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

PlaneScale, UVScale and ARGBScale test 3x and 4x down sample.

Browse Source 
Intel SkylakeX
UVTest3x (1925 ms)
UVTest4x (2915 ms)
PlaneTest3x (2040 ms)
PlaneTest4x (4292 ms)
ARGBTest3x (2079 ms)
ARGBTest4x (1854 ms)

Pixel 2
ARGBTest3x (3602 ms)
ARGBTest4x (4064 ms)
PlaneTest3x (3331 ms)
PlaneTest4x (8977 ms)
UVTest3x (3473 ms)
UVTest4x (6970 ms)

Bug: b/171798872, b/171884264
Change-Id: Iebc70fed907857b6cb71a9baf2aba9861ef1e3f7
Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/2505601
Reviewed-by: richard winterton <rrwinterton@gmail.com>
Commit-Queue: Frank Barchard <fbarchard@chromium.org>
This commit is contained in:
Frank Barchard 2020-10-28 11:31:10 -07:00 committed by Commit Bot
parent 5c4dc242f4
commit cec28e7088
11 changed files with 262 additions and 99 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
BUILD.gn
View File

@ -376,7 +376,9 @@ if (libyuv_include_tests) {
# sources # sources
"util/i444tonv12_eg.cc", "util/i444tonv12_eg.cc",
] ]
deps = [ ":libyuv" ] deps = [
":libyuv",
]
} }
executable("cpuid") { executable("cpuid") {

2
README.chromium
View File

@ -1,6 +1,6 @@
Name: libyuv Name: libyuv
URL: http://code.google.com/p/libyuv/ URL: http://code.google.com/p/libyuv/
Version: 1766 Version: 1767
License: BSD License: BSD
License File: LICENSE License File: LICENSE

13
include/libyuv/convert.h
View File

@ -460,9 +460,16 @@ int MJPGToNV21(const uint8_t* sample,
// JPEG to NV12 // JPEG to NV12
LIBYUV_API LIBYUV_API
int MJPGToNV12(const uint8_t* sample, size_t sample_size, uint8_t* dst_y, int MJPGToNV12(const uint8_t* sample,
int dst_stride_y, uint8_t* dst_uv, int dst_stride_uv, size_t sample_size,
int src_width, int src_height, int dst_width, int dst_height); uint8_t* dst_y,
int dst_stride_y,
uint8_t* dst_uv,
int dst_stride_uv,
int src_width,
int src_height,
int dst_width,
int dst_height);
// Query size of MJPG in pixels. // Query size of MJPG in pixels.
LIBYUV_API LIBYUV_API

2
include/libyuv/convert_from_argb.h
View File

@ -79,7 +79,7 @@ int ARGBToAR30(const uint8_t* src_argb,
// Aliases // Aliases
#define ABGRToRGB24 ARGBToRAW #define ABGRToRGB24 ARGBToRAW
#define ABGRToRAW ARGBToRGB24 #define ABGRToRAW ARGBToRGB24
// Convert ARGB To RGB24. // Convert ARGB To RGB24.
LIBYUV_API LIBYUV_API

52
include/libyuv/scale_row.h
View File

@ -82,7 +82,7 @@ extern "C" {
// The following are available for gcc/clang x86 platforms, but // The following are available for gcc/clang x86 platforms, but
// require clang 3.4 or gcc 4.7. // require clang 3.4 or gcc 4.7.
// TODO(fbarchard): Port to Visual C // TODO(fbarchard): Port to Visual C
#if !defined(LIBYUV_DISABLE_X86) && \ #if !defined(LIBYUV_DISABLE_X86) && \
(defined(__x86_64__) || defined(__i386__)) && !defined(_MSC_VER) && \ (defined(__x86_64__) || defined(__i386__)) && !defined(_MSC_VER) && \
(defined(CLANG_HAS_AVX2) || defined(GCC_HAS_AVX2)) (defined(CLANG_HAS_AVX2) || defined(GCC_HAS_AVX2))
#define HAS_SCALEUVROWDOWN2BOX_AVX2 #define HAS_SCALEUVROWDOWN2BOX_AVX2
@ -848,17 +848,17 @@ void ScaleARGBRowDownEvenBox_Any_MMI(const uint8_t* src_ptr,
// UV Row functions // UV Row functions
void ScaleUVRowDown2_SSSE3(const uint8_t* src_ptr, void ScaleUVRowDown2_SSSE3(const uint8_t* src_ptr,
ptrdiff_t src_stride, ptrdiff_t src_stride,
uint8_t* dst_uv, uint8_t* dst_uv,
int dst_width); int dst_width);
void ScaleUVRowDown2Linear_SSSE3(const uint8_t* src_ptr, void ScaleUVRowDown2Linear_SSSE3(const uint8_t* src_ptr,
ptrdiff_t src_stride, ptrdiff_t src_stride,
uint8_t* dst_uv, uint8_t* dst_uv,
int dst_width); int dst_width);
void ScaleUVRowDown2Box_SSSE3(const uint8_t* src_ptr, void ScaleUVRowDown2Box_SSSE3(const uint8_t* src_ptr,
ptrdiff_t src_stride, ptrdiff_t src_stride,
uint8_t* dst_uv, uint8_t* dst_uv,
int dst_width); int dst_width);
void ScaleUVRowDown2Box_AVX2(const uint8_t* src_ptr, void ScaleUVRowDown2Box_AVX2(const uint8_t* src_ptr,
ptrdiff_t src_stride, ptrdiff_t src_stride,
uint8_t* dst_uv, uint8_t* dst_uv,
@ -952,15 +952,15 @@ void ScaleUVRowDown2Box_Any_MMI(const uint8_t* src_ptr,
uint8_t* dst_ptr, uint8_t* dst_ptr,
int dst_width); int dst_width);
void ScaleUVRowDownEven_SSSE3(const uint8_t* src_ptr, void ScaleUVRowDownEven_SSSE3(const uint8_t* src_ptr,
ptrdiff_t src_stride, ptrdiff_t src_stride,
int src_stepx, int src_stepx,
uint8_t* dst_uv, uint8_t* dst_uv,
int dst_width); int dst_width);
void ScaleUVRowDownEvenBox_SSSE3(const uint8_t* src_ptr, void ScaleUVRowDownEvenBox_SSSE3(const uint8_t* src_ptr,
ptrdiff_t src_stride, ptrdiff_t src_stride,
int src_stepx, int src_stepx,
uint8_t* dst_uv, uint8_t* dst_uv,
int dst_width); int dst_width);
void ScaleUVRowDownEven_NEON(const uint8_t* src_ptr, void ScaleUVRowDownEven_NEON(const uint8_t* src_ptr,
ptrdiff_t src_stride, ptrdiff_t src_stride,
int src_stepx, int src_stepx,
@ -992,15 +992,15 @@ void ScaleUVRowDownEvenBox_MMI(const uint8_t* src_ptr,
uint8_t* dst_uv, uint8_t* dst_uv,
int dst_width); int dst_width);
void ScaleUVRowDownEven_Any_SSSE3(const uint8_t* src_ptr, void ScaleUVRowDownEven_Any_SSSE3(const uint8_t* src_ptr,
ptrdiff_t src_stride, ptrdiff_t src_stride,
int src_stepx, int src_stepx,
uint8_t* dst_ptr, uint8_t* dst_ptr,
int dst_width); int dst_width);
void ScaleUVRowDownEvenBox_Any_SSSE3(const uint8_t* src_ptr, void ScaleUVRowDownEvenBox_Any_SSSE3(const uint8_t* src_ptr,
ptrdiff_t src_stride, ptrdiff_t src_stride,
int src_stepx, int src_stepx,
uint8_t* dst_ptr, uint8_t* dst_ptr,
int dst_width); int dst_width);
void ScaleUVRowDownEven_Any_NEON(const uint8_t* src_ptr, void ScaleUVRowDownEven_Any_NEON(const uint8_t* src_ptr,
ptrdiff_t src_stride, ptrdiff_t src_stride,
int src_stepx, int src_stepx,

2
include/libyuv/version.h
View File

@ -11,6 +11,6 @@
#ifndef INCLUDE_LIBYUV_VERSION_H_ #ifndef INCLUDE_LIBYUV_VERSION_H_
#define INCLUDE_LIBYUV_VERSION_H_ #define INCLUDE_LIBYUV_VERSION_H_
#define LIBYUV_VERSION 1766 #define LIBYUV_VERSION 1767
#endif // INCLUDE_LIBYUV_VERSION_H_ #endif // INCLUDE_LIBYUV_VERSION_H_

37
source/convert_jpeg.cc
View File

@ -328,8 +328,10 @@ int MJPGToNV21(const uint8_t* src_mjpg,
return ret ? 0 : 1; return ret ? 0 : 1;
} }
static void JpegI420ToNV12(void* opaque, const uint8_t* const* data, static void JpegI420ToNV12(void* opaque,
const int* strides, int rows) { const uint8_t* const* data,
const int* strides,
int rows) {
NV21Buffers* dest = (NV21Buffers*)(opaque); NV21Buffers* dest = (NV21Buffers*)(opaque);
// Use NV21 with VU swapped. // Use NV21 with VU swapped.
I420ToNV21(data[0], strides[0], data[2], strides[2], data[1], strides[1], I420ToNV21(data[0], strides[0], data[2], strides[2], data[1], strides[1],
@ -339,8 +341,10 @@ static void JpegI420ToNV12(void* opaque, const uint8_t* const* data,
dest->h -= rows; dest->h -= rows;
} }
static void JpegI422ToNV12(void* opaque, const uint8_t* const* data, static void JpegI422ToNV12(void* opaque,
const int* strides, int rows) { const uint8_t* const* data,
const int* strides,
int rows) {
NV21Buffers* dest = (NV21Buffers*)(opaque); NV21Buffers* dest = (NV21Buffers*)(opaque);
// Use NV21 with VU swapped. // Use NV21 with VU swapped.
I422ToNV21(data[0], strides[0], data[2], strides[2], data[1], strides[1], I422ToNV21(data[0], strides[0], data[2], strides[2], data[1], strides[1],
@ -350,8 +354,10 @@ static void JpegI422ToNV12(void* opaque, const uint8_t* const* data,
dest->h -= rows; dest->h -= rows;
} }
static void JpegI444ToNV12(void* opaque, const uint8_t* const* data, static void JpegI444ToNV12(void* opaque,
const int* strides, int rows) { const uint8_t* const* data,
const int* strides,
int rows) {
NV21Buffers* dest = (NV21Buffers*)(opaque); NV21Buffers* dest = (NV21Buffers*)(opaque);
// Use NV21 with VU swapped. // Use NV21 with VU swapped.
I444ToNV21(data[0], strides[0], data[2], strides[2], data[1], strides[1], I444ToNV21(data[0], strides[0], data[2], strides[2], data[1], strides[1],
@ -361,8 +367,10 @@ static void JpegI444ToNV12(void* opaque, const uint8_t* const* data,
dest->h -= rows; dest->h -= rows;
} }
static void JpegI400ToNV12(void* opaque, const uint8_t* const* data, static void JpegI400ToNV12(void* opaque,
const int* strides, int rows) { const uint8_t* const* data,
const int* strides,
int rows) {
NV21Buffers* dest = (NV21Buffers*)(opaque); NV21Buffers* dest = (NV21Buffers*)(opaque);
// Use NV21 since there is no UV plane. // Use NV21 since there is no UV plane.
I400ToNV21(data[0], strides[0], dest->y, dest->y_stride, dest->vu, I400ToNV21(data[0], strides[0], dest->y, dest->y_stride, dest->vu,
@ -374,9 +382,16 @@ static void JpegI400ToNV12(void* opaque, const uint8_t* const* data,
// MJPG (Motion JPEG) to NV12. // MJPG (Motion JPEG) to NV12.
LIBYUV_API LIBYUV_API
int MJPGToNV12(const uint8_t* sample, size_t sample_size, uint8_t* dst_y, int MJPGToNV12(const uint8_t* sample,
int dst_stride_y, uint8_t* dst_uv, int dst_stride_uv, size_t sample_size,
int src_width, int src_height, int dst_width, int dst_height) { uint8_t* dst_y,
int dst_stride_y,
uint8_t* dst_uv,
int dst_stride_uv,
int src_width,
int src_height,
int dst_width,
int dst_height) {
if (sample_size == kUnknownDataSize) { if (sample_size == kUnknownDataSize) {
// ERROR: MJPEG frame size unknown // ERROR: MJPEG frame size unknown
return -1; return -1;

6
source/scale_common.cc
View File

@ -1063,9 +1063,11 @@ void ScaleUVRowDown2Box_C(const uint8_t* src_uv,
int x; int x;
for (x = 0; x < dst_width; ++x) { for (x = 0; x < dst_width; ++x) {
dst_uv[0] = (src_uv[0] + src_uv[2] + src_uv[src_stride] + dst_uv[0] = (src_uv[0] + src_uv[2] + src_uv[src_stride] +
src_uv[src_stride + 2] + 2) >> 2; src_uv[src_stride + 2] + 2) >>
2;
dst_uv[1] = (src_uv[1] + src_uv[3] + src_uv[src_stride + 1] + dst_uv[1] = (src_uv[1] + src_uv[3] + src_uv[src_stride + 1] +
src_uv[src_stride + 3] + 2) >> 2; src_uv[src_stride + 3] + 2) >>
2;
src_uv += 4; src_uv += 4;
dst_uv += 2; dst_uv += 2;
} }

33
source/scale_gcc.cc
View File

@ -1370,8 +1370,9 @@ int FixedDiv1_X86(int num, int div) {
// Shuffle table for splitting UV into upper and lower part of register. // Shuffle table for splitting UV into upper and lower part of register.
static const uvec8 kShuffleSplitUV = {0u, 2u, 4u, 6u, 8u, 10u, 12u, 14u, static const uvec8 kShuffleSplitUV = {0u, 2u, 4u, 6u, 8u, 10u, 12u, 14u,
1u, 3u, 5u, 7u, 9u, 11u, 13u, 15u}; 1u, 3u, 5u, 7u, 9u, 11u, 13u, 15u};
static const uvec8 kShuffleMergeUV = {0u, 8u, 2u, 10u, 4u, 12u, 6u, 14u, static const uvec8 kShuffleMergeUV = {0u, 8u, 2u, 10u, 4u, 12u,
0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80}; 6u, 14u, 0x80, 0x80, 0x80, 0x80,
0x80, 0x80, 0x80, 0x80};
void ScaleUVRowDown2Box_SSSE3(const uint8_t* src_ptr, void ScaleUVRowDown2Box_SSSE3(const uint8_t* src_ptr,
ptrdiff_t src_stride, ptrdiff_t src_stride,
@ -1402,15 +1403,15 @@ void ScaleUVRowDown2Box_SSSE3(const uint8_t* src_ptr,
"lea 0x8(%1),%1 \n" // 4 UV "lea 0x8(%1),%1 \n" // 4 UV
"sub $0x4,%2 \n" "sub $0x4,%2 \n"
"jg 1b \n" "jg 1b \n"
: "+r"(src_ptr), // %0 : "+r"(src_ptr), // %0
"+r"(dst_ptr), // %1 "+r"(dst_ptr), // %1
"+r"(dst_width) // %2 "+r"(dst_width) // %2
: "r"((intptr_t)(src_stride)), // %3 : "r"((intptr_t)(src_stride)), // %3
"m"(kShuffleSplitUV), // %4 "m"(kShuffleSplitUV), // %4
"m"(kShuffleMergeUV) // %5 "m"(kShuffleMergeUV) // %5
: "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5"); : "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5");
} }
#endif // HAS_SCALEUVROWDOWN2BOX_SSSE3 #endif // HAS_SCALEUVROWDOWN2BOX_SSSE3
#ifdef HAS_SCALEUVROWDOWN2BOX_AVX2 #ifdef HAS_SCALEUVROWDOWN2BOX_AVX2
void ScaleUVRowDown2Box_AVX2(const uint8_t* src_ptr, void ScaleUVRowDown2Box_AVX2(const uint8_t* src_ptr,
@ -1444,15 +1445,15 @@ void ScaleUVRowDown2Box_AVX2(const uint8_t* src_ptr,
"sub $0x8,%2 \n" "sub $0x8,%2 \n"
"jg 1b \n" "jg 1b \n"
"vzeroupper \n" "vzeroupper \n"
: "+r"(src_ptr), // %0 : "+r"(src_ptr), // %0
"+r"(dst_ptr), // %1 "+r"(dst_ptr), // %1
"+r"(dst_width) // %2 "+r"(dst_width) // %2
: "r"((intptr_t)(src_stride)), // %3 : "r"((intptr_t)(src_stride)), // %3
"m"(kShuffleSplitUV), // %4 "m"(kShuffleSplitUV), // %4
"m"(kShuffleMergeUV) // %5 "m"(kShuffleMergeUV) // %5
: "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5"); : "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5");
} }
#endif // HAS_SCALEUVROWDOWN2BOX_AVX2 #endif // HAS_SCALEUVROWDOWN2BOX_AVX2
#endif // defined(__x86_64__) || defined(__i386__) #endif // defined(__x86_64__) || defined(__i386__)

75
unit_test/scale_argb_test.cc
View File

@ -456,4 +456,79 @@ TEST_F(LibYUVScaleTest, YUVToRGBScaleDown) {
EXPECT_LE(diff, 10); EXPECT_LE(diff, 10);
} }
TEST_F(LibYUVScaleTest, ARGBTest3x) {
const int kSrcStride = 48 * 4;
const int kDstStride = 16 * 4;
const int kSize = kSrcStride * 3;
align_buffer_page_end(orig_pixels, kSize);
for (int i = 0; i < 48 * 3; ++i) {
orig_pixels[i * 4 + 0] = i;
orig_pixels[i * 4 + 1] = 255 - i;
orig_pixels[i * 4 + 2] = i + 1;
orig_pixels[i * 4 + 3] = i + 10;
}
align_buffer_page_end(dest_pixels, kDstStride);
int iterations16 =
benchmark_width_ * benchmark_height_ / (16 * 1) * benchmark_iterations_;
for (int i = 0; i < iterations16; ++i) {
ARGBScale(orig_pixels, kSrcStride, 48, 3, dest_pixels, kDstStride, 16, 1,
kFilterBilinear);
}
EXPECT_EQ(49, dest_pixels[0]);
EXPECT_EQ(255 - 49, dest_pixels[1]);
EXPECT_EQ(50, dest_pixels[2]);
EXPECT_EQ(59, dest_pixels[3]);
ARGBScale(orig_pixels, kSrcStride, 48, 3, dest_pixels, kDstStride, 16, 1,
kFilterNone);
EXPECT_EQ(49, dest_pixels[0]);
EXPECT_EQ(255 - 49, dest_pixels[1]);
EXPECT_EQ(50, dest_pixels[2]);
EXPECT_EQ(59, dest_pixels[3]);
free_aligned_buffer_page_end(dest_pixels);
free_aligned_buffer_page_end(orig_pixels);
}
TEST_F(LibYUVScaleTest, ARGBTest4x) {
const int kSrcStride = 64 * 4;
const int kDstStride = 16 * 4;
const int kSize = kSrcStride * 4;
align_buffer_page_end(orig_pixels, kSize);
for (int i = 0; i < 64 * 4; ++i) {
orig_pixels[i * 4 + 0] = i;
orig_pixels[i * 4 + 1] = 255 - i;
orig_pixels[i * 4 + 2] = i + 1;
orig_pixels[i * 4 + 3] = i + 10;
}
align_buffer_page_end(dest_pixels, kDstStride);
int iterations16 =
benchmark_width_ * benchmark_height_ / (16 * 1) * benchmark_iterations_;
for (int i = 0; i < iterations16; ++i) {
ARGBScale(orig_pixels, kSrcStride, 64, 4, dest_pixels, kDstStride, 16, 1,
kFilterBilinear);
}
EXPECT_NEAR((65 + 66 + 129 + 130 + 2) / 4, dest_pixels[0], 4);
EXPECT_NEAR((255 - 65 + 255 - 66 + 255 - 129 + 255 - 130 + 2) / 4,
dest_pixels[1], 4);
EXPECT_NEAR((1 * 4 + 65 + 66 + 129 + 130 + 2) / 4, dest_pixels[2], 4);
EXPECT_NEAR((10 * 4 + 65 + 66 + 129 + 130 + 2) / 4, dest_pixels[3], 4);
ARGBScale(orig_pixels, kSrcStride, 64, 4, dest_pixels, kDstStride, 16, 1,
kFilterNone);
EXPECT_EQ(130, dest_pixels[0]);
EXPECT_EQ(255 - 130, dest_pixels[1]);
EXPECT_EQ(130 + 1, dest_pixels[2]);
EXPECT_EQ(130 + 10, dest_pixels[3]);
free_aligned_buffer_page_end(dest_pixels);
free_aligned_buffer_page_end(orig_pixels);
}
} // namespace libyuv } // namespace libyuv

135
unit_test/scale_uv_test.cc
View File

@ -22,13 +22,13 @@ namespace libyuv {
// Test scaling with C vs Opt and return maximum pixel difference. 0 = exact. // Test scaling with C vs Opt and return maximum pixel difference. 0 = exact.
static int UVTestFilter(int src_width, static int UVTestFilter(int src_width,
int src_height, int src_height,
int dst_width, int dst_width,
int dst_height, int dst_height,
FilterMode f, FilterMode f,
int benchmark_iterations, int benchmark_iterations,
int disable_cpu_flags, int disable_cpu_flags,
int benchmark_cpu_info) { int benchmark_cpu_info) {
if (!SizeValid(src_width, src_height, dst_width, dst_height)) { if (!SizeValid(src_width, src_height, dst_width, dst_height)) {
return 0; return 0;
} }
@ -46,8 +46,7 @@ static int UVTestFilter(int src_width,
} }
MemRandomize(src_uv, src_uv_plane_size); MemRandomize(src_uv, src_uv_plane_size);
int64_t dst_uv_plane_size = int64_t dst_uv_plane_size = (dst_width + b * 2) * (dst_height + b * 2) * 2LL;
(dst_width + b * 2) * (dst_height + b * 2) * 2LL;
int dst_stride_uv = (b * 2 + dst_width) * 2; int dst_stride_uv = (b * 2 + dst_width) * 2;
align_buffer_page_end(dst_uv_c, dst_uv_plane_size); align_buffer_page_end(dst_uv_c, dst_uv_plane_size);
@ -61,29 +60,28 @@ static int UVTestFilter(int src_width,
// Warm up both versions for consistent benchmarks. // Warm up both versions for consistent benchmarks.
MaskCpuFlags(disable_cpu_flags); // Disable all CPU optimization. MaskCpuFlags(disable_cpu_flags); // Disable all CPU optimization.
UVScale(src_uv + (src_stride_uv * b) + b * 2, src_stride_uv, UVScale(src_uv + (src_stride_uv * b) + b * 2, src_stride_uv, src_width,
src_width, src_height, dst_uv_c + (dst_stride_uv * b) + b * 2, src_height, dst_uv_c + (dst_stride_uv * b) + b * 2, dst_stride_uv,
dst_stride_uv, dst_width, dst_height, f); dst_width, dst_height, f);
MaskCpuFlags(benchmark_cpu_info); // Enable all CPU optimization. MaskCpuFlags(benchmark_cpu_info); // Enable all CPU optimization.
UVScale(src_uv + (src_stride_uv * b) + b * 2, src_stride_uv, UVScale(src_uv + (src_stride_uv * b) + b * 2, src_stride_uv, src_width,
src_width, src_height, dst_uv_opt + (dst_stride_uv * b) + b * 2, src_height, dst_uv_opt + (dst_stride_uv * b) + b * 2, dst_stride_uv,
dst_stride_uv, dst_width, dst_height, f); dst_width, dst_height, f);
MaskCpuFlags(disable_cpu_flags); // Disable all CPU optimization. MaskCpuFlags(disable_cpu_flags); // Disable all CPU optimization.
double c_time = get_time(); double c_time = get_time();
UVScale(src_uv + (src_stride_uv * b) + b * 2, src_stride_uv, UVScale(src_uv + (src_stride_uv * b) + b * 2, src_stride_uv, src_width,
src_width, src_height, dst_uv_c + (dst_stride_uv * b) + b * 2, src_height, dst_uv_c + (dst_stride_uv * b) + b * 2, dst_stride_uv,
dst_stride_uv, dst_width, dst_height, f); dst_width, dst_height, f);
c_time = (get_time() - c_time); c_time = (get_time() - c_time);
MaskCpuFlags(benchmark_cpu_info); // Enable all CPU optimization. MaskCpuFlags(benchmark_cpu_info); // Enable all CPU optimization.
double opt_time = get_time(); double opt_time = get_time();
for (i = 0; i < benchmark_iterations; ++i) { for (i = 0; i < benchmark_iterations; ++i) {
UVScale(src_uv + (src_stride_uv * b) + b * 2, src_stride_uv, UVScale(src_uv + (src_stride_uv * b) + b * 2, src_stride_uv, src_width,
src_width, src_height, src_height, dst_uv_opt + (dst_stride_uv * b) + b * 2, dst_stride_uv,
dst_uv_opt + (dst_stride_uv * b) + b * 2, dst_stride_uv, dst_width, dst_height, f);
dst_width, dst_height, f);
} }
opt_time = (get_time() - opt_time) / benchmark_iterations; opt_time = (get_time() - opt_time) / benchmark_iterations;
@ -118,8 +116,8 @@ static int UVTestFilter(int src_width,
#define SX(x, nom, denom) static_cast<int>((x / nom) * denom) #define SX(x, nom, denom) static_cast<int>((x / nom) * denom)
#define TEST_FACTOR1(name, filter, nom, denom, max_diff) \ #define TEST_FACTOR1(name, filter, nom, denom, max_diff) \
TEST_F(LibYUVScaleTest, UVScaleDownBy##name##_##filter) { \ TEST_F(LibYUVScaleTest, UVScaleDownBy##name##_##filter) { \
int diff = UVTestFilter( \ int diff = UVTestFilter( \
SX(benchmark_width_, nom, denom), SX(benchmark_height_, nom, denom), \ SX(benchmark_width_, nom, denom), SX(benchmark_height_, nom, denom), \
DX(benchmark_width_, nom, denom), DX(benchmark_height_, nom, denom), \ DX(benchmark_width_, nom, denom), DX(benchmark_height_, nom, denom), \
kFilter##filter, benchmark_iterations_, disable_cpu_flags_, \ kFilter##filter, benchmark_iterations_, disable_cpu_flags_, \
@ -146,19 +144,19 @@ TEST_FACTOR(3, 1, 3)
#undef SX #undef SX
#undef DX #undef DX
#define TEST_SCALETO1(name, width, height, filter, max_diff) \ #define TEST_SCALETO1(name, width, height, filter, max_diff) \
TEST_F(LibYUVScaleTest, name##To##width##x##height##_##filter) { \ TEST_F(LibYUVScaleTest, name##To##width##x##height##_##filter) { \
int diff = UVTestFilter(benchmark_width_, benchmark_height_, width, \ int diff = UVTestFilter(benchmark_width_, benchmark_height_, width, \
height, kFilter##filter, benchmark_iterations_, \ height, kFilter##filter, benchmark_iterations_, \
disable_cpu_flags_, benchmark_cpu_info_); \ disable_cpu_flags_, benchmark_cpu_info_); \
EXPECT_LE(diff, max_diff); \ EXPECT_LE(diff, max_diff); \
} \ } \
TEST_F(LibYUVScaleTest, name##From##width##x##height##_##filter) { \ TEST_F(LibYUVScaleTest, name##From##width##x##height##_##filter) { \
int diff = UVTestFilter(width, height, Abs(benchmark_width_), \ int diff = UVTestFilter(width, height, Abs(benchmark_width_), \
Abs(benchmark_height_), kFilter##filter, \ Abs(benchmark_height_), kFilter##filter, \
benchmark_iterations_, disable_cpu_flags_, \ benchmark_iterations_, disable_cpu_flags_, \
benchmark_cpu_info_); \ benchmark_cpu_info_); \
EXPECT_LE(diff, max_diff); \ EXPECT_LE(diff, max_diff); \
} }
/// Test scale to a specified size with all 4 filters. /// Test scale to a specified size with all 4 filters.
@ -178,4 +176,67 @@ TEST_SCALETO(UVScale, 1920, 1080)
#undef TEST_SCALETO1 #undef TEST_SCALETO1
#undef TEST_SCALETO #undef TEST_SCALETO
TEST_F(LibYUVScaleTest, UVTest3x) {
const int kSrcStride = 48 * 2;
const int kDstStride = 16 * 2;
const int kSize = kSrcStride * 3;
align_buffer_page_end(orig_pixels, kSize);
for (int i = 0; i < 48 * 3; ++i) {
orig_pixels[i * 2 + 0] = i;
orig_pixels[i * 2 + 1] = 255 - i;
}
align_buffer_page_end(dest_pixels, kDstStride);
int iterations16 =
benchmark_width_ * benchmark_height_ / (16 * 1) * benchmark_iterations_;
for (int i = 0; i < iterations16; ++i) {
UVScale(orig_pixels, kSrcStride, 48, 3, dest_pixels, kDstStride, 16, 1,
kFilterBilinear);
}
EXPECT_EQ(49, dest_pixels[0]);
EXPECT_EQ(255 - 49, dest_pixels[1]);
UVScale(orig_pixels, kSrcStride, 48, 3, dest_pixels, kDstStride, 16, 1,
kFilterNone);
EXPECT_EQ(49, dest_pixels[0]);
EXPECT_EQ(255 - 49, dest_pixels[1]);
free_aligned_buffer_page_end(dest_pixels);
free_aligned_buffer_page_end(orig_pixels);
}
TEST_F(LibYUVScaleTest, UVTest4x) {
const int kSrcStride = 64 * 2;
const int kDstStride = 16 * 2;
const int kSize = kSrcStride * 4;
align_buffer_page_end(orig_pixels, kSize);
for (int i = 0; i < 64 * 4; ++i) {
orig_pixels[i * 2 + 0] = i;
orig_pixels[i * 2 + 1] = 255 - i;
}
align_buffer_page_end(dest_pixels, kDstStride);
int iterations16 =
benchmark_width_ * benchmark_height_ / (16 * 1) * benchmark_iterations_;
for (int i = 0; i < iterations16; ++i) {
UVScale(orig_pixels, kSrcStride, 64, 4, dest_pixels, kDstStride, 16, 1,
kFilterBilinear);
}
EXPECT_EQ((65 + 66 + 129 + 130 + 2) / 4, dest_pixels[0]);
EXPECT_EQ((255 - 65 + 255 - 66 + 255 - 129 + 255 - 130 + 2) / 4,
dest_pixels[1]);
UVScale(orig_pixels, kSrcStride, 64, 4, dest_pixels, kDstStride, 16, 1,
kFilterNone);
EXPECT_EQ(130, dest_pixels[0]); // expect the 3rd pixel of the 3rd row
EXPECT_EQ(255 - 130, dest_pixels[1]);
free_aligned_buffer_page_end(dest_pixels);
free_aligned_buffer_page_end(orig_pixels);
}
} // namespace libyuv } // namespace libyuv