most of the performance tests was rewritten in google-test manner

modules/ocl/perf/main.cpp

@@ -42,7 +42,7 @@
 
 #include "perf_precomp.hpp"
 
-int main(int argc, const char *argv[])
+static int old_main(int argc, const char *argv[])
 {
     const char *keys =
         "{ h | help    | false | print help message }"
@@ -99,7 +99,7 @@ int main(int argc, const char *argv[])
 
     // 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);
@@ -162,3 +162,33 @@ END_DEV:
 
     return 0;
 }
+
+const char * impls[] =
+{
+    "ocl",
+    "plain",
+#ifdef HAVE_OPENCV_GPU
+    "gpu"
+#endif
+};
+
+int main(int argc, char **argv)
+{
+    // temp solution: if no '--gtest_' and '--perf_' args switch to old behavior
+    bool useGTest = false;
+
+    for(int i=1; i<argc; i++)
+    {
+        std::string arg( argv[i] );
+        if( arg.find("--gtest_")==0 || arg.find("--perf_")==0 )
+            useGTest = true;
+
+//        if (arg == "--perf_verify_sanity")
+//            argv[i] = (char*)"--perf_no_verify_sanity";
+    }
+
+    if( !useGTest )
+        return old_main(argc, (const char**)argv);
+
+    CV_PERF_TEST_MAIN_INTERNALS(ocl, impls)
+}

modules/ocl/perf/perf_arithm.cpp

@@ -45,1153 +45,1081 @@
 //M*/
 
 #include "perf_precomp.hpp"
+
+using namespace perf;
+using std::tr1::get;
+using std::tr1::tuple;
+
 ///////////// Lut ////////////////////////
-PERFTEST(lut)
-{
-    Mat src, lut, dst, ocl_dst;
-    ocl::oclMat d_src, d_lut, d_dst;
 
-    int all_type[] = {CV_8UC1, CV_8UC3};
-    std::string type_name[] = {"CV_8UC1", "CV_8UC3"};
+CV_ENUM(LUTMatTypes, CV_8UC1, CV_8UC3)
 
-    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];
+typedef tuple<Size, LUTMatTypes> LUTParams;
+typedef TestBaseWithParam<LUTParams> LUTFixture;
 
-            gen(src, size, size, all_type[j], 0, 256);
-            gen(lut, 1, 256, CV_8UC1, 0, 1);
+PERF_TEST_P(LUTFixture, LUT,
+          ::testing::Combine(OCL_TYPICAL_MAT_SIZES,
+                             LUTMatTypes::all()))
+{
+    // getting params
+    LUTParams params = GetParam();
+    const Size srcSize = get<0>(params);
+    const int type = get<1>(params);
 
-            LUT(src, lut, dst);
+    const std::string impl = getSelectedImpl();
 
-            CPU_ON;
-            LUT(src, lut, dst);
-            CPU_OFF;
+    // creating src data
+    Mat src(srcSize, type), lut(1, 256, CV_8UC1);
+    int dstType = CV_MAKETYPE(lut.depth(), src.channels());
+    Mat dst(srcSize, dstType);
 
-            d_src.upload(src);
-            d_lut.upload(lut);
+    randu(lut, 0, 2);
+    declare.in(src, WARMUP_RNG).in(lut).out(dst);
 
-            WARMUP_ON;
-            ocl::LUT(d_src, d_lut, d_dst);
-            WARMUP_OFF;
+    // select implementation
+    if (impl == "ocl")
+    {
+        ocl::oclMat oclSrc(src), oclLut(lut), oclDst(srcSize, dstType);
 
-            GPU_ON;
-            ocl::LUT(d_src, d_lut, d_dst);
-            GPU_OFF;
+        TEST_CYCLE() cv::ocl::LUT(oclSrc, oclLut, oclDst);
 
-            GPU_FULL_ON;
-            d_src.upload(src);
-            d_lut.upload(lut);
-            ocl::LUT(d_src, d_lut, d_dst);
-            d_dst.download(ocl_dst);
-            GPU_FULL_OFF;
+        oclDst.download(dst);
 
-            TestSystem::instance().ExpectedMatNear(ocl_dst, dst, 0);
-        }
+        SANITY_CHECK(dst);
+    }
+    else if (impl == "plain")
+    {
+        TEST_CYCLE() cv::LUT(src, lut, dst);
 
+        SANITY_CHECK(dst);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }
 
 ///////////// Exp ////////////////////////
-PERFTEST(Exp)
-{
-    Mat src, dst, ocl_dst;
-    ocl::oclMat d_src, d_dst;
 
-    for (int size = Min_Size; size <= Max_Size; size *= Multiple)
-    {
-        SUBTEST << size << 'x' << size << "; CV_32FC1";
+typedef TestBaseWithParam<Size> ExpFixture;
+
+PERF_TEST_P(ExpFixture, Exp, OCL_TYPICAL_MAT_SIZES)
+{
+    // getting params
+    const Size srcSize = GetParam();
 
-        gen(src, size, size, CV_32FC1, 5, 16);
+    const std::string impl = getSelectedImpl();
 
-        exp(src, dst);
+    // creating src data
+    Mat src(srcSize, CV_32FC1), dst(srcSize, CV_32FC1);
+    declare.in(src).out(dst);
+    randu(src, 5, 16);
 
-        CPU_ON;
-        exp(src, dst);
-        CPU_OFF;
-        d_src.upload(src);
+    // select implementation
+    if (impl == "ocl")
+    {
+        ocl::oclMat oclSrc(src), oclDst(srcSize, src.type());
 
-        WARMUP_ON;
-        ocl::exp(d_src, d_dst);
-        WARMUP_OFF;
+        TEST_CYCLE() cv::ocl::exp(oclSrc, oclDst);
 
-        GPU_ON;
-        ocl::exp(d_src, d_dst);
-        GPU_OFF;
+        oclDst.download(dst);
 
-        GPU_FULL_ON;
-        d_src.upload(src);
-        ocl::exp(d_src, d_dst);
-        d_dst.download(ocl_dst);
-        GPU_FULL_OFF;
+        SANITY_CHECK(dst, 0.3);
+    }
+    else if (impl == "plain")
+    {
+        TEST_CYCLE() cv::exp(src, dst);
 
-        TestSystem::instance().ExpectedMatNear(ocl_dst, dst, 2);
+        SANITY_CHECK(dst, 0.3);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }
 
 ///////////// LOG ////////////////////////
-PERFTEST(Log)
-{
-    Mat src, dst, ocl_dst;
-    ocl::oclMat d_src, d_dst;
 
-    for (int size = Min_Size; size <= Max_Size; size *= Multiple)
-    {
-        SUBTEST << size << 'x' << size << "; 32F";
+typedef TestBaseWithParam<Size> LogFixture;
 
-        gen(src, size, size, CV_32F, 1, 10);
+PERF_TEST_P(LogFixture, Log, OCL_TYPICAL_MAT_SIZES)
+{
+    // getting params
+    const Size srcSize = GetParam();
+    const std::string impl = getSelectedImpl();
 
-        log(src, dst);
+    // creating src data
+    Mat src(srcSize, CV_32F), dst(srcSize, src.type());
+    randu(src, 1, 10);
+    declare.in(src).out(dst);
 
-        CPU_ON;
-        log(src, dst);
-        CPU_OFF;
-        d_src.upload(src);
+    if (srcSize == OCL_SIZE_4000)
+        declare.time(3.6);
 
-        WARMUP_ON;
-        ocl::log(d_src, d_dst);
-        WARMUP_OFF;
+    // select implementation
+    if (impl == "ocl")
+    {
+        ocl::oclMat oclSrc(src), oclDst(srcSize, src.type());
 
-        GPU_ON;
-        ocl::log(d_src, d_dst);
-        GPU_OFF;
+        TEST_CYCLE() cv::ocl::log(oclSrc, oclDst);
 
-        GPU_FULL_ON;
-        d_src.upload(src);
-        ocl::log(d_src, d_dst);
-        d_dst.download(ocl_dst);
-        GPU_FULL_OFF;
+        oclDst.download(dst);
 
-        TestSystem::instance().ExpectedMatNear(ocl_dst, dst, 1);
+        SANITY_CHECK(dst);
     }
-}
-
-///////////// Add ////////////////////////
-PERFTEST(Add)
-{
-    Mat src1, src2, dst, ocl_dst;
-    ocl::oclMat d_src1, d_src2, d_dst;
-
-    int all_type[] = {CV_8UC1, CV_32FC1};
-    std::string type_name[] = {"CV_8UC1", "CV_32FC1"};
-
-    for (int size = Min_Size; size <= Max_Size; size *= Multiple)
+    else if (impl == "plain")
     {
-        for (size_t j = 0; j < sizeof(all_type) / sizeof(int); j++)
-        {
-            SUBTEST << size << 'x' << size << "; " << type_name[j];
+        TEST_CYCLE() cv::log(src, dst);
 
-            gen(src1, size, size, all_type[j], 0, 1);
-            gen(src2, size, size, all_type[j], 0, 1);
+        SANITY_CHECK(dst);
+    }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
+}
 
-            add(src1, src2, dst);
+///////////// Add ////////////////////////
 
-            CPU_ON;
-            add(src1, src2, dst);
-            CPU_OFF;
+CV_ENUM(AddMatTypes, CV_8UC1, CV_32FC1)
 
-            d_src1.upload(src1);
-            d_src2.upload(src2);
+typedef tuple<Size, AddMatTypes> AddParams;
+typedef TestBaseWithParam<AddParams> AddFixture;
 
-            WARMUP_ON;
-            ocl::add(d_src1, d_src2, d_dst);
-            WARMUP_OFF;
+PERF_TEST_P(AddFixture, Add,
+            ::testing::Combine(OCL_TYPICAL_MAT_SIZES,
+                               AddMatTypes::all()))
+{
+    // getting params
+    AddParams params = GetParam();
+    const Size srcSize = get<0>(params);
+    const int type = get<1>(params);
+    const string impl = getSelectedImpl();
+
+    // creating src data
+    Mat src1(srcSize, type), src2(srcSize, type), dst(srcSize, type);
+    randu(src1, 0, 1);
+    randu(src2, 0, 1);
+    declare.in(src1, src2).out(dst);
+
+    // select implementation
+    if (impl == "ocl")
+    {
+        ocl::oclMat oclSrc1(src1), oclSrc2(src2), oclDst(srcSize, type);
 
-            GPU_ON;
-            ocl::add(d_src1, d_src2, d_dst);
-            GPU_OFF;
+        TEST_CYCLE() cv::ocl::add(oclSrc1, oclSrc2, oclDst);
 
-            GPU_FULL_ON;
-            d_src1.upload(src1);
-            d_src2.upload(src2);
-            ocl::add(d_src1, d_src2, d_dst);
-            d_dst.download(ocl_dst);
-            GPU_FULL_OFF;
+        oclDst.download(dst);
 
-            TestSystem::instance().ExpectedMatNear(ocl_dst, dst, 0.0);
-        }
+        SANITY_CHECK(dst);
+    }
+    else if (impl == "plain")
+    {
+        TEST_CYCLE() cv::add(src1, src2, dst);
 
+        SANITY_CHECK(dst);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }
 
 ///////////// Mul ////////////////////////
-PERFTEST(Mul)
-{
-    Mat src1, src2, dst, ocl_dst;
-    ocl::oclMat d_src1, d_src2, 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] ;
-
-            gen(src1, size, size, all_type[j], 0, 256);
-            gen(src2, size, size, all_type[j], 0, 256);
 
-            multiply(src1, src2, dst);
+CV_ENUM(MulMatTypes, CV_8UC1, CV_8UC4)
 
-            CPU_ON;
-            multiply(src1, src2, dst);
-            CPU_OFF;
-            d_src1.upload(src1);
-            d_src2.upload(src2);
+typedef tuple<Size, MulMatTypes> MulParams;
+typedef TestBaseWithParam<MulParams> MulFixture;
 
-            WARMUP_ON;
-            ocl::multiply(d_src1, d_src2, d_dst);
-            WARMUP_OFF;
+PERF_TEST_P(MulFixture, Mul, ::testing::Combine(OCL_TYPICAL_MAT_SIZES,
+                                       MulMatTypes::all()))
+{
+    // getting params
+    MulParams params = GetParam();
+    const Size srcSize = get<0>(params);
+    const int type = get<1>(params);
+    const string impl = getSelectedImpl();
+
+    // creating src data
+    Mat src1(srcSize, type), src2(srcSize, type), dst(srcSize, type);
+    randu(src1, 0, 256);
+    randu(src2, 0, 256);
+    declare.in(src1, src2).out(dst);
+
+    // select implementation
+    if (impl == "ocl")
+    {
+        ocl::oclMat oclSrc1(src1), oclSrc2(src2), oclDst(srcSize, type);
 
-            GPU_ON;
-            ocl::multiply(d_src1, d_src2, d_dst);
-            GPU_OFF;
+        TEST_CYCLE() cv::ocl::multiply(oclSrc1, oclSrc2, oclDst);
 
-            GPU_FULL_ON;
-            d_src1.upload(src1);
-            d_src2.upload(src2);
-            ocl::multiply(d_src1, d_src2, d_dst);
-            d_dst.download(ocl_dst);
-            GPU_FULL_OFF;
+        oclDst.download(dst);
 
-            TestSystem::instance().ExpectedMatNear(ocl_dst, dst, 0.0);
-        }
+        SANITY_CHECK(dst);
+    }
+    else if (impl == "plain")
+    {
+        TEST_CYCLE() cv::multiply(src1, src2, dst);
 
+        SANITY_CHECK(dst);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }
 
 ///////////// Div ////////////////////////
-PERFTEST(Div)
-{
-    Mat src1, src2, dst, ocl_dst;
-    ocl::oclMat d_src1, d_src2, 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];
+typedef MulMatTypes DivMatTypes;
+typedef tuple<Size, DivMatTypes> DivParams;
+typedef TestBaseWithParam<DivParams> DivFixture;
 
-            gen(src1, size, size, all_type[j], 0, 256);
-            gen(src2, size, size, all_type[j], 0, 256);
-
-            divide(src1, src2, dst);
-
-            CPU_ON;
-            divide(src1, src2, dst);
-            CPU_OFF;
-
-            d_src1.upload(src1);
-            d_src2.upload(src2);
-
-            WARMUP_ON;
-            ocl::divide(d_src1, d_src2, d_dst);
-            WARMUP_OFF;
+PERF_TEST_P(DivFixture, Div, ::testing::Combine(OCL_TYPICAL_MAT_SIZES,
+                                       DivMatTypes::all()))
+{
+    // getting params
+    DivParams params = GetParam();
+    const Size srcSize = get<0>(params);
+    const int type = get<1>(params);
+    const string impl = getSelectedImpl();
+
+    // creating src data
+    Mat src1(srcSize, type), src2(srcSize, type), dst(srcSize, type);
+    declare.in(src1, src2).out(dst);
+    randu(src1, 0, 256);
+    randu(src2, 0, 256);
+
+    if ((srcSize == OCL_SIZE_4000 && type == CV_8UC1) ||
+            (srcSize == OCL_SIZE_2000 && type == CV_8UC4))
+        declare.time(4.2);
+    else if (srcSize == OCL_SIZE_4000 && type == CV_8UC4)
+        declare.time(16.6);
+
+    // select implementation
+    if (impl == "ocl")
+    {
+        ocl::oclMat oclSrc1(src1), oclSrc2(src2), oclDst(srcSize, type);
 
-            GPU_ON;
-            ocl::divide(d_src1, d_src2, d_dst);
-            GPU_OFF;
+        TEST_CYCLE() cv::ocl::divide(oclSrc1, oclSrc2, oclDst);
 
-            GPU_FULL_ON;
-            d_src1.upload(src1);
-            d_src2.upload(src2);
-            ocl::divide(d_src1, d_src2, d_dst);
-            d_dst.download(ocl_dst);
-            GPU_FULL_OFF;
+        oclDst.download(dst);
 
-            TestSystem::instance().ExpectedMatNear(ocl_dst, dst, 1);
-        }
+        SANITY_CHECK(dst);
+    }
+    else if (impl == "plain")
+    {
+        TEST_CYCLE() cv::divide(src1, src2, dst);
 
+        SANITY_CHECK(dst);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }
 
 ///////////// Absdiff ////////////////////////
-PERFTEST(Absdiff)
-{
-    Mat src1, src2, dst, ocl_dst;
-    ocl::oclMat d_src1, d_src2, d_dst;
 
-    int all_type[] = {CV_8UC1, CV_8UC4};
-    std::string type_name[] = {"CV_8UC1", "CV_8UC4"};
+typedef MulMatTypes AbsDiffMatTypes;
+typedef tuple<Size, AbsDiffMatTypes> AbsDiffParams;
+typedef TestBaseWithParam<AbsDiffParams> AbsDiffFixture;
 
-    for (int size = Min_Size; size <= Max_Size; size *= Multiple)
+PERF_TEST_P(AbsDiffFixture, Absdiff, ::testing::Combine(OCL_TYPICAL_MAT_SIZES,
+                                               AbsDiffMatTypes::all()))
+{
+    // getting params
+    AbsDiffParams params = GetParam();
+    const Size srcSize = get<0>(params);
+    const int type = get<1>(params);
+    const string impl = getSelectedImpl();
+
+    // creating src data
+    Mat src1(srcSize, type), src2(srcSize, type), dst(srcSize, type);
+    declare.in(src1, src2).in(dst);
+    randu(src1, 0, 256);
+    randu(src2, 0, 256);
+
+    // select implementation
+    if (impl == "ocl")
     {
-        for (size_t j = 0; j < sizeof(all_type) / sizeof(int); j++)
-        {
-            SUBTEST << size << 'x' << size << "; " << type_name[j] ;
-
-            gen(src1, size, size, all_type[j], 0, 256);
-            gen(src2, size, size, all_type[j], 0, 256);
-            gen(dst, size, size, all_type[j], 0, 256);
-
-            absdiff(src1, src2, dst);
-
-            CPU_ON;
-            absdiff(src1, src2, dst);
-            CPU_OFF;
-
-            d_src1.upload(src1);
-            d_src2.upload(src2);
+        ocl::oclMat oclSrc1(src1), oclSrc2(src2), oclDst(srcSize, type);
 
-            WARMUP_ON;
-            ocl::absdiff(d_src1, d_src2, d_dst);
-            WARMUP_OFF;
+        TEST_CYCLE() cv::ocl::absdiff(oclSrc1, oclSrc2, oclDst);
 
-            GPU_ON;
-            ocl::absdiff(d_src1, d_src2, d_dst);
-            GPU_OFF;
+        oclDst.download(dst);
 
-            GPU_FULL_ON;
-            d_src1.upload(src1);
-            d_src2.upload(src2);
-            ocl::absdiff(d_src1, d_src2, d_dst);
-            d_dst.download(ocl_dst);
-            GPU_FULL_OFF;
-
-            TestSystem::instance().ExpectedMatNear(ocl_dst, dst, 0.0);
-        }
+        SANITY_CHECK(dst);
+    }
+    else if (impl == "plain")
+    {
+        TEST_CYCLE() cv::absdiff(src1, src2, dst);
 
+        SANITY_CHECK(dst);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }
 
 ///////////// CartToPolar ////////////////////////
-PERFTEST(CartToPolar)
-{
-    Mat src1, src2, dst, dst1, ocl_dst, ocl_dst1;
-    ocl::oclMat d_src1, d_src2, d_dst, d_dst1;
 
-    int all_type[] = {CV_32FC1};
-    std::string type_name[] = {"CV_32FC1"};
+typedef TestBaseWithParam<Size> CartToPolarFixture;
 
-    for (int size = Min_Size; size <= Max_Size; size *= Multiple)
+PERF_TEST_P(CartToPolarFixture, CartToPolar, OCL_TYPICAL_MAT_SIZES)
+{
+    // getting params
+    const Size srcSize = GetParam();
+    const string impl = getSelectedImpl();
+
+    // creating src data
+    Mat src1(srcSize, CV_32FC1), src2(srcSize, CV_32FC1),
+            dst1(srcSize, CV_32FC1), dst2(srcSize, CV_32FC1);
+    declare.in(src1, src2).out(dst1, dst2);
+    randu(src1, 0, 256);
+    randu(src2, 0, 256);
+
+    if (srcSize == OCL_SIZE_4000)
+        declare.time(3.6);
+
+    // select implementation
+    if (impl == "ocl")
     {
-        for (size_t j = 0; j < sizeof(all_type) / sizeof(int); j++)
-        {
-            SUBTEST << size << 'x' << size << "; " << type_name[j];
-
-            gen(src1, size, size, all_type[j], 0, 256);
-            gen(src2, size, size, all_type[j], 0, 256);
-            gen(dst, size, size, all_type[j], 0, 256);
-            gen(dst1, size, size, all_type[j], 0, 256);
-
+        ocl::oclMat oclSrc1(src1), oclSrc2(src2),
+                oclDst1(srcSize, src1.type()), oclDst2(srcSize, src1.type());
 
-            cartToPolar(src1, src2, dst, dst1, 1);
+        TEST_CYCLE() cv::ocl::cartToPolar(oclSrc1, oclSrc2, oclDst1, oclDst2);
 
-            CPU_ON;
-            cartToPolar(src1, src2, dst, dst1, 1);
-            CPU_OFF;
+        oclDst1.download(dst1);
+        oclDst2.download(dst2);
 
-            d_src1.upload(src1);
-            d_src2.upload(src2);
-
-            WARMUP_ON;
-            ocl::cartToPolar(d_src1, d_src2, d_dst, d_dst1, 1);
-            WARMUP_OFF;
-
-            GPU_ON;
-            ocl::cartToPolar(d_src1, d_src2, d_dst, d_dst1, 1);
-            GPU_OFF;
-
-            GPU_FULL_ON;
-            d_src1.upload(src1);
-            d_src2.upload(src2);
-            ocl::cartToPolar(d_src1, d_src2, d_dst, d_dst1, 1);
-            d_dst.download(ocl_dst);
-            d_dst1.download(ocl_dst1);
-            GPU_FULL_OFF;
-
-            double diff1 = checkNorm(ocl_dst1, dst1);
-            double diff2 = checkNorm(ocl_dst, dst);
-            double max_diff = max(diff1, diff2);
-            TestSystem::instance().setAccurate(max_diff<=.5?1:0, max_diff);
-
-        }
+        SANITY_CHECK(dst1, 5e-3);
+        SANITY_CHECK(dst2, 5e-3);
+    }
+    else if (impl == "plain")
+    {
+        TEST_CYCLE() cv::cartToPolar(src1, src2, dst1, dst2);
 
+        SANITY_CHECK(dst1, 5e-3);
+        SANITY_CHECK(dst2, 5e-3);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }
 
 ///////////// PolarToCart ////////////////////////
-PERFTEST(PolarToCart)
-{
-    Mat src1, src2, dst, dst1, ocl_dst, ocl_dst1;
-    ocl::oclMat d_src1, d_src2, d_dst, d_dst1;
 
-    int all_type[] = {CV_32FC1};
-    std::string type_name[] = {"CV_32FC1"};
+typedef TestBaseWithParam<Size> PolarToCartFixture;
 
-    for (int size = Min_Size; size <= Max_Size; size *= Multiple)
+PERF_TEST_P(PolarToCartFixture, PolarToCart, OCL_TYPICAL_MAT_SIZES)
+{
+    // getting params
+    const Size srcSize = GetParam();
+    const string impl = getSelectedImpl();
+
+    // creating src data
+    Mat src1(srcSize, CV_32FC1), src2(srcSize, CV_32FC1),
+            dst1(srcSize, CV_32FC1), dst2(srcSize, CV_32FC1);
+    declare.in(src1, src2).out(dst1, dst2);
+    randu(src1, 0, 256);
+    randu(src2, 0, 256);
+
+    if (srcSize == OCL_SIZE_4000)
+        declare.time(5.4);
+
+    // select implementation
+    if (impl == "ocl")
     {
-        for (size_t j = 0; j < sizeof(all_type) / sizeof(int); j++)
-        {
-            SUBTEST << size << 'x' << size << "; " << type_name[j] ;
-
-            gen(src1, size, size, all_type[j], 0, 256);
-            gen(src2, size, size, all_type[j], 0, 256);
-            gen(dst, size, size, all_type[j], 0, 256);
-            gen(dst1, size, size, all_type[j], 0, 256);
-
-
-            polarToCart(src1, src2, dst, dst1, 1);
-
-            CPU_ON;
-            polarToCart(src1, src2, dst, dst1, 1);
-            CPU_OFF;
-            d_src1.upload(src1);
-            d_src2.upload(src2);
-
-            WARMUP_ON;
-            ocl::polarToCart(d_src1, d_src2, d_dst, d_dst1, 1);
-            WARMUP_OFF;
+        ocl::oclMat oclSrc1(src1), oclSrc2(src2),
+                oclDst1(srcSize, src1.type()), oclDst2(srcSize, src1.type());
 
-            GPU_ON;
-            ocl::polarToCart(d_src1, d_src2, d_dst, d_dst1, 1);
-            GPU_OFF;
+        TEST_CYCLE() cv::ocl::polarToCart(oclSrc1, oclSrc2, oclDst1, oclDst2);
 
-            GPU_FULL_ON;
-            d_src1.upload(src1);
-            d_src2.upload(src2);
-            ocl::polarToCart(d_src1, d_src2, d_dst, d_dst1, 1);
-            d_dst.download(ocl_dst);
-            d_dst1.download(ocl_dst1);
-            GPU_FULL_OFF;
+        oclDst1.download(dst1);
+        oclDst2.download(dst2);
 
-            double diff1 = checkNorm(ocl_dst1, dst1);
-            double diff2 = checkNorm(ocl_dst, dst);
-            double max_diff = max(diff1, diff2);
-            TestSystem::instance().setAccurate(max_diff<=.5?1:0, max_diff);
-
-        }
+        SANITY_CHECK(dst1, 5e-5);
+        SANITY_CHECK(dst2, 5e-5);
+    }
+    else if (impl == "plain")
+    {
+        TEST_CYCLE() cv::polarToCart(src1, src2, dst1, dst2);
 
+        SANITY_CHECK(dst1, 5e-5);
+        SANITY_CHECK(dst2, 5e-5);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }
 
 ///////////// Magnitude ////////////////////////
-PERFTEST(magnitude)
-{
-    Mat x, y, mag, ocl_mag;
-    ocl::oclMat d_x, d_y, d_mag;
 
-    int all_type[] = {CV_32FC1};
-    std::string type_name[] = {"CV_32FC1"};
+typedef TestBaseWithParam<Size> MagnitudeFixture;
 
-    for (int size = Min_Size; size <= Max_Size; size *= Multiple)
+PERF_TEST_P(MagnitudeFixture, Magnitude, OCL_TYPICAL_MAT_SIZES)
+{
+    // getting params
+    const Size srcSize = GetParam();
+    const string impl = getSelectedImpl();
+
+    // creating src data
+    Mat src1(srcSize, CV_32FC1), src2(srcSize, CV_32FC1),
+            dst(srcSize, CV_32FC1);
+    randu(src1, 0, 1);
+    randu(src2, 0, 1);
+    declare.in(src1, src2).out(dst);
+
+    // select implementation
+    if (impl == "ocl")
     {
-        for (size_t j = 0; j < sizeof(all_type) / sizeof(int); j++)
-        {
-            SUBTEST << size << 'x' << size << "; " << type_name[j];
-
-            gen(x, size, size, all_type[j], 0, 1);
-            gen(y, size, size, all_type[j], 0, 1);
+        ocl::oclMat oclSrc1(src1), oclSrc2(src2),
+                oclDst(srcSize, src1.type());
 
-            magnitude(x, y, mag);
+        TEST_CYCLE() cv::ocl::magnitude(oclSrc1, oclSrc2, oclDst);
 
-            CPU_ON;
-            magnitude(x, y, mag);
-            CPU_OFF;
-            d_x.upload(x);
-            d_y.upload(y);
+        oclDst.download(dst);
 
-            WARMUP_ON;
-            ocl::magnitude(d_x, d_y, d_mag);
-            WARMUP_OFF;
-
-            GPU_ON;
-            ocl::magnitude(d_x, d_y, d_mag);
-            GPU_OFF;
-
-            GPU_FULL_ON;
-            d_x.upload(x);
-            d_y.upload(y);
-            ocl::magnitude(d_x, d_y, d_mag);
-            d_mag.download(ocl_mag);
-            GPU_FULL_OFF;
-
-            TestSystem::instance().ExpectedMatNear(ocl_mag, mag, 1e-5);
-        }
+        SANITY_CHECK(dst, 1e-6);
+    }
+    else if (impl == "plain")
+    {
+        TEST_CYCLE() cv::magnitude(src1, src2, dst);
 
+        SANITY_CHECK(dst, 1e-6);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }
 
 ///////////// Transpose ////////////////////////
-PERFTEST(Transpose)
-{
-    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"};
+typedef MulMatTypes TransposeMatTypes;
+typedef tuple<Size, TransposeMatTypes> TransposeParams;
+typedef TestBaseWithParam<TransposeParams> TransposeFixture;
 
-    for (int size = Min_Size; size <= Max_Size; size *= Multiple)
+PERF_TEST_P(TransposeFixture, Transpose,
+            ::testing::Combine(OCL_TYPICAL_MAT_SIZES,
+                               TransposeMatTypes::all()))
+{
+    // getting params
+    TransposeParams params = GetParam();
+    const Size srcSize = get<0>(params);
+    const int type = get<1>(params);
+    const string impl = getSelectedImpl();
+
+    // creating src data
+    Mat src(srcSize, type), dst(srcSize, type);
+    declare.in(src, WARMUP_RNG).out(dst);
+
+    // select implementation
+    if (impl == "ocl")
     {
-        for (size_t j = 0; j < sizeof(all_type) / sizeof(int); j++)
-        {
-            SUBTEST << size << 'x' << size << "; " << type_name[j];
+        ocl::oclMat oclSrc(src), oclDst(srcSize, type);
 
-            gen(src, size, size, all_type[j], 0, 256);
-            gen(dst, size, size, all_type[j], 0, 256);
+        TEST_CYCLE() cv::ocl::transpose(oclSrc, oclDst);
 
-            transpose(src, dst);
+        oclDst.download(dst);
 
-            CPU_ON;
-            transpose(src, dst);
-            CPU_OFF;
-            d_src.upload(src);
-
-            WARMUP_ON;
-            ocl::transpose(d_src, d_dst);
-            WARMUP_OFF;
-
-            GPU_ON;
-            ocl::transpose(d_src, d_dst);
-            GPU_OFF;
-
-            GPU_FULL_ON;
-            d_src.upload(src);
-            ocl::transpose(d_src, d_dst);
-            d_dst.download(ocl_dst);
-            GPU_FULL_OFF;
-
-            TestSystem::instance().ExpectedMatNear(ocl_dst, dst, 1e-5);
-        }
+        SANITY_CHECK(dst);
+    }
+    else if (impl == "plain")
+    {
+        TEST_CYCLE() cv::transpose(src, dst);
 
+        SANITY_CHECK(dst);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }
 
 ///////////// Flip ////////////////////////
-PERFTEST(Flip)
-{
-    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"};
+typedef MulMatTypes FlipMatTypes;
+typedef tuple<Size, FlipMatTypes> FlipParams;
+typedef TestBaseWithParam<FlipParams> FlipFixture;
 
-    for (int size = Min_Size; size <= Max_Size; size *= Multiple)
+PERF_TEST_P(FlipFixture, Flip,
+            ::testing::Combine(OCL_TYPICAL_MAT_SIZES,
+                                         FlipMatTypes::all()))
+{
+    // getting params
+    TransposeParams params = GetParam();
+    const Size srcSize = get<0>(params);
+    const int type = get<1>(params);
+    const string impl = getSelectedImpl();
+
+    // creating src data
+    Mat src(srcSize, type), dst(srcSize, type);
+    declare.in(src, WARMUP_RNG).out(dst);
+
+    // select implementation
+    if (impl == "ocl")
     {
-        for (size_t j = 0; j < sizeof(all_type) / sizeof(int); j++)
-        {
-            SUBTEST << size << 'x' << size << "; " << type_name[j] << " ; FLIP_BOTH";
-
-            gen(src, size, size, all_type[j], 0, 256);
-            gen(dst, size, size, all_type[j], 0, 256);
-
-            flip(src, dst, 0);
+        ocl::oclMat oclSrc(src), oclDst(srcSize, type);
 
-            CPU_ON;
-            flip(src, dst, 0);
-            CPU_OFF;
-            d_src.upload(src);
+        TEST_CYCLE() cv::ocl::flip(oclSrc, oclDst, 0);
 
-            WARMUP_ON;
-            ocl::flip(d_src, d_dst, 0);
-            WARMUP_OFF;
+        oclDst.download(dst);
 
-            GPU_ON;
-            ocl::flip(d_src, d_dst, 0);
-            GPU_OFF;
-
-            GPU_FULL_ON;
-            d_src.upload(src);
-            ocl::flip(d_src, d_dst, 0);
-            d_dst.download(ocl_dst);
-            GPU_FULL_OFF;
-
-            TestSystem::instance().ExpectedMatNear(ocl_dst, dst, 1e-5);
-        }
+        SANITY_CHECK(dst);
+    }
+    else if (impl == "plain")
+    {
+        TEST_CYCLE() cv::flip(src, dst, 0);
 
+        SANITY_CHECK(dst);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }
 
 ///////////// minMax ////////////////////////
-PERFTEST(minMax)
-{
-    Mat src;
-    ocl::oclMat d_src;
-
-    double min_val = 0.0, max_val = 0.0;
-    double min_val_ = 0.0, max_val_ = 0.0;
-    Point min_loc, max_loc;
-    int all_type[] = {CV_8UC1, CV_32FC1};
-    std::string type_name[] = {"CV_8UC1", "CV_32FC1"};
 
-    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];
+typedef AddMatTypes minMaxMatTypes;
+typedef tuple<Size, minMaxMatTypes> minMaxParams;
+typedef TestBaseWithParam<minMaxParams> minMaxFixture;
 
-            gen(src, size, size, all_type[j], 0, 256);
+PERF_TEST_P(minMaxFixture, minMax,
+            ::testing::Combine(OCL_TYPICAL_MAT_SIZES,
+                               minMaxMatTypes::all()))
+{
+    // getting params
+    minMaxParams params = GetParam();
+    const Size srcSize = get<0>(params);
+    const int type = get<1>(params);
+    const string impl = getSelectedImpl();
 
-            CPU_ON;
-            minMaxLoc(src, &min_val, &max_val, &min_loc, &max_loc);
-            CPU_OFF;
-            d_src.upload(src);
+    // creating src data
+    Mat src(srcSize, type);
+    declare.in(src, WARMUP_RNG);
 
-            WARMUP_ON;
-            ocl::minMax(d_src, &min_val_, &max_val_);
-            WARMUP_OFF;
+    double min_val = 0.0, max_val = 0.0;
 
-            if(EeceptDoubleEQ<double>(max_val_, max_val) && EeceptDoubleEQ<double>(min_val_, min_val))
-                TestSystem::instance().setAccurate(1, max(fabs(max_val_-max_val), fabs(min_val_-min_val)));
-            else
-                TestSystem::instance().setAccurate(0, max(fabs(max_val_-max_val), fabs(min_val_-min_val)));
+    // select implementation
+    if (impl == "ocl")
+    {
+        ocl::oclMat oclSrc(src);
 
-            GPU_ON;
-            ocl::minMax(d_src, &min_val, &max_val);
-            GPU_OFF;
+        TEST_CYCLE() cv::ocl::minMax(oclSrc, &min_val, &max_val);
 
-            GPU_FULL_ON;
-            d_src.upload(src);
-            ocl::minMax(d_src, &min_val, &max_val);
-            GPU_FULL_OFF;
+        ASSERT_GE(max_val, min_val);
+        SANITY_CHECK(min_val);
+        SANITY_CHECK(max_val);
+    }
+    else if (impl == "plain")
+    {
+        Point min_loc, max_loc;
 
-        }
+        TEST_CYCLE() cv::minMaxLoc(src, &min_val, &max_val, &min_loc, &max_loc);
 
+        ASSERT_GE(max_val, min_val);
+        SANITY_CHECK(min_val);
+        SANITY_CHECK(max_val);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }
 
 ///////////// minMaxLoc ////////////////////////
-PERFTEST(minMaxLoc)
+
+typedef AddMatTypes minMaxLocMatTypes;
+typedef tuple<Size, minMaxMatTypes> minMaxLocParams;
+typedef TestBaseWithParam<minMaxLocParams> minMaxLocFixture;
+
+PERF_TEST_P(minMaxLocFixture, minMaxLoc,
+            ::testing::Combine(OCL_TYPICAL_MAT_SIZES,
+                               minMaxLocMatTypes::all()))
 {
-    Mat src;
-    ocl::oclMat d_src;
+    // getting params
+    minMaxLocParams params = GetParam();
+    const Size srcSize = get<0>(params);
+    const int type = get<1>(params);
+    const string impl = getSelectedImpl();
+
+    // creating src data
+    Mat src(srcSize, type);
+    randu(src, 0, 1);
+    declare.in(src);
 
     double min_val = 0.0, max_val = 0.0;
-    double min_val_ = 0.0, max_val_ = 0.0;
     Point min_loc, max_loc;
-    Point min_loc_, max_loc_;
-    int all_type[] = {CV_8UC1, CV_32FC1};
-    std::string type_name[] = {"CV_8UC1", "CV_32FC1"};
 
-    for (int size = Min_Size; size <= Max_Size; size *= Multiple)
+    // select implementation
+    if (impl == "ocl")
     {
-        for (size_t j = 0; j < sizeof(all_type) / sizeof(int); j++)
-        {
-            SUBTEST << size << 'x' << size << "; " << type_name[j] ;
+        ocl::oclMat oclSrc(src);
 
-            gen(src, size, size, all_type[j], 0, 1);
+        TEST_CYCLE() cv::ocl::minMaxLoc(oclSrc, &min_val, &max_val, &min_loc, &max_loc);
 
-            CPU_ON;
-            minMaxLoc(src, &min_val, &max_val, &min_loc, &max_loc);
-            CPU_OFF;
-            d_src.upload(src);
-
-            WARMUP_ON;
-            ocl::minMaxLoc(d_src, &min_val_, &max_val_, &min_loc_, &max_loc_);
-            WARMUP_OFF;
-
-            double error0 = 0., error1 = 0., minlocVal = 0., minlocVal_ = 0., maxlocVal = 0., maxlocVal_ = 0.;
-            if(src.depth() == 0)
-            {
-                minlocVal = src.at<unsigned char>(min_loc);
-                minlocVal_ = src.at<unsigned char>(min_loc_);
-                maxlocVal = src.at<unsigned char>(max_loc);
-                maxlocVal_ = src.at<unsigned char>(max_loc_);
-            }
-            if(src.depth() == 1)
-            {
-                minlocVal = src.at<signed char>(min_loc);
-                minlocVal_ = src.at<signed char>(min_loc_);
-                maxlocVal = src.at<signed char>(max_loc);
-                maxlocVal_ = src.at<signed char>(max_loc_);
-            }
-            if(src.depth() == 2)
-            {
-                minlocVal = src.at<unsigned short>(min_loc);
-                minlocVal_ = src.at<unsigned short>(min_loc_);
-                maxlocVal = src.at<unsigned short>(max_loc);
-                maxlocVal_ = src.at<unsigned short>(max_loc_);
-            }
-            if(src.depth() == 3)
-            {
-                minlocVal = src.at<signed short>(min_loc);
-                minlocVal_ = src.at<signed short>(min_loc_);
-                maxlocVal = src.at<signed short>(max_loc);
-                maxlocVal_ = src.at<signed short>(max_loc_);
-            }
-            if(src.depth() == 4)
-            {
-                minlocVal = src.at<int>(min_loc);
-                minlocVal_ = src.at<int>(min_loc_);
-                maxlocVal = src.at<int>(max_loc);
-                maxlocVal_ = src.at<int>(max_loc_);
-            }
-            if(src.depth() == 5)
-            {
-                minlocVal = src.at<float>(min_loc);
-                minlocVal_ = src.at<float>(min_loc_);
-                maxlocVal = src.at<float>(max_loc);
-                maxlocVal_ = src.at<float>(max_loc_);
-            }
-            if(src.depth() == 6)
-            {
-                minlocVal = src.at<double>(min_loc);
-                minlocVal_ = src.at<double>(min_loc_);
-                maxlocVal = src.at<double>(max_loc);
-                maxlocVal_ = src.at<double>(max_loc_);
-            }
-            error0 = ::abs(minlocVal_ - minlocVal);
-            error1 = ::abs(maxlocVal_ - maxlocVal);
-            if( EeceptDoubleEQ<double>(maxlocVal_, maxlocVal)
-                &&EeceptDoubleEQ<double>(minlocVal_, minlocVal)
-                &&EeceptDoubleEQ<double>(max_val_, max_val)
-                &&EeceptDoubleEQ<double>(min_val_, min_val))
-                TestSystem::instance().setAccurate(1, 0.);
-            else
-                TestSystem::instance().setAccurate(0, max(error0, error1));
-
-            GPU_ON;
-            ocl::minMaxLoc(d_src, &min_val, &max_val, &min_loc, &max_loc);
-            GPU_OFF;
-
-            GPU_FULL_ON;
-            d_src.upload(src);
-            ocl::minMaxLoc(d_src, &min_val, &max_val, &min_loc, &max_loc);
-            GPU_FULL_OFF;
-        }
+        ASSERT_GE(max_val, min_val);
+        SANITY_CHECK(min_val);
+        SANITY_CHECK(max_val);
+    }
+    else if (impl == "plain")
+    {
+        TEST_CYCLE() cv::minMaxLoc(src, &min_val, &max_val, &min_loc, &max_loc);
 
+        ASSERT_GE(max_val, min_val);
+        SANITY_CHECK(min_val);
+        SANITY_CHECK(max_val);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }
 
 ///////////// Sum ////////////////////////
-PERFTEST(Sum)
-{
-    Mat src;
-    Scalar cpures, gpures;
-    ocl::oclMat d_src;
-
-    int all_type[] = {CV_8UC1, CV_32SC1};
-    std::string type_name[] = {"CV_8UC1", "CV_32SC1"};
-
-    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] ;
 
-            gen(src, size, size, all_type[j], 0, 60);
+CV_ENUM(SumMatTypes, CV_8UC1, CV_32SC1)
 
-            cpures = sum(src);
+typedef tuple<Size, SumMatTypes> SumParams;
+typedef TestBaseWithParam<SumParams> SumFixture;
 
-            CPU_ON;
-            cpures = sum(src);
-            CPU_OFF;
-            d_src.upload(src);
-
-            WARMUP_ON;
-            gpures = ocl::sum(d_src);
-            WARMUP_OFF;
-
-            vector<double> diffs(4);
-            diffs[3] = fabs(cpures[3] - gpures[3]);
-            diffs[2] = fabs(cpures[2] - gpures[2]);
-            diffs[1] = fabs(cpures[1] - gpures[1]);
-            diffs[0] = fabs(cpures[0] - gpures[0]);
-            double max_diff = *max_element(diffs.begin(), diffs.end());
-            TestSystem::instance().setAccurate(max_diff<0.1?1:0, max_diff);
+PERF_TEST_P(SumFixture, Sum,
+            ::testing::Combine(OCL_TYPICAL_MAT_SIZES,
+                               SumMatTypes::all()))
+{
+    // getting params
+    SumParams params = GetParam();
+    const Size srcSize = get<0>(params);
+    const int type = get<1>(params);
+    const string impl = getSelectedImpl();
+
+    // creating src data
+    Mat src(srcSize, type);
+    Scalar result;
+    randu(src, 0, 60);
+    declare.in(src);
+
+    // select implementation
+    if (impl == "ocl")
+    {
+        ocl::oclMat oclSrc(src);
 
-            GPU_ON;
-            gpures = ocl::sum(d_src);
-            GPU_OFF;
+        TEST_CYCLE() result = cv::ocl::sum(oclSrc);
 
-            GPU_FULL_ON;
-            d_src.upload(src);
-            gpures = ocl::sum(d_src);
-            GPU_FULL_OFF;
-        }
+        SANITY_CHECK(result);
+    }
+    else if (impl == "plain")
+    {
+        TEST_CYCLE() result = cv::sum(src);
 
+        SANITY_CHECK(result);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }
 
 ///////////// countNonZero ////////////////////////
-PERFTEST(countNonZero)
-{
-    Mat src;
-    ocl::oclMat d_src;
 
-    int all_type[] = {CV_8UC1, CV_32FC1};
-    std::string type_name[] = {"CV_8UC1", "CV_32FC1"};
+CV_ENUM(countNonZeroMatTypes, CV_8UC1, CV_32FC1)
 
-    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] ;
-
-            gen(src, size, size, all_type[j], 0, 256);
-
-            countNonZero(src);
-
-            int cpures = 0, gpures = 0;
-            CPU_ON;
-            cpures = countNonZero(src);
-            CPU_OFF;
-            d_src.upload(src);
+typedef tuple<Size, countNonZeroMatTypes> countNonZeroParams;
+typedef TestBaseWithParam<countNonZeroParams> countNonZeroFixture;
 
-            WARMUP_ON;
-            gpures = ocl::countNonZero(d_src);
-            WARMUP_OFF;
-
-            int diff = abs(cpures - gpures);
-            if(diff == 0)
-                TestSystem::instance().setAccurate(1, 0);
-            else
-                TestSystem::instance().setAccurate(0, diff);
+PERF_TEST_P(countNonZeroFixture, countNonZero,
+            ::testing::Combine(OCL_TYPICAL_MAT_SIZES,
+                               countNonZeroMatTypes::all()))
+{
+    // getting params
+    countNonZeroParams params = GetParam();
+    const Size srcSize = get<0>(params);
+    const int type = get<1>(params);
+    const string impl = getSelectedImpl();
+
+    // creating src data
+    Mat src(srcSize, type);
+    int result = 0;
+    randu(src, 0, 256);
+    declare.in(src);
+
+    // select implementation
+    if (impl == "ocl")
+    {
+        ocl::oclMat oclSrc(src);
 
-            GPU_ON;
-            ocl::countNonZero(d_src);
-            GPU_OFF;
+        TEST_CYCLE() result = cv::ocl::countNonZero(oclSrc);
 
-            GPU_FULL_ON;
-            d_src.upload(src);
-            ocl::countNonZero(d_src);
-            GPU_FULL_OFF;
-        }
+        SANITY_CHECK(result);
+    }
+    else if (impl == "plain")
+    {
+        TEST_CYCLE() result = cv::countNonZero(src);
 
+        SANITY_CHECK(result);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }
 
 ///////////// Phase ////////////////////////
-PERFTEST(Phase)
-{
-    Mat src1, src2, dst, ocl_dst;
-    ocl::oclMat d_src1, d_src2, d_dst;
 
-    int all_type[] = {CV_32FC1};
-    std::string type_name[] = {"CV_32FC1"};
+typedef TestBaseWithParam<Size> PhaseFixture;
 
-    for (int size = Min_Size; size <= Max_Size; size *= Multiple)
+PERF_TEST_P(PhaseFixture, Phase, OCL_TYPICAL_MAT_SIZES)
+{
+    // getting params
+    const Size srcSize = GetParam();
+    const string impl = getSelectedImpl();
+
+    // creating src data
+    Mat src1(srcSize, CV_32FC1), src2(srcSize, CV_32FC1),
+            dst(srcSize, CV_32FC1);
+    declare.in(src1, src2).out(dst);
+    randu(src1, 0, 256);
+    randu(src2, 0, 256);
+
+    // select implementation
+    if (impl == "ocl")
     {
-        for (size_t j = 0; j < sizeof(all_type) / sizeof(int); j++)
-        {
-            SUBTEST << size << 'x' << size << "; " << type_name[j] ;
-
-            gen(src1, size, size, all_type[j], 0, 256);
-            gen(src2, size, size, all_type[j], 0, 256);
-            gen(dst, size, size, all_type[j], 0, 256);
-
-            phase(src1, src2, dst, 1);
-
-            CPU_ON;
-            phase(src1, src2, dst, 1);
-            CPU_OFF;
+        ocl::oclMat oclSrc1(src1), oclSrc2(src2),
+                oclDst(srcSize, src1.type());
 
-            d_src1.upload(src1);
-            d_src2.upload(src2);
+        TEST_CYCLE() cv::ocl::phase(oclSrc1, oclSrc2, oclDst, 1);
 
-            WARMUP_ON;
-            ocl::phase(d_src1, d_src2, d_dst, 1);
-            WARMUP_OFF;
+        oclDst.download(dst);
 
-            GPU_ON;
-            ocl::phase(d_src1, d_src2, d_dst, 1);
-            GPU_OFF;
-
-            GPU_FULL_ON;
-            d_src1.upload(src1);
-            d_src2.upload(src2);
-            ocl::phase(d_src1, d_src2, d_dst, 1);
-            d_dst.download(ocl_dst);
-            GPU_FULL_OFF;
-
-            TestSystem::instance().ExpectedMatNear(ocl_dst, dst, 1e-2);
-        }
+        SANITY_CHECK(dst, 1e-2);
+    }
+    else if (impl == "plain")
+    {
+        TEST_CYCLE() cv::phase(src1, src2, dst, 1);
 
+        SANITY_CHECK(dst, 1e-2);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }
 
 ///////////// bitwise_and////////////////////////
-PERFTEST(bitwise_and)
-{
-    Mat src1, src2, dst, ocl_dst;
-    ocl::oclMat d_src1, d_src2, d_dst;
 
-    int all_type[] = {CV_8UC1, CV_32SC1};
-    std::string type_name[] = {"CV_8UC1", "CV_32SC1"};
+typedef SumMatTypes BitwiseAndMatTypes;
+typedef tuple<Size, BitwiseAndMatTypes> BitwiseAndParams;
+typedef TestBaseWithParam<BitwiseAndParams> BitwiseAndFixture;
 
-    for (int size = Min_Size; size <= Max_Size; size *= Multiple)
+PERF_TEST_P(BitwiseAndFixture, bitwise_and,
+            ::testing::Combine(OCL_TYPICAL_MAT_SIZES,
+                               BitwiseAndMatTypes::all()))
+{
+    // getting params
+    BitwiseAndParams params = GetParam();
+    const Size srcSize = get<0>(params);
+    const int type = get<1>(params);
+    const string impl = getSelectedImpl();
+
+    // creating src data
+    Mat src1(srcSize, type), src2(srcSize, type), dst(srcSize, type);
+    declare.in(src1, src2).out(dst);
+    randu(src1, 0, 256);
+    randu(src2, 0, 256);
+
+    // select implementation
+    if (impl == "ocl")
     {
-        for (size_t j = 0; j < sizeof(all_type) / sizeof(int); j++)
-        {
-            SUBTEST << size << 'x' << size << "; " << type_name[j] ;
+        ocl::oclMat oclSrc1(src1), oclSrc2(src2), oclDst(srcSize, src1.type());
 
-            gen(src1, size, size, all_type[j], 0, 256);
-            gen(src2, size, size, all_type[j], 0, 256);
-            gen(dst, size, size, all_type[j], 0, 256);
+        TEST_CYCLE() cv::ocl::bitwise_and(oclSrc1, oclSrc2, oclDst);
 
-            bitwise_and(src1, src2, dst);
+        oclDst.download(dst);
 
-            CPU_ON;
-            bitwise_and(src1, src2, dst);
-            CPU_OFF;
-            d_src1.upload(src1);
-            d_src2.upload(src2);
-
-            WARMUP_ON;
-            ocl::bitwise_and(d_src1, d_src2, d_dst);
-            WARMUP_OFF;
-
-            GPU_ON;
-            ocl::bitwise_and(d_src1, d_src2, d_dst);
-            GPU_OFF;
-
-            GPU_FULL_ON;
-            d_src1.upload(src1);
-            d_src2.upload(src2);
-            ocl::bitwise_and(d_src1, d_src2, d_dst);
-            d_dst.download(ocl_dst);
-            GPU_FULL_OFF;
-
-            TestSystem::instance().ExpectedMatNear(ocl_dst, dst, 0.0);
-        }
+        SANITY_CHECK(dst);
+    }
+    else if (impl == "plain")
+    {
+        TEST_CYCLE() cv::bitwise_and(src1, src2, dst);
 
+        SANITY_CHECK(dst);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }
 
 ///////////// bitwise_not////////////////////////
-PERFTEST(bitwise_not)
-{
-    Mat src1, dst, ocl_dst;
-    ocl::oclMat d_src1, d_dst;
 
-    int all_type[] = {CV_8UC1, CV_32SC1};
-    std::string type_name[] = {"CV_8UC1", "CV_32SC1"};
+typedef SumMatTypes BitwiseNotMatTypes;
+typedef tuple<Size, BitwiseNotMatTypes> BitwiseNotParams;
+typedef TestBaseWithParam<BitwiseNotParams> BitwiseNotFixture;
 
-    for (int size = Min_Size; size <= Max_Size; size *= Multiple)
+PERF_TEST_P(BitwiseAndFixture, bitwise_not,
+            ::testing::Combine(OCL_TYPICAL_MAT_SIZES,
+                               BitwiseAndMatTypes::all()))
+{
+    // getting params
+    BitwiseNotParams params = GetParam();
+    const Size srcSize = get<0>(params);
+    const int type = get<1>(params);
+    const string impl = getSelectedImpl();
+
+    // creating src data
+    Mat src(srcSize, type), dst(srcSize, type);
+    declare.in(src, WARMUP_RNG).out(dst);
+
+    // select implementation
+    if (impl == "ocl")
     {
-        for (size_t j = 0; j < sizeof(all_type) / sizeof(int); j++)
-        {
-            SUBTEST << size << 'x' << size << "; " << type_name[j] ;
-
-            gen(src1, size, size, all_type[j], 0, 256);
-            gen(dst, size, size, all_type[j], 0, 256);
-
-            bitwise_not(src1, dst);
-
-            CPU_ON;
-            bitwise_not(src1, dst);
-            CPU_OFF;
-            d_src1.upload(src1);
+        ocl::oclMat oclSrc(src), oclDst(srcSize, type);
 
-            WARMUP_ON;
-            ocl::bitwise_not(d_src1, d_dst);
-            WARMUP_OFF;
+        TEST_CYCLE() cv::ocl::bitwise_not(oclSrc, oclDst);
 
-            GPU_ON;
-            ocl::bitwise_not(d_src1, d_dst);
-            GPU_OFF;
+        oclDst.download(dst);
 
-            GPU_FULL_ON;
-            d_src1.upload(src1);
-            ocl::bitwise_not(d_src1, d_dst);
-            d_dst.download(ocl_dst);
-            GPU_FULL_OFF;
-
-            TestSystem::instance().ExpectedMatNear(ocl_dst, dst, 0.0);
-        }
+        SANITY_CHECK(dst);
+    }
+    else if (impl == "plain")
+    {
+        TEST_CYCLE() cv::bitwise_not(src, dst);
 
+        SANITY_CHECK(dst);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }
 
 ///////////// compare////////////////////////
-PERFTEST(compare)
-{
-    Mat src1, src2, dst, ocl_dst;
-    ocl::oclMat d_src1, d_src2, d_dst;
 
-    int CMP_EQ = 0;
-    int all_type[] = {CV_8UC1, CV_32FC1};
-    std::string type_name[] = {"CV_8UC1", "CV_32FC1"};
+typedef countNonZeroMatTypes CompareMatTypes;
+typedef tuple<Size, CompareMatTypes> CompareParams;
+typedef TestBaseWithParam<CompareParams> CompareFixture;
 
-    for (int size = Min_Size; size <= Max_Size; size *= Multiple)
+PERF_TEST_P(CompareFixture, compare,
+            ::testing::Combine(OCL_TYPICAL_MAT_SIZES,
+                               CompareMatTypes::all()))
+{
+    // getting params
+    CompareParams params = GetParam();
+    const Size srcSize = get<0>(params);
+    const int type = get<1>(params);
+    const string impl = getSelectedImpl();
+
+    // creating src data
+    Mat src1(srcSize, type), src2(srcSize, type), dst(srcSize, CV_8UC1);
+    declare.in(src1, src2, WARMUP_RNG).out(dst);
+
+    // select implementation
+    if (impl == "ocl")
     {
-        for (size_t j = 0; j < sizeof(all_type) / sizeof(int); j++)
-        {
-            SUBTEST << size << 'x' << size << "; " << type_name[j] ;
-
-            gen(src1, size, size, all_type[j], 0, 256);
-            gen(src2, size, size, all_type[j], 0, 256);
-            gen(dst, size, size, all_type[j], 0, 256);
-
-            compare(src1, src2, dst, CMP_EQ);
-
-            CPU_ON;
-            compare(src1, src2, dst, CMP_EQ);
-            CPU_OFF;
-
-            d_src1.upload(src1);
-            d_src2.upload(src2);
-
-            WARMUP_ON;
-            ocl::compare(d_src1, d_src2, d_dst, CMP_EQ);
-            WARMUP_OFF;
+        ocl::oclMat oclSrc1(src1), oclSrc2(src2), oclDst(srcSize, CV_8UC1);
 
-            GPU_ON;
-            ocl::compare(d_src1, d_src2, d_dst, CMP_EQ);
-            GPU_OFF;
+        TEST_CYCLE() cv::ocl::compare(oclSrc1, oclSrc2, oclDst, CMP_EQ);
 
-            GPU_FULL_ON;
-            d_src1.upload(src1);
-            d_src2.upload(src2);
-            ocl::compare(d_src1, d_src2, d_dst, CMP_EQ);
-            d_dst.download(ocl_dst);
-            GPU_FULL_OFF;
+        oclDst.download(dst);
 
-            TestSystem::instance().ExpectedMatNear(ocl_dst, dst, 0.0);
-        }
+        SANITY_CHECK(dst);
+    }
+    else if (impl == "plain")
+    {
+        TEST_CYCLE() cv::compare(src1, src2, dst, CMP_EQ);
 
+        SANITY_CHECK(dst);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }
 
 ///////////// pow ////////////////////////
-PERFTEST(pow)
-{
-    Mat src, dst, ocl_dst;
-    ocl::oclMat d_src, d_dst;
-
-    int all_type[] = {CV_32FC1};
-    std::string type_name[] = {"CV_32FC1"};
-
-    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] ;
 
-            gen(src, size, size, all_type[j], 5, 16);
+typedef TestBaseWithParam<Size> PowFixture;
 
-            pow(src, -2.0, dst);
+PERF_TEST_P(PowFixture, pow, OCL_TYPICAL_MAT_SIZES)
+{
+    // getting params
+    const Size srcSize = GetParam();
+    const string impl = getSelectedImpl();
 
-            CPU_ON;
-            pow(src, -2.0, dst);
-            CPU_OFF;
-            d_src.upload(src);
-            d_dst.upload(dst);
+    // creating src data
+    Mat src(srcSize, CV_32F), dst(srcSize, CV_32F);
+    declare.in(src, WARMUP_RNG).out(dst);
 
-            WARMUP_ON;
-            ocl::pow(d_src, -2.0, d_dst);
-            WARMUP_OFF;
+    // select implementation
+    if (impl == "ocl")
+    {
+        ocl::oclMat oclSrc(src), oclDst(srcSize, src.type());
 
-            GPU_ON;
-            ocl::pow(d_src, -2.0, d_dst);
-            GPU_OFF;
+        TEST_CYCLE() cv::ocl::pow(oclSrc, -2.0, oclDst);
 
-            GPU_FULL_ON;
-            d_src.upload(src);
-            ocl::pow(d_src, -2.0, d_dst);
-            d_dst.download(ocl_dst);
-            GPU_FULL_OFF;
+        oclDst.download(dst);
 
-            TestSystem::instance().ExpectedMatNear(ocl_dst, dst, 1.0);
-        }
+        SANITY_CHECK(dst, 5e-2);
+    }
+    else if (impl == "plain")
+    {
+        TEST_CYCLE() cv::pow(src, -2.0, dst);
 
+        SANITY_CHECK(dst, 5e-2);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }
 
 ///////////// MagnitudeSqr////////////////////////
-PERFTEST(MagnitudeSqr)
-{
-    Mat src1, src2, dst, ocl_dst;
-    ocl::oclMat d_src1, d_src2, d_dst;
-
-    int all_type[] = {CV_32FC1};
-    std::string type_name[] = {"CV_32FC1"};
 
-    for (int size = Min_Size; size <= Max_Size; size *= Multiple)
-    {
-        for (size_t t = 0; t < sizeof(all_type) / sizeof(int); t++)
-        {
-            SUBTEST << size << 'x' << size << "; " << type_name[t];
+typedef TestBaseWithParam<Size> MagnitudeSqrFixture;
 
-            gen(src1, size, size, all_type[t], 0, 256);
-            gen(src2, size, size, all_type[t], 0, 256);
-            gen(dst, size, size, all_type[t], 0, 256);
-
-            CPU_ON;
-            for (int i = 0; i < src1.rows; ++i)
-                for (int j = 0; j < src1.cols; ++j)
-                {
-                    float val1 = src1.at<float>(i, j);
-                    float val2 = src2.at<float>(i, j);
-                    ((float *)(dst.data))[i * dst.step / 4 + j] = val1 * val1 + val2 * val2;
+PERF_TEST_P(MagnitudeSqrFixture, MagnitudeSqr, OCL_TYPICAL_MAT_SIZES)
+{
+    // getting params
+    const Size srcSize = GetParam();
+    const string impl = getSelectedImpl();
 
-                }
-            CPU_OFF;
+    // creating src data
+    Mat src1(srcSize, CV_32FC1), src2(srcSize, CV_32FC1),
+            dst(srcSize, CV_32FC1);
+    declare.in(src1, src2, WARMUP_RNG).out(dst);
 
-            d_src1.upload(src1);
-            d_src2.upload(src2);
+    // select implementation
+    if (impl == "ocl")
+    {
+        ocl::oclMat oclSrc1(src1), oclSrc2(src2), oclDst(srcSize, src1.type());
 
-            WARMUP_ON;
-            ocl::magnitudeSqr(d_src1, d_src2, d_dst);
-            WARMUP_OFF;
+        TEST_CYCLE() cv::ocl::magnitudeSqr(oclSrc1, oclSrc2, oclDst);
 
-            GPU_ON;
-            ocl::magnitudeSqr(d_src1, d_src2, d_dst);
-            GPU_OFF;
+        oclDst.download(dst);
 
-            GPU_FULL_ON;
-            d_src1.upload(src1);
-            d_src2.upload(src2);
-            ocl::magnitudeSqr(d_src1, d_src2, d_dst);
-            d_dst.download(ocl_dst);
-            GPU_FULL_OFF;
+        SANITY_CHECK(dst, 1e-6, ERROR_RELATIVE);
+    }
+    else if (impl == "plain")
+    {
+        ASSERT_EQ(1, src1.channels());
 
-            TestSystem::instance().ExpectedMatNear(ocl_dst, dst, 1.0);
+        TEST_CYCLE()
+        {
+            for (int y = 0; y < srcSize.height; ++y)
+            {
+                const float * const src1Data = reinterpret_cast<float *>(src1.data + src1.step * y);
+                const float * const src2Data = reinterpret_cast<float *>(src2.data + src2.step * y);
+                float * const dstData = reinterpret_cast<float *>(dst.data + dst.step * y);
+                for (int x = 0; x < srcSize.width; ++x)
+                {
+                    float t0 = src1Data[x] * src1Data[x];
+                    float t1 = src2Data[x] * src2Data[x];
+                    dstData[x] = t0 + t1;
+                }
+            }
         }
 
+        SANITY_CHECK(dst, 1e-6, ERROR_RELATIVE);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }
 
 ///////////// AddWeighted////////////////////////
-PERFTEST(AddWeighted)
-{
-    Mat src1, src2, dst, ocl_dst;
-    ocl::oclMat d_src1, d_src2, d_dst;
 
+typedef countNonZeroMatTypes AddWeightedMatTypes;
+typedef tuple<Size, AddWeightedMatTypes> AddWeightedParams;
+typedef TestBaseWithParam<AddWeightedParams> AddWeightedFixture;
+
+PERF_TEST_P(AddWeightedFixture, AddWeighted,
+            ::testing::Combine(OCL_TYPICAL_MAT_SIZES,
+                               AddWeightedMatTypes::all()))
+{
+    // getting params
+    AddWeightedParams params = GetParam();
+    const Size srcSize = get<0>(params);
+    const int type = get<1>(params);
+    const string impl = getSelectedImpl();
+
+    // creating src data
+    Mat src1(srcSize, type), src2(srcSize, type), dst(srcSize, type);
+    declare.in(src1, src2, WARMUP_RNG).out(dst);
     double alpha = 2.0, beta = 1.0, gama = 3.0;
-    int all_type[] = {CV_8UC1, CV_32FC1};
-    std::string type_name[] = {"CV_8UC1", "CV_32FC1"};
 
-    for (int size = Min_Size; size <= Max_Size; size *= Multiple)
+    // select implementation
+    if (impl == "ocl")
     {
-        for (size_t j = 0; j < sizeof(all_type) / sizeof(int); j++)
-        {
-            SUBTEST << size << 'x' << size << "; " << type_name[j] ;
+        ocl::oclMat oclSrc1(src1), oclSrc2(src2), oclDst(srcSize, type);
 
-            gen(src1, size, size, all_type[j], 0, 256);
-            gen(src2, size, size, all_type[j], 0, 256);
-            gen(dst, size, size, all_type[j], 0, 256);
+        TEST_CYCLE() cv::ocl::addWeighted(oclSrc1, alpha, oclSrc2, beta, gama, oclDst);
 
+        oclDst.download(dst);
 
-            addWeighted(src1, alpha, src2, beta, gama, dst);
-
-            CPU_ON;
-            addWeighted(src1, alpha, src2, beta, gama, dst);
-            CPU_OFF;
-            d_src1.upload(src1);
-            d_src2.upload(src2);
-
-            WARMUP_ON;
-            ocl::addWeighted(d_src1, alpha, d_src2, beta, gama, d_dst);
-            WARMUP_OFF;
-
-            GPU_ON;
-            ocl::addWeighted(d_src1, alpha, d_src2, beta, gama, d_dst);
-            GPU_OFF;
-
-            GPU_FULL_ON;
-            d_src1.upload(src1);
-            d_src2.upload(src2);
-            ocl::addWeighted(d_src1, alpha, d_src2, beta, gama, d_dst);
-            d_dst.download(ocl_dst);
-            GPU_FULL_OFF;
-
-            TestSystem::instance().ExpectedMatNear(ocl_dst, dst, 1e-5);
-        }
+        SANITY_CHECK(dst);
+    }
+    else if (impl == "plain")
+    {
+        TEST_CYCLE() cv::addWeighted(src1, alpha, src2, beta, gama, dst);
 
+        SANITY_CHECK(dst);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }

modules/ocl/perf/perf_blend.cpp

@@ -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 +69,54 @@ 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;
+    // getting params
+    const Size srcSize = GetParam();
+    const int type = CV_8UC1;
+    const std::string impl = getSelectedImpl();
+
+    // creating src data
+    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)
+    // select implementation
+    if (impl == "ocl")
     {
-        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 (impl == "plain")
+    {
+        TEST_CYCLE() blendLinearGold<uchar>(src1, src2, weights1, weights2, dst);
+
+        SANITY_CHECK(dst);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }

modules/ocl/perf/perf_brute_force_matcher.cpp

@@ -45,123 +45,242 @@
 //M*/
 #include "perf_precomp.hpp"
 
+using namespace perf;
+
+#define OCL_BFMATCHER_TYPICAL_MAT_SIZES ::testing::Values(cv::Size(128, 500), cv::Size(128, 1000), cv::Size(128, 2000))
+
 //////////////////// BruteForceMatch /////////////////
-PERFTEST(BruteForceMatcher)
+
+typedef TestBaseWithParam<Size> BruteForceMatcherFixture;
+
+PERF_TEST_P(BruteForceMatcherFixture, match,
+            OCL_BFMATCHER_TYPICAL_MAT_SIZES)
 {
-    Mat trainIdx_cpu;
-    Mat distance_cpu;
-    Mat allDist_cpu;
-    Mat nMatches_cpu;
+    const Size srcSize = GetParam();
+    const string impl = getSelectedImpl();
 
-    for (int size = Min_Size; size <= Max_Size; size *= Multiple)
-    {
-        // Init CPU matcher
-        int desc_len = 64;
+    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);
 
+    if (impl == "plain")
+    {
         BFMatcher matcher(NORM_L2);
+        TEST_CYCLE() matcher.match(query, train, matches);
 
-        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);
+        SANITY_CHECK_MATCHES(matches);
+    }
+    else if (impl == "ocl")
+    {
         // Init GPU matcher
-        ocl::BruteForceMatcher_OCL_base d_matcher(ocl::BruteForceMatcher_OCL_base::L2Dist);
+        ocl::BruteForceMatcher_OCL_base oclMatcher(ocl::BruteForceMatcher_OCL_base::L2Dist);
+        ocl::oclMat oclQuery(query), oclTrain(train);
 
-        ocl::oclMat d_query(query);
-        ocl::oclMat d_train(train);
+        TEST_CYCLE() oclMatcher.match(oclQuery, oclTrain, matches);
 
-        ocl::oclMat d_trainIdx, d_distance, d_allDist, d_nMatches;
+        SANITY_CHECK_MATCHES(matches);
+    }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
+}
+
+//PERF_TEST_P(BruteForceMatcherFixture, matchSingle,
+//            OCL_BFMATCHER_TYPICAL_MAT_SIZES)
+//{
+//    const Size srcSize = GetParam();
+//    const string impl = getSelectedImpl();
+
+//    Mat query(srcSize, CV_32F), train(srcSize, CV_32F);
+//    Mat trainIdx, distance;
 
-        SUBTEST << size << "; match";
+//    randu(query, 0.0f, 1.0f);
+//    randu(train, 0.0f, 1.0f);
 
-        matcher.match(query, train, matches[0]);
+//    if (impl == "plain")
+//        CV_TEST_FAIL_NO_IMPL();
+//    else if (impl == "ocl")
+//    {
+//        ocl::oclMat oclQuery(query), oclTrain(train), oclTrainIdx, oclDistance;
 
-        CPU_ON;
-        matcher.match(query, train, matches[0]);
-        CPU_OFF;
+//        TEST_CYCLE() oclMatcher->matchSingle(oclQuery, oclTrain, oclTrainIdx, oclDistance);
+
+//        oclTrainIdx.download(trainIdx);
+//        oclDistance.download(distance);
+
+//        SANITY_CHECK(trainIdx);
+//        SANITY_CHECK(distance);
+//    }
+//#ifdef HAVE_OPENCV_GPU
+//    else if (impl == "gpu")
+//        CV_TEST_FAIL_NO_IMPL();
+//#endif
+//    else
+//        CV_TEST_FAIL_NO_IMPL();
+//}
+
+PERF_TEST_P(BruteForceMatcherFixture, knnMatch,
+            OCL_BFMATCHER_TYPICAL_MAT_SIZES)
+{
+    const Size srcSize = GetParam();
+    const string impl = getSelectedImpl();
 
-        WARMUP_ON;
-        d_matcher.matchSingle(d_query, d_train, d_trainIdx, d_distance);
-        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.matchSingle(d_query, d_train, d_trainIdx, d_distance);
-        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.match(d_query, d_train, d_matches[0]);
-        GPU_FULL_OFF;
+    if (impl == "plain")
+    {
+        BFMatcher matcher (NORM_L2);
+        TEST_CYCLE() matcher.knnMatch(query, train, matches, 2);
 
-        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);
+        std::vector<DMatch> & matches0 = matches[0], & matches1 = matches[1];
+        SANITY_CHECK_MATCHES(matches0);
+        SANITY_CHECK_MATCHES(matches1);
+    }
+    else if (impl == "ocl")
+    {
+        ocl::BruteForceMatcher_OCL_base oclMatcher(ocl::BruteForceMatcher_OCL_base::L2Dist);
+        ocl::oclMat oclQuery(query), oclTrain(train);
 
-        SUBTEST << size << "; knnMatch";
+        TEST_CYCLE() oclMatcher.knnMatch(oclQuery, oclTrain, matches, 2);
 
-        matcher.knnMatch(query, train, matches, 2);
+        std::vector<DMatch> & matches0 = matches[0], & matches1 = matches[1];
+        SANITY_CHECK_MATCHES(matches0);
+        SANITY_CHECK_MATCHES(matches1);
+    }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
+}
 
-        CPU_ON;
-        matcher.knnMatch(query, train, matches, 2);
-        CPU_OFF;
+//PERF_TEST_P(BruteForceMatcherFixture, knnMatchSingle,
+//            OCL_BFMATCHER_TYPICAL_MAT_SIZES)
+//{
+//    const Size srcSize = GetParam();
+//    const string impl = getSelectedImpl();
 
-        WARMUP_ON;
-        d_matcher.knnMatchSingle(d_query, d_train, d_trainIdx, d_distance, d_allDist, 2);
-        WARMUP_OFF;
+//    Mat query(srcSize, CV_32F), train(srcSize, CV_32F);
+//    Mat trainIdx, distance, allDist;
 
-        GPU_ON;
-        d_matcher.knnMatchSingle(d_query, d_train, d_trainIdx, d_distance, d_allDist, 2);
-        GPU_OFF;
+//    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.knnMatch(d_query, d_train, d_matches, 2);
-        GPU_FULL_OFF;
+//    if (impl == "plain")
+//        CV_TEST_FAIL_NO_IMPL();
+//    else if (impl == "ocl")
+//    {
+//        ocl::oclMat oclQuery(query), oclTrain(train), oclTrainIdx, oclDistance, oclAllDist;
 
-        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);
+//        TEST_CYCLE() oclMatcher->knnMatchSingle(oclQuery, oclTrain, oclTrainIdx, oclDistance, oclAllDist, 2);
 
-        SUBTEST << size << "; radiusMatch";
+//        oclTrainIdx.download(trainIdx);
+//        oclDistance.download(distance);
+//        oclAllDist.download(allDist);
 
-        float max_distance = 2.0f;
+//        SANITY_CHECK(trainIdx);
+//        SANITY_CHECK(distance);
+//        SANITY_CHECK(allDist);
+//    }
+//#ifdef HAVE_OPENCV_GPU
+//    else if (impl == "gpu")
+//        CV_TEST_FAIL_NO_IMPL();
+//#endif
+//    else
+//        CV_TEST_FAIL_NO_IMPL();
+//}
 
-        matcher.radiusMatch(query, train, matches, max_distance);
+PERF_TEST_P(BruteForceMatcherFixture, DISABLED_radiusMatch,
+            OCL_BFMATCHER_TYPICAL_MAT_SIZES)
+{
+    const Size srcSize = GetParam();
+    const string impl = getSelectedImpl();
 
-        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 (impl == "plain")
+    {
+        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 (impl == "ocl")
+    {
+        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);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }
+
+//PERF_TEST_P(BruteForceMatcherFixture, radiusMatchSingle,
+//            OCL_BFMATCHER_TYPICAL_MAT_SIZES)
+//{
+//    const Size srcSize = GetParam();
+//    const string impl = getSelectedImpl();
+
+//    const float max_distance = 2.0f;
+//    Mat query(srcSize, CV_32F), train(srcSize, CV_32F);
+//    Mat trainIdx, distance, nMatches;
+
+//    randu(query, 0.0f, 1.0f);
+//    randu(train, 0.0f, 1.0f);
+
+//    if (impl == "plain")
+//        CV_TEST_FAIL_NO_IMPL();
+//    else if (impl == "ocl")
+//    {
+//        ocl::oclMat oclQuery(query), oclTrain(train), oclTrainIdx, oclDistance, oclNMatches;
+
+//        TEST_CYCLE() oclMatcher->radiusMatchSingle(oclQuery, oclTrain, oclTrainIdx, oclDistance, oclNMatches, max_distance);
+
+//        oclTrainIdx.download(trainIdx);
+//        oclDistance.download(distance);
+//        oclNMatches.download(nMatches);
+
+//        SANITY_CHECK(trainIdx);
+//        SANITY_CHECK(distance);
+//        SANITY_CHECK(nMatches);
+//    }
+//#ifdef HAVE_OPENCV_GPU
+//    else if (impl == "gpu")
+//        CV_TEST_FAIL_NO_IMPL();
+//#endif
+//    else
+//        CV_TEST_FAIL_NO_IMPL();
+//}
+
+#undef OCL_BFMATCHER_TYPICAL_MAT_SIZES

modules/ocl/perf/perf_calib3d.cpp

@@ -45,48 +45,49 @@
 //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)
+{
+    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 std::string impl = getSelectedImpl();
+    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 (impl == "ocl")
+    {
+        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 (impl == "plain")
+    {
+        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);
+    }
+#ifdef HAVE_OPENCV_GPU
+        else if (impl == "gpu")
+            CV_TEST_FAIL_NO_IMPL();
+    #endif
+        else
+            CV_TEST_FAIL_NO_IMPL();
 }

modules/ocl/perf/perf_canny.cpp

@@ -45,41 +45,38 @@
 //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, Canny)
+{
+    Mat img = imread(getDataPath("gpu/stereobm/aloe-L.png"), cv::IMREAD_GRAYSCALE),
+            edges(img.size(), CV_8UC1);
+    ASSERT_TRUE(!img.empty()) << "can't open aloeL.jpg";
 
-    ocl::oclMat d_img(img);
-    ocl::oclMat d_edges;
-    ocl::CannyBuf d_buf;
+    const std::string impl = getSelectedImpl();
+    declare.in(img).out(edges);
 
-    WARMUP_ON;
-    ocl::Canny(d_img, d_buf, d_edges, 50.0, 100.0);
-    WARMUP_OFF;
+    if (impl == "ocl")
+    {
+        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() 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 (impl == "plain")
+    {
+        TEST_CYCLE() Canny(img, edges, 50.0, 100.0);
 
-    TestSystem::instance().ExceptedMatSimilar(edges, ocl_edges, 2e-2);
+        SANITY_CHECK(edges);
+    }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }

modules/ocl/perf/perf_color.cpp

@@ -45,49 +45,39 @@
 //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();
+    const std::string impl = getSelectedImpl();
 
-            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).out(dst);
 
-            TestSystem::instance().ExceptedMatSimilar(dst, ocl_dst, 1e-5);
-        }
+    if (impl == "ocl")
+    {
+        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 (impl == "plain")
+    {
+        TEST_CYCLE() cv::cvtColor(src, dst, CV_RGBA2GRAY, 4);
 
-
+        SANITY_CHECK(dst);
+    }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }

modules/ocl/perf/perf_fft.cpp

@@ -45,47 +45,43 @@
 //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"};
-
-    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";
+using namespace perf;
 
-            gen(src, size, size, all_type[j], Scalar::all(0), Scalar::all(1));
+///////////// dft ////////////////////////
 
-            dft(src, dst);
+typedef TestBaseWithParam<Size> dftFixture;
 
-            CPU_ON;
-            dft(src, dst);
-            CPU_OFF;
+PERF_TEST_P(dftFixture, DISABLED_dft, OCL_TYPICAL_MAT_SIZES)
+{
+    const std::string impl = getSelectedImpl();
+    Size srcSize = GetParam();
 
-            d_src.upload(src);
+    Mat src(srcSize, CV_32FC2), dst;
+    randu(src, 0.0f, 1.0f);
+    declare.in(src);
 
-            WARMUP_ON;
-            ocl::dft(d_src, d_dst, Size(size, size));
-            WARMUP_OFF;
+    if (impl == "ocl")
+    {
+        ocl::oclMat oclSrc(src), oclDst;
 
-            GPU_ON;
-            ocl::dft(d_src, d_dst, Size(size, size));
-            GPU_OFF;
+        EXPECT_NO_THROW({
+            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 (impl == "plain")
+    {
+        TEST_CYCLE() cv::dft(src, dst);
 
+        SANITY_CHECK(dst);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }

modules/ocl/perf/perf_filters.cpp

@@ -45,333 +45,344 @@
 //M*/
 #include "perf_precomp.hpp"
 
+using namespace perf;
+using std::tr1::get;
+using std::tr1::tuple;
+
 ///////////// Blur////////////////////////
-PERFTEST(Blur)
-{
-    Mat src1, dst, ocl_dst;
-    ocl::oclMat d_src1, d_dst;
 
-    Size ksize = Size(3, 3);
-    int bordertype = BORDER_CONSTANT;
-    int all_type[] = {CV_8UC1, CV_8UC4};
-    std::string type_name[] = {"CV_8UC1", "CV_8UC4"};
+CV_ENUM(BlurMatType, 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] ;
+typedef tuple<Size, BlurMatType> BlurParams;
+typedef TestBaseWithParam<BlurParams> BlurFixture;
 
-            gen(src1, size, size, all_type[j], 0, 256);
-            gen(dst, size, size, all_type[j], 0, 256);
+PERF_TEST_P(BlurFixture, Blur,
+            ::testing::Combine(OCL_TYPICAL_MAT_SIZES,
+                               BlurMatType::all()))
+{
+    // getting params
+    BlurParams params = GetParam();
+    const Size srcSize = get<0>(params), ksize(3, 3);
+    const int type = get<1>(params), bordertype = BORDER_CONSTANT;
 
-            blur(src1, dst, ksize, Point(-1, -1), bordertype);
+    const std::string impl = getSelectedImpl();
 
-            CPU_ON;
-            blur(src1, dst, ksize, Point(-1, -1), bordertype);
-            CPU_OFF;
+    Mat src(srcSize, type), dst(srcSize, type);
+    declare.in(src, WARMUP_RNG).out(dst);
 
-            d_src1.upload(src1);
+    if (srcSize == OCL_SIZE_4000 && type == CV_8UC4)
+        declare.time(5);
 
-            WARMUP_ON;
-            ocl::blur(d_src1, d_dst, ksize, Point(-1, -1), bordertype);
-            WARMUP_OFF;
+    if (impl == "ocl")
+    {
+        ocl::oclMat oclSrc(src), oclDst(srcSize, type);
 
-            GPU_ON;
-            ocl::blur(d_src1, d_dst, ksize, Point(-1, -1), bordertype);
-            GPU_OFF;
+        TEST_CYCLE() cv::ocl::blur(oclSrc, oclDst, ksize, Point(-1, -1), bordertype);
 
-            GPU_FULL_ON;
-            d_src1.upload(src1);
-            ocl::blur(d_src1, d_dst, ksize, Point(-1, -1), bordertype);
-            d_dst.download(ocl_dst);
-            GPU_FULL_OFF;
+        oclDst.download(dst);
 
-            TestSystem::instance().ExpectedMatNear(ocl_dst, dst, 1.0);
-        }
+        SANITY_CHECK(dst, 1 + DBL_EPSILON);
+    }
+    else if (impl == "plain")
+    {
+        TEST_CYCLE() cv::blur(src, dst, ksize, Point(-1, -1), bordertype);
 
+        SANITY_CHECK(dst, 1 + DBL_EPSILON);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }
-///////////// Laplacian////////////////////////
-PERFTEST(Laplacian)
-{
-    Mat src1, dst, ocl_dst;
-    ocl::oclMat d_src1, d_dst;
 
-    int ksize = 3;
-    int all_type[] = {CV_8UC1, CV_8UC4};
-    std::string type_name[] = {"CV_8UC1", "CV_8UC4"};
+///////////// Laplacian////////////////////////
 
-    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] ;
+typedef BlurMatType LaplacianMatType;
+typedef tuple<Size, LaplacianMatType> LaplacianParams;
+typedef TestBaseWithParam<LaplacianParams> LaplacianFixture;
 
-            gen(src1, size, size, all_type[j], 0, 256);
-            gen(dst, size, size, all_type[j], 0, 256);
+PERF_TEST_P(LaplacianFixture, Laplacian,
+            ::testing::Combine(OCL_TYPICAL_MAT_SIZES,
+                               LaplacianMatType::all()))
+{
+    // getting params
+    LaplacianParams params = GetParam();
+    const Size srcSize = get<0>(params);
+    const int type = get<1>(params), ksize = 3;
 
-            Laplacian(src1, dst, -1, ksize, 1);
+    const std::string impl = getSelectedImpl();
 
-            CPU_ON;
-            Laplacian(src1, dst, -1, ksize, 1);
-            CPU_OFF;
+    Mat src(srcSize, type), dst(srcSize, type);
+    declare.in(src, WARMUP_RNG).out(dst);
 
-            d_src1.upload(src1);
+    if (srcSize == OCL_SIZE_4000 && type == CV_8UC4)
+        declare.time(6);
 
-            WARMUP_ON;
-            ocl::Laplacian(d_src1, d_dst, -1, ksize, 1);
-            WARMUP_OFF;
+    if (impl == "ocl")
+    {
+        ocl::oclMat oclSrc(src), oclDst(srcSize, type);
 
-            GPU_ON;
-            ocl::Laplacian(d_src1, d_dst, -1, ksize, 1);
-            GPU_OFF;
+        TEST_CYCLE() cv::ocl::Laplacian(oclSrc, oclDst, -1, ksize, 1);
 
-            GPU_FULL_ON;
-            d_src1.upload(src1);
-            ocl::Laplacian(d_src1, d_dst, -1, ksize, 1);
-            d_dst.download(ocl_dst);
-            GPU_FULL_OFF;
+        oclDst.download(dst);
 
-            TestSystem::instance().ExpectedMatNear(ocl_dst, dst, 1e-5);
-        }
+        SANITY_CHECK(dst);
+    }
+    else if (impl == "plain")
+    {
+        TEST_CYCLE() cv::Laplacian(src, dst, -1, ksize, 1);
 
+        SANITY_CHECK(dst);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }
 
 ///////////// Erode ////////////////////
-PERFTEST(Erode)
-{
-    Mat src, dst, ker, ocl_dst;
-    ocl::oclMat d_src, d_dst;
 
-    int all_type[] = {CV_8UC1, CV_8UC4, CV_32FC1, CV_32FC4};
-    std::string type_name[] = {"CV_8UC1", "CV_8UC4", "CV_32FC1", "CV_32FC4"};
+CV_ENUM(ErodeMatType, CV_8UC1, CV_8UC4, CV_32FC1, CV_32FC4)
 
-    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] ;
+typedef tuple<Size, ErodeMatType> ErodeParams;
+typedef TestBaseWithParam<ErodeParams> ErodeFixture;
 
-            gen(src, size, size, all_type[j], Scalar::all(0), Scalar::all(256));
-            ker = getStructuringElement(MORPH_RECT, Size(3, 3));
+PERF_TEST_P(ErodeFixture, Erode,
+            ::testing::Combine(OCL_TYPICAL_MAT_SIZES,
+                               ErodeMatType::all()))
+{
+    // getting params
+    ErodeParams params = GetParam();
+    const Size srcSize = get<0>(params);
+    const int type = get<1>(params), ksize = 3;
+    const Mat ker = getStructuringElement(MORPH_RECT, Size(ksize, ksize));
 
-            erode(src, dst, ker);
+    const std::string impl = getSelectedImpl();
 
-            CPU_ON;
-            erode(src, dst, ker);
-            CPU_OFF;
+    Mat src(srcSize, type), dst(srcSize, type);
+    declare.in(src, WARMUP_RNG).out(dst).in(ker);
 
-            d_src.upload(src);
+    if (srcSize == OCL_SIZE_4000 && type == CV_8UC4)
+        declare.time(5);
 
-            WARMUP_ON;
-            ocl::erode(d_src, d_dst, ker);
-            WARMUP_OFF;
+    if (impl == "ocl")
+    {
+        ocl::oclMat oclSrc(src), oclDst(srcSize, type), oclKer(ker);
 
-            GPU_ON;
-            ocl::erode(d_src, d_dst, ker);
-            GPU_OFF;
+        TEST_CYCLE() cv::ocl::erode(oclSrc, oclDst, oclKer);
 
-            GPU_FULL_ON;
-            d_src.upload(src);
-            ocl::erode(d_src, d_dst, ker);
-            d_dst.download(ocl_dst);
-            GPU_FULL_OFF;
+        oclDst.download(dst);
 
-            TestSystem::instance().ExpectedMatNear(ocl_dst, dst, 1e-5);
-        }
+        SANITY_CHECK(dst);
+    }
+    else if (impl == "plain")
+    {
+        TEST_CYCLE() cv::erode(src, dst, ker);
 
+        SANITY_CHECK(dst);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }
 
 ///////////// Sobel ////////////////////////
-PERFTEST(Sobel)
-{
-    Mat src, dst, ocl_dst;
-    ocl::oclMat d_src, d_dst;
-
-    int dx = 1;
-    int dy = 1;
-    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] ;
+typedef BlurMatType SobelMatType;
+typedef tuple<Size, SobelMatType> SobelMatParams;
+typedef TestBaseWithParam<SobelMatParams> SobelFixture;
 
-            gen(src, size, size, all_type[j], 0, 256);
+PERF_TEST_P(SobelFixture, Sobel,
+            ::testing::Combine(OCL_TYPICAL_MAT_SIZES,
+                               SobelMatType::all()))
+{
+    // getting params
+    SobelMatParams params = GetParam();
+    const Size srcSize = get<0>(params);
+    const int type = get<1>(params), dx = 1, dy = 1;
 
-            Sobel(src, dst, -1, dx, dy);
+    const std::string impl = getSelectedImpl();
 
-            CPU_ON;
-            Sobel(src, dst, -1, dx, dy);
-            CPU_OFF;
+    Mat src(srcSize, type), dst(srcSize, type);
+    declare.in(src, WARMUP_RNG).out(dst);
 
-            d_src.upload(src);
+    if ((srcSize == OCL_SIZE_2000 && type == CV_8UC4) ||
+            (srcSize == OCL_SIZE_4000 && type == CV_8UC1))
+        declare.time(5.5);
+    else if (srcSize == OCL_SIZE_4000 && type == CV_8UC4)
+        declare.time(20);
 
-            WARMUP_ON;
-            ocl::Sobel(d_src, d_dst, -1, dx, dy);
-            WARMUP_OFF;
+    if (impl == "ocl")
+    {
+        ocl::oclMat oclSrc(src), oclDst(srcSize, type);
 
-            GPU_ON;
-            ocl::Sobel(d_src, d_dst, -1, dx, dy);
-            GPU_OFF;
+        TEST_CYCLE() cv::ocl::Sobel(oclSrc, oclDst, -1, dx, dy);
 
-            GPU_FULL_ON;
-            d_src.upload(src);
-            ocl::Sobel(d_src, d_dst, -1, dx, dy);
-            d_dst.download(ocl_dst);
-            GPU_FULL_OFF;
+        oclDst.download(dst);
 
-            TestSystem::instance().ExpectedMatNear(ocl_dst, dst, 1);
-        }
+        SANITY_CHECK(dst);
+    }
+    else if (impl == "plain")
+    {
+        TEST_CYCLE() cv::Sobel(src, dst, -1, dx, dy);
 
+        SANITY_CHECK(dst);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }
+
 ///////////// Scharr ////////////////////////
-PERFTEST(Scharr)
-{
-    Mat src, dst, ocl_dst;
-    ocl::oclMat d_src, d_dst;
 
-    int dx = 1;
-    int dy = 0;
-    int all_type[] = {CV_8UC1, CV_8UC4};
-    std::string type_name[] = {"CV_8UC1", "CV_8UC4"};
+typedef BlurMatType ScharrMatType;
+typedef tuple<Size, ScharrMatType> ScharrParams;
+typedef TestBaseWithParam<ScharrParams> ScharrFixture;
 
-    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] ;
-
-            gen(src, size, size, all_type[j], 0, 256);
+PERF_TEST_P(ScharrFixture, Scharr,
+            ::testing::Combine(OCL_TYPICAL_MAT_SIZES,
+                               ScharrMatType::all()))
+{
+    // getting params
+    ScharrParams params = GetParam();
+    const Size srcSize = get<0>(params);
+    const int type = get<1>(params), dx = 1, dy = 0;
 
-            Scharr(src, dst, -1, dx, dy);
+    const std::string impl = getSelectedImpl();
 
-            CPU_ON;
-            Scharr(src, dst, -1, dx, dy);
-            CPU_OFF;
+    Mat src(srcSize, type), dst(srcSize, type);
+    declare.in(src, WARMUP_RNG).out(dst);
 
-            d_src.upload(src);
+    if ((srcSize == OCL_SIZE_2000 && type == CV_8UC4) ||
+            (srcSize == OCL_SIZE_4000 && type == CV_8UC1))
+        declare.time(5.5);
+    else if (srcSize == OCL_SIZE_4000 && type == CV_8UC4)
+        declare.time(21);
 
-            WARMUP_ON;
-            ocl::Scharr(d_src, d_dst, -1, dx, dy);
-            WARMUP_OFF;
+    if (impl == "ocl")
+    {
+        ocl::oclMat oclSrc(src), oclDst(srcSize, type);
 
-            GPU_ON;
-            ocl::Scharr(d_src, d_dst, -1, dx, dy);
-            GPU_OFF;
+        TEST_CYCLE() cv::ocl::Scharr(oclSrc, oclDst, -1, dx, dy);
 
-            GPU_FULL_ON;
-            d_src.upload(src);
-            ocl::Scharr(d_src, d_dst, -1, dx, dy);
-            d_dst.download(ocl_dst);
-            GPU_FULL_OFF;
+        oclDst.download(dst);
 
-            TestSystem::instance().ExpectedMatNear(ocl_dst, dst, 1);
-        }
+        SANITY_CHECK(dst);
+    }
+    else if (impl == "plain")
+    {
+        TEST_CYCLE() cv::Scharr(src, dst, -1, dx, dy);
 
+        SANITY_CHECK(dst);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }
 
 ///////////// GaussianBlur ////////////////////////
-PERFTEST(GaussianBlur)
-{
-    Mat src, dst, ocl_dst;
-    int all_type[] = {CV_8UC1, CV_8UC4, CV_32FC1, CV_32FC4};
-    std::string type_name[] = {"CV_8UC1", "CV_8UC4", "CV_32FC1", "CV_32FC4"};
-    const int ksize = 7;
 
-    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] ;
+typedef ErodeMatType GaussianBlurMatType;
+typedef tuple<Size, GaussianBlurMatType> GaussianBlurParams;
+typedef TestBaseWithParam<GaussianBlurParams> GaussianBlurFixture;
 
-            gen(src, size, size, all_type[j], 0, 256);
+PERF_TEST_P(GaussianBlurFixture, GaussianBlur,
+            ::testing::Combine(::testing::Values(OCL_SIZE_1000, OCL_SIZE_2000),
+                               GaussianBlurMatType::all()))
+{
+    // getting params
+    GaussianBlurParams params = GetParam();
+    const Size srcSize = get<0>(params);
+    const int type = get<1>(params), ksize = 7;
 
-            GaussianBlur(src, dst, Size(ksize, ksize), 0);
+    const std::string impl = getSelectedImpl();
 
-            CPU_ON;
-            GaussianBlur(src, dst, Size(ksize, ksize), 0);
-            CPU_OFF;
+    Mat src(srcSize, type), dst(srcSize, type);
+    declare.in(src, WARMUP_RNG).out(dst);
 
-            ocl::oclMat d_src(src);
-            ocl::oclMat d_dst;
+    const double eps = src.depth() == CV_8U ? 1 + DBL_EPSILON : 3e-4;
 
-            WARMUP_ON;
-            ocl::GaussianBlur(d_src, d_dst, Size(ksize, ksize), 0);
-            WARMUP_OFF;
+    if (impl == "ocl")
+    {
+        ocl::oclMat oclSrc(src), oclDst(srcSize, type);
 
-            GPU_ON;
-            ocl::GaussianBlur(d_src, d_dst, Size(ksize, ksize), 0);
-            GPU_OFF;
+        TEST_CYCLE() cv::ocl::GaussianBlur(oclSrc, oclDst, Size(ksize, ksize), 0);
 
-            GPU_FULL_ON;
-            d_src.upload(src);
-            ocl::GaussianBlur(d_src, d_dst, Size(ksize, ksize), 0);
-            d_dst.download(ocl_dst);
-            GPU_FULL_OFF;
+        oclDst.download(dst);
 
-            TestSystem::instance().ExpectedMatNear(ocl_dst, dst, 1.0);
-        }
+        SANITY_CHECK(dst, eps);
+    }
+    else if (impl == "plain")
+    {
+        TEST_CYCLE() cv::GaussianBlur(src, dst, Size(ksize, ksize), 0);
 
+        SANITY_CHECK(dst, eps);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }
 
 ///////////// filter2D////////////////////////
-PERFTEST(filter2D)
-{
-    Mat src;
-
-    for (int size = Min_Size; size <= Max_Size; size *= Multiple)
-    {
-        int all_type[] = {CV_8UC1, CV_8UC4};
-        std::string type_name[] = {"CV_8UC1", "CV_8UC4"};
-
-        for (size_t j = 0; j < sizeof(all_type) / sizeof(int); j++)
-        {
-            gen(src, size, size, all_type[j], 0, 256);
-
-            const int ksize = 3;
-
-            SUBTEST << "ksize = " << ksize << "; " << size << 'x' << size << "; " << type_name[j] ;
 
-            Mat kernel;
-            gen(kernel, ksize, ksize, CV_32SC1, -3.0, 3.0);
+typedef BlurMatType filter2DMatType;
+typedef tuple<Size, filter2DMatType> filter2DParams;
+typedef TestBaseWithParam<filter2DParams> filter2DFixture;
 
-            Mat dst, ocl_dst;
-
-            cv::filter2D(src, dst, -1, kernel);
-
-            CPU_ON;
-            cv::filter2D(src, dst, -1, kernel);
-            CPU_OFF;
+PERF_TEST_P(filter2DFixture, filter2D,
+            ::testing::Combine(OCL_TYPICAL_MAT_SIZES,
+                               filter2DMatType::all()))
+{
+    // getting params
+    filter2DParams params = GetParam();
+    const Size srcSize = get<0>(params);
+    const int type = get<1>(params), ksize = 3;
 
-            ocl::oclMat d_src(src), d_dst;
+    const std::string impl = getSelectedImpl();
 
-            WARMUP_ON;
-            ocl::filter2D(d_src, d_dst, -1, kernel);
-            WARMUP_OFF;
+    Mat src(srcSize, type), dst(srcSize, type), kernel(ksize, ksize, CV_32SC1);
+    declare.in(src, WARMUP_RNG).in(kernel).out(dst);
+    randu(kernel, -3.0, 3.0);
 
-            GPU_ON;
-            ocl::filter2D(d_src, d_dst, -1, kernel);
-            GPU_OFF;
+    if (srcSize == OCL_SIZE_4000 && type == CV_8UC4)
+        declare.time(8);
 
-            GPU_FULL_ON;
-            d_src.upload(src);
-            ocl::filter2D(d_src, d_dst, -1, kernel);
-            d_dst.download(ocl_dst);
-            GPU_FULL_OFF;
+    if (impl == "ocl")
+    {
+        ocl::oclMat oclSrc(src), oclDst(srcSize, type), oclKernel(kernel);
 
-            TestSystem::instance().ExpectedMatNear(ocl_dst, dst, 1e-5);
+        TEST_CYCLE() cv::ocl::filter2D(oclSrc, oclDst, -1, oclKernel);
 
-        }
+        oclDst.download(dst);
 
+        SANITY_CHECK(dst);
+    }
+    else if (impl == "plain")
+    {
+        TEST_CYCLE() cv::filter2D(src, dst, -1, kernel);
 
+        SANITY_CHECK(dst);
     }
+ #ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }

modules/ocl/perf/perf_gemm.cpp

@@ -45,46 +45,46 @@
 //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)
+{
+    // getting params
+    const Size srcSize = GetParam();
+    const std::string impl = getSelectedImpl();
 
-        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 (impl == "ocl")
+    {
+        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 (impl == "plain")
+    {
+        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);
     }
+ #ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }

modules/ocl/perf/perf_gftt.cpp

@@ -46,56 +46,54 @@
 
 #include "perf_precomp.hpp"
 
-///////////// GoodFeaturesToTrack ////////////////////////
-PERFTEST(GoodFeaturesToTrack)
-{
-    using namespace cv;
-
-    int maxCorners = 2000;
-    double qualityLevel = 0.01;
-
-    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());
+using namespace perf;
+using std::tr1::tuple;
+using std::tr1::get;
 
-        for(float minDistance = 0; minDistance < 4; minDistance += 3.0)
-        {
-            SUBTEST << "image = " << images[imgIdx] << "; ";
-            SUBTEST << "minDistance = " << minDistance << "; ";
+///////////// GoodFeaturesToTrack ////////////////////////
 
-            cv::goodFeaturesToTrack(frame, pts_gold, maxCorners, qualityLevel, minDistance);
+typedef tuple<string, double> GoodFeaturesToTrackParams;
+typedef TestBaseWithParam<GoodFeaturesToTrackParams> GoodFeaturesToTrackFixture;
 
-            CPU_ON;
-            cv::goodFeaturesToTrack(frame, pts_gold, maxCorners, qualityLevel, minDistance);
-            CPU_OFF;
+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)))
+{
+    std::vector<cv::Point2f> pts_gold;
 
-            cv::ocl::GoodFeaturesToTrackDetector_OCL detector(maxCorners, qualityLevel, minDistance);
+    // getting params
+    GoodFeaturesToTrackParams param = GetParam();
+    const string fileName = getDataPath(get<0>(param)), impl = getSelectedImpl();
+    const int maxCorners = 2000;
+    const double qualityLevel = 0.01, minDistance = get<1>(param);
 
-            ocl::oclMat frame_ocl(frame), pts_oclmat;
+    Mat frame = imread(fileName, IMREAD_GRAYSCALE);
+    declare.in(frame);
+    ASSERT_TRUE(!frame.empty()) << "no input image";
 
-            WARMUP_ON;
-            detector(frame_ocl, pts_oclmat);
-            WARMUP_OFF;
+    if (impl == "ocl")
+    {
+        ocl::oclMat oclFrame(frame), pts_oclmat;
+        cv::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 (impl == "plain")
+    {
+        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);
     }
+ #ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }

modules/ocl/perf/perf_haar.cpp

@@ -45,6 +45,8 @@
 //M*/
 #include "perf_precomp.hpp"
 
+using namespace perf;
+
 ///////////// Haar ////////////////////////
 namespace cv
 {
@@ -83,61 +85,45 @@ public:
 
 }
 }
-PERFTEST(Haar)
-{
-    Mat img = imread(abspath("basketball1.png"), CV_LOAD_IMAGE_GRAYSCALE);
 
-    if (img.empty())
-    {
-        throw runtime_error("can't open basketball1.png");
-    }
+PERF_TEST(HaarFixture, Haar)
+{
+    const std::string impl = getSelectedImpl();
+    vector<Rect> faces;
 
-    CascadeClassifier faceCascadeCPU;
+    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 (!faceCascadeCPU.load(abspath("haarcascade_frontalface_alt.xml")))
+    if (impl == "plain")
     {
-        throw runtime_error("can't load haarcascade_frontalface_alt.xml");
-    }
+        CascadeClassifier faceCascade;
+        ASSERT_TRUE(faceCascade.load(getDataPath("gpu/haarcascade/haarcascade_frontalface_alt.xml")))
+                << "can't load haarcascade_frontalface_alt.xml";
 
-    vector<Rect> faces;
+        TEST_CYCLE() faceCascade.detectMultiScale(img, faces,
+                                                     1.1, 2, 0 | CV_HAAR_SCALE_IMAGE, Size(30, 30));
 
-    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;
+        SANITY_CHECK(faces, 4 + 1e-4);
+    }
+    else if (impl == "ocl")
+    {
+        ocl::CascadeClassifier_GPU faceCascade;
+        ocl::oclMat oclImg(img);
 
+        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(oclImg, 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);
     }
 
-    ocl::oclMat d_img(img);
-
-    WARMUP_ON;
-    faceCascade.detectMultiScale(d_img, oclfaces,
-                                 1.1, 2, 0 | CV_HAAR_SCALE_IMAGE, Size(30, 30));
-    WARMUP_OFF;
-
-    if(faces.size() == oclfaces.size())
-        TestSystem::instance().setAccurate(1, 0);
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
     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;
+        CV_TEST_FAIL_NO_IMPL();
 }

modules/ocl/perf/perf_hog.cpp

@@ -45,50 +45,42 @@
 //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");
-    }
+    Mat src = imread(getDataPath("gpu/hog/road.png"), cv::IMREAD_GRAYSCALE);
+    ASSERT_TRUE(!src.empty()) << "can't open input image road.png";
 
-    cv::HOGDescriptor hog;
-    hog.setSVMDetector(hog.getDefaultPeopleDetector());
+    const std::string impl = getSelectedImpl();
     std::vector<cv::Rect> found_locations;
-    std::vector<cv::Rect> d_found_locations;
-
-    SUBTEST << src.cols << 'x' << src.rows << "; road.png";
+    declare.in(src).time(5);
 
-    hog.detectMultiScale(src, found_locations);
+    if (impl == "plain")
+    {
+        cv::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 (impl == "ocl")
+    {
+        cv::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);
+    }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
     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;
+        CV_TEST_FAIL_NO_IMPL();
 }

modules/ocl/perf/perf_imgproc.cpp

@@ -45,295 +45,299 @@
 //M*/
 #include "perf_precomp.hpp"
 
-///////////// equalizeHist ////////////////////////
-PERFTEST(equalizeHist)
-{
-    Mat src, dst, ocl_dst;
-    int all_type[] = {CV_8UC1};
-    std::string type_name[] = {"CV_8UC1"};
+using namespace perf;
+using std::tr1::tuple;
+using std::tr1::get;
 
-    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] ;
-
-            gen(src, size, size, all_type[j], 0, 256);
-
-            equalizeHist(src, dst);
+///////////// equalizeHist ////////////////////////
 
-            CPU_ON;
-            equalizeHist(src, dst);
-            CPU_OFF;
+typedef TestBaseWithParam<Size> equalizeHistFixture;
 
-            ocl::oclMat d_src(src);
-            ocl::oclMat d_dst;
-            ocl::oclMat d_hist;
-            ocl::oclMat d_buf;
+PERF_TEST_P(equalizeHistFixture, equalizeHist, OCL_TYPICAL_MAT_SIZES)
+{
+    // getting params
+    const Size srcSize = GetParam();
+    const string impl = getSelectedImpl();
 
-            WARMUP_ON;
-            ocl::equalizeHist(d_src, d_dst);
-            WARMUP_OFF;
+    // creating src data
+    Mat src(srcSize, CV_8UC1), dst(srcSize, CV_8UC1);
+    declare.in(src, WARMUP_RNG).out(dst);
 
-            GPU_ON;
-            ocl::equalizeHist(d_src, d_dst);
-            GPU_OFF;
+    // select implementation
+    if (impl == "ocl")
+    {
+        ocl::oclMat oclSrc(src), oclDst(srcSize, src.type());
 
-            GPU_FULL_ON;
-            d_src.upload(src);
-            ocl::equalizeHist(d_src, d_dst);
-            d_dst.download(ocl_dst);
-            GPU_FULL_OFF;
+        TEST_CYCLE() cv::ocl::equalizeHist(oclSrc, oclDst);
 
-            TestSystem::instance().ExpectedMatNear(dst, ocl_dst, 1.1);
-        }
+        oclDst.download(dst);
 
+        SANITY_CHECK(dst, 1 + DBL_EPSILON);
     }
-}
-/////////// CopyMakeBorder //////////////////////
-PERFTEST(CopyMakeBorder)
-{
-    Mat src, dst, ocl_dst;
-    ocl::oclMat d_dst;
-
-    int bordertype = BORDER_CONSTANT;
-    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 (impl == "plain")
     {
-        for (size_t j = 0; j < sizeof(all_type) / sizeof(int); j++)
-        {
-            SUBTEST << size << 'x' << size << "; " << type_name[j] ;
+        TEST_CYCLE() cv::equalizeHist(src, dst);
 
+        SANITY_CHECK(dst, 1 + DBL_EPSILON);
+    }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
+}
 
-            gen(src, size, size, all_type[j], 0, 256);
-
-            copyMakeBorder(src, dst, 7, 5, 5, 7, bordertype, cv::Scalar(1.0));
+/////////// CopyMakeBorder //////////////////////
 
-            CPU_ON;
-            copyMakeBorder(src, dst, 7, 5, 5, 7, bordertype, cv::Scalar(1.0));
-            CPU_OFF;
+CV_ENUM(CopyMakeBorderMatType, CV_8UC1, CV_8UC4)
 
-            ocl::oclMat d_src(src);
+typedef tuple<Size, CopyMakeBorderMatType> CopyMakeBorderParams;
+typedef TestBaseWithParam<CopyMakeBorderParams> CopyMakeBorderFixture;
 
-            WARMUP_ON;
-            ocl::copyMakeBorder(d_src, d_dst, 7, 5, 5, 7, bordertype, cv::Scalar(1.0));
-            WARMUP_OFF;
+PERF_TEST_P(CopyMakeBorderFixture, CopyMakeBorder,
+            ::testing::Combine(OCL_TYPICAL_MAT_SIZES,
+                               CopyMakeBorderMatType::all()))
+{
+    // getting params
+    CopyMakeBorderParams params = GetParam();
+    const Size srcSize = get<0>(params);
+    const int type = get<1>(params), borderType = BORDER_CONSTANT;
+    const string impl = getSelectedImpl();
+
+    // creating src data
+    Mat src(srcSize, type), dst;
+    const Size dstSize = srcSize + Size(12, 12);
+    dst.create(dstSize, type);
+    declare.in(src, WARMUP_RNG).out(dst);
+
+    // select implementation
+    if (impl == "ocl")
+    {
+        ocl::oclMat oclSrc(src), oclDst(dstSize, type);
 
-            GPU_ON;
-            ocl::copyMakeBorder(d_src, d_dst, 7, 5, 5, 7, bordertype, cv::Scalar(1.0));
-            GPU_OFF;
+        TEST_CYCLE() cv::ocl::copyMakeBorder(oclSrc, oclDst, 7, 5, 5, 7, borderType, cv::Scalar(1.0));
 
-            GPU_FULL_ON;
-            d_src.upload(src);
-            ocl::copyMakeBorder(d_src, d_dst, 7, 5, 5, 7, bordertype, cv::Scalar(1.0));
-            d_dst.download(ocl_dst);
-            GPU_FULL_OFF;
+        oclDst.download(dst);
 
-            TestSystem::instance().ExpectedMatNear(dst, ocl_dst, 0.0);
-        }
+        SANITY_CHECK(dst);
+    }
+    else if (impl == "plain")
+    {
+        TEST_CYCLE() cv::copyMakeBorder(src, dst, 7, 5, 5, 7, borderType, cv::Scalar(1.0));
 
+        SANITY_CHECK(dst);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }
-///////////// cornerMinEigenVal ////////////////////////
-PERFTEST(cornerMinEigenVal)
-{
-    Mat src, dst, ocl_dst;
-    ocl::oclMat d_dst;
 
-    int blockSize = 7, apertureSize = 1 + 2 * (rand() % 4);
-    int borderType = BORDER_REFLECT;
-    int all_type[] = {CV_8UC1, CV_32FC1};
-    std::string type_name[] = {"CV_8UC1", "CV_32FC1"};
+///////////// cornerMinEigenVal ////////////////////////
 
-    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] ;
+CV_ENUM(cornerMinEigenValMatType, CV_8UC1, CV_32FC1)
 
-            gen(src, size, size, all_type[j], 0, 256);
+typedef tuple<Size, cornerMinEigenValMatType> cornerMinEigenValParams;
+typedef TestBaseWithParam<cornerMinEigenValParams> cornerMinEigenValFixture;
 
-            cornerMinEigenVal(src, dst, blockSize, apertureSize, borderType);
+PERF_TEST_P(cornerMinEigenValFixture, cornerMinEigenVal,
+            ::testing::Combine(OCL_TYPICAL_MAT_SIZES,
+                               cornerMinEigenValMatType::all()))
+{
+    // getting params
+    cornerMinEigenValParams params = GetParam();
+    const Size srcSize = get<0>(params);
+    const int type = get<1>(params), borderType = BORDER_REFLECT;
+    const int blockSize = 7, apertureSize = 1 + 2 * 3;
 
-            CPU_ON;
-            cornerMinEigenVal(src, dst, blockSize, apertureSize, borderType);
-            CPU_OFF;
+    const string impl = getSelectedImpl();
 
-            ocl::oclMat d_src(src);
+    // creating src data
+    Mat src(srcSize, type), dst(srcSize, CV_32FC1);
+    declare.in(src, WARMUP_RNG).out(dst)
+            .time(srcSize == OCL_SIZE_4000 ? 20 : srcSize == OCL_SIZE_2000 ? 5 : 3);
 
-            WARMUP_ON;
-            ocl::cornerMinEigenVal(d_src, d_dst, blockSize, apertureSize, borderType);
-            WARMUP_OFF;
+    const int depth = CV_MAT_DEPTH(type);
+    const ERROR_TYPE errorType = depth == CV_8U ? ERROR_ABSOLUTE : ERROR_RELATIVE;
 
-            GPU_ON;
-            ocl::cornerMinEigenVal(d_src, d_dst, blockSize, apertureSize, borderType);
-            GPU_OFF;
+    // select implementation
+    if (impl == "ocl")
+    {
+        ocl::oclMat oclSrc(src), oclDst(srcSize, CV_32FC1);
 
-            GPU_FULL_ON;
-            d_src.upload(src);
-            ocl::cornerMinEigenVal(d_src, d_dst, blockSize, apertureSize, borderType);
-            d_dst.download(ocl_dst);
-            GPU_FULL_OFF;
+        TEST_CYCLE() cv::ocl::cornerMinEigenVal(oclSrc, oclDst, blockSize, apertureSize, borderType);
 
-            TestSystem::instance().ExpectedMatNear(dst, ocl_dst, 1.0);
-        }
+        oclDst.download(dst);
 
+        SANITY_CHECK(dst, 1e-6, errorType);
     }
-}
-///////////// cornerHarris ////////////////////////
-PERFTEST(cornerHarris)
-{
-    Mat src, dst, ocl_dst;
-    ocl::oclMat d_src, d_dst;
-
-    int all_type[] = {CV_8UC1, CV_32FC1};
-    std::string type_name[] = {"CV_8UC1", "CV_32FC1"};
-
-    for (int size = Min_Size; size <= Max_Size; size *= Multiple)
+    else if (impl == "plain")
     {
-        for (size_t j = 0; j < sizeof(all_type) / sizeof(int); j++)
-        {
-            SUBTEST << size << 'x' << size << "; " << type_name[j] << " ; BORDER_REFLECT";
+        TEST_CYCLE() cv::cornerMinEigenVal(src, dst, blockSize, apertureSize, borderType);
 
-            gen(src, size, size, all_type[j], 0, 1);
+        SANITY_CHECK(dst, 1e-6, errorType);
+    }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
+}
 
-            cornerHarris(src, dst, 5, 7, 0.1, BORDER_REFLECT);
+///////////// cornerHarris ////////////////////////
 
-            CPU_ON;
-            cornerHarris(src, dst, 5, 7, 0.1, BORDER_REFLECT);
-            CPU_OFF;
+typedef cornerMinEigenValMatType cornerHarrisMatType;
+typedef tuple<Size, cornerHarrisMatType> cornerHarrisParams;
+typedef TestBaseWithParam<cornerHarrisParams> cornerHarrisFixture;
 
-            d_src.upload(src);
+PERF_TEST_P(cornerHarrisFixture, cornerHarris,
+            ::testing::Combine(OCL_TYPICAL_MAT_SIZES,
+                               cornerHarrisMatType::all()))
+{
+    // getting params
+    cornerHarrisParams params = GetParam();
+    const Size srcSize = get<0>(params);
+    const int type = get<1>(params), borderType = BORDER_REFLECT;
 
-            WARMUP_ON;
-            ocl::cornerHarris(d_src, d_dst, 5, 7, 0.1, BORDER_REFLECT);
-            WARMUP_OFF;
+    const string impl = getSelectedImpl();
 
-            GPU_ON;
-            ocl::cornerHarris(d_src, d_dst, 5, 7, 0.1, BORDER_REFLECT);
-            GPU_OFF;
+    // creating src data
+    Mat src(srcSize, type), dst(srcSize, CV_32FC1);
+    randu(src, 0, 1);
+    declare.in(src).out(dst)
+            .time(srcSize == OCL_SIZE_4000 ? 20 : srcSize == OCL_SIZE_2000 ? 5 : 3);
 
-            GPU_FULL_ON;
-            d_src.upload(src);
-            ocl::cornerHarris(d_src, d_dst, 5, 7, 0.1, BORDER_REFLECT);
-            d_dst.download(ocl_dst);
-            GPU_FULL_OFF;
+    // select implementation
+    if (impl == "ocl")
+    {
+        ocl::oclMat oclSrc(src), oclDst(srcSize, CV_32FC1);
 
-            TestSystem::instance().ExpectedMatNear(dst, ocl_dst, 1.0);
-        }
+        TEST_CYCLE() cv::ocl::cornerHarris(oclSrc, oclDst, 5, 7, 0.1, borderType);
 
+        oclDst.download(dst);
 
+        SANITY_CHECK(dst, 3e-5);
     }
-}
-///////////// integral ////////////////////////
-PERFTEST(integral)
-{
-    Mat src, sum, ocl_sum;
-    ocl::oclMat d_src, d_sum, d_buf;
-
-    int all_type[] = {CV_8UC1};
-    std::string type_name[] = {"CV_8UC1"};
-
-    for (int size = Min_Size; size <= Max_Size; size *= Multiple)
+    else if (impl == "plain")
     {
-        for (size_t j = 0; j < sizeof(all_type) / sizeof(int); j++)
-        {
-            SUBTEST << size << 'x' << size << "; " << type_name[j]  ;
+        TEST_CYCLE() cv::cornerHarris(src, dst, 5, 7, 0.1, borderType);
 
-            gen(src, size, size, all_type[j], 0, 256);
+        SANITY_CHECK(dst, 3e-5);
+    }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
+}
 
-            integral(src, sum);
+///////////// integral ////////////////////////
 
-            CPU_ON;
-            integral(src, sum);
-            CPU_OFF;
+typedef TestBaseWithParam<Size> integralFixture;
 
-            d_src.upload(src);
+PERF_TEST_P(integralFixture, DISABLED_integral, OCL_TYPICAL_MAT_SIZES)
+{
+    // getting params
+    const Size srcSize = GetParam();
+    const string impl = getSelectedImpl();
 
-            WARMUP_ON;
-            ocl::integral(d_src, d_sum);
-            WARMUP_OFF;
+    // creating src data
+    Mat src(srcSize, CV_8UC1), dst;
+    declare.in(src, WARMUP_RNG);
 
-            GPU_ON;
-            ocl::integral(d_src, d_sum);
-            GPU_OFF;
+    // select implementation
+    if (impl == "ocl")
+    {
+        ocl::oclMat oclSrc(src), oclDst;
 
-            GPU_FULL_ON;
-            d_src.upload(src);
-            ocl::integral(d_src, d_sum);
-            d_sum.download(ocl_sum);
-            GPU_FULL_OFF;
+        TEST_CYCLE() cv::ocl::integral(oclSrc, oclDst);
 
-            if(sum.type() == ocl_sum.type()) //we won't test accuracy when cpu function overlow
-                TestSystem::instance().ExpectedMatNear(sum, ocl_sum, 0.0);
+        oclDst.download(dst);
 
-        }
+        SANITY_CHECK(dst);
+    }
+    else if (impl == "plain")
+    {
+        TEST_CYCLE() cv::integral(src, dst);
 
+        SANITY_CHECK(dst);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }
+
 ///////////// WarpAffine ////////////////////////
-PERFTEST(WarpAffine)
-{
-    Mat src, dst, ocl_dst;
-    ocl::oclMat d_src, d_dst;
 
+typedef CopyMakeBorderMatType WarpAffineMatType;
+typedef tuple<Size, WarpAffineMatType> WarpAffineParams;
+typedef TestBaseWithParam<WarpAffineParams> WarpAffineFixture;
+
+PERF_TEST_P(WarpAffineFixture, WarpAffine,
+            ::testing::Combine(OCL_TYPICAL_MAT_SIZES,
+                               WarpAffineMatType::all()))
+{
     static const double coeffs[2][3] =
     {
-        {cos(CV_PI / 6), -sin(CV_PI / 6), 100.0},
-        {sin(CV_PI / 6), cos(CV_PI / 6), -100.0}
+        { cos(CV_PI / 6), -sin(CV_PI / 6), 100.0 },
+        { sin(CV_PI / 6), cos(CV_PI / 6), -100.0 }
     };
     Mat M(2, 3, CV_64F, (void *)coeffs);
-    int interpolation = INTER_NEAREST;
+    const int interpolation = INTER_NEAREST;
 
-    int all_type[] = {CV_8UC1, CV_8UC4};
-    std::string type_name[] = {"CV_8UC1", "CV_8UC4"};
+    // getting params
+    WarpAffineParams params = GetParam();
+    const Size srcSize = get<0>(params);
+    const int type = get<1>(params);
+    const string impl = getSelectedImpl();
 
+    // creating src data
+    Mat src(srcSize, type), dst(srcSize, type);
+    declare.in(src, WARMUP_RNG).out(dst);
 
-    for (int size = Min_Size; size <= Max_Size; size *= Multiple)
+    // select implementation
+    if (impl == "ocl")
     {
-        for (size_t j = 0; j < sizeof(all_type) / sizeof(int); j++)
-        {
-            SUBTEST << size << 'x' << size << "; " << type_name[j] ;
-
-            gen(src, size, size, all_type[j], 0, 256);
-            gen(dst, size, size, all_type[j], 0, 256);
-            Size size1 = Size(size, size);
-
-            warpAffine(src, dst, M, size1, interpolation);
-
-            CPU_ON;
-            warpAffine(src, dst, M, size1, interpolation);
-            CPU_OFF;
-
-            d_src.upload(src);
+        ocl::oclMat oclSrc(src), oclDst(srcSize, type);
 
-            WARMUP_ON;
-            ocl::warpAffine(d_src, d_dst, M, size1, interpolation);
-            WARMUP_OFF;
+        TEST_CYCLE() cv::ocl::warpAffine(oclSrc, oclDst, M, srcSize, interpolation);
 
-            GPU_ON;
-            ocl::warpAffine(d_src, d_dst, M, size1, interpolation);
-            GPU_OFF;
+        oclDst.download(dst);
 
-            GPU_FULL_ON;
-            d_src.upload(src);
-            ocl::warpAffine(d_src, d_dst, M, size1, interpolation);
-            d_dst.download(ocl_dst);
-            GPU_FULL_OFF;
-
-            TestSystem::instance().ExpectedMatNear(dst, ocl_dst, 1.0);
-        }
+        SANITY_CHECK(dst);
+    }
+    else if (impl == "plain")
+    {
+        TEST_CYCLE() cv::warpAffine(src, dst, M, srcSize, interpolation);
 
+        SANITY_CHECK(dst);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }
+
 ///////////// WarpPerspective ////////////////////////
-PERFTEST(WarpPerspective)
-{
-    Mat src, dst, ocl_dst;
-    ocl::oclMat d_src, d_dst;
 
+typedef CopyMakeBorderMatType WarpPerspectiveMatType;
+typedef tuple<Size, WarpPerspectiveMatType> WarpPerspectiveParams;
+typedef TestBaseWithParam<WarpPerspectiveParams> WarpPerspectiveFixture;
+
+PERF_TEST_P(WarpPerspectiveFixture, WarpPerspective,
+            ::testing::Combine(OCL_TYPICAL_MAT_SIZES,
+                               WarpPerspectiveMatType::all()))
+{
     static const double coeffs[3][3] =
     {
         {cos(CV_PI / 6), -sin(CV_PI / 6), 100.0},
@@ -341,199 +345,154 @@ PERFTEST(WarpPerspective)
         {0.0, 0.0, 1.0}
     };
     Mat M(3, 3, CV_64F, (void *)coeffs);
-    int interpolation = INTER_LINEAR;
-
-    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] ;
-
-            gen(src, size, size, all_type[j], 0, 256);
-            gen(dst, size, size, all_type[j], 0, 256);
-            Size size1 = Size(size, size);
-
-            warpPerspective(src, dst, M, size1, interpolation);
+    const int interpolation = INTER_LINEAR;
 
-            CPU_ON;
-            warpPerspective(src, dst, M, size1, interpolation);
-            CPU_OFF;
+    // getting params
+    WarpPerspectiveParams params = GetParam();
+    const Size srcSize = get<0>(params);
+    const int type = get<1>(params);
+    const string impl = getSelectedImpl();
 
-            d_src.upload(src);
+    // creating src data
+    Mat src(srcSize, type), dst(srcSize, type);
+    declare.in(src, WARMUP_RNG).out(dst)
+            .time(srcSize == OCL_SIZE_4000 ? 18 : srcSize == OCL_SIZE_2000 ? 5 : 2);
 
-            WARMUP_ON;
-            ocl::warpPerspective(d_src, d_dst, M, size1, interpolation);
-            WARMUP_OFF;
+    // select implementation
+    if (impl == "ocl")
+    {
+        ocl::oclMat oclSrc(src), oclDst(srcSize, type);
 
-            GPU_ON;
-            ocl::warpPerspective(d_src, d_dst, M, size1, interpolation);
-            GPU_OFF;
+        TEST_CYCLE() cv::ocl::warpPerspective(oclSrc, oclDst, M, srcSize, interpolation);
 
-            GPU_FULL_ON;
-            d_src.upload(src);
-            ocl::warpPerspective(d_src, d_dst, M, size1, interpolation);
-            d_dst.download(ocl_dst);
-            GPU_FULL_OFF;
+        oclDst.download(dst);
 
-            TestSystem::instance().ExpectedMatNear(dst, ocl_dst, 1.0);
-        }
+        SANITY_CHECK(dst);
+    }
+    else if (impl == "plain")
+    {
+        TEST_CYCLE() cv::warpPerspective(src, dst, M, srcSize, interpolation);
 
+        SANITY_CHECK(dst);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }
 
 ///////////// resize ////////////////////////
-PERFTEST(resize)
-{
-    Mat src, dst, ocl_dst;
-    ocl::oclMat d_src, d_dst;
 
+CV_ENUM(resizeInterType, INTER_NEAREST, INTER_LINEAR)
 
-    int all_type[] = {CV_8UC1, CV_8UC4};
-    std::string type_name[] = {"CV_8UC1", "CV_8UC4"};
+typedef CopyMakeBorderMatType resizeMatType;
+typedef tuple<Size, resizeMatType, resizeInterType, double> resizeParams;
+typedef TestBaseWithParam<resizeParams> resizeFixture;
 
-    for (int size = Min_Size; size <= Max_Size; size *= Multiple)
+PERF_TEST_P(resizeFixture, resize,
+            ::testing::Combine(OCL_TYPICAL_MAT_SIZES,
+                               resizeMatType::all(),
+                               resizeInterType::all(),
+                               ::testing::Values(0.5, 2.0)))
+{
+    // getting params
+    resizeParams params = GetParam();
+    const Size srcSize = get<0>(params);
+    const int type = get<1>(params), interType = get<2>(params);
+    double scale = get<3>(params);
+    const string impl = getSelectedImpl();
+
+    // creating src data
+    Mat src(srcSize, type), dst;
+    const Size dstSize(cvRound(srcSize.width * scale), cvRound(srcSize.height * scale));
+    dst.create(dstSize, type);
+    declare.in(src, WARMUP_RNG).out(dst);
+    if (interType == INTER_LINEAR && type == CV_8UC4 && OCL_SIZE_4000 == srcSize)
+        declare.time(11);
+
+    // select implementation
+    if (impl == "ocl")
     {
-        for (size_t j = 0; j < sizeof(all_type) / sizeof(int); j++)
-        {
-            SUBTEST << size << 'x' << size << "; " << type_name[j] << " ; up";
-
-            gen(src, size, size, all_type[j], 0, 256);
+        ocl::oclMat oclSrc(src), oclDst(dstSize, type);
 
-            resize(src, dst, Size(), 2.0, 2.0);
+        TEST_CYCLE() cv::ocl::resize(oclSrc, oclDst, Size(), scale, scale, interType);
 
-            CPU_ON;
-            resize(src, dst, Size(), 2.0, 2.0);
-            CPU_OFF;
-
-            d_src.upload(src);
-
-            WARMUP_ON;
-            ocl::resize(d_src, d_dst, Size(), 2.0, 2.0);
-            WARMUP_OFF;
-
-            GPU_ON;
-            ocl::resize(d_src, d_dst, Size(), 2.0, 2.0);
-            GPU_OFF;
-
-            GPU_FULL_ON;
-            d_src.upload(src);
-            ocl::resize(d_src, d_dst, Size(), 2.0, 2.0);
-            d_dst.download(ocl_dst);
-            GPU_FULL_OFF;
-
-            TestSystem::instance().ExpectedMatNear(dst, ocl_dst, 1.0);
-        }
+        oclDst.download(dst);
 
+        SANITY_CHECK(dst, 1 + DBL_EPSILON);
     }
-
-    for (int size = Min_Size; size <= Max_Size; size *= Multiple)
+    else if (impl == "plain")
     {
-        for (size_t j = 0; j < sizeof(all_type) / sizeof(int); j++)
-        {
-            SUBTEST << size << 'x' << size << "; " << type_name[j] << " ; down";
-
-            gen(src, size, size, all_type[j], 0, 256);
-
-            resize(src, dst, Size(), 0.5, 0.5);
-
-            CPU_ON;
-            resize(src, dst, Size(), 0.5, 0.5);
-            CPU_OFF;
-
-            d_src.upload(src);
-
-            WARMUP_ON;
-            ocl::resize(d_src, d_dst, Size(), 0.5, 0.5);
-            WARMUP_OFF;
-
-            GPU_ON;
-            ocl::resize(d_src, d_dst, Size(), 0.5, 0.5);
-            GPU_OFF;
-
-            GPU_FULL_ON;
-            d_src.upload(src);
-            ocl::resize(d_src, d_dst, Size(), 0.5, 0.5);
-            d_dst.download(ocl_dst);
-            GPU_FULL_OFF;
-
-            TestSystem::instance().ExpectedMatNear(dst, ocl_dst, 1.0);
-        }
+        TEST_CYCLE() cv::resize(src, dst, Size(), scale, scale, interType);
 
+        SANITY_CHECK(dst, 1 + DBL_EPSILON);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }
-///////////// threshold////////////////////////
-PERFTEST(threshold)
-{
-    Mat src, dst, ocl_dst;
-    ocl::oclMat d_src, d_dst;
-
-    for (int size = Min_Size; size <= Max_Size; size *= Multiple)
-    {
-        SUBTEST << size << 'x' << size << "; 8UC1; THRESH_BINARY";
 
-        gen(src, size, size, CV_8U, 0, 100);
-
-        threshold(src, dst, 50.0, 0.0, THRESH_BINARY);
+///////////// threshold////////////////////////
 
-        CPU_ON;
-        threshold(src, dst, 50.0, 0.0, THRESH_BINARY);
-        CPU_OFF;
+CV_ENUM(ThreshType, THRESH_BINARY, THRESH_TRUNC)
 
-        d_src.upload(src);
+typedef tuple<Size, ThreshType> ThreshParams;
+typedef TestBaseWithParam<ThreshParams> ThreshFixture;
 
-        WARMUP_ON;
-        ocl::threshold(d_src, d_dst, 50.0, 0.0, THRESH_BINARY);
-        WARMUP_OFF;
+PERF_TEST_P(ThreshFixture, threshold,
+            ::testing::Combine(OCL_TYPICAL_MAT_SIZES,
+                               ThreshType::all()))
+{
+    // getting params
+    ThreshParams params = GetParam();
+    const Size srcSize = get<0>(params);
+    const int threshType = get<1>(params);
+    const string impl = getSelectedImpl();
+
+    // creating src data
+    Mat src(srcSize, CV_8U), dst(srcSize, CV_8U);
+    randu(src, 0, 100);
+    declare.in(src).out(dst);
+
+    // select implementation
+    if (impl == "ocl")
+    {
+        ocl::oclMat oclSrc(src), oclDst(srcSize, CV_8U);
 
-        GPU_ON;
-        ocl::threshold(d_src, d_dst, 50.0, 0.0, THRESH_BINARY);
-        GPU_OFF;
+        TEST_CYCLE() cv::ocl::threshold(oclSrc, oclDst, 50.0, 0.0, threshType);
 
-        GPU_FULL_ON;
-        d_src.upload(src);
-        ocl::threshold(d_src, d_dst, 50.0, 0.0, THRESH_BINARY);
-        d_dst.download(ocl_dst);
-        GPU_FULL_OFF;
+        oclDst.download(dst);
 
-        TestSystem::instance().ExpectedMatNear(dst, ocl_dst, 1.0);
+        SANITY_CHECK(dst);
     }
-
-    for (int size = Min_Size; size <= Max_Size; size *= Multiple)
+    else if (impl == "plain")
     {
-        SUBTEST << size << 'x' << size << "; 32FC1; THRESH_TRUNC [NPP]";
-
-        gen(src, size, size, CV_32FC1, 0, 100);
-
-        threshold(src, dst, 50.0, 0.0, THRESH_TRUNC);
-
-        CPU_ON;
-        threshold(src, dst, 50.0, 0.0, THRESH_TRUNC);
-        CPU_OFF;
-
-        d_src.upload(src);
+        TEST_CYCLE() cv::threshold(src, dst, 50.0, 0.0, threshType);
 
-        WARMUP_ON;
-        ocl::threshold(d_src, d_dst, 50.0, 0.0, THRESH_TRUNC);
-        WARMUP_OFF;
-
-        GPU_ON;
-        ocl::threshold(d_src, d_dst, 50.0, 0.0, THRESH_TRUNC);
-        GPU_OFF;
-
-        GPU_FULL_ON;
-        d_src.upload(src);
-        ocl::threshold(d_src, d_dst, 50.0, 0.0, THRESH_TRUNC);
-        d_dst.download(ocl_dst);
-        GPU_FULL_OFF;
-
-        TestSystem::instance().ExpectedMatNear(dst, ocl_dst, 1.0);
+        SANITY_CHECK(dst);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }
+
 ///////////// meanShiftFiltering////////////////////////
-COOR do_meanShift(int x0, int y0, uchar *sptr, uchar *dptr, int sstep, cv::Size size, int sp, int sr, int maxIter, float eps, int *tab)
+
+typedef struct
+{
+    short x;
+    short y;
+} COOR;
+
+static COOR do_meanShift(int x0, int y0, uchar *sptr, uchar *dptr, int sstep, cv::Size size, int sp, int sr, int maxIter, float eps, int *tab)
 {
 
     int isr2 = sr * sr;
@@ -715,48 +674,46 @@ static void meanShiftFiltering_(const Mat &src_roi, Mat &dst_roi, int sp, int sr
     }
 }
 
-PERFTEST(meanShiftFiltering)
+typedef TestBaseWithParam<Size> meanShiftFilteringFixture;
+
+PERF_TEST_P(meanShiftFilteringFixture, meanShiftFiltering,
+            OCL_TYPICAL_MAT_SIZES)
 {
-    int sp = 5, sr = 6;
-    Mat src, dst, ocl_dst;
+    const Size srcSize = GetParam();
+    const int sp = 5, sr = 6;
+    const string impl = getSelectedImpl();
+    cv::TermCriteria crit(cv::TermCriteria::COUNT + cv::TermCriteria::EPS, 5, 1);
 
-    ocl::oclMat d_src, d_dst;
+    Mat src(srcSize, CV_8UC4), dst(srcSize, CV_8UC4);
+    declare.in(src, WARMUP_RNG).out(dst)
+            .time(srcSize == OCL_SIZE_4000 ?
+                      56 : srcSize == OCL_SIZE_2000 ? 15 : 3.8);
 
-    for (int size = Min_Size; size <= Max_Size; size *= Multiple)
+    if (impl == "plain")
     {
-        SUBTEST << size << 'x' << size << "; 8UC3 vs 8UC4";
-
-        gen(src, size, size, CV_8UC4, Scalar::all(0), Scalar::all(256));
-
-        cv::TermCriteria crit(cv::TermCriteria::COUNT + cv::TermCriteria::EPS, 5, 1);
-
-        meanShiftFiltering_(src, dst, sp, sr, crit);
+        TEST_CYCLE() meanShiftFiltering_(src, dst, sp, sr, crit);
 
-        CPU_ON;
-        meanShiftFiltering_(src, dst, sp, sr, crit);
-        CPU_OFF;
-
-        d_src.upload(src);
-
-        WARMUP_ON;
-        ocl::meanShiftFiltering(d_src, d_dst, sp, sr, crit);
-        WARMUP_OFF;
+        SANITY_CHECK(dst);
+    }
+    else if (impl == "ocl")
+    {
+        ocl::oclMat oclSrc(src), oclDst(srcSize, CV_8UC4);
 
-        GPU_ON;
-        ocl::meanShiftFiltering(d_src, d_dst, sp, sr, crit);
-        GPU_OFF;
+        TEST_CYCLE() ocl::meanShiftFiltering(oclSrc, oclDst, sp, sr, crit);
 
-        GPU_FULL_ON;
-        d_src.upload(src);
-        ocl::meanShiftFiltering(d_src, d_dst, sp, sr, crit);
-        d_dst.download(ocl_dst);
-        GPU_FULL_OFF;
+        oclDst.download(dst);
 
-        TestSystem::instance().ExpectedMatNear(dst, ocl_dst, 0.0);
+        SANITY_CHECK(dst);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }
 
-void meanShiftProc_(const Mat &src_roi, Mat &dst_roi, Mat &dstCoor_roi, int sp, int sr, cv::TermCriteria crit)
+static void meanShiftProc_(const Mat &src_roi, Mat &dst_roi, Mat &dstCoor_roi, int sp, int sr, cv::TermCriteria crit)
 {
     if (src_roi.empty())
     {
@@ -814,200 +771,213 @@ void meanShiftProc_(const Mat &src_roi, Mat &dst_roi, Mat &dstCoor_roi, int sp,
     }
 
 }
-PERFTEST(meanShiftProc)
-{
-    Mat src;
-    vector<Mat> dst(2), ocl_dst(2);
-    ocl::oclMat d_src, d_dst, d_dstCoor;
 
-    TermCriteria crit(TermCriteria::COUNT + TermCriteria::EPS, 5, 1);
-
-    for (int size = Min_Size; size <= Max_Size; size *= Multiple)
-    {
-        SUBTEST << size << 'x' << size << "; 8UC4 and CV_16SC2 ";
-
-        gen(src, size, size, CV_8UC4, Scalar::all(0), Scalar::all(256));
+typedef TestBaseWithParam<Size> meanShiftProcFixture;
 
-        meanShiftProc_(src, dst[0], dst[1], 5, 6, crit);
+PERF_TEST_P(meanShiftProcFixture, meanShiftProc,
+            OCL_TYPICAL_MAT_SIZES)
+{
+    const Size srcSize = GetParam();
+    const string impl = getSelectedImpl();
+    TermCriteria crit(TermCriteria::COUNT + TermCriteria::EPS, 5, 1);
 
-        CPU_ON;
-        meanShiftProc_(src, dst[0], dst[1], 5, 6, crit);
-        CPU_OFF;
+    Mat src(srcSize, CV_8UC4), dst1(srcSize, CV_8UC4),
+            dst2(srcSize, CV_16SC2);
+    declare.in(src, WARMUP_RNG).out(dst1, dst2)
+            .time(srcSize == OCL_SIZE_4000 ?
+                      56 : srcSize == OCL_SIZE_2000 ? 15 : 3.8);;
 
-        d_src.upload(src);
+    if (impl == "plain")
+    {
+        TEST_CYCLE() meanShiftProc_(src, dst1, dst2, 5, 6, crit);
 
-        WARMUP_ON;
-        ocl::meanShiftProc(d_src, d_dst, d_dstCoor, 5, 6, crit);
-        WARMUP_OFF;
+        SANITY_CHECK(dst1);
+        SANITY_CHECK(dst2);
+    }
+    else if (impl == "ocl")
+    {
+        ocl::oclMat oclSrc(src), oclDst1(srcSize, CV_8UC4),
+                oclDst2(srcSize, CV_16SC2);
 
-        GPU_ON;
-        ocl::meanShiftProc(d_src, d_dst, d_dstCoor, 5, 6, crit);
-        GPU_OFF;
+        TEST_CYCLE() ocl::meanShiftProc(oclSrc, oclDst1, oclDst2, 5, 6, crit);
 
-        GPU_FULL_ON;
-        d_src.upload(src);
-        ocl::meanShiftProc(d_src, d_dst, d_dstCoor, 5, 6, crit);
-        d_dst.download(ocl_dst[0]);
-        d_dstCoor.download(ocl_dst[1]);
-        GPU_FULL_OFF;
+        oclDst1.download(dst1);
+        oclDst2.download(dst2);
 
-        vector<double> eps(2, 0.);
-        TestSystem::instance().ExpectMatsNear(dst, ocl_dst, eps);
+        SANITY_CHECK(dst1);
+        SANITY_CHECK(dst2);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }
 
 ///////////// remap////////////////////////
-PERFTEST(remap)
-{
-    Mat src, dst, xmap, ymap, ocl_dst;
-    ocl::oclMat d_src, d_dst, d_xmap, d_ymap;
 
-    int all_type[] = {CV_8UC1, CV_8UC4};
-    std::string type_name[] = {"CV_8UC1", "CV_8UC4"};
+CV_ENUM(RemapInterType, INTER_NEAREST, INTER_LINEAR)
 
-    int interpolation = INTER_LINEAR;
-    int borderMode = BORDER_CONSTANT;
+typedef CopyMakeBorderMatType remapMatType;
+typedef tuple<Size, remapMatType, RemapInterType> remapParams;
+typedef TestBaseWithParam<remapParams> remapFixture;
 
-    for (int size = Min_Size; size <= Max_Size; size *= Multiple)
-    {
-        for (size_t t = 0; t < sizeof(all_type) / sizeof(int); t++)
-        {
-            SUBTEST << size << 'x' << size << "; src " << type_name[t] << "; map CV_32FC1";
-
-            gen(src, size, size, all_type[t], 0, 256);
+PERF_TEST_P(remapFixture, remap,
+            ::testing::Combine(OCL_TYPICAL_MAT_SIZES,
+                               remapMatType::all(),
+                               RemapInterType::all()))
+{
+    // getting params
+    remapParams params = GetParam();
+    const Size srcSize = get<0>(params);
+    const int type = get<1>(params), interpolation = get<2>(params);
+    const string impl = getSelectedImpl();
 
-            xmap.create(size, size, CV_32FC1);
-            dst.create(size, size, CV_32FC1);
-            ymap.create(size, size, CV_32FC1);
+    // creating src data
+    Mat src(srcSize, type), dst(srcSize, type);
+    declare.in(src, WARMUP_RNG).out(dst);
 
-            for (int i = 0; i < size; ++i)
-            {
-                float *xmap_row = xmap.ptr<float>(i);
-                float *ymap_row = ymap.ptr<float>(i);
+    if (srcSize == OCL_SIZE_4000 && interpolation == INTER_LINEAR)
+        declare.time(9);
 
-                for (int j = 0; j < size; ++j)
-                {
-                    xmap_row[j] = (j - size * 0.5f) * 0.75f + size * 0.5f;
-                    ymap_row[j] = (i - size * 0.5f) * 0.75f + size * 0.5f;
-                }
-            }
+    Mat xmap, ymap;
+    xmap.create(srcSize, CV_32FC1);
+    ymap.create(srcSize, CV_32FC1);
 
-            remap(src, dst, xmap, ymap, interpolation, borderMode);
+    for (int i = 0; i < srcSize.height; ++i)
+    {
+        float * const xmap_row = xmap.ptr<float>(i);
+        float * const ymap_row = ymap.ptr<float>(i);
 
-            CPU_ON;
-            remap(src, dst, xmap, ymap, interpolation, borderMode);
-            CPU_OFF;
+        for (int j = 0; j < srcSize.width; ++j)
+        {
+            xmap_row[j] = (j - srcSize.width * 0.5f) * 0.75f + srcSize.width * 0.5f;
+            ymap_row[j] = (i - srcSize.height * 0.5f) * 0.75f + srcSize.height * 0.5f;
+        }
+    }
 
-            d_src.upload(src);
-            d_dst.upload(dst);
-            d_xmap.upload(xmap);
-            d_ymap.upload(ymap);
+    const int borderMode = BORDER_CONSTANT;
 
-            WARMUP_ON;
-            ocl::remap(d_src, d_dst, d_xmap, d_ymap, interpolation, borderMode);
-            WARMUP_OFF;
+    // select implementation
+    if (impl == "ocl")
+    {
+        ocl::oclMat oclSrc(src), oclDst(srcSize, type);
+        ocl::oclMat oclXMap(xmap), oclYMap(ymap);
 
-            GPU_ON;
-            ocl::remap(d_src, d_dst, d_xmap, d_ymap, interpolation, borderMode);
-            GPU_OFF;
+        TEST_CYCLE() cv::ocl::remap(oclSrc, oclDst, oclXMap, oclYMap, interpolation, borderMode);
 
-            GPU_FULL_ON;
-            d_src.upload(src);
-            ocl::remap(d_src, d_dst, d_xmap, d_ymap, interpolation, borderMode);
-            d_dst.download(ocl_dst);
-            GPU_FULL_OFF;
+        oclDst.download(dst);
 
-            TestSystem::instance().ExpectedMatNear(dst, ocl_dst, 2.0);
-        }
+        SANITY_CHECK(dst, 1 + DBL_EPSILON);
+    }
+    else if (impl == "plain")
+    {
+        TEST_CYCLE() cv::remap(src, dst, xmap, ymap, interpolation, borderMode);
 
+        SANITY_CHECK(dst, 1 + DBL_EPSILON);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }
-///////////// CLAHE ////////////////////////
-PERFTEST(CLAHE)
-{
-    Mat src, dst, ocl_dst;
-    cv::ocl::oclMat d_src, d_dst;
-    int all_type[] = {CV_8UC1};
-    std::string type_name[] = {"CV_8UC1"};
-
-    double clipLimit = 40.0;
 
-    cv::Ptr<cv::CLAHE> clahe   = cv::createCLAHE(clipLimit);
-    cv::Ptr<cv::CLAHE> d_clahe = cv::ocl::createCLAHE(clipLimit);
+///////////// CLAHE ////////////////////////
 
-    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] ;
+typedef TestBaseWithParam<Size> CLAHEFixture;
 
-            gen(src, size, size, all_type[j], 0, 256);
+PERF_TEST_P(CLAHEFixture, CLAHE, OCL_TYPICAL_MAT_SIZES)
+{
+    // getting params
+    const Size srcSize = GetParam();
+    const string impl = getSelectedImpl();
 
-            CPU_ON;
-            clahe->apply(src, dst);
-            CPU_OFF;
+    // creating src data
+    Mat src(srcSize, CV_8UC1), dst;
+    const double clipLimit = 40.0;
+    declare.in(src, WARMUP_RNG);
 
-            d_src.upload(src);
+    if (srcSize == OCL_SIZE_4000)
+        declare.time(11);
 
-            WARMUP_ON;
-            d_clahe->apply(d_src, d_dst);
-            WARMUP_OFF;
+    // select implementation
+    if (impl == "ocl")
+    {
+        ocl::oclMat oclSrc(src), oclDst;
+        cv::Ptr<cv::CLAHE> oclClahe = cv::ocl::createCLAHE(clipLimit);
 
-            ocl_dst = d_dst;
+        TEST_CYCLE() oclClahe->apply(oclSrc, oclDst);
 
-            TestSystem::instance().ExpectedMatNear(dst, ocl_dst, 1.0);
+        oclDst.download(dst);
 
-            GPU_ON;
-            d_clahe->apply(d_src, d_dst);
-            GPU_OFF;
+        SANITY_CHECK(dst);
+    }
+    else if (impl == "plain")
+    {
+        cv::Ptr<cv::CLAHE> clahe = cv::createCLAHE(clipLimit);
+        TEST_CYCLE() clahe->apply(src, dst);
 
-            GPU_FULL_ON;
-            d_src.upload(src);
-            d_clahe->apply(d_src, d_dst);
-            d_dst.download(dst);
-            GPU_FULL_OFF;
-        }
+        SANITY_CHECK(dst);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }
 
 ///////////// columnSum////////////////////////
-PERFTEST(columnSum)
+
+typedef TestBaseWithParam<Size> columnSumFixture;
+
+static void columnSumPerfTest(const Mat & src, Mat & dst)
 {
-    Mat src, dst, ocl_dst;
-    ocl::oclMat d_src, d_dst;
+    for (int j = 0; j < src.cols; j++)
+        dst.at<float>(0, j) = src.at<float>(0, j);
 
-    for (int size = Min_Size; size <= Max_Size; size *= Multiple)
-    {
-        SUBTEST << size << 'x' << size << "; CV_32FC1";
+    for (int i = 1; i < src.rows; ++i)
+        for (int j = 0; j < src.cols; ++j)
+            dst.at<float>(i, j) = dst.at<float>(i - 1 , j) + src.at<float>(i , j);
+}
 
-        gen(src, size, size, CV_32FC1, 0, 256);
+PERF_TEST_P(columnSumFixture, columnSum, OCL_TYPICAL_MAT_SIZES)
+{
+    // getting params
+    const Size srcSize = GetParam();
+    const string impl = getSelectedImpl();
 
-        CPU_ON;
-        dst.create(src.size(), src.type());
-        for (int j = 0; j < src.cols; j++)
-            dst.at<float>(0, j) = src.at<float>(0, j);
+    // creating src data
+    Mat src(srcSize, CV_32FC1), dst(srcSize, CV_32FC1);
+    declare.in(src, WARMUP_RNG).out(dst);
 
-        for (int i = 1; i < src.rows; ++i)
-            for (int j = 0; j < src.cols; ++j)
-                dst.at<float>(i, j) = dst.at<float>(i - 1 , j) + src.at<float>(i , j);
-        CPU_OFF;
+    if (srcSize == OCL_SIZE_4000)
+        declare.time(5);
 
-        d_src.upload(src);
+    // select implementation
+    if (impl == "ocl")
+    {
+        ocl::oclMat oclSrc(src), oclDst(srcSize, CV_32FC1);
 
-        WARMUP_ON;
-        ocl::columnSum(d_src, d_dst);
-        WARMUP_OFF;
+        TEST_CYCLE() cv::ocl::columnSum(oclSrc, oclDst);
 
-        GPU_ON;
-        ocl::columnSum(d_src, d_dst);
-        GPU_OFF;
+        oclDst.download(dst);
 
-        GPU_FULL_ON;
-        d_src.upload(src);
-        ocl::columnSum(d_src, d_dst);
-        d_dst.download(ocl_dst);
-        GPU_FULL_OFF;
+        SANITY_CHECK(dst);
+    }
+    else if (impl == "plain")
+    {
+        TEST_CYCLE() columnSumPerfTest(src, dst);
 
-        TestSystem::instance().ExpectedMatNear(dst, ocl_dst, 5e-1);
+        SANITY_CHECK(dst);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }

modules/ocl/perf/perf_match_template.cpp

@@ -45,101 +45,97 @@
 //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";
-
-                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;
+using namespace perf;
+using std::tr1::tuple;
+using std::tr1::get;
 
-                GPU_ON;
-                ocl::matchTemplate(d_src, d_templ, d_dst, CV_TM_CCORR);
-                GPU_OFF;
+/////////// matchTemplate ////////////////////////
 
-                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;
+CV_ENUM(CV_TM_CCORRMatType, CV_32FC1, CV_32FC4)
 
-                TestSystem::instance().ExpectedMatNear(dst, ocl_dst, templ.rows * templ.cols * 1e-1);
-            }
-        }
+typedef tuple<Size, CV_TM_CCORRMatType> CV_TM_CCORRParams;
+typedef TestBaseWithParam<CV_TM_CCORRParams> CV_TM_CCORRFixture;
 
-        int all_type_8U[] = {CV_8UC1};
-        std::string type_name_8U[] = {"CV_8UC1"};
+PERF_TEST_P(CV_TM_CCORRFixture, matchTemplate,
+            ::testing::Combine(::testing::Values(OCL_SIZE_1000, OCL_SIZE_2000),
+                               CV_TM_CCORRMatType::all()))
+{
+    // getting params
+    CV_TM_CCORRParams params = GetParam();
+    const Size srcSize = get<0>(params), templSize(5, 5);
+    const int type = get<1>(params);
+
+    std::string impl = getSelectedImpl();
+
+    // creating src data
+    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);
+
+    // select implementation
+    if (impl == "ocl")
+    {
+        ocl::oclMat oclSrc(src), oclTempl(templ), oclDst(dstSize, CV_32F);
 
-        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";
+        TEST_CYCLE() cv::ocl::matchTemplate(oclSrc, oclTempl, oclDst, CV_TM_CCORR);
 
-                gen(src, size, size, all_type_8U[j], 0, 255);
+        oclDst.download(dst);
 
-                gen(templ, templ_size, templ_size, all_type_8U[j], 0, 255);
+        SANITY_CHECK(dst, 1e-4);
+    }
+    else if (impl == "plain")
+    {
+        TEST_CYCLE() cv::matchTemplate(src, templ, dst, CV_TM_CCORR);
 
-                matchTemplate(src, templ, dst, CV_TM_CCORR_NORMED);
+        SANITY_CHECK(dst, 1e-4);
+    }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
+}
 
-                CPU_ON;
-                matchTemplate(src, templ, dst, CV_TM_CCORR_NORMED);
-                CPU_OFF;
+typedef TestBaseWithParam<Size> CV_TM_CCORR_NORMEDFixture;
 
-                ocl::oclMat d_src(src);
-                ocl::oclMat d_templ(templ), d_dst;
+PERF_TEST_P(CV_TM_CCORR_NORMEDFixture, matchTemplate, OCL_TYPICAL_MAT_SIZES)
+{
+    // getting params
+    const Size srcSize = GetParam(), templSize(5, 5);
+    const std::string impl = getSelectedImpl();
+
+    // creating src data
+    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);
+
+    // select implementation
+    if (impl == "ocl")
+    {
+        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 (impl == "plain")
+    {
+        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);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }

modules/ocl/perf/perf_matrix_operation.cpp

@@ -45,142 +45,147 @@
 //M*/
 #include "perf_precomp.hpp"
 
-///////////// ConvertTo////////////////////////
-PERFTEST(ConvertTo)
-{
-    Mat src, dst, ocl_dst;
-    ocl::oclMat d_src, d_dst;
+using namespace perf;
+using std::tr1::tuple;
+using std::tr1::get;
 
-    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";
+///////////// ConvertTo////////////////////////
 
-            gen(src, size, size, all_type[j], 0, 256);
-            //gen(dst, size, size, all_type[j], 0, 256);
+CV_ENUM(ConvertToMatType, CV_8UC1, CV_8UC4)
 
-            //d_dst.upload(dst);
+typedef tuple<Size, ConvertToMatType> ConvertToParams;
+typedef TestBaseWithParam<ConvertToParams> ConvertToFixture;
 
-            src.convertTo(dst, CV_32FC1);
+PERF_TEST_P(ConvertToFixture, ConvertTo,
+            ::testing::Combine(OCL_TYPICAL_MAT_SIZES,
+                               ConvertToMatType::all()))
+{
+    // getting params
+    ConvertToParams params = GetParam();
+    const Size srcSize = get<0>(params);
+    const int type = get<1>(params);
 
-            CPU_ON;
-            src.convertTo(dst, CV_32FC1);
-            CPU_OFF;
+    std::string impl = getSelectedImpl();
 
-            d_src.upload(src);
+    // creating src data
+    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);
 
-            WARMUP_ON;
-            d_src.convertTo(d_dst, CV_32FC1);
-            WARMUP_OFF;
+    // select implementation
+    if (impl == "ocl")
+    {
+        ocl::oclMat oclSrc(src), oclDst(srcSize, dstType);
 
-            GPU_ON;
-            d_src.convertTo(d_dst, CV_32FC1);
-            GPU_OFF;
+        TEST_CYCLE() oclSrc.convertTo(oclDst, dstType);
 
-            GPU_FULL_ON;
-            d_src.upload(src);
-            d_src.convertTo(d_dst, CV_32FC1);
-            d_dst.download(ocl_dst);
-            GPU_FULL_OFF;
+        oclDst.download(dst);
 
-            TestSystem::instance().ExpectedMatNear(dst, ocl_dst, 0.0);
-        }
+        SANITY_CHECK(dst);
+    }
+    else if (impl == "plain")
+    {
+        TEST_CYCLE() src.convertTo(dst, dstType);
 
+        SANITY_CHECK(dst);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }
-///////////// 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)
-    {
-        for (size_t j = 0; j < sizeof(all_type) / sizeof(int); j++)
-        {
-            SUBTEST << size << 'x' << size << "; " << type_name[j] ;
-
-            gen(src, size, size, all_type[j], 0, 256);
-            //gen(dst, size, size, all_type[j], 0, 256);
+///////////// copyTo////////////////////////
 
-            //d_dst.upload(dst);
+typedef ConvertToMatType copyToMatType;
+typedef tuple<Size, copyToMatType> copyToParams;
+typedef TestBaseWithParam<copyToParams> copyToFixture;
 
-            src.copyTo(dst);
+PERF_TEST_P(copyToFixture, copyTo,
+            ::testing::Combine(OCL_TYPICAL_MAT_SIZES,
+                               copyToMatType::all()))
+{
+    // getting params
+    copyToParams params = GetParam();
+    const Size srcSize = get<0>(params);
+    const int type = get<1>(params);
 
-            CPU_ON;
-            src.copyTo(dst);
-            CPU_OFF;
+    std::string impl = getSelectedImpl();
 
-            d_src.upload(src);
+    // creating src data
+    Mat src(srcSize, type), dst(srcSize, type);
+    declare.in(src, WARMUP_RNG).out(dst);
 
-            WARMUP_ON;
-            d_src.copyTo(d_dst);
-            WARMUP_OFF;
+    // select implementation
+    if (impl == "ocl")
+    {
+        ocl::oclMat oclSrc(src), oclDst(srcSize, type);
 
-            GPU_ON;
-            d_src.copyTo(d_dst);
-            GPU_OFF;
+        TEST_CYCLE() oclSrc.copyTo(oclDst);
 
-            GPU_FULL_ON;
-            d_src.upload(src);
-            d_src.copyTo(d_dst);
-            d_dst.download(ocl_dst);
-            GPU_FULL_OFF;
+        oclDst.download(dst);
 
-            TestSystem::instance().ExpectedMatNear(dst, ocl_dst, 0.0);
-        }
+        SANITY_CHECK(dst);
+    }
+    else if (impl == "plain")
+    {
+        TEST_CYCLE() src.copyTo(dst);
 
+        SANITY_CHECK(dst);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }
-///////////// 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)
-    {
-        for (size_t j = 0; j < sizeof(all_type) / sizeof(int); j++)
-        {
-            SUBTEST << size << 'x' << size << "; " << type_name[j] ;
-
-            gen(src, size, size, all_type[j], 0, 256);
+///////////// setTo////////////////////////
 
-            src.setTo(val);
+typedef ConvertToMatType setToMatType;
+typedef tuple<Size, setToMatType> setToParams;
+typedef TestBaseWithParam<setToParams> setToFixture;
 
-            CPU_ON;
-            src.setTo(val);
-            CPU_OFF;
+PERF_TEST_P(setToFixture, setTo,
+            ::testing::Combine(OCL_TYPICAL_MAT_SIZES,
+                               setToMatType::all()))
+{
+    // getting params
+    setToParams params = GetParam();
+    const Size srcSize = get<0>(params);
+    const int type = get<1>(params);
+    const Scalar val(1, 2, 3, 4);
 
-            d_src.upload(src);
+    std::string impl = getSelectedImpl();
 
-            WARMUP_ON;
-            d_src.setTo(val);
-            WARMUP_OFF;
+    // creating src data
+    Mat src(srcSize, type);
+    declare.in(src);
 
-            d_src.download(ocl_src);
-            TestSystem::instance().ExpectedMatNear(src, ocl_src, 1.0);
+    // select implementation
+    if (impl == "ocl")
+    {
+        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 (impl == "plain")
+    {
+        TEST_CYCLE() src.setTo(val);
 
+        SANITY_CHECK(src);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }

modules/ocl/perf/perf_moments.cpp

@@ -43,50 +43,59 @@
 // 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"};
+using namespace perf;
+using std::tr1::tuple;
+using std::tr1::get;
 
-    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];
+///////////// Moments ////////////////////////
 
-            gen(src, size, size, all_type[j], 0, 256);
+CV_ENUM(MomentsMatType, CV_8UC1, CV_16SC1, CV_32FC1, CV_64FC1)
 
-            cv::Moments CvMom = moments(src, binaryImage);
+typedef tuple<Size, MomentsMatType> MomentsParams;
+typedef TestBaseWithParam<MomentsParams> MomentsFixture;
 
-            CPU_ON;
-            moments(src, binaryImage);
-            CPU_OFF;
+PERF_TEST_P(MomentsFixture, DISABLED_Moments,
+            ::testing::Combine(OCL_TYPICAL_MAT_SIZES,
+                               MomentsMatType::all()))
+{
+    // getting params
+    MomentsParams params = GetParam();
+    const Size srcSize = get<0>(params);
+    const int type = get<1>(params);
 
-            cv::Moments oclMom;
-            WARMUP_ON;
-            oclMom = ocl::ocl_moments(src, binaryImage);
-            WARMUP_OFF;
+    std::string impl = getSelectedImpl();
 
-            Mat gpu_dst, cpu_dst;
-            HuMoments(CvMom, cpu_dst);
-            HuMoments(oclMom, gpu_dst);
+    // creating src data
+    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;
+    // select implementation
+    if (impl == "ocl")
+    {
+        ocl::oclMat oclSrc(src);
 
-            TestSystem::instance().ExpectedMatNear(gpu_dst, cpu_dst, .5);
+        TEST_CYCLE() mom = cv::ocl::ocl_moments(oclSrc, binaryImage);
+        cv::HuMoments(mom, dst);
 
-        }
+        SANITY_CHECK(dst);
+    }
+    else if (impl == "plain")
+    {
+        TEST_CYCLE() mom = cv::moments(src, binaryImage);
+        cv::HuMoments(mom, dst);
 
+        SANITY_CHECK(dst);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }

modules/ocl/perf/perf_norm.cpp

@@ -45,43 +45,46 @@
 //M*/
 #include "perf_precomp.hpp"
 
-///////////// norm////////////////////////
-PERFTEST(norm)
-{
-    Mat src1, src2, ocl_src1;
-    ocl::oclMat d_src1, d_src2;
+using namespace perf;
+using std::tr1::tuple;
+using std::tr1::get;
 
-    for (int size = Min_Size; size <= Max_Size; size *= Multiple)
-    {
-        SUBTEST << size << 'x' << size << "; CV_8UC1; NORM_INF";
-
-        gen(src1, size, size, CV_8UC1, Scalar::all(0), Scalar::all(1));
-        gen(src2, size, size, CV_8UC1, Scalar::all(0), Scalar::all(1));
+///////////// norm////////////////////////
 
-        norm(src1, src2, NORM_INF);
+typedef TestBaseWithParam<Size> normFixture;
 
-        CPU_ON;
-        norm(src1, src2, NORM_INF);
-        CPU_OFF;
+PERF_TEST_P(normFixture, DISABLED_norm, OCL_TYPICAL_MAT_SIZES)
+{
+    // getting params
+    const Size srcSize = GetParam();
+    const std::string impl = getSelectedImpl();
+    double value = 0.0;
 
-        d_src1.upload(src1);
-        d_src2.upload(src2);
+    // creating src data
+    Mat src1(srcSize, CV_8UC1), src2(srcSize, CV_8UC1);
+    declare.in(src1, src2);
+    randu(src1, 0, 1);
+    randu(src2, 0, 1);
 
-        WARMUP_ON;
-        ocl::norm(d_src1, d_src2, NORM_INF);
-        WARMUP_OFF;
+    // select implementation
+    if (impl == "ocl")
+    {
+        ocl::oclMat oclSrc1(src1), oclSrc2(src2);
 
-        d_src1.download(ocl_src1);
-        TestSystem::instance().ExpectedMatNear(src1, ocl_src1, .5);
+        TEST_CYCLE() value = cv::ocl::norm(oclSrc1, oclSrc2, NORM_INF);
 
-        GPU_ON;
-        ocl::norm(d_src1, d_src2, NORM_INF);
-        GPU_OFF;
+        SANITY_CHECK(value);
+    }
+    else if (impl == "plain")
+    {
+        TEST_CYCLE() value = cv::norm(src1, src2, NORM_INF);
 
-        GPU_FULL_ON;
-        d_src1.upload(src1);
-        d_src2.upload(src2);
-        ocl::norm(d_src1, d_src2, NORM_INF);
-        GPU_FULL_OFF;
+        SANITY_CHECK(value);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }

modules/ocl/perf/perf_opticalflow.cpp

@@ -46,117 +46,130 @@
 #include "perf_precomp.hpp"
 
 ///////////// PyrLKOpticalFlow ////////////////////////
-PERFTEST(PyrLKOpticalFlow)
-{
-    std::string images1[] = {"rubberwhale1.png", "aloeL.jpg"};
-    std::string images2[] = {"rubberwhale2.png", "aloeR.jpg"};
-
-    for (size_t i = 0; i < sizeof(images1) / sizeof(std::string); i++)
-    {
-        Mat frame0 = imread(abspath(images1[i]), i == 0 ? IMREAD_COLOR : IMREAD_GRAYSCALE);
-
-        if (frame0.empty())
-        {
-            std::string errstr = "can't open " + images1[i];
-            throw runtime_error(errstr);
-        }
-
-        Mat frame1 = imread(abspath(images2[i]), i == 0 ? IMREAD_COLOR : IMREAD_GRAYSCALE);
-
-        if (frame1.empty())
-        {
-            std::string errstr = "can't open " + images2[i];
-            throw runtime_error(errstr);
-        }
-
-        Mat gray_frame;
-
-        if (i == 0)
-        {
-            cvtColor(frame0, gray_frame, COLOR_BGR2GRAY);
-        }
-
-        for (int points = Min_Size; points <= Max_Size; points *= Multiple)
-        {
-            if (i == 0)
-                SUBTEST << frame0.cols << "x" << frame0.rows << "; color; " << points << " points";
-            else
-                SUBTEST << frame0.cols << "x" << frame0.rows << "; gray; " << points << " points";
-            Mat ocl_nextPts;
-            Mat ocl_status;
-
-            vector<Point2f> pts;
-            goodFeaturesToTrack(i == 0 ? gray_frame : frame0, pts, points, 0.01, 0.0);
-
-            vector<Point2f> nextPts;
-            vector<unsigned char> status;
 
-            vector<float> err;
+using namespace perf;
+using std::tr1::get;
+using std::tr1::tuple;
+using std::tr1::make_tuple;
 
-            calcOpticalFlowPyrLK(frame0, frame1, pts, nextPts, status, err);
-
-            CPU_ON;
-            calcOpticalFlowPyrLK(frame0, frame1, pts, nextPts, status, err);
-            CPU_OFF;
-
-            ocl::PyrLKOpticalFlow d_pyrLK;
-
-            ocl::oclMat d_frame0(frame0);
-            ocl::oclMat d_frame1(frame1);
-
-            ocl::oclMat d_pts;
-            Mat pts_mat(1, (int)pts.size(), CV_32FC2, (void *)&pts[0]);
-            d_pts.upload(pts_mat);
-
-            ocl::oclMat d_nextPts;
-            ocl::oclMat d_status;
-            ocl::oclMat d_err;
+template <typename T>
+static vector<T> & MatToVector(const ocl::oclMat & oclSrc, vector<T> & instance)
+{
+    Mat src;
+    oclSrc.download(src);
 
-            WARMUP_ON;
-            d_pyrLK.sparse(d_frame0, d_frame1, d_pts, d_nextPts, d_status, &d_err);
-            WARMUP_OFF;
+    for (int i = 0; i < src.cols; ++i)
+        instance.push_back(src.at<T>(0, i));
 
-            GPU_ON;
-            d_pyrLK.sparse(d_frame0, d_frame1, d_pts, d_nextPts, d_status, &d_err);
-            GPU_OFF;
+    return instance;
+}
 
-            GPU_FULL_ON;
-            d_frame0.upload(frame0);
-            d_frame1.upload(frame1);
-            d_pts.upload(pts_mat);
-            d_pyrLK.sparse(d_frame0, d_frame1, d_pts, d_nextPts, d_status, &d_err);
+CV_ENUM(LoadMode, IMREAD_GRAYSCALE, IMREAD_COLOR)
+
+typedef tuple<int, tuple<string, string, LoadMode> > PyrLKOpticalFlowParamType;
+typedef TestBaseWithParam<PyrLKOpticalFlowParamType> PyrLKOpticalFlowFixture;
+
+PERF_TEST_P(PyrLKOpticalFlowFixture,
+            PyrLKOpticalFlow,
+            ::testing::Combine(
+                ::testing::Values(1000, 2000, 4000),
+                ::testing::Values(
+                    make_tuple<string, string, LoadMode>
+                    (
+                        string("gpu/opticalflow/rubberwhale1.png"),
+                        string("gpu/opticalflow/rubberwhale1.png"),
+                        LoadMode(IMREAD_COLOR)
+                        )
+//                    , make_tuple<string, string, LoadMode>
+//                    (
+//                        string("gpu/stereobm/aloe-L.png"),
+//                        string("gpu/stereobm/aloe-R.png"),
+//                        LoadMode(IMREAD_GRAYSCALE)
+//                        )
+                    )
+                )
+            )
+{
+    PyrLKOpticalFlowParamType params = GetParam();
+    tuple<string, string, LoadMode> fileParam = get<1>(params);
+    const int pointsCount = get<0>(params);
+    const int openMode = static_cast<int>(get<2>(fileParam));
+    const string fileName0 = get<0>(fileParam), fileName1 = get<1>(fileParam);
+    Mat frame0 = imread(getDataPath(fileName0), openMode);
+    Mat frame1 = imread(getDataPath(fileName1), openMode);
+    const string impl = getSelectedImpl();
+
+    ASSERT_FALSE(frame0.empty()) << "can't load " << fileName0;
+    ASSERT_FALSE(frame1.empty()) << "can't load " << fileName1;
+
+    Mat grayFrame;
+    if (openMode == IMREAD_COLOR)
+        cvtColor(frame0, grayFrame, COLOR_BGR2GRAY);
+    else
+        grayFrame = frame0;
+
+    // initialization
+    vector<Point2f> pts, nextPts;
+    vector<unsigned char> status;
+    vector<float> err;
+    goodFeaturesToTrack(grayFrame, pts, pointsCount, 0.01, 0.0);
+
+    // selecting implementation
+    if (impl == "plain")
+    {
+        TEST_CYCLE()
+                cv::calcOpticalFlowPyrLK(frame0, frame1, pts, nextPts, status, err);
 
-            if (!d_nextPts.empty())
-                d_nextPts.download(ocl_nextPts);
+        SANITY_CHECK(nextPts);
+        SANITY_CHECK(status);
+        SANITY_CHECK(err);
+    }
+    else if (impl == "ocl")
+    {
+        ocl::PyrLKOpticalFlow oclPyrLK;
+        ocl::oclMat oclFrame0(frame0), oclFrame1(frame1);
+        ocl::oclMat oclPts(1, static_cast<int>(pts.size()), CV_32FC2, (void *)&pts[0]);
+        ocl::oclMat oclNextPts, oclStatus, oclErr;
 
-            if (!d_status.empty())
-                d_status.download(ocl_status);
-            GPU_FULL_OFF;
+        TEST_CYCLE()
+                oclPyrLK.sparse(oclFrame0, oclFrame1, oclPts, oclNextPts, oclStatus, &oclErr);
 
-            size_t mismatch = 0;
-            for (int i = 0; i < (int)nextPts.size(); ++i)
-            {
-                if(status[i] != ocl_status.at<unsigned char>(0, i))
-                {
-                    mismatch++;
-                    continue;
-                }
-                if(status[i])
-                {
-                    Point2f gpu_rst = ocl_nextPts.at<Point2f>(0, i);
-                    Point2f cpu_rst = nextPts[i];
-                    if(fabs(gpu_rst.x - cpu_rst.x) >= 1. || fabs(gpu_rst.y - cpu_rst.y) >= 1.)
-                        mismatch++;
-                }
-            }
-            double ratio = (double)mismatch / (double)nextPts.size();
-            if(ratio < .02)
-                TestSystem::instance().setAccurate(1, ratio);
-            else
-                TestSystem::instance().setAccurate(0, ratio);
-        }
+        MatToVector(oclNextPts, nextPts);
+        MatToVector(oclStatus, status);
+        MatToVector(oclErr, err);
 
+        SANITY_CHECK(nextPts);
+        SANITY_CHECK(status);
+        SANITY_CHECK(err);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
+
+//    size_t mismatch = 0;
+//    for (int i = 0; i < (int)nextPts.size(); ++i)
+//    {
+//        if(status[i] != ocl_status.at<unsigned char>(0, i))
+//        {
+//            mismatch++;
+//            continue;
+//        }
+//        if(status[i])
+//        {
+//            Point2f gpu_rst = ocl_nextPts.at<Point2f>(0, i);
+//            Point2f cpu_rst = nextPts[i];
+//            if(fabs(gpu_rst.x - cpu_rst.x) >= 1. || fabs(gpu_rst.y - cpu_rst.y) >= 1.)
+//                mismatch++;
+//        }
+//    }
+//    double ratio = (double)mismatch / (double)nextPts.size();
+//    if(ratio < .02)
+//        TestSystem::instance().setAccurate(1, ratio);
+//    else
+//        TestSystem::instance().setAccurate(0, ratio);
 }
 
 

modules/ocl/perf/perf_precomp.hpp

@@ -40,6 +40,15 @@
 //
 //M*/
 
+#ifdef __GNUC__
+#  pragma GCC diagnostic ignored "-Wmissing-declarations"
+#  pragma GCC diagnostic ignored "-Wunused-function"
+#  if defined __clang__ || defined __APPLE__
+#    pragma GCC diagnostic ignored "-Wmissing-prototypes"
+#    pragma GCC diagnostic ignored "-Wextra"
+#  endif
+#endif
+
 #ifndef __OPENCV_PERF_PRECOMP_HPP__
 #define __OPENCV_PERF_PRECOMP_HPP__
 
@@ -50,6 +59,7 @@
 #include <cstdio>
 #include <vector>
 #include <numeric>
+
 #include "opencv2/core/core.hpp"
 #include "opencv2/imgproc/imgproc.hpp"
 #include "opencv2/highgui/highgui.hpp"
@@ -59,9 +69,12 @@
 #include "opencv2/features2d/features2d.hpp"
 #include "opencv2/ocl/ocl.hpp"
 #include "opencv2/ts/ts.hpp"
-#include "opencv2/ts/ts_perf.hpp"
-#include "opencv2/ts/ts_gtest.h"
 
+#define OCL_SIZE_1000 cv::Size(1000, 1000)
+#define OCL_SIZE_2000 cv::Size(2000, 2000)
+#define OCL_SIZE_4000 cv::Size(4000, 4000)
+
+#define OCL_TYPICAL_MAT_SIZES ::testing::Values(OCL_SIZE_1000, OCL_SIZE_2000, OCL_SIZE_4000)
 
 #define Min_Size 1000
 #define Max_Size 4000
@@ -76,15 +89,15 @@ void gen(Mat &mat, int rows, int cols, int type, int low, int high, int n);
 
 string abspath(const string &relpath);
 int CV_CDECL cvErrorCallback(int, const char *, const char *, const char *, int, void *);
-typedef struct
-{
-    short x;
-    short y;
-} COOR;
-COOR do_meanShift(int x0, int y0, uchar *sptr, uchar *dptr, int sstep,
-                  cv::Size size, int sp, int sr, int maxIter, float eps, int *tab);
-void meanShiftProc_(const Mat &src_roi, Mat &dst_roi, Mat &dstCoor_roi,
-                    int sp, int sr, cv::TermCriteria crit);
+//typedef struct
+//{
+//    short x;
+//    short y;
+//} COOR;
+//COOR do_meanShift(int x0, int y0, uchar *sptr, uchar *dptr, int sstep,
+//                  cv::Size size, int sp, int sr, int maxIter, float eps, int *tab);
+//void meanShiftProc_(const Mat &src_roi, Mat &dst_roi, Mat &dstCoor_roi,
+//                    int sp, int sr, cv::TermCriteria crit);
 
 
 template<class T1, class T2>

modules/ocl/perf/perf_pyramid.cpp

@@ -45,88 +45,103 @@
 //M*/
 #include "perf_precomp.hpp"
 
+using namespace perf;
+using std::tr1::tuple;
+using std::tr1::get;
+
 ///////////// pyrDown //////////////////////
-PERFTEST(pyrDown)
-{
-    Mat src, dst, ocl_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] ;
+CV_ENUM(pyrDownMatType, CV_8UC1, CV_8UC4)
 
-            gen(src, size, size, all_type[j], 0, 256);
+typedef tuple<Size, pyrDownMatType> pyrDownParams;
+typedef TestBaseWithParam<pyrDownParams> pyrDownFixture;
 
-            pyrDown(src, dst);
+PERF_TEST_P(pyrDownFixture, pyrDown,
+            ::testing::Combine(OCL_TYPICAL_MAT_SIZES,
+                               pyrDownMatType::all()))
+{
+    // getting params
+    pyrDownParams params = GetParam();
+    const Size srcSize = get<0>(params);
+    const int type = get<1>(params);
 
-            CPU_ON;
-            pyrDown(src, dst);
-            CPU_OFF;
+    std::string impl = getSelectedImpl();
 
-            ocl::oclMat d_src(src);
-            ocl::oclMat d_dst;
+    // creating src data
+    Mat src(srcSize, type), dst;
+    Size dstSize((srcSize.height + 1) >> 1, (srcSize.width + 1) >> 1);
+    dst.create(dstSize, type);
+    declare.in(src).out(dst);
 
-            WARMUP_ON;
-            ocl::pyrDown(d_src, d_dst);
-            WARMUP_OFF;
+    // select implementation
+    if (impl == "ocl")
+    {
+        ocl::oclMat oclSrc(src), oclDst(dstSize, type);
+
+        TEST_CYCLE() cv::ocl::pyrDown(oclSrc, oclDst);
 
-            GPU_ON;
-            ocl::pyrDown(d_src, d_dst);
-            GPU_OFF;
+        oclDst.download(dst);
 
-            GPU_FULL_ON;
-            d_src.upload(src);
-            ocl::pyrDown(d_src, d_dst);
-            d_dst.download(ocl_dst);
-            GPU_FULL_OFF;
+        SANITY_CHECK(dst);
+    }
+    else if (impl == "plain")
+    {
+        TEST_CYCLE() cv::pyrDown(src, dst);
 
-            TestSystem::instance().ExpectedMatNear(dst, ocl_dst, dst.depth() == CV_32F ? 1e-4f : 1.0f);
-        }
+        SANITY_CHECK(dst);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }
 
 ///////////// pyrUp ////////////////////////
-PERFTEST(pyrUp)
-{
-    Mat src, dst, ocl_dst;
-    int all_type[] = {CV_8UC1, CV_8UC4};
-    std::string type_name[] = {"CV_8UC1", "CV_8UC4"};
 
-    for (int size = 500; size <= 2000; size *= 2)
-    {
-        for (size_t j = 0; j < sizeof(all_type) / sizeof(int); j++)
-        {
-            SUBTEST << size << 'x' << size << "; " << type_name[j] ;
+typedef pyrDownMatType pyrUpMatType;
+typedef tuple<Size, pyrUpMatType> pyrUpParams;
+typedef TestBaseWithParam<pyrUpParams> pyrUpFixture;
 
-            gen(src, size, size, all_type[j], 0, 256);
+PERF_TEST_P(pyrUpFixture, pyrUp,
+            ::testing::Combine(OCL_TYPICAL_MAT_SIZES,
+                               pyrUpMatType::all()))
+{
+    // getting params
+    pyrUpParams params = GetParam();
+    const Size srcSize = get<0>(params);
+    const int type = get<1>(params);
 
-            pyrUp(src, dst);
+    std::string impl = getSelectedImpl();
 
-            CPU_ON;
-            pyrUp(src, dst);
-            CPU_OFF;
+    // creating src data
+    Mat src(srcSize, type), dst;
+    Size dstSize(srcSize.height << 1, srcSize.width << 1);
+    dst.create(dstSize, type);
+    declare.in(src).out(dst);
 
-            ocl::oclMat d_src(src);
-            ocl::oclMat d_dst;
+    // select implementation
+    if (impl == "ocl")
+    {
+        ocl::oclMat oclSrc(src), oclDst(dstSize, type);
 
-            WARMUP_ON;
-            ocl::pyrUp(d_src, d_dst);
-            WARMUP_OFF;
+        TEST_CYCLE() cv::ocl::pyrDown(oclSrc, oclDst);
 
-            GPU_ON;
-            ocl::pyrUp(d_src, d_dst);
-            GPU_OFF;
+        oclDst.download(dst);
 
-            GPU_FULL_ON;
-            d_src.upload(src);
-            ocl::pyrUp(d_src, d_dst);
-            d_dst.download(ocl_dst);
-            GPU_FULL_OFF;
+        SANITY_CHECK(dst);
+    }
+    else if (impl == "plain")
+    {
+        TEST_CYCLE() cv::pyrDown(src, dst);
 
-            TestSystem::instance().ExpectedMatNear(dst, ocl_dst, (src.depth() == CV_32F ? 1e-4f : 1.0));
-        }
+        SANITY_CHECK(dst);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }

modules/ocl/perf/perf_split_merge.cpp

@@ -45,110 +45,120 @@
 //M*/
 #include "perf_precomp.hpp"
 
+using namespace perf;
+using std::tr1::tuple;
+using std::tr1::get;
+
 ///////////// Merge////////////////////////
-PERFTEST(Merge)
-{
-    Mat dst, ocl_dst;
-    ocl::oclMat d_dst;
 
-    int channels = 4;
-    int all_type[] = {CV_8UC1, CV_32FC1};
-    std::string type_name[] = {"CV_8UC1", "CV_32FC1"};
+CV_ENUM(MergeMatType, CV_8U, CV_32F)
 
-    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] ;
-            Size size1 = Size(size, size);
-            std::vector<Mat> src(channels);
-
-            for (int i = 0; i < channels; ++i)
-            {
-                src[i] = Mat(size1, all_type[j], cv::Scalar::all(i));
-            }
-
-            merge(src, dst);
-
-            CPU_ON;
-            merge(src, dst);
-            CPU_OFF;
-
-            std::vector<ocl::oclMat> d_src(channels);
-
-            for (int i = 0; i < channels; ++i)
-            {
-                d_src[i] = ocl::oclMat(size1, all_type[j], cv::Scalar::all(i));
-            }
-
-            WARMUP_ON;
-            ocl::merge(d_src, d_dst);
-            WARMUP_OFF;
-
-            GPU_ON;
-            ocl::merge(d_src, d_dst);
-            GPU_OFF;
-
-            GPU_FULL_ON;
-            for (int i = 0; i < channels; ++i)
-            {
-                d_src[i] = ocl::oclMat(size1, all_type[j], cv::Scalar::all(i));
-            }
-            ocl::merge(d_src, d_dst);
-            d_dst.download(ocl_dst);
-            GPU_FULL_OFF;
-
-            TestSystem::instance().ExpectedMatNear(dst, ocl_dst, 0.0);
-        }
+typedef tuple<Size, MergeMatType> MergeParams;
+typedef TestBaseWithParam<MergeParams> MergeFixture;
 
-    }
-}
-
-///////////// Split////////////////////////
-PERFTEST(Split)
+PERF_TEST_P(MergeFixture, Merge,
+            ::testing::Combine(::testing::Values(OCL_SIZE_1000, OCL_SIZE_2000),
+                               MergeMatType::all()))
 {
-    //int channels = 4;
-    int all_type[] = {CV_8UC1, CV_32FC1};
-    std::string type_name[] = {"CV_8UC1", "CV_32FC1"};
+    // getting params
+    MergeParams params = GetParam();
+    const Size srcSize = get<0>(params);
+    const int depth = get<1>(params), channels = 3;
+
+    std::string impl = getSelectedImpl();
+
+    // creating src data
+    const int dstType = CV_MAKE_TYPE(depth, channels);
+    Mat dst(srcSize, dstType);
+    vector<Mat> src(channels);
+    for (vector<Mat>::iterator i = src.begin(), end = src.end(); i != end; ++i)
+    {
+        i->create(srcSize, CV_MAKE_TYPE(depth, 1));
+        declare.in(*i, WARMUP_RNG);
+    }
+    declare.out(dst);
 
-    for (int size = Min_Size; size <= Max_Size; size *= Multiple)
+    // select implementation
+    if (impl == "ocl")
     {
-        for (size_t j = 0; j < sizeof(all_type) / sizeof(int); j++)
-        {
-            SUBTEST << size << 'x' << size << "; " << type_name[j];
-            Size size1 = Size(size, size);
+        ocl::oclMat oclDst(srcSize, dstType);
+        vector<ocl::oclMat> oclSrc(src.size());
+        for (vector<ocl::oclMat>::size_type i = 0, end = src.size(); i < end; ++i)
+            oclSrc[i] = src[i];
 
-            Mat src(size1, CV_MAKE_TYPE(all_type[j], 4), cv::Scalar(1, 2, 3, 4));
+        TEST_CYCLE() cv::ocl::merge(oclSrc, oclDst);
 
-            std::vector<cv::Mat> dst, ocl_dst(4);
+        oclDst.download(dst);
 
-            split(src, dst);
+        SANITY_CHECK(dst);
+    }
+    else if (impl == "plain")
+    {
+        TEST_CYCLE() cv::merge(src, dst);
 
-            CPU_ON;
-            split(src, dst);
-            CPU_OFF;
+        SANITY_CHECK(dst);
+    }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
+}
 
-            ocl::oclMat d_src(size1, CV_MAKE_TYPE(all_type[j], 4), cv::Scalar(1, 2, 3, 4));
-            std::vector<cv::ocl::oclMat> d_dst;
+///////////// Split////////////////////////
 
-            WARMUP_ON;
-            ocl::split(d_src, d_dst);
-            WARMUP_OFF;
+typedef MergeMatType SplitMatType;
+typedef tuple<Size, SplitMatType> SplitParams;
+typedef TestBaseWithParam<SplitParams> SplitFixture;
 
-            GPU_ON;
-            ocl::split(d_src, d_dst);
-            GPU_OFF;
+PERF_TEST_P(SplitFixture, Split,
+            ::testing::Combine(OCL_TYPICAL_MAT_SIZES,
+                               SplitMatType::all()))
+{
+    // getting params
+    MergeParams params = GetParam();
+    const Size srcSize = get<0>(params);
+    const int depth = get<1>(params), channels = 3;
 
-            GPU_FULL_ON;
-            d_src.upload(src);
-            ocl::split(d_src, d_dst);
-            for(size_t i = 0; i < dst.size(); i++)
-                d_dst[i].download(ocl_dst[i]);
-            GPU_FULL_OFF;
+    std::string impl = getSelectedImpl();
 
-            vector<double> eps(4, 0.);
-            TestSystem::instance().ExpectMatsNear(dst, ocl_dst, eps);
-        }
+    // creating src data
+    Mat src(srcSize, CV_MAKE_TYPE(depth, channels));
+    declare.in(src, WARMUP_RNG);
 
+    // select implementation
+    if (impl == "ocl")
+    {
+        ocl::oclMat oclSrc(src);
+        vector<ocl::oclMat> oclDst(channels, ocl::oclMat(srcSize, CV_MAKE_TYPE(depth, 1)));
+
+        TEST_CYCLE() cv::ocl::split(oclSrc, oclDst);
+
+        AssertEQ(channels, 3);
+        Mat dst0, dst1, dst2;
+        oclDst[0].download(dst0);
+        oclDst[1].download(dst1);
+        oclDst[2].download(dst2);
+        SANITY_CHECK(dst0);
+        SANITY_CHECK(dst1);
+        SANITY_CHECK(dst2);
+    }
+    else if (impl == "plain")
+    {
+        vector<Mat> dst(channels, Mat(srcSize, CV_MAKE_TYPE(depth, 1)));
+        TEST_CYCLE() cv::split(src, dst);
+
+        AssertEQ(channels, 3);
+        Mat & dst0 = dst[0], & dst1 = dst[1], & dst2 = dst[2];
+        SANITY_CHECK(dst0);
+        SANITY_CHECK(dst1);
+        SANITY_CHECK(dst2);
     }
+#ifdef HAVE_OPENCV_GPU
+    else if (impl == "gpu")
+        CV_TEST_FAIL_NO_IMPL();
+#endif
+    else
+        CV_TEST_FAIL_NO_IMPL();
 }