added perf4au app
authorVladislav Vinogradov <vlad.vinogradov@itseez.com>
Wed, 13 Feb 2013 11:58:24 +0000 (15:58 +0400)
committerVladislav Vinogradov <vlad.vinogradov@itseez.com>
Wed, 13 Feb 2013 11:58:24 +0000 (15:58 +0400)
modules/gpu/CMakeLists.txt
modules/gpu/perf4au/CMakeLists.txt [new file with mode: 0644]
modules/gpu/perf4au/im1_1280x800.jpg [new file with mode: 0644]
modules/gpu/perf4au/im2_1280x800.jpg [new file with mode: 0644]
modules/gpu/perf4au/main.cpp [new file with mode: 0644]

index 0c9f709..26bf624 100644 (file)
@@ -21,7 +21,7 @@ source_group("Src\\Cuda"      FILES ${lib_cuda} ${lib_cuda_hdrs})
 source_group("Device"         FILES ${lib_device_hdrs})
 source_group("Device\\Detail" FILES ${lib_device_hdrs_detail})
 
-if (HAVE_CUDA)
+if(HAVE_CUDA)
   file(GLOB_RECURSE ncv_srcs "src/nvidia/*.cpp" "src/nvidia/*.h*")
   file(GLOB_RECURSE ncv_cuda "src/nvidia/*.cu")
   set(ncv_files ${ncv_srcs} ${ncv_cuda})
@@ -104,3 +104,7 @@ ocv_add_accuracy_tests(FILES "Include" ${test_hdrs}
                        FILES "Src" ${test_srcs}
                        ${nvidia})
 ocv_add_perf_tests()
+
+if(HAVE_CUDA)
+  add_subdirectory(perf4au)
+endif()
diff --git a/modules/gpu/perf4au/CMakeLists.txt b/modules/gpu/perf4au/CMakeLists.txt
new file mode 100644 (file)
index 0000000..7452203
--- /dev/null
@@ -0,0 +1,28 @@
+set(PERF4AU_REQUIRED_DEPS opencv_core opencv_imgproc opencv_highgui opencv_video opencv_legacy opencv_gpu opencv_ts)
+
+ocv_check_dependencies(${PERF4AU_REQUIRED_DEPS})
+
+set(the_target gpu_perf4au)
+project(${the_target})
+
+ocv_include_modules(${PERF4AU_REQUIRED_DEPS})
+
+if(CMAKE_COMPILER_IS_GNUCXX AND NOT ENABLE_NOISY_WARNINGS)
+    set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -Wno-unused-function")
+endif()
+
+file(GLOB srcs RELATIVE ${CMAKE_CURRENT_SOURCE_DIR} *.cpp *.h *.hpp)
+add_executable(${the_target} ${srcs})
+
+target_link_libraries(${the_target} ${OPENCV_LINKER_LIBS} ${PERF4AU_REQUIRED_DEPS})
+
+if(ENABLE_SOLUTION_FOLDERS)
+  set_target_properties(${the_target} PROPERTIES FOLDER "tests performance")
+endif()
+
+if(WIN32)
+    if(MSVC AND NOT BUILD_SHARED_LIBS)
+        set_target_properties(${the_target} PROPERTIES LINK_FLAGS "/NODEFAULTLIB:atlthunk.lib /NODEFAULTLIB:atlsd.lib /DEBUG")
+    endif()
+endif()
+
diff --git a/modules/gpu/perf4au/im1_1280x800.jpg b/modules/gpu/perf4au/im1_1280x800.jpg
new file mode 100644 (file)
index 0000000..bdbbd4a
Binary files /dev/null and b/modules/gpu/perf4au/im1_1280x800.jpg differ
diff --git a/modules/gpu/perf4au/im2_1280x800.jpg b/modules/gpu/perf4au/im2_1280x800.jpg
new file mode 100644 (file)
index 0000000..ae49640
Binary files /dev/null and b/modules/gpu/perf4au/im2_1280x800.jpg differ
diff --git a/modules/gpu/perf4au/main.cpp b/modules/gpu/perf4au/main.cpp
new file mode 100644 (file)
index 0000000..80d97ea
--- /dev/null
@@ -0,0 +1,490 @@
+#include <cstdio>
+#ifdef HAVE_CVCONFIG_H
+#include "cvconfig.h"
+#endif
+#include "opencv2/core/core.hpp"
+#include "opencv2/gpu/gpu.hpp"
+#include "opencv2/highgui/highgui.hpp"
+#include "opencv2/video/video.hpp"
+#include "opencv2/legacy/legacy.hpp"
+#include "opencv2/ts/ts.hpp"
+#include "opencv2/ts/ts_perf.hpp"
+
+static void printOsInfo()
+{
+#if defined _WIN32
+#   if defined _WIN64
+        printf("[----------]\n[ GPU INFO ] \tRun on OS Windows x64.\n[----------]\n"); fflush(stdout);
+#   else
+        printf("[----------]\n[ GPU INFO ] \tRun on OS Windows x32.\n[----------]\n"); fflush(stdout);
+#   endif
+#elif defined linux
+#   if defined _LP64
+        printf("[----------]\n[ GPU INFO ] \tRun on OS Linux x64.\n[----------]\n"); fflush(stdout);
+#   else
+        printf("[----------]\n[ GPU INFO ] \tRun on OS Linux x32.\n[----------]\n"); fflush(stdout);
+#   endif
+#elif defined __APPLE__
+#   if defined _LP64
+        printf("[----------]\n[ GPU INFO ] \tRun on OS Apple x64.\n[----------]\n"); fflush(stdout);
+#   else
+        printf("[----------]\n[ GPU INFO ] \tRun on OS Apple x32.\n[----------]\n"); fflush(stdout);
+#   endif
+#endif
+}
+
+static void printCudaInfo()
+{
+    const int deviceCount = cv::gpu::getCudaEnabledDeviceCount();
+
+    printf("[----------]\n"); fflush(stdout);
+    printf("[ GPU INFO ] \tCUDA device count:: %d.\n", deviceCount); fflush(stdout);
+    printf("[----------]\n"); fflush(stdout);
+
+    for (int i = 0; i < deviceCount; ++i)
+    {
+        cv::gpu::DeviceInfo info(i);
+
+        printf("[----------]\n"); fflush(stdout);
+        printf("[ DEVICE   ] \t# %d %s.\n", i, info.name().c_str()); fflush(stdout);
+        printf("[          ] \tCompute capability: %d.%d\n", info.majorVersion(), info.minorVersion()); fflush(stdout);
+        printf("[          ] \tMulti Processor Count:  %d\n", info.multiProcessorCount()); fflush(stdout);
+        printf("[          ] \tTotal memory: %d Mb\n", static_cast<int>(static_cast<int>(info.totalMemory() / 1024.0) / 1024.0)); fflush(stdout);
+        printf("[          ] \tFree  memory: %d Mb\n", static_cast<int>(static_cast<int>(info.freeMemory()  / 1024.0) / 1024.0)); fflush(stdout);
+        if (!info.isCompatible())
+            printf("[ GPU INFO ] \tThis device is NOT compatible with current GPU module build\n");
+        printf("[----------]\n"); fflush(stdout);
+    }
+}
+
+int main(int argc, char* argv[])
+{
+    printOsInfo();
+    printCudaInfo();
+
+    perf::Regression::Init("nv_perf_test");
+    perf::TestBase::Init(argc, argv);
+    testing::InitGoogleTest(&argc, argv);
+
+    return RUN_ALL_TESTS();
+}
+
+#define DEF_PARAM_TEST(name, ...) typedef ::perf::TestBaseWithParam< std::tr1::tuple< __VA_ARGS__ > > name
+#define DEF_PARAM_TEST_1(name, param_type) typedef ::perf::TestBaseWithParam< param_type > name
+
+//////////////////////////////////////////////////////////
+// HoughLinesP
+
+DEF_PARAM_TEST_1(Image, std::string);
+
+PERF_TEST_P(Image, HoughLinesP, testing::Values(std::string("im1_1280x800.jpg")))
+{
+    declare.time(30.0);
+
+    std::string fileName = GetParam();
+
+    const float rho = 1.f;
+    const float theta = 1.f;
+    const int threshold = 40;
+    const int minLineLenght = 20;
+    const int maxLineGap = 5;
+
+    cv::Mat image = cv::imread(fileName, cv::IMREAD_GRAYSCALE);
+
+    if (PERF_RUN_GPU())
+    {
+        cv::gpu::GpuMat d_image(image);
+        cv::gpu::GpuMat d_lines;
+        cv::gpu::HoughLinesBuf d_buf;
+
+        cv::gpu::HoughLinesP(d_image, d_lines, d_buf, rho, theta, minLineLenght, maxLineGap);
+
+        TEST_CYCLE()
+        {
+            cv::gpu::HoughLinesP(d_image, d_lines, d_buf, rho, theta, minLineLenght, maxLineGap);
+        }
+    }
+    else
+    {
+        cv::Mat mask;
+        cv::Canny(image, mask, 50, 100);
+
+        std::vector<cv::Vec4i> lines;
+        cv::HoughLinesP(mask, lines, rho, theta, threshold, minLineLenght, maxLineGap);
+
+        TEST_CYCLE()
+        {
+            cv::HoughLinesP(mask, lines, rho, theta, threshold, minLineLenght, maxLineGap);
+        }
+    }
+
+    SANITY_CHECK(0);
+}
+
+//////////////////////////////////////////////////////////
+// GoodFeaturesToTrack
+
+DEF_PARAM_TEST(Image_Depth, std::string, perf::MatDepth);
+
+PERF_TEST_P(Image_Depth, GoodFeaturesToTrack,
+                testing::Combine(
+                testing::Values(std::string("im1_1280x800.jpg")),
+                testing::Values(CV_8U, CV_16U)
+                ))
+{
+    declare.time(60);
+
+    const std::string fileName = std::tr1::get<0>(GetParam());
+    const int depth = std::tr1::get<1>(GetParam());
+
+    const int maxCorners = 5000;
+    const double qualityLevel = 0.05;
+    const int minDistance = 5;
+    const int blockSize = 3;
+    const bool useHarrisDetector = true;
+    const double k = 0.05;
+
+    cv::Mat src = cv::imread(fileName, cv::IMREAD_GRAYSCALE);
+    if (src.empty())
+        FAIL() << "Unable to load source image [" << fileName << "]";
+
+    if (depth != CV_8U)
+        src.convertTo(src, depth);
+
+    cv::Mat mask(src.size(), CV_8UC1, cv::Scalar::all(1));
+    mask(cv::Rect(0, 0, 100, 100)).setTo(cv::Scalar::all(0));
+
+    if (PERF_RUN_GPU())
+    {
+        cv::gpu::GoodFeaturesToTrackDetector_GPU d_detector(maxCorners, qualityLevel, minDistance, blockSize, useHarrisDetector, k);
+
+        cv::gpu::GpuMat d_src(src);
+        cv::gpu::GpuMat d_mask(mask);
+        cv::gpu::GpuMat d_pts;
+
+        d_detector(d_src, d_pts, d_mask);
+
+        TEST_CYCLE()
+        {
+            d_detector(d_src, d_pts, d_mask);
+        }
+    }
+    else
+    {
+        if (depth != CV_8U)
+            FAIL() << "Unsupported depth";
+
+        cv::Mat pts;
+
+        cv::goodFeaturesToTrack(src, pts, maxCorners, qualityLevel, minDistance, mask, blockSize, useHarrisDetector, k);
+
+        TEST_CYCLE()
+        {
+            cv::goodFeaturesToTrack(src, pts, maxCorners, qualityLevel, minDistance, mask, blockSize, useHarrisDetector, k);
+        }
+    }
+
+    SANITY_CHECK(0);
+}
+
+//////////////////////////////////////////////////////////
+// OpticalFlowPyrLKSparse
+
+typedef std::pair<std::string, std::string> string_pair;
+
+DEF_PARAM_TEST(ImagePair_Depth_GraySource, string_pair, perf::MatDepth, bool);
+
+PERF_TEST_P(ImagePair_Depth_GraySource, OpticalFlowPyrLKSparse,
+                testing::Combine(
+                    testing::Values(string_pair("im1_1280x800.jpg", "im2_1280x800.jpg")),
+                    testing::Values(CV_8U, CV_16U),
+                    testing::Bool()
+                    ))
+{
+    declare.time(60);
+
+    const string_pair fileNames = std::tr1::get<0>(GetParam());
+    const int depth = std::tr1::get<1>(GetParam());
+    const bool graySource = std::tr1::get<2>(GetParam());
+
+    // PyrLK params
+    const cv::Size winSize(15, 15);
+    const int maxLevel = 5;
+    const cv::TermCriteria criteria(cv::TermCriteria::COUNT + cv::TermCriteria::EPS, 30, 0.01);
+
+    // GoodFeaturesToTrack params
+    const int maxCorners = 5000;
+    const double qualityLevel = 0.05;
+    const int minDistance = 5;
+    const int blockSize = 3;
+    const bool useHarrisDetector = true;
+    const double k = 0.05;
+
+    cv::Mat src1 = cv::imread(fileNames.first, graySource ? cv::IMREAD_GRAYSCALE : cv::IMREAD_COLOR);
+    if (src1.empty())
+        FAIL() << "Unable to load source image [" << fileNames.first << "]";
+
+    cv::Mat src2 = cv::imread(fileNames.second, graySource ? cv::IMREAD_GRAYSCALE : cv::IMREAD_COLOR);
+    if (src2.empty())
+        FAIL() << "Unable to load source image [" << fileNames.second << "]";
+
+    cv::Mat gray_src;
+    if (graySource)
+        gray_src = src1;
+    else
+        cv::cvtColor(src1, gray_src, cv::COLOR_BGR2GRAY);
+
+    cv::Mat pts;
+    cv::goodFeaturesToTrack(gray_src, pts, maxCorners, qualityLevel, minDistance, cv::noArray(), blockSize, useHarrisDetector, k);
+
+    if (depth != CV_8U)
+    {
+        src1.convertTo(src1, depth);
+        src2.convertTo(src2, depth);
+    }
+
+    if (PERF_RUN_GPU())
+    {
+        cv::gpu::GpuMat d_src1(src1);
+        cv::gpu::GpuMat d_src2(src2);
+        cv::gpu::GpuMat d_pts(pts.reshape(2, 1));
+        cv::gpu::GpuMat d_nextPts;
+        cv::gpu::GpuMat d_status;
+
+        cv::gpu::PyrLKOpticalFlow d_pyrLK;
+        d_pyrLK.winSize = winSize;
+        d_pyrLK.maxLevel = maxLevel;
+        d_pyrLK.iters = criteria.maxCount;
+        d_pyrLK.useInitialFlow = false;
+
+        d_pyrLK.sparse(d_src1, d_src2, d_pts, d_nextPts, d_status);
+
+        TEST_CYCLE()
+        {
+            d_pyrLK.sparse(d_src1, d_src2, d_pts, d_nextPts, d_status);
+        }
+    }
+    else
+    {
+        if (depth != CV_8U)
+            FAIL() << "Unsupported depth";
+
+        cv::Mat nextPts;
+        cv::Mat status;
+
+        cv::calcOpticalFlowPyrLK(src1, src2, pts, nextPts, status, cv::noArray(), winSize, maxLevel, criteria);
+
+        TEST_CYCLE()
+        {
+            cv::calcOpticalFlowPyrLK(src1, src2, pts, nextPts, status, cv::noArray(), winSize, maxLevel, criteria);
+        }
+    }
+
+    SANITY_CHECK(0);
+}
+
+//////////////////////////////////////////////////////////
+// OpticalFlowFarneback
+
+DEF_PARAM_TEST(ImagePair_Depth, string_pair, perf::MatDepth);
+
+PERF_TEST_P(ImagePair_Depth, OpticalFlowFarneback,
+                testing::Combine(
+                    testing::Values(string_pair("im1_1280x800.jpg", "im2_1280x800.jpg")),
+                    testing::Values(CV_8U, CV_16U)
+                    ))
+{
+    declare.time(500);
+
+    const string_pair fileNames = std::tr1::get<0>(GetParam());
+    const int depth = std::tr1::get<1>(GetParam());
+
+    const double pyrScale = 0.5;
+    const int numLevels = 6;
+    const int winSize = 7;
+    const int numIters = 15;
+    const int polyN = 7;
+    const double polySigma = 1.5;
+    const int flags = cv::OPTFLOW_USE_INITIAL_FLOW;
+
+    cv::Mat src1 = cv::imread(fileNames.first, cv::IMREAD_GRAYSCALE);
+    if (src1.empty())
+        FAIL() << "Unable to load source image [" << fileNames.first << "]";
+
+    cv::Mat src2 = cv::imread(fileNames.second, cv::IMREAD_GRAYSCALE);
+    if (src2.empty())
+        FAIL() << "Unable to load source image [" << fileNames.second << "]";
+
+    if (depth != CV_8U)
+    {
+        src1.convertTo(src1, depth);
+        src2.convertTo(src2, depth);
+    }
+
+    if (PERF_RUN_GPU())
+    {
+        cv::gpu::GpuMat d_src1(src1);
+        cv::gpu::GpuMat d_src2(src2);
+        cv::gpu::GpuMat d_u(src1.size(), CV_32FC1, cv::Scalar::all(0));
+        cv::gpu::GpuMat d_v(src1.size(), CV_32FC1, cv::Scalar::all(0));
+
+        cv::gpu::FarnebackOpticalFlow d_farneback;
+        d_farneback.pyrScale = pyrScale;
+        d_farneback.numLevels = numLevels;
+        d_farneback.winSize = winSize;
+        d_farneback.numIters = numIters;
+        d_farneback.polyN = polyN;
+        d_farneback.polySigma = polySigma;
+        d_farneback.flags = flags;
+
+        d_farneback(d_src1, d_src2, d_u, d_v);
+
+        TEST_CYCLE_N(10)
+        {
+            d_farneback(d_src1, d_src2, d_u, d_v);
+        }
+    }
+    else
+    {
+        if (depth != CV_8U)
+            FAIL() << "Unsupported depth";
+
+        cv::Mat flow(src1.size(), CV_32FC2, cv::Scalar::all(0));
+
+        cv::calcOpticalFlowFarneback(src1, src2, flow, pyrScale, numLevels, winSize, numIters, polyN, polySigma, flags);
+
+        TEST_CYCLE_N(10)
+        {
+            cv::calcOpticalFlowFarneback(src1, src2, flow, pyrScale, numLevels, winSize, numIters, polyN, polySigma, flags);
+        }
+    }
+
+    SANITY_CHECK(0);
+}
+
+//////////////////////////////////////////////////////////
+// OpticalFlowBM
+
+void calcOpticalFlowBM(const cv::Mat& prev, const cv::Mat& curr,
+                       cv::Size bSize, cv::Size shiftSize, cv::Size maxRange, int usePrevious,
+                       cv::Mat& velx, cv::Mat& vely)
+{
+    cv::Size sz((curr.cols - bSize.width + shiftSize.width)/shiftSize.width, (curr.rows - bSize.height + shiftSize.height)/shiftSize.height);
+
+    velx.create(sz, CV_32FC1);
+    vely.create(sz, CV_32FC1);
+
+    CvMat cvprev = prev;
+    CvMat cvcurr = curr;
+
+    CvMat cvvelx = velx;
+    CvMat cvvely = vely;
+
+    cvCalcOpticalFlowBM(&cvprev, &cvcurr, bSize, shiftSize, maxRange, usePrevious, &cvvelx, &cvvely);
+}
+
+DEF_PARAM_TEST(ImagePair_BlockSize_ShiftSize_MaxRange, string_pair, cv::Size, cv::Size, cv::Size);
+
+PERF_TEST_P(ImagePair_BlockSize_ShiftSize_MaxRange, OpticalFlowBM,
+                testing::Combine(
+                    testing::Values(string_pair("im1_1280x800.jpg", "im2_1280x800.jpg")),
+                    testing::Values(cv::Size(16, 16)),
+                    testing::Values(cv::Size(2, 2)),
+                    testing::Values(cv::Size(16, 16))
+                    ))
+{
+    declare.time(3000);
+
+    const string_pair fileNames = std::tr1::get<0>(GetParam());
+    const cv::Size block_size = std::tr1::get<1>(GetParam());
+    const cv::Size shift_size = std::tr1::get<2>(GetParam());
+    const cv::Size max_range = std::tr1::get<3>(GetParam());
+
+    cv::Mat src1 = cv::imread(fileNames.first, cv::IMREAD_GRAYSCALE);
+    if (src1.empty())
+        FAIL() << "Unable to load source image [" << fileNames.first << "]";
+
+    cv::Mat src2 = cv::imread(fileNames.second, cv::IMREAD_GRAYSCALE);
+    if (src2.empty())
+        FAIL() << "Unable to load source image [" << fileNames.second << "]";
+
+    if (PERF_RUN_GPU())
+    {
+        cv::gpu::GpuMat d_src1(src1);
+        cv::gpu::GpuMat d_src2(src2);
+        cv::gpu::GpuMat d_velx, d_vely, buf;
+
+        cv::gpu::calcOpticalFlowBM(d_src1, d_src2, block_size, shift_size, max_range, false, d_velx, d_vely, buf);
+
+        TEST_CYCLE_N(10)
+        {
+            cv::gpu::calcOpticalFlowBM(d_src1, d_src2, block_size, shift_size, max_range, false, d_velx, d_vely, buf);
+        }
+    }
+    else
+    {
+        cv::Mat velx, vely;
+
+        calcOpticalFlowBM(src1, src2, block_size, shift_size, max_range, false, velx, vely);
+
+        TEST_CYCLE_N(10)
+        {
+            calcOpticalFlowBM(src1, src2, block_size, shift_size, max_range, false, velx, vely);
+        }
+    }
+
+    SANITY_CHECK(0);
+}
+
+PERF_TEST_P(ImagePair_BlockSize_ShiftSize_MaxRange, FastOpticalFlowBM,
+                testing::Combine(
+                    testing::Values(string_pair("im1_1280x800.jpg", "im2_1280x800.jpg")),
+                    testing::Values(cv::Size(16, 16)),
+                    testing::Values(cv::Size(1, 1)),
+                    testing::Values(cv::Size(16, 16))
+                    ))
+{
+    declare.time(3000);
+
+    const string_pair fileNames = std::tr1::get<0>(GetParam());
+    const cv::Size block_size = std::tr1::get<1>(GetParam());
+    const cv::Size shift_size = std::tr1::get<2>(GetParam());
+    const cv::Size max_range = std::tr1::get<3>(GetParam());
+
+    cv::Mat src1 = cv::imread(fileNames.first, cv::IMREAD_GRAYSCALE);
+    if (src1.empty())
+        FAIL() << "Unable to load source image [" << fileNames.first << "]";
+
+    cv::Mat src2 = cv::imread(fileNames.second, cv::IMREAD_GRAYSCALE);
+    if (src2.empty())
+        FAIL() << "Unable to load source image [" << fileNames.second << "]";
+
+    if (PERF_RUN_GPU())
+    {
+        cv::gpu::GpuMat d_src1(src1);
+        cv::gpu::GpuMat d_src2(src2);
+        cv::gpu::GpuMat d_velx, d_vely;
+
+        cv::gpu::FastOpticalFlowBM fastBM;
+
+        fastBM(d_src1, d_src2, d_velx, d_vely, max_range.width, block_size.width);
+
+        TEST_CYCLE_N(10)
+        {
+            fastBM(d_src1, d_src2, d_velx, d_vely, max_range.width, block_size.width);
+        }
+    }
+    else
+    {
+        cv::Mat velx, vely;
+
+        calcOpticalFlowBM(src1, src2, block_size, shift_size, max_range, false, velx, vely);
+
+        TEST_CYCLE_N(10)
+        {
+            calcOpticalFlowBM(src1, src2, block_size, shift_size, max_range, false, velx, vely);
+        }
+    }
+
+    SANITY_CHECK(0);
+}