Skip to content

L
libyuv
Commit 11c6d32a authored by fbarchard@google.com's avatar fbarchard@google.com
Browse files
Options

I420ToARGB1555, I420ToARGB4444, I420ToRGB565, ARGBToARGB1555, ARGBToARGB4444,… 

I420ToARGB1555, I420ToARGB4444, I420ToRGB565, ARGBToARGB1555, ARGBToARGB4444, and ARGBToRGB565 done with shared macro for conversion in 1 step NEON.
BUG=139
TEST=libyuv_unittest --gtest_filter=*I420To*RGB???*_*
Review URL: https://webrtc-codereview.appspot.com/928013

git-svn-id: http://libyuv.googlecode.com/svn/trunk@460 16f28f9a-4ce2-e073-06de-1de4eb20be90
parent af1aa56f
Hide whitespace changes
Inline Side-by-side
Showing with 127 additions and 72 deletions
source/convert_from.cc
View file @ 11c6d32a
......@@ -939,7 +939,7 @@ int I420ToARGB1555(const uint8* src_y, int src_stride_y,
uint8* rgb_buf,
int width) = I422ToARGB1555Row_C;
#if defined(HAS_I422TOARGB1555ROW_SSSE3)
if (TestCpuFlag(kCpuHasSSSE3) && width >= 8) {
if (TestCpuFlag(kCpuHasSSSE3) && width >= 8 && width <= kMaxStride * 4) {
I422ToARGB1555Row = I422ToARGB1555Row_Any_SSSE3;
if (IS_ALIGNED(width, 8)) {
I422ToARGB1555Row = I422ToARGB1555Row_SSSE3;
......@@ -990,7 +990,7 @@ int I420ToARGB4444(const uint8* src_y, int src_stride_y,
uint8* rgb_buf,
int width) = I422ToARGB4444Row_C;
#if defined(HAS_I422TOARGB4444ROW_SSSE3)
if (TestCpuFlag(kCpuHasSSSE3) && width >= 8) {
if (TestCpuFlag(kCpuHasSSSE3) && width >= 8 && width <= kMaxStride * 4) {
I422ToARGB4444Row = I422ToARGB4444Row_Any_SSSE3;
if (IS_ALIGNED(width, 8)) {
I422ToARGB4444Row = I422ToARGB4444Row_SSSE3;
......@@ -1040,7 +1040,11 @@ int I420ToRGB565(const uint8* src_y, int src_stride_y,
uint8* rgb_buf,
int width) = I422ToRGB565Row_C;
#if defined(HAS_I422TORGB565ROW_SSSE3)
if (TestCpuFlag(kCpuHasSSSE3) && width >= 8) {
if (TestCpuFlag(kCpuHasSSSE3) && width >= 8
#if defined(__x86_64__) || defined(__i386__)
&& width <= kMaxStride * 4
#endif
) {
I422ToRGB565Row = I422ToRGB565Row_Any_SSSE3;
if (IS_ALIGNED(width, 8)) {
I422ToRGB565Row = I422ToRGB565Row_SSSE3;
......
source/row_common.cc
View file @ 11c6d32a
......@@ -1275,7 +1275,10 @@ void I422ToUYVYRow_C(const uint8* src_y,
dst_frame[3] = src_y[0]; // duplicate last y
}
}
#if !defined(YUV_DISABLE_ASM)
// row_win.cc has asm version, but GCC uses 2 step wrapper. 5% slower.
// TODO(fbarchard): Handle width > kMaxStride here instead of calling code.
#if defined(__x86_64__) || defined(__i386__)
void I422ToRGB565Row_SSSE3(const uint8* y_buf,
const uint8* u_buf,
......@@ -1309,28 +1312,8 @@ void I422ToARGB4444Row_SSSE3(const uint8* y_buf,
ARGBToARGB4444Row_SSE2(row, rgb_buf, width);
}
#endif // defined(_M_IX86) || defined(__x86_64__) || defined(__i386__)
#if defined(__ARM_NEON__)
void I422ToARGB1555Row_NEON(const uint8* y_buf,
const uint8* u_buf,
const uint8* v_buf,
uint8* rgb_buf,
int width) {
SIMD_ALIGNED(uint8 row[kMaxStride]);
I422ToARGBRow_NEON(y_buf, u_buf, v_buf, row, width);
ARGBToARGB1555Row_NEON(row, rgb_buf, width);
}
void I422ToARGB4444Row_NEON(const uint8* y_buf,
const uint8* u_buf,
const uint8* v_buf,
uint8* rgb_buf,
int width) {
SIMD_ALIGNED(uint8 row[kMaxStride]);
I422ToARGBRow_NEON(y_buf, u_buf, v_buf, row, width);
ARGBToARGB4444Row_NEON(row, rgb_buf, width);
}
#endif // defined(__ARM_NEON__)
#endif // !defined(YUV_DISABLE_ASM)
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv
......
source/row_neon.cc
View file @ 11c6d32a
......@@ -274,6 +274,18 @@ void I422ToRAWRow_NEON(const uint8* src_y,
#endif // HAS_I422TORAWROW_NEON
#ifdef HAS_I422TORGB565ROW_NEON
#define ARGBTORGB565 \
"vshr.u8 d20, d20, #3 \n" /* B */ \
"vshr.u8 d21, d21, #2 \n" /* G */ \
"vshr.u8 d22, d22, #3 \n" /* R */ \
"vmovl.u8 q8, d20 \n" /* B */ \
"vmovl.u8 q9, d21 \n" /* G */ \
"vmovl.u8 q10, d22 \n" /* R */ \
"vshl.u16 q9, q9, #5 \n" /* G */ \
"vshl.u16 q10, q10, #11 \n" /* R */ \
"vorr q0, q8, q9 \n" /* BG */ \
"vorr q0, q0, q10 \n" /* BGR */
void I422ToRGB565Row_NEON(const uint8* src_y,
const uint8* src_u,
const uint8* src_v,
......@@ -290,16 +302,7 @@ void I422ToRGB565Row_NEON(const uint8* src_y,
READYUV422
YUV422TORGB
"subs %4, %4, #8 \n"
"vshr.u8 d20, d20, #3 \n" // B
"vshr.u8 d21, d21, #2 \n" // G
"vshr.u8 d22, d22, #3 \n" // R
"vmovl.u8 q8, d20 \n" // B
"vmovl.u8 q9, d21 \n" // G
"vmovl.u8 q10, d22 \n" // R
"vshl.u16 q9, q9, #5 \n" // G
"vshl.u16 q10, q10, #11 \n" // R
"vorr q0, q8, q9 \n" // BG
"vorr q0, q0, q10 \n" // BGR
ARGBTORGB565
"vst1.8 {q0}, [%3]! \n" // store 8 pixels RGB565.
"bgt 1b \n"
: "+r"(src_y), // %0
......@@ -315,6 +318,99 @@ void I422ToRGB565Row_NEON(const uint8* src_y,
}
#endif // HAS_I422TORGB565ROW_NEON
#ifdef HAS_I422TOARGB1555ROW_NEON
#define ARGBTOARGB1555 \
"vshr.u8 q10, q10, #3 \n" /* B */ \
"vshr.u8 d22, d22, #3 \n" /* R */ \
"vshr.u8 d23, d23, #7 \n" /* A */ \
"vmovl.u8 q8, d20 \n" /* B */ \
"vmovl.u8 q9, d21 \n" /* G */ \
"vmovl.u8 q10, d22 \n" /* R */ \
"vmovl.u8 q11, d23 \n" /* A */ \
"vshl.u16 q9, q9, #5 \n" /* G */ \
"vshl.u16 q10, q10, #10 \n" /* R */ \
"vshl.u16 q11, q11, #15 \n" /* A */ \
"vorr q0, q8, q9 \n" /* BG */ \
"vorr q1, q10, q11 \n" /* RA */ \
"vorr q0, q0, q1 \n" /* BGRA */
void I422ToARGB1555Row_NEON(const uint8* src_y,
const uint8* src_u,
const uint8* src_v,
uint8* dst_argb1555,
int width) {
asm volatile (
"vld1.u8 {d24}, [%5] \n"
"vld1.u8 {d25}, [%6] \n"
"vmov.u8 d26, #128 \n"
"vmov.u16 q14, #74 \n"
"vmov.u16 q15, #16 \n"
".p2align 2 \n"
"1: \n"
READYUV422
YUV422TORGB
"subs %4, %4, #8 \n"
"vmov.u8 d23, #255 \n"
ARGBTOARGB1555
"vst1.8 {q0}, [%3]! \n" // store 8 pixels ARGB1555.
"bgt 1b \n"
: "+r"(src_y), // %0
"+r"(src_u), // %1
"+r"(src_v), // %2
"+r"(dst_argb1555), // %3
"+r"(width) // %4
: "r"(&kUVToRB), // %5
"r"(&kUVToG) // %6
: "cc", "memory", "q0", "q1", "q2", "q3",
"q8", "q9", "q10", "q11", "q12", "q13", "q14", "q15"
);
}
#endif // HAS_I422TOARGB1555ROW_NEON
#ifdef HAS_I422TOARGB4444ROW_NEON
#define ARGBTOARGB4444 \
"vshr.u8 d20, d20, #4 \n" /* B */ \
"vbic.32 d21, d21, d4 \n" /* G */ \
"vshr.u8 d22, d22, #4 \n" /* R */ \
"vbic.32 d23, d23, d4 \n" /* A */ \
"vorr d0, d20, d21 \n" /* BG */ \
"vorr d1, d22, d23 \n" /* RA */ \
"vzip.u8 d0, d1 \n" /* BGRA */
void I422ToARGB4444Row_NEON(const uint8* src_y,
const uint8* src_u,
const uint8* src_v,
uint8* dst_argb4444,
int width) {
asm volatile (
"vld1.u8 {d24}, [%5] \n"
"vld1.u8 {d25}, [%6] \n"
"vmov.u8 d26, #128 \n"
"vmov.u16 q14, #74 \n"
"vmov.u16 q15, #16 \n"
"vmov.u8 d4, #0x0f \n" // bits to clear with vbic.
".p2align 2 \n"
"1: \n"
READYUV422
YUV422TORGB
"subs %4, %4, #8 \n"
"vmov.u8 d23, #255 \n"
ARGBTOARGB4444
"vst1.8 {q0}, [%3]! \n" // store 8 pixels ARGB4444.
"bgt 1b \n"
: "+r"(src_y), // %0
"+r"(src_u), // %1
"+r"(src_v), // %2
"+r"(dst_argb4444), // %3
"+r"(width) // %4
: "r"(&kUVToRB), // %5
"r"(&kUVToG) // %6
: "cc", "memory", "q0", "q1", "q2", "q3",
"q8", "q9", "q10", "q11", "q12", "q13", "q14", "q15"
);
}
#endif // HAS_I422TOARGB4444ROW_NEON
#ifdef HAS_NV12TOARGBROW_NEON
void NV12ToARGBRow_NEON(const uint8* src_y,
const uint8* src_uv,
......@@ -1020,25 +1116,16 @@ void ARGBToRGB565Row_NEON(const uint8* src_argb, uint8* dst_rgb565, int pix) {
asm volatile (
".p2align 2 \n"
"1: \n"
"vld4.8 {d0, d1, d2, d3}, [%0]! \n" // load 8 pixels of ARGB.
"vld4.8 {d20, d21, d22, d23}, [%0]! \n" // load 8 pixels of ARGB.
"subs %2, %2, #8 \n" // 8 processed per loop.
"vshr.u8 d0, d0, #3 \n" // B
"vshr.u8 d1, d1, #2 \n" // G
"vshr.u8 d2, d2, #3 \n" // R
"vmovl.u8 q8, d0 \n" // B
"vmovl.u8 q9, d1 \n" // G
"vmovl.u8 q10, d2 \n" // R
"vshl.u16 q9, q9, #5 \n" // G
"vshl.u16 q10, q10, #11 \n" // R
"vorr q0, q8, q9 \n" // BG
"vorr q0, q0, q10 \n" // BGR
ARGBTORGB565
"vst1.8 {q0}, [%1]! \n" // store 8 pixels RGB565.
"bgt 1b \n"
: "+r"(src_argb), // %0
"+r"(dst_rgb565), // %1
"+r"(pix) // %2
:
: "memory", "cc", "d0", "d1", "d2", "d3", "q8", "q9", "q10"
: "memory", "cc", "q0", "q8", "q9", "q10", "q11"
);
}
#endif // HAS_ARGBTORGB565ROW_NEON
......@@ -1049,29 +1136,16 @@ void ARGBToARGB1555Row_NEON(const uint8* src_argb, uint8* dst_argb1555,
asm volatile (
".p2align 2 \n"
"1: \n"
"vld4.8 {d0, d1, d2, d3}, [%0]! \n" // load 8 pixels of ARGB.
"vld4.8 {d20, d21, d22, d23}, [%0]! \n" // load 8 pixels of ARGB.
"subs %2, %2, #8 \n" // 8 processed per loop.
"vshr.u8 d0, d0, #3 \n" // B
"vshr.u8 d1, d1, #3 \n" // G
"vshr.u8 d2, d2, #3 \n" // R
"vshr.u8 d3, d3, #7 \n" // A
"vmovl.u8 q8, d0 \n" // B
"vmovl.u8 q9, d1 \n" // G
"vmovl.u8 q10, d2 \n" // R
"vmovl.u8 q11, d3 \n" // A
"vshl.u16 q9, q9, #5 \n" // G
"vshl.u16 q10, q10, #10 \n" // R
"vshl.u16 q11, q11, #15 \n" // A
"vorr q0, q8, q9 \n" // BG
"vorr q1, q10, q11 \n" // RA
"vorr q0, q0, q1 \n" // BGRA
ARGBTOARGB1555
"vst1.8 {q0}, [%1]! \n" // store 8 pixels ARGB1555.
"bgt 1b \n"
: "+r"(src_argb), // %0
"+r"(dst_argb1555), // %1
"+r"(pix) // %2
:
: "memory", "cc", "d0", "d1", "d2", "d3", "q8", "q9", "q10", "q11"
: "memory", "cc", "q0", "q8", "q9", "q10", "q11"
);
}
#endif // HAS_ARGBTOARGB1555ROW_NEON
......@@ -1083,22 +1157,16 @@ void ARGBToARGB4444Row_NEON(const uint8* src_argb, uint8* dst_argb4444,
"vmov.u8 d4, #0x0f \n" // bits to clear with vbic.
".p2align 2 \n"
"1: \n"
"vld4.8 {d0, d1, d2, d3}, [%0]! \n" // load 8 pixels of ARGB.
"vld4.8 {d20, d21, d22, d23}, [%0]! \n" // load 8 pixels of ARGB.
"subs %2, %2, #8 \n" // 8 processed per loop.
"vshr.u8 d0, d0, #4 \n" // B
"vbic.32 d1, d1, d4 \n" // G
"vshr.u8 d2, d2, #4 \n" // R
"vbic.32 d3, d3, d4 \n" // A
"vorr d0, d0, d1 \n" // BG
"vorr d1, d2, d3 \n" // RA
"vzip.u8 d0, d1 \n" // BGRA weaved together.
ARGBTOARGB4444
"vst1.8 {q0}, [%1]! \n" // store 8 pixels ARGB4444.
"bgt 1b \n"
: "+r"(src_argb), // %0
"+r"(dst_argb4444), // %1
"+r"(pix) // %2
:
: "memory", "cc", "d0", "d1", "d2", "d3", "d4" // Clobber List
: "memory", "cc", "q0", "q8", "q9", "q10", "q11"
);
}
#endif // HAS_ARGBTOARGB4444ROW_NEON
......
unit_test/version_test.cc
View file @ 11c6d32a
......@@ -32,7 +32,7 @@ TEST_F(libyuvTest, TestVersion) {
}
int svn_revision = atoi(ver); // NOLINT
printf("LIBYUV_SVNREVISION %d\n", svn_revision);
EXPECT_NEAR(LIBYUV_VERSION, svn_revision, 3); // Allow version to be close.
EXPECT_NEAR(LIBYUV_VERSION, svn_revision, 20); // Allow version to be close.
if (LIBYUV_VERSION != svn_revision) {
printf("WARNING - Versions do not match.\n");
}
......
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