potok/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2026-07-30 15:53:03 +04:00
Code Packages Releases Wiki Activity

Merge pull request #13070 from elatkin:el/gapi_perf_sobel

Browse Source 
GAPI (fluid): optimization of Sobel 3x3 (#13070)

* GAPI: performance test for Sobel

* GAPI: performance test for Sobel w/FP32 input

* GAPI: Sobel speedup: 2.5x (U8) up to 10x (float)

* GAPI: Sobel 3x3 to support U8 into S16

* GAPI (fluid): Sobel 3x3 speedup: 10% (uchar), 1.5x (float)

* GAPI (fluid): Sobel 3x3 speedup: +10x (uchar), but -20% (float)

* GAPI (fluid): Sobel 3x3 speedup: +10% (float)

* GAPI (fluid): Sobel 3x3 speedup: +15% (float), +10% (uchar)

* GAPI (fluid): Sobel 3x3: address GCC warnings

* GAPI (fluid): Sobel 3x3: separate *.cpp file w/SIMD code

* GAPI (fluid): Sobel 3x3: fixed AVX2 code, AVX2 speedup 20-50% (uchar), 10-20% (float)

* GAPI (fluid): Sobel 3x3: fix CV_SIMD code for AVX2

* GAPI (fluid): Sobel 3x3: refactor
This commit is contained in:
Evgeny Latkin
2018-11-13 15:04:37 +03:00
committed by Alexander Alekhin
parent a456b968cf
commit 4e40e5bb88
10 changed files with 420 additions and 51 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/gapi/perf/common/gapi_imgproc_perf_tests_inl.hpp
+2 -3
View File
@@ -476,7 +476,7 @@ PERF_TEST_P_(SobelPerfTest, TestPerformance)
// G-API code //////////////////////////////////////////////////////////////
cv::GMat in;
auto out = cv::gapi::Sobel(in, dtype, dx, dy, kernSize );
auto out = cv::gapi::Sobel(in, dtype, dx, dy, kernSize);
cv::GComputation c(in, out);
// Warm-up graph engine:
@@ -484,7 +484,7 @@ PERF_TEST_P_(SobelPerfTest, TestPerformance)
TEST_CYCLE()
{
c.apply(in_mat1, out_mat_gapi, std::move(compile_args));
c.apply(in_mat1, out_mat_gapi);
}
// Comparison //////////////////////////////////////////////////////////////
@@ -494,7 +494,6 @@ PERF_TEST_P_(SobelPerfTest, TestPerformance)
}
SANITY_CHECK_NOTHING();
}
//------------------------------------------------------------------------------
modules/gapi/perf/cpu/gapi_imgproc_perf_tests_cpu.cpp
+12 -4
View File
@@ -31,8 +31,6 @@ INSTANTIATE_TEST_CASE_P(SepFilterPerfTestCPU_other, SepFilterPerfTest,
Values(-1, CV_32F),
Values(cv::compile_args(IMGPROC_CPU))));
INSTANTIATE_TEST_CASE_P(Filter2DPerfTestCPU, Filter2DPerfTest,
Combine(Values(AbsExact().to_compare_f()),
Values(CV_8UC1, CV_8UC3, CV_16UC1, CV_16SC1, CV_32FC1),
@@ -109,10 +107,20 @@ INSTANTIATE_TEST_CASE_P(Dilate3x3PerfTestCPU, Dilate3x3PerfTest,
INSTANTIATE_TEST_CASE_P(SobelPerfTestCPU, SobelPerfTest,
Combine(Values(AbsExact().to_compare_f()),
Values(CV_8UC1, CV_8UC3, CV_16UC1, CV_16SC1, CV_32FC1),
Values(CV_8UC1, CV_8UC3, CV_16UC1, CV_16SC1),
Values(3, 5),
Values(szVGA, sz720p, sz1080p),
Values(-1, CV_32F),
Values(-1, CV_16S, CV_32F),
Values(0, 1),
Values(1, 2),
Values(cv::compile_args(IMGPROC_CPU))));
INSTANTIATE_TEST_CASE_P(SobelPerfTestCPU32F, SobelPerfTest,
Combine(Values(AbsExact().to_compare_f()),
Values(CV_32FC1),
Values(3, 5),
Values(szVGA, sz720p, sz1080p),
Values(CV_32F),
Values(0, 1),
Values(1, 2),
Values(cv::compile_args(IMGPROC_CPU))));
modules/gapi/perf/cpu/gapi_imgproc_perf_tests_fluid.cpp
+38
View File
@@ -0,0 +1,38 @@
// This file is part of OpenCV project.
// It is subject to the license terms in the LICENSE file found in the top-level directory
// of this distribution and at http://opencv.org/license.html.
//
// Copyright (C) 2018 Intel Corporation
#include "../perf_precomp.hpp"
#include "../common/gapi_imgproc_perf_tests.hpp"
#include "../../src/backends/fluid/gfluidimgproc.hpp"
#define IMGPROC_FLUID cv::gapi::imgproc::fluid::kernels()
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
Values(3), // add 5x5 once supported
Values(szVGA, sz720p, sz1080p),
Values(-1, CV_16S, CV_32F),
Values(0, 1),
Values(1, 2),
Values(cv::compile_args(IMGPROC_FLUID))));
INSTANTIATE_TEST_CASE_P(SobelPerfTestFluid32F, SobelPerfTest,
Combine(Values(AbsToleranceSobel(1e-3).to_compare_f()),
Values(CV_32FC1),
Values(3), // add 5x5 once supported
Values(szVGA, sz720p, sz1080p),
Values(CV_32F),
Values(0, 1),
Values(1, 2),
Values(cv::compile_args(IMGPROC_FLUID))));
}