Skip to content

O
opencv
Commit b43890a7 authored by Roman Donchenko's avatar Roman Donchenko Committed by OpenCV Buildbot
Browse files
Options

Merge pull request #1326 from ilya-lavrenov:perf_ocl

parents d44adcd6 d02ccc95
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 535 additions and 752 deletions
modules/ocl/perf/main.cpp
View file @ b43890a7
......@@ -42,123 +42,54 @@
#include "perf_precomp.hpp"
int main(int argc, const char *argv[])
const char * impls[] =
{
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 }"
"{ c | cpu_ocl | false | use cpu as ocl device}"
"{ 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}";
IMPL_OCL,
IMPL_PLAIN,
#ifdef HAVE_OPENCV_GPU
IMPL_GPU
#endif
};
int main(int argc, char ** argv)
{
const char * keys =
"{ h | help | false | print help message }"
"{ t | type | gpu | set device type:cpu or gpu}"
"{ p | platform | 0 | set platform id }"
"{ d | device | 0 | set device id }";
redirectError(cvErrorCallback);
CommandLineParser cmd(argc, argv, keys);
if (cmd.get<bool>("help"))
{
cout << "Avaible options:" << endl;
cout << "Available options besides google test option:" << endl;
cmd.printParams();
return 0;
}
// get ocl devices
bool use_cpu = cmd.get<bool>("c");
vector<ocl::Info> oclinfo;
int num_devices = 0;
if(use_cpu)
num_devices = getDevice(oclinfo, ocl::CVCL_DEVICE_TYPE_CPU);
else
num_devices = getDevice(oclinfo);
if (num_devices < 1)
{
cerr << "no device found\n";
return -1;
}
// show device info
int devidx = 0;
for (size_t i = 0; i < oclinfo.size(); i++)
{
for (size_t j = 0; j < oclinfo[i].DeviceName.size(); j++)
{
cout << "device " << devidx++ << ": " << oclinfo[i].DeviceName[j] << endl;
}
}
string type = cmd.get<string>("type");
unsigned int pid = cmd.get<unsigned int>("platform");
int device = cmd.get<int>("device");
if (device < 0 || device >= num_devices)
{
cerr << "Invalid device ID" << endl;
return -1;
}
// set this to overwrite binary cache every time the test starts
ocl::setBinaryDiskCache(ocl::CACHE_UPDATE);
if (cmd.get<bool>("verify"))
{
TestSystem::instance().setNumIters(1);
TestSystem::instance().setGPUWarmupIters(0);
TestSystem::instance().setCPUIters(0);
}
int flag = type == "cpu" ? cv::ocl::CVCL_DEVICE_TYPE_CPU :
cv::ocl::CVCL_DEVICE_TYPE_GPU;
devidx = 0;
for (size_t i = 0; i < oclinfo.size(); i++)
std::vector<cv::ocl::Info> oclinfo;
int devnums = cv::ocl::getDevice(oclinfo, flag);
if (devnums <= device || device < 0)
{
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]);
cout << "use " << devidx << ": " <<oclinfo[i].DeviceName[j] << endl;
goto END_DEV;
}
}
}
END_DEV:
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 (!filter.empty())
{
TestSystem::instance().setTestFilter(filter);
}
if (!workdir.empty())
{
if (workdir[workdir.size() - 1] != '/' && workdir[workdir.size() - 1] != '\\')
{
workdir += '/';
}
TestSystem::instance().setWorkingDir(workdir);
std::cout << "device invalid\n";
return -1;
}
if (list)
if (pid >= oclinfo.size())
{
TestSystem::instance().setListMode(true);
std::cout << "platform invalid\n";
return -1;
}
TestSystem::instance().setNumIters(iters);
TestSystem::instance().setGPUWarmupIters(wu_iters);
TestSystem::instance().run();
cv::ocl::setDevice(oclinfo[pid], device);
cv::ocl::setBinaryDiskCache(cv::ocl::CACHE_UPDATE);
return 0;
CV_PERF_TEST_MAIN_INTERNALS(ocl, impls)
}
modules/ocl/perf/perf_arithm.cpp
View file @ b43890a7
This diff is collapsed.
modules/ocl/perf/perf_blend.cpp
View file @ b43890a7
......@@ -45,9 +45,15 @@
//M*/
#include "perf_precomp.hpp"
using namespace perf;
///////////// 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)
static void blendLinearGold(const cv::Mat &img1, const cv::Mat &img2,
const cv::Mat &weights1, const cv::Mat &weights2,
cv::Mat &result_gold)
{
result_gold.create(img1.size(), img1.type());
......@@ -63,60 +69,46 @@ void blendLinearGold(const cv::Mat &img1, const cv::Mat &img2, const cv::Mat &we
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));
int x1 = x * cn;
float w1 = weights1_row[x];
float w2 = weights2_row[x];
result_gold_row[x] = static_cast<T>((img1_row[x1] * w1
+ img2_row[x1] * w2) / (w1 + w2 + 1e-5f));
}
}
}
PERFTEST(blend)
typedef TestBaseWithParam<Size> blendLinearFixture;
PERF_TEST_P(blendLinearFixture, blendLinear, OCL_TYPICAL_MAT_SIZES)
{
Mat src1, src2, weights1, weights2, dst, ocl_dst;
ocl::oclMat d_src1, d_src2, d_weights1, d_weights2, d_dst;
const Size srcSize = GetParam();
const int type = CV_8UC1;
Mat src1(srcSize, type), src2(srcSize, CV_8UC1), dst;
Mat weights1(srcSize, CV_32FC1), weights2(srcSize, CV_32FC1);
int all_type[] = {CV_8UC1, CV_8UC4};
std::string type_name[] = {"CV_8UC1", "CV_8UC4"};
declare.in(src1, src2, WARMUP_RNG);
randu(weights1, 0.0f, 1.0f);
randu(weights2, 0.0f, 1.0f);
for (int size = Min_Size; size <= Max_Size; size *= Multiple)
if (RUN_OCL_IMPL)
{
for (size_t j = 0; j < sizeof(all_type) / sizeof(int); j++)
{
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(ocl_dst);
GPU_FULL_OFF;
TestSystem::instance().ExpectedMatNear(dst, ocl_dst, 1.f);
}
ocl::oclMat oclSrc1(src1), oclSrc2(src2), oclDst;
ocl::oclMat oclWeights1(weights1), oclWeights2(weights2);
TEST_CYCLE() cv::ocl::blendLinear(oclSrc1, oclSrc2, oclWeights1, oclWeights2, oclDst);
oclDst.download(dst);
SANITY_CHECK(dst);
}
else if (RUN_PLAIN_IMPL)
{
TEST_CYCLE() blendLinearGold<uchar>(src1, src2, weights1, weights2, dst);
SANITY_CHECK(dst);
}
else
OCL_PERF_ELSE
}
modules/ocl/perf/perf_brute_force_matcher.cpp
View file @ b43890a7
......@@ -45,123 +45,119 @@
//M*/
#include "perf_precomp.hpp"
//////////////////// BruteForceMatch /////////////////
PERFTEST(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);
vector< vector<DMatch> > d_matches(2);
// Init GPU matcher
ocl::BruteForceMatcher_OCL_base d_matcher(ocl::BruteForceMatcher_OCL_base::L2Dist);
using namespace perf;
ocl::oclMat d_query(query);
ocl::oclMat d_train(train);
#define OCL_BFMATCHER_TYPICAL_MAT_SIZES ::testing::Values(cv::Size(128, 500), cv::Size(128, 1000), cv::Size(128, 2000))
ocl::oclMat d_trainIdx, d_distance, d_allDist, d_nMatches;
SUBTEST << size << "; match";
matcher.match(query, train, matches[0]);
//////////////////// BruteForceMatch /////////////////
CPU_ON;
matcher.match(query, train, matches[0]);
CPU_OFF;
typedef TestBaseWithParam<Size> BruteForceMatcherFixture;
WARMUP_ON;
d_matcher.matchSingle(d_query, d_train, d_trainIdx, d_distance);
WARMUP_OFF;
PERF_TEST_P(BruteForceMatcherFixture, DISABLED_match,
OCL_BFMATCHER_TYPICAL_MAT_SIZES) // TODO too big difference between implementations
{
const Size srcSize = GetParam();
GPU_ON;
d_matcher.matchSingle(d_query, d_train, d_trainIdx, d_distance);
GPU_OFF;
vector<DMatch> matches;
Mat query(srcSize, CV_32F), train(srcSize, CV_32F);
declare.in(query, train).time(srcSize.height == 2000 ? 8 : 4 );
randu(query, 0.0f, 1.0f);
randu(train, 0.0f, 1.0f);
GPU_FULL_ON;
d_query.upload(query);
d_train.upload(train);
d_matcher.match(d_query, d_train, d_matches[0]);
GPU_FULL_OFF;
if (RUN_PLAIN_IMPL)
{
BFMatcher matcher(NORM_L2);
TEST_CYCLE() matcher.match(query, train, matches);
int diff = abs((int)d_matches[0].size() - (int)matches[0].size());
if(diff == 0)
TestSystem::instance().setAccurate(1, 0);
else
TestSystem::instance().setAccurate(0, diff);
SANITY_CHECK_MATCHES(matches);
}
else if (RUN_OCL_IMPL)
{
ocl::BruteForceMatcher_OCL_base oclMatcher(ocl::BruteForceMatcher_OCL_base::L2Dist);
ocl::oclMat oclQuery(query), oclTrain(train);
SUBTEST << size << "; knnMatch";
TEST_CYCLE() oclMatcher.match(oclQuery, oclTrain, matches);
matcher.knnMatch(query, train, matches, 2);
SANITY_CHECK_MATCHES(matches);
}
else
OCL_PERF_ELSE
}
CPU_ON;
matcher.knnMatch(query, train, matches, 2);
CPU_OFF;
PERF_TEST_P(BruteForceMatcherFixture, DISABLED_knnMatch,
OCL_BFMATCHER_TYPICAL_MAT_SIZES) // TODO too many outliers
{
const Size srcSize = GetParam();
WARMUP_ON;
d_matcher.knnMatchSingle(d_query, d_train, d_trainIdx, d_distance, d_allDist, 2);
WARMUP_OFF;
vector<vector<DMatch> > matches(2);
Mat query(srcSize, CV_32F), train(srcSize, CV_32F);
randu(query, 0.0f, 1.0f);
randu(train, 0.0f, 1.0f);
GPU_ON;
d_matcher.knnMatchSingle(d_query, d_train, d_trainIdx, d_distance, d_allDist, 2);
GPU_OFF;
declare.in(query, train);
if (srcSize.height == 2000)
declare.time(8);
GPU_FULL_ON;
d_query.upload(query);
d_train.upload(train);
d_matcher.knnMatch(d_query, d_train, d_matches, 2);
GPU_FULL_OFF;
if (RUN_PLAIN_IMPL)
{
BFMatcher matcher (NORM_L2);
TEST_CYCLE() matcher.knnMatch(query, train, matches, 2);
diff = abs((int)d_matches[0].size() - (int)matches[0].size());
if(diff == 0)
TestSystem::instance().setAccurate(1, 0);
else
TestSystem::instance().setAccurate(0, diff);
std::vector<DMatch> & matches0 = matches[0], & matches1 = matches[1];
SANITY_CHECK_MATCHES(matches0);
SANITY_CHECK_MATCHES(matches1);
}
else if (RUN_OCL_IMPL)
{
ocl::BruteForceMatcher_OCL_base oclMatcher(ocl::BruteForceMatcher_OCL_base::L2Dist);
ocl::oclMat oclQuery(query), oclTrain(train);
SUBTEST << size << "; radiusMatch";
TEST_CYCLE() oclMatcher.knnMatch(oclQuery, oclTrain, matches, 2);
float max_distance = 2.0f;
std::vector<DMatch> & matches0 = matches[0], & matches1 = matches[1];
SANITY_CHECK_MATCHES(matches0);
SANITY_CHECK_MATCHES(matches1);
}
else
OCL_PERF_ELSE
}
matcher.radiusMatch(query, train, matches, max_distance);
PERF_TEST_P(BruteForceMatcherFixture, DISABLED_radiusMatch,
OCL_BFMATCHER_TYPICAL_MAT_SIZES) // TODO too many outliers
{
const Size srcSize = GetParam();
CPU_ON;
matcher.radiusMatch(query, train, matches, max_distance);
CPU_OFF;
const float max_distance = 2.0f;
vector<vector<DMatch> > matches(2);
Mat query(srcSize, CV_32F), train(srcSize, CV_32F);
declare.in(query, train);
Mat trainIdx, distance, allDist;
d_trainIdx.release();
randu(query, 0.0f, 1.0f);
randu(train, 0.0f, 1.0f);
WARMUP_ON;
d_matcher.radiusMatchSingle(d_query, d_train, d_trainIdx, d_distance, d_nMatches, max_distance);
WARMUP_OFF;
if (RUN_PLAIN_IMPL)
{
BFMatcher matcher (NORM_L2);
TEST_CYCLE() matcher.radiusMatch(query, matches, max_distance);
GPU_ON;
d_matcher.radiusMatchSingle(d_query, d_train, d_trainIdx, d_distance, d_nMatches, max_distance);
GPU_OFF;
std::vector<DMatch> & matches0 = matches[0], & matches1 = matches[1];
SANITY_CHECK_MATCHES(matches0);
SANITY_CHECK_MATCHES(matches1);
}
else if (RUN_OCL_IMPL)
{
ocl::oclMat oclQuery(query), oclTrain(train);
ocl::BruteForceMatcher_OCL_base oclMatcher(ocl::BruteForceMatcher_OCL_base::L2Dist);
GPU_FULL_ON;
d_query.upload(query);
d_train.upload(train);
d_matcher.radiusMatch(d_query, d_train, d_matches, max_distance);
GPU_FULL_OFF;
TEST_CYCLE() oclMatcher.radiusMatch(oclQuery, oclTrain, matches, max_distance);
diff = abs((int)d_matches[0].size() - (int)matches[0].size());
if(diff == 0)
TestSystem::instance().setAccurate(1, 0);
else
TestSystem::instance().setAccurate(0, diff);
std::vector<DMatch> & matches0 = matches[0], & matches1 = matches[1];
SANITY_CHECK_MATCHES(matches0);
SANITY_CHECK_MATCHES(matches1);
}
else
OCL_PERF_ELSE
}
#undef OCL_BFMATCHER_TYPICAL_MAT_SIZES
modules/ocl/perf/perf_calib3d.cpp
View file @ b43890a7
......@@ -45,48 +45,44 @@
//M*/
#include "perf_precomp.hpp"
///////////// StereoMatchBM ////////////////////////
PERFTEST(StereoMatchBM)
{
Mat left_image = imread(abspath("aloeL.jpg"), cv::IMREAD_GRAYSCALE);
Mat right_image = imread(abspath("aloeR.jpg"), cv::IMREAD_GRAYSCALE);
Mat disp,dst;
ocl::oclMat d_left, d_right,d_disp;
int n_disp= 128;
int winSize =19;
SUBTEST << left_image.cols << 'x' << left_image.rows << "; aloeL.jpg ;"<< right_image.cols << 'x' << right_image.rows << "; aloeR.jpg ";
PERF_TEST(StereoMatchBMFixture, DISABLED_StereoMatchBM) // TODO doesn't work properly
{
Mat left_image = imread(getDataPath("gpu/stereobm/aloe-L.png"), cv::IMREAD_GRAYSCALE);
Mat right_image = imread(getDataPath("gpu/stereobm/aloe-R.png"), cv::IMREAD_GRAYSCALE);
StereoBM bm(0, n_disp, winSize);
bm(left_image, right_image, dst);
ASSERT_TRUE(!left_image.empty()) << "no input image";
ASSERT_TRUE(!right_image.empty()) << "no input image";
ASSERT_TRUE(right_image.size() == left_image.size());
ASSERT_TRUE(right_image.size() == left_image.size());
CPU_ON;
bm(left_image, right_image, dst);
CPU_OFF;
const int n_disp = 128, winSize = 19;
Mat disp(left_image.size(), CV_16SC1);
d_left.upload(left_image);
d_right.upload(right_image);
declare.in(left_image, right_image).out(disp);
ocl::StereoBM_OCL d_bm(0, n_disp, winSize);
if (RUN_OCL_IMPL)
{
ocl::oclMat oclLeft(left_image), oclRight(right_image),
oclDisp(left_image.size(), CV_16SC1);
ocl::StereoBM_OCL oclBM(0, n_disp, winSize);
WARMUP_ON;
d_bm(d_left, d_right, d_disp);
WARMUP_OFF;
TEST_CYCLE() oclBM(oclLeft, oclRight, oclDisp);
cv::Mat ocl_mat;
d_disp.download(ocl_mat);
ocl_mat.convertTo(ocl_mat, dst.type());
oclDisp.download(disp);
GPU_ON;
d_bm(d_left, d_right, d_disp);
GPU_OFF;
SANITY_CHECK(disp);
}
else if (RUN_PLAIN_IMPL)
{
StereoBM bm(0, n_disp, winSize);
GPU_FULL_ON;
d_left.upload(left_image);
d_right.upload(right_image);
d_bm(d_left, d_right, d_disp);
d_disp.download(disp);
GPU_FULL_OFF;
TEST_CYCLE() bm(left_image, right_image, disp);
TestSystem::instance().setAccurate(-1, 0.);
SANITY_CHECK(disp);
}
else
OCL_PERF_ELSE
}
modules/ocl/perf/perf_canny.cpp
View file @ b43890a7
......@@ -45,41 +45,33 @@
//M*/
#include "perf_precomp.hpp"
///////////// Canny ////////////////////////
PERFTEST(Canny)
{
Mat img = imread(abspath("aloeL.jpg"), CV_LOAD_IMAGE_GRAYSCALE);
if (img.empty())
{
throw runtime_error("can't open aloeL.jpg");
}
SUBTEST << img.cols << 'x' << img.rows << "; aloeL.jpg" << "; edges" << "; CV_8UC1";
using namespace perf;
Mat edges(img.size(), CV_8UC1), ocl_edges;
///////////// Canny ////////////////////////
CPU_ON;
Canny(img, edges, 50.0, 100.0);
CPU_OFF;
PERF_TEST(CannyFixture, DISABLED_Canny) // TODO difference between implmentations
{
Mat img = imread(getDataPath("gpu/stereobm/aloe-L.png"), cv::IMREAD_GRAYSCALE),
edges(img.size(), CV_8UC1);
ASSERT_TRUE(!img.empty()) << "can't open aloe-L.png";
ocl::oclMat d_img(img);
ocl::oclMat d_edges;
ocl::CannyBuf d_buf;
declare.in(img).out(edges);
WARMUP_ON;
ocl::Canny(d_img, d_buf, d_edges, 50.0, 100.0);
WARMUP_OFF;
if (RUN_OCL_IMPL)
{
ocl::oclMat oclImg(img), oclEdges(img.size(), CV_8UC1);
GPU_ON;
ocl::Canny(d_img, d_buf, d_edges, 50.0, 100.0);
GPU_OFF;
TEST_CYCLE() ocl::Canny(oclImg, oclEdges, 50.0, 100.0);
oclEdges.download(edges);
GPU_FULL_ON;
d_img.upload(img);
ocl::Canny(d_img, d_buf, d_edges, 50.0, 100.0);
d_edges.download(ocl_edges);
GPU_FULL_OFF;
SANITY_CHECK(edges);
}
else if (RUN_PLAIN_IMPL)
{
TEST_CYCLE() Canny(img, edges, 50.0, 100.0);
TestSystem::instance().ExceptedMatSimilar(edges, ocl_edges, 2e-2);
SANITY_CHECK(edges);
}
else
OCL_PERF_ELSE
}
modules/ocl/perf/perf_color.cpp
View file @ b43890a7
......@@ -45,49 +45,34 @@
//M*/
#include "perf_precomp.hpp"
///////////// cvtColor////////////////////////
PERFTEST(cvtColor)
{
Mat src, dst, ocl_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;
using namespace perf;
d_src.upload(src);
///////////// cvtColor////////////////////////
WARMUP_ON;
ocl::cvtColor(d_src, d_dst, CV_RGBA2GRAY, 4);
WARMUP_OFF;
typedef TestBaseWithParam<Size> cvtColorFixture;
GPU_ON;
ocl::cvtColor(d_src, d_dst, CV_RGBA2GRAY, 4);
GPU_OFF;
PERF_TEST_P(cvtColorFixture, cvtColor, OCL_TYPICAL_MAT_SIZES)
{
const Size srcSize = GetParam();
GPU_FULL_ON;
d_src.upload(src);
ocl::cvtColor(d_src, d_dst, CV_RGBA2GRAY, 4);
d_dst.download(ocl_dst);
GPU_FULL_OFF;
Mat src(srcSize, CV_8UC4), dst(srcSize, CV_8UC4);
declare.in(src, WARMUP_RNG).out(dst);
TestSystem::instance().ExceptedMatSimilar(dst, ocl_dst, 1e-5);
}
if (RUN_OCL_IMPL)
{
ocl::oclMat oclSrc(src), oclDst(src.size(), CV_8UC4);
TEST_CYCLE() ocl::cvtColor(oclSrc, oclDst, CV_RGBA2GRAY, 4);
oclDst.download(dst);
SANITY_CHECK(dst);
}
else if (RUN_PLAIN_IMPL)
{
TEST_CYCLE() cv::cvtColor(src, dst, CV_RGBA2GRAY, 4);
SANITY_CHECK(dst);
}
else
OCL_PERF_ELSE
}
modules/ocl/perf/perf_fft.cpp
View file @ b43890a7
......@@ -45,47 +45,39 @@
//M*/
#include "perf_precomp.hpp"
///////////// dft ////////////////////////
PERFTEST(dft)
{
Mat src, dst, ocl_dst;
ocl::oclMat d_src, d_dst;
int all_type[] = {CV_32FC2};
std::string type_name[] = {"CV_32FC2"};
using namespace perf;
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";
///////////// dft ////////////////////////
gen(src, size, size, all_type[j], Scalar::all(0), Scalar::all(1));
typedef TestBaseWithParam<Size> dftFixture;
dft(src, dst);
PERF_TEST_P(dftFixture, DISABLED_dft, OCL_TYPICAL_MAT_SIZES) // TODO not implemented
{
const Size srcSize = GetParam();
CPU_ON;
dft(src, dst);
CPU_OFF;
Mat src(srcSize, CV_32FC2), dst;
randu(src, 0.0f, 1.0f);
declare.in(src);
d_src.upload(src);
if (srcSize == OCL_SIZE_4000)
declare.time(7.4);
WARMUP_ON;
ocl::dft(d_src, d_dst, Size(size, size));
WARMUP_OFF;
if (RUN_OCL_IMPL)
{
ocl::oclMat oclSrc(src), oclDst;
GPU_ON;
ocl::dft(d_src, d_dst, Size(size, size));
GPU_OFF;
TEST_CYCLE() cv::ocl::dft(oclSrc, oclDst);
GPU_FULL_ON;
d_src.upload(src);
ocl::dft(d_src, d_dst, Size(size, size));
d_dst.download(ocl_dst);
GPU_FULL_OFF;
oclDst.download(dst);
TestSystem::instance().ExpectedMatNear(dst, ocl_dst, src.size().area() * 1e-4);
}
SANITY_CHECK(dst);
}
else if (RUN_PLAIN_IMPL)
{
TEST_CYCLE() cv::dft(src, dst);
SANITY_CHECK(dst);
}
else
OCL_PERF_ELSE
}
modules/ocl/perf/perf_filters.cpp
View file @ b43890a7
This diff is collapsed.
modules/ocl/perf/perf_gemm.cpp
View file @ b43890a7
......@@ -45,46 +45,40 @@
//M*/
#include "perf_precomp.hpp"
///////////// gemm ////////////////////////
PERFTEST(gemm)
{
Mat src1, src2, src3, dst, ocl_dst;
ocl::oclMat d_src1, d_src2, d_src3, d_dst;
using namespace perf;
for (int size = Min_Size; size <= Max_Size; size *= Multiple)
{
SUBTEST << size << 'x' << size;
///////////// gemm ////////////////////////
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));
typedef TestBaseWithParam<Size> gemmFixture;
gemm(src1, src2, 1.0, src3, 1.0, dst);
PERF_TEST_P(gemmFixture, DISABLED_gemm, OCL_TYPICAL_MAT_SIZES) // TODO not implemented
{
const Size srcSize = GetParam();
CPU_ON;
gemm(src1, src2, 1.0, src3, 1.0, dst);
CPU_OFF;
Mat src1(srcSize, CV_32FC1), src2(srcSize, CV_32FC1),
src3(srcSize, CV_32FC1), dst(srcSize, CV_32FC1);
declare.in(src1, src2, src3).out(dst);
randu(src1, -10.0f, 10.0f);
randu(src2, -10.0f, 10.0f);
randu(src3, -10.0f, 10.0f);
d_src1.upload(src1);
d_src2.upload(src2);
d_src3.upload(src3);
if (RUN_OCL_IMPL)
{
ocl::oclMat oclSrc1(src1), oclSrc2(src2),
oclSrc3(src3), oclDst(srcSize, CV_32FC1);
WARMUP_ON;
ocl::gemm(d_src1, d_src2, 1.0, d_src3, 1.0, d_dst);
WARMUP_OFF;
TEST_CYCLE() cv::ocl::gemm(oclSrc1, oclSrc2, 1.0, oclSrc3, 1.0, oclDst);
GPU_ON;
ocl::gemm(d_src1, d_src2, 1.0, d_src3, 1.0, d_dst);
GPU_OFF;
oclDst.download(dst);
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(ocl_dst);
GPU_FULL_OFF;
SANITY_CHECK(dst);
}
else if (RUN_PLAIN_IMPL)
{
TEST_CYCLE() cv::gemm(src1, src2, 1.0, src3, 1.0, dst);
TestSystem::instance().ExpectedMatNear(ocl_dst, dst, src1.cols * src1.rows * 1e-4);
SANITY_CHECK(dst);
}
else
OCL_PERF_ELSE
}
modules/ocl/perf/perf_gftt.cpp
View file @ b43890a7
......@@ -46,56 +46,50 @@
#include "perf_precomp.hpp"
///////////// GoodFeaturesToTrack ////////////////////////
PERFTEST(GoodFeaturesToTrack)
{
using namespace cv;
int maxCorners = 2000;
double qualityLevel = 0.01;
using namespace perf;
using std::tr1::tuple;
using std::tr1::get;
std::string images[] = { "rubberwhale1.png", "aloeL.jpg" };
std::vector<cv::Point2f> pts_gold, pts_ocl;
for(size_t imgIdx = 0; imgIdx < (sizeof(images)/sizeof(std::string)); ++imgIdx)
{
Mat frame = imread(abspath(images[imgIdx]), IMREAD_GRAYSCALE);
CV_Assert(!frame.empty());
///////////// GoodFeaturesToTrack ////////////////////////
for(float minDistance = 0; minDistance < 4; minDistance += 3.0)
{
SUBTEST << "image = " << images[imgIdx] << "; ";
SUBTEST << "minDistance = " << minDistance << "; ";
typedef tuple<string, double> GoodFeaturesToTrackParams;
typedef TestBaseWithParam<GoodFeaturesToTrackParams> GoodFeaturesToTrackFixture;
cv::goodFeaturesToTrack(frame, pts_gold, maxCorners, qualityLevel, minDistance);
PERF_TEST_P(GoodFeaturesToTrackFixture, GoodFeaturesToTrack,
::testing::Combine(::testing::Values(string("gpu/opticalflow/rubberwhale1.png"),
string("gpu/stereobm/aloe-L.png")),
::testing::Range(0.0, 4.0, 3.0)))
{
CPU_ON;
cv::goodFeaturesToTrack(frame, pts_gold, maxCorners, qualityLevel, minDistance);
CPU_OFF;
const GoodFeaturesToTrackParams param = GetParam();
const string fileName = getDataPath(get<0>(param));
const int maxCorners = 2000;
const double qualityLevel = 0.01, minDistance = get<1>(param);
cv::ocl::GoodFeaturesToTrackDetector_OCL detector(maxCorners, qualityLevel, minDistance);
Mat frame = imread(fileName, IMREAD_GRAYSCALE);
ASSERT_TRUE(!frame.empty()) << "no input image";
ocl::oclMat frame_ocl(frame), pts_oclmat;
vector<Point2f> pts_gold;
declare.in(frame);
WARMUP_ON;
detector(frame_ocl, pts_oclmat);
WARMUP_OFF;
if (RUN_OCL_IMPL)
{
ocl::oclMat oclFrame(frame), pts_oclmat;
ocl::GoodFeaturesToTrackDetector_OCL detector(maxCorners, qualityLevel, minDistance);
detector.downloadPoints(pts_oclmat, pts_ocl);
TEST_CYCLE() detector(oclFrame, pts_oclmat);
double diff = abs(static_cast<float>(pts_gold.size() - pts_ocl.size()));
TestSystem::instance().setAccurate(diff == 0.0, diff);
detector.downloadPoints(pts_oclmat, pts_gold);
GPU_ON;
detector(frame_ocl, pts_oclmat);
GPU_OFF;
SANITY_CHECK(pts_gold);
}
else if (RUN_PLAIN_IMPL)
{
TEST_CYCLE() cv::goodFeaturesToTrack(frame, pts_gold,
maxCorners, qualityLevel, minDistance);
GPU_FULL_ON;
frame_ocl.upload(frame);
detector(frame_ocl, pts_oclmat);
detector.downloadPoints(pts_oclmat, pts_ocl);
GPU_FULL_OFF;
}
SANITY_CHECK(pts_gold);
}
else
OCL_PERF_ELSE
}
modules/ocl/perf/perf_haar.cpp
View file @ b43890a7
......@@ -45,6 +45,8 @@
//M*/
#include "perf_precomp.hpp"
using namespace perf;
///////////// Haar ////////////////////////
namespace cv
{
......@@ -83,61 +85,39 @@ public:
}
}
PERFTEST(Haar)
{
Mat img = imread(abspath("basketball1.png"), CV_LOAD_IMAGE_GRAYSCALE);
if (img.empty())
{
throw runtime_error("can't open basketball1.png");
}
CascadeClassifier faceCascadeCPU;
if (!faceCascadeCPU.load(abspath("haarcascade_frontalface_alt.xml")))
{
throw runtime_error("can't load haarcascade_frontalface_alt.xml");
}
PERF_TEST(HaarFixture, Haar)
{
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;
Mat img = imread(getDataPath("gpu/haarcascade/basketball1.png"), CV_LOAD_IMAGE_GRAYSCALE);
ASSERT_TRUE(!img.empty()) << "can't open basketball1.png";
declare.in(img);
if (RUN_PLAIN_IMPL)
{
CascadeClassifier faceCascade;
ASSERT_TRUE(faceCascade.load(getDataPath("gpu/haarcascade/haarcascade_frontalface_alt.xml")))
<< "can't load haarcascade_frontalface_alt.xml";
vector<Rect> oclfaces;
ocl::CascadeClassifier_GPU faceCascade;
TEST_CYCLE() faceCascade.detectMultiScale(img, faces,
1.1, 2, 0 | CV_HAAR_SCALE_IMAGE, Size(30, 30));
if (!faceCascade.load(abspath("haarcascade_frontalface_alt.xml")))
{
throw runtime_error("can't load haarcascade_frontalface_alt.xml");
SANITY_CHECK(faces, 4 + 1e-4);
}
else if (RUN_OCL_IMPL)
{
ocl::CascadeClassifier_GPU faceCascade;
ocl::oclMat oclImg(img);
ocl::oclMat d_img(img);
ASSERT_TRUE(faceCascade.load(getDataPath("gpu/haarcascade/haarcascade_frontalface_alt.xml")))
<< "can't load haarcascade_frontalface_alt.xml";
WARMUP_ON;
faceCascade.detectMultiScale(d_img, oclfaces,
1.1, 2, 0 | CV_HAAR_SCALE_IMAGE, Size(30, 30));
WARMUP_OFF;
TEST_CYCLE() faceCascade.detectMultiScale(oclImg, faces,
1.1, 2, 0 | CV_HAAR_SCALE_IMAGE, Size(30, 30));
if(faces.size() == oclfaces.size())
TestSystem::instance().setAccurate(1, 0);
SANITY_CHECK(faces, 4 + 1e-4);
}
else
TestSystem::instance().setAccurate(0, abs((int)faces.size() - (int)oclfaces.size()));
faces.clear();
GPU_ON;
faceCascade.detectMultiScale(d_img, oclfaces,
1.1, 2, 0 | CV_HAAR_SCALE_IMAGE, Size(30, 30));
GPU_OFF;
GPU_FULL_ON;
d_img.upload(img);
faceCascade.detectMultiScale(d_img, oclfaces,
1.1, 2, 0 | CV_HAAR_SCALE_IMAGE, Size(30, 30));
GPU_FULL_OFF;
OCL_PERF_ELSE
}
modules/ocl/perf/perf_hog.cpp
View file @ b43890a7
......@@ -45,50 +45,37 @@
//M*/
#include "perf_precomp.hpp"
using namespace perf;
///////////// HOG////////////////////////
PERFTEST(HOG)
PERF_TEST(HOGFixture, HOG)
{
Mat src = imread(abspath("road.png"), cv::IMREAD_GRAYSCALE);
if (src.empty())
{
throw runtime_error("can't open road.png");
}
cv::HOGDescriptor hog;
hog.setSVMDetector(hog.getDefaultPeopleDetector());
std::vector<cv::Rect> found_locations;
std::vector<cv::Rect> d_found_locations;
Mat src = imread(getDataPath("gpu/hog/road.png"), cv::IMREAD_GRAYSCALE);
ASSERT_TRUE(!src.empty()) << "can't open input image road.png";
SUBTEST << src.cols << 'x' << src.rows << "; road.png";
vector<cv::Rect> found_locations;
declare.in(src).time(5);
hog.detectMultiScale(src, found_locations);
if (RUN_PLAIN_IMPL)
{
HOGDescriptor hog;
hog.setSVMDetector(hog.getDefaultPeopleDetector());
CPU_ON;
hog.detectMultiScale(src, found_locations);
CPU_OFF;
TEST_CYCLE() hog.detectMultiScale(src, found_locations);
cv::ocl::HOGDescriptor ocl_hog;
ocl_hog.setSVMDetector(ocl_hog.getDefaultPeopleDetector());
ocl::oclMat d_src;
d_src.upload(src);
SANITY_CHECK(found_locations, 1 + DBL_EPSILON);
}
else if (RUN_OCL_IMPL)
{
ocl::HOGDescriptor ocl_hog;
ocl_hog.setSVMDetector(ocl_hog.getDefaultPeopleDetector());
ocl::oclMat oclSrc(src);
WARMUP_ON;
ocl_hog.detectMultiScale(d_src, d_found_locations);
WARMUP_OFF;
TEST_CYCLE() ocl_hog.detectMultiScale(oclSrc, found_locations);
if(d_found_locations.size() == found_locations.size())
TestSystem::instance().setAccurate(1, 0);
SANITY_CHECK(found_locations, 1 + DBL_EPSILON);
}
else
TestSystem::instance().setAccurate(0, abs((int)found_locations.size() - (int)d_found_locations.size()));
GPU_ON;
ocl_hog.detectMultiScale(d_src, found_locations);
GPU_OFF;
GPU_FULL_ON;
d_src.upload(src);
ocl_hog.detectMultiScale(d_src, found_locations);
GPU_FULL_OFF;
OCL_PERF_ELSE
}
modules/ocl/perf/perf_imgproc.cpp
View file @ b43890a7
This diff is collapsed.
modules/ocl/perf/perf_match_template.cpp
View file @ b43890a7
......@@ -45,101 +45,77 @@
//M*/
#include "perf_precomp.hpp"
/////////// 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);
//}
PERFTEST(matchTemplate)
{
//InitMatchTemplate();
Mat src, templ, dst, ocl_dst;
int templ_size = 5;
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"};
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);
SUBTEST << src.cols << 'x' << src.rows << "; " << type_name[j] << "; templ " << templ_size << 'x' << templ_size << "; CCORR";
using namespace perf;
using std::tr1::tuple;
using std::tr1::get;
gen(templ, templ_size, templ_size, all_type[j], 0, 1);
matchTemplate(src, templ, dst, CV_TM_CCORR);
CPU_ON;
matchTemplate(src, templ, dst, CV_TM_CCORR);
CPU_OFF;
ocl::oclMat d_src(src), d_templ(templ), d_dst;
WARMUP_ON;
ocl::matchTemplate(d_src, d_templ, d_dst, CV_TM_CCORR);
WARMUP_OFF;
/////////// matchTemplate ////////////////////////
GPU_ON;
ocl::matchTemplate(d_src, d_templ, d_dst, CV_TM_CCORR);
GPU_OFF;
typedef Size_MatType CV_TM_CCORRFixture;
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(ocl_dst);
GPU_FULL_OFF;
PERF_TEST_P(CV_TM_CCORRFixture, matchTemplate,
::testing::Combine(::testing::Values(OCL_SIZE_1000, OCL_SIZE_2000),
OCL_PERF_ENUM(CV_32FC1, CV_32FC4)))
{
const Size_MatType_t params = GetParam();
const Size srcSize = get<0>(params), templSize(5, 5);
const int type = get<1>(params);
Mat src(srcSize, type), templ(templSize, type);
const Size dstSize(src.cols - templ.cols + 1, src.rows - templ.rows + 1);
Mat dst(dstSize, CV_32F);
randu(src, 0.0f, 1.0f);
randu(templ, 0.0f, 1.0f);
declare.time(srcSize == OCL_SIZE_2000 ? 20 : 6).in(src, templ).out(dst);
if (RUN_OCL_IMPL)
{
ocl::oclMat oclSrc(src), oclTempl(templ), oclDst(dstSize, CV_32F);
TestSystem::instance().ExpectedMatNear(dst, ocl_dst, templ.rows * templ.cols * 1e-1);
}
}
TEST_CYCLE() cv::ocl::matchTemplate(oclSrc, oclTempl, oclDst, CV_TM_CCORR);
int all_type_8U[] = {CV_8UC1};
std::string type_name_8U[] = {"CV_8UC1"};
oclDst.download(dst);
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";
SANITY_CHECK(dst, 1e-4);
}
else if (RUN_PLAIN_IMPL)
{
TEST_CYCLE() cv::matchTemplate(src, templ, dst, CV_TM_CCORR);
gen(src, size, size, all_type_8U[j], 0, 255);
SANITY_CHECK(dst, 1e-4);
}
else
OCL_PERF_ELSE
}
gen(templ, templ_size, templ_size, all_type_8U[j], 0, 255);
typedef TestBaseWithParam<Size> CV_TM_CCORR_NORMEDFixture;
matchTemplate(src, templ, dst, CV_TM_CCORR_NORMED);
PERF_TEST_P(CV_TM_CCORR_NORMEDFixture, matchTemplate, OCL_TYPICAL_MAT_SIZES)
{
const Size srcSize = GetParam(), templSize(5, 5);
CPU_ON;
matchTemplate(src, templ, dst, CV_TM_CCORR_NORMED);
CPU_OFF;
Mat src(srcSize, CV_8UC1), templ(templSize, CV_8UC1), dst;
const Size dstSize(src.cols - templ.cols + 1, src.rows - templ.rows + 1);
dst.create(dstSize, CV_8UC1);
declare.in(src, templ, WARMUP_RNG).out(dst)
.time(srcSize == OCL_SIZE_2000 ? 10 : srcSize == OCL_SIZE_4000 ? 23 : 2);
ocl::oclMat d_src(src);
ocl::oclMat d_templ(templ), d_dst;
if (RUN_OCL_IMPL)
{
ocl::oclMat oclSrc(src), oclTempl(templ), oclDst(dstSize, CV_8UC1);
WARMUP_ON;
ocl::matchTemplate(d_src, d_templ, d_dst, CV_TM_CCORR_NORMED);
WARMUP_OFF;
TEST_CYCLE() cv::ocl::matchTemplate(oclSrc, oclTempl, oclDst, CV_TM_CCORR_NORMED);
GPU_ON;
ocl::matchTemplate(d_src, d_templ, d_dst, CV_TM_CCORR_NORMED);
GPU_OFF;
oclDst.download(dst);
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(ocl_dst);
GPU_FULL_OFF;
SANITY_CHECK(dst, 2e-2);
}
else if (RUN_PLAIN_IMPL)
{
TEST_CYCLE() cv::matchTemplate(src, templ, dst, CV_TM_CCORR_NORMED);
TestSystem::instance().ExpectedMatNear(dst, ocl_dst, templ.rows * templ.cols * 1e-1);
}
}
SANITY_CHECK(dst, 2e-2);
}
else
OCL_PERF_ELSE
}
modules/ocl/perf/perf_matrix_operation.cpp
View file @ b43890a7
......@@ -45,142 +45,113 @@
//M*/
#include "perf_precomp.hpp"
///////////// ConvertTo////////////////////////
PERFTEST(ConvertTo)
{
Mat src, dst, ocl_dst;
ocl::oclMat d_src, d_dst;
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)
{
for (size_t j = 0; j < sizeof(all_type) / sizeof(int); j++)
{
SUBTEST << size << 'x' << size << "; " << type_name[j] << " to 32FC1";
using namespace perf;
using std::tr1::tuple;
using std::tr1::get;
gen(src, size, size, all_type[j], 0, 256);
//gen(dst, size, size, all_type[j], 0, 256);
//d_dst.upload(dst);
src.convertTo(dst, CV_32FC1);
///////////// ConvertTo////////////////////////
CPU_ON;
src.convertTo(dst, CV_32FC1);
CPU_OFF;
typedef Size_MatType ConvertToFixture;
d_src.upload(src);
PERF_TEST_P(ConvertToFixture, ConvertTo,
::testing::Combine(OCL_TYPICAL_MAT_SIZES,
OCL_PERF_ENUM(CV_8UC1, CV_8UC4)))
{
const Size_MatType_t params = GetParam();
const Size srcSize = get<0>(params);
const int type = get<1>(params);
WARMUP_ON;
d_src.convertTo(d_dst, CV_32FC1);
WARMUP_OFF;
Mat src(srcSize, type), dst;
const int dstType = CV_MAKE_TYPE(CV_32F, src.channels());
dst.create(srcSize, dstType);
declare.in(src, WARMUP_RNG).out(dst);
GPU_ON;
d_src.convertTo(d_dst, CV_32FC1);
GPU_OFF;
if (RUN_OCL_IMPL)
{
ocl::oclMat oclSrc(src), oclDst(srcSize, dstType);
GPU_FULL_ON;
d_src.upload(src);
d_src.convertTo(d_dst, CV_32FC1);
d_dst.download(ocl_dst);
GPU_FULL_OFF;
TEST_CYCLE() oclSrc.convertTo(oclDst, dstType);
TestSystem::instance().ExpectedMatNear(dst, ocl_dst, 0.0);
}
oclDst.download(dst);
SANITY_CHECK(dst);
}
}
///////////// copyTo////////////////////////
PERFTEST(copyTo)
{
Mat src, dst, ocl_dst;
ocl::oclMat d_src, d_dst;
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)
else if (RUN_PLAIN_IMPL)
{
for (size_t j = 0; j < sizeof(all_type) / sizeof(int); j++)
{
SUBTEST << size << 'x' << size << "; " << type_name[j] ;
TEST_CYCLE() src.convertTo(dst, dstType);
gen(src, size, size, all_type[j], 0, 256);
//gen(dst, size, size, all_type[j], 0, 256);
//d_dst.upload(dst);
SANITY_CHECK(dst);
}
else
OCL_PERF_ELSE
}
src.copyTo(dst);
///////////// copyTo////////////////////////
CPU_ON;
src.copyTo(dst);
CPU_OFF;
typedef Size_MatType copyToFixture;
d_src.upload(src);
PERF_TEST_P(copyToFixture, copyTo,
::testing::Combine(OCL_TYPICAL_MAT_SIZES,
OCL_PERF_ENUM(CV_8UC1, CV_8UC4)))
{
const Size_MatType_t params = GetParam();
const Size srcSize = get<0>(params);
const int type = get<1>(params);
WARMUP_ON;
d_src.copyTo(d_dst);
WARMUP_OFF;
Mat src(srcSize, type), dst(srcSize, type);
declare.in(src, WARMUP_RNG).out(dst);
GPU_ON;
d_src.copyTo(d_dst);
GPU_OFF;
if (RUN_OCL_IMPL)
{
ocl::oclMat oclSrc(src), oclDst(srcSize, type);
GPU_FULL_ON;
d_src.upload(src);
d_src.copyTo(d_dst);
d_dst.download(ocl_dst);
GPU_FULL_OFF;
TEST_CYCLE() oclSrc.copyTo(oclDst);
TestSystem::instance().ExpectedMatNear(dst, ocl_dst, 0.0);
}
oclDst.download(dst);
SANITY_CHECK(dst);
}
}
///////////// setTo////////////////////////
PERFTEST(setTo)
{
Mat src, ocl_src;
Scalar val(1, 2, 3, 4);
ocl::oclMat d_src;
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)
else if (RUN_PLAIN_IMPL)
{
for (size_t j = 0; j < sizeof(all_type) / sizeof(int); j++)
{
SUBTEST << size << 'x' << size << "; " << type_name[j] ;
TEST_CYCLE() src.copyTo(dst);
gen(src, size, size, all_type[j], 0, 256);
SANITY_CHECK(dst);
}
else
OCL_PERF_ELSE
}
src.setTo(val);
///////////// setTo////////////////////////
CPU_ON;
src.setTo(val);
CPU_OFF;
typedef Size_MatType setToFixture;
d_src.upload(src);
PERF_TEST_P(setToFixture, setTo,
::testing::Combine(OCL_TYPICAL_MAT_SIZES,
OCL_PERF_ENUM(CV_8UC1, CV_8UC4)))
{
const Size_MatType_t params = GetParam();
const Size srcSize = get<0>(params);
const int type = get<1>(params);
const Scalar val(1, 2, 3, 4);
WARMUP_ON;
d_src.setTo(val);
WARMUP_OFF;
Mat src(srcSize, type);
declare.in(src);
d_src.download(ocl_src);
TestSystem::instance().ExpectedMatNear(src, ocl_src, 1.0);
if (RUN_OCL_IMPL)
{
ocl::oclMat oclSrc(srcSize, type);
GPU_ON;;
d_src.setTo(val);
GPU_OFF;
TEST_CYCLE() oclSrc.setTo(val);
oclSrc.download(src);
GPU_FULL_ON;
d_src.upload(src);
d_src.setTo(val);
GPU_FULL_OFF;
}
SANITY_CHECK(src);
}
else if (RUN_PLAIN_IMPL)
{
TEST_CYCLE() src.setTo(val);
SANITY_CHECK(src);
}
else
OCL_PERF_ELSE
}
modules/ocl/perf/perf_moments.cpp
View file @ b43890a7
......@@ -43,50 +43,47 @@
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "perf_precomp.hpp"
///////////// Moments ////////////////////////
PERFTEST(Moments)
{
Mat src;
bool binaryImage = 0;
int all_type[] = {CV_8UC1, CV_16SC1, CV_32FC1, CV_64FC1};
std::string type_name[] = {"CV_8UC1", "CV_16SC1", "CV_32FC1", "CV_64FC1"};
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];
#include "perf_precomp.hpp"
gen(src, size, size, all_type[j], 0, 256);
using namespace perf;
using std::tr1::tuple;
using std::tr1::get;
cv::Moments CvMom = moments(src, binaryImage);
///////////// Moments ////////////////////////
CPU_ON;
moments(src, binaryImage);
CPU_OFF;
typedef Size_MatType MomentsFixture;
cv::Moments oclMom;
WARMUP_ON;
oclMom = ocl::ocl_moments(src, binaryImage);
WARMUP_OFF;
PERF_TEST_P(MomentsFixture, DISABLED_Moments,
::testing::Combine(OCL_TYPICAL_MAT_SIZES,
OCL_PERF_ENUM(CV_8UC1, CV_16SC1, CV_32FC1, CV_64FC1))) // TODO does not work properly (see below)
{
const Size_MatType_t params = GetParam();
const Size srcSize = get<0>(params);
const int type = get<1>(params);
Mat gpu_dst, cpu_dst;
HuMoments(CvMom, cpu_dst);
HuMoments(oclMom, gpu_dst);
Mat src(srcSize, type), dst(7, 1, CV_64F);
const bool binaryImage = false;
cv::Moments mom;
GPU_ON;
ocl::ocl_moments(src, binaryImage);
GPU_OFF;
declare.in(src, WARMUP_RNG).out(dst);
GPU_FULL_ON;
ocl::ocl_moments(src, binaryImage);
GPU_FULL_OFF;
if (RUN_OCL_IMPL)
{
ocl::oclMat oclSrc(src);
TestSystem::instance().ExpectedMatNear(gpu_dst, cpu_dst, .5);
TEST_CYCLE() mom = cv::ocl::ocl_moments(oclSrc, binaryImage); // TODO Use oclSrc
cv::HuMoments(mom, dst);
}
SANITY_CHECK(dst);
}
else if (RUN_PLAIN_IMPL)
{
TEST_CYCLE() mom = cv::moments(src, binaryImage);
cv::HuMoments(mom, dst);
SANITY_CHECK(dst);
}
else
OCL_PERF_ELSE
}
modules/ocl/perf/perf_norm.cpp
View file @ b43890a7
This diff is collapsed.
modules/ocl/perf/perf_opticalflow.cpp
View file @ b43890a7
This diff is collapsed.
modules/ocl/perf/perf_precomp.cpp
View file @ b43890a7
This diff is collapsed.
modules/ocl/perf/perf_precomp.hpp
View file @ b43890a7
This diff is collapsed.
modules/ocl/perf/perf_pyramid.cpp
View file @ b43890a7
This diff is collapsed.
modules/ocl/perf/perf_split_merge.cpp
View file @ b43890a7
This diff is collapsed.
modules/ocl/src/arithm.cpp
View file @ b43890a7
......@@ -2336,7 +2336,7 @@ void cv::ocl::pow(const oclMat &x, double p, oclMat &y)
return;
}
CV_Assert((x.type() == y.type() && x.size() == y.size() && x.depth() == CV_32F) || x.depth() == CV_64F);
CV_Assert(x.depth() == CV_32F || x.depth() == CV_64F);
y.create(x.size(), x.type());
string kernelName = "arithm_pow";
......
modules/ocl/test/test_norm.cpp 0 → 100644
View file @ b43890a7
This diff is collapsed.
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment