added first version of gpu performance tests

modules/gpu/src/match_template.cpp

@@ -178,7 +178,7 @@ namespace
             if (depth == CV_8U) return 300;
             break;
         case CV_TM_SQDIFF:
-            if (depth == CV_8U) return 500;
+            if (depth == CV_8U) return 300;
             break;
         }
         CV_Error(CV_StsBadArg, "getTemplateThreshold: unsupported match template mode");

samples/gpu/CMakeLists.txt

@@ -47,7 +47,9 @@ if (BUILD_EXAMPLES)
     foreach(sample_filename ${gpu_samples})
         get_filename_component(sample ${sample_filename} NAME_WE)
         MY_DEFINE_EXAMPLE(${sample}  ${sample_filename})
-    endforeach()
+    endforeach()   
+    
+    include("performance/CMakeLists.txt")
 endif(BUILD_EXAMPLES)
 
 if (NOT WIN32)

samples/gpu/performance/CMakeLists.txt

+set(the_target "example_gpu_performance")
+
+file(GLOB sources "performance/*.cpp")
+file(GLOB headers "performance/*.h")
+
+add_executable(${the_target} ${sources} ${headers})
+
+set_target_properties(${the_target} PROPERTIES
+    OUTPUT_NAME "performance_gpu"
+    PROJECT_LABEL "(EXAMPLE_GPU) performance")
+    
+add_dependencies(${the_target} opencv_core opencv_flann opencv_imgproc opencv_highgui
+    opencv_ml opencv_video opencv_objdetect opencv_features2d
+    opencv_calib3d opencv_legacy opencv_contrib opencv_gpu)
+    
+target_link_libraries(${the_target} ${OPENCV_LINKER_LIBS} opencv_core
+    opencv_flann opencv_imgproc opencv_highgui opencv_ml opencv_video opencv_objdetect
+    opencv_features2d opencv_calib3d opencv_legacy opencv_contrib opencv_gpu)
+
+if(WIN32)
+    install(TARGETS ${the_target} RUNTIME DESTINATION "samples/gpu" COMPONENT main)
+endif()
+
+if(NOT WIN32)
+    file(GLOB GPU_FILES performance/*.cpp performance/*.h)
+    install(FILES ${GPU_FILES}
+            DESTINATION share/opencv/samples/gpu/performance
+            PERMISSIONS OWNER_READ GROUP_READ WORLD_READ)
+endif()
\ No newline at end of file

samples/gpu/performance/performance.cpp

+#include <iomanip>
+#include "performance.h"
+
+using namespace std;
+using namespace cv;
+
+
+void Test::gen(Mat& mat, int rows, int cols, int type, Scalar low, Scalar high)
+{   
+    mat.create(rows, cols, type);
+
+    RNG rng(0);
+    rng.fill(mat, RNG::UNIFORM, low, high);
+}
+
+
+void Test::gen(Mat& mat, int rows, int cols, int type)
+{   
+    mat.create(rows, cols, type);
+
+    Mat mat8u(rows, cols * mat.elemSize(), CV_8U, mat.data, mat.step);
+
+    RNG rng(0);
+    rng.fill(mat, RNG::UNIFORM, Scalar(0), Scalar(256));
+}
+
+
+void TestSystem::run()
+{
+    cout << setiosflags(ios_base::left);
+    cout << "    " << setw(10) << "CPU, ms" << setw(10) << "GPU, ms" 
+        << setw(10) << "SPEEDUP" << "DESCRIPTION\n";
+    cout << resetiosflags(ios_base::left);
+
+    vector<Test*>::iterator it = tests_.begin();
+    for (; it != tests_.end(); ++it)
+    {
+        can_flush_ = false;
+        Test* test = *it;
+
+        cout << endl << test->name() << ":\n";
+        test->run();
+
+        flush();
+    }
+
+    cout << setiosflags(ios_base::fixed | ios_base::left);
+    cout << "\nCPU Total: " << setprecision(3) << cpu_total_ / getTickFrequency() << " sec\n";
+    cout << "GPU Total: " << setprecision(3) << gpu_total_ / getTickFrequency() << " sec (x" 
+        << setprecision(3) << (double)cpu_total_ / gpu_total_ << ")\n";
+    cout << resetiosflags(ios_base::fixed | ios_base::left);
+}
+
+
+void TestSystem::flush()
+{
+    if (!can_flush_)
+        return;
+
+    int cpu_time = static_cast<int>(cpu_elapsed_ / getTickFrequency() * 1000.0);
+    int gpu_time = static_cast<int>(gpu_elapsed_ / getTickFrequency() * 1000.0);
+    double speedup = (double)cpu_time / gpu_time;
+
+    cpu_elapsed_ = 0;
+    gpu_elapsed_ = 0;
+
+    cout << "    " << setiosflags(ios_base::fixed | ios_base::left);
+
+    stringstream stream;
+    stream << cpu_time;
+    cout << setw(10) << stream.str();
+
+    stream.str("");
+    stream << gpu_time;
+    cout << setw(10) << stream.str();
+
+    stream.str("");
+    stream << "x" << setprecision(3) << speedup;
+    cout << setw(10) << stream.str();
+
+    cout << description_.str();
+    description_.str("");
+
+    cout << resetiosflags(ios_base::fixed | ios_base::left) << endl;
+    can_flush_ = false;
+}
+
+
+int main()
+{
+    TestSystem::instance()->run();
+    return 0;
+}
\ No newline at end of file

samples/gpu/performance/performance.h

+#include <iostream>
+#include <cstdio>
+#include <vector>
+#include <string>
+#include <opencv2/core/core.hpp>
+
+class Test
+{
+public:
+    Test(const std::string& name): name_(name) {}
+
+    const std::string& name() const { return name_; }
+
+    void gen(cv::Mat& mat, int rows, int cols, int type,
+             cv::Scalar low, cv::Scalar high);
+
+    void gen(cv::Mat& mat, int rows, int cols, int type);
+
+    virtual void run() = 0;
+
+private:
+    std::string name_;
+};
+
+
+class TestSystem
+{
+public:
+    static TestSystem* instance()
+    {
+        static TestSystem me;
+        return &me;
+    }
+
+    void add(Test* test) { tests_.push_back(test); }
+
+    void run();
+
+    void cpuOn() { cpu_started_ = cv::getTickCount(); }
+
+    void cpuOff() 
+    {
+        int64 delta = cv::getTickCount() - cpu_started_;
+        cpu_elapsed_ += delta;
+        cpu_total_ += delta; 
+        can_flush_ = true;
+    }  
+
+    void gpuOn() { gpu_started_ = cv::getTickCount(); }
+
+    void gpuOff() 
+    {
+        int64 delta = cv::getTickCount() - gpu_started_;
+        gpu_elapsed_ += delta;
+        gpu_total_ += delta; 
+        can_flush_ = true;
+    }
+
+    // Ends current subtest and starts new one
+    std::stringstream& subtest()
+    {
+        flush();
+        return description_;
+    }
+
+private:
+    TestSystem(): can_flush_(false),
+                  cpu_elapsed_(0), cpu_total_(0), 
+                  gpu_elapsed_(0), gpu_total_(0) {};
+
+    void flush();
+
+    std::vector<Test*> tests_;
+
+    // Current test (subtest) description
+    std::stringstream description_;
+
+    bool can_flush_;
+
+    int64 cpu_started_, cpu_elapsed_, cpu_total_;
+    int64 gpu_started_, gpu_elapsed_, gpu_total_;
+};
+
+
+#define TEST(name) \
+    struct name##_test: public Test \
+    { \
+        name##_test(): Test(#name) { TestSystem::instance()->add(this); } \
+        void run(); \
+    } name##_test_instance; \
+    void name##_test::run()
+
+
+#define CPU_ON TestSystem::instance()->cpuOn()
+#define GPU_ON TestSystem::instance()->gpuOn()
+#define CPU_OFF TestSystem::instance()->cpuOff()
+#define GPU_OFF TestSystem::instance()->gpuOff()
+#define SUBTEST TestSystem::instance()->subtest()
+#define DESCRIPTION TestSystem::instance()->subtest()
\ No newline at end of file

samples/gpu/performance/tests.cpp

+#include <opencv2/imgproc/imgproc.hpp>
+#include <opencv2/gpu/gpu.hpp>
+#include "performance.h"
+
+using namespace std;
+using namespace cv;
+
+TEST(matchTemplate)
+{
+    for (int templ_size = 5; templ_size <= 1000; templ_size *= 2)
+    {
+        SUBTEST << "img 3000, templ " << templ_size << ", 8U, SQDIFF";
+
+        Mat image; gen(image, 3000, 3000, CV_8U);
+        Mat templ; gen(templ, templ_size, templ_size, CV_8U);
+        Mat result;
+
+        CPU_ON;
+        matchTemplate(image, templ, result, CV_TM_SQDIFF);
+        CPU_OFF;
+
+        gpu::GpuMat d_image(image);
+        gpu::GpuMat d_templ(templ);
+        gpu::GpuMat d_result;
+
+        GPU_ON;
+        gpu::matchTemplate(d_image, d_templ, d_result, CV_TM_SQDIFF);
+        GPU_OFF;
+    }
+}