added some performance tests

diff --git a/modules/core/perf/opencl/perf_arithm.cpp b/modules/core/perf/opencl/perf_arithm.cpp
index cb06ac4..5bc2a5a 100644 (file)
--- a/modules/core/perf/opencl/perf_arithm.cpp
+++ b/modules/core/perf/opencl/perf_arithm.cpp
@@ -814,6 +814,82 @@ OCL_PERF_TEST_P(NormalizeFixture, Normalize,
     SANITY_CHECK(dst, 5e-2);
 }
 
+///////////// ConvertScaleAbs ////////////////////////
+
+typedef Size_MatType ConvertScaleAbsFixture;
+
+OCL_PERF_TEST_P(ConvertScaleAbsFixture, ConvertScaleAbs,
+                ::testing::Combine(OCL_TEST_SIZES, OCL_TEST_TYPES))
+{
+    const Size_MatType_t params = GetParam();
+    const Size srcSize = get<0>(params);
+    const int type = get<1>(params), cn = CV_MAT_CN(type);
+
+    checkDeviceMaxMemoryAllocSize(srcSize, type);
+
+    UMat src(srcSize, type), dst(srcSize, CV_8UC(cn));
+    declare.in(src, WARMUP_RNG).out(dst);
+
+    OCL_TEST_CYCLE() cv::convertScaleAbs(src, dst, 0.5, 2);
+
+    SANITY_CHECK(dst);
+}
+
+///////////// PatchNaNs ////////////////////////
+
+typedef Size_MatType PatchNaNsFixture;
+
+OCL_PERF_TEST_P(PatchNaNsFixture, PatchNaNs,
+                ::testing::Combine(OCL_TEST_SIZES, OCL_PERF_ENUM(CV_32FC1, CV_32FC4)))
+{
+    const Size_MatType_t params = GetParam();
+    Size srcSize = get<0>(params);
+    const int type = get<1>(params), cn = CV_MAT_CN(type);
+
+    checkDeviceMaxMemoryAllocSize(srcSize, type);
+
+    UMat src(srcSize, type);
+    declare.in(src, WARMUP_RNG).out(src);
+
+    // generating NaNs
+    {
+        Mat src_ = src.getMat(ACCESS_RW);
+        srcSize.width *= cn;
+        for (int y = 0; y < srcSize.height; ++y)
+        {
+            float * const ptr = src_.ptr<float>(y);
+            for (int x = 0; x < srcSize.width; ++x)
+                ptr[x] = (x + y) % 2 == 0 ? std::numeric_limits<float>::quiet_NaN() : ptr[x];
+        }
+    }
+
+    OCL_TEST_CYCLE() cv::patchNaNs(src, 17.7);
+
+    SANITY_CHECK(src);
+}
+
+
+///////////// ScaleAdd ////////////////////////
+
+typedef Size_MatType ScaleAddFixture;
+
+OCL_PERF_TEST_P(ScaleAddFixture, ScaleAdd,
+                ::testing::Combine(OCL_TEST_SIZES, OCL_TEST_TYPES))
+{
+    const Size_MatType_t params = GetParam();
+    const Size srcSize = get<0>(params);
+    const int type = get<1>(params);
+
+    checkDeviceMaxMemoryAllocSize(srcSize, type);
+
+    UMat src1(srcSize, type), src2(srcSize, type), dst(srcSize, type);
+    declare.in(src1, src2, WARMUP_RNG).out(dst);
+
+    OCL_TEST_CYCLE() cv::scaleAdd(src1, 0.6, src2, dst);
+
+    SANITY_CHECK(dst, 1e-6);
+}
+
 } } // namespace cvtest::ocl
 
 #endif // HAVE_OPENCL

diff --git a/modules/core/src/mathfuncs.cpp b/modules/core/src/mathfuncs.cpp
index 90e0d74..15bd93e 100644 (file)
--- a/modules/core/src/mathfuncs.cpp
+++ b/modules/core/src/mathfuncs.cpp
@@ -2375,7 +2375,7 @@ static bool ocl_patchNaNs( InputOutputArray _a, float value )
     int cn = a.channels();
 
     k.args(ocl::KernelArg::ReadOnlyNoSize(a),
-           ocl::KernelArg::WriteOnly(a), (float)value);
+           ocl::KernelArg::WriteOnly(a, cn), (float)value);
 
     size_t globalsize[2] = { a.cols * cn, a.rows };
     return k.run(2, globalsize, NULL, false);