ARGBCopy(const uint8* src_argb, int src_stride_argb,
uint8* dst_argb, int dst_stride_argb,
int width, int height) {
if (!src_argb ||
!dst_argb ||
width <= 0 || height == 0) {
return -1;
}
// Negative height means invert the image.
if (height < 0) {
height = -height;
src_argb = src_argb + (height - 1) * src_stride_argb;
src_stride_argb = -src_stride_argb;
}
CopyPlane(src_argb, src_stride_argb, dst_argb, dst_stride_argb,
width * 4, height);
return 0;
}
int I420Mirror(const uint8* src_y, int src_stride_y, int I420Mirror(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u, const uint8* src_u, int src_stride_u,
const uint8* src_v, int src_stride_v, const uint8* src_v, int src_stride_v,
...@@ -1673,109 +1719,111 @@ static void SetRow8_NEON(uint8* dst, uint32 v32, int count) { ...@@ -1673,109 +1719,111 @@ static void SetRow8_NEON(uint8* dst, uint32 v32, int count) {
); );
} }
static void SetRow32_NEON(uint8* dst, uint32 v32, int count) { // TODO(fbarchard): Make fully assembler
asm volatile ( static void SetRows32_NEON(uint8* dst, uint32 v32, int width,
"vdup.u32 q0, %2 \n" // duplicate 4 ints int dst_stride, int height) {
"1: \n" for (int y = 0; y < height; ++y) {
"vst1.u32 {q0}, [%0]! \n" // store SetRow8_NEON(dst, v32, width << 2);
"subs %1, %1, #4 \n" // 4 pixels per loop dst += dst_stride;
"bhi 1b \n" }
: "+r"(dst), // %0
"+r"(count) // %1
: "r"(v32) // %2
: "q0", "memory", "cc"
);
} }
#elif defined(_M_IX86) && !defined(YUV_DISABLE_ASM) #elif defined(_M_IX86) && !defined(YUV_DISABLE_ASM)
#define HAS_SETROW_SSE2 #define HAS_SETROW_X86
__declspec(naked) __declspec(naked)
static void SetRow8_SSE2(uint8* dst, uint32 v32, int count) { static void SetRow8_X86(uint8* dst, uint32 v32, int count) {
__asm { __asm {
mov eax, [esp + 4] // dst push edi
movd xmm5, [esp + 8] // v32 mov edi, [esp + 4 + 4] // dst
mov ecx, [esp + 12] // count mov eax, [esp + 4 + 8] // v32
pshufd xmm5, xmm5, 0 mov ecx, [esp + 4 + 12] // count
shr ecx, 2
convertloop: rep stosd
movdqa [eax], xmm5 pop edi
lea eax, [eax + 16]
sub ecx, 16
ja convertloop
ret ret
} }
} }
__declspec(naked) __declspec(naked)
static void SetRow32_SSE2(uint8* dst, uint32 v32, int count) { static void SetRows32_X86(uint8* dst, uint32 v32, int width,
int dst_stride, int height) {
__asm { __asm {
mov eax, [esp + 4] // dst push edi
movd xmm5, [esp + 8] // v32 push ebp
mov ecx, [esp + 12] // count mov edi, [esp + 8 + 4] // dst
pshufd xmm5, xmm5, 0 mov eax, [esp + 8 + 8] // v32
mov ebp, [esp + 8 + 12] // width
mov edx, [esp + 8 + 16] // dst_stride
mov ebx, [esp + 8 + 20] // height
lea ecx, [ebp * 4]
sub edx, ecx // stride - width * 4
convertloop: convertloop:
movdqa [eax], xmm5 mov ecx, ebp
lea eax, [eax + 16] rep stosd
sub ecx, 4 add edi, edx
sub ebx, 1
ja convertloop ja convertloop
pop ebp
pop edi
ret ret
} }
} }
#elif (defined(__x86_64__) || defined(__i386__)) && !defined(YUV_DISABLE_ASM) #elif (defined(__x86_64__) || defined(__i386__)) && !defined(YUV_DISABLE_ASM)
#define HAS_SETROW_X86
#define HAS_SETROW_SSE2 static void SetRow8_X86(uint8* dst, uint32 v32, int width) {
static void SetRow8_SSE2(uint8* dst, uint32 v32, int count) { size_t width_tmp = static_cast<size_t>(width);
asm volatile ( asm volatile (
"movd %2, %%xmm5 \n" "shr $0x2,%1 \n"
"pshufd $0x0,%%xmm5,%%xmm5 \n" "rep stos %2,(%0) \n"
"1: \n" : "+D"(dst), // %0
"movdqa %%xmm5,(%0) \n" "+c"(width_tmp) // %1
"lea 0x10(%0),%0 \n" : "a"(v32) // %2
"sub $0x10,%1 \n"
"ja 1b \n"
: "+r"(dst), // %0
"+r"(count) // %1
: "r"(v32) // %2
: "memory", "cc" : "memory", "cc"
#if defined(__SSE2__)
, "xmm5"
#endif
); );
} }
static void SetRow32_SSE2(uint8* dst, uint32 v32, int count) { static void SetRows32_X86(uint8* dst, uint32 v32, int width,
asm volatile ( int dst_stride, int height) {
"movd %2, %%xmm5 \n" for (int y = 0; y < height; ++y) {
"pshufd $0x0,%%xmm5,%%xmm5 \n" size_t width_tmp = static_cast<size_t>(width);
"1: \n" uint32* d = reinterpret_cast<uint32*>(dst);
"movdqa %%xmm5,(%0) \n" asm volatile (
"lea 0x10(%0),%0 \n" "rep stos %2,(%0) \n"
"sub $0x4,%1 \n" : "+D"(d), // %0
"ja 1b \n" "+c"(width_tmp) // %1
: "+r"(dst), // %0 : "a"(v32) // %2
"+r"(count) // %1 : "memory", "cc"
: "r"(v32) // %2 );
: "memory", "cc" dst += dst_stride;
#if defined(__SSE2__) }
, "xmm5"
#endif
);
} }
#endif #endif
#if !defined(HAS_SETROW_X86)
static void SetRow8_C(uint8* dst, uint32 v8, int count) { static void SetRow8_C(uint8* dst, uint32 v8, int count) {
#ifdef _MSC_VER
for (int x = 0; x < count; ++x) {
dst[x] = v8;
}
#else
memset(dst, v8, count); memset(dst, v8, count);
#endif
} }
// count measured in bytes static void SetRows32_C(uint8* dst, uint32 v32, int width,
static void SetRow32_C(uint8* dst, uint32 v32, int count) { int dst_stride, int height) {
uint32* d = reinterpret_cast<uint32*>(dst); for (int y = 0; y < height; ++y) {
for (int x = 0; x < count; ++x) { uint32* d = reinterpret_cast<uint32*>(dst);
d[x] = v32; for (int x = 0; x < width; ++x) {
d[x] = v32;
}
dst += dst_stride;
} }
} }
#endif
static void SetPlane(uint8* dst_y, int dst_stride_y, static void SetPlane(uint8* dst_y, int dst_stride_y,
int width, int height, int width, int height,
...@@ -1795,7 +1843,11 @@ static void SetPlane(uint8* dst_y, int dst_stride_y, ...@@ -1795,7 +1843,11 @@ static void SetPlane(uint8* dst_y, int dst_stride_y,
} else } else
#endif #endif
{ {
#if defined(HAS_SETROW_X86)
SetRow = SetRow8_X86;
#else
SetRow = SetRow8_C; SetRow = SetRow8_C;
#endif
} }
uint32 v32 = value | (value << 8) | (value << 16) | (value << 24); uint32 v32 = value | (value << 8) | (value << 16) | (value << 24);
...@@ -1844,27 +1896,23 @@ int ARGBRect(uint8* dst_argb, int dst_stride_argb, ...@@ -1844,27 +1896,23 @@ int ARGBRect(uint8* dst_argb, int dst_stride_argb,
return -1; return -1;
} }
uint8* dst = dst_argb + dst_y * dst_stride_argb + dst_x * 4; uint8* dst = dst_argb + dst_y * dst_stride_argb + dst_x * 4;
void (*SetRow)(uint8* dst, uint32 value, int count); void (*SetRows)(uint8* dst, uint32 value, int width,
int dst_stride, int height);
#if defined(HAS_SETROW_NEON) #if defined(HAS_SETROW_NEON)
if (TestCpuFlag(kCpuHasNEON) && if (TestCpuFlag(kCpuHasNEON) &&
IS_ALIGNED(width, 16) && IS_ALIGNED(width, 16) &&
IS_ALIGNED(dst, 16) && IS_ALIGNED(dst_stride_argb, 16)) { IS_ALIGNED(dst, 16) && IS_ALIGNED(dst_stride_argb, 16)) {
SetRow = SetRow32_NEON; SetRows = SetRows32_NEON;
} else
#elif defined(HAS_SETROW_SSE2)
if (TestCpuFlag(kCpuHasSSE2) &&
IS_ALIGNED(width, 16) &&
IS_ALIGNED(dst, 16) && IS_ALIGNED(dst_stride_argb, 16)) {
SetRow = SetRow32_SSE2;
} else } else
#endif #endif
{ {
SetRow = SetRow32_C; #if defined(HAS_SETROW_X86)
} SetRows = SetRows32_X86;
for (int y = 0; y < height; ++y) { #else
SetRow(dst, value, width); SetRows = SetRows32_C;
dst += dst_stride_argb; #endif
} }
SetRows(dst, value, width, dst_stride_argb, height);
return 0; return 0;
} }
......
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