Skip to content

L
libyuv
Commit 9de8867a authored by fbarchard@google.com's avatar fbarchard@google.com
Browse files
Options

I422ToYUY2_NEON 

BUG=none
TEST=convert_test *I422ToYUY2*
Review URL: https://webrtc-codereview.appspot.com/869008

git-svn-id: http://libyuv.googlecode.com/svn/trunk@413 16f28f9a-4ce2-e073-06de-1de4eb20be90
parent 7d3786c6
Hide whitespace changes
Inline Side-by-side
Showing with 375 additions and 292 deletions
README.chromium
View file @ 9de8867a
Name: libyuv
URL: http://code.google.com/p/libyuv/
Version: 410
Version: 413
License: BSD
License File: LICENSE
......
include/libyuv/row.h
View file @ 9de8867a
......@@ -84,6 +84,8 @@ extern "C" {
#define HAS_YUY2TOUV422ROW_SSE2
#define HAS_YUY2TOUVROW_SSE2
#define HAS_YUY2TOYROW_SSE2
#define HAS_I422TOYUY2ROW_SSE2
#define HAS_I422TOUYVYROW_SSE2
// Effects
#define HAS_ARGBAFFINEROW_SSE2
......@@ -119,6 +121,9 @@ extern "C" {
#if !defined(YUV_DISABLE_ASM) && (defined(__ARM_NEON__) || defined(LIBYUV_NEON))
#define HAS_ABGRTOARGBROW_NEON
#define HAS_ARGBTOBAYERROW_NEON
#define HAS_ARGBTORAWROW_NEON
#define HAS_ARGBTORGB24ROW_NEON
#define HAS_ARGBTORGBAROW_NEON
#define HAS_BGRATOARGBROW_NEON
#define HAS_COPYROW_NEON
#define HAS_HALFROW_NEON
......@@ -130,6 +135,8 @@ extern "C" {
#define HAS_I422TORGBAROW_NEON
#define HAS_MIRRORROW_NEON
#define HAS_MIRRORROWUV_NEON
#define HAS_NV12TOARGBROW_NEON
#define HAS_NV21TOARGBROW_NEON
#define HAS_RAWTOARGBROW_NEON
#define HAS_RGB24TOARGBROW_NEON
#define HAS_RGBATOARGBROW_NEON
......@@ -141,13 +148,9 @@ extern "C" {
#define HAS_YUY2TOUV422ROW_NEON
#define HAS_YUY2TOUVROW_NEON
#define HAS_YUY2TOYROW_NEON
#define HAS_I422TOYUY2ROW_NEON
#define HAS_I422TOUYVYROW_NEON
// TODO(fbarchard): Hook these up to calling functions.
#define HAS_ARGBTORAWROW_NEON
#define HAS_ARGBTORGB24ROW_NEON
#define HAS_ARGBTORGBAROW_NEON
#define HAS_NV12TOARGBROW_NEON
#define HAS_NV21TOARGBROW_NEON
#endif
#if defined(_MSC_VER) && !defined(__CLR_VER)
......@@ -768,6 +771,31 @@ void ARGBToBayerRow_SSSE3(const uint8* src_argb,
void ARGBToBayerRow_NEON(const uint8* src_argb,
uint8* dst_bayer, uint32 selector, int pix);
void I422ToYUY2Row_C(const uint8* src_y,
const uint8* src_u,
const uint8* src_v,
uint8* dst_frame, int width);
void I422ToUYVYRow_C(const uint8* src_y,
const uint8* src_u,
const uint8* src_v,
uint8* dst_frame, int width);
void I422ToYUY2Row_SSE2(const uint8* src_y,
const uint8* src_u,
const uint8* src_v,
uint8* dst_frame, int width);
void I422ToUYVYRow_SSE2(const uint8* src_y,
const uint8* src_u,
const uint8* src_v,
uint8* dst_frame, int width);
void I422ToYUY2Row_NEON(const uint8* src_y,
const uint8* src_u,
const uint8* src_v,
uint8* dst_frame, int width);
void I422ToUYVYRow_NEON(const uint8* src_y,
const uint8* src_u,
const uint8* src_v,
uint8* dst_frame, int width);
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
......
include/libyuv/version.h
View file @ 9de8867a
......@@ -11,6 +11,6 @@
#ifndef INCLUDE_LIBYUV_VERSION_H_ // NOLINT
#define INCLUDE_LIBYUV_VERSION_H_
#define LIBYUV_VERSION 410
#define LIBYUV_VERSION 413
#endif // INCLUDE_LIBYUV_VERSION_H_ NOLINT
source/convert_from.cc
View file @ 9de8867a
......@@ -222,208 +222,6 @@ int I400Copy(const uint8* src_y, int src_stride_y,
return 0;
}
// YUY2 - Macro-pixel = 2 image pixels
// Y0U0Y1V0....Y2U2Y3V2...Y4U4Y5V4....
// UYVY - Macro-pixel = 2 image pixels
// U0Y0V0Y1
#if !defined(YUV_DISABLE_ASM) && defined(_M_IX86)
#define HAS_I422TOYUY2ROW_SSE2
__declspec(naked) __declspec(align(16))
static void I422ToYUY2Row_SSE2(const uint8* src_y,
const uint8* src_u,
const uint8* src_v,
uint8* dst_frame, int width) {
__asm {
push esi
push edi
mov eax, [esp + 8 + 4] // src_y
mov esi, [esp + 8 + 8] // src_u
mov edx, [esp + 8 + 12] // src_v
mov edi, [esp + 8 + 16] // dst_frame
mov ecx, [esp + 8 + 20] // width
sub edx, esi
align 16
convertloop:
movq xmm2, qword ptr [esi] // U
movq xmm3, qword ptr [esi + edx] // V
lea esi, [esi + 8]
punpcklbw xmm2, xmm3 // UV
movdqa xmm0, [eax] // Y
lea eax, [eax + 16]
movdqa xmm1, xmm0
punpcklbw xmm0, xmm2 // YUYV
punpckhbw xmm1, xmm2
movdqa [edi], xmm0
movdqa [edi + 16], xmm1
lea edi, [edi + 32]
sub ecx, 16
jg convertloop
pop edi
pop esi
ret
}
}
#define HAS_I422TOUYVYROW_SSE2
__declspec(naked) __declspec(align(16))
static void I422ToUYVYRow_SSE2(const uint8* src_y,
const uint8* src_u,
const uint8* src_v,
uint8* dst_frame, int width) {
__asm {
push esi
push edi
mov eax, [esp + 8 + 4] // src_y
mov esi, [esp + 8 + 8] // src_u
mov edx, [esp + 8 + 12] // src_v
mov edi, [esp + 8 + 16] // dst_frame
mov ecx, [esp + 8 + 20] // width
sub edx, esi
align 16
convertloop:
movq xmm2, qword ptr [esi] // U
movq xmm3, qword ptr [esi + edx] // V
lea esi, [esi + 8]
punpcklbw xmm2, xmm3 // UV
movdqa xmm0, [eax] // Y
movdqa xmm1, xmm2
lea eax, [eax + 16]
punpcklbw xmm1, xmm0 // UYVY
punpckhbw xmm2, xmm0
movdqa [edi], xmm1
movdqa [edi + 16], xmm2
lea edi, [edi + 32]
sub ecx, 16
jg convertloop
pop edi
pop esi
ret
}
}
#elif !defined(YUV_DISABLE_ASM) && (defined(__x86_64__) || defined(__i386__))
#define HAS_I422TOYUY2ROW_SSE2
static void I422ToYUY2Row_SSE2(const uint8* src_y,
const uint8* src_u,
const uint8* src_v,
uint8* dst_frame, int width) {
asm volatile (
"sub %1,%2 \n"
".p2align 4 \n"
"1: \n"
"movq (%1),%%xmm2 \n"
"movq (%1,%2,1),%%xmm3 \n"
"lea 0x8(%1),%1 \n"
"punpcklbw %%xmm3,%%xmm2 \n"
"movdqa (%0),%%xmm0 \n"
"lea 0x10(%0),%0 \n"
"movdqa %%xmm0,%%xmm1 \n"
"punpcklbw %%xmm2,%%xmm0 \n"
"punpckhbw %%xmm2,%%xmm1 \n"
"movdqa %%xmm0,(%3) \n"
"movdqa %%xmm1,0x10(%3) \n"
"lea 0x20(%3),%3 \n"
"sub $0x10,%4 \n"
"jg 1b \n"
: "+r"(src_y), // %0
"+r"(src_u), // %1
"+r"(src_v), // %2
"+r"(dst_frame), // %3
"+rm"(width) // %4
:
: "memory", "cc"
#if defined(__SSE2__)
, "xmm0", "xmm1", "xmm2", "xmm3"
#endif
);
}
#define HAS_I422TOUYVYROW_SSE2
static void I422ToUYVYRow_SSE2(const uint8* src_y,
const uint8* src_u,
const uint8* src_v,
uint8* dst_frame, int width) {
asm volatile (
"sub %1,%2 \n"
".p2align 4 \n"
"1: \n"
"movq (%1),%%xmm2 \n"
"movq (%1,%2,1),%%xmm3 \n"
"lea 0x8(%1),%1 \n"
"punpcklbw %%xmm3,%%xmm2 \n"
"movdqa (%0),%%xmm0 \n"
"movdqa %%xmm2,%%xmm1 \n"
"lea 0x10(%0),%0 \n"
"punpcklbw %%xmm0,%%xmm1 \n"
"punpckhbw %%xmm0,%%xmm2 \n"
"movdqa %%xmm1,(%3) \n"
"movdqa %%xmm2,0x10(%3) \n"
"lea 0x20(%3),%3 \n"
"sub $0x10,%4 \n"
"jg 1b \n"
: "+r"(src_y), // %0
"+r"(src_u), // %1
"+r"(src_v), // %2
"+r"(dst_frame), // %3
"+rm"(width) // %4
:
: "memory", "cc"
#if defined(__SSE2__)
, "xmm0", "xmm1", "xmm2", "xmm3"
#endif
);
}
#endif
static void I422ToYUY2Row_C(const uint8* src_y,
const uint8* src_u,
const uint8* src_v,
uint8* dst_frame, int width) {
for (int x = 0; x < width - 1; x += 2) {
dst_frame[0] = src_y[0];
dst_frame[1] = src_u[0];
dst_frame[2] = src_y[1];
dst_frame[3] = src_v[0];
dst_frame += 4;
src_y += 2;
src_u += 1;
src_v += 1;
}
if (width & 1) {
dst_frame[0] = src_y[0];
dst_frame[1] = src_u[0];
dst_frame[2] = src_y[0]; // duplicate last y
dst_frame[3] = src_v[0];
}
}
static void I422ToUYVYRow_C(const uint8* src_y,
const uint8* src_u,
const uint8* src_v,
uint8* dst_frame, int width) {
for (int x = 0; x < width - 1; x += 2) {
dst_frame[0] = src_u[0];
dst_frame[1] = src_y[0];
dst_frame[2] = src_v[0];
dst_frame[3] = src_y[1];
dst_frame += 4;
src_y += 2;
src_u += 1;
src_v += 1;
}
if (width & 1) {
dst_frame[0] = src_u[0];
dst_frame[1] = src_y[0];
dst_frame[2] = src_v[0];
dst_frame[3] = src_y[0]; // duplicate last y
}
}
// Visual C x86 or GCC little endian.
#if defined(__x86_64__) || defined(_M_X64) || \
defined(__i386__) || defined(_M_IX86) || \
......@@ -463,7 +261,6 @@ static void UYVYToV210Row_C(const uint8* src_uyvy, uint8* dst_v210, int width) {
}
}
// TODO(fbarchard): Deprecate, move or expand 422 support?
LIBYUV_API
int I422ToYUY2(const uint8* src_y, int src_stride_y,
const uint8* src_u, int src_stride_u,
......@@ -490,6 +287,10 @@ int I422ToYUY2(const uint8* src_y, int src_stride_y,
IS_ALIGNED(dst_frame, 16) && IS_ALIGNED(dst_stride_frame, 16)) {
I422ToYUY2Row = I422ToYUY2Row_SSE2;
}
#elif defined(HAS_I422TOYUY2ROW_NEON)
if (TestCpuFlag(kCpuHasNEON) && IS_ALIGNED(width, 16)) {
I422ToYUY2Row = I422ToYUY2Row_NEON;
}
#endif
for (int y = 0; y < height; ++y) {
......@@ -528,6 +329,10 @@ int I420ToYUY2(const uint8* src_y, int src_stride_y,
IS_ALIGNED(dst_frame, 16) && IS_ALIGNED(dst_stride_frame, 16)) {
I422ToYUY2Row = I422ToYUY2Row_SSE2;
}
#elif defined(HAS_I422TOYUY2ROW_NEON)
if (TestCpuFlag(kCpuHasNEON) && IS_ALIGNED(width, 16)) {
I422ToYUY2Row = I422ToYUY2Row_NEON;
}
#endif
for (int y = 0; y < height - 1; y += 2) {
......@@ -572,6 +377,10 @@ int I422ToUYVY(const uint8* src_y, int src_stride_y,
IS_ALIGNED(dst_frame, 16) && IS_ALIGNED(dst_stride_frame, 16)) {
I422ToUYVYRow = I422ToUYVYRow_SSE2;
}
#elif defined(HAS_I422TOUYVYROW_NEON)
if (TestCpuFlag(kCpuHasNEON) && IS_ALIGNED(width, 16)) {
I422ToUYVYRow = I422ToUYVYRow_NEON;
}
#endif
for (int y = 0; y < height; ++y) {
......@@ -610,6 +419,10 @@ int I420ToUYVY(const uint8* src_y, int src_stride_y,
IS_ALIGNED(dst_frame, 16) && IS_ALIGNED(dst_stride_frame, 16)) {
I422ToUYVYRow = I422ToUYVYRow_SSE2;
}
#elif defined(HAS_I422TOUYVYROW_NEON)
if (TestCpuFlag(kCpuHasNEON) && IS_ALIGNED(width, 16)) {
I422ToUYVYRow = I422ToUYVYRow_NEON;
}
#endif
for (int y = 0; y < height - 1; y += 2) {
......
source/row_common.cc
View file @ 9de8867a
......@@ -1279,6 +1279,50 @@ void ARGBToBayerRow_C(const uint8* src_argb,
}
}
void I422ToYUY2Row_C(const uint8* src_y,
const uint8* src_u,
const uint8* src_v,
uint8* dst_frame, int width) {
for (int x = 0; x < width - 1; x += 2) {
dst_frame[0] = src_y[0];
dst_frame[1] = src_u[0];
dst_frame[2] = src_y[1];
dst_frame[3] = src_v[0];
dst_frame += 4;
src_y += 2;
src_u += 1;
src_v += 1;
}
if (width & 1) {
dst_frame[0] = src_y[0];
dst_frame[1] = src_u[0];
dst_frame[2] = src_y[0]; // duplicate last y
dst_frame[3] = src_v[0];
}
}
void I422ToUYVYRow_C(const uint8* src_y,
const uint8* src_u,
const uint8* src_v,
uint8* dst_frame, int width) {
for (int x = 0; x < width - 1; x += 2) {
dst_frame[0] = src_u[0];
dst_frame[1] = src_y[0];
dst_frame[2] = src_v[0];
dst_frame[3] = src_y[1];
dst_frame += 4;
src_y += 2;
src_u += 1;
src_v += 1;
}
if (width & 1) {
dst_frame[0] = src_u[0];
dst_frame[1] = src_y[0];
dst_frame[2] = src_v[0];
dst_frame[3] = src_y[0]; // duplicate last y
}
}
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
......
source/row_neon.cc
View file @ 9de8867a
......@@ -75,10 +75,10 @@ static const vec8 kUVToG = { -25, -25, -25, -25, -52, -52, -52, -52,
#endif
#ifdef HAS_I422TOARGBROW_NEON
void I422ToARGBRow_NEON(const uint8* y_buf,
const uint8* u_buf,
const uint8* v_buf,
uint8* rgb_buf,
void I422ToARGBRow_NEON(const uint8* src_y,
const uint8* src_u,
const uint8* src_v,
uint8* dst_argb,
int width) {
asm volatile (
"vld1.u8 {d24}, [%5] \n"
......@@ -94,13 +94,13 @@ void I422ToARGBRow_NEON(const uint8* y_buf,
"vmov.u8 d23, #255 \n"
"vst4.8 {d20, d21, d22, d23}, [%3]! \n"
"bgt 1b \n"
: "+r"(y_buf), // %0
"+r"(u_buf), // %1
"+r"(v_buf), // %2
"+r"(rgb_buf), // %3
"+r"(width) // %4
: "r"(&kUVToRB), // %5
"r"(&kUVToG) // %6
: "+r"(src_y), // %0
"+r"(src_u), // %1
"+r"(src_v), // %2
"+r"(dst_argb), // %3
"+r"(width) // %4
: "r"(&kUVToRB), // %5
"r"(&kUVToG) // %6
: "cc", "memory", "q0", "q1", "q2", "q3",
"q8", "q9", "q10", "q11", "q12", "q13", "q14", "q15"
);
......@@ -108,10 +108,10 @@ void I422ToARGBRow_NEON(const uint8* y_buf,
#endif // HAS_I422TOARGBROW_NEON
#ifdef HAS_I422TOBGRAROW_NEON
void I422ToBGRARow_NEON(const uint8* y_buf,
const uint8* u_buf,
const uint8* v_buf,
uint8* rgb_buf,
void I422ToBGRARow_NEON(const uint8* src_y,
const uint8* src_u,
const uint8* src_v,
uint8* dst_bgra,
int width) {
asm volatile (
"vld1.u8 {d24}, [%5] \n"
......@@ -128,13 +128,13 @@ void I422ToBGRARow_NEON(const uint8* y_buf,
"vmov.u8 d19, #255 \n"
"vst4.8 {d19, d20, d21, d22}, [%3]! \n"
"bgt 1b \n"
: "+r"(y_buf), // %0
"+r"(u_buf), // %1
"+r"(v_buf), // %2
"+r"(rgb_buf), // %3
"+r"(width) // %4
: "r"(&kUVToRB), // %5
"r"(&kUVToG) // %6
: "+r"(src_y), // %0
"+r"(src_u), // %1
"+r"(src_v), // %2
"+r"(dst_bgra), // %3
"+r"(width) // %4
: "r"(&kUVToRB), // %5
"r"(&kUVToG) // %6
: "cc", "memory", "q0", "q1", "q2", "q3",
"q8", "q9", "q10", "q11", "q12", "q13", "q14", "q15"
);
......@@ -142,10 +142,10 @@ void I422ToBGRARow_NEON(const uint8* y_buf,
#endif // HAS_I422TOBGRAROW_NEON
#ifdef HAS_I422TOABGRROW_NEON
void I422ToABGRRow_NEON(const uint8* y_buf,
const uint8* u_buf,
const uint8* v_buf,
uint8* rgb_buf,
void I422ToABGRRow_NEON(const uint8* src_y,
const uint8* src_u,
const uint8* src_v,
uint8* dst_abgr,
int width) {
asm volatile (
"vld1.u8 {d24}, [%5] \n"
......@@ -162,13 +162,13 @@ void I422ToABGRRow_NEON(const uint8* y_buf,
"vmov.u8 d23, #255 \n"
"vst4.8 {d20, d21, d22, d23}, [%3]! \n"
"bgt 1b \n"
: "+r"(y_buf), // %0
"+r"(u_buf), // %1
"+r"(v_buf), // %2
"+r"(rgb_buf), // %3
"+r"(width) // %4
: "r"(&kUVToRB), // %5
"r"(&kUVToG) // %6
: "+r"(src_y), // %0
"+r"(src_u), // %1
"+r"(src_v), // %2
"+r"(dst_abgr), // %3
"+r"(width) // %4
: "r"(&kUVToRB), // %5
"r"(&kUVToG) // %6
: "cc", "memory", "q0", "q1", "q2", "q3",
"q8", "q9", "q10", "q11", "q12", "q13", "q14", "q15"
);
......@@ -176,10 +176,10 @@ void I422ToABGRRow_NEON(const uint8* y_buf,
#endif // HAS_I422TOABGRROW_NEON
#ifdef HAS_I422TORGBAROW_NEON
void I422ToRGBARow_NEON(const uint8* y_buf,
const uint8* u_buf,
const uint8* v_buf,
uint8* rgb_buf,
void I422ToRGBARow_NEON(const uint8* src_y,
const uint8* src_u,
const uint8* src_v,
uint8* dst_rgba,
int width) {
asm volatile (
"vld1.u8 {d24}, [%5] \n"
......@@ -195,13 +195,13 @@ void I422ToRGBARow_NEON(const uint8* y_buf,
"vmov.u8 d19, #255 \n"
"vst4.8 {d19, d20, d21, d22}, [%3]! \n"
"bgt 1b \n"
: "+r"(y_buf), // %0
"+r"(u_buf), // %1
"+r"(v_buf), // %2
"+r"(rgb_buf), // %3
"+r"(width) // %4
: "r"(&kUVToRB), // %5
"r"(&kUVToG) // %6
: "+r"(src_y), // %0
"+r"(src_u), // %1
"+r"(src_v), // %2
"+r"(dst_rgba), // %3
"+r"(width) // %4
: "r"(&kUVToRB), // %5
"r"(&kUVToG) // %6
: "cc", "memory", "q0", "q1", "q2", "q3",
"q8", "q9", "q10", "q11", "q12", "q13", "q14", "q15"
);
......@@ -209,10 +209,10 @@ void I422ToRGBARow_NEON(const uint8* y_buf,
#endif // HAS_I422TORGBAROW_NEON
#ifdef HAS_I422TORGB24ROW_NEON
void I422ToRGB24Row_NEON(const uint8* y_buf,
const uint8* u_buf,
const uint8* v_buf,
uint8* rgb_buf,
void I422ToRGB24Row_NEON(const uint8* src_y,
const uint8* src_u,
const uint8* src_v,
uint8* dst_rgb24,
int width) {
asm volatile (
"vld1.u8 {d24}, [%5] \n"
......@@ -227,13 +227,13 @@ void I422ToRGB24Row_NEON(const uint8* y_buf,
"subs %4, %4, #8 \n"
"vst3.8 {d20, d21, d22}, [%3]! \n"
"bgt 1b \n"
: "+r"(y_buf), // %0
"+r"(u_buf), // %1
"+r"(v_buf), // %2
"+r"(rgb_buf), // %3
"+r"(width) // %4
: "r"(&kUVToRB), // %5
"r"(&kUVToG) // %6
: "+r"(src_y), // %0
"+r"(src_u), // %1
"+r"(src_v), // %2
"+r"(dst_rgb24), // %3
"+r"(width) // %4
: "r"(&kUVToRB), // %5
"r"(&kUVToG) // %6
: "cc", "memory", "q0", "q1", "q2", "q3",
"q8", "q9", "q10", "q11", "q12", "q13", "q14", "q15"
);
......@@ -241,10 +241,10 @@ void I422ToRGB24Row_NEON(const uint8* y_buf,
#endif // HAS_I422TORGB24ROW_NEON
#ifdef HAS_I422TORAWROW_NEON
void I422ToRAWRow_NEON(const uint8* y_buf,
const uint8* u_buf,
const uint8* v_buf,
uint8* rgb_buf,
void I422ToRAWRow_NEON(const uint8* src_y,
const uint8* src_u,
const uint8* src_v,
uint8* dst_raw,
int width) {
asm volatile (
"vld1.u8 {d24}, [%5] \n"
......@@ -260,10 +260,10 @@ void I422ToRAWRow_NEON(const uint8* y_buf,
"vswp.u8 d20, d22 \n"
"vst3.8 {d20, d21, d22}, [%3]! \n"
"bgt 1b \n"
: "+r"(y_buf), // %0
"+r"(u_buf), // %1
"+r"(v_buf), // %2
"+r"(rgb_buf), // %3
: "+r"(src_y), // %0
"+r"(src_u), // %1
"+r"(src_v), // %2
"+r"(dst_raw), // %3
"+r"(width) // %4
: "r"(&kUVToRB), // %5
"r"(&kUVToG) // %6
......@@ -274,9 +274,9 @@ void I422ToRAWRow_NEON(const uint8* y_buf,
#endif // HAS_I422TORAWROW_NEON
#ifdef HAS_NV12TOARGBROW_NEON
void NV12ToARGBRow_NEON(const uint8* y_buf,
const uint8* uv_buf,
uint8* rgb_buf,
void NV12ToARGBRow_NEON(const uint8* src_y,
const uint8* src_uv,
uint8* dst_argb,
int width) {
asm volatile (
"vld1.u8 {d24}, [%4] \n"
......@@ -292,12 +292,12 @@ void NV12ToARGBRow_NEON(const uint8* y_buf,
"vmov.u8 d23, #255 \n"
"vst4.8 {d20, d21, d22, d23}, [%2]! \n"
"bgt 1b \n"
: "+r"(y_buf), // %0
"+r"(uv_buf), // %1
"+r"(rgb_buf), // %2
"+r"(width) // %3
: "r"(&kUVToRB), // %4
"r"(&kUVToG) // %5
: "+r"(src_y), // %0
"+r"(src_uv), // %1
"+r"(dst_argb), // %2
"+r"(width) // %3
: "r"(&kUVToRB), // %4
"r"(&kUVToG) // %5
: "cc", "memory", "q0", "q1", "q2", "q3",
"q8", "q9", "q10", "q11", "q12", "q13", "q14", "q15"
);
......@@ -305,9 +305,9 @@ void NV12ToARGBRow_NEON(const uint8* y_buf,
#endif // HAS_NV12TOARGBROW_NEON
#ifdef HAS_NV21TOARGBROW_NEON
void NV21ToARGBRow_NEON(const uint8* y_buf,
const uint8* uv_buf,
uint8* rgb_buf,
void NV21ToARGBRow_NEON(const uint8* src_y,
const uint8* src_uv,
uint8* dst_argb,
int width) {
asm volatile (
"vld1.u8 {d24}, [%4] \n"
......@@ -323,12 +323,12 @@ void NV21ToARGBRow_NEON(const uint8* y_buf,
"vmov.u8 d23, #255 \n"
"vst4.8 {d20, d21, d22, d23}, [%2]! \n"
"bgt 1b \n"
: "+r"(y_buf), // %0
"+r"(uv_buf), // %1
"+r"(rgb_buf), // %2
"+r"(width) // %3
: "r"(&kUVToRB), // %4
"r"(&kUVToG) // %5
: "+r"(src_y), // %0
"+r"(src_uv), // %1
"+r"(dst_argb), // %2
"+r"(width) // %3
: "r"(&kUVToRB), // %4
"r"(&kUVToG) // %5
: "cc", "memory", "q0", "q1", "q2", "q3",
"q8", "q9", "q10", "q11", "q12", "q13", "q14", "q15"
);
......@@ -862,6 +862,52 @@ void ARGBToBayerRow_NEON(const uint8* src_argb,
);
}
void I422ToYUY2Row_NEON(const uint8* src_y,
const uint8* src_u,
const uint8* src_v,
uint8* dst_yuy2, int width) {
asm volatile (
".p2align 2 \n"
"1: \n"
"vld2.8 {d0, d2}, [%0]! \n" // load 16 Ys
"vld1.8 {d1}, [%1]! \n" // load 8 Us
"vld1.8 {d3}, [%2]! \n" // load 8 Vs
"subs %4, %4, #16 \n" // 16 pixels
"vst4.u8 {d0, d1, d2, d3}, [%3]! \n" // Store 8 YUY2/16 pixels.
"bgt 1b \n"
: "+r"(src_y), // %0
"+r"(src_u), // %1
"+r"(src_v), // %2
"+r"(dst_yuy2), // %3
"+r"(width) // %4
:
: "cc", "memory", "d0", "d1", "d2", "d3"
);
}
void I422ToUYVYRow_NEON(const uint8* src_y,
const uint8* src_u,
const uint8* src_v,
uint8* dst_uyvy, int width) {
asm volatile (
".p2align 2 \n"
"1: \n"
"vld2.8 {d1, d3}, [%0]! \n" // load 16 Ys
"vld1.8 {d0}, [%1]! \n" // load 8 Us
"vld1.8 {d2}, [%2]! \n" // load 8 Vs
"subs %4, %4, #16 \n" // 16 pixels
"vst4.u8 {d0, d1, d2, d3}, [%3]! \n" // Store 8 UYVY/16 pixels.
"bgt 1b \n"
: "+r"(src_y), // %0
"+r"(src_u), // %1
"+r"(src_v), // %2
"+r"(dst_uyvy), // %3
"+r"(width) // %4
:
: "cc", "memory", "d0", "d1", "d2", "d3"
);
}
#endif // __ARM_NEON__
#ifdef __cplusplus
......
source/row_posix.cc
View file @ 9de8867a
......@@ -4109,6 +4109,77 @@ void ARGBToBayerRow_SSSE3(const uint8* src_argb, uint8* dst_bayer,
#endif
);
}
void I422ToYUY2Row_SSE2(const uint8* src_y,
const uint8* src_u,
const uint8* src_v,
uint8* dst_frame, int width) {
asm volatile (
"sub %1,%2 \n"
".p2align 4 \n"
"1: \n"
"movq (%1),%%xmm2 \n"
"movq (%1,%2,1),%%xmm3 \n"
"lea 0x8(%1),%1 \n"
"punpcklbw %%xmm3,%%xmm2 \n"
"movdqa (%0),%%xmm0 \n"
"lea 0x10(%0),%0 \n"
"movdqa %%xmm0,%%xmm1 \n"
"punpcklbw %%xmm2,%%xmm0 \n"
"punpckhbw %%xmm2,%%xmm1 \n"
"movdqa %%xmm0,(%3) \n"
"movdqa %%xmm1,0x10(%3) \n"
"lea 0x20(%3),%3 \n"
"sub $0x10,%4 \n"
"jg 1b \n"
: "+r"(src_y), // %0
"+r"(src_u), // %1
"+r"(src_v), // %2
"+r"(dst_frame), // %3
"+rm"(width) // %4
:
: "memory", "cc"
#if defined(__SSE2__)
, "xmm0", "xmm1", "xmm2", "xmm3"
#endif
);
}
void I422ToUYVYRow_SSE2(const uint8* src_y,
const uint8* src_u,
const uint8* src_v,
uint8* dst_frame, int width) {
asm volatile (
"sub %1,%2 \n"
".p2align 4 \n"
"1: \n"
"movq (%1),%%xmm2 \n"
"movq (%1,%2,1),%%xmm3 \n"
"lea 0x8(%1),%1 \n"
"punpcklbw %%xmm3,%%xmm2 \n"
"movdqa (%0),%%xmm0 \n"
"movdqa %%xmm2,%%xmm1 \n"
"lea 0x10(%0),%0 \n"
"punpcklbw %%xmm0,%%xmm1 \n"
"punpckhbw %%xmm0,%%xmm2 \n"
"movdqa %%xmm1,(%3) \n"
"movdqa %%xmm2,0x10(%3) \n"
"lea 0x20(%3),%3 \n"
"sub $0x10,%4 \n"
"jg 1b \n"
: "+r"(src_y), // %0
"+r"(src_u), // %1
"+r"(src_v), // %2
"+r"(dst_frame), // %3
"+rm"(width) // %4
:
: "memory", "cc"
#if defined(__SSE2__)
, "xmm0", "xmm1", "xmm2", "xmm3"
#endif
);
}
#endif // defined(__x86_64__) || defined(__i386__)
#ifdef __cplusplus
......
source/row_win.cc
View file @ 9de8867a
......@@ -4239,6 +4239,87 @@ void ARGBToBayerRow_SSSE3(const uint8* src_argb,
}
}
// YUY2 - Macro-pixel = 2 image pixels
// Y0U0Y1V0....Y2U2Y3V2...Y4U4Y5V4....
// UYVY - Macro-pixel = 2 image pixels
// U0Y0V0Y1
__declspec(naked) __declspec(align(16))
void I422ToYUY2Row_SSE2(const uint8* src_y,
const uint8* src_u,
const uint8* src_v,
uint8* dst_frame, int width) {
__asm {
push esi
push edi
mov eax, [esp + 8 + 4] // src_y
mov esi, [esp + 8 + 8] // src_u
mov edx, [esp + 8 + 12] // src_v
mov edi, [esp + 8 + 16] // dst_frame
mov ecx, [esp + 8 + 20] // width
sub edx, esi
align 16
convertloop:
movq xmm2, qword ptr [esi] // U
movq xmm3, qword ptr [esi + edx] // V
lea esi, [esi + 8]
punpcklbw xmm2, xmm3 // UV
movdqa xmm0, [eax] // Y
lea eax, [eax + 16]
movdqa xmm1, xmm0
punpcklbw xmm0, xmm2 // YUYV
punpckhbw xmm1, xmm2
movdqa [edi], xmm0
movdqa [edi + 16], xmm1
lea edi, [edi + 32]
sub ecx, 16
jg convertloop
pop edi
pop esi
ret
}
}
__declspec(naked) __declspec(align(16))
void I422ToUYVYRow_SSE2(const uint8* src_y,
const uint8* src_u,
const uint8* src_v,
uint8* dst_frame, int width) {
__asm {
push esi
push edi
mov eax, [esp + 8 + 4] // src_y
mov esi, [esp + 8 + 8] // src_u
mov edx, [esp + 8 + 12] // src_v
mov edi, [esp + 8 + 16] // dst_frame
mov ecx, [esp + 8 + 20] // width
sub edx, esi
align 16
convertloop:
movq xmm2, qword ptr [esi] // U
movq xmm3, qword ptr [esi + edx] // V
lea esi, [esi + 8]
punpcklbw xmm2, xmm3 // UV
movdqa xmm0, [eax] // Y
movdqa xmm1, xmm2
lea eax, [eax + 16]
punpcklbw xmm1, xmm0 // UYVY
punpckhbw xmm2, xmm0
movdqa [edi], xmm1
movdqa [edi + 16], xmm2
lea edi, [edi + 32]
sub ecx, 16
jg convertloop
pop edi
pop esi
ret
}
}
#endif // _M_IX86
#ifdef __cplusplus
......
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