generic-library/opencv
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

update gpu perf tests

Browse Source
This commit is contained in:
Vladislav Vinogradov 2011-12-28 12:53:08 +00:00
parent eccfc90b77
commit 457b8d7bff
12 changed files with 1635 additions and 1427 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

905
modules/gpu/perf/perf_arithm.cpp
View File

File diff suppressed because it is too large Load Diff

181
modules/gpu/perf/perf_calib3d.cpp
View File

@ -1,93 +1,101 @@
#include "perf_precomp.hpp" #include "perf_precomp.hpp"
PERF_TEST_P(DevInfo, transformPoints, testing::ValuesIn(devices())) #ifdef HAVE_CUDA
//////////////////////////////////////////////////////////////////////
// TransformPoints
GPU_PERF_TEST_1(TransformPoints, cv::gpu::DeviceInfo)
{ {
DeviceInfo devInfo = GetParam(); cv::gpu::DeviceInfo devInfo = GetParam();
setDevice(devInfo.deviceID()); cv::gpu::setDevice(devInfo.deviceID());
Mat src_host(1, 10000, CV_32FC3); cv::Mat src_host(1, 10000, CV_32FC3);
declare.in(src_host, WARMUP_RNG); declare.in(src_host, WARMUP_RNG);
GpuMat src(src_host); cv::gpu::GpuMat src(src_host);
GpuMat dst; cv::gpu::GpuMat dst;
TEST_CYCLE(100) TEST_CYCLE(100)
{ {
transformPoints(src, Mat::ones(1, 3, CV_32FC1), Mat::ones(1, 3, CV_32FC1), dst); cv::gpu::transformPoints(src, cv::Mat::ones(1, 3, CV_32FC1), cv::Mat::ones(1, 3, CV_32FC1), dst);
}
} }
Mat dst_host(dst); INSTANTIATE_TEST_CASE_P(Calib3D, TransformPoints, ALL_DEVICES);
SANITY_CHECK(dst_host); //////////////////////////////////////////////////////////////////////
} // ProjectPoints
PERF_TEST_P(DevInfo, projectPoints, testing::ValuesIn(devices())) GPU_PERF_TEST_1(ProjectPoints, cv::gpu::DeviceInfo)
{ {
DeviceInfo devInfo = GetParam(); cv::gpu::DeviceInfo devInfo = GetParam();
setDevice(devInfo.deviceID()); cv::gpu::setDevice(devInfo.deviceID());
Mat src_host(1, 10000, CV_32FC3); cv::Mat src_host(1, 10000, CV_32FC3);
declare.in(src_host, WARMUP_RNG); declare.in(src_host, WARMUP_RNG);
GpuMat src(src_host); cv::gpu::GpuMat src(src_host);
GpuMat dst; cv::gpu::GpuMat dst;
TEST_CYCLE(100) TEST_CYCLE(100)
{ {
projectPoints(src, Mat::ones(1, 3, CV_32FC1), Mat::ones(1, 3, CV_32FC1), Mat::ones(3, 3, CV_32FC1), Mat(), dst); cv::gpu::projectPoints(src, cv::Mat::ones(1, 3, CV_32FC1), cv::Mat::ones(1, 3, CV_32FC1), cv::Mat::ones(3, 3, CV_32FC1), cv::Mat(), dst);
}
} }
Mat dst_host(dst); INSTANTIATE_TEST_CASE_P(Calib3D, ProjectPoints, ALL_DEVICES);
SANITY_CHECK(dst_host); //////////////////////////////////////////////////////////////////////
} // SolvePnPRansac
PERF_TEST_P(DevInfo, solvePnPRansac, testing::ValuesIn(devices())) GPU_PERF_TEST_1(SolvePnPRansac, cv::gpu::DeviceInfo)
{ {
DeviceInfo devInfo = GetParam(); cv::gpu::DeviceInfo devInfo = GetParam();
setDevice(devInfo.deviceID()); cv::gpu::setDevice(devInfo.deviceID());
Mat object(1, 10000, CV_32FC3); cv::Mat object(1, 10000, CV_32FC3);
Mat image(1, 10000, CV_32FC2); cv::Mat image(1, 10000, CV_32FC2);
declare.in(object, image, WARMUP_RNG); declare.in(object, image, WARMUP_RNG);
Mat rvec, tvec; cv::Mat rvec, tvec;
declare.time(3.0); declare.time(3.0);
TEST_CYCLE(100) TEST_CYCLE(100)
{ {
solvePnPRansac(object, image, Mat::ones(3, 3, CV_32FC1), Mat(1, 8, CV_32F, Scalar::all(0)), rvec, tvec); cv::gpu::solvePnPRansac(object, image, cv::Mat::ones(3, 3, CV_32FC1), cv::Mat(1, 8, CV_32F, cv::Scalar::all(0)), rvec, tvec);
}
} }
SANITY_CHECK(rvec); INSTANTIATE_TEST_CASE_P(Calib3D, SolvePnPRansac, ALL_DEVICES);
SANITY_CHECK(tvec);
}
PERF_TEST_P(DevInfo, StereoBM, testing::ValuesIn(devices())) //////////////////////////////////////////////////////////////////////
// StereoBM
GPU_PERF_TEST_1(StereoBM, cv::gpu::DeviceInfo)
{ {
DeviceInfo devInfo = GetParam(); cv::gpu::DeviceInfo devInfo = GetParam();
setDevice(devInfo.deviceID()); cv::gpu::setDevice(devInfo.deviceID());
Mat img_l_host = readImage("gpu/perf/aloe.jpg", CV_LOAD_IMAGE_GRAYSCALE); cv::Mat img_l_host = readImage("gpu/perf/aloe.jpg", cv::IMREAD_GRAYSCALE);
Mat img_r_host = readImage("gpu/perf/aloeR.jpg", CV_LOAD_IMAGE_GRAYSCALE); cv::Mat img_r_host = readImage("gpu/perf/aloeR.jpg", cv::IMREAD_GRAYSCALE);
ASSERT_FALSE(img_l_host.empty()); ASSERT_FALSE(img_l_host.empty());
ASSERT_FALSE(img_r_host.empty()); ASSERT_FALSE(img_r_host.empty());
GpuMat img_l(img_l_host); cv::gpu::GpuMat img_l(img_l_host);
GpuMat img_r(img_r_host); cv::gpu::GpuMat img_r(img_r_host);
cv::gpu::GpuMat dst;
GpuMat dst; cv::gpu::StereoBM_GPU bm(0, 256);
StereoBM_GPU bm(0, 256);
declare.time(5.0); declare.time(5.0);
@ -95,30 +103,30 @@ PERF_TEST_P(DevInfo, StereoBM, testing::ValuesIn(devices()))
{ {
bm(img_l, img_r, dst); bm(img_l, img_r, dst);
} }
Mat dst_host(dst);
SANITY_CHECK(dst_host);
} }
PERF_TEST_P(DevInfo, StereoBeliefPropagation, testing::ValuesIn(devices())) INSTANTIATE_TEST_CASE_P(Calib3D, StereoBM, ALL_DEVICES);
//////////////////////////////////////////////////////////////////////
// StereoBeliefPropagation
GPU_PERF_TEST_1(StereoBeliefPropagation, cv::gpu::DeviceInfo)
{ {
DeviceInfo devInfo = GetParam(); cv::gpu::DeviceInfo devInfo = GetParam();
setDevice(devInfo.deviceID()); cv::gpu::setDevice(devInfo.deviceID());
Mat img_l_host = readImage("gpu/stereobp/aloe-L.png", CV_LOAD_IMAGE_GRAYSCALE); cv::Mat img_l_host = readImage("gpu/stereobp/aloe-L.png");
Mat img_r_host = readImage("gpu/stereobp/aloe-R.png", CV_LOAD_IMAGE_GRAYSCALE); cv::Mat img_r_host = readImage("gpu/stereobp/aloe-R.png");
ASSERT_FALSE(img_l_host.empty()); ASSERT_FALSE(img_l_host.empty());
ASSERT_FALSE(img_r_host.empty()); ASSERT_FALSE(img_r_host.empty());
GpuMat img_l(img_l_host); cv::gpu::GpuMat img_l(img_l_host);
GpuMat img_r(img_r_host); cv::gpu::GpuMat img_r(img_r_host);
cv::gpu::GpuMat dst;
GpuMat dst; cv::gpu::StereoBeliefPropagation bp(64);
StereoBeliefPropagation bp(128);
declare.time(10.0); declare.time(10.0);
@ -126,30 +134,30 @@ PERF_TEST_P(DevInfo, StereoBeliefPropagation, testing::ValuesIn(devices()))
{ {
bp(img_l, img_r, dst); bp(img_l, img_r, dst);
} }
Mat dst_host(dst);
SANITY_CHECK(dst_host);
} }
PERF_TEST_P(DevInfo, StereoConstantSpaceBP, testing::ValuesIn(devices())) INSTANTIATE_TEST_CASE_P(Calib3D, StereoBeliefPropagation, ALL_DEVICES);
//////////////////////////////////////////////////////////////////////
// StereoConstantSpaceBP
GPU_PERF_TEST_1(StereoConstantSpaceBP, cv::gpu::DeviceInfo)
{ {
DeviceInfo devInfo = GetParam(); cv::gpu::DeviceInfo devInfo = GetParam();
setDevice(devInfo.deviceID()); cv::gpu::setDevice(devInfo.deviceID());
Mat img_l_host = readImage("gpu/stereocsbp/aloe.jpg", CV_LOAD_IMAGE_GRAYSCALE); cv::Mat img_l_host = readImage("gpu/stereobm/aloe-L.png", cv::IMREAD_GRAYSCALE);
Mat img_r_host = readImage("gpu/stereocsbp/aloeR.jpg", CV_LOAD_IMAGE_GRAYSCALE); cv::Mat img_r_host = readImage("gpu/stereobm/aloe-R.png", cv::IMREAD_GRAYSCALE);
ASSERT_FALSE(img_l_host.empty()); ASSERT_FALSE(img_l_host.empty());
ASSERT_FALSE(img_r_host.empty()); ASSERT_FALSE(img_r_host.empty());
GpuMat img_l(img_l_host); cv::gpu::GpuMat img_l(img_l_host);
GpuMat img_r(img_r_host); cv::gpu::GpuMat img_r(img_r_host);
cv::gpu::GpuMat dst;
GpuMat dst; cv::gpu::StereoConstantSpaceBP bp(128);
StereoConstantSpaceBP bp(128);
declare.time(10.0); declare.time(10.0);
@ -157,37 +165,38 @@ PERF_TEST_P(DevInfo, StereoConstantSpaceBP, testing::ValuesIn(devices()))
{ {
bp(img_l, img_r, dst); bp(img_l, img_r, dst);
} }
Mat dst_host(dst);
SANITY_CHECK(dst_host);
} }
PERF_TEST_P(DevInfo, DisparityBilateralFilter, testing::ValuesIn(devices())) INSTANTIATE_TEST_CASE_P(Calib3D, StereoConstantSpaceBP, ALL_DEVICES);
//////////////////////////////////////////////////////////////////////
// DisparityBilateralFilter
GPU_PERF_TEST_1(DisparityBilateralFilter, cv::gpu::DeviceInfo)
{ {
DeviceInfo devInfo = GetParam(); cv::gpu::DeviceInfo devInfo = GetParam();
setDevice(devInfo.deviceID()); cv::gpu::setDevice(devInfo.deviceID());
Mat img_host = readImage("gpu/stereobm/aloe-L.png", CV_LOAD_IMAGE_GRAYSCALE); cv::Mat img_host = readImage("gpu/stereobm/aloe-L.png", cv::IMREAD_GRAYSCALE);
Mat disp_host = readImage("gpu/stereobm/aloe-disp.png", CV_LOAD_IMAGE_GRAYSCALE); cv::Mat disp_host = readImage("gpu/stereobm/aloe-disp.png", cv::IMREAD_GRAYSCALE);
ASSERT_FALSE(img_host.empty()); ASSERT_FALSE(img_host.empty());
ASSERT_FALSE(disp_host.empty()); ASSERT_FALSE(disp_host.empty());
GpuMat img(img_host); cv::gpu::GpuMat img(img_host);
GpuMat disp(disp_host); cv::gpu::GpuMat disp(disp_host);
cv::gpu::GpuMat dst;
GpuMat dst; cv::gpu::DisparityBilateralFilter f(128);
DisparityBilateralFilter f(128);
TEST_CYCLE(100) TEST_CYCLE(100)
{ {
f(disp, img, dst); f(disp, img, dst);
} }
Mat dst_host(dst);
SANITY_CHECK(dst_host);
} }
INSTANTIATE_TEST_CASE_P(Calib3D, DisparityBilateralFilter, ALL_DEVICES);
#endif

182
modules/gpu/perf/perf_features2d.cpp
View File

@ -1,23 +1,27 @@
#include "perf_precomp.hpp" #include "perf_precomp.hpp"
PERF_TEST_P(DevInfo_DescSize, BruteForceMatcher_match, testing::Combine(testing::ValuesIn(devices()), #ifdef HAVE_CUDA
testing::Values(64, 128, 256)))
//////////////////////////////////////////////////////////////////////
// BruteForceMatcher_match
GPU_PERF_TEST(BruteForceMatcher_match, cv::gpu::DeviceInfo, int)
{ {
DeviceInfo devInfo = std::tr1::get<0>(GetParam()); cv::gpu::DeviceInfo devInfo = GET_PARAM(0);
int desc_size = std::tr1::get<1>(GetParam()); int desc_size = GET_PARAM(1);
setDevice(devInfo.deviceID()); cv::gpu::setDevice(devInfo.deviceID());
Mat query_host(3000, desc_size, CV_32FC1); cv::Mat query_host(3000, desc_size, CV_32FC1);
Mat train_host(3000, desc_size, CV_32FC1); cv::Mat train_host(3000, desc_size, CV_32FC1);
declare.in(query_host, train_host, WARMUP_RNG); declare.in(query_host, train_host, WARMUP_RNG);
GpuMat query(query_host); cv::gpu::GpuMat query(query_host);
GpuMat train(train_host); cv::gpu::GpuMat train(train_host);
GpuMat trainIdx, distance; cv::gpu::GpuMat trainIdx, distance;
BruteForceMatcher_GPU< L2<float> > matcher; cv::gpu::BruteForceMatcher_GPU< cv::L2<float> > matcher;
declare.time(3.0); declare.time(3.0);
@ -25,34 +29,33 @@ PERF_TEST_P(DevInfo_DescSize, BruteForceMatcher_match, testing::Combine(testing:
{ {
matcher.matchSingle(query, train, trainIdx, distance); matcher.matchSingle(query, train, trainIdx, distance);
} }
Mat trainIdx_host(trainIdx);
Mat distance_host(distance);
SANITY_CHECK(trainIdx_host);
SANITY_CHECK(distance_host);
} }
PERF_TEST_P(DevInfo_K_DescSize, BruteForceMatcher_knnMatch, testing::Combine(testing::ValuesIn(devices()), INSTANTIATE_TEST_CASE_P(Features2D, BruteForceMatcher_match, testing::Combine(
testing::Values(2, 3), ALL_DEVICES,
testing::Values(64, 128, 256))) testing::Values(64, 128, 256)));
//////////////////////////////////////////////////////////////////////
// BruteForceMatcher_knnMatch
GPU_PERF_TEST(BruteForceMatcher_knnMatch, cv::gpu::DeviceInfo, int, int)
{ {
DeviceInfo devInfo = std::tr1::get<0>(GetParam()); cv::gpu::DeviceInfo devInfo = GET_PARAM(0);
int k = std::tr1::get<1>(GetParam()); int desc_size = GET_PARAM(1);
int desc_size = std::tr1::get<2>(GetParam()); int k = GET_PARAM(2);
setDevice(devInfo.deviceID()); cv::gpu::setDevice(devInfo.deviceID());
Mat query_host(3000, desc_size, CV_32FC1); cv::Mat query_host(3000, desc_size, CV_32FC1);
Mat train_host(3000, desc_size, CV_32FC1); cv::Mat train_host(3000, desc_size, CV_32FC1);
declare.in(query_host, train_host, WARMUP_RNG); declare.in(query_host, train_host, WARMUP_RNG);
GpuMat query(query_host); cv::gpu::GpuMat query(query_host);
GpuMat train(train_host); cv::gpu::GpuMat train(train_host);
GpuMat trainIdx, distance, allDist; cv::gpu::GpuMat trainIdx, distance, allDist;
BruteForceMatcher_GPU< L2<float> > matcher; cv::gpu::BruteForceMatcher_GPU< cv::L2<float> > matcher;
declare.time(3.0); declare.time(3.0);
@ -60,30 +63,34 @@ PERF_TEST_P(DevInfo_K_DescSize, BruteForceMatcher_knnMatch, testing::Combine(tes
{ {
matcher.knnMatchSingle(query, train, trainIdx, distance, allDist, k); matcher.knnMatchSingle(query, train, trainIdx, distance, allDist, k);
} }
Mat trainIdx_host(trainIdx);
Mat distance_host(distance);
SANITY_CHECK(trainIdx_host);
SANITY_CHECK(distance_host);
} }
PERF_TEST_P(DevInfo_DescSize, BruteForceMatcher_radiusMatch, testing::Combine(testing::ValuesIn(devices(SHARED_ATOMICS)), INSTANTIATE_TEST_CASE_P(Features2D, BruteForceMatcher_knnMatch, testing::Combine(
testing::Values(64, 128, 256))) ALL_DEVICES,
testing::Values(64, 128, 256),
testing::Values(2, 3)));
//////////////////////////////////////////////////////////////////////
// BruteForceMatcher_radiusMatch
GPU_PERF_TEST(BruteForceMatcher_radiusMatch, cv::gpu::DeviceInfo, int)
{ {
DeviceInfo devInfo = std::tr1::get<0>(GetParam()); cv::gpu::DeviceInfo devInfo = GET_PARAM(0);
int desc_size = std::tr1::get<1>(GetParam()); int desc_size = GET_PARAM(1);
setDevice(devInfo.deviceID()); cv::gpu::setDevice(devInfo.deviceID());
Mat query_host = cvtest::randomMat(theRNG(), Size(desc_size, 3000), CV_32FC1, 0, 1, false); cv::Mat query_host(3000, desc_size, CV_32FC1);
Mat train_host = cvtest::randomMat(theRNG(), Size(desc_size, 3000), CV_32FC1, 0, 1, false); cv::Mat train_host(3000, desc_size, CV_32FC1);
GpuMat query(query_host); fill(query_host, 0, 1);
GpuMat train(train_host); fill(train_host, 0, 1);
GpuMat trainIdx, nMatches, distance;
BruteForceMatcher_GPU< L2<float> > matcher; cv::gpu::GpuMat query(query_host);
cv::gpu::GpuMat train(train_host);
cv::gpu::GpuMat trainIdx, nMatches, distance;
cv::gpu::BruteForceMatcher_GPU< cv::L2<float> > matcher;
declare.time(3.0); declare.time(3.0);
@ -91,81 +98,90 @@ PERF_TEST_P(DevInfo_DescSize, BruteForceMatcher_radiusMatch, testing::Combine(te
{ {
matcher.radiusMatchSingle(query, train, trainIdx, distance, nMatches, 2.0); matcher.radiusMatchSingle(query, train, trainIdx, distance, nMatches, 2.0);
} }
Mat trainIdx_host(trainIdx);
Mat nMatches_host(nMatches);
Mat distance_host(distance);
SANITY_CHECK(trainIdx_host);
SANITY_CHECK(nMatches_host);
SANITY_CHECK(distance_host);
} }
PERF_TEST_P(DevInfo, SURF, testing::ValuesIn(devices())) INSTANTIATE_TEST_CASE_P(Features2D, BruteForceMatcher_radiusMatch, testing::Combine(
ALL_DEVICES,
testing::Values(64, 128, 256)));
//////////////////////////////////////////////////////////////////////
// SURF
GPU_PERF_TEST_1(SURF, cv::gpu::DeviceInfo)
{ {
DeviceInfo devInfo = GetParam(); cv::gpu::DeviceInfo devInfo = GetParam();
setDevice(devInfo.deviceID()); cv::gpu::setDevice(devInfo.deviceID());
Mat img_host = readImage("gpu/perf/aloe.jpg", CV_LOAD_IMAGE_GRAYSCALE); cv::Mat img_host = readImage("gpu/perf/aloe.jpg", cv::IMREAD_GRAYSCALE);
ASSERT_FALSE(img_host.empty()); ASSERT_FALSE(img_host.empty());
GpuMat img(img_host); cv::gpu::GpuMat img(img_host);
GpuMat keypoints, descriptors; cv::gpu::GpuMat keypoints, descriptors;
SURF_GPU surf; cv::gpu::SURF_GPU surf;
declare.time(2.0); declare.time(2.0);
TEST_CYCLE(100) TEST_CYCLE(100)
{ {
surf(img, GpuMat(), keypoints, descriptors); surf(img, cv::gpu::GpuMat(), keypoints, descriptors);
} }
} }
PERF_TEST_P(DevInfo, FAST, testing::ValuesIn(devices())) INSTANTIATE_TEST_CASE_P(Features2D, SURF, DEVICES(cv::gpu::GLOBAL_ATOMICS));
//////////////////////////////////////////////////////////////////////
// FAST
GPU_PERF_TEST_1(FAST, cv::gpu::DeviceInfo)
{ {
DeviceInfo devInfo = GetParam(); cv::gpu::DeviceInfo devInfo = GetParam();
setDevice(devInfo.deviceID()); cv::gpu::setDevice(devInfo.deviceID());
Mat img_host = readImage("gpu/perf/aloe.jpg", CV_LOAD_IMAGE_GRAYSCALE); cv::Mat img_host = readImage("gpu/perf/aloe.jpg", cv::IMREAD_GRAYSCALE);
ASSERT_FALSE(img_host.empty()); ASSERT_FALSE(img_host.empty());
GpuMat img(img_host); cv::gpu::GpuMat img(img_host);
GpuMat keypoints; cv::gpu::GpuMat keypoints, descriptors;
FAST_GPU fastGPU(20); cv::gpu::FAST_GPU fastGPU(20);
declare.time(2.0);
TEST_CYCLE(100) TEST_CYCLE(100)
{ {
fastGPU(img, GpuMat(), keypoints); fastGPU(img, cv::gpu::GpuMat(), keypoints);
} }
} }
PERF_TEST_P(DevInfo, ORB, testing::ValuesIn(devices())) INSTANTIATE_TEST_CASE_P(Features2D, FAST, DEVICES(cv::gpu::GLOBAL_ATOMICS));
//////////////////////////////////////////////////////////////////////
// ORB
GPU_PERF_TEST_1(ORB, cv::gpu::DeviceInfo)
{ {
DeviceInfo devInfo = GetParam(); cv::gpu::DeviceInfo devInfo = GetParam();
setDevice(devInfo.deviceID()); cv::gpu::setDevice(devInfo.deviceID());
Mat img_host = readImage("gpu/perf/aloe.jpg", CV_LOAD_IMAGE_GRAYSCALE); cv::Mat img_host = readImage("gpu/perf/aloe.jpg", cv::IMREAD_GRAYSCALE);
ASSERT_FALSE(img_host.empty()); ASSERT_FALSE(img_host.empty());
GpuMat img(img_host); cv::gpu::GpuMat img(img_host);
GpuMat keypoints, descriptors; cv::gpu::GpuMat keypoints, descriptors;
ORB_GPU orbGPU(4000); cv::gpu::ORB_GPU orbGPU(4000);
declare.time(2.0);
TEST_CYCLE(100) TEST_CYCLE(100)
{ {
orbGPU(img, GpuMat(), keypoints, descriptors); orbGPU(img, cv::gpu::GpuMat(), keypoints, descriptors);
} }
} }
INSTANTIATE_TEST_CASE_P(Features2D, ORB, DEVICES(cv::gpu::GLOBAL_ATOMICS));
#endif

176
modules/gpu/perf/perf_filters.cpp
View File

@ -1,89 +1,95 @@
#include "perf_precomp.hpp" #include "perf_precomp.hpp"
PERF_TEST_P(DevInfo_Size_MatType_KernelSize, boxFilter, testing::Combine(testing::ValuesIn(devices()), #ifdef HAVE_CUDA
testing::Values(GPU_TYPICAL_MAT_SIZES),
testing::Values(CV_8UC1, CV_8UC4), //////////////////////////////////////////////////////////////////////
testing::Values(3, 5))) // BoxFilter
GPU_PERF_TEST(BoxFilter, cv::gpu::DeviceInfo, cv::Size, perf::MatType, int)
{ {
DeviceInfo devInfo = std::tr1::get<0>(GetParam()); cv::gpu::DeviceInfo devInfo = GET_PARAM(0);
Size size = std::tr1::get<1>(GetParam()); cv::Size size = GET_PARAM(1);
int type = std::tr1::get<2>(GetParam()); int type = GET_PARAM(2);
int ksize = std::tr1::get<3>(GetParam()); int ksize = GET_PARAM(3);
setDevice(devInfo.deviceID()); cv::gpu::setDevice(devInfo.deviceID());
Mat src_host(size, type); cv::Mat src_host(size, type);
declare.in(src_host, WARMUP_RNG); declare.in(src_host, WARMUP_RNG);
GpuMat src(src_host); cv::gpu::GpuMat src(src_host);
GpuMat dst(size, type); cv::gpu::GpuMat dst;
Ptr<FilterEngine_GPU> filter = createBoxFilter_GPU(type, type, Size(ksize, ksize)); cv::Ptr<cv::gpu::FilterEngine_GPU> filter = cv::gpu::createBoxFilter_GPU(type, type, cv::Size(ksize, ksize));
TEST_CYCLE(100) TEST_CYCLE(100)
{ {
filter->apply(src, dst); filter->apply(src, dst);
} }
Mat dst_host(dst);
SANITY_CHECK(dst_host);
} }
PERF_TEST_P(DevInfo_Size_MatType_MorphOp_KernelSize, morphologyFilter, testing::Combine(testing::ValuesIn(devices()), INSTANTIATE_TEST_CASE_P(Filter, BoxFilter, testing::Combine(
testing::Values(GPU_TYPICAL_MAT_SIZES), ALL_DEVICES,
GPU_TYPICAL_MAT_SIZES,
testing::Values(CV_8UC1, CV_8UC4), testing::Values(CV_8UC1, CV_8UC4),
testing::Values((int)MORPH_ERODE, (int)MORPH_DILATE), testing::Values(3, 5)));
testing::Values(3, 5)))
//////////////////////////////////////////////////////////////////////
// MorphologyFilter
GPU_PERF_TEST(MorphologyFilter, cv::gpu::DeviceInfo, cv::Size, perf::MatType, MorphOp, int)
{ {
DeviceInfo devInfo = std::tr1::get<0>(GetParam()); cv::gpu::DeviceInfo devInfo = GET_PARAM(0);
Size size = std::tr1::get<1>(GetParam()); cv::Size size = GET_PARAM(1);
int type = std::tr1::get<2>(GetParam()); int type = GET_PARAM(2);
int op = std::tr1::get<3>(GetParam()); int op = GET_PARAM(3);
int ksize = std::tr1::get<4>(GetParam()); int ksize = GET_PARAM(4);
setDevice(devInfo.deviceID()); cv::gpu::setDevice(devInfo.deviceID());
Mat src_host(size, type); cv::Mat src_host(size, type);
declare.in(src_host, WARMUP_RNG); declare.in(src_host, WARMUP_RNG);
GpuMat src(src_host); cv::gpu::GpuMat src(src_host);
GpuMat dst(size, type); cv::gpu::GpuMat dst;
Ptr<FilterEngine_GPU> filter = createMorphologyFilter_GPU(op, type, Mat::ones(ksize, ksize, CV_8U)); cv::Ptr<cv::gpu::FilterEngine_GPU> filter = cv::gpu::createMorphologyFilter_GPU(op, type, cv::Mat::ones(ksize, ksize, CV_8U));
TEST_CYCLE(100) TEST_CYCLE(100)
{ {
filter->apply(src, dst); filter->apply(src, dst);
} }
Mat dst_host(dst);
SANITY_CHECK(dst_host);
} }
PERF_TEST_P(DevInfo_Size_MatType_KernelSize, linearFilter, testing::Combine(testing::ValuesIn(devices()), INSTANTIATE_TEST_CASE_P(Filter, MorphologyFilter, testing::Combine(
testing::Values(GPU_TYPICAL_MAT_SIZES), ALL_DEVICES,
GPU_TYPICAL_MAT_SIZES,
testing::Values(CV_8UC1, CV_8UC4), testing::Values(CV_8UC1, CV_8UC4),
testing::Values(3, 5))) testing::Values((int) cv::MORPH_ERODE, (int) cv::MORPH_DILATE),
testing::Values(3, 5)));
//////////////////////////////////////////////////////////////////////
// LinearFilter
GPU_PERF_TEST(LinearFilter, cv::gpu::DeviceInfo, cv::Size, perf::MatType, int)
{ {
DeviceInfo devInfo = std::tr1::get<0>(GetParam()); cv::gpu::DeviceInfo devInfo = GET_PARAM(0);
Size size = std::tr1::get<1>(GetParam()); cv::Size size = GET_PARAM(1);
int type = std::tr1::get<2>(GetParam()); int type = GET_PARAM(2);
int ksize = std::tr1::get<3>(GetParam()); int ksize = GET_PARAM(3);
setDevice(devInfo.deviceID()); cv::gpu::setDevice(devInfo.deviceID());
Mat src_host(size, type); cv::Mat src_host(size, type);
declare.in(src_host, WARMUP_RNG); declare.in(src_host, WARMUP_RNG);
GpuMat src(src_host); cv::gpu::GpuMat src(src_host);
GpuMat dst(size, type); cv::gpu::GpuMat dst;
Ptr<FilterEngine_GPU> filter = createLinearFilter_GPU(type, type, Mat::ones(ksize, ksize, CV_8U)); cv::Ptr<cv::gpu::FilterEngine_GPU> filter = cv::gpu::createLinearFilter_GPU(type, type, cv::Mat::ones(ksize, ksize, CV_8U));
declare.time(1.0); declare.time(1.0);
@ -91,42 +97,48 @@ PERF_TEST_P(DevInfo_Size_MatType_KernelSize, linearFilter, testing::Combine(test
{ {
filter->apply(src, dst); filter->apply(src, dst);
} }
Mat dst_host(dst);
SANITY_CHECK(dst_host);
} }
PERF_TEST_P(DevInfo_Size_MatType_KernelSize, separableLinearFilter, testing::Combine(testing::ValuesIn(devices()), INSTANTIATE_TEST_CASE_P(Filter, LinearFilter, testing::Combine(
testing::Values(GPU_TYPICAL_MAT_SIZES), ALL_DEVICES,
GPU_TYPICAL_MAT_SIZES,
testing::Values(CV_8UC1, CV_8UC4),
testing::Values(3, 5)));
//////////////////////////////////////////////////////////////////////
// SeparableLinearFilter
GPU_PERF_TEST(SeparableLinearFilter, cv::gpu::DeviceInfo, cv::Size, perf::MatType, int)
{
cv::gpu::DeviceInfo devInfo = GET_PARAM(0);
cv::Size size = GET_PARAM(1);
int type = GET_PARAM(2);
int ksize = GET_PARAM(3);
cv::gpu::setDevice(devInfo.deviceID());
cv::Mat src_host(size, type);
declare.in(src_host, WARMUP_RNG);
cv::gpu::GpuMat src(src_host);
cv::gpu::GpuMat dst;
cv::Mat kernel = cv::getGaussianKernel(ksize, 0.5, CV_32F);
cv::Ptr<cv::gpu::FilterEngine_GPU> filter = cv::gpu::createSeparableLinearFilter_GPU(type, type, kernel, kernel);
declare.time(1.0);
TEST_CYCLE(100)
{
filter->apply(src, dst, cv::Rect(0, 0, src.cols, src.rows));
}
}
INSTANTIATE_TEST_CASE_P(Filter, SeparableLinearFilter, testing::Combine(
ALL_DEVICES,
GPU_TYPICAL_MAT_SIZES,
testing::Values(CV_8UC1, CV_8UC4, CV_32FC1), testing::Values(CV_8UC1, CV_8UC4, CV_32FC1),
testing::Values(3, 5))) testing::Values(3, 5)));
{
DeviceInfo devInfo = std::tr1::get<0>(GetParam());
Size size = std::tr1::get<1>(GetParam());
int type = std::tr1::get<2>(GetParam());
int ksize = std::tr1::get<3>(GetParam());
setDevice(devInfo.deviceID()); #endif
Mat src_host(size, type);
declare.in(src_host, WARMUP_RNG);
GpuMat src(src_host);
GpuMat dst(size, type);
Mat kernel = getGaussianKernel(ksize, 0.5, CV_32F);
Ptr<FilterEngine_GPU> filter = createSeparableLinearFilter_GPU(type, type, kernel, kernel, Point(-1,-1));
declare.time(1.0);
TEST_CYCLE(100)
{
filter->apply(src, dst, Rect(0, 0, src.cols, src.rows));
}
Mat dst_host(dst);
SANITY_CHECK(dst_host);
}

1167
modules/gpu/perf/perf_imgproc.cpp
View File

File diff suppressed because it is too large Load Diff

3
modules/gpu/perf/perf_main.cpp
View File

@ -5,8 +5,7 @@
int main(int argc, char **argv) int main(int argc, char **argv)
{ {
testing::InitGoogleTest(&argc, argv); testing::InitGoogleTest(&argc, argv);
Regression::Init("gpu"); perf::TestBase::Init(argc, argv);
TestBase::Init(argc, argv);
return RUN_ALL_TESTS(); return RUN_ALL_TESTS();
} }

207
modules/gpu/perf/perf_matop.cpp
View File

@ -1,176 +1,185 @@
#include "perf_precomp.hpp" #include "perf_precomp.hpp"
PERF_TEST_P(DevInfo_Size_MatType, merge, testing::Combine(testing::ValuesIn(devices()), #ifdef HAVE_CUDA
testing::Values(GPU_TYPICAL_MAT_SIZES),
testing::Values(CV_8UC1, CV_16UC1, CV_32FC1)))
{
DeviceInfo devInfo = std::tr1::get<0>(GetParam());
Size size = std::tr1::get<1>(GetParam());
int type = std::tr1::get<2>(GetParam());
setDevice(devInfo.deviceID()); //////////////////////////////////////////////////////////////////////
// Merge
GPU_PERF_TEST(Merge, cv::gpu::DeviceInfo, cv::Size, perf::MatType)
{
cv::gpu::DeviceInfo devInfo = GET_PARAM(0);
cv::Size size = GET_PARAM(1);
int type = GET_PARAM(2);
cv::gpu::setDevice(devInfo.deviceID());
const int num_channels = 4; const int num_channels = 4;
vector<GpuMat> src(num_channels); std::vector<cv::gpu::GpuMat> src(num_channels);
for (int i = 0; i < num_channels; ++i) for (int i = 0; i < num_channels; ++i)
src[i] = GpuMat(size, type, cv::Scalar::all(i)); src[i] = cv::gpu::GpuMat(size, type, cv::Scalar::all(i));
GpuMat dst(size, CV_MAKETYPE(type, num_channels)); cv::gpu::GpuMat dst;
TEST_CYCLE(100) TEST_CYCLE(100)
{ {
merge(src, dst); cv::gpu::merge(src, dst);
}
} }
Mat dst_host(dst); INSTANTIATE_TEST_CASE_P(MatOp, Merge, testing::Combine(
ALL_DEVICES,
GPU_TYPICAL_MAT_SIZES,
testing::Values(CV_8UC1, CV_16UC1, CV_32FC1)));
SANITY_CHECK(dst_host); //////////////////////////////////////////////////////////////////////
} // Split
PERF_TEST_P(DevInfo_Size_MatType, split, testing::Combine(testing::ValuesIn(devices()), GPU_PERF_TEST(Split, cv::gpu::DeviceInfo, cv::Size, perf::MatType)
testing::Values(GPU_TYPICAL_MAT_SIZES),
testing::Values(CV_8UC1, CV_16UC1, CV_32FC1)))
{ {
DeviceInfo devInfo = std::tr1::get<0>(GetParam()); cv::gpu::DeviceInfo devInfo = GET_PARAM(0);
Size size = std::tr1::get<1>(GetParam()); cv::Size size = GET_PARAM(1);
int type = std::tr1::get<2>(GetParam()); int type = GET_PARAM(2);
setDevice(devInfo.deviceID()); cv::gpu::setDevice(devInfo.deviceID());
const int num_channels = 4; const int num_channels = 4;
GpuMat src(size, CV_MAKETYPE(type, num_channels), cv::Scalar(1, 2, 3, 4)); cv::gpu::GpuMat src(size, CV_MAKETYPE(type, num_channels), cv::Scalar(1, 2, 3, 4));
vector<GpuMat> dst(num_channels); std::vector<cv::gpu::GpuMat> dst(num_channels);
for (int i = 0; i < num_channels; ++i) for (int i = 0; i < num_channels; ++i)
dst[i] = GpuMat(size, type); dst[i] = cv::gpu::GpuMat(size, type);
TEST_CYCLE(100) TEST_CYCLE(100)
{ {
split(src, dst); cv::gpu::split(src, dst);
}
} }
vector<Mat> dst_host(dst.size()); INSTANTIATE_TEST_CASE_P(MatOp, Split, testing::Combine(
for (size_t i = 0; i < dst.size(); ++i) ALL_DEVICES,
dst[i].download(dst_host[i]); GPU_TYPICAL_MAT_SIZES,
testing::Values(CV_8UC1, CV_16UC1, CV_32FC1)));
SANITY_CHECK(dst_host); //////////////////////////////////////////////////////////////////////
} // SetTo
PERF_TEST_P(DevInfo_Size_MatType, setTo, testing::Combine(testing::ValuesIn(devices()), GPU_PERF_TEST(SetTo, cv::gpu::DeviceInfo, cv::Size, perf::MatType)
testing::Values(GPU_TYPICAL_MAT_SIZES),
testing::Values(CV_8UC1, CV_8UC4, CV_32FC1)))
{ {
DeviceInfo devInfo = std::tr1::get<0>(GetParam()); cv::gpu::DeviceInfo devInfo = GET_PARAM(0);
Size size = std::tr1::get<1>(GetParam()); cv::Size size = GET_PARAM(1);
int type = std::tr1::get<2>(GetParam()); int type = GET_PARAM(2);
setDevice(devInfo.deviceID()); cv::gpu::setDevice(devInfo.deviceID());
GpuMat src(size, type); cv::gpu::GpuMat src(size, type);
Scalar val(1, 2, 3, 4); cv::Scalar val(1, 2, 3, 4);
TEST_CYCLE(100) TEST_CYCLE(100)
{ {
src.setTo(val); src.setTo(val);
} }
Mat src_host(src);
SANITY_CHECK(src_host);
} }
PERF_TEST_P(DevInfo_Size_MatType, setToMasked, testing::Combine(testing::ValuesIn(devices()), INSTANTIATE_TEST_CASE_P(MatOp, SetTo, testing::Combine(
testing::Values(GPU_TYPICAL_MAT_SIZES), ALL_DEVICES,
testing::Values(CV_8UC1, CV_8UC4, CV_32FC1))) GPU_TYPICAL_MAT_SIZES,
testing::Values(CV_8UC1, CV_8UC3, CV_8UC4, CV_16UC1, CV_16UC3, CV_16UC4, CV_32FC1, CV_32FC3, CV_32FC4)));
//////////////////////////////////////////////////////////////////////
// SetToMasked
GPU_PERF_TEST(SetToMasked, cv::gpu::DeviceInfo, cv::Size, perf::MatType)
{ {
DeviceInfo devInfo = std::tr1::get<0>(GetParam()); cv::gpu::DeviceInfo devInfo = GET_PARAM(0);
Size size = std::tr1::get<1>(GetParam()); cv::Size size = GET_PARAM(1);
int type = std::tr1::get<2>(GetParam()); int type = GET_PARAM(2);
setDevice(devInfo.deviceID()); cv::gpu::setDevice(devInfo.deviceID());
Mat src_host(size, type); cv::Mat src_host(size, type);
cv::Mat mask_host(size, CV_8UC1);
declare.in(src_host, WARMUP_RNG); declare.in(src_host, WARMUP_RNG);
fill(mask_host, 0, 2);
GpuMat src(src_host); cv::gpu::GpuMat src(src_host);
Scalar val(1, 2, 3, 4); cv::Scalar val(1, 2, 3, 4);
cv::gpu::GpuMat mask(mask_host);
Mat mask_host(size, CV_8UC1);
randu(mask_host, 0.0, 2.0);
GpuMat mask(mask_host);
TEST_CYCLE(100) TEST_CYCLE(100)
{ {
src.setTo(val, mask); src.setTo(val, mask);
} }
src.download(src_host);
SANITY_CHECK(src_host);
} }
PERF_TEST_P(DevInfo_Size_MatType, copyToMasked, testing::Combine(testing::ValuesIn(devices()), INSTANTIATE_TEST_CASE_P(MatOp, SetToMasked, testing::Combine(
testing::Values(GPU_TYPICAL_MAT_SIZES), ALL_DEVICES,
testing::Values(CV_8UC1, CV_8UC4, CV_32FC1))) GPU_TYPICAL_MAT_SIZES,
testing::Values(CV_8UC1, CV_8UC3, CV_8UC4, CV_16UC1, CV_16UC3, CV_16UC4, CV_32FC1, CV_32FC3, CV_32FC4)));
//////////////////////////////////////////////////////////////////////
// CopyToMasked
GPU_PERF_TEST(CopyToMasked, cv::gpu::DeviceInfo, cv::Size, perf::MatType)
{ {
DeviceInfo devInfo = std::tr1::get<0>(GetParam()); cv::gpu::DeviceInfo devInfo = GET_PARAM(0);
Size size = std::tr1::get<1>(GetParam()); cv::Size size = GET_PARAM(1);
int type = std::tr1::get<2>(GetParam()); int type = GET_PARAM(2);
setDevice(devInfo.deviceID()); cv::gpu::setDevice(devInfo.deviceID());
Mat src_host(size, type); cv::Mat src_host(size, type);
cv::Mat mask_host(size, CV_8UC1);
declare.in(src_host, WARMUP_RNG); declare.in(src_host, WARMUP_RNG);
fill(mask_host, 0, 2);
GpuMat src(src_host); cv::gpu::GpuMat src(src_host);
GpuMat dst(size, type); cv::gpu::GpuMat mask(mask_host);
cv::gpu::GpuMat dst;
Mat mask_host(size, CV_8UC1);
randu(mask_host, 0.0, 2.0);
GpuMat mask(mask_host);
TEST_CYCLE(100) TEST_CYCLE(100)
{ {
src.copyTo(dst, mask); src.copyTo(dst, mask);
} }
Mat dst_host(dst);
SANITY_CHECK(dst_host);
} }
PERF_TEST_P(DevInfo_Size_MatType_MatType, convertTo, testing::Combine(testing::ValuesIn(devices()), INSTANTIATE_TEST_CASE_P(MatOp, CopyToMasked, testing::Combine(
testing::Values(GPU_TYPICAL_MAT_SIZES), ALL_DEVICES,
testing::Values(CV_8UC1, CV_16UC1, CV_32FC1), GPU_TYPICAL_MAT_SIZES,
testing::Values(CV_8UC1, CV_16UC1, CV_32FC1))) testing::Values(CV_8UC1, CV_8UC3, CV_8UC4, CV_16UC1, CV_16UC3, CV_16UC4, CV_32FC1, CV_32FC3, CV_32FC4)));
//////////////////////////////////////////////////////////////////////
// ConvertTo
GPU_PERF_TEST(ConvertTo, cv::gpu::DeviceInfo, cv::Size, perf::MatType, perf::MatType)
{ {
DeviceInfo devInfo = std::tr1::get<0>(GetParam()); cv::gpu::DeviceInfo devInfo = GET_PARAM(0);
Size size = std::tr1::get<1>(GetParam()); cv::Size size = GET_PARAM(1);
int type1 = std::tr1::get<2>(GetParam()); int type1 = GET_PARAM(2);
int type2 = std::tr1::get<3>(GetParam()); int type2 = GET_PARAM(3);
setDevice(devInfo.deviceID()); cv::gpu::setDevice(devInfo.deviceID());
Mat src_host(size, type1); cv::Mat src_host(size, type1);
declare.in(src_host, WARMUP_RNG); declare.in(src_host, WARMUP_RNG);
GpuMat src(src_host); cv::gpu::GpuMat src(src_host);
GpuMat dst(size, type2); cv::gpu::GpuMat dst;
double a = 0.5;
double b = 1.0;
TEST_CYCLE(100) TEST_CYCLE(100)
{ {
src.convertTo(dst, type2, a, b); src.convertTo(dst, type2, 0.5, 1.0);
}
} }
Mat dst_host(dst); INSTANTIATE_TEST_CASE_P(MatOp, ConvertTo, testing::Combine(
ALL_DEVICES,
GPU_TYPICAL_MAT_SIZES,
testing::Values(CV_8UC1, CV_16UC1, CV_32FC1),
testing::Values(CV_8UC1, CV_16UC1, CV_32FC1)));
SANITY_CHECK(dst_host); #endif
}

22
modules/gpu/perf/perf_objdetect.cpp
View File

@ -1,21 +1,27 @@
#include "perf_precomp.hpp" #include "perf_precomp.hpp"
PERF_TEST_P(DevInfo, HOGDescriptor, testing::ValuesIn(devices())) #ifdef HAVE_CUDA
GPU_PERF_TEST_1(HOG, cv::gpu::DeviceInfo)
{ {
DeviceInfo devInfo = GetParam(); cv::gpu::DeviceInfo devInfo = GetParam();
setDevice(devInfo.deviceID()); cv::gpu::setDevice(devInfo.deviceID());
Mat img_host = readImage("gpu/hog/road.png", CV_LOAD_IMAGE_GRAYSCALE); cv::Mat img_host = readImage("gpu/hog/road.png", cv::IMREAD_GRAYSCALE);
GpuMat img(img_host); cv::gpu::GpuMat img(img_host);
vector<Rect> found_locations; std::vector<cv::Rect> found_locations;
gpu::HOGDescriptor hog; cv::gpu::HOGDescriptor hog;
hog.setSVMDetector(gpu::HOGDescriptor::getDefaultPeopleDetector()); hog.setSVMDetector(cv::gpu::HOGDescriptor::getDefaultPeopleDetector());
TEST_CYCLE(100) TEST_CYCLE(100)
{ {
hog.detectMultiScale(img, found_locations); hog.detectMultiScale(img, found_locations);
} }
} }
INSTANTIATE_TEST_CASE_P(ObjDetect, HOG, ALL_DEVICES);
#endif

1
modules/gpu/perf/perf_precomp.hpp
View File

@ -5,6 +5,7 @@
#include <iostream> #include <iostream>
#include "cvconfig.h" #include "cvconfig.h"
#include "opencv2/ts/ts.hpp" #include "opencv2/ts/ts.hpp"
#include "opencv2/ts/ts_perf.hpp"
#include "opencv2/core/core.hpp" #include "opencv2/core/core.hpp"
#include "opencv2/highgui/highgui.hpp" #include "opencv2/highgui/highgui.hpp"
#include "opencv2/gpu/gpu.hpp" #include "opencv2/gpu/gpu.hpp"

127
modules/gpu/perf/perf_utility.cpp
View File

@ -1,70 +1,16 @@
#include "perf_precomp.hpp" #include "perf_precomp.hpp"
using namespace std; using namespace std;
using namespace cv;
using namespace cv::gpu; using namespace cv::gpu;
Mat readImage(const string& fileName, int flags) void fill(Mat& m, double a, double b)
{ {
return imread(::perf::TestBase::getDataPath(fileName), flags); RNG rng(123456789);
rng.fill(m, RNG::UNIFORM, a, b);
} }
bool supportFeature(const DeviceInfo& info, FeatureSet feature) void PrintTo(const CvtColorInfo& info, ostream* os)
{
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;
}
void cv::gpu::PrintTo(const DeviceInfo& info, ostream* os)
{
*os << info.name();
}
void PrintTo(const CvtColorInfo& info, ::std::ostream* os)
{ {
static const char* str[] = static const char* str[] =
{ {
@ -190,3 +136,66 @@ void PrintTo(const CvtColorInfo& info, ::std::ostream* os)
*os << str[info.code]; *os << str[info.code];
} }
void cv::gpu::PrintTo(const DeviceInfo& info, ostream* os)
{
*os << info.name();
}
Mat readImage(const string& fileName, int flags)
{
return imread(perf::TestBase::getDataPath(fileName), flags);
}
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;
}

92
modules/gpu/perf/perf_utility.hpp
View File

@ -1,25 +1,16 @@
#ifndef __OPENCV_PERF_GPU_UTILITY_HPP__ #ifndef __OPENCV_PERF_GPU_UTILITY_HPP__
#define __OPENCV_PERF_GPU_UTILITY_HPP__ #define __OPENCV_PERF_GPU_UTILITY_HPP__
#include <iosfwd> void fill(cv::Mat& m, double a, double b);
#include "opencv2/ts/ts.hpp"
#include "opencv2/highgui/highgui.hpp"
#include "opencv2/gpu/gpu.hpp"
using namespace std;
using namespace cv;
using namespace cv::gpu;
using namespace perf;
enum {HORIZONTAL_AXIS = 0, VERTICAL_AXIS = 1, BOTH_AXIS = -1}; enum {HORIZONTAL_AXIS = 0, VERTICAL_AXIS = 1, BOTH_AXIS = -1};
CV_ENUM(MorphOp, MORPH_ERODE, MORPH_DILATE) CV_ENUM(MorphOp, cv::MORPH_ERODE, cv::MORPH_DILATE)
CV_ENUM(BorderMode, BORDER_REFLECT101, BORDER_REPLICATE, BORDER_CONSTANT, BORDER_REFLECT, BORDER_WRAP) CV_ENUM(BorderMode, cv::BORDER_REFLECT101, cv::BORDER_REPLICATE, cv::BORDER_CONSTANT, cv::BORDER_REFLECT, cv::BORDER_WRAP)
CV_ENUM(FlipCode, HORIZONTAL_AXIS, VERTICAL_AXIS, BOTH_AXIS) CV_ENUM(FlipCode, HORIZONTAL_AXIS, VERTICAL_AXIS, BOTH_AXIS)
CV_ENUM(CmpOp, CMP_EQ, CMP_GT, CMP_GE, CMP_LT, CMP_LE, CMP_NE) CV_ENUM(Interpolation, cv::INTER_NEAREST, cv::INTER_LINEAR, cv::INTER_CUBIC)
CV_ENUM(Interpolation, INTER_NEAREST, INTER_LINEAR, INTER_CUBIC) CV_ENUM(MatchMethod, cv::TM_SQDIFF, cv::TM_SQDIFF_NORMED, cv::TM_CCORR, cv::TM_CCORR_NORMED, cv::TM_CCOEFF, cv::TM_CCOEFF_NORMED)
CV_ENUM(MatchMethod, TM_SQDIFF, TM_SQDIFF_NORMED, TM_CCORR, TM_CCORR_NORMED, TM_CCOEFF, TM_CCOEFF_NORMED) CV_ENUM(NormType, cv::NORM_INF, cv::NORM_L1, cv::NORM_L2)
CV_ENUM(NormType, NORM_INF, NORM_L1, NORM_L2)
struct CvtColorInfo struct CvtColorInfo
{ {
@ -30,53 +21,48 @@ struct CvtColorInfo
explicit CvtColorInfo(int scn_=0, int dcn_=0, int code_=0) : scn(scn_), dcn(dcn_), code(code_) {} explicit CvtColorInfo(int scn_=0, int dcn_=0, int code_=0) : scn(scn_), dcn(dcn_), code(code_) {}
}; };
typedef TestBaseWithParam<DeviceInfo> DevInfo; void PrintTo(const CvtColorInfo& info, std::ostream* os);
typedef TestBaseWithParam< std::tr1::tuple<DeviceInfo, Size> > DevInfo_Size;
typedef TestBaseWithParam< std::tr1::tuple<DeviceInfo, int, int> > DevInfo_Int_Int;
typedef TestBaseWithParam< std::tr1::tuple<DeviceInfo, MatType> > DevInfo_MatType;
typedef TestBaseWithParam< std::tr1::tuple<DeviceInfo, Size, MatType> > DevInfo_Size_MatType;
typedef TestBaseWithParam< std::tr1::tuple<DeviceInfo, Size, MatType, MatType> > DevInfo_Size_MatType_MatType;
typedef TestBaseWithParam< std::tr1::tuple<DeviceInfo, Size, MatType, int> > DevInfo_Size_MatType_KernelSize;
typedef TestBaseWithParam< std::tr1::tuple<DeviceInfo, Size, MatType, MorphOp, int> > DevInfo_Size_MatType_MorphOp_KernelSize;
typedef TestBaseWithParam< std::tr1::tuple<DeviceInfo, Size, MatType, int, BorderMode> > DevInfo_Size_MatType_KernelSize_BorderMode;
typedef TestBaseWithParam< std::tr1::tuple<DeviceInfo, Size, MatType, FlipCode> > DevInfo_Size_MatType_FlipCode;
typedef TestBaseWithParam< std::tr1::tuple<DeviceInfo, Size, MatType, CmpOp> > DevInfo_Size_MatType_CmpOp;
typedef TestBaseWithParam< std::tr1::tuple<DeviceInfo, Size, MatType, Interpolation> > DevInfo_Size_MatType_Interpolation;
typedef TestBaseWithParam< std::tr1::tuple<DeviceInfo, Size, MatType, Interpolation, double> > DevInfo_Size_MatType_Interpolation_SizeCoeff;
typedef TestBaseWithParam< std::tr1::tuple<DeviceInfo, Size, MatType, Interpolation, BorderMode> > DevInfo_Size_MatType_Interpolation_BorderMode;
typedef TestBaseWithParam< std::tr1::tuple<DeviceInfo, Size, MatType, CvtColorInfo> > DevInfo_Size_MatType_CvtColorInfo;
typedef TestBaseWithParam< std::tr1::tuple<DeviceInfo, Size, MatType, MatchMethod> > DevInfo_Size_MatType_MatchMethod;
typedef TestBaseWithParam< std::tr1::tuple<DeviceInfo, Size, NormType> > DevInfo_Size_NormType;
typedef TestBaseWithParam< std::tr1::tuple<DeviceInfo, Size, MatType, NormType> > DevInfo_Size_MatType_NormType;
typedef TestBaseWithParam< std::tr1::tuple<DeviceInfo, int> > DevInfo_DescSize;
typedef TestBaseWithParam< std::tr1::tuple<DeviceInfo, int, int> > DevInfo_K_DescSize;
typedef TestBaseWithParam< std::tr1::tuple<DeviceInfo, Size, MatType, BorderMode> > DevInfo_Size_MatType_BorderMode;
const cv::Size sz1800x1500 = cv::Size(1800, 1500); namespace cv { namespace gpu
const cv::Size sz4700x3000 = cv::Size(4700, 3000); {
void PrintTo(const cv::gpu::DeviceInfo& info, std::ostream* os);
}}
//#define GPU_TYPICAL_MAT_SIZES szXGA, szSXGA, sz720p, sz1080p, sz1800x1500, sz4700x3000 #define GPU_PERF_TEST(name, ...) \
#define GPU_TYPICAL_MAT_SIZES szSXGA, sz1080p, sz4700x3000 struct name : perf::TestBaseWithParam< std::tr1::tuple< __VA_ARGS__ > > \
{ \
public: \
name() {} \
protected: \
void PerfTestBody(); \
}; \
TEST_P(name, perf){ RunPerfTestBody(); } \
void name :: PerfTestBody()
//! read image from testdata folder. #define GPU_PERF_TEST_1(name, param_type) \
Mat readImage(const string& fileName, int flags = CV_LOAD_IMAGE_COLOR); struct name : perf::TestBaseWithParam< param_type > \
{ \
public: \
name() {} \
protected: \
void PerfTestBody(); \
}; \
TEST_P(name, perf){ RunPerfTestBody(); } \
void name :: PerfTestBody()
#define GPU_TYPICAL_MAT_SIZES testing::Values(perf::szSXGA, perf::sz1080p, cv::Size(1800, 1500))
cv::Mat readImage(const std::string& fileName, int flags = cv::IMREAD_COLOR);
//! 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); bool supportFeature(const cv::gpu::DeviceInfo& info, cv::gpu::FeatureSet feature);
//! return all devices compatible with current gpu module build.
const std::vector<cv::gpu::DeviceInfo>& devices(); const std::vector<cv::gpu::DeviceInfo>& devices();
//! return all devices compatible with current gpu module build which support specified feature.
std::vector<cv::gpu::DeviceInfo> devices(cv::gpu::FeatureSet feature); std::vector<cv::gpu::DeviceInfo> devices(cv::gpu::FeatureSet feature);
namespace cv #define ALL_DEVICES testing::ValuesIn(devices())
{ #define DEVICES(feature) testing::ValuesIn(devices(feature))
namespace gpu
{
void PrintTo(const DeviceInfo& info, ::std::ostream* os);
}
}
void PrintTo(const CvtColorInfo& info, ::std::ostream* os); #define GET_PARAM(k) std::tr1::get< k >(GetParam())
#endif // __OPENCV_PERF_GPU_UTILITY_HPP__ #endif // __OPENCV_PERF_GPU_UTILITY_HPP__

81
modules/gpu/perf/perf_video.cpp Normal file
View File

@ -0,0 +1,81 @@
#include "perf_precomp.hpp"
#ifdef HAVE_CUDA
//////////////////////////////////////////////////////
// BroxOpticalFlow
GPU_PERF_TEST_1(BroxOpticalFlow, cv::gpu::DeviceInfo)
{
cv::gpu::DeviceInfo devInfo = GetParam();
cv::gpu::setDevice(devInfo.deviceID());
cv::Mat frame0_host = readImage("gpu/perf/aloe.jpg", cv::IMREAD_GRAYSCALE);
cv::Mat frame1_host = readImage("gpu/perf/aloeR.jpg", cv::IMREAD_GRAYSCALE);
ASSERT_FALSE(frame0_host.empty());
ASSERT_FALSE(frame1_host.empty());
frame0_host.convertTo(frame0_host, CV_32FC1, 1.0 / 255.0);
frame1_host.convertTo(frame1_host, CV_32FC1, 1.0 / 255.0);
cv::gpu::GpuMat frame0(frame0_host);
cv::gpu::GpuMat frame1(frame1_host);
cv::gpu::GpuMat u;
cv::gpu::GpuMat v;
cv::gpu::BroxOpticalFlow d_flow(0.197f /*alpha*/, 50.0f /*gamma*/, 0.8f /*scale_factor*/,
10 /*inner_iterations*/, 77 /*outer_iterations*/, 10 /*solver_iterations*/);
declare.time(10);
TEST_CYCLE(100)
{
d_flow(frame0, frame1, u, v);
}
}
INSTANTIATE_TEST_CASE_P(Video, BroxOpticalFlow, ALL_DEVICES);
//////////////////////////////////////////////////////
// InterpolateFrames
GPU_PERF_TEST_1(InterpolateFrames, cv::gpu::DeviceInfo)
{
cv::gpu::DeviceInfo devInfo = GetParam();
cv::gpu::setDevice(devInfo.deviceID());
cv::Mat frame0_host = readImage("gpu/perf/aloe.jpg", cv::IMREAD_GRAYSCALE);
cv::Mat frame1_host = readImage("gpu/perf/aloeR.jpg", cv::IMREAD_GRAYSCALE);
ASSERT_FALSE(frame0_host.empty());
ASSERT_FALSE(frame1_host.empty());
frame0_host.convertTo(frame0_host, CV_32FC1, 1.0 / 255.0);
frame1_host.convertTo(frame1_host, CV_32FC1, 1.0 / 255.0);
cv::gpu::GpuMat frame0(frame0_host);
cv::gpu::GpuMat frame1(frame1_host);
cv::gpu::GpuMat fu, fv;
cv::gpu::GpuMat bu, bv;
cv::gpu::BroxOpticalFlow d_flow(0.197f /*alpha*/, 50.0f /*gamma*/, 0.8f /*scale_factor*/,
10 /*inner_iterations*/, 77 /*outer_iterations*/, 10 /*solver_iterations*/);
d_flow(frame0, frame1, fu, fv);
d_flow(frame1, frame0, bu, bv);
cv::gpu::GpuMat newFrame;
cv::gpu::GpuMat buf;
TEST_CYCLE(100)
{
cv::gpu::interpolateFrames(frame0, frame1, fu, fv, bu, bv, 0.5f, newFrame, buf);
}
}
INSTANTIATE_TEST_CASE_P(Video, InterpolateFrames, ALL_DEVICES);
#endif