added more GPU perf. tests, refactored

samples/gpu/performance/performance.cpp

@@ -9,11 +9,14 @@ void TestSystem::run()
     // Run initializers
     vector<Runnable*>::iterator it = inits_.begin();
     for (; it != inits_.end(); ++it)
+    {
         (*it)->run();
+    }
 
     cout << setiosflags(ios_base::left);
     cout << TAB << setw(10) << "CPU, ms" << setw(10) << "GPU, ms" 
-        << setw(10) << "SPEEDUP" << "DESCRIPTION\n";
+        << setw(10) << "SPEEDUP" 
+        << "DESCRIPTION\n";
     cout << resetiosflags(ios_base::left);
 
     // Run tests
@@ -24,30 +27,23 @@ void TestSystem::run()
         try
         {
             (*it)->run();
-            flush_subtest_data();
+            flushSubtestData();
         }
-        catch (const cv::Exception& e)
+        catch (const cv::Exception&)
         {
-            cout << TAB << "error";
-            switch (e.code)
-            {
-                case CV_StsNoMem: cout << ": out of memory"; break;
-            }
-            if (!description_.str().empty())
-                cout << " [" << description_.str() << "]";
-            cout << endl;
-            reset_subtest_data();
+            resetSubtestData();
         }
     }
 
-    cout << setiosflags(ios_base::fixed | ios_base::left);
-    cout << "\naverage GPU speedup: x" << setprecision(3) 
-        << speedup_total_ / num_subtests_called_ << endl;
-    cout << resetiosflags(ios_base::fixed | ios_base::left);
+    cout << setiosflags(ios_base::fixed);
+    cout << "\naverage GPU speedup: x" 
+        << setprecision(3) << speedup_total_ / num_subtests_called_ 
+        << endl;
+    cout << resetiosflags(ios_base::fixed);
 }
 
 
-void TestSystem::flush_subtest_data()
+void TestSystem::flushSubtestData()
 {
     if (!can_flush_)
         return;
@@ -58,9 +54,10 @@ void TestSystem::flush_subtest_data()
     double speedup = static_cast<double>(cpu_time) / std::max(1, gpu_time);
     speedup_total_ += speedup;
 
-    cout << TAB << setiosflags(ios_base::fixed | ios_base::left);
+    cout << TAB << setiosflags(ios_base::left);
 
     stringstream stream;
+
     stream << cpu_time;
     cout << setw(10) << stream.str();
 
@@ -73,11 +70,10 @@ void TestSystem::flush_subtest_data()
     cout << setw(10) << stream.str();
 
     cout << description_.str();
-
-    cout << resetiosflags(ios_base::fixed | ios_base::left) << endl;
+    cout << resetiosflags(ios_base::left) << endl;
     
     num_subtests_called_++;
-    reset_subtest_data();
+    resetSubtestData();
 }
 
 

samples/gpu/performance/performance.h

@@ -38,6 +38,13 @@ public:
 
     void run();
 
+    // Ends current subtest and starts new one
+    std::stringstream& subtest()
+    {
+        flushSubtestData();
+        return description_;
+    }
+
     void cpuOn() { cpu_started_ = cv::getTickCount(); }
 
     void cpuOff() 
@@ -56,20 +63,13 @@ public:
         can_flush_ = true;
     }
 
-    // Ends current subtest and starts new one
-    std::stringstream& subtest()
-    {
-        flush_subtest_data();
-        return description_;
-    }
-
 private:
     TestSystem(): can_flush_(false), cpu_elapsed_(0), gpu_elapsed_(0), 
                   speedup_total_(0.0), num_subtests_called_(0) {};
 
-    void flush_subtest_data();
+    void flushSubtestData();
 
-    void reset_subtest_data() 
+    void resetSubtestData() 
     {
         cpu_elapsed_ = 0;
         gpu_elapsed_ = 0;
@@ -93,17 +93,6 @@ private:
 };
 
 
-#define TEST(name) \
-    struct name##_test: Runnable \
-    { \
-        name##_test(): Runnable(#name) { \
-            TestSystem::instance()->addTest(this); \
-        } \
-        void run(); \
-    } name##_test_instance; \
-    void name##_test::run()
-
-
 #define INIT(name) \
     struct name##_init: Runnable \
     { \
@@ -115,12 +104,22 @@ private:
     void name##_init::run()
 
 
+#define TEST(name) \
+    struct name##_test: Runnable \
+    { \
+        name##_test(): Runnable(#name) { \
+            TestSystem::instance()->addTest(this); \
+        } \
+        void run(); \
+    } name##_test_instance; \
+    void name##_test::run()
+
+#define SUBTEST TestSystem::instance()->subtest()
+#define DESCRIPTION TestSystem::instance()->subtest()
 #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()
 
 void gen(cv::Mat& mat, int rows, int cols, int type, cv::Scalar low, 
          cv::Scalar high);

samples/gpu/performance/tests.cpp

@@ -6,42 +6,40 @@
 using namespace std;
 using namespace cv;
 
-// This code calls CUFFT DFT and initializes that lib
-INIT(CUFFT_library)
+INIT(matchTemplate)
 {
-    Mat src, templ;
-    gen(src, 500, 500, CV_32F, 0, 1);
-    gen(templ, 500, 500, CV_32F, 0, 1);
+    Mat src; gen(src, 500, 500, CV_32F, 0, 1);
+    Mat templ; gen(templ, 500, 500, CV_32F, 0, 1);
 
-    gpu::GpuMat d_src(src);
-    gpu::GpuMat d_templ(templ);
-    gpu::GpuMat d_result;
+    gpu::GpuMat d_src(src), d_templ(templ), d_dst;
 
-    gpu::matchTemplate(d_src, d_templ, d_result, CV_TM_CCORR);
+    gpu::matchTemplate(d_src, d_templ, d_dst, CV_TM_CCORR);
 }
 
 
 TEST(matchTemplate)
 {
-    Mat src, templ, result;
+    Mat src, templ, dst;
     gen(src, 3000, 3000, CV_32F, 0, 1);
 
-    gpu::GpuMat d_image(src), d_templ, d_result;
+    gpu::GpuMat d_src(src), d_templ, d_dst;
 
-    for (int templ_size = 5; templ_size <= 1000; templ_size *= 2)
+    for (int templ_size = 5; templ_size < 200; templ_size *= 5)
     {
         SUBTEST << "src " << src.rows << ", templ " << templ_size << ", 32F, CCORR";
 
         gen(templ, templ_size, templ_size, CV_32F, 0, 1);
+        dst.create(src.rows - templ.rows + 1, src.cols - templ.cols + 1, CV_32F);
 
         CPU_ON;
-        matchTemplate(src, templ, result, CV_TM_CCORR);
+        matchTemplate(src, templ, dst, CV_TM_CCORR);
         CPU_OFF;
 
         d_templ = templ;
+        d_dst.create(d_src.rows - d_templ.rows + 1, d_src.cols - d_templ.cols + 1, CV_32F);
 
         GPU_ON;
-        gpu::matchTemplate(d_image, d_templ, d_result, CV_TM_CCORR);
+        gpu::matchTemplate(d_src, d_templ, d_dst, CV_TM_CCORR);
         GPU_OFF;
     }
 }
@@ -86,6 +84,7 @@ TEST(remap)
         gen(src, size, size, CV_8UC1, 0, 256);
         gen(xmap, size, size, CV_32F, 0, size);
         gen(ymap, size, size, CV_32F, 0, size);
+        dst.create(xmap.size(), src.type());
 
         CPU_ON;
         remap(src, dst, xmap, ymap, INTER_LINEAR);
@@ -94,6 +93,7 @@ TEST(remap)
         d_src = src;
         d_xmap = xmap;
         d_ymap = ymap;
+        d_dst.create(d_xmap.size(), d_src.type());
 
         GPU_ON;
         gpu::remap(d_src, d_dst, d_xmap, d_ymap);
@@ -107,17 +107,19 @@ TEST(dft)
     Mat src, dst;
     gpu::GpuMat d_src, d_dst;
 
-    for (int size = 1000; size <= 8000; size *= 2)
+    for (int size = 1000; size <= 4000; size *= 2)
     {
         SUBTEST << "size " << size << ", 32FC2, complex-to-complex";
 
         gen(src, size, size, CV_32FC2, Scalar::all(0), Scalar::all(1));
+        dst.create(src.size(), src.type());
 
         CPU_ON;
         dft(src, dst);
         CPU_OFF;
 
         d_src = src;
+        d_dst.create(d_src.size(), d_src.type());
 
         GPU_ON;
         gpu::dft(d_src, d_dst, Size(size, size));
@@ -136,12 +138,14 @@ TEST(cornerHarris)
         SUBTEST << "size " << size << ", 32FC1";
 
         gen(src, size, size, CV_32F, 0, 1);
+        dst.create(src.size(), src.type());
 
         CPU_ON;
         cornerHarris(src, dst, 5, 7, 0.1, BORDER_REFLECT101);
         CPU_OFF;
 
         d_src = src;
+        d_dst.create(src.size(), src.type());
 
         GPU_ON;
         gpu::cornerHarris(d_src, d_dst, 5, 7, 0.1);
@@ -150,22 +154,51 @@ TEST(cornerHarris)
 }
 
 
-TEST(memoryAllocation)
+TEST(integral)
 {
-    Mat mat;
-    gpu::GpuMat d_mat;
+    Mat src, sum;
+    gpu::GpuMat d_src, d_sum;
+
+    for (int size = 1000; size <= 8000; size *= 2)
+    {
+        SUBTEST << "size " << size << ", 8U";
 
-    int begin = 100, end = 8000, step = 100;
+        gen(src, size, size, CV_8U, 0, 256);
+        sum.create(size + 1, size + 1, CV_32S);
 
-    DESCRIPTION << "32F matrices from " << begin << " to " << end;
+        CPU_ON;
+        integral(src, sum);
+        CPU_OFF;
 
-    CPU_ON;
-    for (int size = begin; size <= end; size += step)
-        mat.create(size, size, CV_32FC1);
-    CPU_OFF;
+        d_src = src;
+        d_sum.create(size + 1, size + 1, CV_32S);
 
-    GPU_ON;
-    for (int size = begin; size <= end; size += step)
-        d_mat.create(size, size, CV_32FC1);
-    GPU_OFF;
+        GPU_ON;
+        gpu::integral(d_src, d_sum);
+        GPU_OFF;
+    }
+}
+
+
+TEST(norm)
+{
+    Mat src;
+    gpu::GpuMat d_src;
+
+    for (int size = 1000; size <= 8000; size *= 2)
+    {
+        SUBTEST << "size " << size << ", 8U";
+
+        gen(src, size, size, CV_8U, 0, 256);
+
+        CPU_ON;
+        norm(src);
+        CPU_OFF;
+
+        d_src = src;
+
+        GPU_ON;
+        gpu::norm(d_src);
+        GPU_OFF;
+    }
 }