Chiebot-Mirror
/
opencv
mirror of https://github.com/opencv/opencv.git
8
0
Fork 0
Code Issues Projects Releases Wiki Activity

Merge pull request #713 from bitwangyaoyao:2.4_perf

Browse Source
pull/728/head
Andrey Kamaev 12 years ago committed by OpenCV Buildbot
parent a3954fb223 5539e85a11
commit 2a1cf23fab
26 changed files with 3525 additions and 14039 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Split View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 120
      modules/ocl/perf/interpolation.hpp
  2. 182
      modules/ocl/perf/main.cpp
  3. 4935
      modules/ocl/perf/perf_arithm.cpp
  4. 128
      modules/ocl/perf/perf_blend.cpp
  5. 150
      modules/ocl/perf/perf_brute_force_matcher.cpp
  6. 122
      modules/ocl/perf/perf_canny.cpp
  7. 91
      modules/ocl/perf/perf_color.cpp
  8. 112
      modules/ocl/perf/perf_columnsum.cpp
  9. 99
      modules/ocl/perf/perf_fft.cpp
  10. 1281
      modules/ocl/perf/perf_filters.cpp
  11. 103
      modules/ocl/perf/perf_gemm.cpp
  12. 178
      modules/ocl/perf/perf_haar.cpp
  13. 140
      modules/ocl/perf/perf_hog.cpp
  14. 2465
      modules/ocl/perf/perf_imgproc.cpp
  15. 234
      modules/ocl/perf/perf_match_template.cpp
  16. 743
      modules/ocl/perf/perf_matrix_operation.cpp
  17. 84
      modules/ocl/perf/perf_norm.cpp
  18. 114
      modules/ocl/perf/perf_pyrdown.cpp
  19. 143
      modules/ocl/perf/perf_pyrlk.cpp
  20. 97
      modules/ocl/perf/perf_pyrup.cpp
  21. 483
      modules/ocl/perf/perf_split_merge.cpp
  22. 330
      modules/ocl/perf/precomp.cpp
  23. 386
      modules/ocl/perf/precomp.hpp
  24. 265
      modules/ocl/perf/utility.cpp
  25. 182
      modules/ocl/perf/utility.hpp
  26. 4397
      samples/ocl/performance.cpp

120
modules/ocl/perf/interpolation.hpp
Unescape View File

@ -1,120 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// Intel License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000, Intel Corporation, all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of Intel Corporation may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#ifndef __OPENCV_TEST_INTERPOLATION_HPP__
#define __OPENCV_TEST_INTERPOLATION_HPP__
template <typename T> T readVal(const cv::Mat &src, int y, int x, int c, int border_type, cv::Scalar borderVal = cv::Scalar())
{
if (border_type == cv::BORDER_CONSTANT)
return (y >= 0 && y < src.rows && x >= 0 && x < src.cols) ? src.at<T>(y, x * src.channels() + c) : cv::saturate_cast<T>(borderVal.val[c]);
return src.at<T>(cv::borderInterpolate(y, src.rows, border_type), cv::borderInterpolate(x, src.cols, border_type) * src.channels() + c);
}
template <typename T> struct NearestInterpolator
{
static T getValue(const cv::Mat &src, float y, float x, int c, int border_type, cv::Scalar borderVal = cv::Scalar())
{
return readVal<T>(src, cvFloor(y), cvFloor(x), c, border_type, borderVal);
}
};
template <typename T> struct LinearInterpolator
{
static T getValue(const cv::Mat &src, float y, float x, int c, int border_type, cv::Scalar borderVal = cv::Scalar())
{
x -= 0.5f;
y -= 0.5f;
int x1 = cvFloor(x);
int y1 = cvFloor(y);
int x2 = x1 + 1;
int y2 = y1 + 1;
float res = 0;
res += readVal<T>(src, y1, x1, c, border_type, borderVal) * ((x2 - x) * (y2 - y));
res += readVal<T>(src, y1, x2, c, border_type, borderVal) * ((x - x1) * (y2 - y));
res += readVal<T>(src, y2, x1, c, border_type, borderVal) * ((x2 - x) * (y - y1));
res += readVal<T>(src, y2, x2, c, border_type, borderVal) * ((x - x1) * (y - y1));
return cv::saturate_cast<T>(res);
}
};
template <typename T> struct CubicInterpolator
{
static float getValue(float p[4], float x)
{
return p[1] + 0.5 * x * (p[2] - p[0] + x * (2.0 * p[0] - 5.0 * p[1] + 4.0 * p[2] - p[3] + x * (3.0 * (p[1] - p[2]) + p[3] - p[0])));
}
static float getValue(float p[4][4], float x, float y)
{
float arr[4];
arr[0] = getValue(p[0], x);
arr[1] = getValue(p[1], x);
arr[2] = getValue(p[2], x);
arr[3] = getValue(p[3], x);
return getValue(arr, y);
}
static T getValue(const cv::Mat &src, float y, float x, int c, int border_type, cv::Scalar borderVal = cv::Scalar())
{
int ix = cvRound(x);
int iy = cvRound(y);
float vals[4][4] =
{
{readVal<T>(src, iy - 2, ix - 2, c, border_type, borderVal), readVal<T>(src, iy - 2, ix - 1, c, border_type, borderVal), readVal<T>(src, iy - 2, ix, c, border_type, borderVal), readVal<T>(src, iy - 2, ix + 1, c, border_type, borderVal)},
{readVal<T>(src, iy - 1, ix - 2, c, border_type, borderVal), readVal<T>(src, iy - 1, ix - 1, c, border_type, borderVal), readVal<T>(src, iy - 1, ix, c, border_type, borderVal), readVal<T>(src, iy - 1, ix + 1, c, border_type, borderVal)},
{readVal<T>(src, iy , ix - 2, c, border_type, borderVal), readVal<T>(src, iy , ix - 1, c, border_type, borderVal), readVal<T>(src, iy , ix, c, border_type, borderVal), readVal<T>(src, iy , ix + 1, c, border_type, borderVal)},
{readVal<T>(src, iy + 1, ix - 2, c, border_type, borderVal), readVal<T>(src, iy + 1, ix - 1, c, border_type, borderVal), readVal<T>(src, iy + 1, ix, c, border_type, borderVal), readVal<T>(src, iy + 1, ix + 1, c, border_type, borderVal)},
};
return cv::saturate_cast<T>(getValue(vals, (x - ix + 2.0) / 4.0, (y - iy + 2.0) / 4.0));
}
};
#endif // __OPENCV_TEST_INTERPOLATION_HPP__

182
modules/ocl/perf/main.cpp
Unescape View File

@ -7,12 +7,13 @@
// copy or use the software.
//
//
// Intel License Agreement
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000, Intel Corporation, all rights reserved.
// Copyright (C) 2010-2012, Multicoreware, Inc., all rights reserved.
// Copyright (C) 2010-2012, Advanced Micro Devices, Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
@ -21,12 +22,12 @@
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
// and/or other oclMaterials provided with the distribution.
//
// * The name of Intel Corporation may not be used to endorse or promote products
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// This software is provided by the copyright holders and contributors as is and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
@ -41,129 +42,118 @@
#include "precomp.hpp"
#ifdef HAVE_OPENCL
using namespace std;
using namespace cv;
using namespace cv::ocl;
using namespace cvtest;
using namespace testing;
void print_info()
{
printf("\n");
#if defined _WIN32
# if defined _WIN64
puts("OS: Windows 64");
# else
puts("OS: Windows 32");
# endif
#elif defined linux
# if defined _LP64
puts("OS: Linux 64");
# else
puts("OS: Linux 32");
# endif
#elif defined __APPLE__
# if defined _LP64
puts("OS: Apple 64");
# else
puts("OS: Apple 32");
# endif
#endif
}
std::string workdir;
int main(int argc, char **argv)
int main(int argc, const char *argv[])
{
TS::ptr()->init("ocl");
InitGoogleTest(&argc, argv);
const char *keys =
"{ h | help | false | print help message }"
"{ w | workdir | ../../../samples/c/| set working directory }"
vector<ocl::Info> oclinfo;
int num_devices = getDevice(oclinfo);
"{ t | type | gpu | set device type:cpu or gpu}"
if (num_devices < 1)
{
cerr << "no device found\n";
return -1;
}
"{ p | platform | 0 | set platform id }"
int devidx = 0;
"{ d | device | 0 | set device id }";
for (size_t i = 0; i < oclinfo.size(); i++)
{
for (size_t j = 0; j < oclinfo[i].DeviceName.size(); j++)
{
printf("device %d: %s\n", devidx++, oclinfo[i].DeviceName[j].c_str());
}
}
redirectError(cvErrorCallback);
const char *keys =
"{ h | help | false | print help message }"
"{ f | filter | | filter for test }"
"{ w | workdir | | set working directory }"
"{ l | list | false | show all tests }"
"{ d | device | 0 | device id }"
"{ i | iters | 10 | iteration count }"
"{ m | warmup | 1 | gpu warm up iteration count}"
"{ t | xtop | 1.1 | xfactor top boundary}"
"{ b | xbottom | 0.9 | xfactor bottom boundary}"
"{ v | verify | false | only run gpu once to verify if problems occur}";
CommandLineParser cmd(argc, argv, keys);
if (cmd.get<bool>("help"))
{
cout << "Avaible options besides goole test option:" << endl;
cout << "Avaible options:" << endl;
cmd.printParams();
return 0;
}
workdir = cmd.get<string>("workdir");
string type = cmd.get<string>("type");
unsigned int pid = cmd.get<unsigned int>("platform");
int device = cmd.get<int>("device");
print_info();
// int flag = CVCL_DEVICE_TYPE_GPU;
// if(type == "cpu")
// {
// flag = CVCL_DEVICE_TYPE_CPU;
// }
std::vector<cv::ocl::Info> oclinfo;
int devnums = getDevice(oclinfo);
if(devnums <= device || device < 0)
if (device < 0 || device >= num_devices)
{
std::cout << "device invalid\n";
cerr << "Invalid device ID" << endl;
return -1;
}
if (cmd.get<bool>("verify"))
{
TestSystem::instance().setNumIters(1);
TestSystem::instance().setGPUWarmupIters(0);
TestSystem::instance().setCPUIters(0);
}
if(pid >= oclinfo.size())
devidx = 0;
for (size_t i = 0; i < oclinfo.size(); i++)
{
for (size_t j = 0; j < oclinfo[i].DeviceName.size(); j++, devidx++)
{
if (device == devidx)
{
ocl::setDevice(oclinfo[i], (int)j);
TestSystem::instance().setRecordName(oclinfo[i].DeviceName[j]);
printf("\nuse %d: %s\n", devidx, oclinfo[i].DeviceName[j].c_str());
goto END_DEV;
}
}
}
std::cout << "platform invalid\n";
END_DEV:
return -1;
string filter = cmd.get<string>("filter");
string workdir = cmd.get<string>("workdir");
bool list = cmd.get<bool>("list");
int iters = cmd.get<int>("iters");
int wu_iters = cmd.get<int>("warmup");
double x_top = cmd.get<double>("xtop");
double x_bottom = cmd.get<double>("xbottom");
}
TestSystem::instance().setTopThreshold(x_top);
TestSystem::instance().setBottomThreshold(x_bottom);
if(pid != 0 || device != 0)
if (!filter.empty())
{
TestSystem::instance().setTestFilter(filter);
}
if (!workdir.empty())
{
if (workdir[workdir.size() - 1] != '/' && workdir[workdir.size() - 1] != '\\')
{
workdir += '/';
}
setDevice(oclinfo[pid], device);
TestSystem::instance().setWorkingDir(workdir);
}
if (list)
{
TestSystem::instance().setListMode(true);
}
cout << "Device type:" << type << endl << "Device name:" << oclinfo[pid].DeviceName[device] << endl;
setBinpath(CLBINPATH);
return RUN_ALL_TESTS();
}
TestSystem::instance().setNumIters(iters);
TestSystem::instance().setGPUWarmupIters(wu_iters);
#else // DON'T HAVE_OPENCL
TestSystem::instance().run();
int main()
{
printf("OpenCV was built without OpenCL support\n");
return 0;
}
#endif // HAVE_OPENCL
}

4935
modules/ocl/perf/perf_arithm.cpp
View File

File diff suppressed because it is too large Load Diff

128
modules/ocl/perf/perf_blend.cpp
Unescape View File

@ -44,79 +44,77 @@
//M*/
#include "precomp.hpp"
#include <iomanip>
#ifdef HAVE_OPENCL
using namespace cv;
using namespace cv::ocl;
using namespace cvtest;
using namespace testing;
using namespace std;
PARAM_TEST_CASE(Blend, MatType, int)
///////////// blend ////////////////////////
template <typename T>
void blendLinearGold(const cv::Mat &img1, const cv::Mat &img2, const cv::Mat &weights1, const cv::Mat &weights2, cv::Mat &result_gold)
{
int type;
int channels;
std::vector<cv::ocl::Info> oclinfo;
result_gold.create(img1.size(), img1.type());
int cn = img1.channels();
virtual void SetUp()
for (int y = 0; y < img1.rows; ++y)
{
const float *weights1_row = weights1.ptr<float>(y);
const float *weights2_row = weights2.ptr<float>(y);
const T *img1_row = img1.ptr<T>(y);
const T *img2_row = img2.ptr<T>(y);
T *result_gold_row = result_gold.ptr<T>(y);
type = GET_PARAM(0);
channels = GET_PARAM(1);
//int devnums = getDevice(oclinfo);
//CV_Assert(devnums > 0);
//cv::ocl::setBinpath(CLBINPATH);
for (int x = 0; x < img1.cols * cn; ++x)
{
float w1 = weights1_row[x / cn];
float w2 = weights2_row[x / cn];
result_gold_row[x] = static_cast<T>((img1_row[x] * w1 + img2_row[x] * w2) / (w1 + w2 + 1e-5f));
}
}
};
TEST_P(Blend, Performance)
}
TEST(blend)
{
cv::Size size(MWIDTH, MHEIGHT);
cv::Mat img1_host = randomMat(size, CV_MAKETYPE(type, channels), 0, type == CV_8U ? 255.0 : 1.0);
cv::Mat img2_host = randomMat(size, CV_MAKETYPE(type, channels), 0, type == CV_8U ? 255.0 : 1.0);
cv::Mat weights1 = randomMat(size, CV_32F, 0, 1);
cv::Mat weights2 = randomMat(size, CV_32F, 0, 1);
cv::ocl::oclMat gimg1(size, CV_MAKETYPE(type, channels)), gimg2(size, CV_MAKETYPE(type, channels)), gweights1(size, CV_32F), gweights2(size, CV_32F);
cv::ocl::oclMat gdst(size, CV_MAKETYPE(type, channels));
Mat src1, src2, weights1, weights2, dst;
ocl::oclMat d_src1, d_src2, d_weights1, d_weights2, d_dst;
int all_type[] = {CV_8UC1, CV_8UC4};
std::string type_name[] = {"CV_8UC1", "CV_8UC4"};
double totalgputick_all = 0;
double totalgputick_kernel = 0;
double t1 = 0;
double t2 = 0;
for (int j = 0; j < LOOP_TIMES + 1; j ++) //LOOP_TIMES=100
for (int size = Min_Size; size <= Max_Size; size *= Multiple)
{
t1 = (double)cvGetTickCount();
cv::ocl::oclMat gimg1 = cv::ocl::oclMat(img1_host);
cv::ocl::oclMat gimg2 = cv::ocl::oclMat(img2_host);
cv::ocl::oclMat gweights1 = cv::ocl::oclMat(weights1);
cv::ocl::oclMat gweights2 = cv::ocl::oclMat(weights1);
t2 = (double)cvGetTickCount();
cv::ocl::blendLinear(gimg1, gimg2, gweights1, gweights2, gdst);
t2 = (double)cvGetTickCount() - t2;
cv::Mat m;
gdst.download(m);
t1 = (double)cvGetTickCount() - t1;
if (j == 0)
for (size_t j = 0; j < sizeof(all_type) / sizeof(int); j++)
{
continue;
SUBTEST << size << 'x' << size << "; " << type_name[j] << " and CV_32FC1";
gen(src1, size, size, all_type[j], 0, 256);
gen(src2, size, size, all_type[j], 0, 256);
gen(weights1, size, size, CV_32FC1, 0, 1);
gen(weights2, size, size, CV_32FC1, 0, 1);
blendLinearGold<uchar>(src1, src2, weights1, weights2, dst);
CPU_ON;
blendLinearGold<uchar>(src1, src2, weights1, weights2, dst);
CPU_OFF;
d_src1.upload(src1);
d_src2.upload(src2);
d_weights1.upload(weights1);
d_weights2.upload(weights2);
WARMUP_ON;
ocl::blendLinear(d_src1, d_src2, d_weights1, d_weights2, d_dst);
WARMUP_OFF;
GPU_ON;
ocl::blendLinear(d_src1, d_src2, d_weights1, d_weights2, d_dst);
;
GPU_OFF;
GPU_FULL_ON;
d_src1.upload(src1);
d_src2.upload(src2);
d_weights1.upload(weights1);
d_weights2.upload(weights2);
ocl::blendLinear(d_src1, d_src2, d_weights1, d_weights2, d_dst);
d_dst.download(dst);
GPU_FULL_OFF;
}
totalgputick_all = t1 + totalgputick_all;
totalgputick_kernel = t2 + totalgputick_kernel;
};
cout << "average gpu total runtime is " << totalgputick_all / ((double)cvGetTickFrequency()* LOOP_TIMES * 1000.) << "ms" << endl;
cout << "average gpu runtime without data transfering is " << totalgputick_kernel / ((double)cvGetTickFrequency()* LOOP_TIMES * 1000.) << "ms" << endl;
}
INSTANTIATE_TEST_CASE_P(GPU_ImgProc, Blend, Combine(
Values(CV_8U, CV_32F), Values(1, 4)));
#endif
}
}

150
modules/ocl/perf/perf_brute_force_matcher.cpp
Unescape View File

@ -0,0 +1,150 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2010-2012, Multicoreware, Inc., all rights reserved.
// Copyright (C) 2010-2012, Advanced Micro Devices, Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// @Authors
// Fangfang Bai, fangfang@multicorewareinc.com
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other oclMaterials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors as is and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "precomp.hpp"
//////////////////// BruteForceMatch /////////////////
TEST(BruteForceMatcher)
{
Mat trainIdx_cpu;
Mat distance_cpu;
Mat allDist_cpu;
Mat nMatches_cpu;
for (int size = Min_Size; size <= Max_Size; size *= Multiple)
{
// Init CPU matcher
int desc_len = 64;
BFMatcher matcher(NORM_L2);
Mat query;
gen(query, size, desc_len, CV_32F, 0, 1);
Mat train;
gen(train, size, desc_len, CV_32F, 0, 1);
// Output
vector< vector<DMatch> > matches(2);
// Init GPU matcher
ocl::BruteForceMatcher_OCL_base d_matcher(ocl::BruteForceMatcher_OCL_base::L2Dist);
ocl::oclMat d_query(query);
ocl::oclMat d_train(train);
ocl::oclMat d_trainIdx, d_distance, d_allDist, d_nMatches;
SUBTEST << size << "; match";
matcher.match(query, train, matches[0]);
CPU_ON;
matcher.match(query, train, matches[0]);
CPU_OFF;
WARMUP_ON;
d_matcher.matchSingle(d_query, d_train, d_trainIdx, d_distance);
WARMUP_OFF;
GPU_ON;
d_matcher.matchSingle(d_query, d_train, d_trainIdx, d_distance);
;
GPU_OFF;
GPU_FULL_ON;
d_query.upload(query);
d_train.upload(train);
d_matcher.match(d_query, d_train, matches[0]);
GPU_FULL_OFF;
SUBTEST << size << "; knnMatch";
matcher.knnMatch(query, train, matches, 2);
CPU_ON;
matcher.knnMatch(query, train, matches, 2);
CPU_OFF;
WARMUP_ON;
d_matcher.knnMatchSingle(d_query, d_train, d_trainIdx, d_distance, d_allDist, 2);
WARMUP_OFF;
GPU_ON;
d_matcher.knnMatchSingle(d_query, d_train, d_trainIdx, d_distance, d_allDist, 2);
;
GPU_OFF;
GPU_FULL_ON;
d_query.upload(query);
d_train.upload(train);
d_matcher.knnMatch(d_query, d_train, matches, 2);
GPU_FULL_OFF;
SUBTEST << size << "; radiusMatch";
float max_distance = 2.0f;
matcher.radiusMatch(query, train, matches, max_distance);
CPU_ON;
matcher.radiusMatch(query, train, matches, max_distance);
CPU_OFF;
d_trainIdx.release();
WARMUP_ON;
d_matcher.radiusMatchSingle(d_query, d_train, d_trainIdx, d_distance, d_nMatches, max_distance);
WARMUP_OFF;
GPU_ON;
d_matcher.radiusMatchSingle(d_query, d_train, d_trainIdx, d_distance, d_nMatches, max_distance);
;
GPU_OFF;
GPU_FULL_ON;
d_query.upload(query);
d_train.upload(train);
d_matcher.radiusMatch(d_query, d_train, matches, max_distance);
GPU_FULL_OFF;
}
}

122
modules/ocl/perf/perf_canny.cpp
Unescape View File

@ -42,112 +42,42 @@
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "precomp.hpp"
#include <iomanip>
#ifdef HAVE_OPENCL
using namespace cv;
using namespace cv::ocl;
using namespace cvtest;
using namespace testing;
using namespace std;
#ifndef MWC_TEST_UTILITY
#define MWC_TEST_UTILITY
// Param class
#ifndef IMPLEMENT_PARAM_CLASS
#define IMPLEMENT_PARAM_CLASS(name, type) \
class name \
{ \
public: \
name ( type arg = type ()) : val_(arg) {} \
operator type () const {return val_;} \
private: \
type val_; \
}; \
inline void PrintTo( name param, std::ostream* os) \
{ \
*os << #name << "(" << testing::PrintToString(static_cast< type >(param)) << ")"; \
}
IMPLEMENT_PARAM_CLASS(Channels, int)
#endif // IMPLEMENT_PARAM_CLASS
#endif // MWC_TEST_UTILITY
////////////////////////////////////////////////////////
// Canny1
extern std::string workdir;
IMPLEMENT_PARAM_CLASS(AppertureSize, int);
IMPLEMENT_PARAM_CLASS(L2gradient, bool);
PARAM_TEST_CASE(Canny1, AppertureSize, L2gradient)
{
int apperture_size;
bool useL2gradient;
//std::vector<cv::ocl::Info> oclinfo;
virtual void SetUp()
{
apperture_size = GET_PARAM(0);
useL2gradient = GET_PARAM(1);
//int devnums = getDevice(oclinfo);
//CV_Assert(devnums > 0);
}
};
TEST_P(Canny1, Performance)
///////////// Canny ////////////////////////
TEST(Canny)
{
cv::Mat img = readImage(workdir + "fruits.jpg", cv::IMREAD_GRAYSCALE);
ASSERT_FALSE(img.empty());
Mat img = imread(abspath("aloeL.jpg"), CV_LOAD_IMAGE_GRAYSCALE);
double low_thresh = 100.0;
double high_thresh = 150.0;
cv::Mat edges_gold;
cv::ocl::oclMat edges;
double totalgputick = 0;
double totalgputick_kernel = 0;
double t1 = 0;
double t2 = 0;
for(int j = 0; j < LOOP_TIMES + 1; j ++)
if (img.empty())
{
t1 = (double)cvGetTickCount();//gpu start1
cv::ocl::oclMat ocl_img = cv::ocl::oclMat(img);//upload
t2 = (double)cvGetTickCount(); //kernel
cv::ocl::Canny(ocl_img, edges, low_thresh, high_thresh, apperture_size, useL2gradient);
t2 = (double)cvGetTickCount() - t2;//kernel
cv::Mat cpu_dst;
edges.download (cpu_dst);//download
t1 = (double)cvGetTickCount() - t1;//gpu end1
if(j == 0)
continue;
totalgputick = t1 + totalgputick;
totalgputick_kernel = t2 + totalgputick_kernel;
throw runtime_error("can't open aloeL.jpg");
}
cout << "average gpu runtime is " << totalgputick / ((double)cvGetTickFrequency()* LOOP_TIMES * 1000.) << "ms" << endl;
cout << "average gpu runtime without data transfer is " << totalgputick_kernel / ((double)cvGetTickFrequency()* LOOP_TIMES * 1000.) << "ms" << endl;
SUBTEST << img.cols << 'x' << img.rows << "; aloeL.jpg" << "; edges" << "; CV_8UC1";
Mat edges(img.size(), CV_8UC1);
}
CPU_ON;
Canny(img, edges, 50.0, 100.0);
CPU_OFF;
INSTANTIATE_TEST_CASE_P(GPU_ImgProc, Canny1, testing::Combine(
testing::Values(AppertureSize(3), AppertureSize(5)),
testing::Values(L2gradient(false), L2gradient(true))));
ocl::oclMat d_img(img);
ocl::oclMat d_edges;
ocl::CannyBuf d_buf;
WARMUP_ON;
ocl::Canny(d_img, d_buf, d_edges, 50.0, 100.0);
WARMUP_OFF;
GPU_ON;
ocl::Canny(d_img, d_buf, d_edges, 50.0, 100.0);
;
GPU_OFF;
#endif //Have opencl
GPU_FULL_ON;
d_img.upload(img);
ocl::Canny(d_img, d_buf, d_edges, 50.0, 100.0);
d_edges.download(edges);
GPU_FULL_OFF;
}

91
modules/ocl/perf/perf_color.cpp
Unescape View File

@ -0,0 +1,91 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2010-2012, Multicoreware, Inc., all rights reserved.
// Copyright (C) 2010-2012, Advanced Micro Devices, Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// @Authors
// Fangfang Bai, fangfang@multicorewareinc.com
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other oclMaterials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors as is and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "precomp.hpp"
///////////// cvtColor////////////////////////
TEST(cvtColor)
{
Mat src, dst;
ocl::oclMat d_src, d_dst;
int all_type[] = {CV_8UC4};
std::string type_name[] = {"CV_8UC4"};
for (int size = Min_Size; size <= Max_Size; size *= Multiple)
{
for (size_t j = 0; j < sizeof(all_type) / sizeof(int); j++)
{
gen(src, size, size, all_type[j], 0, 256);
SUBTEST << size << "x" << size << "; " << type_name[j] << " ; CV_RGBA2GRAY";
cvtColor(src, dst, CV_RGBA2GRAY, 4);
CPU_ON;
cvtColor(src, dst, CV_RGBA2GRAY, 4);
CPU_OFF;
d_src.upload(src);
WARMUP_ON;
ocl::cvtColor(d_src, d_dst, CV_RGBA2GRAY, 4);
WARMUP_OFF;
GPU_ON;
ocl::cvtColor(d_src, d_dst, CV_RGBA2GRAY, 4);
;
GPU_OFF;
GPU_FULL_ON;
d_src.upload(src);
ocl::cvtColor(d_src, d_dst, CV_RGBA2GRAY, 4);
d_dst.download(dst);
GPU_FULL_OFF;
}
}
}

112
modules/ocl/perf/perf_columnsum.cpp
Unescape View File

@ -15,8 +15,7 @@
// Third party copyrights are property of their respective owners.
//
// @Authors
// Fangfang Bai fangfang@multicorewareinc.com
//
// Fangfang Bai, fangfang@multicorewareinc.com
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
@ -31,7 +30,7 @@
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// This software is provided by the copyright holders and contributors as is and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
@ -43,78 +42,47 @@
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "precomp.hpp"
#include <iomanip>
using namespace cv;
using namespace cv::ocl;
using namespace cvtest;
using namespace testing;
using namespace std;
///////////////////////////////////////////////////////////////////////////////
/// ColumnSum
#ifdef HAVE_OPENCL
////////////////////////////////////////////////////////////////////////
// ColumnSum
PARAM_TEST_CASE(ColumnSum)
///////////// columnSum////////////////////////
TEST(columnSum)
{
cv::Mat src;
//std::vector<cv::ocl::Info> oclinfo;
Mat src, dst;
ocl::oclMat d_src, d_dst;
virtual void SetUp()
for (int size = Min_Size; size <= Max_Size; size *= Multiple)
{
//int devnums = getDevice(oclinfo);
//CV_Assert(devnums > 0);
SUBTEST << size << 'x' << size << "; CV_32FC1";
gen(src, size, size, CV_32FC1, 0, 256);
CPU_ON;
dst.create(src.size(), src.type());
for (int i = 1; i < src.rows; ++i)
{
for (int j = 0; j < src.cols; ++j)
{
dst.at<float>(i, j) = src.at<float>(i, j) += src.at<float>(i - 1, j);
}
}
CPU_OFF;
d_src.upload(src);
WARMUP_ON;
ocl::columnSum(d_src, d_dst);
WARMUP_OFF;
GPU_ON;
ocl::columnSum(d_src, d_dst);
;
GPU_OFF;
GPU_FULL_ON;
d_src.upload(src);
ocl::columnSum(d_src, d_dst);
d_dst.download(dst);
GPU_FULL_OFF;
}
};
TEST_F(ColumnSum, Performance)
{
cv::Size size(MWIDTH, MHEIGHT);
cv::Mat src = randomMat(size, CV_32FC1);
cv::ocl::oclMat d_dst;
double totalgputick = 0;
double totalgputick_kernel = 0;
double t1 = 0;
double t2 = 0;
for(int j = 0; j < LOOP_TIMES + 1; j ++)
{
t1 = (double)cvGetTickCount();//gpu start1
cv::ocl::oclMat d_src(src);
t2 = (double)cvGetTickCount(); //kernel
cv::ocl::columnSum(d_src, d_dst);
t2 = (double)cvGetTickCount() - t2;//kernel
cv::Mat cpu_dst;
d_dst.download (cpu_dst);//download
t1 = (double)cvGetTickCount() - t1;//gpu end1
if(j == 0)
continue;
totalgputick = t1 + totalgputick;
totalgputick_kernel = t2 + totalgputick_kernel;
}
cout << "average gpu runtime is " << totalgputick / ((double)cvGetTickFrequency()* LOOP_TIMES * 1000.) << "ms" << endl;
cout << "average gpu runtime without data transfer is " << totalgputick_kernel / ((double)cvGetTickFrequency()* LOOP_TIMES * 1000.) << "ms" << endl;
}
#endif
}

99
modules/ocl/perf/perf_fft.cpp
Unescape View File

@ -15,7 +15,7 @@
// Third party copyrights are property of their respective owners.
//
// @Authors
// Fangfangbai, fangfang@multicorewareinc.com
// Fangfang Bai, fangfang@multicorewareinc.com
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
@ -42,85 +42,48 @@
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "precomp.hpp"
using namespace std;
#ifdef HAVE_CLAMDFFT
////////////////////////////////////////////////////////////////////////////
// Dft
PARAM_TEST_CASE(Dft, cv::Size, bool)
{
cv::Size dft_size;
bool dft_rows;
vector<cv::ocl::Info> info;
virtual void SetUp()
{
dft_size = GET_PARAM(0);
dft_rows = GET_PARAM(1);
cv::ocl::getDevice(info);
}
};
TEST_P(Dft, C2C)
///////////// dft ////////////////////////
TEST(dft)
{
cv::Mat a = randomMat(dft_size, CV_32FC2, 0.0, 10.0);
int flags = 0;
flags |= dft_rows ? cv::DFT_ROWS : 0;
Mat src, dst;
ocl::oclMat d_src, d_dst;
cv::ocl::oclMat d_b;
int all_type[] = {CV_32FC1, CV_32FC2};
std::string type_name[] = {"CV_32FC1", "CV_32FC2"};
double totalgputick = 0;
double totalgputick_kernel = 0;
double t1 = 0;
double t2 = 0;
for(int j = 0; j < LOOP_TIMES + 1; j ++)
for (int size = Min_Size; size <= Max_Size; size *= Multiple)
{
for (size_t j = 0; j < sizeof(all_type) / sizeof(int); j++)
{
SUBTEST << size << 'x' << size << "; " << type_name[j] << " ; complex-to-complex";
t1 = (double)cvGetTickCount();//gpu start1
gen(src, size, size, all_type[j], Scalar::all(0), Scalar::all(1));
cv::ocl::oclMat ga = cv::ocl::oclMat(a); //upload
dft(src, dst);
t2 = (double)cvGetTickCount(); //kernel
cv::ocl::dft(ga, d_b, a.size(), flags);
t2 = (double)cvGetTickCount() - t2;//kernel
CPU_ON;
dft(src, dst);
CPU_OFF;
cv::Mat cpu_dst;
d_b.download (cpu_dst);//download
d_src.upload(src);
t1 = (double)cvGetTickCount() - t1;//gpu end1
WARMUP_ON;
ocl::dft(d_src, d_dst, Size(size, size));
WARMUP_OFF;
if(j == 0)
continue;
GPU_ON;
ocl::dft(d_src, d_dst, Size(size, size));
;
GPU_OFF;
totalgputick = t1 + totalgputick;
totalgputick_kernel = t2 + totalgputick_kernel;
GPU_FULL_ON;
d_src.upload(src);
ocl::dft(d_src, d_dst, Size(size, size));
d_dst.download(dst);
GPU_FULL_OFF;
}
}
cout << "average gpu runtime is " << totalgputick / ((double)cvGetTickFrequency()* LOOP_TIMES * 1000.) << "ms" << endl;
cout << "average gpu runtime without data transfer is " << totalgputick_kernel / ((double)cvGetTickFrequency()* LOOP_TIMES * 1000.) << "ms" << endl;
}
TEST_P(Dft, R2CthenC2R)
{
cv::Mat a = randomMat(dft_size, CV_32FC1, 0.0, 10.0);
int flags = 0;
//flags |= dft_rows ? cv::DFT_ROWS : 0; // not supported yet
cv::ocl::oclMat d_b, d_c;
cv::ocl::dft(cv::ocl::oclMat(a), d_b, a.size(), flags);
cv::ocl::dft(d_b, d_c, a.size(), flags + cv::DFT_INVERSE + cv::DFT_REAL_OUTPUT);
EXPECT_MAT_NEAR(a, d_c, a.size().area() * 1e-4, "");
}
//INSTANTIATE_TEST_CASE_P(ocl_DFT, Dft, testing::Combine(
// testing::Values(cv::Size(1280, 1024), cv::Size(1920, 1080),cv::Size(1800, 1500)),
// testing::Values(false, true)));
#endif // HAVE_CLAMDFFT
}

1281
modules/ocl/perf/perf_filters.cpp
View File

File diff suppressed because it is too large Load Diff

103
modules/ocl/perf/perf_gemm.cpp
Unescape View File

@ -16,6 +16,7 @@
//
// @Authors
// Fangfang Bai, fangfang@multicorewareinc.com
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
@ -41,73 +42,47 @@
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "precomp.hpp"
using namespace std;
#ifdef HAVE_CLAMDBLAS
////////////////////////////////////////////////////////////////////////////
// GEMM
PARAM_TEST_CASE(Gemm, int, cv::Size, int)
{
int type;
cv::Size mat_size;
int flags;
vector<cv::ocl::Info> info;
virtual void SetUp()
{
type = GET_PARAM(0);
mat_size = GET_PARAM(1);
flags = GET_PARAM(2);
cv::ocl::getDevice(info);
}
};
TEST_P(Gemm, Performance)
///////////// gemm ////////////////////////
TEST(gemm)
{
cv::Mat a = randomMat(mat_size, type, 0.0, 10.0);
cv::Mat b = randomMat(mat_size, type, 0.0, 10.0);
cv::Mat c = randomMat(mat_size, type, 0.0, 10.0);
cv::ocl::oclMat ocl_dst;
Mat src1, src2, src3, dst;
ocl::oclMat d_src1, d_src2, d_src3, d_dst;
double totalgputick = 0;
double totalgputick_kernel = 0;
double t1 = 0;
double t2 = 0;
for(int j = 0; j < LOOP_TIMES + 1; j ++)
for (int size = Min_Size; size <= Max_Size; size *= Multiple)
{
t1 = (double)cvGetTickCount();//gpu start1
cv::ocl::oclMat ga = cv::ocl::oclMat(a);//upload
cv::ocl::oclMat gb = cv::ocl::oclMat(b);//upload
cv::ocl::oclMat gc = cv::ocl::oclMat(c);//upload
t2 = (double)cvGetTickCount(); //kernel
cv::ocl::gemm(ga, gb, 1.0, gc, 1.0, ocl_dst, flags);
t2 = (double)cvGetTickCount() - t2;//kernel
cv::Mat cpu_dst;
ocl_dst.download (cpu_dst);//download
t1 = (double)cvGetTickCount() - t1;//gpu end
if(j == 0)
continue;
totalgputick = t1 + totalgputick;
totalgputick_kernel = t2 + totalgputick_kernel;
SUBTEST << size << 'x' << size;
gen(src1, size, size, CV_32FC1, Scalar::all(-10), Scalar::all(10));
gen(src2, size, size, CV_32FC1, Scalar::all(-10), Scalar::all(10));
gen(src3, size, size, CV_32FC1, Scalar::all(-10), Scalar::all(10));
gemm(src1, src2, 1.0, src3, 1.0, dst);
CPU_ON;
gemm(src1, src2, 1.0, src3, 1.0, dst);
CPU_OFF;
d_src1.upload(src1);
d_src2.upload(src2);
d_src3.upload(src3);
WARMUP_ON;
ocl::gemm(d_src1, d_src2, 1.0, d_src3, 1.0, d_dst);
WARMUP_OFF;
GPU_ON;
ocl::gemm(d_src1, d_src2, 1.0, d_src3, 1.0, d_dst);
;
GPU_OFF;
GPU_FULL_ON;
d_src1.upload(src1);
d_src2.upload(src2);
d_src3.upload(src3);
ocl::gemm(d_src1, d_src2, 1.0, d_src3, 1.0, d_dst);
d_dst.download(dst);
GPU_FULL_OFF;
}
cout << "average gpu runtime is " << totalgputick / ((double)cvGetTickFrequency()* LOOP_TIMES * 1000.) << "ms" << endl;
cout << "average gpu runtime without data transfer is " << totalgputick_kernel / ((double)cvGetTickFrequency()* LOOP_TIMES * 1000.) << "ms" << endl;
}
INSTANTIATE_TEST_CASE_P(ocl_gemm, Gemm, testing::Combine(
testing::Values(CV_32FC1, CV_32FC2/* , CV_64FC1, CV_64FC2*/),
testing::Values(cv::Size(512, 512), cv::Size(1024, 1024)),
testing::Values(0, (int)cv::GEMM_1_T, (int)cv::GEMM_2_T, (int)(cv::GEMM_1_T + cv::GEMM_2_T))));
#endif
}

178
modules/ocl/perf/perf_haar.cpp
Unescape View File

@ -10,12 +10,12 @@
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2010-2012, Institute Of Software Chinese Academy Of Science, all rights reserved.
// Copyright (C) 2010-2012, Multicoreware, Inc., all rights reserved.
// Copyright (C) 2010-2012, Advanced Micro Devices, Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// @Authors
// Jia Haipeng, jiahaipeng95@gmail.com
// Fangfang Bai, fangfang@multicorewareinc.com
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
@ -30,7 +30,7 @@
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// This software is provided by the copyright holders and contributors as is and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
@ -42,133 +42,97 @@
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "opencv2/objdetect/objdetect.hpp"
#include "precomp.hpp"
#ifdef HAVE_OPENCL
///////////// Haar ////////////////////////
namespace cv
{
namespace ocl
{
using namespace cvtest;
using namespace testing;
using namespace std;
using namespace cv;
extern std::string workdir;
struct getRect
{
Rect operator ()(const CvAvgComp &e) const
Rect operator()(const CvAvgComp &e) const
{
return e.rect;
}
};
PARAM_TEST_CASE(HaarTestBase, int, int)
class CascadeClassifier_GPU : public OclCascadeClassifier
{
//std::vector<cv::ocl::Info> oclinfo;
cv::ocl::OclCascadeClassifier cascade, nestedCascade;
cv::CascadeClassifier cpucascade, cpunestedCascade;
// Mat img;
double scale;
int index;
virtual void SetUp()
public:
void detectMultiScale(oclMat &image,
CV_OUT std::vector<cv::Rect>& faces,
double scaleFactor = 1.1,
int minNeighbors = 3, int flags = 0,
Size minSize = Size(),
Size maxSize = Size())
{
scale = 1.0;
index = 0;
string cascadeName = "../../../data/haarcascades/haarcascade_frontalface_alt.xml";
if( (!cascade.load( cascadeName )) || (!cpucascade.load(cascadeName)))
{
cout << "ERROR: Could not load classifier cascade" << endl;
return;
}
//int devnums = getDevice(oclinfo);
//CV_Assert(devnums>0);
////if you want to use undefault device, set it here
////setDevice(oclinfo[0]);
//cv::ocl::setBinpath("E:\\");
(void)maxSize;
MemStorage storage(cvCreateMemStorage(0));
//CvMat img=image;
CvSeq *objs = oclHaarDetectObjects(image, storage, scaleFactor, minNeighbors, flags, minSize);
vector<CvAvgComp> vecAvgComp;
Seq<CvAvgComp>(objs).copyTo(vecAvgComp);
faces.resize(vecAvgComp.size());
std::transform(vecAvgComp.begin(), vecAvgComp.end(), faces.begin(), getRect());
}
};
////////////////////////////////faceDetect/////////////////////////////////////////////////
struct Haar : HaarTestBase {};
};
TEST_F(Haar, FaceDetect)
}
}
TEST(Haar)
{
string imgName = workdir + "lena.jpg";
Mat img = imread( imgName, 1 );
Mat img = imread(abspath("basketball1.png"), CV_LOAD_IMAGE_GRAYSCALE);
if(img.empty())
if (img.empty())
{
std::cout << imgName << std::endl;
return ;
throw runtime_error("can't open basketball1.png");
}
//int i = 0;
double t = 0;
vector<Rect> faces, oclfaces;
// const static Scalar colors[] = { CV_RGB(0, 0, 255),
// CV_RGB(0, 128, 255),
// CV_RGB(0, 255, 255),
// CV_RGB(0, 255, 0),
// CV_RGB(255, 128, 0),
// CV_RGB(255, 255, 0),
// CV_RGB(255, 0, 0),
// CV_RGB(255, 0, 255)
// } ;
Mat gray, smallImg(cvRound (img.rows / scale), cvRound(img.cols / scale), CV_8UC1 );
MemStorage storage(cvCreateMemStorage(0));
cvtColor( img, gray, CV_BGR2GRAY );
resize( gray, smallImg, smallImg.size(), 0, 0, INTER_LINEAR );
equalizeHist( smallImg, smallImg );
t = (double)cvGetTickCount();
for(int k = 0; k < LOOP_TIMES; k++)
CascadeClassifier faceCascadeCPU;
if (!faceCascadeCPU.load(abspath("haarcascade_frontalface_alt.xml")))
{
cpucascade.detectMultiScale( smallImg, faces, 1.1,
3, 0
| CV_HAAR_SCALE_IMAGE
, Size(30, 30), Size(0, 0) );
throw runtime_error("can't load haarcascade_frontalface_alt.xml");
}
t = (double)cvGetTickCount() - t ;
printf( "cpudetection time = %g ms\n", t / (LOOP_TIMES * (double)cvGetTickFrequency() * 1000.) );
cv::ocl::oclMat image;
CvSeq *_objects=NULL;
t = (double)cvGetTickCount();
for(int k = 0; k < LOOP_TIMES; k++)
vector<Rect> faces;
SUBTEST << img.cols << "x" << img.rows << "; scale image";
CPU_ON;
faceCascadeCPU.detectMultiScale(img, faces,
1.1, 2, 0 | CV_HAAR_SCALE_IMAGE, Size(30, 30));
CPU_OFF;
ocl::CascadeClassifier_GPU faceCascade;
if (!faceCascade.load(abspath("haarcascade_frontalface_alt.xml")))
{
image.upload(smallImg);
_objects = cascade.oclHaarDetectObjects( image, storage, 1.1,
3, 0
| CV_HAAR_SCALE_IMAGE
, Size(30, 30), Size(0, 0) );
throw runtime_error("can't load haarcascade_frontalface_alt.xml");
}
t = (double)cvGetTickCount() - t ;
printf( "ocldetection time = %g ms\n", t / (LOOP_TIMES * (double)cvGetTickFrequency() * 1000.) );
vector<CvAvgComp> vecAvgComp;
Seq<CvAvgComp>(_objects).copyTo(vecAvgComp);
oclfaces.resize(vecAvgComp.size());
std::transform(vecAvgComp.begin(), vecAvgComp.end(), oclfaces.begin(), getRect());
//for( vector<Rect>::const_iterator r = faces.begin(); r != faces.end(); r++, i++ )
//{
// Mat smallImgROI;
// Point center;
// Scalar color = colors[i%8];
// int radius;
// center.x = cvRound((r->x + r->width*0.5)*scale);
// center.y = cvRound((r->y + r->height*0.5)*scale);
// radius = cvRound((r->width + r->height)*0.25*scale);
// circle( img, center, radius, color, 3, 8, 0 );
//}
//namedWindow("result");
//imshow("result",img);
//waitKey(0);
//destroyAllWindows();
}
#endif // HAVE_OPENCL
ocl::oclMat d_img(img);
faces.clear();
WARMUP_ON;
faceCascade.detectMultiScale(d_img, faces,
1.1, 2, 0 | CV_HAAR_SCALE_IMAGE, Size(30, 30));
WARMUP_OFF;
faces.clear();
GPU_ON;
faceCascade.detectMultiScale(d_img, faces,
1.1, 2, 0 | CV_HAAR_SCALE_IMAGE, Size(30, 30));
;
GPU_OFF;
GPU_FULL_ON;
d_img.upload(img);
faceCascade.detectMultiScale(d_img, faces,
1.1, 2, 0 | CV_HAAR_SCALE_IMAGE, Size(30, 30));
GPU_FULL_OFF;
}

140
modules/ocl/perf/perf_hog.cpp
Unescape View File

@ -15,7 +15,7 @@
// Third party copyrights are property of their respective owners.
//
// @Authors
// Peng Xiao, pengxiao@multicorewareinc.com
// Fangfang Bai, fangfang@multicorewareinc.com
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
@ -42,125 +42,47 @@
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "precomp.hpp"
#include <iomanip>
#ifdef HAVE_OPENCL
using namespace cv;
using namespace cv::ocl;
using namespace cvtest;
using namespace testing;
using namespace std;
extern std::string workdir;
#ifndef MWC_TEST_UTILITY
#define MWC_TEST_UTILITY
// Param class
#ifndef IMPLEMENT_PARAM_CLASS
#define IMPLEMENT_PARAM_CLASS(name, type) \
class name \
{ \
public: \
name ( type arg = type ()) : val_(arg) {} \
operator type () const {return val_;} \
private: \
type val_; \
}; \
inline void PrintTo( name param, std::ostream* os) \
{ \
*os << #name << "(" << testing::PrintToString(static_cast< type >(param)) << ")"; \
}
#endif // IMPLEMENT_PARAM_CLASS
#endif // MWC_TEST_UTILITY
IMPLEMENT_PARAM_CLASS(WinSizw48, bool);
PARAM_TEST_CASE(HOG, WinSizw48, bool)
{
bool is48;
vector<float> detector;
virtual void SetUp()
{
is48 = GET_PARAM(0);
if(is48)
{
detector = cv::ocl::HOGDescriptor::getPeopleDetector48x96();
}
else
{
detector = cv::ocl::HOGDescriptor::getPeopleDetector64x128();
}
}
};
TEST_P(HOG, Performance)
///////////// HOG////////////////////////
TEST(HOG)
{
cv::Mat img = readImage(workdir + "lena.jpg", cv::IMREAD_GRAYSCALE);
ASSERT_FALSE(img.empty());
// define HOG related arguments
float scale = 1.05f;
//int nlevels = 13;
int gr_threshold = 8;
float hit_threshold = 1.4f;
//bool hit_threshold_auto = true;
Mat src = imread(abspath("road.png"), cv::IMREAD_GRAYSCALE);
int win_width = is48 ? 48 : 64;
int win_stride_width = 8;
int win_stride_height = 8;
bool gamma_corr = true;
Size win_size(win_width, win_width * 2); //(64, 128) or (48, 96)
Size win_stride(win_stride_width, win_stride_height);
cv::ocl::HOGDescriptor gpu_hog(win_size, Size(16, 16), Size(8, 8), Size(8, 8), 9,
cv::ocl::HOGDescriptor::DEFAULT_WIN_SIGMA, 0.2, gamma_corr,
cv::ocl::HOGDescriptor::DEFAULT_NLEVELS);
gpu_hog.setSVMDetector(detector);
double totalgputick = 0;
double totalgputick_kernel = 0;
double t1 = 0;
double t2 = 0;
for(int j = 0; j < LOOP_TIMES + 1; j ++)
if (src.empty())
{
t1 = (double)cvGetTickCount();//gpu start1
ocl::oclMat d_src(img);//upload
t2 = (double)cvGetTickCount(); //kernel
vector<Rect> found;
gpu_hog.detectMultiScale(d_src, found, hit_threshold, win_stride,
Size(0, 0), scale, gr_threshold);
t2 = (double)cvGetTickCount() - t2;//kernel
// no download time for HOG
throw runtime_error("can't open road.png");
}
t1 = (double)cvGetTickCount() - t1;//gpu end1
if(j == 0)
continue;
cv::HOGDescriptor hog;
hog.setSVMDetector(hog.getDefaultPeopleDetector());
std::vector<cv::Rect> found_locations;
totalgputick = t1 + totalgputick;
SUBTEST << 768 << 'x' << 576 << "; road.png";
totalgputick_kernel = t2 + totalgputick_kernel;
hog.detectMultiScale(src, found_locations);
}
CPU_ON;
hog.detectMultiScale(src, found_locations);
CPU_OFF;
cout << "average gpu runtime is " << totalgputick / ((double)cvGetTickFrequency()* LOOP_TIMES * 1000.) << "ms" << endl;
cout << "average gpu runtime without data transfer is " << totalgputick_kernel / ((double)cvGetTickFrequency()* LOOP_TIMES * 1000.) << "ms" << endl;
}
cv::ocl::HOGDescriptor ocl_hog;
ocl_hog.setSVMDetector(ocl_hog.getDefaultPeopleDetector());
ocl::oclMat d_src;
d_src.upload(src);
WARMUP_ON;
ocl_hog.detectMultiScale(d_src, found_locations);
WARMUP_OFF;
INSTANTIATE_TEST_CASE_P(GPU_ObjDetect, HOG, testing::Combine(testing::Values(WinSizw48(false), WinSizw48(true)), testing::Values(false)));
GPU_ON;
ocl_hog.detectMultiScale(d_src, found_locations);
;
GPU_OFF;
#endif //Have opencl
GPU_FULL_ON;
d_src.upload(src);
ocl_hog.detectMultiScale(d_src, found_locations);
GPU_FULL_OFF;
}

2465
modules/ocl/perf/perf_imgproc.cpp
View File

File diff suppressed because it is too large Load Diff

234
modules/ocl/perf/perf_match_template.cpp
Unescape View File

@ -42,191 +42,105 @@
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "precomp.hpp"
#include <iomanip>
#ifdef HAVE_OPENCL
using namespace cv;
using namespace cv::ocl;
using namespace cvtest;
using namespace testing;
using namespace std;
#ifndef MWC_TEST_UTILITY
#define MWC_TEST_UTILITY
//////// Utility
#ifndef DIFFERENT_SIZES
#else
#undef DIFFERENT_SIZES
#endif
#define DIFFERENT_SIZES testing::Values(cv::Size(256, 256), cv::Size(3000, 3000))
// Param class
#ifndef IMPLEMENT_PARAM_CLASS
#define IMPLEMENT_PARAM_CLASS(name, type) \
class name \
{ \
public: \
name ( type arg = type ()) : val_(arg) {} \
operator type () const {return val_;} \
private: \
type val_; \
}; \
inline void PrintTo( name param, std::ostream* os) \
{ \
*os << #name << "(" << testing::PrintToString(static_cast< type >(param)) << ")"; \
}
IMPLEMENT_PARAM_CLASS(Channels, int)
#endif // IMPLEMENT_PARAM_CLASS
#endif // MWC_TEST_UTILITY
////////////////////////////////////////////////////////////////////////////////
// MatchTemplate
#define ALL_TEMPLATE_METHODS testing::Values(TemplateMethod(cv::TM_SQDIFF), TemplateMethod(cv::TM_CCORR), TemplateMethod(cv::TM_CCOEFF), TemplateMethod(cv::TM_SQDIFF_NORMED), TemplateMethod(cv::TM_CCORR_NORMED), TemplateMethod(cv::TM_CCOEFF_NORMED))
IMPLEMENT_PARAM_CLASS(TemplateSize, cv::Size);
const char *TEMPLATE_METHOD_NAMES[6] = {"TM_SQDIFF", "TM_SQDIFF_NORMED", "TM_CCORR", "TM_CCORR_NORMED", "TM_CCOEFF", "TM_CCOEFF_NORMED"};
PARAM_TEST_CASE(MatchTemplate, cv::Size, TemplateSize, Channels, TemplateMethod)
{
cv::Size size;
cv::Size templ_size;
int cn;
int method;
//vector<cv::ocl::Info> oclinfo;
virtual void SetUp()
{
size = GET_PARAM(0);
templ_size = GET_PARAM(1);
cn = GET_PARAM(2);
method = GET_PARAM(3);
//int devnums = getDevice(oclinfo);
//CV_Assert(devnums > 0);
}
};
struct MatchTemplate8U : MatchTemplate {};
TEST_P(MatchTemplate8U, Performance)
/////////// matchTemplate ////////////////////////
//void InitMatchTemplate()
//{
// Mat src; gen(src, 500, 500, CV_32F, 0, 1);
// Mat templ; gen(templ, 500, 500, CV_32F, 0, 1);
// ocl::oclMat d_src(src), d_templ(templ), d_dst;
// ocl::matchTemplate(d_src, d_templ, d_dst, CV_TM_CCORR);
//}
TEST(matchTemplate)
{
std::cout << "Method: " << TEMPLATE_METHOD_NAMES[method] << std::endl;
std::cout << "Image Size: (" << size.width << ", " << size.height << ")" << std::endl;
std::cout << "Template Size: (" << templ_size.width << ", " << templ_size.height << ")" << std::endl;
std::cout << "Channels: " << cn << std::endl;
cv::Mat image = randomMat(size, CV_MAKETYPE(CV_8U, cn));
cv::Mat templ = randomMat(templ_size, CV_MAKETYPE(CV_8U, cn));
cv::Mat dst_gold;
cv::ocl::oclMat dst;
//InitMatchTemplate();
Mat src, templ, dst;
int templ_size = 5;
double totalgputick = 0;
double totalgputick_kernel = 0;
double t1 = 0;
double t2 = 0;
for(int j = 0; j < LOOP_TIMES + 1; j ++)
for (int size = Min_Size; size <= Max_Size; size *= Multiple)
{
int all_type[] = {CV_32FC1, CV_32FC4};
std::string type_name[] = {"CV_32FC1", "CV_32FC4"};
t1 = (double)cvGetTickCount();//gpu start1
for (size_t j = 0; j < sizeof(all_type) / sizeof(int); j++)
{
for(templ_size = 5; templ_size <= 5; templ_size *= 5)
{
gen(src, size, size, all_type[j], 0, 1);
cv::ocl::oclMat ocl_image = cv::ocl::oclMat(image);//upload
cv::ocl::oclMat ocl_templ = cv::ocl::oclMat(templ);//upload
SUBTEST << src.cols << 'x' << src.rows << "; " << type_name[j] << "; templ " << templ_size << 'x' << templ_size << "; CCORR";
t2 = (double)cvGetTickCount(); //kernel
cv::ocl::matchTemplate(ocl_image, ocl_templ, dst, method);
t2 = (double)cvGetTickCount() - t2;//kernel
gen(templ, templ_size, templ_size, all_type[j], 0, 1);
cv::Mat cpu_dst;
dst.download (cpu_dst);//download
matchTemplate(src, templ, dst, CV_TM_CCORR);
t1 = (double)cvGetTickCount() - t1;//gpu end1
CPU_ON;
matchTemplate(src, templ, dst, CV_TM_CCORR);
CPU_OFF;
if(j == 0)
continue;
ocl::oclMat d_src(src), d_templ, d_dst;
totalgputick = t1 + totalgputick;
totalgputick_kernel = t2 + totalgputick_kernel;
d_templ.upload(templ);
}
WARMUP_ON;
ocl::matchTemplate(d_src, d_templ, d_dst, CV_TM_CCORR);
WARMUP_OFF;
cout << "average gpu runtime is " << totalgputick / ((double)cvGetTickFrequency()* LOOP_TIMES * 1000.) << "ms" << endl;
cout << "average gpu runtime without data transfer is " << totalgputick_kernel / ((double)cvGetTickFrequency()* LOOP_TIMES * 1000.) << "ms" << endl;
GPU_ON;
ocl::matchTemplate(d_src, d_templ, d_dst, CV_TM_CCORR);
;
GPU_OFF;
GPU_FULL_ON;
d_src.upload(src);
d_templ.upload(templ);
ocl::matchTemplate(d_src, d_templ, d_dst, CV_TM_CCORR);
d_dst.download(dst);
GPU_FULL_OFF;
}
}
}
int all_type_8U[] = {CV_8UC1};
std::string type_name_8U[] = {"CV_8UC1"};
for (size_t j = 0; j < sizeof(all_type_8U) / sizeof(int); j++)
{
for(templ_size = 5; templ_size <= 5; templ_size *= 5)
{
SUBTEST << src.cols << 'x' << src.rows << "; " << type_name_8U[j] << "; templ " << templ_size << 'x' << templ_size << "; CCORR_NORMED";
struct MatchTemplate32F : MatchTemplate {};
TEST_P(MatchTemplate32F, Performance)
{
std::cout << "Method: " << TEMPLATE_METHOD_NAMES[method] << std::endl;
std::cout << "Image Size: (" << size.width << ", " << size.height << ")" << std::endl;
std::cout << "Template Size: (" << templ_size.width << ", " << templ_size.height << ")" << std::endl;
std::cout << "Channels: " << cn << std::endl;
cv::Mat image = randomMat(size, CV_MAKETYPE(CV_32F, cn));
cv::Mat templ = randomMat(templ_size, CV_MAKETYPE(CV_32F, cn));
gen(src, size, size, all_type_8U[j], 0, 255);
cv::Mat dst_gold;
cv::ocl::oclMat dst;
gen(templ, templ_size, templ_size, all_type_8U[j], 0, 255);
matchTemplate(src, templ, dst, CV_TM_CCORR_NORMED);
CPU_ON;
matchTemplate(src, templ, dst, CV_TM_CCORR_NORMED);
CPU_OFF;
ocl::oclMat d_src(src);
ocl::oclMat d_templ(templ), d_dst;
double totalgputick = 0;
double totalgputick_kernel = 0;
double t1 = 0;
double t2 = 0;
for(int j = 0; j < LOOP_TIMES; j ++)
{
t1 = (double)cvGetTickCount();//gpu start1
WARMUP_ON;
ocl::matchTemplate(d_src, d_templ, d_dst, CV_TM_CCORR_NORMED);
WARMUP_OFF;
cv::ocl::oclMat ocl_image = cv::ocl::oclMat(image);//upload
cv::ocl::oclMat ocl_templ = cv::ocl::oclMat(templ);//upload
t2 = (double)cvGetTickCount(); //kernel
cv::ocl::matchTemplate(ocl_image, ocl_templ, dst, method);
t2 = (double)cvGetTickCount() - t2;//kernel
cv::Mat cpu_dst;
dst.download (cpu_dst);//download
t1 = (double)cvGetTickCount() - t1;//gpu end1
totalgputick = t1 + totalgputick;
totalgputick_kernel = t2 + totalgputick_kernel;
GPU_ON;
ocl::matchTemplate(d_src, d_templ, d_dst, CV_TM_CCORR_NORMED);
;
GPU_OFF;
GPU_FULL_ON;
d_src.upload(src);
d_templ.upload(templ);
ocl::matchTemplate(d_src, d_templ, d_dst, CV_TM_CCORR_NORMED);
d_dst.download(dst);
GPU_FULL_OFF;
}
}
}
cout << "average gpu runtime is " << totalgputick / ((double)cvGetTickFrequency()* LOOP_TIMES * 1000.) << "ms" << endl;
cout << "average gpu runtime without data transfer is " << totalgputick_kernel / ((double)cvGetTickFrequency()* LOOP_TIMES * 1000.) << "ms" << endl;
}
INSTANTIATE_TEST_CASE_P(GPU_ImgProc, MatchTemplate8U,
testing::Combine(
testing::Values(cv::Size(1280, 1024), cv::Size(MWIDTH, MHEIGHT), cv::Size(1800, 1500)),
testing::Values(TemplateSize(cv::Size(5, 5)), TemplateSize(cv::Size(16, 16))/*, TemplateSize(cv::Size(30, 30))*/),
testing::Values(Channels(1), Channels(4)/*, Channels(3)*/),
ALL_TEMPLATE_METHODS
)
);
INSTANTIATE_TEST_CASE_P(GPU_ImgProc, MatchTemplate32F, testing::Combine(
testing::Values(cv::Size(1280, 1024), cv::Size(MWIDTH, MHEIGHT), cv::Size(1800, 1500)),
testing::Values(TemplateSize(cv::Size(5, 5)), TemplateSize(cv::Size(16, 16))/*, TemplateSize(cv::Size(30, 30))*/),
testing::Values(Channels(1), Channels(4) /*, Channels(3)*/),
testing::Values(TemplateMethod(cv::TM_SQDIFF), TemplateMethod(cv::TM_CCORR))));
#endif //HAVE_OPENCL
}

743
modules/ocl/perf/perf_matrix_operation.cpp
Unescape View File

@ -10,12 +10,12 @@
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2010-2012, Institute Of Software Chinese Academy Of Science, all rights reserved.
// Copyright (C) 2010-2012, Multicoreware, Inc., all rights reserved.
// Copyright (C) 2010-2012, Advanced Micro Devices, Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// @Authors
// Jia Haipeng, jiahaipeng95@gmail.com
// Fangfang Bai, fangfang@multicorewareinc.com
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
@ -30,7 +30,7 @@
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// This software is provided by the copyright holders and contributors as is and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
@ -42,697 +42,140 @@
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "precomp.hpp"
#ifdef HAVE_OPENCL
using namespace cvtest;
using namespace testing;
using namespace std;
using namespace cv::ocl;
////////////////////////////////converto/////////////////////////////////////////////////
PARAM_TEST_CASE(ConvertToTestBase, MatType, MatType)
///////////// ConvertTo////////////////////////
TEST(ConvertTo)
{
int type;
int dst_type;
//src mat
cv::Mat mat;
cv::Mat dst;
// set up roi
int roicols;
int roirows;
int srcx;
int srcy;
int dstx;
int dsty;
//src mat with roi
cv::Mat mat_roi;
cv::Mat dst_roi;
//std::vector<cv::ocl::Info> oclinfo;
//ocl dst mat for testing
cv::ocl::oclMat gdst_whole;
//ocl mat with roi
cv::ocl::oclMat gmat;
cv::ocl::oclMat gdst;
virtual void SetUp()
{
type = GET_PARAM(0);
dst_type = GET_PARAM(1);
cv::RNG &rng = TS::ptr()->get_rng();
cv::Size size(MWIDTH, MHEIGHT);
mat = randomMat(rng, size, type, 5, 16, false);
dst = randomMat(rng, size, type, 5, 16, false);
//int devnums = getDevice(oclinfo);
//CV_Assert(devnums > 0);
////if you want to use undefault device, set it here
////setDevice(oclinfo[0]);
//setBinpath(CLBINPATH);
}
Mat src, dst;
ocl::oclMat d_src, d_dst;
void Has_roi(int b)
int all_type[] = {CV_8UC1, CV_8UC4};
std::string type_name[] = {"CV_8UC1", "CV_8UC4"};
for (int size = Min_Size; size <= Max_Size; size *= Multiple)
{
//cv::RNG& rng = TS::ptr()->get_rng();
if(b)
{
//randomize ROI
roicols = mat.cols - 1; //start
roirows = mat.rows - 1;
srcx = 1;
srcy = 1;
dstx = 1;
dsty = 1;
}
else
for (size_t j = 0; j < sizeof(all_type) / sizeof(int); j++)
{
roicols = mat.cols;
roirows = mat.rows;
srcx = 0;
srcy = 0;
dstx = 0;
dsty = 0;
};
SUBTEST << size << 'x' << size << "; " << type_name[j] << " to 32FC1";
mat_roi = mat(Rect(srcx, srcy, roicols, roirows));
dst_roi = dst(Rect(dstx, dsty, roicols, roirows));
gen(src, size, size, all_type[j], 0, 256);
//gen(dst, size, size, all_type[j], 0, 256);
//gdst_whole = dst;
//gdst = gdst_whole(Rect(dstx,dsty,roicols,roirows));
//d_dst.upload(dst);
//gmat = mat_roi;
}
};
src.convertTo(dst, CV_32FC1);
CPU_ON;
src.convertTo(dst, CV_32FC1);
CPU_OFF;
struct ConvertTo : ConvertToTestBase {};
d_src.upload(src);
TEST_P(ConvertTo, Accuracy)
{
#ifndef PRINT_KERNEL_RUN_TIME
double totalcputick = 0;
double totalgputick = 0;
double totalgputick_kernel = 0;
double t0 = 0;
double t1 = 0;
double t2 = 0;
for(int k = LOOPROISTART; k < LOOPROIEND; k++)
{
totalcputick = 0;
totalgputick = 0;
totalgputick_kernel = 0;
for(int j = 0; j < LOOP_TIMES + 1; j ++)
{
Has_roi(k);
t0 = (double)cvGetTickCount();//cpu start
mat_roi.convertTo(dst_roi, dst_type);
t0 = (double)cvGetTickCount() - t0;//cpu end
t1 = (double)cvGetTickCount();//gpu start1
gdst_whole = dst;
gdst = gdst_whole(Rect(dstx, dsty, roicols, roirows));
gmat = mat_roi;
t2 = (double)cvGetTickCount(); //kernel
gmat.convertTo(gdst, dst_type);
t2 = (double)cvGetTickCount() - t2;//kernel
cv::Mat cpu_dst;
gdst_whole.download (cpu_dst);//download
t1 = (double)cvGetTickCount() - t1;//gpu end1
if(j == 0)
continue;
totalgputick = t1 + totalgputick;
totalcputick = t0 + totalcputick;
totalgputick_kernel = t2 + totalgputick_kernel;
WARMUP_ON;
d_src.convertTo(d_dst, CV_32FC1);
WARMUP_OFF;
}
if(k == 0)
{
cout << "no roi\n";
}
else
{
cout << "with roi\n";
};
cout << "average cpu runtime is " << totalcputick / ((double)cvGetTickFrequency()* LOOP_TIMES * 1000.) << "ms" << endl;
cout << "average gpu runtime is " << totalgputick / ((double)cvGetTickFrequency()* LOOP_TIMES * 1000.) << "ms" << endl;
cout << "average gpu runtime without data transfer is " << totalgputick_kernel / ((double)cvGetTickFrequency()* LOOP_TIMES * 1000.) << "ms" << endl;
}
#else
for(int j = LOOPROISTART; j < LOOPROIEND; j ++)
{
Has_roi(j);
gdst_whole = dst;
gdst = gdst_whole(Rect(dstx, dsty, roicols, roirows));
GPU_ON;
d_src.convertTo(d_dst, CV_32FC1);
;
GPU_OFF;
gmat = mat_roi;
if(j == 0)
{
cout << "no roi:";
GPU_FULL_ON;
d_src.upload(src);
d_src.convertTo(d_dst, CV_32FC1);
d_dst.download(dst);
GPU_FULL_OFF;
}
else
{
cout << "\nwith roi:";
};
gmat.convertTo(gdst, dst_type);
};
#endif
}
}
///////////////////////////////////////////copyto/////////////////////////////////////////////////////////////
PARAM_TEST_CASE(CopyToTestBase, MatType, bool)
///////////// copyTo////////////////////////
TEST(copyTo)
{
int type;
cv::Mat mat;
cv::Mat mask;
cv::Mat dst;
// set up roi
int roicols;
int roirows;
int srcx;
int srcy;
int dstx;
int dsty;
int maskx;
int masky;
//src mat with roi
cv::Mat mat_roi;
cv::Mat mask_roi;
cv::Mat dst_roi;
//std::vector<cv::ocl::Info> oclinfo;
//ocl dst mat for testing
cv::ocl::oclMat gdst_whole;
//ocl mat with roi
cv::ocl::oclMat gmat;
cv::ocl::oclMat gdst;
cv::ocl::oclMat gmask;
virtual void SetUp()
{
type = GET_PARAM(0);
cv::RNG &rng = TS::ptr()->get_rng();
cv::Size size(MWIDTH, MHEIGHT);
Mat src, dst;
ocl::oclMat d_src, d_dst;
mat = randomMat(rng, size, type, 5, 16, false);
dst = randomMat(rng, size, type, 5, 16, false);
mask = randomMat(rng, size, CV_8UC1, 0, 2, false);
int all_type[] = {CV_8UC1, CV_8UC4};
std::string type_name[] = {"CV_8UC1", "CV_8UC4"};
cv::threshold(mask, mask, 0.5, 255., CV_8UC1);
//int devnums = getDevice(oclinfo);
//CV_Assert(devnums > 0);
////if you want to use undefault device, set it here
////setDevice(oclinfo[0]);
//setBinpath(CLBINPATH);
}
void Has_roi(int b)
for (int size = Min_Size; size <= Max_Size; size *= Multiple)
{
//cv::RNG& rng = TS::ptr()->get_rng();
if(b)
{
//randomize ROI
roicols = mat.cols - 1; //start
roirows = mat.rows - 1;
srcx = 1;
srcy = 1;
dstx = 1;
dsty = 1;
maskx = 1;
masky = 1;
}
else
for (size_t j = 0; j < sizeof(all_type) / sizeof(int); j++)
{
roicols = mat.cols;
roirows = mat.rows;
srcx = 0;
srcy = 0;
dstx = 0;
dsty = 0;
maskx = 0;
masky = 0;
};
mat_roi = mat(Rect(srcx, srcy, roicols, roirows));
mask_roi = mask(Rect(maskx, masky, roicols, roirows));
dst_roi = dst(Rect(dstx, dsty, roicols, roirows));
//gdst_whole = dst;
//gdst = gdst_whole(Rect(dstx,dsty,roicols,roirows));
//gmat = mat_roi;
//gmask = mask_roi;
}
};
SUBTEST << size << 'x' << size << "; " << type_name[j] ;
struct CopyTo : CopyToTestBase {};
gen(src, size, size, all_type[j], 0, 256);
//gen(dst, size, size, all_type[j], 0, 256);
TEST_P(CopyTo, Without_mask)
{
#ifndef PRINT_KERNEL_RUN_TIME
double totalcputick = 0;
double totalgputick = 0;
double totalgputick_kernel = 0;
double t0 = 0;
double t1 = 0;
double t2 = 0;
for(int k = LOOPROISTART; k < LOOPROIEND; k++)
{
totalcputick = 0;
totalgputick = 0;
totalgputick_kernel = 0;
for(int j = 0; j < LOOP_TIMES + 1; j ++)
{
Has_roi(k);
t0 = (double)cvGetTickCount();//cpu start
mat_roi.copyTo(dst_roi);
t0 = (double)cvGetTickCount() - t0;//cpu end
t1 = (double)cvGetTickCount();//gpu start1
gdst_whole = dst;
gdst = gdst_whole(Rect(dstx, dsty, roicols, roirows));
gmat = mat_roi;
t2 = (double)cvGetTickCount(); //kernel
gmat.copyTo(gdst);
t2 = (double)cvGetTickCount() - t2;//kernel
cv::Mat cpu_dst;
gdst_whole.download (cpu_dst);//download
t1 = (double)cvGetTickCount() - t1;//gpu end1
if(j == 0)
continue;
totalgputick = t1 + totalgputick;
totalcputick = t0 + totalcputick;
totalgputick_kernel = t2 + totalgputick_kernel;
//d_dst.upload(dst);
}
if(k == 0)
{
cout << "no roi\n";
}
else
{
cout << "with roi\n";
};
cout << "average cpu runtime is " << totalcputick / ((double)cvGetTickFrequency()* LOOP_TIMES * 1000.) << "ms" << endl;
cout << "average gpu runtime is " << totalgputick / ((double)cvGetTickFrequency()* LOOP_TIMES * 1000.) << "ms" << endl;
cout << "average gpu runtime without data transfer is " << totalgputick_kernel / ((double)cvGetTickFrequency()* LOOP_TIMES * 1000.) << "ms" << endl;
}
#else
for(int j = LOOPROISTART; j < LOOPROIEND; j ++)
{
Has_roi(j);
gdst_whole = dst;
gdst = gdst_whole(Rect(dstx, dsty, roicols, roirows));
src.copyTo(dst);
gmat = mat_roi;
CPU_ON;
src.copyTo(dst);
CPU_OFF;
if(j == 0)
{
cout << "no roi:";
}
else
{
cout << "\nwith roi:";
};
gmat.copyTo(gdst);
};
#endif
}
d_src.upload(src);
TEST_P(CopyTo, With_mask)
{
#ifndef PRINT_KERNEL_RUN_TIME
double totalcputick = 0;
double totalgputick = 0;
double totalgputick_kernel = 0;
double t0 = 0;
double t1 = 0;
double t2 = 0;
for(int k = LOOPROISTART; k < LOOPROIEND; k++)
{
totalcputick = 0;
totalgputick = 0;
totalgputick_kernel = 0;
for(int j = 0; j < LOOP_TIMES + 1; j ++)
{
Has_roi(k);
t0 = (double)cvGetTickCount();//cpu start
mat_roi.copyTo(dst_roi, mask_roi);
t0 = (double)cvGetTickCount() - t0;//cpu end
t1 = (double)cvGetTickCount();//gpu start1
gdst_whole = dst;
gdst = gdst_whole(Rect(dstx, dsty, roicols, roirows));
gmat = mat_roi;
gmask = mask_roi;
t2 = (double)cvGetTickCount(); //kernel
gmat.copyTo(gdst, gmask);
t2 = (double)cvGetTickCount() - t2;//kernel
cv::Mat cpu_dst;
gdst_whole.download (cpu_dst);//download
t1 = (double)cvGetTickCount() - t1;//gpu end1
if(j == 0)
continue;
totalgputick = t1 + totalgputick;
totalcputick = t0 + totalcputick;
totalgputick_kernel = t2 + totalgputick_kernel;
WARMUP_ON;
d_src.copyTo(d_dst);
WARMUP_OFF;
}
if(k == 0)
{
cout << "no roi\n";
}
else
{
cout << "with roi\n";
};
cout << "average cpu runtime is " << totalcputick / ((double)cvGetTickFrequency()* LOOP_TIMES * 1000.) << "ms" << endl;
cout << "average gpu runtime is " << totalgputick / ((double)cvGetTickFrequency()* LOOP_TIMES * 1000.) << "ms" << endl;
cout << "average gpu runtime without data transfer is " << totalgputick_kernel / ((double)cvGetTickFrequency()* LOOP_TIMES * 1000.) << "ms" << endl;
}
#else
for(int j = LOOPROISTART; j < LOOPROIEND; j ++)
{
Has_roi(j);
gdst_whole = dst;
gdst = gdst_whole(Rect(dstx, dsty, roicols, roirows));
gmat = mat_roi;
gmask = mask_roi;
GPU_ON;
d_src.copyTo(d_dst);
;
GPU_OFF;
if(j == 0)
{
cout << "no roi:";
GPU_FULL_ON;
d_src.upload(src);
d_src.copyTo(d_dst);
d_dst.download(dst);
GPU_FULL_OFF;
}
else
{
cout << "\nwith roi:";
};
gmat.copyTo(gdst, gmask);
};
#endif
}
///////////////////////////////////////////copyto/////////////////////////////////////////////////////////////
PARAM_TEST_CASE(SetToTestBase, MatType, bool)
}
}
///////////// setTo////////////////////////
TEST(setTo)
{
int type;
cv::Scalar val;
cv::Mat mat;
cv::Mat mask;
// set up roi
int roicols;
int roirows;
int srcx;
int srcy;
int maskx;
int masky;
//src mat with roi
cv::Mat mat_roi;
cv::Mat mask_roi;
//std::vector<cv::ocl::Info> oclinfo;
//ocl dst mat for testing
cv::ocl::oclMat gmat_whole;
//ocl mat with roi
cv::ocl::oclMat gmat;
cv::ocl::oclMat gmask;
virtual void SetUp()
{
type = GET_PARAM(0);
cv::RNG &rng = TS::ptr()->get_rng();
cv::Size size(MWIDTH, MHEIGHT);
Mat src, dst;
Scalar val(1, 2, 3, 4);
ocl::oclMat d_src, d_dst;
mat = randomMat(rng, size, type, 5, 16, false);
mask = randomMat(rng, size, CV_8UC1, 0, 2, false);
int all_type[] = {CV_8UC1, CV_8UC4};
std::string type_name[] = {"CV_8UC1", "CV_8UC4"};
cv::threshold(mask, mask, 0.5, 255., CV_8UC1);
val = cv::Scalar(rng.uniform(-10.0, 10.0), rng.uniform(-10.0, 10.0), rng.uniform(-10.0, 10.0), rng.uniform(-10.0, 10.0));
//int devnums = getDevice(oclinfo);
//CV_Assert(devnums > 0);
////if you want to use undefault device, set it here
////setDevice(oclinfo[0]);
//setBinpath(CLBINPATH);
}
void Has_roi(int b)
for (int size = Min_Size; size <= Max_Size; size *= Multiple)
{
//cv::RNG& rng = TS::ptr()->get_rng();
if(b)
for (size_t j = 0; j < sizeof(all_type) / sizeof(int); j++)
{
//randomize ROI
roicols = mat.cols - 1; //start
roirows = mat.rows - 1;
srcx = 1;
srcy = 1;
maskx = 1;
masky = 1;
}
else
{
roicols = mat.cols;
roirows = mat.rows;
srcx = 0;
srcy = 0;
maskx = 0;
masky = 0;
};
mat_roi = mat(Rect(srcx, srcy, roicols, roirows));
mask_roi = mask(Rect(maskx, masky, roicols, roirows));
//gmat_whole = mat;
//gmat = gmat_whole(Rect(srcx,srcy,roicols,roirows));
//gmask = mask_roi;
}
};
struct SetTo : SetToTestBase {};
SUBTEST << size << 'x' << size << "; " << type_name[j] ;
TEST_P(SetTo, Without_mask)
{
#ifndef PRINT_KERNEL_RUN_TIME
double totalcputick = 0;
double totalgputick = 0;
double totalgputick_kernel = 0;
double t0 = 0;
double t1 = 0;
double t2 = 0;
for(int k = LOOPROISTART; k < LOOPROIEND; k++)
{
totalcputick = 0;
totalgputick = 0;
totalgputick_kernel = 0;
for(int j = 0; j < LOOP_TIMES + 1; j ++)
{
Has_roi(k);
t0 = (double)cvGetTickCount();//cpu start
mat_roi.setTo(val);
t0 = (double)cvGetTickCount() - t0;//cpu end
t1 = (double)cvGetTickCount();//gpu start1
gmat_whole = mat;
gmat = gmat_whole(Rect(srcx, srcy, roicols, roirows));
t2 = (double)cvGetTickCount(); //kernel
gmat.setTo(val);
t2 = (double)cvGetTickCount() - t2;//kernel
cv::Mat cpu_dst;
gmat_whole.download(cpu_dst);//download
t1 = (double)cvGetTickCount() - t1;//gpu end1
if(j == 0)
continue;
totalgputick = t1 + totalgputick;
totalcputick = t0 + totalcputick;
totalgputick_kernel = t2 + totalgputick_kernel;
gen(src, size, size, all_type[j], 0, 256);
}
if(k == 0)
{
cout << "no roi\n";
}
else
{
cout << "with roi\n";
};
cout << "average cpu runtime is " << totalcputick / ((double)cvGetTickFrequency()* LOOP_TIMES * 1000.) << "ms" << endl;
cout << "average gpu runtime is " << totalgputick / ((double)cvGetTickFrequency()* LOOP_TIMES * 1000.) << "ms" << endl;
cout << "average gpu runtime without data transfer is " << totalgputick_kernel / ((double)cvGetTickFrequency()* LOOP_TIMES * 1000.) << "ms" << endl;
}
#else
for(int j = LOOPROISTART; j < LOOPROIEND; j ++)
{
Has_roi(j);
gmat_whole = mat;
gmat = gmat_whole(Rect(srcx, srcy, roicols, roirows));
src.setTo(val);
if(j == 0)
{
cout << "no roi:";
}
else
{
cout << "\nwith roi:";
};
gmat.setTo(val);
};
#endif
}
CPU_ON;
src.setTo(val);
CPU_OFF;
TEST_P(SetTo, With_mask)
{
#ifndef PRINT_KERNEL_RUN_TIME
double totalcputick = 0;
double totalgputick = 0;
double totalgputick_kernel = 0;
double t0 = 0;
double t1 = 0;
double t2 = 0;
for(int k = LOOPROISTART; k < LOOPROIEND; k++)
{
totalcputick = 0;
totalgputick = 0;
totalgputick_kernel = 0;
for(int j = 0; j < LOOP_TIMES + 1; j ++)
{
Has_roi(k);
t0 = (double)cvGetTickCount();//cpu start
mat_roi.setTo(val, mask_roi);
t0 = (double)cvGetTickCount() - t0;//cpu end
t1 = (double)cvGetTickCount();//gpu start1
gmat_whole = mat;
gmat = gmat_whole(Rect(srcx, srcy, roicols, roirows));
gmask = mask_roi;
t2 = (double)cvGetTickCount(); //kernel
gmat.setTo(val, gmask);
t2 = (double)cvGetTickCount() - t2;//kernel
cv::Mat cpu_dst;
gmat_whole.download(cpu_dst);//download
t1 = (double)cvGetTickCount() - t1;//gpu end1
if(j == 0)
continue;
totalgputick = t1 + totalgputick;
totalcputick = t0 + totalcputick;
totalgputick_kernel = t2 + totalgputick_kernel;
d_src.upload(src);
}
if(k == 0)
{
cout << "no roi\n";
}
else
{
cout << "with roi\n";
};
cout << "average cpu runtime is " << totalcputick / ((double)cvGetTickFrequency()* LOOP_TIMES * 1000.) << "ms" << endl;
cout << "average gpu runtime is " << totalgputick / ((double)cvGetTickFrequency()* LOOP_TIMES * 1000.) << "ms" << endl;
cout << "average gpu runtime without data transfer is " << totalgputick_kernel / ((double)cvGetTickFrequency()* LOOP_TIMES * 1000.) << "ms" << endl;
}
#else
for(int j = LOOPROISTART; j < LOOPROIEND; j ++)
{
Has_roi(j);
gmat_whole = mat;
gmat = gmat_whole(Rect(srcx, srcy, roicols, roirows));
WARMUP_ON;
d_src.setTo(val);
WARMUP_OFF;
gmask = mask_roi;
GPU_ON;
d_src.setTo(val);
;
GPU_OFF;
if(j == 0)
{
cout << "no roi:";
GPU_FULL_ON;
d_src.upload(src);
d_src.setTo(val);
GPU_FULL_OFF;
}
else
{
cout << "\nwith roi:";
};
gmat.setTo(val, gmask);
};
#endif
}
PARAM_TEST_CASE(DataTransfer, MatType, bool)
{
int type;
cv::Mat mat;
cv::ocl::oclMat gmat_whole;
virtual void SetUp()
{
type = GET_PARAM(0);
cv::RNG &rng = TS::ptr()->get_rng();
cv::Size size(MWIDTH, MHEIGHT);
mat = randomMat(rng, size, type, 5, 16, false);
}
};
TEST_P(DataTransfer, perf)
{
double totaluploadtick = 0;
double totaldownloadtick = 0;
double totaltick = 0;
double t0 = 0;
double t1 = 0;
cv::Mat cpu_dst;
for(int j = 0; j < LOOP_TIMES + 1; j ++)
{
t0 = (double)cvGetTickCount();
gmat_whole.upload(mat);//upload
t0 = (double)cvGetTickCount() - t0;
t1 = (double)cvGetTickCount();
gmat_whole.download(cpu_dst);//download
t1 = (double)cvGetTickCount() - t1;
if(j == 0)
continue;
totaluploadtick = t0 + totaluploadtick;
totaldownloadtick = t1 + totaldownloadtick;
}
totaltick = totaluploadtick + totaldownloadtick;
cout << "average upload time is " << totaluploadtick / ((double)cvGetTickFrequency()* LOOP_TIMES * 1000.) << "ms" << endl;
cout << "average download time is " << totaldownloadtick / ((double)cvGetTickFrequency()* LOOP_TIMES * 1000.) << "ms" << endl;
cout << "average data transfer time is " << totaltick / ((double)cvGetTickFrequency()* LOOP_TIMES * 1000.) << "ms" << endl;
}
//**********test************
INSTANTIATE_TEST_CASE_P(MatrixOperation, ConvertTo, Combine(
Values(CV_8UC1, CV_8UC4, CV_32FC1, CV_32FC4),
Values(CV_8UC1, CV_8UC4, CV_32FC1, CV_32FC4)));
INSTANTIATE_TEST_CASE_P(MatrixOperation, CopyTo, Combine(
Values(CV_8UC1, CV_8UC4, CV_32FC1, CV_32FC4),
Values(false))); // Values(false) is the reserved parameter
INSTANTIATE_TEST_CASE_P(MatrixOperation, SetTo, Combine(
Values(CV_8UC1, CV_8UC4, CV_32FC1, CV_32FC4),
Values(false))); // Values(false) is the reserved parameter
INSTANTIATE_TEST_CASE_P(MatrixOperation, DataTransfer, Combine(
Values(CV_8UC1, CV_8UC3, CV_8UC4, CV_32FC1, CV_32FC3, CV_32FC4),
Values(false))); // Values(false) is the reserved parameter
#endif
}

84
modules/ocl/perf/perf_norm.cpp
Unescape View File

@ -0,0 +1,84 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2010-2012, Multicoreware, Inc., all rights reserved.
// Copyright (C) 2010-2012, Advanced Micro Devices, Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// @Authors
// Fangfang Bai, fangfang@multicorewareinc.com
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other oclMaterials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors as is and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "precomp.hpp"
///////////// norm////////////////////////
TEST(norm)
{
Mat src, buf;
ocl::oclMat d_src, d_buf;
for (int size = Min_Size; size <= Max_Size; size *= Multiple)
{
SUBTEST << size << 'x' << size << "; CV_8UC1; NORM_INF";
gen(src, size, size, CV_8UC1, Scalar::all(0), Scalar::all(1));
gen(buf, size, size, CV_8UC1, Scalar::all(0), Scalar::all(1));
norm(src, NORM_INF);
CPU_ON;
norm(src, NORM_INF);
CPU_OFF;
d_src.upload(src);
d_buf.upload(buf);
WARMUP_ON;
ocl::norm(d_src, d_buf, NORM_INF);
WARMUP_OFF;
GPU_ON;
ocl::norm(d_src, d_buf, NORM_INF);
;
GPU_OFF;
GPU_FULL_ON;
d_src.upload(src);
ocl::norm(d_src, d_buf, NORM_INF);
GPU_FULL_OFF;
}
}

114
modules/ocl/perf/perf_pyrdown.cpp
Unescape View File

@ -1,4 +1,4 @@
///////////////////////////////////////////////////////////////////////////////////////
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
@ -15,7 +15,7 @@
// Third party copyrights are property of their respective owners.
//
// @Authors
// fangfang bai, fangfang@multicorewareinc.com
// Fangfang Bai, fangfang@multicorewareinc.com
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
@ -30,7 +30,7 @@
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// This software is provided by the copyright holders and contributors as is and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
@ -42,96 +42,46 @@
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "precomp.hpp"
#include <iomanip>
#ifdef HAVE_OPENCL
using namespace cv;
using namespace cv::ocl;
using namespace cvtest;
using namespace testing;
using namespace std;
PARAM_TEST_CASE(PyrDown, MatType, int)
///////////// pyrDown //////////////////////
TEST(pyrDown)
{
int type;
int channels;
//src mat
cv::Mat mat1;
cv::Mat dst;
//std::vector<cv::ocl::Info> oclinfo;
//ocl dst mat for testing
Mat src, dst;
int all_type[] = {CV_8UC1, CV_8UC4};
std::string type_name[] = {"CV_8UC1", "CV_8UC4"};
cv::ocl::oclMat gmat1;
cv::ocl::oclMat gdst;
virtual void SetUp()
for (int size = Min_Size; size <= Max_Size; size *= Multiple)
{
type = GET_PARAM(0);
channels = GET_PARAM(1);
//int devnums = getDevice(oclinfo);
//CV_Assert(devnums > 0);
}
};
#define VARNAME(A) string(#A);
////////////////////////////////PyrDown/////////////////////////////////////////////////
TEST_P(PyrDown, Mat)
{
cv::Size size(MWIDTH, MHEIGHT);
cv::RNG &rng = TS::ptr()->get_rng();
mat1 = randomMat(rng, size, CV_MAKETYPE(type, channels), 5, 16, false);
cv::ocl::oclMat gdst;
double totalgputick = 0;
double totalgputick_kernel = 0;
double t1 = 0;
double t2 = 0;
for (size_t j = 0; j < sizeof(all_type) / sizeof(int); j++)
{
SUBTEST << size << 'x' << size << "; " << type_name[j] ;
for (int j = 0; j < LOOP_TIMES + 1; j ++)
{
gen(src, size, size, all_type[j], 0, 256);
t1 = (double)cvGetTickCount();//gpu start1
pyrDown(src, dst);
cv::ocl::oclMat gmat1(mat1);
CPU_ON;
pyrDown(src, dst);
CPU_OFF;
t2 = (double)cvGetTickCount(); //kernel
cv::ocl::pyrDown(gmat1, gdst);
t2 = (double)cvGetTickCount() - t2;//kernel
ocl::oclMat d_src(src);
ocl::oclMat d_dst;
cv::Mat cpu_dst;
gdst.download(cpu_dst);
WARMUP_ON;
ocl::pyrDown(d_src, d_dst);
WARMUP_OFF;
t1 = (double)cvGetTickCount() - t1;//gpu end1
GPU_ON;
ocl::pyrDown(d_src, d_dst);
;
GPU_OFF;
if (j == 0)
{
continue;
GPU_FULL_ON;
d_src.upload(src);
ocl::pyrDown(d_src, d_dst);
d_dst.download(dst);
GPU_FULL_OFF;
}
totalgputick = t1 + totalgputick;
totalgputick_kernel = t2 + totalgputick_kernel;
}
cout << "average gpu runtime is " << totalgputick / ((double)cvGetTickFrequency()* LOOP_TIMES * 1000.) << "ms" << endl;
cout << "average gpu runtime without data transfer is " << totalgputick_kernel / ((double)cvGetTickFrequency()* LOOP_TIMES * 1000.) << "ms" << endl;
}
//********test****************
INSTANTIATE_TEST_CASE_P(GPU_ImgProc, PyrDown, Combine(
Values(CV_8U, CV_32F), Values(1, 4)));
#endif // HAVE_OPENCL
}

143
modules/ocl/perf/perf_pyrlk.cpp
Unescape View File

@ -0,0 +1,143 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2010-2012, Multicoreware, Inc., all rights reserved.
// Copyright (C) 2010-2012, Advanced Micro Devices, Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// @Authors
// Fangfang Bai, fangfang@multicorewareinc.com
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other oclMaterials provided with the distribution.
//
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors as is and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "precomp.hpp"
///////////// PyrLKOpticalFlow ////////////////////////
TEST(PyrLKOpticalFlow)
{
std::string images1[] = {"rubberwhale1.png", "aloeL.jpg"};
std::string images2[] = {"rubberwhale2.png", "aloeR.jpg"};
for (size_t i = 0; i < sizeof(images1) / sizeof(std::string); i++)
{
Mat frame0 = imread(abspath(images1[i]), i == 0 ? IMREAD_COLOR : IMREAD_GRAYSCALE);
if (frame0.empty())
{
std::string errstr = "can't open " + images1[i];
throw runtime_error(errstr);
}
Mat frame1 = imread(abspath(images2[i]), i == 0 ? IMREAD_COLOR : IMREAD_GRAYSCALE);
if (frame1.empty())
{
std::string errstr = "can't open " + images2[i];
throw runtime_error(errstr);
}
Mat gray_frame;
if (i == 0)
{
cvtColor(frame0, gray_frame, COLOR_BGR2GRAY);
}
for (int points = Min_Size; points <= Max_Size; points *= Multiple)
{
if (i == 0)
SUBTEST << frame0.cols << "x" << frame0.rows << "; color; " << points << " points";
else
SUBTEST << frame0.cols << "x" << frame0.rows << "; gray; " << points << " points";
Mat nextPts_cpu;
Mat status_cpu;
vector<Point2f> pts;
goodFeaturesToTrack(i == 0 ? gray_frame : frame0, pts, points, 0.01, 0.0);
vector<Point2f> nextPts;
vector<unsigned char> status;
vector<float> err;
calcOpticalFlowPyrLK(frame0, frame1, pts, nextPts, status, err);
CPU_ON;
calcOpticalFlowPyrLK(frame0, frame1, pts, nextPts, status, err);
CPU_OFF;
ocl::PyrLKOpticalFlow d_pyrLK;
ocl::oclMat d_frame0(frame0);
ocl::oclMat d_frame1(frame1);
ocl::oclMat d_pts;
Mat pts_mat(1, (int)pts.size(), CV_32FC2, (void *)&pts[0]);
d_pts.upload(pts_mat);
ocl::oclMat d_nextPts;
ocl::oclMat d_status;
ocl::oclMat d_err;
WARMUP_ON;
d_pyrLK.sparse(d_frame0, d_frame1, d_pts, d_nextPts, d_status, &d_err);
WARMUP_OFF;
GPU_ON;
d_pyrLK.sparse(d_frame0, d_frame1, d_pts, d_nextPts, d_status, &d_err);
;
GPU_OFF;
GPU_FULL_ON;
d_frame0.upload(frame0);
d_frame1.upload(frame1);
d_pts.upload(pts_mat);
d_pyrLK.sparse(d_frame0, d_frame1, d_pts, d_nextPts, d_status, &d_err);
if (!d_nextPts.empty())
{
d_nextPts.download(nextPts_cpu);
}
if (!d_status.empty())
{
d_status.download(status_cpu);
}
GPU_FULL_OFF;
}
}
}

97
modules/ocl/perf/perf_pyrup.cpp
Unescape View File

@ -15,7 +15,7 @@
// Third party copyrights are property of their respective owners.
//
// @Authors
// fangfang bai fangfang@multicorewareinc.com
// Fangfang Bai, fangfang@multicorewareinc.com
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
@ -30,7 +30,7 @@
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// This software is provided by the copyright holders and contributors as is and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
@ -42,81 +42,46 @@
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "opencv2/core/core.hpp"
#include "precomp.hpp"
#include <iomanip>
#ifdef HAVE_OPENCL
using namespace cv;
using namespace cv::ocl;
using namespace cvtest;
using namespace testing;
using namespace std;
PARAM_TEST_CASE(PyrUp, MatType, int)
///////////// pyrUp ////////////////////////
TEST(pyrUp)
{
int type;
int channels;
//std::vector<cv::ocl::Info> oclinfo;
Mat src, dst;
int all_type[] = {CV_8UC1, CV_8UC4};
std::string type_name[] = {"CV_8UC1", "CV_8UC4"};
virtual void SetUp()
for (int size = 500; size <= 2000; size *= 2)
{
type = GET_PARAM(0);
channels = GET_PARAM(1);
//int devnums = getDevice(oclinfo);
//CV_Assert(devnums > 0);
}
};
TEST_P(PyrUp, Performance)
{
cv::Size size(MWIDTH, MHEIGHT);
cv::Mat src = randomMat(size, CV_MAKETYPE(type, channels));
cv::Mat dst_gold;
cv::ocl::oclMat dst;
for (size_t j = 0; j < sizeof(all_type) / sizeof(int); j++)
{
SUBTEST << size << 'x' << size << "; " << type_name[j] ;
double totalgputick = 0;
double totalgputick_kernel = 0;
gen(src, size, size, all_type[j], 0, 256);
double t1 = 0;
double t2 = 0;
pyrUp(src, dst);
for (int j = 0; j < LOOP_TIMES + 1; j ++)
{
t1 = (double)cvGetTickCount();//gpu start1
CPU_ON;
pyrUp(src, dst);
CPU_OFF;
cv::ocl::oclMat srcMat = cv::ocl::oclMat(src);//upload
ocl::oclMat d_src(src);
ocl::oclMat d_dst;
t2 = (double)cvGetTickCount(); //kernel
cv::ocl::pyrUp(srcMat, dst);
t2 = (double)cvGetTickCount() - t2;//kernel
WARMUP_ON;
ocl::pyrUp(d_src, d_dst);
WARMUP_OFF;
cv::Mat cpu_dst;
dst.download(cpu_dst); //download
GPU_ON;
ocl::pyrUp(d_src, d_dst);
;
GPU_OFF;
t1 = (double)cvGetTickCount() - t1;//gpu end1
if (j == 0)
{
continue;
GPU_FULL_ON;
d_src.upload(src);
ocl::pyrUp(d_src, d_dst);
d_dst.download(dst);
GPU_FULL_OFF;
}
totalgputick = t1 + totalgputick;
totalgputick_kernel = t2 + totalgputick_kernel;
}
cout << "average gpu runtime is " << totalgputick / ((double)cvGetTickFrequency()* LOOP_TIMES * 1000.) << "ms" << endl;
cout << "average gpu runtime without data transfer is " << totalgputick_kernel / ((double)cvGetTickFrequency()* LOOP_TIMES * 1000.) << "ms" << endl;
}
INSTANTIATE_TEST_CASE_P(GPU_ImgProc, PyrUp, Combine(
Values(CV_8U, CV_32F), Values(1, 4)));
#endif // HAVE_OPENCL
}

483
modules/ocl/perf/perf_split_merge.cpp
Unescape View File

@ -10,12 +10,12 @@
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2010-2012, Institute Of Software Chinese Academy Of Science, all rights reserved.
// Copyright (C) 2010-2012, Multicoreware, Inc., all rights reserved.
// Copyright (C) 2010-2012, Advanced Micro Devices, Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// @Authors
// Jia Haipeng, jiahaipeng95@gmail.com
// Fangfang Bai, fangfang@multicorewareinc.com
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
@ -30,7 +30,7 @@
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// This software is provided by the copyright holders and contributors as is and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
@ -42,446 +42,109 @@
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "precomp.hpp"
#ifdef HAVE_OPENCL
using namespace cvtest;
using namespace testing;
using namespace std;
using namespace cv::ocl;
PARAM_TEST_CASE(MergeTestBase, MatType, int)
///////////// Merge////////////////////////
TEST(Merge)
{
int type;
int channels;
//src mat
cv::Mat mat1;
cv::Mat mat2;
cv::Mat mat3;
cv::Mat mat4;
//dst mat
cv::Mat dst;
// set up roi
int roicols;
int roirows;
int src1x;
int src1y;
int src2x;
int src2y;
int src3x;
int src3y;
int src4x;
int src4y;
int dstx;
int dsty;
//src mat with roi
cv::Mat mat1_roi;
cv::Mat mat2_roi;
cv::Mat mat3_roi;
cv::Mat mat4_roi;
//dst mat with roi
cv::Mat dst_roi;
//std::vector<cv::ocl::Info> oclinfo;
//ocl dst mat for testing
cv::ocl::oclMat gdst_whole;
//ocl mat with roi
cv::ocl::oclMat gmat1;
cv::ocl::oclMat gmat2;
cv::ocl::oclMat gmat3;
cv::ocl::oclMat gmat4;
cv::ocl::oclMat gdst;
Mat dst;
ocl::oclMat d_dst;
virtual void SetUp()
{
type = GET_PARAM(0);
channels = GET_PARAM(1);
cv::RNG &rng = TS::ptr()->get_rng();
cv::Size size(MWIDTH, MHEIGHT);
int channels = 4;
int all_type[] = {CV_8UC1, CV_32FC1};
std::string type_name[] = {"CV_8UC1", "CV_32FC1"};
mat1 = randomMat(rng, size, CV_MAKETYPE(type, 1), 5, 16, false);
mat2 = randomMat(rng, size, CV_MAKETYPE(type, 1), 5, 16, false);
mat3 = randomMat(rng, size, CV_MAKETYPE(type, 1), 5, 16, false);
mat4 = randomMat(rng, size, CV_MAKETYPE(type, 1), 5, 16, false);
dst = randomMat(rng, size, CV_MAKETYPE(type, channels), 5, 16, false);
//int devnums = getDevice(oclinfo);
//CV_Assert(devnums > 0);
////if you want to use undefault device, set it here
////setDevice(oclinfo[0]);
//setBinpath(CLBINPATH);
}
void Has_roi(int b)
for (int size = Min_Size; size <= Max_Size; size *= Multiple)
{
//cv::RNG& rng = TS::ptr()->get_rng();
if(b)
for (size_t j = 0; j < sizeof(all_type) / sizeof(int); j++)
{
//randomize ROI
roicols = mat1.cols - 1; //start
roirows = mat1.rows - 1;
src1x = 1;
src1y = 1;
src2x = 1;
src2y = 1;
src3x = 1;
src3y = 1;
src4x = 1;
src4y = 1;
dstx = 1;
dsty = 1;
SUBTEST << size << 'x' << size << "; " << type_name[j] ;
Size size1 = Size(size, size);
std::vector<Mat> src(channels);
}
else
{
roicols = mat1.cols;
roirows = mat1.rows;
src1x = 0;
src1y = 0;
src2x = 0;
src2y = 0;
src3x = 0;
src3y = 0;
src4x = 0;
src4y = 0;
dstx = 0;
dsty = 0;
};
for (int i = 0; i < channels; ++i)
{
src[i] = Mat(size1, all_type[j], cv::Scalar::all(i));
}
mat1_roi = mat1(Rect(src1x, src1y, roicols, roirows));
mat2_roi = mat2(Rect(src2x, src2y, roicols, roirows));
mat3_roi = mat3(Rect(src3x, src3y, roicols, roirows));
mat4_roi = mat4(Rect(src4x, src4y, roicols, roirows));
merge(src, dst);
CPU_ON;
merge(src, dst);
CPU_OFF;
dst_roi = dst(Rect(dstx, dsty, roicols, roirows));
}
std::vector<ocl::oclMat> d_src(channels);
};
for (int i = 0; i < channels; ++i)
{
d_src[i] = ocl::oclMat(size1, all_type[j], cv::Scalar::all(i));
}
struct Merge : MergeTestBase {};
WARMUP_ON;
ocl::merge(d_src, d_dst);
WARMUP_OFF;
TEST_P(Merge, Accuracy)
{
#ifndef PRINT_KERNEL_RUN_TIME
double totalcputick = 0;
double totalgputick = 0;
double totalgputick_kernel = 0;
double t0 = 0;
double t1 = 0;
double t2 = 0;
for(int k = LOOPROISTART; k < LOOPROIEND; k++)
{
totalcputick = 0;
totalgputick = 0;
totalgputick_kernel = 0;
for(int j = 0; j < LOOP_TIMES + 1; j ++)
{
Has_roi(k);
std::vector<cv::Mat> dev_src;
dev_src.push_back(mat1_roi);
dev_src.push_back(mat2_roi);
dev_src.push_back(mat3_roi);
dev_src.push_back(mat4_roi);
t0 = (double)cvGetTickCount();//cpu start
cv::merge(dev_src, dst_roi);
t0 = (double)cvGetTickCount() - t0;//cpu end
t1 = (double)cvGetTickCount();//gpu start1 ]
gmat1 = mat1_roi;
gmat2 = mat2_roi;
gmat3 = mat3_roi;
gmat4 = mat4_roi;
gdst_whole = dst;
gdst = gdst_whole(Rect(dstx, dsty, roicols, roirows));
std::vector<cv::ocl::oclMat> dev_gsrc;
dev_gsrc.push_back(gmat1);
dev_gsrc.push_back(gmat2);
dev_gsrc.push_back(gmat3);
dev_gsrc.push_back(gmat4);
t2 = (double)cvGetTickCount(); //kernel
cv::ocl::merge(dev_gsrc, gdst);
t2 = (double)cvGetTickCount() - t2;//kernel
cv::Mat cpu_dst;
gdst_whole.download (cpu_dst);//download
t1 = (double)cvGetTickCount() - t1;//gpu end1
GPU_ON;
ocl::merge(d_src, d_dst);
;
GPU_OFF;
if(j == 0)
continue;
GPU_FULL_ON;
totalgputick = t1 + totalgputick;
totalcputick = t0 + totalcputick;
totalgputick_kernel = t2 + totalgputick_kernel;
for (int i = 0; i < channels; ++i)
{
d_src[i] = ocl::oclMat(size1, CV_8U, cv::Scalar::all(i));
}
ocl::merge(d_src, d_dst);
d_dst.download(dst);
GPU_FULL_OFF;
}
if(k == 0)
{
cout << "no roi\n";
}
else
{
cout << "with roi\n";
};
cout << "average cpu runtime is " << totalcputick / ((double)cvGetTickFrequency()* LOOP_TIMES * 1000.) << "ms" << endl;
cout << "average gpu runtime is " << totalgputick / ((double)cvGetTickFrequency()* LOOP_TIMES * 1000.) << "ms" << endl;
cout << "average gpu runtime without data transfer is " << totalgputick_kernel / ((double)cvGetTickFrequency()* LOOP_TIMES * 1000.) << "ms" << endl;
}
#else
for(int j = LOOPROISTART; j < LOOPROIEND; j ++)
{
Has_roi(j);
gmat1 = mat1_roi;
gmat2 = mat2_roi;
gmat3 = mat3_roi;
gmat4 = mat4_roi;
gdst_whole = dst;
gdst = gdst_whole(Rect(dstx, dsty, roicols, roirows));
std::vector<cv::ocl::oclMat> dev_gsrc;
dev_gsrc.push_back(gmat1);
dev_gsrc.push_back(gmat2);
dev_gsrc.push_back(gmat3);
dev_gsrc.push_back(gmat4);
if(j == 0)
{
cout << "no roi:";
}
else
{
cout << "\nwith roi:";
};
cv::ocl::merge(dev_gsrc, gdst);
};
#endif
}
}
PARAM_TEST_CASE(SplitTestBase, MatType, int)
///////////// Split////////////////////////
TEST(Split)
{
int type;
int channels;
//src mat
cv::Mat mat;
//dstmat
cv::Mat dst1;
cv::Mat dst2;
cv::Mat dst3;
cv::Mat dst4;
// set up roi
int roicols;
int roirows;
int srcx;
int srcy;
int dst1x;
int dst1y;
int dst2x;
int dst2y;
int dst3x;
int dst3y;
int dst4x;
int dst4y;
//int channels = 4;
int all_type[] = {CV_8UC1, CV_32FC1};
std::string type_name[] = {"CV_8UC1", "CV_32FC1"};
//src mat with roi
cv::Mat mat_roi;
//dst mat with roi
cv::Mat dst1_roi;
cv::Mat dst2_roi;
cv::Mat dst3_roi;
cv::Mat dst4_roi;
//std::vector<cv::ocl::Info> oclinfo;
//ocl dst mat for testing
cv::ocl::oclMat gdst1_whole;
cv::ocl::oclMat gdst2_whole;
cv::ocl::oclMat gdst3_whole;
cv::ocl::oclMat gdst4_whole;
//ocl mat with roi
cv::ocl::oclMat gmat;
cv::ocl::oclMat gdst1;
cv::ocl::oclMat gdst2;
cv::ocl::oclMat gdst3;
cv::ocl::oclMat gdst4;
virtual void SetUp()
{
type = GET_PARAM(0);
channels = GET_PARAM(1);
cv::RNG &rng = TS::ptr()->get_rng();
cv::Size size(MWIDTH, MHEIGHT);
mat = randomMat(rng, size, CV_MAKETYPE(type, channels), 5, 16, false);
dst1 = randomMat(rng, size, CV_MAKETYPE(type, 1), 5, 16, false);
dst2 = randomMat(rng, size, CV_MAKETYPE(type, 1), 5, 16, false);
dst3 = randomMat(rng, size, CV_MAKETYPE(type, 1), 5, 16, false);
dst4 = randomMat(rng, size, CV_MAKETYPE(type, 1), 5, 16, false);
//int devnums = getDevice(oclinfo);
//CV_Assert(devnums > 0);
////if you want to use undefault device, set it here
////setDevice(oclinfo[0]);
//setBinpath(CLBINPATH);
}
void Has_roi(int b)
for (int size = Min_Size; size <= Max_Size; size *= Multiple)
{
//cv::RNG& rng = TS::ptr()->get_rng();
if(b)
for (size_t j = 0; j < sizeof(all_type) / sizeof(int); j++)
{
//randomize ROI
roicols = mat.cols - 1; //start
roirows = mat.rows - 1;
srcx = 1;
srcx = 1;
dst1x = 1;
dst1y = 1;
dst2x = 1;
dst2y = 1;
dst3x = 1;
dst3y = 1;
dst4x = 1;
dst4y = 1;
}
else
{
roicols = mat.cols;
roirows = mat.rows;
srcx = 0;
srcy = 0;
dst1x = 0;
dst1y = 0;
dst2x = 0;
dst2y = 0;
dst3x = 0;
dst3y = 0;
dst4x = 0;
dst4y = 0;
};
mat_roi = mat(Rect(srcx, srcy, roicols, roirows));
dst1_roi = dst1(Rect(dst1x, dst1y, roicols, roirows));
dst2_roi = dst2(Rect(dst2x, dst2y, roicols, roirows));
dst3_roi = dst3(Rect(dst3x, dst3y, roicols, roirows));
dst4_roi = dst4(Rect(dst4x, dst4y, roicols, roirows));
}
SUBTEST << size << 'x' << size << "; " << type_name[j];
Size size1 = Size(size, size);
};
Mat src(size1, CV_MAKE_TYPE(all_type[j], 4), cv::Scalar(1, 2, 3, 4));
struct Split : SplitTestBase {};
std::vector<cv::Mat> dst;
TEST_P(Split, Accuracy)
{
#ifndef PRINT_KERNEL_RUN_TIME
double totalcputick = 0;
double totalgputick = 0;
double totalgputick_kernel = 0;
double t0 = 0;
double t1 = 0;
double t2 = 0;
for(int k = LOOPROISTART; k < LOOPROIEND; k++)
{
totalcputick = 0;
totalgputick = 0;
totalgputick_kernel = 0;
for(int j = 0; j < LOOP_TIMES + 1; j ++)
{
Has_roi(k);
cv::Mat dev_dst[4] = {dst1_roi, dst2_roi, dst3_roi, dst4_roi};
cv::ocl::oclMat dev_gdst[4] = {gdst1, gdst2, gdst3, gdst4};
t0 = (double)cvGetTickCount();//cpu start
cv::split(mat_roi, dev_dst);
t0 = (double)cvGetTickCount() - t0;//cpu end
split(src, dst);
t1 = (double)cvGetTickCount();//gpu start1
gdst1_whole = dst1;
gdst1 = gdst1_whole(Rect(dst1x, dst1y, roicols, roirows));
CPU_ON;
split(src, dst);
CPU_OFF;
gdst2_whole = dst2;
gdst2 = gdst2_whole(Rect(dst2x, dst2y, roicols, roirows));
ocl::oclMat d_src(size1, CV_MAKE_TYPE(all_type[j], 4), cv::Scalar(1, 2, 3, 4));
std::vector<cv::ocl::oclMat> d_dst;
gdst3_whole = dst3;
gdst3 = gdst3_whole(Rect(dst3x, dst3y, roicols, roirows));
WARMUP_ON;
ocl::split(d_src, d_dst);
WARMUP_OFF;
gdst4_whole = dst4;
gdst4 = gdst4_whole(Rect(dst4x, dst4y, roicols, roirows));
GPU_ON;
ocl::split(d_src, d_dst);
;
GPU_OFF;
gmat = mat_roi;
t2 = (double)cvGetTickCount(); //kernel
cv::ocl::split(gmat, dev_gdst);
t2 = (double)cvGetTickCount() - t2;//kernel
cv::Mat cpu_dst1;
cv::Mat cpu_dst2;
cv::Mat cpu_dst3;
cv::Mat cpu_dst4;
gdst1_whole.download(cpu_dst1);
gdst2_whole.download(cpu_dst2);
gdst3_whole.download(cpu_dst3);
gdst4_whole.download(cpu_dst4);
t1 = (double)cvGetTickCount() - t1;//gpu end1
if(j == 0)
continue;
totalgputick = t1 + totalgputick;
totalcputick = t0 + totalcputick;
totalgputick_kernel = t2 + totalgputick_kernel;
}
if(k == 0)
{
cout << "no roi\n";
GPU_FULL_ON;
d_src.upload(src);
ocl::split(d_src, d_dst);
GPU_FULL_OFF;
}
else
{
cout << "with roi\n";
};
cout << "average cpu runtime is " << totalcputick / ((double)cvGetTickFrequency()* LOOP_TIMES * 1000.) << "ms" << endl;
cout << "average gpu runtime is " << totalgputick / ((double)cvGetTickFrequency()* LOOP_TIMES * 1000.) << "ms" << endl;
cout << "average gpu runtime without data transfer is " << totalgputick_kernel / ((double)cvGetTickFrequency()* LOOP_TIMES * 1000.) << "ms" << endl;
}
#else
for(int j = LOOPROISTART; j < LOOPROIEND; j ++)
{
Has_roi(j);
//cv::Mat dev_dst[4] = {dst1_roi, dst2_roi, dst3_roi, dst4_roi};
cv::ocl::oclMat dev_gdst[4] = {gdst1, gdst2, gdst3, gdst4};
gdst1_whole = dst1;
gdst1 = gdst1_whole(Rect(dst1x, dst1y, roicols, roirows));
gdst2_whole = dst2;
gdst2 = gdst2_whole(Rect(dst2x, dst2y, roicols, roirows));
gdst3_whole = dst3;
gdst3 = gdst3_whole(Rect(dst3x, dst3y, roicols, roirows));
gdst4_whole = dst4;
gdst4 = gdst4_whole(Rect(dst4x, dst4y, roicols, roirows));
gmat = mat_roi;
if(j == 0)
{
cout << "no roi:";
}
else
{
cout << "\nwith roi:";
};
cv::ocl::split(gmat, dev_gdst);
};
#endif
}
}
//*************test*****************
INSTANTIATE_TEST_CASE_P(SplitMerge, Merge, Combine(
Values(CV_8UC4, CV_32FC4), Values(1, 4)));
INSTANTIATE_TEST_CASE_P(SplitMerge, Split , Combine(
Values(CV_8U, CV_32S, CV_32F), Values(1, 4)));
#endif // HAVE_OPENCL

330
modules/ocl/perf/precomp.cpp
Unescape View File

@ -7,12 +7,13 @@
// copy or use the software.
//
//
// Intel License Agreement
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000, Intel Corporation, all rights reserved.
// Copyright (C) 2010-2012, Multicoreware, Inc., all rights reserved.
// Copyright (C) 2010-2012, Advanced Micro Devices, Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
@ -21,12 +22,12 @@
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
// and/or other oclMaterials provided with the distribution.
//
// * The name of Intel Corporation may not be used to endorse or promote products
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// This software is provided by the copyright holders and contributors as is and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
@ -41,4 +42,321 @@
#include "precomp.hpp"
// This program test most of the functions in ocl module and generate data metrix of x-factor in .csv files
// All images needed in this test are in samples/gpu folder.
// For haar template, haarcascade_frontalface_alt.xml shouold be in working directory
void TestSystem::run()
{
if (is_list_mode_)
{
for (vector<Runnable *>::iterator it = tests_.begin(); it != tests_.end(); ++it)
{
cout << (*it)->name() << endl;
}
return;
}
// Run test initializers
for (vector<Runnable *>::iterator it = inits_.begin(); it != inits_.end(); ++it)
{
if ((*it)->name().find(test_filter_, 0) != string::npos)
{
(*it)->run();
}
}
printHeading();
writeHeading();
// Run tests
for (vector<Runnable *>::iterator it = tests_.begin(); it != tests_.end(); ++it)
{
try
{
if ((*it)->name().find(test_filter_, 0) != string::npos)
{
cout << endl << (*it)->name() << ":\n";
setCurrentTest((*it)->name());
//fprintf(record_,"%s\n",(*it)->name().c_str());
(*it)->run();
finishCurrentSubtest();
}
}
catch (const Exception &)
{
// Message is printed via callback
resetCurrentSubtest();
}
catch (const runtime_error &e)
{
printError(e.what());
resetCurrentSubtest();
}
}
printSummary();
writeSummary();
}
void TestSystem::finishCurrentSubtest()
{
if (cur_subtest_is_empty_)
// There is no need to print subtest statistics
{
return;
}
double cpu_time = cpu_elapsed_ / getTickFrequency() * 1000.0;
double gpu_time = gpu_elapsed_ / getTickFrequency() * 1000.0;
double gpu_full_time = gpu_full_elapsed_ / getTickFrequency() * 1000.0;
double speedup = static_cast<double>(cpu_elapsed_) / std::max(1.0, gpu_elapsed_);
speedup_total_ += speedup;
double fullspeedup = static_cast<double>(cpu_elapsed_) / std::max(1.0, gpu_full_elapsed_);
speedup_full_total_ += fullspeedup;
if (speedup > top_)
{
speedup_faster_count_++;
}
else if (speedup < bottom_)
{
speedup_slower_count_++;
}
else
{
speedup_equal_count_++;
}
if (fullspeedup > top_)
{
speedup_full_faster_count_++;
}
else if (fullspeedup < bottom_)
{
speedup_full_slower_count_++;
}
else
{
speedup_full_equal_count_++;
}
// compute min, max and
std::sort(gpu_times_.begin(), gpu_times_.end());
double gpu_min = gpu_times_.front() / getTickFrequency() * 1000.0;
double gpu_max = gpu_times_.back() / getTickFrequency() * 1000.0;
double deviation = 0;
if (gpu_times_.size() > 1)
{
double sum = 0;
for (size_t i = 0; i < gpu_times_.size(); i++)
{
int64 diff = gpu_times_[i] - static_cast<int64>(gpu_elapsed_);
double diff_time = diff * 1000 / getTickFrequency();
sum += diff_time * diff_time;
}
deviation = std::sqrt(sum / gpu_times_.size());
}
printMetrics(cpu_time, gpu_time, gpu_full_time, speedup, fullspeedup);
writeMetrics(cpu_time, gpu_time, gpu_full_time, speedup, fullspeedup, gpu_min, gpu_max, deviation);
num_subtests_called_++;
resetCurrentSubtest();
}
double TestSystem::meanTime(const vector<int64> &samples)
{
double sum = accumulate(samples.begin(), samples.end(), 0.);
return sum / samples.size();
}
void TestSystem::printHeading()
{
cout << endl;
cout << setiosflags(ios_base::left);
cout << TAB << setw(10) << "CPU, ms" << setw(10) << "GPU, ms"
<< setw(14) << "SPEEDUP" << setw(14) << "GPUTOTAL, ms" << setw(14) << "TOTALSPEEDUP"
<< "DESCRIPTION\n";
cout << resetiosflags(ios_base::left);
}
void TestSystem::writeHeading()
{
if (!record_)
{
recordname_ += "_OCL.csv";
record_ = fopen(recordname_.c_str(), "w");
}
fprintf(record_, "NAME,DESCRIPTION,CPU (ms),GPU (ms),SPEEDUP,GPUTOTAL (ms),TOTALSPEEDUP,GPU Min (ms),GPU Max (ms), Standard deviation (ms)\n");
fflush(record_);
}
void TestSystem::printSummary()
{
cout << setiosflags(ios_base::fixed);
cout << "\naverage GPU speedup: x"
<< setprecision(3) << speedup_total_ / std::max(1, num_subtests_called_)
<< endl;
cout << "\nGPU exceeded: "
<< setprecision(3) << speedup_faster_count_
<< "\nGPU passed: "
<< setprecision(3) << speedup_equal_count_
<< "\nGPU failed: "
<< setprecision(3) << speedup_slower_count_
<< endl;
cout << "\nGPU exceeded rate: "
<< setprecision(3) << (float)speedup_faster_count_ / std::max(1, num_subtests_called_) * 100
<< "%"
<< "\nGPU passed rate: "
<< setprecision(3) << (float)speedup_equal_count_ / std::max(1, num_subtests_called_) * 100
<< "%"
<< "\nGPU failed rate: "
<< setprecision(3) << (float)speedup_slower_count_ / std::max(1, num_subtests_called_) * 100
<< "%"
<< endl;
cout << "\naverage GPUTOTAL speedup: x"
<< setprecision(3) << speedup_full_total_ / std::max(1, num_subtests_called_)
<< endl;
cout << "\nGPUTOTAL exceeded: "
<< setprecision(3) << speedup_full_faster_count_
<< "\nGPUTOTAL passed: "
<< setprecision(3) << speedup_full_equal_count_
<< "\nGPUTOTAL failed: "
<< setprecision(3) << speedup_full_slower_count_
<< endl;
cout << "\nGPUTOTAL exceeded rate: "
<< setprecision(3) << (float)speedup_full_faster_count_ / std::max(1, num_subtests_called_) * 100
<< "%"
<< "\nGPUTOTAL passed rate: "
<< setprecision(3) << (float)speedup_full_equal_count_ / std::max(1, num_subtests_called_) * 100
<< "%"
<< "\nGPUTOTAL failed rate: "
<< setprecision(3) << (float)speedup_full_slower_count_ / std::max(1, num_subtests_called_) * 100
<< "%"
<< endl;
cout << resetiosflags(ios_base::fixed);
}
void TestSystem::printMetrics(double cpu_time, double gpu_time, double gpu_full_time, double speedup, double fullspeedup)
{
cout << TAB << setiosflags(ios_base::left);
stringstream stream;
stream << cpu_time;
cout << setw(10) << stream.str();
stream.str("");
stream << gpu_time;
cout << setw(10) << stream.str();
stream.str("");
stream << "x" << setprecision(3) << speedup;
cout << setw(14) << stream.str();
stream.str("");
stream << gpu_full_time;
cout << setw(14) << stream.str();
stream.str("");
stream << "x" << setprecision(3) << fullspeedup;
cout << setw(14) << stream.str();
cout << cur_subtest_description_.str();
cout << resetiosflags(ios_base::left) << endl;
}
void TestSystem::writeMetrics(double cpu_time, double gpu_time, double gpu_full_time, double speedup, double fullspeedup, double gpu_min, double gpu_max, double std_dev)
{
if (!record_)
{
recordname_ += ".csv";
record_ = fopen(recordname_.c_str(), "w");
}
fprintf(record_, "%s,%s,%.3f,%.3f,%.3f,%.3f,%.3f,%.3f,%.3f,%.3f\n", itname_changed_ ? itname_.c_str() : "",
cur_subtest_description_.str().c_str(),
cpu_time, gpu_time, speedup, gpu_full_time, fullspeedup,
gpu_min, gpu_max, std_dev);
if (itname_changed_)
{
itname_changed_ = false;
}
fflush(record_);
}
void TestSystem::writeSummary()
{
if (!record_)
{
recordname_ += ".csv";
record_ = fopen(recordname_.c_str(), "w");
}
fprintf(record_, "\nAverage GPU speedup: %.3f\n"
"exceeded: %d (%.3f%%)\n"
"passed: %d (%.3f%%)\n"
"failed: %d (%.3f%%)\n"
"\nAverage GPUTOTAL speedup: %.3f\n"
"exceeded: %d (%.3f%%)\n"
"passed: %d (%.3f%%)\n"
"failed: %d (%.3f%%)\n",
speedup_total_ / std::max(1, num_subtests_called_),
speedup_faster_count_, (float)speedup_faster_count_ / std::max(1, num_subtests_called_) * 100,
speedup_equal_count_, (float)speedup_equal_count_ / std::max(1, num_subtests_called_) * 100,
speedup_slower_count_, (float)speedup_slower_count_ / std::max(1, num_subtests_called_) * 100,
speedup_full_total_ / std::max(1, num_subtests_called_),
speedup_full_faster_count_, (float)speedup_full_faster_count_ / std::max(1, num_subtests_called_) * 100,
speedup_full_equal_count_, (float)speedup_full_equal_count_ / std::max(1, num_subtests_called_) * 100,
speedup_full_slower_count_, (float)speedup_full_slower_count_ / std::max(1, num_subtests_called_) * 100
);
fflush(record_);
}
void TestSystem::printError(const std::string &msg)
{
if(msg != "CL_INVALID_BUFFER_SIZE")
{
cout << TAB << "[error: " << msg << "] " << cur_subtest_description_.str() << endl;
}
}
void gen(Mat &mat, int rows, int cols, int type, Scalar low, Scalar high)
{
mat.create(rows, cols, type);
RNG rng(0);
rng.fill(mat, RNG::UNIFORM, low, high);
}
string abspath(const string &relpath)
{
return TestSystem::instance().workingDir() + relpath;
}
int CV_CDECL cvErrorCallback(int /*status*/, const char * /*func_name*/,
const char *err_msg, const char * /*file_name*/,
int /*line*/, void * /*userdata*/)
{
TestSystem::instance().printError(err_msg);
return 0;
}

386
modules/ocl/perf/precomp.hpp
Unescape View File

@ -7,12 +7,13 @@
// copy or use the software.
//
//
// Intel License Agreement
// License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000, Intel Corporation, all rights reserved.
// Copyright (C) 2010-2012, Multicoreware, Inc., all rights reserved.
// Copyright (C) 2010-2012, Advanced Micro Devices, Inc., all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
@ -21,12 +22,12 @@
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
// and/or other oclMaterials provided with the distribution.
//
// * The name of Intel Corporation may not be used to endorse or promote products
// * The name of the copyright holders may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// This software is provided by the copyright holders and contributors as is and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
@ -39,43 +40,352 @@
//
//M*/
#ifdef __GNUC__
# pragma GCC diagnostic ignored "-Wmissing-declarations"
# if defined __clang__ || defined __APPLE__
# pragma GCC diagnostic ignored "-Wmissing-prototypes"
# pragma GCC diagnostic ignored "-Wextra"
# endif
#endif
#ifndef __OPENCV_TEST_PRECOMP_HPP__
#define __OPENCV_TEST_PRECOMP_HPP__
#include <cmath>
#include <cstdio>
#include <iostream>
#include <fstream>
#include <sstream>
#include <iomanip>
#include <stdexcept>
#include <string>
#include <limits>
#include <algorithm>
#include <iterator>
#include <string>
#include <cstdarg>
#include "cvconfig.h"
#include <iostream>
#include <cstdio>
#include <vector>
#include <numeric>
#include "opencv2/core/core.hpp"
#include "opencv2/highgui/highgui.hpp"
//#include "opencv2/calib3d/calib3d.hpp"
#include "opencv2/imgproc/imgproc.hpp"
#include "opencv2/highgui/highgui.hpp"
#include "opencv2/video/video.hpp"
#include "opencv2/ts/ts.hpp"
#include "opencv2/ts/ts_perf.hpp"
#include "opencv2/objdetect/objdetect.hpp"
#include "opencv2/features2d/features2d.hpp"
#include "opencv2/ocl/ocl.hpp"
//#include "opencv2/nonfree/nonfree.hpp"
#include "utility.hpp"
#include "interpolation.hpp"
//#include "add_test_info.h"
//#define PERF_TEST_OCL 1
#define Min_Size 1000
#define Max_Size 4000
#define Multiple 2
#define TAB " "
using namespace std;
using namespace cv;
void gen(Mat &mat, int rows, int cols, int type, Scalar low, Scalar high);
string abspath(const string &relpath);
int CV_CDECL cvErrorCallback(int, const char *, const char *, const char *, int, void *);
typedef struct
{
short x;
short y;
} COOR;
COOR do_meanShift(int x0, int y0, uchar *sptr, uchar *dptr, int sstep,
cv::Size size, int sp, int sr, int maxIter, float eps, int *tab);
void meanShiftProc_(const Mat &src_roi, Mat &dst_roi, Mat &dstCoor_roi,
int sp, int sr, cv::TermCriteria crit);
class Runnable
{
public:
explicit Runnable(const std::string &runname): name_(runname) {}
virtual ~Runnable() {}
const std::string &name() const
{
return name_;
}
virtual void run() = 0;
private:
std::string name_;
};
class TestSystem
{
public:
static TestSystem &instance()
{
static TestSystem me;
return me;
}
void setWorkingDir(const std::string &val)
{
working_dir_ = val;
}
const std::string &workingDir() const
{
return working_dir_;
}
void setTestFilter(const std::string &val)
{
test_filter_ = val;
}
const std::string &testFilter() const
{
return test_filter_;
}
void setNumIters(int num_iters)
{
num_iters_ = num_iters;
}
void setGPUWarmupIters(int num_iters)
{
gpu_warmup_iters_ = num_iters;
}
void setCPUIters(int num_iters)
{
cpu_num_iters_ = num_iters;
}
void setTopThreshold(double top)
{
top_ = top;
}
void setBottomThreshold(double bottom)
{
bottom_ = bottom;
}
void addInit(Runnable *init)
{
inits_.push_back(init);
}
void addTest(Runnable *test)
{
tests_.push_back(test);
}
void run();
// It's public because OpenCV callback uses it
void printError(const std::string &msg);
std::stringstream &startNewSubtest()
{
finishCurrentSubtest();
return cur_subtest_description_;
}
bool stop() const
{
return cur_iter_idx_ >= num_iters_;
}
bool cpu_stop() const
{
return cur_iter_idx_ >= cpu_num_iters_;
}
bool warmupStop()
{
return cur_warmup_idx_++ >= gpu_warmup_iters_;
}
void warmupComplete()
{
cur_warmup_idx_ = 0;
}
void cpuOn()
{
cpu_started_ = cv::getTickCount();
}
void cpuOff()
{
int64 delta = cv::getTickCount() - cpu_started_;
cpu_times_.push_back(delta);
++cur_iter_idx_;
}
void cpuComplete()
{
cpu_elapsed_ += meanTime(cpu_times_);
cur_subtest_is_empty_ = false;
cur_iter_idx_ = 0;
}
void gpuOn()
{
gpu_started_ = cv::getTickCount();
}
void gpuOff()
{
int64 delta = cv::getTickCount() - gpu_started_;
gpu_times_.push_back(delta);
++cur_iter_idx_;
}
void gpuComplete()
{
gpu_elapsed_ += meanTime(gpu_times_);
cur_subtest_is_empty_ = false;
cur_iter_idx_ = 0;
}
void gpufullOn()
{
gpu_full_started_ = cv::getTickCount();
}
void gpufullOff()
{
int64 delta = cv::getTickCount() - gpu_full_started_;
gpu_full_times_.push_back(delta);
++cur_iter_idx_;
}
void gpufullComplete()
{
gpu_full_elapsed_ += meanTime(gpu_full_times_);
cur_subtest_is_empty_ = false;
cur_iter_idx_ = 0;
}
bool isListMode() const
{
return is_list_mode_;
}
void setListMode(bool value)
{
is_list_mode_ = value;
}
void setRecordName(const std::string &name)
{
recordname_ = name;
}
void setCurrentTest(const std::string &name)
{
itname_ = name;
itname_changed_ = true;
}
private:
TestSystem():
cur_subtest_is_empty_(true), cpu_elapsed_(0),
gpu_elapsed_(0), gpu_full_elapsed_(0), speedup_total_(0.0),
num_subtests_called_(0),
speedup_faster_count_(0), speedup_slower_count_(0), speedup_equal_count_(0),
speedup_full_faster_count_(0), speedup_full_slower_count_(0), speedup_full_equal_count_(0), is_list_mode_(false),
num_iters_(10), cpu_num_iters_(2),
gpu_warmup_iters_(1), cur_iter_idx_(0), cur_warmup_idx_(0),
record_(0), recordname_("performance"), itname_changed_(true)
{
cpu_times_.reserve(num_iters_);
gpu_times_.reserve(num_iters_);
gpu_full_times_.reserve(num_iters_);
}
void finishCurrentSubtest();
void resetCurrentSubtest()
{
cpu_elapsed_ = 0;
gpu_elapsed_ = 0;
gpu_full_elapsed_ = 0;
cur_subtest_description_.str("");
cur_subtest_is_empty_ = true;
cur_iter_idx_ = 0;
cpu_times_.clear();
gpu_times_.clear();
gpu_full_times_.clear();
}
double meanTime(const std::vector<int64> &samples);
void printHeading();
void printSummary();
void printMetrics(double cpu_time, double gpu_time = 0.0f, double gpu_full_time = 0.0f, double speedup = 0.0f, double fullspeedup = 0.0f);
void writeHeading();
void writeSummary();
void writeMetrics(double cpu_time, double gpu_time = 0.0f, double gpu_full_time = 0.0f,
double speedup = 0.0f, double fullspeedup = 0.0f,
double gpu_min = 0.0f, double gpu_max = 0.0f, double std_dev = 0.0f);
std::string working_dir_;
std::string test_filter_;
std::vector<Runnable *> inits_;
std::vector<Runnable *> tests_;
std::stringstream cur_subtest_description_;
bool cur_subtest_is_empty_;
int64 cpu_started_;
int64 gpu_started_;
int64 gpu_full_started_;
double cpu_elapsed_;
double gpu_elapsed_;
double gpu_full_elapsed_;
double speedup_total_;
double speedup_full_total_;
int num_subtests_called_;
int speedup_faster_count_;
int speedup_slower_count_;
int speedup_equal_count_;
int speedup_full_faster_count_;
int speedup_full_slower_count_;
int speedup_full_equal_count_;
bool is_list_mode_;
double top_;
double bottom_;
int num_iters_;
int cpu_num_iters_; //there's no need to set cpu running same times with gpu
int gpu_warmup_iters_; //gpu warm up times, default is 1
int cur_iter_idx_;
int cur_warmup_idx_; //current gpu warm up times
std::vector<int64> cpu_times_;
std::vector<int64> gpu_times_;
std::vector<int64> gpu_full_times_;
FILE *record_;
std::string recordname_;
std::string itname_;
bool itname_changed_;
};
#define GLOBAL_INIT(name) \
struct name##_init: Runnable { \
name##_init(): Runnable(#name) { \
TestSystem::instance().addInit(this); \
} \
void run(); \
} name##_init_instance; \
void name##_init::run()
#define TEST(name) \
struct name##_test: Runnable { \
name##_test(): Runnable(#name) { \
TestSystem::instance().addTest(this); \
} \
void run(); \
} name##_test_instance; \
void name##_test::run()
#define SUBTEST TestSystem::instance().startNewSubtest()
#define CPU_ON \
while (!TestSystem::instance().cpu_stop()) { \
TestSystem::instance().cpuOn()
#define CPU_OFF \
TestSystem::instance().cpuOff(); \
} TestSystem::instance().cpuComplete()
#define GPU_ON \
while (!TestSystem::instance().stop()) { \
TestSystem::instance().gpuOn()
#define GPU_OFF \
TestSystem::instance().gpuOff(); \
} TestSystem::instance().gpuComplete()
#endif
#define GPU_FULL_ON \
while (!TestSystem::instance().stop()) { \
TestSystem::instance().gpufullOn()
#define GPU_FULL_OFF \
TestSystem::instance().gpufullOff(); \
} TestSystem::instance().gpufullComplete()
#define WARMUP_ON \
while (!TestSystem::instance().warmupStop()) {
#define WARMUP_OFF \
} TestSystem::instance().warmupComplete()

265
modules/ocl/perf/utility.cpp
Unescape View File

@ -1,265 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// Intel License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000, Intel Corporation, all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of Intel Corporation may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "precomp.hpp"
#define VARNAME(A) #A
using namespace std;
using namespace cv;
using namespace cv::gpu;
using namespace cvtest;
//std::string generateVarList(int first,...)
//{
// vector<std::string> varname;
//
// va_list argp;
// string s;
// stringstream ss;
// va_start(argp,first);
// int i=first;
// while(i!=-1)
// {
// ss<<i<<",";
// i=va_arg(argp,int);
// };
// s=ss.str();
// va_end(argp);
// return s;
//};
//std::string generateVarList(int& p1,int& p2)
//{
// stringstream ss;
// ss<<VARNAME(p1)<<":"<<src1x<<","<<VARNAME(p2)<<":"<<src1y;
// return ss.str();
//};
int randomInt(int minVal, int maxVal)
{
RNG &rng = TS::ptr()->get_rng();
return rng.uniform(minVal, maxVal);
}
double randomDouble(double minVal, double maxVal)
{
RNG &rng = TS::ptr()->get_rng();
return rng.uniform(minVal, maxVal);
}
Size randomSize(int minVal, int maxVal)
{
return cv::Size(randomInt(minVal, maxVal), randomInt(minVal, maxVal));
}
Scalar randomScalar(double minVal, double maxVal)
{
return Scalar(randomDouble(minVal, maxVal), randomDouble(minVal, maxVal), randomDouble(minVal, maxVal), randomDouble(minVal, maxVal));
}
Mat randomMat(Size size, int type, double minVal, double maxVal)
{
return randomMat(TS::ptr()->get_rng(), size, type, minVal, maxVal, false);
}
/*
void showDiff(InputArray gold_, InputArray actual_, double eps)
{
Mat gold;
if (gold_.kind() == _InputArray::MAT)
gold = gold_.getMat();
else
gold_.getGpuMat().download(gold);
Mat actual;
if (actual_.kind() == _InputArray::MAT)
actual = actual_.getMat();
else
actual_.getGpuMat().download(actual);
Mat diff;
absdiff(gold, actual, diff);
threshold(diff, diff, eps, 255.0, cv::THRESH_BINARY);
namedWindow("gold", WINDOW_NORMAL);
namedWindow("actual", WINDOW_NORMAL);
namedWindow("diff", WINDOW_NORMAL);
imshow("gold", gold);
imshow("actual", actual);
imshow("diff", diff);
waitKey();
}
*/
/*
bool supportFeature(const DeviceInfo& info, FeatureSet feature)
{
return TargetArchs::builtWith(feature) && info.supports(feature);
}
const vector<DeviceInfo>& devices()
{
static vector<DeviceInfo> devs;
static bool first = true;
if (first)
{
int deviceCount = getCudaEnabledDeviceCount();
devs.reserve(deviceCount);
for (int i = 0; i < deviceCount; ++i)
{
DeviceInfo info(i);
if (info.isCompatible())
devs.push_back(info);
}
first = false;
}
return devs;
}
vector<DeviceInfo> devices(FeatureSet feature)
{
const vector<DeviceInfo>& d = devices();
vector<DeviceInfo> devs_filtered;
if (TargetArchs::builtWith(feature))
{
devs_filtered.reserve(d.size());
for (size_t i = 0, size = d.size(); i < size; ++i)
{
const DeviceInfo& info = d[i];
if (info.supports(feature))
devs_filtered.push_back(info);
}
}
return devs_filtered;
}
*/
vector<MatType> types(int depth_start, int depth_end, int cn_start, int cn_end)
{
vector<MatType> v;
v.reserve((depth_end - depth_start + 1) * (cn_end - cn_start + 1));
for (int depth = depth_start; depth <= depth_end; ++depth)
{
for (int cn = cn_start; cn <= cn_end; ++cn)
{
v.push_back(CV_MAKETYPE(depth, cn));
}
}
return v;
}
const vector<MatType> &all_types()
{
static vector<MatType> v = types(CV_8U, CV_64F, 1, 4);
return v;
}
Mat readImage(const string &fileName, int flags)
{
return imread(string(cvtest::TS::ptr()->get_data_path()) + fileName, flags);
}
Mat readImageType(const string &fname, int type)
{
Mat src = readImage(fname, CV_MAT_CN(type) == 1 ? IMREAD_GRAYSCALE : IMREAD_COLOR);
if (CV_MAT_CN(type) == 4)
{
Mat temp;
cvtColor(src, temp, cv::COLOR_BGR2BGRA);
swap(src, temp);
}
src.convertTo(src, CV_MAT_DEPTH(type));
return src;
}
double checkNorm(const Mat &m)
{
return norm(m, NORM_INF);
}
double checkNorm(const Mat &m1, const Mat &m2)
{
return norm(m1, m2, NORM_INF);
}
double checkSimilarity(const Mat &m1, const Mat &m2)
{
Mat diff;
matchTemplate(m1, m2, diff, CV_TM_CCORR_NORMED);
return std::abs(diff.at<float>(0, 0) - 1.f);
}
/*
void cv::ocl::PrintTo(const DeviceInfo& info, ostream* os)
{
(*os) << info.name();
}
*/
void PrintTo(const Inverse &inverse, std::ostream *os)
{
if (inverse)
(*os) << "inverse";
else
(*os) << "direct";
}

182
modules/ocl/perf/utility.hpp
Unescape View File

@ -1,182 +0,0 @@
/*M///////////////////////////////////////////////////////////////////////////////////////
//
// IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
// By downloading, copying, installing or using the software you agree to this license.
// If you do not agree to this license, do not download, install,
// copy or use the software.
//
//
// Intel License Agreement
// For Open Source Computer Vision Library
//
// Copyright (C) 2000, Intel Corporation, all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
// * Redistribution's of source code must retain the above copyright notice,
// this list of conditions and the following disclaimer.
//
// * Redistribution's in binary form must reproduce the above copyright notice,
// this list of conditions and the following disclaimer in the documentation
// and/or other materials provided with the distribution.
//
// * The name of Intel Corporation may not be used to endorse or promote products
// derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#ifndef __OPENCV_TEST_UTILITY_HPP__
#define __OPENCV_TEST_UTILITY_HPP__
//#define PRINT_KERNEL_RUN_TIME
#ifdef PRINT_KERNEL_RUN_TIME
#define LOOP_TIMES 1
#else
#define LOOP_TIMES 1
#endif
#define MWIDTH 1920
#define MHEIGHT 1080
#define CLBINPATH ".\\"
#define LOOPROISTART 0
#define LOOPROIEND 1
int randomInt(int minVal, int maxVal);
double randomDouble(double minVal, double maxVal);
//std::string generateVarList(int first,...);
std::string generateVarList(int &p1, int &p2);
cv::Size randomSize(int minVal, int maxVal);
cv::Scalar randomScalar(double minVal, double maxVal);
cv::Mat randomMat(cv::Size size, int type, double minVal = 0.0, double maxVal = 255.0);
void showDiff(cv::InputArray gold, cv::InputArray actual, double eps);
//! return true if device supports specified feature and gpu module was built with support the feature.
//bool supportFeature(const cv::gpu::DeviceInfo& info, cv::gpu::FeatureSet feature);
//! return all devices compatible with current gpu module build.
//const std::vector<cv::ocl::DeviceInfo>& devices();
//! return all devices compatible with current gpu module build which support specified feature.
//std::vector<cv::ocl::DeviceInfo> devices(cv::gpu::FeatureSet feature);
//! read image from testdata folder.
cv::Mat readImage(const std::string &fileName, int flags = cv::IMREAD_COLOR);
cv::Mat readImageType(const std::string &fname, int type);
double checkNorm(const cv::Mat &m);
double checkNorm(const cv::Mat &m1, const cv::Mat &m2);
double checkSimilarity(const cv::Mat &m1, const cv::Mat &m2);
#define EXPECT_MAT_NORM(mat, eps) \
{ \
EXPECT_LE(checkNorm(cv::Mat(mat)), eps) \
}
/*#define EXPECT_MAT_NEAR(mat1, mat2, eps) \
{ \
ASSERT_EQ(mat1.type(), mat2.type()); \
ASSERT_EQ(mat1.size(), mat2.size()); \
EXPECT_LE(checkNorm(cv::Mat(mat1), cv::Mat(mat2)), eps); \
}*/
#define EXPECT_MAT_NEAR(mat1, mat2, eps,s) \
{ \
ASSERT_EQ(mat1.type(), mat2.type()); \
ASSERT_EQ(mat1.size(), mat2.size()); \
EXPECT_LE(checkNorm(cv::Mat(mat1), cv::Mat(mat2)), eps)<<s; \
}
#define EXPECT_MAT_SIMILAR(mat1, mat2, eps) \
{ \
ASSERT_EQ(mat1.type(), mat2.type()); \
ASSERT_EQ(mat1.size(), mat2.size()); \
EXPECT_LE(checkSimilarity(cv::Mat(mat1), cv::Mat(mat2)), eps); \
}
namespace cv
{
namespace ocl
{
// void PrintTo(const DeviceInfo& info, std::ostream* os);
}
}
using perf::MatDepth;
using perf::MatType;
//! return vector with types from specified range.
std::vector<MatType> types(int depth_start, int depth_end, int cn_start, int cn_end);
//! return vector with all types (depth: CV_8U-CV_64F, channels: 1-4).
const std::vector<MatType> &all_types();
class Inverse
{
public:
inline Inverse(bool val = false) : val_(val) {}
inline operator bool() const
{
return val_;
}
private:
bool val_;
};
void PrintTo(const Inverse &useRoi, std::ostream *os);
CV_ENUM(CmpCode, cv::CMP_EQ, cv::CMP_GT, cv::CMP_GE, cv::CMP_LT, cv::CMP_LE, cv::CMP_NE)
CV_ENUM(NormCode, cv::NORM_INF, cv::NORM_L1, cv::NORM_L2, cv::NORM_TYPE_MASK, cv::NORM_RELATIVE, cv::NORM_MINMAX)
enum {FLIP_BOTH = 0, FLIP_X = 1, FLIP_Y = -1};
CV_ENUM(FlipCode, FLIP_BOTH, FLIP_X, FLIP_Y)
CV_ENUM(ReduceOp, CV_REDUCE_SUM, CV_REDUCE_AVG, CV_REDUCE_MAX, CV_REDUCE_MIN)
CV_FLAGS(GemmFlags, cv::GEMM_1_T, cv::GEMM_2_T, cv::GEMM_3_T);
CV_ENUM(MorphOp, cv::MORPH_OPEN, cv::MORPH_CLOSE, cv::MORPH_GRADIENT, cv::MORPH_TOPHAT, cv::MORPH_BLACKHAT)
CV_ENUM(ThreshOp, cv::THRESH_BINARY, cv::THRESH_BINARY_INV, cv::THRESH_TRUNC, cv::THRESH_TOZERO, cv::THRESH_TOZERO_INV)
CV_ENUM(Interpolation, cv::INTER_NEAREST, cv::INTER_LINEAR, cv::INTER_CUBIC)
CV_ENUM(Border, cv::BORDER_REFLECT101, cv::BORDER_REPLICATE, cv::BORDER_CONSTANT, cv::BORDER_REFLECT, cv::BORDER_WRAP)
CV_FLAGS(WarpFlags, cv::INTER_NEAREST, cv::INTER_LINEAR, cv::INTER_CUBIC, cv::WARP_INVERSE_MAP)
CV_ENUM(TemplateMethod, cv::TM_SQDIFF, cv::TM_SQDIFF_NORMED, cv::TM_CCORR, cv::TM_CCORR_NORMED, cv::TM_CCOEFF, cv::TM_CCOEFF_NORMED)
CV_FLAGS(DftFlags, cv::DFT_INVERSE, cv::DFT_SCALE, cv::DFT_ROWS, cv::DFT_COMPLEX_OUTPUT, cv::DFT_REAL_OUTPUT)
void run_perf_test();
#define PARAM_TEST_CASE(name, ...) struct name : testing::TestWithParam< std::tr1::tuple< __VA_ARGS__ > >
#define GET_PARAM(k) std::tr1::get< k >(GetParam())
#define ALL_DEVICES testing::ValuesIn(devices())
#define DEVICES(feature) testing::ValuesIn(devices(feature))
#define ALL_TYPES testing::ValuesIn(all_types())
#define TYPES(depth_start, depth_end, cn_start, cn_end) testing::ValuesIn(types(depth_start, depth_end, cn_start, cn_end))
#define DIFFERENT_SIZES testing::Values(cv::Size(128, 128), cv::Size(113, 113))
#define DIRECT_INVERSE testing::Values(Inverse(false), Inverse(true))
#endif // __OPENCV_TEST_UTILITY_HPP__

4397
samples/ocl/performance.cpp
View File

File diff suppressed because it is too large Load Diff
Write Preview
Loading…
Cancel
Save