ocl: perf tests: skip tests if we fail check for maxMemoryAllocSize

modules/ocl/perf/perf_arithm.cpp

@@ -1066,6 +1066,9 @@ PERF_TEST_P(RepeatFixture, Repeat,
     const int nx = 3, ny = 2;
     const Size dstSize(srcSize.width * nx, srcSize.height * ny);
 
+    checkDeviceMaxMemoryAllocSize(srcSize, type);
+    checkDeviceMaxMemoryAllocSize(dstSize, type);
+
     Mat src(srcSize, type), dst(dstSize, type);
     declare.in(src, WARMUP_RNG).out(dst);
 

modules/ocl/perf/perf_filters.cpp

@@ -61,6 +61,8 @@ PERF_TEST_P(BlurFixture, Blur,
     const Size srcSize = get<0>(params), ksize(3, 3);
     const int type = get<1>(params), bordertype = BORDER_CONSTANT;
 
+    checkDeviceMaxMemoryAllocSize(srcSize, type);
+
     Mat src(srcSize, type), dst(srcSize, type);
     declare.in(src, WARMUP_RNG).out(dst);
 
@@ -99,6 +101,8 @@ PERF_TEST_P(LaplacianFixture, Laplacian,
     const Size srcSize = get<0>(params);
     const int type = get<1>(params), ksize = 3;
 
+    checkDeviceMaxMemoryAllocSize(srcSize, type);
+
     Mat src(srcSize, type), dst(srcSize, type);
     declare.in(src, WARMUP_RNG).out(dst);
 
@@ -138,6 +142,8 @@ PERF_TEST_P(ErodeFixture, Erode,
     const int type = get<1>(params), ksize = 3;
     const Mat ker = getStructuringElement(MORPH_RECT, Size(ksize, ksize));
 
+    checkDeviceMaxMemoryAllocSize(srcSize, type);
+
     Mat src(srcSize, type), dst(srcSize, type);
     declare.in(src, WARMUP_RNG).out(dst).in(ker);
 
@@ -176,6 +182,8 @@ PERF_TEST_P(SobelFixture, Sobel,
     const Size srcSize = get<0>(params);
     const int type = get<1>(params), dx = 1, dy = 1;
 
+    checkDeviceMaxMemoryAllocSize(srcSize, type, sizeof(float) * 2);
+
     Mat src(srcSize, type), dst(srcSize, type);
     declare.in(src, WARMUP_RNG).out(dst);
 
@@ -217,6 +225,8 @@ PERF_TEST_P(ScharrFixture, Scharr,
     const Size srcSize = get<0>(params);
     const int type = get<1>(params), dx = 1, dy = 0;
 
+    checkDeviceMaxMemoryAllocSize(srcSize, type, sizeof(float) * 2);
+
     Mat src(srcSize, type), dst(srcSize, type);
     declare.in(src, WARMUP_RNG).out(dst);
 
@@ -258,6 +268,8 @@ PERF_TEST_P(GaussianBlurFixture, GaussianBlur,
     const Size srcSize = get<0>(params);
     const int type = get<1>(params), ksize = 7;
 
+    checkDeviceMaxMemoryAllocSize(srcSize, type);
+
     Mat src(srcSize, type), dst(srcSize, type);
     declare.in(src, WARMUP_RNG).out(dst);
 
@@ -295,6 +307,8 @@ PERF_TEST_P(filter2DFixture, filter2D,
     const Size srcSize = get<0>(params);
     const int type = get<1>(params), ksize = 3;
 
+    checkDeviceMaxMemoryAllocSize(srcSize, type);
+
     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);
@@ -335,6 +349,8 @@ PERF_TEST_P(BilateralFixture, Bilateral,
     const int type = get<1>(params), d = 7;
     const double sigmacolor = 50.0, sigmaspace = 50.0;
 
+    checkDeviceMaxMemoryAllocSize(srcSize, type);
+
     Mat src(srcSize, type), dst(srcSize, type);
     declare.in(src, WARMUP_RNG).out(dst);
 
@@ -374,6 +390,8 @@ PERF_TEST_P(adaptiveBilateralFixture, adaptiveBilateral,
     const double sigmaspace = 10.0;
     Size ksize(9, 9);
 
+    checkDeviceMaxMemoryAllocSize(srcSize, type);
+
     Mat src(srcSize, type), dst(srcSize, type);
     declare.in(src, WARMUP_RNG).out(dst);
 

modules/ocl/perf/perf_imgwarp.cpp

@@ -154,9 +154,12 @@ PERF_TEST_P(resizeFixture, resize,
     const Size srcSize = get<0>(params);
     const int type = get<1>(params), interType = get<2>(params);
     double scale = get<3>(params);
+    const Size dstSize(cvRound(srcSize.width * scale), cvRound(srcSize.height * scale));
+
+    checkDeviceMaxMemoryAllocSize(srcSize, type);
+    checkDeviceMaxMemoryAllocSize(dstSize, type);
 
     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)

modules/ocl/perf/perf_matrix_operation.cpp

@@ -63,6 +63,10 @@ PERF_TEST_P(ConvertToFixture, ConvertTo,
 
     Mat src(srcSize, type), dst;
     const int dstType = CV_MAKE_TYPE(CV_32F, src.channels());
+
+    checkDeviceMaxMemoryAllocSize(srcSize, type);
+    checkDeviceMaxMemoryAllocSize(srcSize, dstType);
+
     dst.create(srcSize, dstType);
     declare.in(src, WARMUP_RNG).out(dst);
 

modules/ocl/perf/perf_precomp.hpp

@@ -103,8 +103,30 @@ using namespace cv;
             CV_TEST_FAIL_NO_IMPL();
 #endif
 
-#define OCL_TEST_CYCLE_N(n) for(declare.iterations(n); startTimer(), next(); ocl::finish(), stopTimer())
-#define OCL_TEST_CYCLE() for(; startTimer(), next(); ocl::finish(), stopTimer())
-#define OCL_TEST_CYCLE_MULTIRUN(runsNum) for(declare.runs(runsNum); startTimer(), next(); stopTimer()) for(int r = 0; r < runsNum; ocl::finish(), ++r)
+#define OCL_TEST_CYCLE_N(n) for(declare.iterations(n); startTimer(), next(); cv::ocl::finish(), stopTimer())
+#define OCL_TEST_CYCLE() for(; startTimer(), next(); cv::ocl::finish(), stopTimer())
+#define OCL_TEST_CYCLE_MULTIRUN(runsNum) for(declare.runs(runsNum); startTimer(), next(); stopTimer()) for(int r = 0; r < runsNum; cv::ocl::finish(), ++r)
+
+namespace cvtest {
+namespace ocl {
+inline void checkDeviceMaxMemoryAllocSize(const Size& size, int type, int factor = 1)
+{
+    assert(factor > 0);
+    if (!(IMPL_OCL == perf::TestBase::getSelectedImpl()))
+        return; // OpenCL devices are not used
+    int cn = CV_MAT_CN(type);
+    int cn_ocl = cn == 3 ? 4 : cn;
+    int type_ocl = CV_MAKE_TYPE(CV_MAT_DEPTH(type), cn_ocl);
+    size_t memSize = size.area() * CV_ELEM_SIZE(type_ocl);
+    const cv::ocl::DeviceInfo& devInfo = cv::ocl::Context::getContext()->getDeviceInfo();
+    if (memSize * factor >= devInfo.maxMemAllocSize)
+    {
+        throw perf::TestBase::PerfSkipTestException();
+    }
+}
+} // namespace cvtest::ocl
+} // namespace cvtest
+
+using namespace cvtest::ocl;
 
 #endif

modules/ocl/perf/perf_pyramid.cpp

@@ -60,9 +60,12 @@ PERF_TEST_P(pyrDownFixture, pyrDown,
     const Size_MatType_t params = GetParam();
     const Size srcSize = get<0>(params);
     const int type = get<1>(params);
+    Size dstSize((srcSize.height + 1) >> 1, (srcSize.width + 1) >> 1);
+
+    checkDeviceMaxMemoryAllocSize(srcSize, type);
+    checkDeviceMaxMemoryAllocSize(dstSize, type);
 
     Mat src(srcSize, type), dst;
-    Size dstSize((srcSize.height + 1) >> 1, (srcSize.width + 1) >> 1);
     dst.create(dstSize, type);
     declare.in(src, WARMUP_RNG).out(dst);
 
@@ -97,9 +100,12 @@ PERF_TEST_P(pyrUpFixture, pyrUp,
     const Size_MatType_t params = GetParam();
     const Size srcSize = get<0>(params);
     const int type = get<1>(params);
+    Size dstSize(srcSize.height << 1, srcSize.width << 1);
+
+    checkDeviceMaxMemoryAllocSize(srcSize, type);
+    checkDeviceMaxMemoryAllocSize(dstSize, type);
 
     Mat src(srcSize, type), dst;
-    Size dstSize(srcSize.height << 1, srcSize.width << 1);
     dst.create(dstSize, type);
     declare.in(src, WARMUP_RNG).out(dst);
 

modules/ocl/perf/perf_split_merge.cpp

@@ -60,8 +60,10 @@ PERF_TEST_P(MergeFixture, Merge,
     const Size_MatType_t params = GetParam();
     const Size srcSize = get<0>(params);
     const int depth = get<1>(params), channels = 3;
-
     const int dstType = CV_MAKE_TYPE(depth, channels);
+
+    checkDeviceMaxMemoryAllocSize(srcSize, dstType);
+
     Mat dst(srcSize, dstType);
     vector<Mat> src(channels);
     for (vector<Mat>::iterator i = src.begin(), end = src.end(); i != end; ++i)
@@ -105,8 +107,11 @@ PERF_TEST_P(SplitFixture, Split,
     const Size_MatType_t params = GetParam();
     const Size srcSize = get<0>(params);
     const int depth = get<1>(params), channels = 3;
+    const int type = CV_MAKE_TYPE(depth, channels);
+
+    checkDeviceMaxMemoryAllocSize(srcSize, type);
 
-    Mat src(srcSize, CV_MAKE_TYPE(depth, channels));
+    Mat src(srcSize, type);
     declare.in(src, WARMUP_RNG);
 
     if (RUN_OCL_IMPL)