Skip to content

O
opencv
Commit f07856ea authored by Evgeny Latkin's avatar Evgeny Latkin Committed by Alexander Alekhin
Browse files
Options

Merge pull request #13221 from elatkin:el/gapi_perf_sepfilter 

GAPI (fluid): optimization of Separable filter (#13221)

* GAPI (fluid): Separable filter: performance test

* GAPI (fluid): enable all performance tests

* GAPI: separable filters: alternative code for Sobel

* GAPI (fluid): hide unused old code for Sobel filter

* GAPI (fluid): especial code for Sobel if U8 into S16

* GAPI (fluid): back to old code for Sobel

* GAPI (fluid): run_sepfilter3x3_impl() with CPU dispatcher

* GAPI (fluid): run_sepfilter3x3_impl(): fix compiler warnings

* GAPI (fluid): new engine for separable filters (but Sobel)

* GAPI (fluid): new performance engine for Sobel

* GAPI (fluid): Sepfilters performance: fixed compilation error
parent dd952f6d
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 191 additions and 99 deletions
modules/gapi/perf/common/gapi_imgproc_perf_tests_inl.hpp
View file @ f07856ea
......@@ -52,7 +52,7 @@ PERF_TEST_P_(SepFilterPerfTest, TestPerformance)
TEST_CYCLE()
{
c.apply(in_mat1, out_mat_gapi, std::move(compile_args));
c.apply(in_mat1, out_mat_gapi);
}
// Comparison //////////////////////////////////////////////////////////////
......@@ -100,7 +100,7 @@ PERF_TEST_P_(Filter2DPerfTest, TestPerformance)
TEST_CYCLE()
{
c.apply(in_mat1, out_mat_gapi, std::move(compile_args));
c.apply(in_mat1, out_mat_gapi);
}
// Comparison //////////////////////////////////////////////////////////////
......@@ -145,7 +145,7 @@ PERF_TEST_P_(BoxFilterPerfTest, TestPerformance)
TEST_CYCLE()
{
c.apply(in_mat1, out_mat_gapi, std::move(compile_args));
c.apply(in_mat1, out_mat_gapi);
}
// Comparison //////////////////////////////////////////////////////////////
......@@ -188,7 +188,7 @@ PERF_TEST_P_(BlurPerfTest, TestPerformance)
TEST_CYCLE()
{
c.apply(in_mat1, out_mat_gapi, std::move(compile_args));
c.apply(in_mat1, out_mat_gapi);
}
// Comparison //////////////////////////////////////////////////////////////
......@@ -230,7 +230,7 @@ PERF_TEST_P_(GaussianBlurPerfTest, TestPerformance)
TEST_CYCLE()
{
c.apply(in_mat1, out_mat_gapi, std::move(compile_args));
c.apply(in_mat1, out_mat_gapi);
}
// Comparison //////////////////////////////////////////////////////////////
......@@ -271,7 +271,7 @@ PERF_TEST_P_(MedianBlurPerfTest, TestPerformance)
TEST_CYCLE()
{
c.apply(in_mat1, out_mat_gapi, std::move(compile_args));
c.apply(in_mat1, out_mat_gapi);
}
// Comparison //////////////////////////////////////////////////////////////
......@@ -314,7 +314,7 @@ PERF_TEST_P_(ErodePerfTest, TestPerformance)
TEST_CYCLE()
{
c.apply(in_mat1, out_mat_gapi, std::move(compile_args));
c.apply(in_mat1, out_mat_gapi);
}
// Comparison //////////////////////////////////////////////////////////////
......@@ -357,7 +357,7 @@ PERF_TEST_P_(Erode3x3PerfTest, TestPerformance)
TEST_CYCLE()
{
c.apply(in_mat1, out_mat_gapi, std::move(compile_args));
c.apply(in_mat1, out_mat_gapi);
}
// Comparison //////////////////////////////////////////////////////////////
......@@ -400,7 +400,7 @@ PERF_TEST_P_(DilatePerfTest, TestPerformance)
TEST_CYCLE()
{
c.apply(in_mat1, out_mat_gapi, std::move(compile_args));
c.apply(in_mat1, out_mat_gapi);
}
// Comparison //////////////////////////////////////////////////////////////
......@@ -443,7 +443,7 @@ PERF_TEST_P_(Dilate3x3PerfTest, TestPerformance)
TEST_CYCLE()
{
c.apply(in_mat1, out_mat_gapi, std::move(compile_args));
c.apply(in_mat1, out_mat_gapi);
}
// Comparison //////////////////////////////////////////////////////////////
......@@ -526,7 +526,7 @@ PERF_TEST_P_(CannyPerfTest, TestPerformance)
TEST_CYCLE()
{
c.apply(in_mat1, out_mat_gapi, std::move(compile_args));
c.apply(in_mat1, out_mat_gapi);
}
// Comparison //////////////////////////////////////////////////////////////
......@@ -564,7 +564,7 @@ PERF_TEST_P_(EqHistPerfTest, TestPerformance)
TEST_CYCLE()
{
c.apply(in_mat1, out_mat_gapi, std::move(compile_args));
c.apply(in_mat1, out_mat_gapi);
}
// Comparison //////////////////////////////////////////////////////////////
......@@ -830,7 +830,7 @@ PERF_TEST_P_(LUV2BGRPerfTest, TestPerformance)
TEST_CYCLE()
{
c.apply(in_mat1, out_mat_gapi, std::move(compile_args));
c.apply(in_mat1, out_mat_gapi);
}
// Comparison //////////////////////////////////////////////////////////////
......
modules/gapi/perf/cpu/gapi_imgproc_perf_tests_fluid.cpp
View file @ f07856ea
......@@ -13,9 +13,101 @@
namespace opencv_test
{
INSTANTIATE_TEST_CASE_P(SobelPerfTestFluid, SobelPerfTest,
Combine(Values(AbsExact().to_compare_f()),
Values(CV_8UC1, CV_8UC3, CV_16UC1, CV_16SC1), // add CV_32FC1 when ready
INSTANTIATE_TEST_CASE_P(SepFilterPerfTestFluid_8U, SepFilterPerfTest,
Combine(Values(ToleranceFilter(1e-4f, 0.01).to_compare_f()),
Values(CV_8UC1, CV_8UC3),
Values(3),
Values(szVGA, sz720p, sz1080p),
Values(-1, CV_16S, CV_32F),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(SepFilterPerfTestFluid_other, SepFilterPerfTest,
Combine(Values(ToleranceFilter(1e-4f, 0.01).to_compare_f()),
Values(CV_16UC1, CV_16SC1, CV_32FC1),
Values(3),
Values(szVGA, sz720p, sz1080p),
Values(-1, CV_32F),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(Filter2DPerfTestFluid, Filter2DPerfTest,
Combine(Values(ToleranceFilter(1e-4f, 0.01).to_compare_f()),
Values(CV_8UC1, CV_8UC3, CV_16UC1, CV_16SC1, CV_32FC1),
Values(3), // add 4, 5, 7 when kernel is ready
Values(szVGA, sz720p, sz1080p),
Values(cv::BORDER_DEFAULT),
Values(-1, CV_32F),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(BoxFilterPerfTestFluid, BoxFilterPerfTest,
Combine(Values(ToleranceFilter(1e-4f, 0.01).to_compare_f()),
Values(CV_8UC1, CV_8UC3, CV_16UC1, CV_16SC1, CV_32FC1),
Values(3), // add size=5, when kernel is ready
Values(szVGA, sz720p, sz1080p),
Values(cv::BORDER_DEFAULT),
Values(-1, CV_32F),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(BlurPerfTestFluid, BlurPerfTest,
Combine(Values(ToleranceFilter(1e-4f, 0.01).to_compare_f()),
Values(CV_8UC1, CV_8UC3, CV_16UC1, CV_16SC1, CV_32FC1),
Values(3), // add size=5, when kernel is ready
Values(szVGA, sz720p, sz1080p),
Values(cv::BORDER_DEFAULT),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(GaussianBlurPerfTestFluid, GaussianBlurPerfTest,
Combine(Values(ToleranceFilter(1e-3f, 0.01).to_compare_f()),
Values(CV_8UC1, CV_8UC3, CV_16UC1, CV_16SC1, CV_32FC1),
Values(3), // add size=5, when kernel is ready
Values(szVGA, sz720p, sz1080p),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(MedianBlurPerfTestFluid, MedianBlurPerfTest,
Combine(Values(AbsExact().to_compare_f()),
Values(CV_8UC1, CV_8UC3, CV_16UC1, CV_16SC1, CV_32FC1),
Values(3), // add size=5, when kernel is ready
Values(szVGA, sz720p, sz1080p),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(ErodePerfTestFluid, ErodePerfTest,
Combine(Values(AbsExact().to_compare_f()),
Values(CV_8UC1, CV_8UC3, CV_16UC1, CV_16SC1, CV_32FC1),
Values(3), // add size=5, when kernel is ready
Values(szVGA, sz720p, sz1080p),
Values(cv::MorphShapes::MORPH_RECT,
cv::MorphShapes::MORPH_CROSS,
cv::MorphShapes::MORPH_ELLIPSE),
Values(cv::compile_args(IMGPROC_FLUID))));
// GAPI/fluid does not support iterations parameter for the Erode kernel
INSTANTIATE_TEST_CASE_P(DISABLED_Erode3x3PerfTestFluid, Erode3x3PerfTest,
Combine(Values(AbsExact().to_compare_f()),
Values(CV_8UC1, CV_8UC3, CV_16UC1, CV_16SC1, CV_32FC1),
Values(szVGA, sz720p, sz1080p),
Values(1, 2, 4),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(DilatePerfTestFluid, DilatePerfTest,
Combine(Values(AbsExact().to_compare_f()),
Values(CV_8UC1, CV_8UC3, CV_16UC1, CV_16SC1, CV_32FC1),
Values(3), // add size=5, when kernel is ready
Values(szVGA, sz720p, sz1080p),
Values(cv::MorphShapes::MORPH_RECT,
cv::MorphShapes::MORPH_CROSS,
cv::MorphShapes::MORPH_ELLIPSE),
Values(cv::compile_args(IMGPROC_FLUID))));
// GAPI/fluid does not support iterations parameter for the Dilate kernel
INSTANTIATE_TEST_CASE_P(DISABLED_Dilate3x3PerfTestFluid, Dilate3x3PerfTest,
Combine(Values(AbsExact().to_compare_f()),
Values(CV_8UC1, CV_8UC3, CV_16UC1, CV_16SC1, CV_32FC1),
Values(szVGA, sz720p, sz1080p),
Values(1, 2, 4),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(SobelPerfTestFluid, SobelPerfTest,
Combine(Values(AbsExact().to_compare_f()),
Values(CV_8UC1, CV_8UC3, CV_16UC1, CV_16SC1),
Values(3), // add 5x5 once supported
Values(szVGA, sz720p, sz1080p),
Values(-1, CV_16S, CV_32F),
......@@ -23,8 +115,8 @@ namespace opencv_test
Values(1, 2),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(SobelPerfTestFluid32F, SobelPerfTest,
Combine(Values(ToleranceFilter(1e-3f, 0.0).to_compare_f()),
INSTANTIATE_TEST_CASE_P(SobelPerfTestFluid32F, SobelPerfTest,
Combine(Values(ToleranceFilter(1e-3f, 0.0).to_compare_f()),
Values(CV_32FC1),
Values(3), // add 5x5 once supported
Values(szVGA, sz720p, sz1080p),
......@@ -33,44 +125,44 @@ namespace opencv_test
Values(1, 2),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(RGB2GrayPerfTestFluid, RGB2GrayPerfTest,
Combine(Values(ToleranceColor(1e-3).to_compare_f()),
Values(szVGA, sz720p, sz1080p),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(BGR2GrayPerfTestFluid, BGR2GrayPerfTest,
Combine(Values(ToleranceColor(1e-3).to_compare_f()),
Values(szVGA, sz720p, sz1080p),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(RGB2YUVPerfTestFluid, RGB2YUVPerfTest,
Combine(Values(ToleranceColor(1e-3).to_compare_f()),
Values(szVGA, sz720p, sz1080p),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(YUV2RGBPerfTestFluid, YUV2RGBPerfTest,
Combine(Values(ToleranceColor(1e-3).to_compare_f()),
Values(szVGA, sz720p, sz1080p),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(BGR2YUVPerfTestFluid, BGR2YUVPerfTest,
Combine(Values(ToleranceColor(1e-3).to_compare_f()),
Values(szVGA, sz720p, sz1080p),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(YUV2BGRPerfTestFluid, YUV2BGRPerfTest,
Combine(Values(ToleranceColor(1e-3).to_compare_f()),
Values(szVGA, sz720p, sz1080p),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(BGR2LUVPerfTestFluid, BGR2LUVPerfTest,
Combine(Values(AbsSimilarPoints(1, 0.05).to_compare_f()),
Values(szVGA, sz720p, sz1080p),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(RGB2LabPerfTestFluid, RGB2LabPerfTest,
Combine(Values(AbsSimilarPoints(1, 0.05).to_compare_f()),
Values(szVGA, sz720p, sz1080p),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(RGB2GrayPerfTestFluid, RGB2GrayPerfTest,
Combine(Values(ToleranceColor(1e-3).to_compare_f()),
Values(szVGA, sz720p, sz1080p),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(BGR2GrayPerfTestFluid, BGR2GrayPerfTest,
Combine(Values(ToleranceColor(1e-3).to_compare_f()),
Values(szVGA, sz720p, sz1080p),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(RGB2YUVPerfTestFluid, RGB2YUVPerfTest,
Combine(Values(ToleranceColor(1e-3).to_compare_f()),
Values(szVGA, sz720p, sz1080p),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(YUV2RGBPerfTestFluid, YUV2RGBPerfTest,
Combine(Values(ToleranceColor(1e-3).to_compare_f()),
Values(szVGA, sz720p, sz1080p),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(BGR2YUVPerfTestFluid, BGR2YUVPerfTest,
Combine(Values(ToleranceColor(1e-3).to_compare_f()),
Values(szVGA, sz720p, sz1080p),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(YUV2BGRPerfTestFluid, YUV2BGRPerfTest,
Combine(Values(ToleranceColor(1e-3).to_compare_f()),
Values(szVGA, sz720p, sz1080p),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(BGR2LUVPerfTestFluid, BGR2LUVPerfTest,
Combine(Values(AbsSimilarPoints(1, 0.05).to_compare_f()),
Values(szVGA, sz720p, sz1080p),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(RGB2LabPerfTestFluid, RGB2LabPerfTest,
Combine(Values(AbsSimilarPoints(1, 0.05).to_compare_f()),
Values(szVGA, sz720p, sz1080p),
Values(cv::compile_args(IMGPROC_FLUID))));
}
modules/gapi/src/backends/fluid/gfluidimgproc_func.dispatch.cpp
View file @ f07856ea
......@@ -57,34 +57,34 @@ void run_yuv2rgb_impl(uchar out[], const uchar in[], int width, const float coef
CV_CPU_DISPATCH(run_yuv2rgb_impl, (out, in, width, coef), CV_CPU_DISPATCH_MODES_ALL);
}
//---------------------
//-------------------------
//
// Fluid kernels: Sobel
// Fluid kernels: sepFilter
//
//---------------------
#define RUN_SOBEL_ROW(DST, SRC) \
void run_sobel_row(DST out[], const SRC *in[], int width, int chan, \
const float kx[], const float ky[], int border, \
float scale, float delta, float *buf[], \
int y, int y0) \
{ \
CV_CPU_DISPATCH(run_sobel_row, \
(out, in, width, chan, kx, ky, border, scale, delta, buf,y, y0), \
CV_CPU_DISPATCH_MODES_ALL); \
//-------------------------
#define RUN_SEPFILTER3X3_IMPL(DST, SRC) \
void run_sepfilter3x3_impl(DST out[], const SRC *in[], int width, int chan, \
const float kx[], const float ky[], int border, \
float scale, float delta, \
float *buf[], int y, int y0) \
{ \
CV_CPU_DISPATCH(run_sepfilter3x3_impl, \
(out, in, width, chan, kx, ky, border, scale, delta, buf,y, y0), \
CV_CPU_DISPATCH_MODES_ALL); \
}
RUN_SOBEL_ROW(uchar , uchar )
RUN_SOBEL_ROW(ushort, ushort)
RUN_SOBEL_ROW( short, uchar )
RUN_SOBEL_ROW( short, ushort)
RUN_SOBEL_ROW( short, short)
RUN_SOBEL_ROW( float, uchar )
RUN_SOBEL_ROW( float, ushort)
RUN_SOBEL_ROW( float, short)
RUN_SOBEL_ROW( float, float)
#undef RUN_SOBEL_ROW
RUN_SEPFILTER3X3_IMPL(uchar , uchar )
RUN_SEPFILTER3X3_IMPL( short, uchar )
RUN_SEPFILTER3X3_IMPL( float, uchar )
RUN_SEPFILTER3X3_IMPL(ushort, ushort)
RUN_SEPFILTER3X3_IMPL( short, ushort)
RUN_SEPFILTER3X3_IMPL( float, ushort)
RUN_SEPFILTER3X3_IMPL( short, short)
RUN_SEPFILTER3X3_IMPL( float, short)
RUN_SEPFILTER3X3_IMPL( float, float)
#undef RUN_SEPFILTER3X3_IMPL
} // namespace fliud
} // namespace gapi
......
modules/gapi/src/backends/fluid/gfluidimgproc_func.hpp
View file @ f07856ea
......@@ -33,29 +33,29 @@ void run_rgb2yuv_impl(uchar out[], const uchar in[], int width, const float coef
void run_yuv2rgb_impl(uchar out[], const uchar in[], int width, const float coef[4]);
//---------------------
//-------------------------
//
// Fluid kernels: Sobel
// Fluid kernels: sepFilter
//
//---------------------
//-------------------------
#define RUN_SOBEL_ROW(DST, SRC) \
void run_sobel_row(DST out[], const SRC *in[], int width, int chan, \
const float kx[], const float ky[], int border, \
float scale, float delta, float *buf[], \
int y, int y0);
#define RUN_SEPFILTER3X3_IMPL(DST, SRC) \
void run_sepfilter3x3_impl(DST out[], const SRC *in[], int width, int chan, \
const float kx[], const float ky[], int border, \
float scale, float delta, \
float *buf[], int y, int y0);
RUN_SOBEL_ROW(uchar , uchar )
RUN_SOBEL_ROW(ushort, ushort)
RUN_SOBEL_ROW( short, uchar )
RUN_SOBEL_ROW( short, ushort)
RUN_SOBEL_ROW( short, short)
RUN_SOBEL_ROW( float, uchar )
RUN_SOBEL_ROW( float, ushort)
RUN_SOBEL_ROW( float, short)
RUN_SOBEL_ROW( float, float)
RUN_SEPFILTER3X3_IMPL(uchar , uchar )
RUN_SEPFILTER3X3_IMPL( short, uchar )
RUN_SEPFILTER3X3_IMPL( float, uchar )
RUN_SEPFILTER3X3_IMPL(ushort, ushort)
RUN_SEPFILTER3X3_IMPL( short, ushort)
RUN_SEPFILTER3X3_IMPL( float, ushort)
RUN_SEPFILTER3X3_IMPL( short, short)
RUN_SEPFILTER3X3_IMPL( float, short)
RUN_SEPFILTER3X3_IMPL( float, float)
#undef RUN_SOBEL_ROW
#undef RUN_SEPFILTER3X3_IMPL
} // namespace fluid
} // namespace gapi
......
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