Skip to content

L
libyuv
Commit 8d37dd5c authored by fbarchard@google.com's avatar fbarchard@google.com
Browse files
Options

Any conversions fix for RGB 3 bytes. Fix for overread valgrind. Avoid memcpy. … 

Any conversions fix for RGB 3 bytes.  Fix for overread valgrind.  Avoid memcpy.  Add _Any unittests for all conversions.
BUG=121
TEST=./libyuv_unittest --gtest_filter=*Any
Review URL: https://webrtc-codereview.appspot.com/873010

git-svn-id: http://libyuv.googlecode.com/svn/trunk@401 16f28f9a-4ce2-e073-06de-1de4eb20be90
parent e91bdaca
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 306 additions and 175 deletions
README.chromium
View file @ 8d37dd5c
Name: libyuv Name: libyuv
URL: http://code.google.com/p/libyuv/ URL: http://code.google.com/p/libyuv/
Version: 399 Version: 401
License: BSD License: BSD
License File: LICENSE License File: LICENSE
......
include/libyuv/row.h
View file @ 8d37dd5c
...@@ -42,6 +42,7 @@ extern "C" { ...@@ -42,6 +42,7 @@ extern "C" {
#define HAS_ARGB4444TOARGBROW_SSE2 #define HAS_ARGB4444TOARGBROW_SSE2
#define HAS_ARGBTOARGB1555ROW_SSE2 #define HAS_ARGBTOARGB1555ROW_SSE2
#define HAS_ARGBTOARGB4444ROW_SSE2 #define HAS_ARGBTOARGB4444ROW_SSE2
#define HAS_ARGBTOBAYERROW_SSSE3
#define HAS_ARGBTORAWROW_SSSE3 #define HAS_ARGBTORAWROW_SSSE3
#define HAS_ARGBTORGB24ROW_SSSE3 #define HAS_ARGBTORGB24ROW_SSSE3
#define HAS_ARGBTORGB565ROW_SSE2 #define HAS_ARGBTORGB565ROW_SSE2
...@@ -94,7 +95,7 @@ extern "C" { ...@@ -94,7 +95,7 @@ extern "C" {
#define HAS_CUMULATIVESUMTOAVERAGE_SSE2 #define HAS_CUMULATIVESUMTOAVERAGE_SSE2
#endif #endif
// The following are Windows only: // The following are Windows only. TODO(fbarchard): Port to gcc.
#if !defined(YUV_DISABLE_ASM) && defined(_M_IX86) #if !defined(YUV_DISABLE_ASM) && defined(_M_IX86)
#define HAS_ABGRTOARGBROW_SSSE3 #define HAS_ABGRTOARGBROW_SSSE3
#define HAS_ARGBCOLORTABLEROW_X86 #define HAS_ARGBCOLORTABLEROW_X86
...@@ -116,6 +117,7 @@ extern "C" { ...@@ -116,6 +117,7 @@ extern "C" {
// The following are available on Neon platforms // The following are available on Neon platforms
#if !defined(YUV_DISABLE_ASM) && (defined(__ARM_NEON__) || defined(LIBYUV_NEON)) #if !defined(YUV_DISABLE_ASM) && (defined(__ARM_NEON__) || defined(LIBYUV_NEON))
#define HAS_ARGBTOBAYERROW_NEON
#define HAS_COPYROW_NEON #define HAS_COPYROW_NEON
#define HAS_HALFROW_NEON #define HAS_HALFROW_NEON
#define HAS_I422TOABGRROW_NEON #define HAS_I422TOABGRROW_NEON
...@@ -759,6 +761,12 @@ void HalfRow_SSE2(const uint8* src_uv, int src_uv_stride, ...@@ -759,6 +761,12 @@ void HalfRow_SSE2(const uint8* src_uv, int src_uv_stride,
void HalfRow_NEON(const uint8* src_uv, int src_uv_stride, void HalfRow_NEON(const uint8* src_uv, int src_uv_stride,
uint8* dst_uv, int pix); uint8* dst_uv, int pix);
void ARGBToBayerRow_C(const uint8* src_argb,
uint8* dst_bayer, uint32 selector, int pix);
void ARGBToBayerRow_SSSE3(const uint8* src_argb,
uint8* dst_bayer, uint32 selector, int pix);
void ARGBToBayerRow_NEON(const uint8* src_argb,
uint8* dst_bayer, uint32 selector, int pix);
#ifdef __cplusplus #ifdef __cplusplus
} // extern "C" } // extern "C"
......
include/libyuv/version.h
View file @ 8d37dd5c
...@@ -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 399 #define LIBYUV_VERSION 401
#endif // INCLUDE_LIBYUV_VERSION_H_ NOLINT #endif // INCLUDE_LIBYUV_VERSION_H_ NOLINT
source/convert_argb.cc
View file @ 8d37dd5c
...@@ -115,14 +115,7 @@ int I422ToARGB(const uint8* src_y, int src_stride_y, ...@@ -115,14 +115,7 @@ int I422ToARGB(const uint8* src_y, int src_stride_y,
const uint8* v_buf, const uint8* v_buf,
uint8* rgb_buf, uint8* rgb_buf,
int width) = I422ToARGBRow_C; int width) = I422ToARGBRow_C;
#if defined(HAS_I422TOARGBROW_NEON) #if defined(HAS_I422TOARGBROW_SSSE3)
if (TestCpuFlag(kCpuHasNEON)) {
I422ToARGBRow = I422ToARGBRow_Any_NEON;
if (IS_ALIGNED(width, 16)) {
I422ToARGBRow = I422ToARGBRow_NEON;
}
}
#elif defined(HAS_I422TOARGBROW_SSSE3)
if (TestCpuFlag(kCpuHasSSSE3) && width >= 8) { if (TestCpuFlag(kCpuHasSSSE3) && width >= 8) {
I422ToARGBRow = I422ToARGBRow_Any_SSSE3; I422ToARGBRow = I422ToARGBRow_Any_SSSE3;
if (IS_ALIGNED(width, 8)) { if (IS_ALIGNED(width, 8)) {
...@@ -132,6 +125,13 @@ int I422ToARGB(const uint8* src_y, int src_stride_y, ...@@ -132,6 +125,13 @@ int I422ToARGB(const uint8* src_y, int src_stride_y,
} }
} }
} }
#elif defined(HAS_I422TOARGBROW_NEON)
if (TestCpuFlag(kCpuHasNEON) && width >= 8) {
I422ToARGBRow = I422ToARGBRow_Any_NEON;
if (IS_ALIGNED(width, 8)) {
I422ToARGBRow = I422ToARGBRow_NEON;
}
}
#endif #endif
for (int y = 0; y < height; ++y) { for (int y = 0; y < height; ++y) {
...@@ -189,7 +189,6 @@ int I411ToARGB(const uint8* src_y, int src_stride_y, ...@@ -189,7 +189,6 @@ int I411ToARGB(const uint8* src_y, int src_stride_y,
return 0; return 0;
} }
// Convert I400 to ARGB. // Convert I400 to ARGB.
LIBYUV_API LIBYUV_API
int I400ToARGB_Reference(const uint8* src_y, int src_stride_y, int I400ToARGB_Reference(const uint8* src_y, int src_stride_y,
...@@ -725,21 +724,23 @@ int YUY2ToARGB(const uint8* src_yuy2, int src_stride_yuy2, ...@@ -725,21 +724,23 @@ int YUY2ToARGB(const uint8* src_yuy2, int src_stride_yuy2,
void (*I422ToARGBRow)(const uint8* y_buf, void (*I422ToARGBRow)(const uint8* y_buf,
const uint8* u_buf, const uint8* u_buf,
const uint8* v_buf, const uint8* v_buf,
uint8* argb_buf, uint8* rgb_buf,
int width) = I422ToARGBRow_C; int width) = I422ToARGBRow_C;
#if defined(HAS_I422TOARGBROW_NEON) #if defined(HAS_I422TOARGBROW_SSSE3)
if (TestCpuFlag(kCpuHasNEON)) {
I422ToARGBRow = I422ToARGBRow_Any_NEON;
if (IS_ALIGNED(width, 16)) {
I422ToARGBRow = I422ToARGBRow_NEON;
}
}
#elif defined(HAS_I422TOARGBROW_SSSE3)
if (TestCpuFlag(kCpuHasSSSE3) && width >= 8) { if (TestCpuFlag(kCpuHasSSSE3) && width >= 8) {
I422ToARGBRow = I422ToARGBRow_Any_SSSE3; I422ToARGBRow = I422ToARGBRow_Any_SSSE3;
if (IS_ALIGNED(width, 8) && if (IS_ALIGNED(width, 8)) {
IS_ALIGNED(dst_argb, 16) && IS_ALIGNED(dst_stride_argb, 16)) { I422ToARGBRow = I422ToARGBRow_Unaligned_SSSE3;
I422ToARGBRow = I422ToARGBRow_SSSE3; if (IS_ALIGNED(dst_argb, 16) && IS_ALIGNED(dst_stride_argb, 16)) {
I422ToARGBRow = I422ToARGBRow_SSSE3;
}
}
}
#elif defined(HAS_I422TOARGBROW_NEON)
if (TestCpuFlag(kCpuHasNEON) && width >= 8) {
I422ToARGBRow = I422ToARGBRow_Any_NEON;
if (IS_ALIGNED(width, 8)) {
I422ToARGBRow = I422ToARGBRow_NEON;
} }
} }
#endif #endif
...@@ -796,21 +797,23 @@ int UYVYToARGB(const uint8* src_uyvy, int src_stride_uyvy, ...@@ -796,21 +797,23 @@ int UYVYToARGB(const uint8* src_uyvy, int src_stride_uyvy,
void (*I422ToARGBRow)(const uint8* y_buf, void (*I422ToARGBRow)(const uint8* y_buf,
const uint8* u_buf, const uint8* u_buf,
const uint8* v_buf, const uint8* v_buf,
uint8* argb_buf, uint8* rgb_buf,
int width) = I422ToARGBRow_C; int width) = I422ToARGBRow_C;
#if defined(HAS_I422TOARGBROW_NEON) #if defined(HAS_I422TOARGBROW_SSSE3)
if (TestCpuFlag(kCpuHasNEON)) {
I422ToARGBRow = I422ToARGBRow_Any_NEON;
if (IS_ALIGNED(width, 16)) {
I422ToARGBRow = I422ToARGBRow_NEON;
}
}
#elif defined(HAS_I422TOARGBROW_SSSE3)
if (TestCpuFlag(kCpuHasSSSE3) && width >= 8) { if (TestCpuFlag(kCpuHasSSSE3) && width >= 8) {
I422ToARGBRow = I422ToARGBRow_Any_SSSE3; I422ToARGBRow = I422ToARGBRow_Any_SSSE3;
if (IS_ALIGNED(width, 8) && if (IS_ALIGNED(width, 8)) {
IS_ALIGNED(dst_argb, 16) && IS_ALIGNED(dst_stride_argb, 16)) { I422ToARGBRow = I422ToARGBRow_Unaligned_SSSE3;
I422ToARGBRow = I422ToARGBRow_SSSE3; if (IS_ALIGNED(dst_argb, 16) && IS_ALIGNED(dst_stride_argb, 16)) {
I422ToARGBRow = I422ToARGBRow_SSSE3;
}
}
}
#elif defined(HAS_I422TOARGBROW_NEON)
if (TestCpuFlag(kCpuHasNEON) && width >= 8) {
I422ToARGBRow = I422ToARGBRow_Any_NEON;
if (IS_ALIGNED(width, 8)) {
I422ToARGBRow = I422ToARGBRow_NEON;
} }
} }
#endif #endif
......
source/convert_from.cc
View file @ 8d37dd5c
This diff is collapsed.
source/format_conversion.cc
View file @ 8d37dd5c
...@@ -20,81 +20,6 @@ namespace libyuv { ...@@ -20,81 +20,6 @@ namespace libyuv {
extern "C" { extern "C" {
#endif #endif
// Note: to do this with Neon vld4.8 would load ARGB values into 4 registers
// and vst would select which 2 components to write. The low level would need
// to be ARGBToBG, ARGBToGB, ARGBToRG, ARGBToGR
#if !defined(YUV_DISABLE_ASM) && defined(_M_IX86)
#define HAS_ARGBTOBAYERROW_SSSE3
__declspec(naked) __declspec(align(16))
static void ARGBToBayerRow_SSSE3(const uint8* src_argb,
uint8* dst_bayer, uint32 selector, int pix) {
__asm {
mov eax, [esp + 4] // src_argb
mov edx, [esp + 8] // dst_bayer
movd xmm5, [esp + 12] // selector
mov ecx, [esp + 16] // pix
pshufd xmm5, xmm5, 0
align 16
wloop:
movdqa xmm0, [eax]
lea eax, [eax + 16]
pshufb xmm0, xmm5
sub ecx, 4
movd [edx], xmm0
lea edx, [edx + 4]
jg wloop
ret
}
}
#elif !defined(YUV_DISABLE_ASM) && (defined(__x86_64__) || defined(__i386__))
#define HAS_ARGBTOBAYERROW_SSSE3
static void ARGBToBayerRow_SSSE3(const uint8* src_argb, uint8* dst_bayer,
uint32 selector, int pix) {
asm volatile (
"movd %3,%%xmm5 \n"
"pshufd $0x0,%%xmm5,%%xmm5 \n"
".p2align 4 \n"
"1: \n"
"movdqa (%0),%%xmm0 \n"
"lea 0x10(%0),%0 \n"
"pshufb %%xmm5,%%xmm0 \n"
"sub $0x4,%2 \n"
"movd %%xmm0,(%1) \n"
"lea 0x4(%1),%1 \n"
"jg 1b \n"
: "+r"(src_argb), // %0
"+r"(dst_bayer), // %1
"+r"(pix) // %2
: "g"(selector) // %3
: "memory", "cc"
#if defined(__SSE2__)
, "xmm0", "xmm5"
#endif
);
}
#endif
static void ARGBToBayerRow_C(const uint8* src_argb,
uint8* dst_bayer, uint32 selector, int pix) {
int index0 = selector & 0xff;
int index1 = (selector >> 8) & 0xff;
// Copy a row of Bayer.
for (int x = 0; x < pix - 1; x += 2) {
dst_bayer[0] = src_argb[index0];
dst_bayer[1] = src_argb[index1];
src_argb += 8;
dst_bayer += 2;
}
if (pix & 1) {
dst_bayer[0] = src_argb[index0];
}
}
// generate a selector mask useful for pshufb // generate a selector mask useful for pshufb
static uint32 GenerateSelector(int select0, int select1) { static uint32 GenerateSelector(int select0, int select1) {
return static_cast<uint32>(select0) | return static_cast<uint32>(select0) |
...@@ -147,11 +72,14 @@ int ARGBToBayer(const uint8* src_argb, int src_stride_argb, ...@@ -147,11 +72,14 @@ int ARGBToBayer(const uint8* src_argb, int src_stride_argb,
void (*ARGBToBayerRow)(const uint8* src_argb, uint8* dst_bayer, void (*ARGBToBayerRow)(const uint8* src_argb, uint8* dst_bayer,
uint32 selector, int pix) = ARGBToBayerRow_C; uint32 selector, int pix) = ARGBToBayerRow_C;
#if defined(HAS_ARGBTOBAYERROW_SSSE3) #if defined(HAS_ARGBTOBAYERROW_SSSE3)
if (TestCpuFlag(kCpuHasSSSE3) && if (TestCpuFlag(kCpuHasSSSE3) && IS_ALIGNED(width, 4) &&
IS_ALIGNED(width, 4) &&
IS_ALIGNED(src_argb, 16) && IS_ALIGNED(src_stride_argb, 16)) { IS_ALIGNED(src_argb, 16) && IS_ALIGNED(src_stride_argb, 16)) {
ARGBToBayerRow = ARGBToBayerRow_SSSE3; ARGBToBayerRow = ARGBToBayerRow_SSSE3;
} }
#elif defined(HAS_ARGBTOBAYERROW_NEON)
if (TestCpuFlag(kCpuHasNEON) && IS_ALIGNED(width, 4)) {
ARGBToBayerRow = ARGBToBayerRow_NEON;
}
#endif #endif
const int blue_index = 0; // Offsets for ARGB format const int blue_index = 0; // Offsets for ARGB format
const int green_index = 1; const int green_index = 1;
...@@ -455,15 +383,22 @@ int I420ToBayer(const uint8* src_y, int src_stride_y, ...@@ -455,15 +383,22 @@ int I420ToBayer(const uint8* src_y, int src_stride_y,
const uint8* v_buf, const uint8* v_buf,
uint8* rgb_buf, uint8* rgb_buf,
int width) = I422ToARGBRow_C; int width) = I422ToARGBRow_C;
#if defined(HAS_I422TOARGBROW_NEON) #if defined(HAS_I422TOARGBROW_SSSE3)
if (TestCpuFlag(kCpuHasNEON)) { if (TestCpuFlag(kCpuHasSSSE3) && width >= 8) {
I422ToARGBRow = I422ToARGBRow_NEON; I422ToARGBRow = I422ToARGBRow_Any_SSSE3;
if (IS_ALIGNED(width, 8)) {
I422ToARGBRow = I422ToARGBRow_Unaligned_SSSE3;
}
} }
#elif defined(HAS_I422TOARGBROW_SSSE3) #elif defined(HAS_I422TOARGBROW_NEON)
if (TestCpuFlag(kCpuHasSSSE3)) { if (TestCpuFlag(kCpuHasNEON) && width >= 8) {
I422ToARGBRow = I422ToARGBRow_SSSE3; I422ToARGBRow = I422ToARGBRow_Any_NEON;
if (IS_ALIGNED(width, 8)) {
I422ToARGBRow = I422ToARGBRow_NEON;
}
} }
#endif #endif
SIMD_ALIGNED(uint8 row[kMaxStride]); SIMD_ALIGNED(uint8 row[kMaxStride]);
void (*ARGBToBayerRow)(const uint8* src_argb, uint8* dst_bayer, void (*ARGBToBayerRow)(const uint8* src_argb, uint8* dst_bayer,
uint32 selector, int pix) = ARGBToBayerRow_C; uint32 selector, int pix) = ARGBToBayerRow_C;
...@@ -471,6 +406,10 @@ int I420ToBayer(const uint8* src_y, int src_stride_y, ...@@ -471,6 +406,10 @@ int I420ToBayer(const uint8* src_y, int src_stride_y,
if (TestCpuFlag(kCpuHasSSSE3) && IS_ALIGNED(width, 4)) { if (TestCpuFlag(kCpuHasSSSE3) && IS_ALIGNED(width, 4)) {
ARGBToBayerRow = ARGBToBayerRow_SSSE3; ARGBToBayerRow = ARGBToBayerRow_SSSE3;
} }
#elif defined(HAS_ARGBTOBAYERROW_NEON)
if (TestCpuFlag(kCpuHasNEON) && IS_ALIGNED(width, 4)) {
ARGBToBayerRow = ARGBToBayerRow_NEON;
}
#endif #endif
const int blue_index = 0; // Offsets for ARGB format const int blue_index = 0; // Offsets for ARGB format
const int green_index = 1; const int green_index = 1;
......
source/planar_functions.cc
View file @ 8d37dd5c
...@@ -626,8 +626,7 @@ int ARGBToRGBA(const uint8* src_argb, int src_stride_argb, ...@@ -626,8 +626,7 @@ int ARGBToRGBA(const uint8* src_argb, int src_stride_argb,
IS_ALIGNED(dst_rgba, 16) && IS_ALIGNED(dst_stride_rgba, 16)) { IS_ALIGNED(dst_rgba, 16) && IS_ALIGNED(dst_stride_rgba, 16)) {
ARGBToRGBARow = ARGBToRGBARow_SSSE3; ARGBToRGBARow = ARGBToRGBARow_SSSE3;
} }
#endif #elif defined(HAS_ARGBTORGBAROW_NEON)
#if defined(HAS_ARGBTORGBAROW_NEON)
if (TestCpuFlag(kCpuHasNEON) && IS_ALIGNED(width, 8)) { if (TestCpuFlag(kCpuHasNEON) && IS_ALIGNED(width, 8)) {
ARGBToRGBARow = ARGBToRGBARow_NEON; ARGBToRGBARow = ARGBToRGBARow_NEON;
} }
...@@ -657,22 +656,17 @@ int ARGBToRGB24(const uint8* src_argb, int src_stride_argb, ...@@ -657,22 +656,17 @@ int ARGBToRGB24(const uint8* src_argb, int src_stride_argb,
void (*ARGBToRGB24Row)(const uint8* src_argb, uint8* dst_rgb, int pix) = void (*ARGBToRGB24Row)(const uint8* src_argb, uint8* dst_rgb, int pix) =
ARGBToRGB24Row_C; ARGBToRGB24Row_C;
#if defined(HAS_ARGBTORGB24ROW_SSSE3) #if defined(HAS_ARGBTORGB24ROW_SSSE3)
if (TestCpuFlag(kCpuHasSSSE3) && if (TestCpuFlag(kCpuHasSSSE3) && width >= 16 &&
IS_ALIGNED(src_argb, 16) && IS_ALIGNED(src_stride_argb, 16)) { IS_ALIGNED(src_argb, 16) && IS_ALIGNED(src_stride_argb, 16) &&
if (width * 3 <= kMaxStride) { IS_ALIGNED(dst_rgb24, 16) && IS_ALIGNED(dst_stride_rgb24, 16)) {
ARGBToRGB24Row = ARGBToRGB24Row_Any_SSSE3; ARGBToRGB24Row = ARGBToRGB24Row_Any_SSSE3;
} if (IS_ALIGNED(width, 16)) {
if (IS_ALIGNED(width, 16) &&
IS_ALIGNED(dst_rgb24, 16) && IS_ALIGNED(dst_stride_rgb24, 16)) {
ARGBToRGB24Row = ARGBToRGB24Row_SSSE3; ARGBToRGB24Row = ARGBToRGB24Row_SSSE3;
} }
} }
#endif #elif defined(HAS_ARGBTORGB24ROW_NEON)
#if defined(HAS_ARGBTORGB24ROW_NEON) if (TestCpuFlag(kCpuHasNEON) && width >= 8) {
if (TestCpuFlag(kCpuHasNEON)) { ARGBToRGB24Row = ARGBToRGB24Row_Any_NEON;
if (width * 3 <= kMaxStride) {
ARGBToRGB24Row = ARGBToRGB24Row_Any_NEON;
}
if (IS_ALIGNED(width, 8)) { if (IS_ALIGNED(width, 8)) {
ARGBToRGB24Row = ARGBToRGB24Row_NEON; ARGBToRGB24Row = ARGBToRGB24Row_NEON;
} }
...@@ -703,22 +697,17 @@ int ARGBToRAW(const uint8* src_argb, int src_stride_argb, ...@@ -703,22 +697,17 @@ int ARGBToRAW(const uint8* src_argb, int src_stride_argb,
void (*ARGBToRAWRow)(const uint8* src_argb, uint8* dst_rgb, int pix) = void (*ARGBToRAWRow)(const uint8* src_argb, uint8* dst_rgb, int pix) =
ARGBToRAWRow_C; ARGBToRAWRow_C;
#if defined(HAS_ARGBTORAWROW_SSSE3) #if defined(HAS_ARGBTORAWROW_SSSE3)
if (TestCpuFlag(kCpuHasSSSE3) && if (TestCpuFlag(kCpuHasSSSE3) && width >= 16 &&
IS_ALIGNED(src_argb, 16) && IS_ALIGNED(src_stride_argb, 16)) { IS_ALIGNED(src_argb, 16) && IS_ALIGNED(src_stride_argb, 16) &&
if (width * 3 <= kMaxStride) { IS_ALIGNED(dst_raw, 16) && IS_ALIGNED(dst_stride_raw, 16)) {
ARGBToRAWRow = ARGBToRAWRow_Any_SSSE3; ARGBToRAWRow = ARGBToRAWRow_Any_SSSE3;
} if (IS_ALIGNED(width, 16)) {
if (IS_ALIGNED(width, 16) &&
IS_ALIGNED(dst_raw, 16) && IS_ALIGNED(dst_stride_raw, 16)) {
ARGBToRAWRow = ARGBToRAWRow_SSSE3; ARGBToRAWRow = ARGBToRAWRow_SSSE3;
} }
} }
#endif #elif defined(HAS_ARGBTORAWROW_NEON)
#if defined(HAS_ARGBTORAWROW_NEON) if (TestCpuFlag(kCpuHasNEON) && width >= 8) {
if (TestCpuFlag(kCpuHasNEON)) { ARGBToRAWRow = ARGBToRAWRow_Any_NEON;
if (width * 3 <= kMaxStride) {
ARGBToRAWRow = ARGBToRAWRow_Any_NEON;
}
if (IS_ALIGNED(width, 8)) { if (IS_ALIGNED(width, 8)) {
ARGBToRAWRow = ARGBToRAWRow_NEON; ARGBToRAWRow = ARGBToRAWRow_NEON;
} }
...@@ -749,11 +738,9 @@ int ARGBToRGB565(const uint8* src_argb, int src_stride_argb, ...@@ -749,11 +738,9 @@ int ARGBToRGB565(const uint8* src_argb, int src_stride_argb,
void (*ARGBToRGB565Row)(const uint8* src_argb, uint8* dst_rgb, int pix) = void (*ARGBToRGB565Row)(const uint8* src_argb, uint8* dst_rgb, int pix) =
ARGBToRGB565Row_C; ARGBToRGB565Row_C;
#if defined(HAS_ARGBTORGB565ROW_SSE2) #if defined(HAS_ARGBTORGB565ROW_SSE2)
if (TestCpuFlag(kCpuHasSSE2) && if (TestCpuFlag(kCpuHasSSE2) && width >= 4 &&
IS_ALIGNED(src_argb, 16) && IS_ALIGNED(src_stride_argb, 16)) { IS_ALIGNED(src_argb, 16) && IS_ALIGNED(src_stride_argb, 16)) {
if (width * 2 <= kMaxStride) { ARGBToRGB565Row = ARGBToRGB565Row_Any_SSE2;
ARGBToRGB565Row = ARGBToRGB565Row_Any_SSE2;
}
if (IS_ALIGNED(width, 4)) { if (IS_ALIGNED(width, 4)) {
ARGBToRGB565Row = ARGBToRGB565Row_SSE2; ARGBToRGB565Row = ARGBToRGB565Row_SSE2;
} }
...@@ -784,11 +771,9 @@ int ARGBToARGB1555(const uint8* src_argb, int src_stride_argb, ...@@ -784,11 +771,9 @@ int ARGBToARGB1555(const uint8* src_argb, int src_stride_argb,
void (*ARGBToARGB1555Row)(const uint8* src_argb, uint8* dst_rgb, int pix) = void (*ARGBToARGB1555Row)(const uint8* src_argb, uint8* dst_rgb, int pix) =
ARGBToARGB1555Row_C; ARGBToARGB1555Row_C;
#if defined(HAS_ARGBTOARGB1555ROW_SSE2) #if defined(HAS_ARGBTOARGB1555ROW_SSE2)
if (TestCpuFlag(kCpuHasSSE2) && if (TestCpuFlag(kCpuHasSSE2) && width >= 4 &&
IS_ALIGNED(src_argb, 16) && IS_ALIGNED(src_stride_argb, 16)) { IS_ALIGNED(src_argb, 16) && IS_ALIGNED(src_stride_argb, 16)) {
if (width * 2 <= kMaxStride) { ARGBToARGB1555Row = ARGBToARGB1555Row_Any_SSE2;
ARGBToARGB1555Row = ARGBToARGB1555Row_Any_SSE2;
}
if (IS_ALIGNED(width, 4)) { if (IS_ALIGNED(width, 4)) {
ARGBToARGB1555Row = ARGBToARGB1555Row_SSE2; ARGBToARGB1555Row = ARGBToARGB1555Row_SSE2;
} }
...@@ -819,11 +804,9 @@ int ARGBToARGB4444(const uint8* src_argb, int src_stride_argb, ...@@ -819,11 +804,9 @@ int ARGBToARGB4444(const uint8* src_argb, int src_stride_argb,
void (*ARGBToARGB4444Row)(const uint8* src_argb, uint8* dst_rgb, int pix) = void (*ARGBToARGB4444Row)(const uint8* src_argb, uint8* dst_rgb, int pix) =
ARGBToARGB4444Row_C; ARGBToARGB4444Row_C;
#if defined(HAS_ARGBTOARGB4444ROW_SSE2) #if defined(HAS_ARGBTOARGB4444ROW_SSE2)
if (TestCpuFlag(kCpuHasSSE2) && if (TestCpuFlag(kCpuHasSSE2) && width >= 4 &&
IS_ALIGNED(src_argb, 16) && IS_ALIGNED(src_stride_argb, 16)) { IS_ALIGNED(src_argb, 16) && IS_ALIGNED(src_stride_argb, 16)) {
if (width * 2 <= kMaxStride) { ARGBToARGB4444Row = ARGBToARGB4444Row_Any_SSE2;
ARGBToARGB4444Row = ARGBToARGB4444Row_Any_SSE2;
}
if (IS_ALIGNED(width, 4)) { if (IS_ALIGNED(width, 4)) {
ARGBToARGB4444Row = ARGBToARGB4444Row_SSE2; ARGBToARGB4444Row = ARGBToARGB4444Row_SSE2;
} }
...@@ -839,7 +822,7 @@ int ARGBToARGB4444(const uint8* src_argb, int src_stride_argb, ...@@ -839,7 +822,7 @@ int ARGBToARGB4444(const uint8* src_argb, int src_stride_argb,
} }
// Convert NV12 to RGB565. // Convert NV12 to RGB565.
// TODO(fbarchard): (Re) Optimize for Neon. // TODO(fbarchard): One pass conversion.
LIBYUV_API LIBYUV_API
int NV12ToRGB565(const uint8* src_y, int src_stride_y, int NV12ToRGB565(const uint8* src_y, int src_stride_y,
const uint8* src_uv, int src_stride_uv, const uint8* src_uv, int src_stride_uv,
...@@ -859,22 +842,26 @@ int NV12ToRGB565(const uint8* src_y, int src_stride_y, ...@@ -859,22 +842,26 @@ int NV12ToRGB565(const uint8* src_y, int src_stride_y,
uint8* rgb_buf, uint8* rgb_buf,
int width) = NV12ToARGBRow_C; int width) = NV12ToARGBRow_C;
#if defined(HAS_NV12TOARGBROW_SSSE3) #if defined(HAS_NV12TOARGBROW_SSSE3)
if (TestCpuFlag(kCpuHasSSSE3) && width * 4 <= kMaxStride) { if (TestCpuFlag(kCpuHasSSSE3) && IS_ALIGNED(width, 8)) {
NV12ToARGBRow = NV12ToARGBRow_SSSE3; NV12ToARGBRow = NV12ToARGBRow_SSSE3;
} }
#endif #elif defined(HAS_NV12TOARGBROW_NEON)
#if defined(HAS_NV12TOARGBROW_NEON) if (TestCpuFlag(kCpuHasNEON) && IS_ALIGNED(width, 8)) {
if (TestCpuFlag(kCpuHasNEON) && width * 4 <= kMaxStride) {
NV12ToARGBRow = NV12ToARGBRow_NEON; NV12ToARGBRow = NV12ToARGBRow_NEON;
} }
#endif #endif
if (width * 4 > kMaxStride) {
return -1;
}
SIMD_ALIGNED(uint8 row[kMaxStride]); SIMD_ALIGNED(uint8 row[kMaxStride]);
void (*ARGBToRGB565Row)(const uint8* src_argb, uint8* dst_rgb, int pix) = void (*ARGBToRGB565Row)(const uint8* src_argb, uint8* dst_rgb, int pix) =
ARGBToRGB565Row_C; ARGBToRGB565Row_C;
#if defined(HAS_ARGBTORGB565ROW_SSE2) #if defined(HAS_ARGBTORGB565ROW_SSE2)
if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(width, 4)) { if (TestCpuFlag(kCpuHasSSE2) && width >= 4) {
ARGBToRGB565Row = ARGBToRGB565Row_SSE2; ARGBToRGB565Row = ARGBToRGB565Row_Any_SSE2;
if (IS_ALIGNED(width, 4)) {
ARGBToRGB565Row = ARGBToRGB565Row_SSE2;
}
} }
#endif #endif
...@@ -893,10 +880,10 @@ int NV12ToRGB565(const uint8* src_y, int src_stride_y, ...@@ -893,10 +880,10 @@ int NV12ToRGB565(const uint8* src_y, int src_stride_y,
// Convert NV21 to RGB565. // Convert NV21 to RGB565.
LIBYUV_API LIBYUV_API
int NV21ToRGB565(const uint8* src_y, int src_stride_y, int NV21ToRGB565(const uint8* src_y, int src_stride_y,
const uint8* src_vu, int src_stride_vu, const uint8* src_uv, int src_stride_uv,
uint8* dst_rgb565, int dst_stride_rgb565, uint8* dst_rgb565, int dst_stride_rgb565,
int width, int height) { int width, int height) {
if (!src_y || !src_vu || !dst_rgb565 || width <= 0 || height == 0) { if (!src_y || !src_uv || !dst_rgb565 || width <= 0 || height == 0) {
return -1; return -1;
} }
// Negative height means invert the image. // Negative height means invert the image.
...@@ -910,27 +897,36 @@ int NV21ToRGB565(const uint8* src_y, int src_stride_y, ...@@ -910,27 +897,36 @@ int NV21ToRGB565(const uint8* src_y, int src_stride_y,
uint8* rgb_buf, uint8* rgb_buf,
int width) = NV21ToARGBRow_C; int width) = NV21ToARGBRow_C;
#if defined(HAS_NV21TOARGBROW_SSSE3) #if defined(HAS_NV21TOARGBROW_SSSE3)
if (TestCpuFlag(kCpuHasSSSE3) && width * 4 <= kMaxStride) { if (TestCpuFlag(kCpuHasSSSE3) && IS_ALIGNED(width, 8)) {
NV21ToARGBRow = NV21ToARGBRow_SSSE3; NV21ToARGBRow = NV21ToARGBRow_SSSE3;
} }
#elif defined(HAS_NV21TOARGBROW_NEON)
if (TestCpuFlag(kCpuHasNEON) && IS_ALIGNED(width, 8)) {
NV21ToARGBRow = NV21ToARGBRow_NEON;
}
#endif #endif
if (width * 4 > kMaxStride) {
return -1;
}
SIMD_ALIGNED(uint8 row[kMaxStride]); SIMD_ALIGNED(uint8 row[kMaxStride]);
void (*ARGBToRGB565Row)(const uint8* src_argb, uint8* dst_rgb, int pix) = void (*ARGBToRGB565Row)(const uint8* src_argb, uint8* dst_rgb, int pix) =
ARGBToRGB565Row_C; ARGBToRGB565Row_C;
#if defined(HAS_ARGBTORGB565ROW_SSE2) #if defined(HAS_ARGBTORGB565ROW_SSE2)
if (TestCpuFlag(kCpuHasSSE2) && IS_ALIGNED(width, 4)) { if (TestCpuFlag(kCpuHasSSE2) && width >= 4) {
ARGBToRGB565Row = ARGBToRGB565Row_SSE2; ARGBToRGB565Row = ARGBToRGB565Row_Any_SSE2;
if (IS_ALIGNED(width, 4)) {
ARGBToRGB565Row = ARGBToRGB565Row_SSE2;
}
} }
#endif #endif
for (int y = 0; y < height; ++y) { for (int y = 0; y < height; ++y) {
NV21ToARGBRow(src_y, src_vu, row, width); NV21ToARGBRow(src_y, src_uv, row, width);
ARGBToRGB565Row(row, dst_rgb565, width); ARGBToRGB565Row(row, dst_rgb565, width);
dst_rgb565 += dst_stride_rgb565; dst_rgb565 += dst_stride_rgb565;
src_y += src_stride_y; src_y += src_stride_y;
if (y & 1) { if (y & 1) {
src_vu += src_stride_vu; src_uv += src_stride_uv;
} }
} }
return 0; return 0;
......
source/row_common.cc
View file @ 8d37dd5c
This diff is collapsed.
source/row_neon.cc
View file @ 8d37dd5c
...@@ -842,6 +842,24 @@ void HalfRow_NEON(const uint8* src_uv, int src_uv_stride, ...@@ -842,6 +842,24 @@ void HalfRow_NEON(const uint8* src_uv, int src_uv_stride,
); );
} }
// Select 2 channels from ARGB on alternating pixels. e.g. BGBGBGBG
// TODO(fbarchard): Neon port.
void ARGBToBayerRow_NEON(const uint8* src_argb,
uint8* dst_bayer, uint32 selector, int pix) {
int index0 = selector & 0xff;
int index1 = (selector >> 8) & 0xff;
// Copy a row of Bayer.
for (int x = 0; x < pix - 1; x += 2) {
dst_bayer[0] = src_argb[index0];
dst_bayer[1] = src_argb[index1];
src_argb += 8;
dst_bayer += 2;
}
if (pix & 1) {
dst_bayer[0] = src_argb[index0];
}
}
#endif // __ARM_NEON__ #endif // __ARM_NEON__
#ifdef __cplusplus #ifdef __cplusplus
......
source/row_posix.cc
View file @ 8d37dd5c
...@@ -118,6 +118,16 @@ CONST uvec8 kShuffleMaskARGBToRAW = { ...@@ -118,6 +118,16 @@ CONST uvec8 kShuffleMaskARGBToRAW = {
2u, 1u, 0u, 6u, 5u, 4u, 10u, 9u, 8u, 14u, 13u, 12u, 128u, 128u, 128u, 128u 2u, 1u, 0u, 6u, 5u, 4u, 10u, 9u, 8u, 14u, 13u, 12u, 128u, 128u, 128u, 128u
}; };
// Shuffle table for converting ARGBToRGB24 for I420ToRGB24. First 8 + next 4
CONST uvec8 kShuffleMaskARGBToRGB24_0 = {
0u, 1u, 2u, 4u, 5u, 6u, 8u, 9u, 128u, 128u, 128u, 128u, 10u, 12u, 13u, 14u
};
// Shuffle table for converting ARGB to RAW.
CONST uvec8 kShuffleMaskARGBToRAW_0 = {
2u, 1u, 0u, 6u, 5u, 4u, 10u, 9u, 128u, 128u, 128u, 128u, 8u, 14u, 13u, 12u
};
void I400ToARGBRow_SSE2(const uint8* src_y, uint8* dst_argb, int pix) { void I400ToARGBRow_SSE2(const uint8* src_y, uint8* dst_argb, int pix) {
asm volatile ( asm volatile (
"pcmpeqb %%xmm5,%%xmm5 \n" "pcmpeqb %%xmm5,%%xmm5 \n"
...@@ -1431,6 +1441,115 @@ void OMITFP I444ToARGBRow_SSSE3(const uint8* y_buf, ...@@ -1431,6 +1441,115 @@ void OMITFP I444ToARGBRow_SSSE3(const uint8* y_buf,
); );
} }
void OMITFP I422ToRGB24Row_SSSE3(const uint8* y_buf,
const uint8* u_buf,
const uint8* v_buf,
uint8* rgb24_buf,
int width) {
// fpic 32 bit gcc 4.2 on OSX runs out of GPR regs.
#ifdef __APPLE__
asm volatile (
"movdqa %[kShuffleMaskARGBToRGB24],%%xmm5 \n"
"movdqa %[kShuffleMaskARGBToRGB24_0],%%xmm6 \n"
:: [kShuffleMaskARGBToRGB24]"m"(kShuffleMaskARGBToRGB24),
[kShuffleMaskARGBToRGB24_0]"m"(kShuffleMaskARGBToRGB24_0));
#endif
asm volatile (
#ifndef __APPLE__
"movdqa %[kShuffleMaskARGBToRGB24],%%xmm5 \n"
"movdqa %[kShuffleMaskARGBToRGB24_0],%%xmm6 \n"
#endif
"sub %[u_buf],%[v_buf] \n"
"pxor %%xmm4,%%xmm4 \n"
".p2align 4 \n"
"1: \n"
READYUV422
YUVTORGB
"punpcklbw %%xmm1,%%xmm0 \n"
"punpcklbw %%xmm2,%%xmm2 \n"
"movdqa %%xmm0,%%xmm1 \n"
"punpcklwd %%xmm2,%%xmm0 \n"
"punpckhwd %%xmm2,%%xmm1 \n"
"pshufb %%xmm5,%%xmm0 \n"
"pshufb %%xmm6,%%xmm1 \n"
"palignr $0xc,%%xmm0,%%xmm1 \n"
"movq %%xmm0,(%[rgb24_buf]) \n"
"movdqu %%xmm1,0x8(%[rgb24_buf]) \n"
"lea 0x18(%[rgb24_buf]),%[rgb24_buf] \n"
"sub $0x8,%[width] \n"
"jg 1b \n"
: [y_buf]"+r"(y_buf), // %[y_buf]
[u_buf]"+r"(u_buf), // %[u_buf]
[v_buf]"+r"(v_buf), // %[v_buf]
[rgb24_buf]"+r"(rgb24_buf), // %[rgb24_buf]
[width]"+rm"(width) // %[width]
: [kYuvConstants]"r"(&kYuvConstants.kUVToB)
#ifndef __APPLE__
, [kShuffleMaskARGBToRGB24]"m"(kShuffleMaskARGBToRGB24),
[kShuffleMaskARGBToRGB24_0]"m"(kShuffleMaskARGBToRGB24_0)
#endif
: "memory", "cc"
#if defined(__SSE2__)
, "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6"
#endif
);
}
void OMITFP I422ToRAWRow_SSSE3(const uint8* y_buf,
const uint8* u_buf,
const uint8* v_buf,
uint8* raw_buf,
int width) {
#ifdef __APPLE__
asm volatile (
"movdqa %[kShuffleMaskARGBToRAW],%%xmm5 \n"
"movdqa %[kShuffleMaskARGBToRAW_0],%%xmm6 \n"
:: [kShuffleMaskARGBToRAW]"m"(kShuffleMaskARGBToRAW),
[kShuffleMaskARGBToRAW_0]"m"(kShuffleMaskARGBToRAW_0));
#endif
asm volatile (
#ifndef __APPLE__
"movdqa %[kShuffleMaskARGBToRAW],%%xmm5 \n"
"movdqa %[kShuffleMaskARGBToRAW_0],%%xmm6 \n"
#endif
"sub %[u_buf],%[v_buf] \n"
"pxor %%xmm4,%%xmm4 \n"
".p2align 4 \n"
"1: \n"
READYUV422
YUVTORGB
"punpcklbw %%xmm1,%%xmm0 \n"
"punpcklbw %%xmm2,%%xmm2 \n"
"movdqa %%xmm0,%%xmm1 \n"
"punpcklwd %%xmm2,%%xmm0 \n"
"punpckhwd %%xmm2,%%xmm1 \n"
"pshufb %%xmm5,%%xmm0 \n"
"pshufb %%xmm6,%%xmm1 \n"
"palignr $0xc,%%xmm0,%%xmm1 \n"
"movq %%xmm0,(%[raw_buf]) \n"
"movdqu %%xmm1,0x8(%[raw_buf]) \n"
"lea 0x18(%[raw_buf]),%[raw_buf] \n"
"sub $0x8,%[width] \n"
"jg 1b \n"
: [y_buf]"+r"(y_buf), // %[y_buf]
[u_buf]"+r"(u_buf), // %[u_buf]
[v_buf]"+r"(v_buf), // %[v_buf]
[raw_buf]"+r"(raw_buf), // %[raw_buf]
[width]"+rm"(width) // %[width]
: [kYuvConstants]"r"(&kYuvConstants.kUVToB)
#ifndef __APPLE__
, [kShuffleMaskARGBToRAW]"m"(kShuffleMaskARGBToRAW),
[kShuffleMaskARGBToRAW_0]"m"(kShuffleMaskARGBToRAW_0)
#endif
: "memory", "cc"
#if defined(__SSE2__)
, "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5"
#endif
);
}
void OMITFP I422ToARGBRow_SSSE3(const uint8* y_buf, void OMITFP I422ToARGBRow_SSSE3(const uint8* y_buf,
const uint8* u_buf, const uint8* u_buf,
const uint8* v_buf, const uint8* v_buf,
...@@ -3751,6 +3870,31 @@ void HalfRow_SSE2(const uint8* src_uv, int src_uv_stride, ...@@ -3751,6 +3870,31 @@ void HalfRow_SSE2(const uint8* src_uv, int src_uv_stride,
#endif #endif
); );
} }
void ARGBToBayerRow_SSSE3(const uint8* src_argb, uint8* dst_bayer,
uint32 selector, int pix) {
asm volatile (
"movd %3,%%xmm5 \n"
"pshufd $0x0,%%xmm5,%%xmm5 \n"
".p2align 4 \n"
"1: \n"
"movdqa (%0),%%xmm0 \n"
"lea 0x10(%0),%0 \n"
"pshufb %%xmm5,%%xmm0 \n"
"sub $0x4,%2 \n"
"movd %%xmm0,(%1) \n"
"lea 0x4(%1),%1 \n"
"jg 1b \n"
: "+r"(src_argb), // %0
"+r"(dst_bayer), // %1
"+r"(pix) // %2
: "g"(selector) // %3
: "memory", "cc"
#if defined(__SSE2__)
, "xmm0", "xmm5"
#endif
);
}
#endif // defined(__x86_64__) || defined(__i386__) #endif // defined(__x86_64__) || defined(__i386__)
#ifdef __cplusplus #ifdef __cplusplus
......
source/row_win.cc
View file @ 8d37dd5c
...@@ -4217,6 +4217,29 @@ void HalfRow_SSE2(const uint8* src_uv, int src_uv_stride, ...@@ -4217,6 +4217,29 @@ void HalfRow_SSE2(const uint8* src_uv, int src_uv_stride,
} }
} }
__declspec(naked) __declspec(align(16))
void ARGBToBayerRow_SSSE3(const uint8* src_argb,
uint8* dst_bayer, uint32 selector, int pix) {
__asm {
mov eax, [esp + 4] // src_argb
mov edx, [esp + 8] // dst_bayer
movd xmm5, [esp + 12] // selector
mov ecx, [esp + 16] // pix
pshufd xmm5, xmm5, 0
align 16
wloop:
movdqa xmm0, [eax]
lea eax, [eax + 16]
pshufb xmm0, xmm5
sub ecx, 4
movd [edx], xmm0
lea edx, [edx + 4]
jg wloop
ret
}
}
#endif // _M_IX86 #endif // _M_IX86
#ifdef __cplusplus #ifdef __cplusplus
......
unit_test/planar_test.cc
View file @ 8d37dd5c
This diff is collapsed.
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