coffee/libyuv
1
0
Fork 0
mirror of https://chromium.googlesource.com/libyuv/libyuv synced 2025-12-06 16:56:55 +08:00
Code Issues Packages Projects Releases Wiki Activity

Add AYUVToNV12 and NV21ToNV12

Browse Source 
BUG=libyuv:832
TESTED=out/Release/libyuv_unittest --gtest_filter=*ToNV12* --libyuv_width=1280 --libyuv_height=720 --libyuv_repeat=1000 --libyuv_flags=-1 --libyuv_cpu_info=-1

R=rrwinterton@gmail.com

Change-Id: Id03b4613211fb6a6e163d10daa7c692fe31e36d8
Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/1560080
Commit-Queue: Frank Barchard <fbarchard@chromium.org>
Reviewed-by: richard winterton <rrwinterton@gmail.com>
Reviewed-by: Frank Barchard <fbarchard@chromium.org>
This commit is contained in:
Frank Barchard 2019-04-12 10:20:44 -07:00 committed by Commit Bot
parent 4bd08cbc0e
commit 413a8d8041
20 changed files with 710 additions and 281 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
README.chromium
View File

@ -1,6 +1,6 @@
Name: libyuv
URL: http://code.google.com/p/libyuv/
Version: 1725
Version: 1727
License: BSD
License File: LICENSE

11
include/libyuv/convert.h
View File

@ -226,6 +226,17 @@ int UYVYToI420(const uint8_t* src_uyvy,
int width,
int height);
// Convert AYUV to NV12.
LIBYUV_API
int AYUVToNV12(const uint8_t* src_ayuv,
int src_stride_ayuv,
uint8_t* dst_y,
int dst_stride_y,
uint8_t* dst_uv,
int dst_stride_uv,
int width,
int height);
// Convert AYUV to NV21.
LIBYUV_API
int AYUVToNV21(const uint8_t* src_ayuv,

13
include/libyuv/planar_functions.h
View File

@ -224,6 +224,19 @@ int UYVYToNV12(const uint8_t* src_uyvy,
int width,
int height);
// Convert NV21 to NV12.
LIBYUV_API
int NV21ToNV12(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_vu,
int src_stride_vu,
uint8_t* dst_y,
int dst_stride_y,
uint8_t* dst_uv,
int dst_stride_uv,
int width,
int height);
LIBYUV_API
int YUY2ToY(const uint8_t* src_yuy2,
int src_stride_yuy2,

28
include/libyuv/row.h
View File

@ -332,6 +332,7 @@ extern "C" {
#define HAS_ARGBTOUVROW_NEON
#define HAS_ARGBTOYJROW_NEON
#define HAS_ARGBTOYROW_NEON
#define HAS_AYUVTOUVROW_NEON
#define HAS_AYUVTOVUROW_NEON
#define HAS_AYUVTOYROW_NEON
#define HAS_BGRATOUVROW_NEON
@ -375,6 +376,7 @@ extern "C" {
#define HAS_SETROW_NEON
#define HAS_SPLITRGBROW_NEON
#define HAS_SPLITUVROW_NEON
#define HAS_UVToVUROW_NEON
#define HAS_UYVYTOARGBROW_NEON
#define HAS_UYVYTOUV422ROW_NEON
#define HAS_UYVYTOUVROW_NEON
@ -3370,17 +3372,34 @@ void UYVYToUV422Row_Any_MMI(const uint8_t* src_ptr,
uint8_t* dst_u,
uint8_t* dst_v,
int width);
void UVToVURow_C(const uint8_t* src_uv, uint8_t* dst_vu, int width);
void UVToVURow_NEON(const uint8_t* src_uv, uint8_t* dst_vu, int width);
void UVToVURow_Any_NEON(const uint8_t* src_uv, uint8_t* dst_vu, int width);
void AYUVToYRow_C(const uint8_t* src_ayuv, uint8_t* dst_y, int width);
void AYUVToVURow_C(const uint8_t* src_ayuv, int stride_ayuv,
void AYUVToUVRow_C(const uint8_t* src_ayuv,
int stride_ayuv,
uint8_t* dst_uv,
int width);
void AYUVToVURow_C(const uint8_t* src_ayuv,
int stride_ayuv,
uint8_t* dst_vu,
int width);
void AYUVToYRow_NEON(const uint8_t* src_ayuv, uint8_t* dst_y, int width);
void AYUVToVURow_NEON(const uint8_t* src_ayuv, int stride_ayuv,
void AYUVToUVRow_NEON(const uint8_t* src_ayuv,
int stride_ayuv,
uint8_t* dst_uv,
int width);
void AYUVToVURow_NEON(const uint8_t* src_ayuv,
int stride_ayuv,
uint8_t* dst_vu,
int width);
void AYUVToYRow_Any_NEON(const uint8_t* src_ayuv, uint8_t* dst_y, int width);
void AYUVToVURow_Any_NEON(const uint8_t* src_ayuv, int stride_ayuv,
void AYUVToUVRow_Any_NEON(const uint8_t* src_ayuv,
int stride_ayuv,
uint8_t* dst_uv,
int width);
void AYUVToVURow_Any_NEON(const uint8_t* src_ayuv,
int stride_ayuv,
uint8_t* dst_vu,
int width);
@ -4010,7 +4029,6 @@ void FloatDivToByteRow_NEON(const float* src_weights,
uint8_t* dst_mask,
int width);
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv

2
include/libyuv/version.h
View File

@ -11,6 +11,6 @@
#ifndef INCLUDE_LIBYUV_VERSION_H_
#define INCLUDE_LIBYUV_VERSION_H_
#define LIBYUV_VERSION 1725
#define LIBYUV_VERSION 1727
#endif // INCLUDE_LIBYUV_VERSION_H_

73
source/convert.cc
View File

@ -880,6 +880,75 @@ int UYVYToI420(const uint8_t* src_uyvy,
return 0;
}
// Convert AYUV to NV12.
LIBYUV_API
int AYUVToNV12(const uint8_t* src_ayuv,
int src_stride_ayuv,
uint8_t* dst_y,
int dst_stride_y,
uint8_t* dst_uv,
int dst_stride_uv,
int width,
int height) {
int y;
void (*AYUVToUVRow)(const uint8_t* src_ayuv, int src_stride_ayuv,
uint8_t* dst_uv, int width) = AYUVToUVRow_C;
void (*AYUVToYRow)(const uint8_t* src_ayuv, uint8_t* dst_y, int width) =
AYUVToYRow_C;
// Negative height means invert the image.
if (height < 0) {
height = -height;
src_ayuv = src_ayuv + (height - 1) * src_stride_ayuv;
src_stride_ayuv = -src_stride_ayuv;
}
// place holders for future intel code
#if defined(HAS_AYUVTOYROW_SSE2)
if (TestCpuFlag(kCpuHasSSE2)) {
AYUVToUVRow = AYUVToUVRow_Any_SSE2;
AYUVToYRow = AYUVToYRow_Any_SSE2;
if (IS_ALIGNED(width, 16)) {
AYUVToUVRow = AYUVToUVRow_SSE2;
AYUVToYRow = AYUVToYRow_SSE2;
}
}
#endif
#if defined(HAS_AYUVTOYROW_AVX2)
if (TestCpuFlag(kCpuHasAVX2)) {
AYUVToUVRow = AYUVToUVRow_Any_AVX2;
AYUVToYRow = AYUVToYRow_Any_AVX2;
if (IS_ALIGNED(width, 32)) {
AYUVToUVRow = AYUVToUVRow_AVX2;
AYUVToYRow = AYUVToYRow_AVX2;
}
}
#endif
#if defined(HAS_AYUVTOYROW_NEON)
if (TestCpuFlag(kCpuHasNEON)) {
AYUVToYRow = AYUVToYRow_Any_NEON;
AYUVToUVRow = AYUVToUVRow_Any_NEON;
if (IS_ALIGNED(width, 16)) {
AYUVToYRow = AYUVToYRow_NEON;
AYUVToUVRow = AYUVToUVRow_NEON;
}
}
#endif
for (y = 0; y < height - 1; y += 2) {
AYUVToUVRow(src_ayuv, src_stride_ayuv, dst_uv, width);
AYUVToYRow(src_ayuv, dst_y, width);
AYUVToYRow(src_ayuv + src_stride_ayuv, dst_y + dst_stride_y, width);
src_ayuv += src_stride_ayuv * 2;
dst_y += dst_stride_y * 2;
dst_uv += dst_stride_uv;
}
if (height & 1) {
AYUVToUVRow(src_ayuv, 0, dst_uv, width);
AYUVToYRow(src_ayuv, dst_y, width);
}
return 0;
}
// Convert AYUV to NV21.
LIBYUV_API
int AYUVToNV21(const uint8_t* src_ayuv,
@ -892,8 +961,7 @@ int AYUVToNV21(const uint8_t* src_ayuv,
int height) {
int y;
void (*AYUVToVURow)(const uint8_t* src_ayuv, int src_stride_ayuv,
uint8_t* dst_vu, int width) =
AYUVToVURow_C;
uint8_t* dst_vu, int width) = AYUVToVURow_C;
void (*AYUVToYRow)(const uint8_t* src_ayuv, uint8_t* dst_y, int width) =
AYUVToYRow_C;
// Negative height means invert the image.
@ -2235,7 +2303,6 @@ int Android420ToI420(const uint8_t* src_y,
return 0;
}
#ifdef __cplusplus
} // extern "C"
} // namespace libyuv

6
source/convert_argb.cc
View File

@ -2008,10 +2008,8 @@ int NV21ToYUV24(const uint8_t* src_y,
int width,
int height) {
int y;
void (*NV21ToYUV24Row)(const uint8_t* src_y,
const uint8_t* src_vu,
uint8_t* dst_yuv24,
int width) = NV21ToYUV24Row_C;
void (*NV21ToYUV24Row)(const uint8_t* src_y, const uint8_t* src_vu,
uint8_t* dst_yuv24, int width) = NV21ToYUV24Row_C;
if (!src_y || !src_vu || !dst_yuv24 || width <= 0 || height == 0) {
return -1;
}

58
source/planar_functions.cc
View File

@ -440,7 +440,6 @@ void MergeUVPlane(const uint8_t* src_u,
int y;
void (*MergeUVRow)(const uint8_t* src_u, const uint8_t* src_v,
uint8_t* dst_uv, int width) = MergeUVRow_C;
// Coalesce rows.
// Negative height means invert the image.
if (height < 0) {
height = -height;
@ -504,6 +503,63 @@ void MergeUVPlane(const uint8_t* src_u,
}
}
// Convert NV21 to NV12.
LIBYUV_API
int NV21ToNV12(const uint8_t* src_y,
int src_stride_y,
const uint8_t* src_vu,
int src_stride_vu,
uint8_t* dst_y,
int dst_stride_y,
uint8_t* dst_uv,
int dst_stride_uv,
int width,
int height) {
int y;
void (*UVToVURow)(const uint8_t* src_uv, uint8_t* dst_vu, int width) =
UVToVURow_C;
int halfwidth = (width + 1) >> 1;
int halfheight = (height + 1) >> 1;
if (!src_vu || !dst_uv || width <= 0 || height == 0) {
return -1;
}
// Negative height means invert the image.
if (height < 0) {
height = -height;
halfheight = (height + 1) >> 1;
src_y = src_y + (height - 1) * src_stride_y;
src_vu = src_vu + (halfheight - 1) * src_stride_vu;
src_stride_y = -src_stride_y;
src_stride_vu = -src_stride_vu;
}
// Coalesce rows.
if (src_stride_vu == halfwidth * 2 && dst_stride_uv == halfwidth * 2) {
halfwidth *= halfheight;
halfheight = 1;
src_stride_vu = dst_stride_uv = 0;
}
#if defined(HAS_UVToVUROW_NEON)
if (TestCpuFlag(kCpuHasNEON)) {
UVToVURow = UVToVURow_Any_NEON;
if (IS_ALIGNED(halfwidth, 16)) {
UVToVURow = UVToVURow_NEON;
}
}
#endif
if (dst_y) {
CopyPlane(src_y, src_stride_y, dst_y, dst_stride_y, width, height);
}
for (y = 0; y < halfheight; ++y) {
UVToVURow(src_vu, dst_uv, halfwidth);
src_vu += src_stride_vu;
dst_uv += dst_stride_uv;
}
return 0;
}
// Support function for NV12 etc RGB channels.
// Width and height are plane sizes (typically half pixel width).
LIBYUV_API

5
source/row_any.cc
View File

@ -707,10 +707,12 @@ ANY11(UYVYToYRow_Any_MSA, UYVYToYRow_MSA, 1, 4, 1, 31)
#ifdef HAS_UYVYTOYROW_MMI
ANY11(UYVYToYRow_Any_MMI, UYVYToYRow_MMI, 1, 4, 1, 15)
#endif
#ifdef HAS_AYUVTOYROW_NEON
ANY11(AYUVToYRow_Any_NEON, AYUVToYRow_NEON, 0, 4, 1, 15)
#endif
#ifdef HAS_AYUVTOYROW_NEON
ANY11(UVToVURow_Any_NEON, UVToVURow_NEON, 0, 2, 2, 15)
#endif
#ifdef HAS_RGB24TOARGBROW_NEON
ANY11(RGB24ToARGBRow_Any_NEON, RGB24ToARGBRow_NEON, 0, 3, 4, 7)
#endif
@ -1416,6 +1418,7 @@ ANY12S(UYVYToUVRow_Any_MMI, UYVYToUVRow_MMI, 1, 4, 15)
}
#ifdef HAS_AYUVTOVUROW_NEON
ANY11S(AYUVToUVRow_Any_NEON, AYUVToUVRow_NEON, 0, 4, 15)
ANY11S(AYUVToVURow_Any_NEON, AYUVToVURow_NEON, 0, 4, 15)
#endif
#undef ANY11S

56
source/row_common.cc
View File

@ -3236,7 +3236,6 @@ void NV21ToYUV24Row_C(const uint8_t* src_y,
const uint8_t* src_vu,
uint8_t* dst_yuv24,
int width) {
int x;
for (x = 0; x < width - 1; x += 2) {
dst_yuv24[0] = src_vu[0]; // V
@ -3256,6 +3255,33 @@ void NV21ToYUV24Row_C(const uint8_t* src_y,
}
}
// Filter 2 rows of AYUV UV's (444) into UV (420).
void AYUVToUVRow_C(const uint8_t* src_ayuv,
int src_stride_ayuv,
uint8_t* dst_uv,
int width) {
// Output a row of UV values, filtering 2x2 rows of AYUV.
int x;
for (x = 0; x < width; x += 2) {
dst_uv[0] = (src_ayuv[1] + src_ayuv[5] + src_ayuv[src_stride_ayuv + 1] +
src_ayuv[src_stride_ayuv + 5] + 2) >>
2;
dst_uv[1] = (src_ayuv[0] + src_ayuv[4] + src_ayuv[src_stride_ayuv + 0] +
src_ayuv[src_stride_ayuv + 4] + 2) >>
2;
src_ayuv += 8;
dst_uv += 2;
}
if (width & 1) {
dst_uv[0] = (src_ayuv[0] + src_ayuv[0] + src_ayuv[src_stride_ayuv + 0] +
src_ayuv[src_stride_ayuv + 0] + 2) >>
2;
dst_uv[1] = (src_ayuv[1] + src_ayuv[1] + src_ayuv[src_stride_ayuv + 1] +
src_ayuv[src_stride_ayuv + 1] + 2) >>
2;
}
}
// Filter 2 rows of AYUV UV's (444) into VU (420).
void AYUVToVURow_C(const uint8_t* src_ayuv,
int src_stride_ayuv,
@ -3264,14 +3290,22 @@ void AYUVToVURow_C(const uint8_t* src_ayuv,
// Output a row of VU values, filtering 2x2 rows of AYUV.
int x;
for (x = 0; x < width; x += 2) {
dst_vu[0] = (src_ayuv[0] + src_ayuv[4] + src_ayuv[src_stride_ayuv + 0] + src_ayuv[src_stride_ayuv + 4] + 2) >> 2;
dst_vu[1] = (src_ayuv[1] + src_ayuv[5] + src_ayuv[src_stride_ayuv + 1] + src_ayuv[src_stride_ayuv + 5] + 2) >> 2;
dst_vu[0] = (src_ayuv[0] + src_ayuv[4] + src_ayuv[src_stride_ayuv + 0] +
src_ayuv[src_stride_ayuv + 4] + 2) >>
2;
dst_vu[1] = (src_ayuv[1] + src_ayuv[5] + src_ayuv[src_stride_ayuv + 1] +
src_ayuv[src_stride_ayuv + 5] + 2) >>
2;
src_ayuv += 8;
dst_vu += 2;
}
if (width & 1) {
dst_vu[0] = (src_ayuv[0] + src_ayuv[0] + src_ayuv[src_stride_ayuv + 0] + src_ayuv[src_stride_ayuv + 0] + 2) >> 2;
dst_vu[1] = (src_ayuv[1] + src_ayuv[1] + src_ayuv[src_stride_ayuv + 1] + src_ayuv[src_stride_ayuv + 1] + 2) >> 2;
dst_vu[0] = (src_ayuv[0] + src_ayuv[0] + src_ayuv[src_stride_ayuv + 0] +
src_ayuv[src_stride_ayuv + 0] + 2) >>
2;
dst_vu[1] = (src_ayuv[1] + src_ayuv[1] + src_ayuv[src_stride_ayuv + 1] +
src_ayuv[src_stride_ayuv + 1] + 2) >>
2;
}
}
@ -3285,6 +3319,18 @@ void AYUVToYRow_C(const uint8_t* src_ayuv, uint8_t* dst_y, int width) {
}
}
void UVToVURow_C(const uint8_t* src_uv, uint8_t* dst_vu, int width) {
int x;
for (x = 0; x < width; ++x) {
uint8_t u = src_uv[0];
uint8_t v = src_uv[1];
dst_vu[0] = v;
dst_vu[1] = u;
src_uv += 2;
dst_vu += 2;
}
}
// divide values by weights and provide mask to indicate weight of 0.
void FloatDivToByteRow_C(const float* src_weights,
const float* src_values,

26
source/row_gcc.cc
View File

@ -6669,7 +6669,6 @@ void ARGBLumaColorTableRow_SSSE3(const uint8_t* src_argb,
}
#endif // HAS_ARGBLUMACOLORTABLEROW_SSSE3
#ifdef HAS_NV21TOYUV24ROW_AVX2
// begin NV21ToYUV24Row_C avx2 constants
@ -6723,7 +6722,6 @@ void NV21ToYUV24Row_AVX2(const uint8_t* src_y,
const uint8_t* src_vu,
uint8_t* dst_yuv24,
int width) {
uint8_t* src_y_ptr;
uint64_t src_offset = 0;
uint64_t width64;
@ -6743,10 +6741,13 @@ void NV21ToYUV24Row_AVX2(const uint8_t* src_y,
"vmovdqu 1(%1,%4), %%ymm4 \n" // src_uv+1
"vmovdqu (%1), %%ymm5 \n" // src_uv
"vpshufb %8, %%ymm3, %%ymm13 \n" // y, kSHUF0 for shuf
"vpshufb %9, %%ymm4, %%ymm14 \n" //uv+1, kSHUF1 for shuf
"vpshufb %10, %%ymm5, %%ymm15 \n" //uv, kSHUF2 for shuf
"vpshufb %9, %%ymm4, %%ymm14 \n" // uv+1, kSHUF1 for
// shuf
"vpshufb %10, %%ymm5, %%ymm15 \n" // uv, kSHUF2 for
// shuf
"vpshufb %11, %%ymm3, %%ymm3 \n" // y kSHUF3 for shuf
"vpshufb %12, %%ymm4, %%ymm4 \n" //uv+1 kSHUF4 for shuf
"vpshufb %12, %%ymm4, %%ymm4 \n" // uv+1 kSHUF4 for
// shuf
"vpblendvb %%ymm0, %%ymm14, %%ymm13, %%ymm12 \n" // blend 0
"vpblendvb %%ymm0, %%ymm13, %%ymm14, %%ymm14 \n" // blend 0
"vpblendvb %%ymm2, %%ymm15, %%ymm12, %%ymm12 \n" // blend 2
@ -6755,16 +6756,20 @@ void NV21ToYUV24Row_AVX2(const uint8_t* src_y,
"vpor %%ymm4, %%ymm3, %%ymm5 \n" // get results
"vmovdqu %%ymm12, 0x20(%2) \n" // store dst_yuv+20h
"vpor %%ymm15, %%ymm5, %%ymm3 \n" // get results
"add $0x20, %4 \n" //add to src buffer ptr
"add $0x20, %4 \n" // add to src buffer
// ptr
"vinserti128 $0x1, %%xmm3, %%ymm13, %%ymm4 \n" // insert
"vperm2i128 $0x31, %%ymm13, %%ymm3, %%ymm5 \n" // insert
"vmovdqu %%ymm4, (%2) \n" // store dst_yuv
"vmovdqu %%ymm5, 0x40(%2) \n" // store dst_yuv+40h
"add $0x60,%2 \n" //add to dst buffer ptr
// "cmp %3, %4 \n" //(width64 - 32 bytes) and src_offset
"add $0x60,%2 \n" // add to dst buffer
// ptr
// "cmp %3, %4 \n" //(width64 -
// 32 bytes) and src_offset
"sub $0x20,%3 \n" // 32 pixels per loop
"jg 1b \n"
"vzeroupper \n" //sse-avx2 transistions
"vzeroupper \n" // sse-avx2
// transistions
: "+r"(src_y), //%0
"+r"(src_vu), //%1
@ -6780,7 +6785,8 @@ void NV21ToYUV24Row_AVX2(const uint8_t* src_y,
"m"(kSHUF3), //%11
"m"(kSHUF4), //%12
"m"(kSHUF5) //%13
: "memory", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm12", "xmm13", "xmm14", "xmm15");
: "memory", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm12",
"xmm13", "xmm14", "xmm15");
}
#endif // HAS_NV21TOYUV24ROW_AVX2

60
source/row_neon.cc
View File

@ -2710,6 +2710,37 @@ void NV21ToYUV24Row_NEON(const uint8_t* src_y,
: "cc", "memory", "q0", "q1", "q2");
}
void AYUVToUVRow_NEON(const uint8_t* src_ayuv,
int src_stride_ayuv,
uint8_t* dst_uv,
int width) {
asm volatile(
"add %1, %0, %1 \n" // src_stride + src_AYUV
"1: \n"
"vld4.8 {d0, d2, d4, d6}, [%0]! \n" // load 8 AYUV pixels.
"vld4.8 {d1, d3, d5, d7}, [%0]! \n" // load next 8 AYUV
// pixels.
"vpaddl.u8 q0, q0 \n" // V 16 bytes -> 8 shorts.
"vpaddl.u8 q1, q1 \n" // U 16 bytes -> 8 shorts.
"vld4.8 {d8, d10, d12, d14}, [%1]! \n" // load 8 more AYUV
// pixels.
"vld4.8 {d9, d11, d13, d15}, [%1]! \n" // load last 8 AYUV
// pixels.
"vpadal.u8 q0, q4 \n" // B 16 bytes -> 8 shorts.
"vpadal.u8 q1, q5 \n" // G 16 bytes -> 8 shorts.
"vqrshrun.s16 d1, q0, #2 \n" // 2x2 average
"vqrshrun.s16 d0, q1, #2 \n"
"subs %3, %3, #16 \n" // 16 processed per loop.
"vst2.8 {d0, d1}, [%2]! \n" // store 8 pixels UV.
"bgt 1b \n"
: "+r"(src_ayuv), // %0
"+r"(src_stride_ayuv), // %1
"+r"(dst_uv), // %2
"+r"(width) // %3
:
: "cc", "memory", "q0", "q1", "q2", "q3", "q4", "q5", "q6", "q7");
}
void AYUVToVURow_NEON(const uint8_t* src_ayuv,
int src_stride_ayuv,
uint8_t* dst_vu,
@ -2718,11 +2749,14 @@ void AYUVToVURow_NEON(const uint8_t* src_ayuv,
"add %1, %0, %1 \n" // src_stride + src_AYUV
"1: \n"
"vld4.8 {d0, d2, d4, d6}, [%0]! \n" // load 8 AYUV pixels.
"vld4.8 {d1, d3, d5, d7}, [%0]! \n" // load next 8 AYUV pixels.
"vld4.8 {d1, d3, d5, d7}, [%0]! \n" // load next 8 AYUV
// pixels.
"vpaddl.u8 q0, q0 \n" // V 16 bytes -> 8 shorts.
"vpaddl.u8 q1, q1 \n" // U 16 bytes -> 8 shorts.
"vld4.8 {d8, d10, d12, d14}, [%1]! \n" // load 8 more AYUV pixels.
"vld4.8 {d9, d11, d13, d15}, [%1]! \n" // load last 8 AYUV pixels.
"vld4.8 {d8, d10, d12, d14}, [%1]! \n" // load 8 more AYUV
// pixels.
"vld4.8 {d9, d11, d13, d15}, [%1]! \n" // load last 8 AYUV
// pixels.
"vpadal.u8 q0, q4 \n" // B 16 bytes -> 8 shorts.
"vpadal.u8 q1, q5 \n" // G 16 bytes -> 8 shorts.
"vqrshrun.s16 d0, q0, #2 \n" // 2x2 average
@ -2735,8 +2769,7 @@ void AYUVToVURow_NEON(const uint8_t* src_ayuv,
"+r"(dst_vu), // %2
"+r"(width) // %3
:
: "cc", "memory", "q0", "q1", "q2", "q3", "q4", "q5", "q6", "q7"
);
: "cc", "memory", "q0", "q1", "q2", "q3", "q4", "q5", "q6", "q7");
}
// Copy row of AYUV Y's into Y.
@ -2756,6 +2789,23 @@ void AYUVToYRow_NEON(const uint8_t* src_ayuv, uint8_t* dst_y, int width) {
: "cc", "memory", "q0", "q1", "q2", "q3");
}
// Convert biplanar UV channel of NV12 to NV21
void UVToVURow_NEON(const uint8_t* src_uv, uint8_t* dst_vu, int width) {
asm volatile(
"1: \n"
"vld2.8 {d0, d2}, [%0]! \n" // load 16 UV values
"vld2.8 {d1, d3}, [%0]! \n"
"vorr.u8 q2, q0, q0 \n" // move U after V
"subs %2, %2, #16 \n" // 16 pixels per loop
"vst2.8 {q1, q2}, [%1]! \n" // store 16 VU pixels
"bgt 1b \n"
: "+r"(src_uv), // %0
"+r"(dst_vu), // %1
"+r"(width) // %2
:
: "cc", "memory", "q0", "q1", "q2");
}
#endif // !defined(LIBYUV_DISABLE_NEON) && defined(__ARM_NEON__)..
#ifdef __cplusplus

53
source/row_neon64.cc
View File

@ -2898,6 +2898,34 @@ void NV21ToYUV24Row_NEON(const uint8_t* src_y,
: "cc", "memory", "v0", "v1", "v2");
}
void AYUVToUVRow_NEON(const uint8_t* src_ayuv,
int src_stride_ayuv,
uint8_t* dst_uv,
int width) {
const uint8_t* src_ayuv_1 = src_ayuv + src_stride_ayuv;
asm volatile(
"1: \n"
"ld4 {v0.16b,v1.16b,v2.16b,v3.16b}, [%0], #64 \n" // load 16
// pixels.
"uaddlp v0.8h, v0.16b \n" // V 16 bytes -> 8 shorts.
"uaddlp v1.8h, v1.16b \n" // U 16 bytes -> 8 shorts.
"ld4 {v4.16b,v5.16b,v6.16b,v7.16b}, [%1], #64 \n" // load next 16
"uadalp v0.8h, v4.16b \n" // V 16 bytes -> 8 shorts.
"uadalp v1.8h, v5.16b \n" // U 16 bytes -> 8 shorts.
"uqrshrn v3.8b, v0.8h, #2 \n" // 2x2 average
"uqrshrn v2.8b, v1.8h, #2 \n"
"subs %w3, %w3, #16 \n" // 16 processed per loop.
"st2 {v2.8b,v3.8b}, [%2], #16 \n" // store 8 pixels UV.
"b.gt 1b \n"
: "+r"(src_ayuv), // %0
"+r"(src_ayuv_1), // %1
"+r"(dst_uv), // %2
"+r"(width) // %3
:
: "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7");
}
void AYUVToVURow_NEON(const uint8_t* src_ayuv,
int src_stride_ayuv,
uint8_t* dst_vu,
@ -2906,7 +2934,8 @@ void AYUVToVURow_NEON(const uint8_t* src_ayuv,
asm volatile(
"1: \n"
"ld4 {v0.16b,v1.16b,v2.16b,v3.16b}, [%0], #64 \n" // load 16 pixels.
"ld4 {v0.16b,v1.16b,v2.16b,v3.16b}, [%0], #64 \n" // load 16
// pixels.
"uaddlp v0.8h, v0.16b \n" // V 16 bytes -> 8 shorts.
"uaddlp v1.8h, v1.16b \n" // U 16 bytes -> 8 shorts.
"ld4 {v4.16b,v5.16b,v6.16b,v7.16b}, [%1], #64 \n" // load next 16
@ -2922,15 +2951,15 @@ void AYUVToVURow_NEON(const uint8_t* src_ayuv,
"+r"(dst_vu), // %2
"+r"(width) // %3
:
: "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7"
);
: "cc", "memory", "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7");
}
// Copy row of AYUV Y's into Y
void AYUVToYRow_NEON(const uint8_t* src_ayuv, uint8_t* dst_y, int width) {
asm volatile(
"1: \n"
"ld4 {v0.16b,v1.16b,v2.16b,v3.16b}, [%0], #64 \n" // load 16 pixels
"ld4 {v0.16b,v1.16b,v2.16b,v3.16b}, [%0], #64 \n" // load 16
// pixels
"subs %w2, %w2, #16 \n" // 16 pixels per loop
"st1 {v2.16b}, [%1], #16 \n" // store 16 Y pixels
"b.gt 1b \n"
@ -2983,6 +3012,22 @@ void FloatDivToByteRow_NEON(const float* src_weights,
: "cc", "memory", "v1", "v2", "v3", "v4", "v5", "v6");
}
// Convert biplanar UV channel of NV12 to NV21
void UVToVURow_NEON(const uint8_t* src_uv, uint8_t* dst_vu, int width) {
asm volatile(
"1: \n"
"ld2 {v0.16b, v1.16b}, [%0], #32 \n" // load 16 UV values
"orr v2.16b, v0.16b, v0.16b \n" // move U after V
"subs %w2, %w2, #16 \n" // 16 pixels per loop
"st2 {v1.16b, v2.16b}, [%1], #32 \n" // store 16 VU pixels
"b.gt 1b \n"
: "+r"(src_uv), // %0
"+r"(dst_vu), // %1
"+r"(width) // %2
:
: "cc", "memory", "v0", "v1", "v2");
}
#endif // !defined(LIBYUV_DISABLE_NEON) && defined(__aarch64__)
#ifdef __cplusplus

139
unit_test/convert_test.cc
View File

@ -311,10 +311,10 @@ int I400ToNV21(const uint8_t* src_y,
SUBSAMPLE(kHeight, SRC_SUBSAMP_Y) + \
OFF); \
align_buffer_page_end(dst_y_c, kWidth* kHeight); \
align_buffer_page_end(dst_uv_c, SUBSAMPLE(kWidth * 2, SUBSAMP_X) * \
align_buffer_page_end(dst_uv_c, SUBSAMPLE(kWidth, SUBSAMP_X) * 2 * \
SUBSAMPLE(kHeight, SUBSAMP_Y)); \
align_buffer_page_end(dst_y_opt, kWidth* kHeight); \
align_buffer_page_end(dst_uv_opt, SUBSAMPLE(kWidth * 2, SUBSAMP_X) * \
align_buffer_page_end(dst_uv_opt, SUBSAMPLE(kWidth, SUBSAMP_X) * 2 * \
SUBSAMPLE(kHeight, SUBSAMP_Y)); \
for (int i = 0; i < kHeight; ++i) \
for (int j = 0; j < kWidth; ++j) \
@ -329,21 +329,21 @@ int I400ToNV21(const uint8_t* src_y,
} \
memset(dst_y_c, 1, kWidth* kHeight); \
memset(dst_uv_c, 2, \
SUBSAMPLE(kWidth * 2, SUBSAMP_X) * SUBSAMPLE(kHeight, SUBSAMP_Y)); \
SUBSAMPLE(kWidth, SUBSAMP_X) * 2 * SUBSAMPLE(kHeight, SUBSAMP_Y)); \
memset(dst_y_opt, 101, kWidth* kHeight); \
memset(dst_uv_opt, 102, \
SUBSAMPLE(kWidth * 2, SUBSAMP_X) * SUBSAMPLE(kHeight, SUBSAMP_Y)); \
SUBSAMPLE(kWidth, SUBSAMP_X) * 2 * SUBSAMPLE(kHeight, SUBSAMP_Y)); \
MaskCpuFlags(disable_cpu_flags_); \
SRC_FMT_PLANAR##To##FMT_PLANAR( \
src_y + OFF, kWidth, src_u + OFF, SUBSAMPLE(kWidth, SRC_SUBSAMP_X), \
src_v + OFF, SUBSAMPLE(kWidth, SRC_SUBSAMP_X), dst_y_c, kWidth, \
dst_uv_c, SUBSAMPLE(kWidth * 2, SUBSAMP_X), kWidth, NEG kHeight); \
dst_uv_c, SUBSAMPLE(kWidth, SUBSAMP_X) * 2, kWidth, NEG kHeight); \
MaskCpuFlags(benchmark_cpu_info_); \
for (int i = 0; i < benchmark_iterations_; ++i) { \
SRC_FMT_PLANAR##To##FMT_PLANAR( \
src_y + OFF, kWidth, src_u + OFF, SUBSAMPLE(kWidth, SRC_SUBSAMP_X), \
src_v + OFF, SUBSAMPLE(kWidth, SRC_SUBSAMP_X), dst_y_opt, kWidth, \
dst_uv_opt, SUBSAMPLE(kWidth * 2, SUBSAMP_X), kWidth, NEG kHeight); \
dst_uv_opt, SUBSAMPLE(kWidth, SUBSAMP_X) * 2, kWidth, NEG kHeight); \
} \
int max_diff = 0; \
for (int i = 0; i < kHeight; ++i) { \
@ -357,12 +357,12 @@ int I400ToNV21(const uint8_t* src_y,
} \
EXPECT_LE(max_diff, 1); \
for (int i = 0; i < SUBSAMPLE(kHeight, SUBSAMP_Y); ++i) { \
for (int j = 0; j < SUBSAMPLE(kWidth * 2, SUBSAMP_X); ++j) { \
for (int j = 0; j < SUBSAMPLE(kWidth, SUBSAMP_X) * 2; ++j) { \
int abs_diff = \
abs(static_cast<int>( \
dst_uv_c[i * SUBSAMPLE(kWidth * 2, SUBSAMP_X) + j]) - \
dst_uv_c[i * SUBSAMPLE(kWidth, SUBSAMP_X) * 2 + j]) - \
static_cast<int>( \
dst_uv_opt[i * SUBSAMPLE(kWidth * 2, SUBSAMP_X) + j])); \
dst_uv_opt[i * SUBSAMPLE(kWidth, SUBSAMP_X) * 2 + j])); \
if (abs_diff > max_diff) { \
max_diff = abs_diff; \
} \
@ -395,6 +395,99 @@ TESTPLANARTOBP(I422, 2, 1, NV21, 2, 2)
TESTPLANARTOBP(I444, 1, 1, NV21, 2, 2)
TESTPLANARTOBP(I400, 2, 2, NV21, 2, 2)
#define TESTBIPLANARTOBPI(SRC_FMT_PLANAR, SRC_SUBSAMP_X, SRC_SUBSAMP_Y, \
FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, W1280, N, NEG, \
OFF) \
TEST_F(LibYUVConvertTest, SRC_FMT_PLANAR##To##FMT_PLANAR##N) { \
const int kWidth = ((W1280) > 0) ? (W1280) : 1; \
const int kHeight = benchmark_height_; \
align_buffer_page_end(src_y, kWidth* kHeight + OFF); \
align_buffer_page_end(src_uv, SUBSAMPLE(kWidth, SRC_SUBSAMP_X) * 2 * \
SUBSAMPLE(kHeight, SRC_SUBSAMP_Y) + \
OFF); \
align_buffer_page_end(dst_y_c, kWidth* kHeight); \
align_buffer_page_end(dst_uv_c, SUBSAMPLE(kWidth, SUBSAMP_X) * 2 * \
SUBSAMPLE(kHeight, SUBSAMP_Y)); \
align_buffer_page_end(dst_y_opt, kWidth* kHeight); \
align_buffer_page_end(dst_uv_opt, SUBSAMPLE(kWidth, SUBSAMP_X) * 2 * \
SUBSAMPLE(kHeight, SUBSAMP_Y)); \
for (int i = 0; i < kHeight; ++i) \
for (int j = 0; j < kWidth; ++j) \
src_y[i * kWidth + j + OFF] = (fastrand() & 0xff); \
for (int i = 0; i < SUBSAMPLE(kHeight, SRC_SUBSAMP_Y); ++i) { \
for (int j = 0; j < SUBSAMPLE(kWidth, SRC_SUBSAMP_X); ++j) { \
src_uv[(i * SUBSAMPLE(kWidth, SRC_SUBSAMP_X)) * 2 + j + 0 + OFF] = \
(fastrand() & 0xff); \
src_uv[(i * SUBSAMPLE(kWidth, SRC_SUBSAMP_X)) * 2 + j + 1 + OFF] = \
(fastrand() & 0xff); \
} \
} \
memset(dst_y_c, 1, kWidth* kHeight); \
memset(dst_uv_c, 2, \
SUBSAMPLE(kWidth, SUBSAMP_X) * 2 * SUBSAMPLE(kHeight, SUBSAMP_Y)); \
memset(dst_y_opt, 101, kWidth* kHeight); \
memset(dst_uv_opt, 102, \
SUBSAMPLE(kWidth, SUBSAMP_X) * 2 * SUBSAMPLE(kHeight, SUBSAMP_Y)); \
MaskCpuFlags(disable_cpu_flags_); \
SRC_FMT_PLANAR##To##FMT_PLANAR( \
src_y + OFF, kWidth, src_uv + OFF, \
SUBSAMPLE(kWidth, SRC_SUBSAMP_X) * 2, dst_y_c, kWidth, dst_uv_c, \
SUBSAMPLE(kWidth, SUBSAMP_X) * 2, kWidth, NEG kHeight); \
MaskCpuFlags(benchmark_cpu_info_); \
for (int i = 0; i < benchmark_iterations_; ++i) { \
SRC_FMT_PLANAR##To##FMT_PLANAR( \
src_y + OFF, kWidth, src_uv + OFF, \
SUBSAMPLE(kWidth, SRC_SUBSAMP_X) * 2, dst_y_opt, kWidth, dst_uv_opt, \
SUBSAMPLE(kWidth, SUBSAMP_X) * 2, kWidth, NEG kHeight); \
} \
int max_diff = 0; \
for (int i = 0; i < kHeight; ++i) { \
for (int j = 0; j < kWidth; ++j) { \
int abs_diff = abs(static_cast<int>(dst_y_c[i * kWidth + j]) - \
static_cast<int>(dst_y_opt[i * kWidth + j])); \
if (abs_diff > max_diff) { \
max_diff = abs_diff; \
} \
} \
} \
EXPECT_LE(max_diff, 1); \
for (int i = 0; i < SUBSAMPLE(kHeight, SUBSAMP_Y); ++i) { \
for (int j = 0; j < SUBSAMPLE(kWidth, SUBSAMP_X) * 2; ++j) { \
int abs_diff = \
abs(static_cast<int>( \
dst_uv_c[i * SUBSAMPLE(kWidth, SUBSAMP_X) * 2 + j]) - \
static_cast<int>( \
dst_uv_opt[i * SUBSAMPLE(kWidth, SUBSAMP_X) * 2 + j])); \
if (abs_diff > max_diff) { \
max_diff = abs_diff; \
} \
} \
} \
EXPECT_LE(max_diff, 1); \
free_aligned_buffer_page_end(dst_y_c); \
free_aligned_buffer_page_end(dst_uv_c); \
free_aligned_buffer_page_end(dst_y_opt); \
free_aligned_buffer_page_end(dst_uv_opt); \
free_aligned_buffer_page_end(src_y); \
free_aligned_buffer_page_end(src_uv); \
}
#define TESTBIPLANARTOBP(SRC_FMT_PLANAR, SRC_SUBSAMP_X, SRC_SUBSAMP_Y, \
FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y) \
TESTBIPLANARTOBPI(SRC_FMT_PLANAR, SRC_SUBSAMP_X, SRC_SUBSAMP_Y, FMT_PLANAR, \
SUBSAMP_X, SUBSAMP_Y, benchmark_width_, _Opt, +, 0) \
TESTBIPLANARTOBPI(SRC_FMT_PLANAR, SRC_SUBSAMP_X, SRC_SUBSAMP_Y, FMT_PLANAR, \
SUBSAMP_X, SUBSAMP_Y, benchmark_width_ - 4, _Any, +, 0) \
TESTBIPLANARTOBPI(SRC_FMT_PLANAR, SRC_SUBSAMP_X, SRC_SUBSAMP_Y, FMT_PLANAR, \
SUBSAMP_X, SUBSAMP_Y, benchmark_width, _Unaligned, +, 1) \
TESTBIPLANARTOBPI(SRC_FMT_PLANAR, SRC_SUBSAMP_X, SRC_SUBSAMP_Y, FMT_PLANAR, \
SUBSAMP_X, SUBSAMP_Y, benchmark_width_, _Invert, -, 0) \
TESTBIPLANARTOBPI(SRC_FMT_PLANAR, SRC_SUBSAMP_X, SRC_SUBSAMP_Y, FMT_PLANAR, \
SUBSAMP_X, SUBSAMP_Y, benchmark_width_, _Opt, +, 0)
// TODO(fbarchard): Fix msan on this unittest
// TESTBIPLANARTOBP(NV21, 2, 2, NV12, 2, 2)
#define TESTBIPLANARTOPI(SRC_FMT_PLANAR, SRC_SUBSAMP_X, SRC_SUBSAMP_Y, \
FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, W1280, N, NEG, OFF, \
DOY) \
@ -680,8 +773,8 @@ TESTPLANARTOB(H420, 2, 2, AR30, 4, 4, 1)
TESTQPLANARTOB(I420Alpha, 2, 2, ARGB, 4, 4, 1, 2)
TESTQPLANARTOB(I420Alpha, 2, 2, ABGR, 4, 4, 1, 2)
#define TESTBIPLANARTOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, FMT_C, BPP_B,\
W1280, DIFF, N, NEG, OFF) \
#define TESTBIPLANARTOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, FMT_C, \
BPP_B, W1280, DIFF, N, NEG, OFF) \
TEST_F(LibYUVConvertTest, FMT_PLANAR##To##FMT_B##N) { \
const int kWidth = ((W1280) > 0) ? (W1280) : 1; \
const int kHeight = benchmark_height_; \
@ -740,8 +833,8 @@ TESTQPLANARTOB(I420Alpha, 2, 2, ABGR, 4, 4, 1, 2)
free_aligned_buffer_page_end(dst_argb32_opt); \
}
#define TESTBIPLANARTOB(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, FMT_C, \
BPP_B, DIFF) \
#define TESTBIPLANARTOB(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, FMT_C, BPP_B, \
DIFF) \
TESTBIPLANARTOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, FMT_C, BPP_B, \
benchmark_width_ - 4, DIFF, _Any, +, 0) \
TESTBIPLANARTOBI(FMT_PLANAR, SUBSAMP_X, SUBSAMP_Y, FMT_B, FMT_C, BPP_B, \
@ -980,6 +1073,7 @@ TESTATOBIPLANAR(ARGB, 1, 4, NV12, 2, 2)
TESTATOBIPLANAR(ARGB, 1, 4, NV21, 2, 2)
TESTATOBIPLANAR(YUY2, 2, 4, NV12, 2, 2)
TESTATOBIPLANAR(UYVY, 2, 4, NV12, 2, 2)
TESTATOBIPLANAR(AYUV, 1, 4, NV12, 2, 2)
TESTATOBIPLANAR(AYUV, 1, 4, NV21, 2, 2)
#define TESTATOBI(FMT_A, BPP_A, STRIDE_A, HEIGHT_A, FMT_B, BPP_B, STRIDE_B, \
@ -1378,14 +1472,15 @@ TEST_F(LibYUVConvertTest, FuzzJpeg) {
orig_pixels[0] = 0xff;
orig_pixels[1] = 0xd8; // SOI.
orig_pixels[kSize - 1] = 0xff;
ValidateJpeg(orig_pixels, kSize); // Failure normally expected.
ValidateJpeg(orig_pixels,
kSize); // Failure normally expected.
free_aligned_buffer_page_end(orig_pixels);
}
}
// Test data created in GIMP. In export jpeg, disable thumbnails etc,
// choose a subsampling, and use low quality (50) to keep size small.
// Generated with xxd -i test.jpg
// Test data created in GIMP. In export jpeg, disable
// thumbnails etc, choose a subsampling, and use low quality
// (50) to keep size small. Generated with xxd -i test.jpg
// test 0 is J400
static const uint8_t kTest0Jpg[] = {
0xff, 0xd8, 0xff, 0xe0, 0x00, 0x10, 0x4a, 0x46, 0x49, 0x46, 0x00, 0x01,
@ -1987,8 +2082,8 @@ TEST_F(LibYUVConvertTest, TestMJPGInfo) {
EXPECT_EQ(1, ShowJPegInfo(kTest1Jpg, kTest1JpgLen));
EXPECT_EQ(1, ShowJPegInfo(kTest2Jpg, kTest2JpgLen));
EXPECT_EQ(1, ShowJPegInfo(kTest3Jpg, kTest3JpgLen));
EXPECT_EQ(1,
ShowJPegInfo(kTest4Jpg, kTest4JpgLen)); // Valid but unsupported.
EXPECT_EQ(1, ShowJPegInfo(kTest4Jpg,
kTest4JpgLen)); // Valid but unsupported.
}
#endif // HAVE_JPEG
@ -2906,7 +3001,8 @@ TEST_F(LibYUVConvertTest, TestH010ToARGB) {
}
// Test 10 bit YUV to 10 bit RGB
// Caveat: Result is near due to float rounding in expected result.
// Caveat: Result is near due to float rounding in expected
// result.
TEST_F(LibYUVConvertTest, TestH010ToAR30) {
const int kSize = 1024;
int histogram_b[1024];
@ -2969,7 +3065,8 @@ TEST_F(LibYUVConvertTest, TestH010ToAR30) {
}
// Test 10 bit YUV to 10 bit RGB
// Caveat: Result is near due to float rounding in expected result.
// Caveat: Result is near due to float rounding in expected
// result.
TEST_F(LibYUVConvertTest, TestH010ToAB30) {
const int kSize = 1024;
int histogram_b[1024];

37
unit_test/planar_test.cc
View File

@ -3287,7 +3287,8 @@ float TestFloatDivToByte(int benchmark_width,
// large values are problematic. audio is really -1 to 1.
for (i = 0; i < kPixels; ++i) {
(reinterpret_cast<float*>(src_weights))[i] = scale;
(reinterpret_cast<float*>(src_values))[i] = sinf(static_cast<float>(i) * 0.1f);
(reinterpret_cast<float*>(src_values))[i] =
sinf(static_cast<float>(i) * 0.1f);
}
memset(dst_out_c, 0, kPixels);
memset(dst_out_opt, 1, kPixels);
@ -3295,24 +3296,24 @@ float TestFloatDivToByte(int benchmark_width,
memset(dst_mask_opt, 3, kPixels);
FloatDivToByteRow_C(reinterpret_cast<float*>(src_weights),
reinterpret_cast<float*>(src_values),
dst_out_c, dst_mask_c, kPixels);
reinterpret_cast<float*>(src_values), dst_out_c,
dst_mask_c, kPixels);
for (j = 0; j < benchmark_iterations; j++) {
if (opt) {
#ifdef HAS_FLOATDIVTOBYTEROW_NEON
FloatDivToByteRow_NEON(reinterpret_cast<float*>(src_weights),
reinterpret_cast<float*>(src_values),
dst_out_opt, dst_mask_opt, kPixels);
reinterpret_cast<float*>(src_values), dst_out_opt,
dst_mask_opt, kPixels);
#else
FloatDivToByteRow_C(reinterpret_cast<float*>(src_weights),
reinterpret_cast<float*>(src_values),
dst_out_opt, dst_mask_opt, kPixels);
reinterpret_cast<float*>(src_values), dst_out_opt,
dst_mask_opt, kPixels);
#endif
} else {
FloatDivToByteRow_C(reinterpret_cast<float*>(src_weights),
reinterpret_cast<float*>(src_values),
dst_out_opt, dst_mask_opt, kPixels);
reinterpret_cast<float*>(src_values), dst_out_opt,
dst_mask_opt, kPixels);
}
}
@ -3347,5 +3348,23 @@ TEST_F(LibYUVPlanarTest, TestFloatDivToByte_Opt) {
EXPECT_EQ(0, diff);
}
TEST_F(LibYUVPlanarTest, UVToVURow) {
const int kPixels = benchmark_width_ * benchmark_height_;
align_buffer_page_end(src_pixels_vu, kPixels * 2);
align_buffer_page_end(dst_pixels_uv, kPixels * 2);
MemRandomize(src_pixels_vu, kPixels * 2);
memset(dst_pixels_uv, 1, kPixels * 2);
UVToVURow_C(src_pixels_vu, dst_pixels_uv, kPixels);
for (int i = 0; i < kPixels; ++i) {
EXPECT_EQ(dst_pixels_uv[i * 2 + 0], src_pixels_vu[i * 2 + 1]);
EXPECT_EQ(dst_pixels_uv[i * 2 + 1], src_pixels_vu[i * 2 + 0]);
}
free_aligned_buffer_page_end(src_pixels_vu);
free_aligned_buffer_page_end(dst_pixels_uv);
}
} // namespace libyuv