diff --git a/modules/core/src/cuda/gpu_mat.cu b/modules/core/src/cuda/gpu_mat.cu
index d9332b7332..af0a4f5d0d 100644
--- a/modules/core/src/cuda/gpu_mat.cu
+++ b/modules/core/src/cuda/gpu_mat.cu
@@ -216,7 +216,7 @@ namespace
     template <typename T>
     void copyWithMask(const GpuMat& src, const GpuMat& dst, const GpuMat& mask, Stream& stream)
     {
-        gridTransform_< CopyToPolicy<sizeof(typename VecTraits<T>::elem_type)> >(globPtr<T>(src), globPtr<T>(dst), identity<T>(), globPtr<uchar>(mask), stream);
+        gridTransformUnary_< CopyToPolicy<sizeof(typename VecTraits<T>::elem_type)> >(globPtr<T>(src), globPtr<T>(dst), identity<T>(), globPtr<uchar>(mask), stream);
     }
 }
 
@@ -268,14 +268,14 @@ namespace
     void setToWithOutMask(const GpuMat& mat, Scalar _scalar, Stream& stream)
     {
         Scalar_<typename VecTraits<T>::elem_type> scalar = _scalar;
-        gridTransform(constantPtr(VecTraits<T>::make(scalar.val), mat.rows, mat.cols), globPtr<T>(mat), identity<T>(), stream);
+        gridTransformUnary(constantPtr(VecTraits<T>::make(scalar.val), mat.rows, mat.cols), globPtr<T>(mat), identity<T>(), stream);
     }
 
     template <typename T>
     void setToWithMask(const GpuMat& mat, const GpuMat& mask, Scalar _scalar, Stream& stream)
     {
         Scalar_<typename VecTraits<T>::elem_type> scalar = _scalar;
-        gridTransform(constantPtr(VecTraits<T>::make(scalar.val), mat.rows, mat.cols), globPtr<T>(mat), identity<T>(), globPtr<uchar>(mask), stream);
+        gridTransformUnary(constantPtr(VecTraits<T>::make(scalar.val), mat.rows, mat.cols), globPtr<T>(mat), identity<T>(), globPtr<uchar>(mask), stream);
     }
 }
 
@@ -382,7 +382,7 @@ namespace
         typedef typename LargerType<src_elem_type, float>::type larger_elem_type;
         typedef typename LargerType<float, dst_elem_type>::type scalar_type;
 
-        gridTransform_< ConvertToPolicy<scalar_type> >(globPtr<T>(src), globPtr<D>(dst), saturate_cast_func<T, D>(), stream);
+        gridTransformUnary_< ConvertToPolicy<scalar_type> >(globPtr<T>(src), globPtr<D>(dst), saturate_cast_func<T, D>(), stream);
     }
 
     template <typename T, typename D, typename S> struct Convertor : unary_function<T, D>
@@ -408,7 +408,7 @@ namespace
         op.alpha = cv::saturate_cast<scalar_type>(alpha);
         op.beta = cv::saturate_cast<scalar_type>(beta);
 
-        gridTransform_< ConvertToPolicy<scalar_type> >(globPtr<T>(src), globPtr<D>(dst), op, stream);
+        gridTransformUnary_< ConvertToPolicy<scalar_type> >(globPtr<T>(src), globPtr<D>(dst), op, stream);
     }
 }
 
diff --git a/modules/cudaarithm/CMakeLists.txt b/modules/cudaarithm/CMakeLists.txt
index 67a7ff922f..f44c283c09 100644
--- a/modules/cudaarithm/CMakeLists.txt
+++ b/modules/cudaarithm/CMakeLists.txt
@@ -6,7 +6,7 @@ set(the_description "CUDA-accelerated Operations on Matrices")
 
 ocv_warnings_disable(CMAKE_CXX_FLAGS /wd4127 /wd4324 /wd4512 -Wundef -Wmissing-declarations)
 
-ocv_add_module(cudaarithm opencv_core OPTIONAL opencv_cudalegacy)
+ocv_add_module(cudaarithm opencv_core OPTIONAL opencv_cudev)
 
 ocv_module_include_directories()
 ocv_glob_module_sources()
diff --git a/modules/cudaarithm/perf/perf_arithm.cpp b/modules/cudaarithm/perf/perf_arithm.cpp
index d0f3e66176..9004155016 100644
--- a/modules/cudaarithm/perf/perf_arithm.cpp
+++ b/modules/cudaarithm/perf/perf_arithm.cpp
@@ -248,60 +248,3 @@ PERF_TEST_P(Sz_KernelSz_Ccorr, Convolve,
         CPU_SANITY_CHECK(dst);
     }
 }
-
-//////////////////////////////////////////////////////////////////////
-// Integral
-
-PERF_TEST_P(Sz, Integral,
-            CUDA_TYPICAL_MAT_SIZES)
-{
-    const cv::Size size = GetParam();
-
-    cv::Mat src(size, CV_8UC1);
-    declare.in(src, WARMUP_RNG);
-
-    if (PERF_RUN_CUDA())
-    {
-        const cv::cuda::GpuMat d_src(src);
-        cv::cuda::GpuMat dst;
-        cv::cuda::GpuMat d_buf;
-
-        TEST_CYCLE() cv::cuda::integral(d_src, dst, d_buf);
-
-        CUDA_SANITY_CHECK(dst);
-    }
-    else
-    {
-        cv::Mat dst;
-
-        TEST_CYCLE() cv::integral(src, dst);
-
-        CPU_SANITY_CHECK(dst);
-    }
-}
-
-//////////////////////////////////////////////////////////////////////
-// IntegralSqr
-
-PERF_TEST_P(Sz, IntegralSqr,
-            CUDA_TYPICAL_MAT_SIZES)
-{
-    const cv::Size size = GetParam();
-
-    cv::Mat src(size, CV_8UC1);
-    declare.in(src, WARMUP_RNG);
-
-    if (PERF_RUN_CUDA())
-    {
-        const cv::cuda::GpuMat d_src(src);
-        cv::cuda::GpuMat dst, buf;
-
-        TEST_CYCLE() cv::cuda::sqrIntegral(d_src, dst, buf);
-
-        CUDA_SANITY_CHECK(dst);
-    }
-    else
-    {
-        FAIL_NO_CPU();
-    }
-}
diff --git a/modules/cudaarithm/perf/perf_reductions.cpp b/modules/cudaarithm/perf/perf_reductions.cpp
index fe72795560..470df48a3f 100644
--- a/modules/cudaarithm/perf/perf_reductions.cpp
+++ b/modules/cudaarithm/perf/perf_reductions.cpp
@@ -373,7 +373,7 @@ PERF_TEST_P(Sz_Depth_Cn_Code_Dim, Reduce,
         const cv::cuda::GpuMat d_src(src);
         cv::cuda::GpuMat dst;
 
-        TEST_CYCLE() cv::cuda::reduce(d_src, dst, dim, reduceOp);
+        TEST_CYCLE() cv::cuda::reduce(d_src, dst, dim, reduceOp, CV_32F);
 
         CUDA_SANITY_CHECK(dst);
     }
@@ -381,7 +381,7 @@ PERF_TEST_P(Sz_Depth_Cn_Code_Dim, Reduce,
     {
         cv::Mat dst;
 
-        TEST_CYCLE() cv::reduce(src, dst, dim, reduceOp);
+        TEST_CYCLE() cv::reduce(src, dst, dim, reduceOp, CV_32F);
 
         CPU_SANITY_CHECK(dst);
     }
@@ -465,3 +465,60 @@ PERF_TEST_P(Sz, MeanStdDev,
         SANITY_CHECK(cpu_stddev);
     }
 }
+
+//////////////////////////////////////////////////////////////////////
+// Integral
+
+PERF_TEST_P(Sz, Integral,
+            CUDA_TYPICAL_MAT_SIZES)
+{
+    const cv::Size size = GetParam();
+
+    cv::Mat src(size, CV_8UC1);
+    declare.in(src, WARMUP_RNG);
+
+    if (PERF_RUN_CUDA())
+    {
+        const cv::cuda::GpuMat d_src(src);
+        cv::cuda::GpuMat dst;
+        cv::cuda::GpuMat d_buf;
+
+        TEST_CYCLE() cv::cuda::integral(d_src, dst, d_buf);
+
+        CUDA_SANITY_CHECK(dst);
+    }
+    else
+    {
+        cv::Mat dst;
+
+        TEST_CYCLE() cv::integral(src, dst);
+
+        CPU_SANITY_CHECK(dst);
+    }
+}
+
+//////////////////////////////////////////////////////////////////////
+// IntegralSqr
+
+PERF_TEST_P(Sz, IntegralSqr,
+            CUDA_TYPICAL_MAT_SIZES)
+{
+    const cv::Size size = GetParam();
+
+    cv::Mat src(size, CV_8UC1);
+    declare.in(src, WARMUP_RNG);
+
+    if (PERF_RUN_CUDA())
+    {
+        const cv::cuda::GpuMat d_src(src);
+        cv::cuda::GpuMat dst, buf;
+
+        TEST_CYCLE() cv::cuda::sqrIntegral(d_src, dst, buf);
+
+        CUDA_SANITY_CHECK(dst);
+    }
+    else
+    {
+        FAIL_NO_CPU();
+    }
+}
diff --git a/modules/cudaarithm/src/arithm.cpp b/modules/cudaarithm/src/arithm.cpp
index 6f7417ae35..4fe38e935e 100644
--- a/modules/cudaarithm/src/arithm.cpp
+++ b/modules/cudaarithm/src/arithm.cpp
@@ -292,95 +292,6 @@ void cv::cuda::gemm(InputArray _src1, InputArray _src2, double alpha, InputArray
 #endif
 }
 
-//////////////////////////////////////////////////////////////////////////////
-// mulSpectrums
-
-#ifdef HAVE_CUFFT
-
-namespace cv { namespace cuda { namespace device
-{
-    void mulSpectrums(const PtrStep<cufftComplex> a, const PtrStep<cufftComplex> b, PtrStepSz<cufftComplex> c, cudaStream_t stream);
-
-    void mulSpectrums_CONJ(const PtrStep<cufftComplex> a, const PtrStep<cufftComplex> b, PtrStepSz<cufftComplex> c, cudaStream_t stream);
-}}}
-
-#endif
-
-void cv::cuda::mulSpectrums(InputArray _src1, InputArray _src2, OutputArray _dst, int flags, bool conjB, Stream& stream)
-{
-#ifndef HAVE_CUFFT
-    (void) _src1;
-    (void) _src2;
-    (void) _dst;
-    (void) flags;
-    (void) conjB;
-    (void) stream;
-    throw_no_cuda();
-#else
-    (void) flags;
-
-    typedef void (*Caller)(const PtrStep<cufftComplex>, const PtrStep<cufftComplex>, PtrStepSz<cufftComplex>, cudaStream_t stream);
-    static Caller callers[] = { device::mulSpectrums, device::mulSpectrums_CONJ };
-
-    GpuMat src1 = _src1.getGpuMat();
-    GpuMat src2 = _src2.getGpuMat();
-
-    CV_Assert( src1.type() == src2.type() && src1.type() == CV_32FC2 );
-    CV_Assert( src1.size() == src2.size() );
-
-    _dst.create(src1.size(), CV_32FC2);
-    GpuMat dst = _dst.getGpuMat();
-
-    Caller caller = callers[(int)conjB];
-    caller(src1, src2, dst, StreamAccessor::getStream(stream));
-#endif
-}
-
-//////////////////////////////////////////////////////////////////////////////
-// mulAndScaleSpectrums
-
-#ifdef HAVE_CUFFT
-
-namespace cv { namespace cuda { namespace device
-{
-    void mulAndScaleSpectrums(const PtrStep<cufftComplex> a, const PtrStep<cufftComplex> b, float scale, PtrStepSz<cufftComplex> c, cudaStream_t stream);
-
-    void mulAndScaleSpectrums_CONJ(const PtrStep<cufftComplex> a, const PtrStep<cufftComplex> b, float scale, PtrStepSz<cufftComplex> c, cudaStream_t stream);
-}}}
-
-#endif
-
-void cv::cuda::mulAndScaleSpectrums(InputArray _src1, InputArray _src2, OutputArray _dst, int flags, float scale, bool conjB, Stream& stream)
-{
-#ifndef HAVE_CUFFT
-    (void) _src1;
-    (void) _src2;
-    (void) _dst;
-    (void) flags;
-    (void) scale;
-    (void) conjB;
-    (void) stream;
-    throw_no_cuda();
-#else
-    (void)flags;
-
-    typedef void (*Caller)(const PtrStep<cufftComplex>, const PtrStep<cufftComplex>, float scale, PtrStepSz<cufftComplex>, cudaStream_t stream);
-    static Caller callers[] = { device::mulAndScaleSpectrums, device::mulAndScaleSpectrums_CONJ };
-
-    GpuMat src1 = _src1.getGpuMat();
-    GpuMat src2 = _src2.getGpuMat();
-
-    CV_Assert( src1.type() == src2.type() && src1.type() == CV_32FC2);
-    CV_Assert( src1.size() == src2.size() );
-
-    _dst.create(src1.size(), CV_32FC2);
-    GpuMat dst = _dst.getGpuMat();
-
-    Caller caller = callers[(int)conjB];
-    caller(src1, src2, scale, dst, StreamAccessor::getStream(stream));
-#endif
-}
-
 //////////////////////////////////////////////////////////////////////////////
 // dft
 
diff --git a/modules/cudaarithm/src/core.cpp b/modules/cudaarithm/src/core.cpp
index e6c14db09f..eb71d6a4ec 100644
--- a/modules/cudaarithm/src/core.cpp
+++ b/modules/cudaarithm/src/core.cpp
@@ -63,163 +63,6 @@ void cv::cuda::copyMakeBorder(InputArray, OutputArray, int, int, int, int, int,
 
 #else /* !defined (HAVE_CUDA) */
 
-////////////////////////////////////////////////////////////////////////
-// merge/split
-
-namespace cv { namespace cuda { namespace device
-{
-    namespace split_merge
-    {
-        void merge(const PtrStepSzb* src, PtrStepSzb& dst, int total_channels, size_t elem_size, const cudaStream_t& stream);
-        void split(const PtrStepSzb& src, PtrStepSzb* dst, int num_channels, size_t elem_size1, const cudaStream_t& stream);
-    }
-}}}
-
-namespace
-{
-    void merge_caller(const GpuMat* src, size_t n, OutputArray _dst, Stream& stream)
-    {
-        CV_Assert( src != 0 );
-        CV_Assert( n > 0 && n <= 4 );
-
-        const int depth = src[0].depth();
-        const Size size = src[0].size();
-
-        for (size_t i = 0; i < n; ++i)
-        {
-            CV_Assert( src[i].size() == size );
-            CV_Assert( src[i].depth() == depth );
-            CV_Assert( src[i].channels() == 1 );
-        }
-
-        if (depth == CV_64F)
-        {
-            if (!deviceSupports(NATIVE_DOUBLE))
-                CV_Error(cv::Error::StsUnsupportedFormat, "The device doesn't support double");
-        }
-
-        if (n == 1)
-        {
-            src[0].copyTo(_dst, stream);
-        }
-        else
-        {
-            _dst.create(size, CV_MAKE_TYPE(depth, (int)n));
-            GpuMat dst = _dst.getGpuMat();
-
-            PtrStepSzb src_as_devmem[4];
-            for(size_t i = 0; i < n; ++i)
-                src_as_devmem[i] = src[i];
-
-            PtrStepSzb dst_as_devmem(dst);
-            cv::cuda::device::split_merge::merge(src_as_devmem, dst_as_devmem, (int)n, CV_ELEM_SIZE(depth), StreamAccessor::getStream(stream));
-        }
-    }
-
-    void split_caller(const GpuMat& src, GpuMat* dst, Stream& stream)
-    {
-        CV_Assert( dst != 0 );
-
-        const int depth = src.depth();
-        const int num_channels = src.channels();
-
-        CV_Assert( num_channels <= 4 );
-
-        if (depth == CV_64F)
-        {
-            if (!deviceSupports(NATIVE_DOUBLE))
-                CV_Error(cv::Error::StsUnsupportedFormat, "The device doesn't support double");
-        }
-
-        if (num_channels == 1)
-        {
-            src.copyTo(dst[0], stream);
-            return;
-        }
-
-        for (int i = 0; i < num_channels; ++i)
-            dst[i].create(src.size(), depth);
-
-        PtrStepSzb dst_as_devmem[4];
-        for (int i = 0; i < num_channels; ++i)
-            dst_as_devmem[i] = dst[i];
-
-        PtrStepSzb src_as_devmem(src);
-        cv::cuda::device::split_merge::split(src_as_devmem, dst_as_devmem, num_channels, src.elemSize1(), StreamAccessor::getStream(stream));
-    }
-}
-
-void cv::cuda::merge(const GpuMat* src, size_t n, OutputArray dst, Stream& stream)
-{
-    merge_caller(src, n, dst, stream);
-}
-
-
-void cv::cuda::merge(const std::vector<GpuMat>& src, OutputArray dst, Stream& stream)
-{
-    merge_caller(&src[0], src.size(), dst, stream);
-}
-
-void cv::cuda::split(InputArray _src, GpuMat* dst, Stream& stream)
-{
-    GpuMat src = _src.getGpuMat();
-    split_caller(src, dst, stream);
-}
-
-void cv::cuda::split(InputArray _src, std::vector<GpuMat>& dst, Stream& stream)
-{
-    GpuMat src = _src.getGpuMat();
-    dst.resize(src.channels());
-    if(src.channels() > 0)
-        split_caller(src, &dst[0], stream);
-}
-
-////////////////////////////////////////////////////////////////////////
-// transpose
-
-namespace arithm
-{
-    template <typename T> void transpose(PtrStepSz<T> src, PtrStepSz<T> dst, cudaStream_t stream);
-}
-
-void cv::cuda::transpose(InputArray _src, OutputArray _dst, Stream& _stream)
-{
-    GpuMat src = _src.getGpuMat();
-
-    CV_Assert( src.elemSize() == 1 || src.elemSize() == 4 || src.elemSize() == 8 );
-
-    _dst.create( src.cols, src.rows, src.type() );
-    GpuMat dst = _dst.getGpuMat();
-
-    cudaStream_t stream = StreamAccessor::getStream(_stream);
-
-    if (src.elemSize() == 1)
-    {
-        NppStreamHandler h(stream);
-
-        NppiSize sz;
-        sz.width  = src.cols;
-        sz.height = src.rows;
-
-        nppSafeCall( nppiTranspose_8u_C1R(src.ptr<Npp8u>(), static_cast<int>(src.step),
-            dst.ptr<Npp8u>(), static_cast<int>(dst.step), sz) );
-
-        if (stream == 0)
-            cudaSafeCall( cudaDeviceSynchronize() );
-    }
-    else if (src.elemSize() == 4)
-    {
-        arithm::transpose<int>(src, dst, stream);
-    }
-    else // if (src.elemSize() == 8)
-    {
-        if (!deviceSupports(NATIVE_DOUBLE))
-            CV_Error(cv::Error::StsUnsupportedFormat, "The device doesn't support double");
-
-        arithm::transpose<double>(src, dst, stream);
-    }
-}
-
 ////////////////////////////////////////////////////////////////////////
 // flip
 
@@ -287,326 +130,4 @@ void cv::cuda::flip(InputArray _src, OutputArray _dst, int flipCode, Stream& str
     funcs[src.depth()][src.channels() - 1](src, dst, flipCode, StreamAccessor::getStream(stream));
 }
 
-////////////////////////////////////////////////////////////////////////
-// LUT
-
-#if (CUDA_VERSION >= 5000)
-
-namespace
-{
-    class LookUpTableImpl : public LookUpTable
-    {
-    public:
-        LookUpTableImpl(InputArray lut);
-
-        void transform(InputArray src, OutputArray dst, Stream& stream = Stream::Null());
-
-    private:
-        int lut_cn;
-
-        int nValues3[3];
-        const Npp32s* pValues3[3];
-        const Npp32s* pLevels3[3];
-
-        GpuMat d_pLevels;
-        GpuMat d_nppLut;
-        GpuMat d_nppLut3[3];
-    };
-
-    LookUpTableImpl::LookUpTableImpl(InputArray _lut)
-    {
-        nValues3[0] = nValues3[1] = nValues3[2] = 256;
-
-        Npp32s pLevels[256];
-        for (int i = 0; i < 256; ++i)
-            pLevels[i] = i;
-
-        d_pLevels.upload(Mat(1, 256, CV_32S, pLevels));
-        pLevels3[0] = pLevels3[1] = pLevels3[2] = d_pLevels.ptr<Npp32s>();
-
-        GpuMat lut;
-        if (_lut.kind() == _InputArray::GPU_MAT)
-        {
-            lut = _lut.getGpuMat();
-        }
-        else
-        {
-            Mat hLut = _lut.getMat();
-            CV_Assert( hLut.total() == 256 && hLut.isContinuous() );
-            lut.upload(Mat(1, 256, hLut.type(), hLut.data));
-        }
-
-        lut_cn = lut.channels();
-
-        CV_Assert( lut.depth() == CV_8U );
-        CV_Assert( lut.rows == 1 && lut.cols == 256 );
-
-        lut.convertTo(d_nppLut, CV_32S);
-
-        if (lut_cn == 1)
-        {
-            pValues3[0] = pValues3[1] = pValues3[2] = d_nppLut.ptr<Npp32s>();
-        }
-        else
-        {
-            cuda::split(d_nppLut, d_nppLut3);
-
-            pValues3[0] = d_nppLut3[0].ptr<Npp32s>();
-            pValues3[1] = d_nppLut3[1].ptr<Npp32s>();
-            pValues3[2] = d_nppLut3[2].ptr<Npp32s>();
-        }
-    }
-
-    void LookUpTableImpl::transform(InputArray _src, OutputArray _dst, Stream& _stream)
-    {
-        GpuMat src = _src.getGpuMat();
-
-        const int cn = src.channels();
-
-        CV_Assert( src.type() == CV_8UC1 || src.type() == CV_8UC3 );
-        CV_Assert( lut_cn == 1 || lut_cn == cn );
-
-        _dst.create(src.size(), src.type());
-        GpuMat dst = _dst.getGpuMat();
-
-        cudaStream_t stream = StreamAccessor::getStream(_stream);
-
-        NppStreamHandler h(stream);
-
-        NppiSize sz;
-        sz.height = src.rows;
-        sz.width = src.cols;
-
-        if (src.type() == CV_8UC1)
-        {
-            nppSafeCall( nppiLUT_Linear_8u_C1R(src.ptr<Npp8u>(), static_cast<int>(src.step),
-                dst.ptr<Npp8u>(), static_cast<int>(dst.step), sz, d_nppLut.ptr<Npp32s>(), d_pLevels.ptr<Npp32s>(), 256) );
-        }
-        else
-        {
-            nppSafeCall( nppiLUT_Linear_8u_C3R(src.ptr<Npp8u>(), static_cast<int>(src.step),
-                dst.ptr<Npp8u>(), static_cast<int>(dst.step), sz, pValues3, pLevels3, nValues3) );
-        }
-
-        if (stream == 0)
-            cudaSafeCall( cudaDeviceSynchronize() );
-    }
-}
-
-#else //  (CUDA_VERSION >= 5000)
-
-namespace
-{
-    class LookUpTableImpl : public LookUpTable
-    {
-    public:
-        LookUpTableImpl(InputArray lut);
-
-        void transform(InputArray src, OutputArray dst, Stream& stream = Stream::Null());
-
-    private:
-        int lut_cn;
-
-        Npp32s pLevels[256];
-        int nValues3[3];
-        const Npp32s* pValues3[3];
-        const Npp32s* pLevels3[3];
-
-        Mat nppLut;
-        Mat nppLut3[3];
-    };
-
-    LookUpTableImpl::LookUpTableImpl(InputArray _lut)
-    {
-        nValues3[0] = nValues3[1] = nValues3[2] = 256;
-
-        for (int i = 0; i < 256; ++i)
-            pLevels[i] = i;
-        pLevels3[0] = pLevels3[1] = pLevels3[2] = pLevels;
-
-        Mat lut;
-        if (_lut.kind() == _InputArray::GPU_MAT)
-        {
-            lut = Mat(_lut.getGpuMat());
-        }
-        else
-        {
-            Mat hLut = _lut.getMat();
-            CV_Assert( hLut.total() == 256 && hLut.isContinuous() );
-            lut = hLut;
-        }
-
-        lut_cn = lut.channels();
-
-        CV_Assert( lut.depth() == CV_8U );
-        CV_Assert( lut.rows == 1 && lut.cols == 256 );
-
-        lut.convertTo(nppLut, CV_32S);
-
-        if (lut_cn == 1)
-        {
-            pValues3[0] = pValues3[1] = pValues3[2] = nppLut.ptr<Npp32s>();
-        }
-        else
-        {
-            cv::split(nppLut, nppLut3);
-
-            pValues3[0] = nppLut3[0].ptr<Npp32s>();
-            pValues3[1] = nppLut3[1].ptr<Npp32s>();
-            pValues3[2] = nppLut3[2].ptr<Npp32s>();
-        }
-    }
-
-    void LookUpTableImpl::transform(InputArray _src, OutputArray _dst, Stream& _stream)
-    {
-        GpuMat src = _src.getGpuMat();
-
-        const int cn = src.channels();
-
-        CV_Assert( src.type() == CV_8UC1 || src.type() == CV_8UC3 );
-        CV_Assert( lut_cn == 1 || lut_cn == cn );
-
-        _dst.create(src.size(), src.type());
-        GpuMat dst = _dst.getGpuMat();
-
-        cudaStream_t stream = StreamAccessor::getStream(_stream);
-
-        NppStreamHandler h(stream);
-
-        NppiSize sz;
-        sz.height = src.rows;
-        sz.width = src.cols;
-
-        if (src.type() == CV_8UC1)
-        {
-            nppSafeCall( nppiLUT_Linear_8u_C1R(src.ptr<Npp8u>(), static_cast<int>(src.step),
-                dst.ptr<Npp8u>(), static_cast<int>(dst.step), sz, nppLut.ptr<Npp32s>(), pLevels, 256) );
-        }
-        else
-        {
-            nppSafeCall( nppiLUT_Linear_8u_C3R(src.ptr<Npp8u>(), static_cast<int>(src.step),
-                dst.ptr<Npp8u>(), static_cast<int>(dst.step), sz, pValues3, pLevels3, nValues3) );
-        }
-
-        if (stream == 0)
-            cudaSafeCall( cudaDeviceSynchronize() );
-    }
-}
-
-#endif //  (CUDA_VERSION >= 5000)
-
-Ptr<LookUpTable> cv::cuda::createLookUpTable(InputArray lut)
-{
-    return makePtr<LookUpTableImpl>(lut);
-}
-
-////////////////////////////////////////////////////////////////////////
-// copyMakeBorder
-
-namespace cv { namespace cuda { namespace device
-{
-    namespace imgproc
-    {
-        template <typename T, int cn> void copyMakeBorder_gpu(const PtrStepSzb& src, const PtrStepSzb& dst, int top, int left, int borderMode, const T* borderValue, cudaStream_t stream);
-    }
-}}}
-
-namespace
-{
-    template <typename T, int cn> void copyMakeBorder_caller(const PtrStepSzb& src, const PtrStepSzb& dst, int top, int left, int borderType, const Scalar& value, cudaStream_t stream)
-    {
-        using namespace ::cv::cuda::device::imgproc;
-
-        Scalar_<T> val(saturate_cast<T>(value[0]), saturate_cast<T>(value[1]), saturate_cast<T>(value[2]), saturate_cast<T>(value[3]));
-
-        copyMakeBorder_gpu<T, cn>(src, dst, top, left, borderType, val.val, stream);
-    }
-}
-
-#if defined __GNUC__ && __GNUC__ > 2 && __GNUC_MINOR__  > 4
-typedef Npp32s __attribute__((__may_alias__)) Npp32s_a;
-#else
-typedef Npp32s Npp32s_a;
-#endif
-
-void cv::cuda::copyMakeBorder(InputArray _src, OutputArray _dst, int top, int bottom, int left, int right, int borderType, Scalar value, Stream& _stream)
-{
-    GpuMat src = _src.getGpuMat();
-
-    CV_Assert( src.depth() <= CV_32F && src.channels() <= 4 );
-    CV_Assert( borderType == BORDER_REFLECT_101 || borderType == BORDER_REPLICATE || borderType == BORDER_CONSTANT || borderType == BORDER_REFLECT || borderType == BORDER_WRAP );
-
-    _dst.create(src.rows + top + bottom, src.cols + left + right, src.type());
-    GpuMat dst = _dst.getGpuMat();
-
-    cudaStream_t stream = StreamAccessor::getStream(_stream);
-
-    if (borderType == BORDER_CONSTANT && (src.type() == CV_8UC1 || src.type() == CV_8UC4 || src.type() == CV_32SC1 || src.type() == CV_32FC1))
-    {
-        NppiSize srcsz;
-        srcsz.width  = src.cols;
-        srcsz.height = src.rows;
-
-        NppiSize dstsz;
-        dstsz.width  = dst.cols;
-        dstsz.height = dst.rows;
-
-        NppStreamHandler h(stream);
-
-        switch (src.type())
-        {
-        case CV_8UC1:
-            {
-                Npp8u nVal = saturate_cast<Npp8u>(value[0]);
-                nppSafeCall( nppiCopyConstBorder_8u_C1R(src.ptr<Npp8u>(), static_cast<int>(src.step), srcsz,
-                    dst.ptr<Npp8u>(), static_cast<int>(dst.step), dstsz, top, left, nVal) );
-                break;
-            }
-        case CV_8UC4:
-            {
-                Npp8u nVal[] = {saturate_cast<Npp8u>(value[0]), saturate_cast<Npp8u>(value[1]), saturate_cast<Npp8u>(value[2]), saturate_cast<Npp8u>(value[3])};
-                nppSafeCall( nppiCopyConstBorder_8u_C4R(src.ptr<Npp8u>(), static_cast<int>(src.step), srcsz,
-                    dst.ptr<Npp8u>(), static_cast<int>(dst.step), dstsz, top, left, nVal) );
-                break;
-            }
-        case CV_32SC1:
-            {
-                Npp32s nVal = saturate_cast<Npp32s>(value[0]);
-                nppSafeCall( nppiCopyConstBorder_32s_C1R(src.ptr<Npp32s>(), static_cast<int>(src.step), srcsz,
-                    dst.ptr<Npp32s>(), static_cast<int>(dst.step), dstsz, top, left, nVal) );
-                break;
-            }
-        case CV_32FC1:
-            {
-                Npp32f val = saturate_cast<Npp32f>(value[0]);
-                Npp32s nVal = *(reinterpret_cast<Npp32s_a*>(&val));
-                nppSafeCall( nppiCopyConstBorder_32s_C1R(src.ptr<Npp32s>(), static_cast<int>(src.step), srcsz,
-                    dst.ptr<Npp32s>(), static_cast<int>(dst.step), dstsz, top, left, nVal) );
-                break;
-            }
-        }
-
-        if (stream == 0)
-            cudaSafeCall( cudaDeviceSynchronize() );
-    }
-    else
-    {
-        typedef void (*caller_t)(const PtrStepSzb& src, const PtrStepSzb& dst, int top, int left, int borderType, const Scalar& value, cudaStream_t stream);
-        static const caller_t callers[6][4] =
-        {
-            {   copyMakeBorder_caller<uchar, 1>  ,    copyMakeBorder_caller<uchar, 2>   ,    copyMakeBorder_caller<uchar, 3>  ,    copyMakeBorder_caller<uchar, 4>},
-            {0/*copyMakeBorder_caller<schar, 1>*/, 0/*copyMakeBorder_caller<schar, 2>*/ , 0/*copyMakeBorder_caller<schar, 3>*/, 0/*copyMakeBorder_caller<schar, 4>*/},
-            {   copyMakeBorder_caller<ushort, 1> , 0/*copyMakeBorder_caller<ushort, 2>*/,    copyMakeBorder_caller<ushort, 3> ,    copyMakeBorder_caller<ushort, 4>},
-            {   copyMakeBorder_caller<short, 1>  , 0/*copyMakeBorder_caller<short, 2>*/ ,    copyMakeBorder_caller<short, 3>  ,    copyMakeBorder_caller<short, 4>},
-            {0/*copyMakeBorder_caller<int,   1>*/, 0/*copyMakeBorder_caller<int,   2>*/ , 0/*copyMakeBorder_caller<int,   3>*/, 0/*copyMakeBorder_caller<int  , 4>*/},
-            {   copyMakeBorder_caller<float, 1>  , 0/*copyMakeBorder_caller<float, 2>*/ ,    copyMakeBorder_caller<float, 3>  ,    copyMakeBorder_caller<float ,4>}
-        };
-
-        caller_t func = callers[src.depth()][src.channels() - 1];
-        CV_Assert(func != 0);
-
-        func(src, dst, top, left, borderType, value, stream);
-    }
-}
-
 #endif /* !defined (HAVE_CUDA) */
diff --git a/modules/cudaarithm/src/cuda/absdiff_mat.cu b/modules/cudaarithm/src/cuda/absdiff_mat.cu
index ffdcbfa380..ec04f12284 100644
--- a/modules/cudaarithm/src/cuda/absdiff_mat.cu
+++ b/modules/cudaarithm/src/cuda/absdiff_mat.cu
@@ -40,43 +40,22 @@
 //
 //M*/
 
-#if !defined CUDA_DISABLER
+#include "opencv2/opencv_modules.hpp"
 
-#include "opencv2/core/cuda/common.hpp"
-#include "opencv2/core/cuda/functional.hpp"
-#include "opencv2/core/cuda/transform.hpp"
-#include "opencv2/core/cuda/saturate_cast.hpp"
-#include "opencv2/core/cuda/simd_functions.hpp"
+#ifndef HAVE_OPENCV_CUDEV
 
-#include "arithm_func_traits.hpp"
+#error "opencv_cudev is required"
 
-using namespace cv::cuda;
-using namespace cv::cuda::device;
+#else
 
-namespace arithm
+#include "opencv2/cudev.hpp"
+
+using namespace cv::cudev;
+
+void absDiffMat(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat&, double, Stream& stream, int);
+
+namespace
 {
-    struct VAbsDiff4 : binary_function<uint, uint, uint>
-    {
-        __device__ __forceinline__ uint operator ()(uint a, uint b) const
-        {
-            return vabsdiff4(a, b);
-        }
-
-        __host__ __device__ __forceinline__ VAbsDiff4() {}
-        __host__ __device__ __forceinline__ VAbsDiff4(const VAbsDiff4&) {}
-    };
-
-    struct VAbsDiff2 : binary_function<uint, uint, uint>
-    {
-        __device__ __forceinline__ uint operator ()(uint a, uint b) const
-        {
-            return vabsdiff2(a, b);
-        }
-
-        __host__ __device__ __forceinline__ VAbsDiff2() {}
-        __host__ __device__ __forceinline__ VAbsDiff2(const VAbsDiff2&) {}
-    };
-
     __device__ __forceinline__ int _abs(int a)
     {
         return ::abs(a);
@@ -90,58 +69,120 @@ namespace arithm
         return ::fabs(a);
     }
 
-    template <typename T> struct AbsDiffMat : binary_function<T, T, T>
+    template <typename T> struct AbsDiffOp1 : binary_function<T, T, T>
     {
         __device__ __forceinline__ T operator ()(T a, T b) const
         {
             return saturate_cast<T>(_abs(a - b));
         }
-
-        __host__ __device__ __forceinline__ AbsDiffMat() {}
-        __host__ __device__ __forceinline__ AbsDiffMat(const AbsDiffMat&) {}
-    };
-}
-
-namespace cv { namespace cuda { namespace device
-{
-    template <> struct TransformFunctorTraits< arithm::VAbsDiff4 > : arithm::ArithmFuncTraits<sizeof(uint), sizeof(uint)>
-    {
     };
 
-    template <> struct TransformFunctorTraits< arithm::VAbsDiff2 > : arithm::ArithmFuncTraits<sizeof(uint), sizeof(uint)>
+    template <typename ScalarDepth> struct TransformPolicy : DefaultTransformPolicy
     {
     };
-
-    template <typename T> struct TransformFunctorTraits< arithm::AbsDiffMat<T> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(T)>
+    template <> struct TransformPolicy<double> : DefaultTransformPolicy
     {
+        enum {
+            shift = 1
+        };
     };
-}}}
-
-namespace arithm
-{
-    void absDiffMat_v4(PtrStepSz<uint> src1, PtrStepSz<uint> src2, PtrStepSz<uint> dst, cudaStream_t stream)
-    {
-        device::transform(src1, src2, dst, VAbsDiff4(), WithOutMask(), stream);
-    }
-
-    void absDiffMat_v2(PtrStepSz<uint> src1, PtrStepSz<uint> src2, PtrStepSz<uint> dst, cudaStream_t stream)
-    {
-        device::transform(src1, src2, dst, VAbsDiff2(), WithOutMask(), stream);
-    }
 
     template <typename T>
-    void absDiffMat(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream)
+    void absDiffMat_v1(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, Stream& stream)
     {
-        device::transform((PtrStepSz<T>) src1, (PtrStepSz<T>) src2, (PtrStepSz<T>) dst, AbsDiffMat<T>(), WithOutMask(), stream);
+        gridTransformBinary_< TransformPolicy<T> >(globPtr<T>(src1), globPtr<T>(src2), globPtr<T>(dst), AbsDiffOp1<T>(), stream);
     }
 
-    template void absDiffMat<uchar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template void absDiffMat<schar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template void absDiffMat<ushort>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template void absDiffMat<short>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template void absDiffMat<int>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template void absDiffMat<float>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template void absDiffMat<double>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
+    struct AbsDiffOp2 : binary_function<uint, uint, uint>
+    {
+        __device__ __forceinline__ uint operator ()(uint a, uint b) const
+        {
+            return vabsdiff2(a, b);
+        }
+    };
+
+    void absDiffMat_v2(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, Stream& stream)
+    {
+        const int vcols = src1.cols >> 1;
+
+        GlobPtrSz<uint> src1_ = globPtr((uint*) src1.data, src1.step, src1.rows, vcols);
+        GlobPtrSz<uint> src2_ = globPtr((uint*) src2.data, src2.step, src1.rows, vcols);
+        GlobPtrSz<uint> dst_ = globPtr((uint*) dst.data, dst.step, src1.rows, vcols);
+
+        gridTransformBinary(src1_, src2_, dst_, AbsDiffOp2(), stream);
+    }
+
+    struct AbsDiffOp4 : binary_function<uint, uint, uint>
+    {
+        __device__ __forceinline__ uint operator ()(uint a, uint b) const
+        {
+            return vabsdiff4(a, b);
+        }
+    };
+
+    void absDiffMat_v4(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, Stream& stream)
+    {
+        const int vcols = src1.cols >> 2;
+
+        GlobPtrSz<uint> src1_ = globPtr((uint*) src1.data, src1.step, src1.rows, vcols);
+        GlobPtrSz<uint> src2_ = globPtr((uint*) src2.data, src2.step, src1.rows, vcols);
+        GlobPtrSz<uint> dst_ = globPtr((uint*) dst.data, dst.step, src1.rows, vcols);
+
+        gridTransformBinary(src1_, src2_, dst_, AbsDiffOp4(), stream);
+    }
 }
 
-#endif // CUDA_DISABLER
+void absDiffMat(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat&, double, Stream& stream, int)
+{
+    typedef void (*func_t)(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, Stream& stream);
+    static const func_t funcs[] =
+    {
+        absDiffMat_v1<uchar>,
+        absDiffMat_v1<schar>,
+        absDiffMat_v1<ushort>,
+        absDiffMat_v1<short>,
+        absDiffMat_v1<int>,
+        absDiffMat_v1<float>,
+        absDiffMat_v1<double>
+    };
+
+    const int depth = src1.depth();
+
+    CV_DbgAssert( depth <= CV_64F );
+
+    GpuMat src1_ = src1.reshape(1);
+    GpuMat src2_ = src2.reshape(1);
+    GpuMat dst_ = dst.reshape(1);
+
+    if (depth == CV_8U || depth == CV_16U)
+    {
+        const intptr_t src1ptr = reinterpret_cast<intptr_t>(src1_.data);
+        const intptr_t src2ptr = reinterpret_cast<intptr_t>(src2_.data);
+        const intptr_t dstptr = reinterpret_cast<intptr_t>(dst_.data);
+
+        const bool isAllAligned = (src1ptr & 31) == 0 && (src2ptr & 31) == 0 && (dstptr & 31) == 0;
+
+        if (isAllAligned)
+        {
+            if (depth == CV_8U && (src1_.cols & 3) == 0)
+            {
+                absDiffMat_v4(src1_, src2_, dst_, stream);
+                return;
+            }
+            else if (depth == CV_16U && (src1_.cols & 1) == 0)
+            {
+                absDiffMat_v2(src1_, src2_, dst_, stream);
+                return;
+            }
+        }
+    }
+
+    const func_t func = funcs[depth];
+
+    if (!func)
+        CV_Error(cv::Error::StsUnsupportedFormat, "Unsupported combination of source and destination types");
+
+    func(src1_, src2_, dst_, stream);
+}
+
+#endif
diff --git a/modules/cudaarithm/src/cuda/absdiff_scalar.cu b/modules/cudaarithm/src/cuda/absdiff_scalar.cu
index ad30bffda6..3ffd0661b7 100644
--- a/modules/cudaarithm/src/cuda/absdiff_scalar.cu
+++ b/modules/cudaarithm/src/cuda/absdiff_scalar.cu
@@ -40,59 +40,71 @@
 //
 //M*/
 
-#if !defined CUDA_DISABLER
+#include "opencv2/opencv_modules.hpp"
 
-#include "opencv2/core/cuda/common.hpp"
-#include "opencv2/core/cuda/functional.hpp"
-#include "opencv2/core/cuda/transform.hpp"
-#include "opencv2/core/cuda/saturate_cast.hpp"
-#include "opencv2/core/cuda/simd_functions.hpp"
+#ifndef HAVE_OPENCV_CUDEV
 
-#include "arithm_func_traits.hpp"
+#error "opencv_cudev is required"
 
-using namespace cv::cuda;
-using namespace cv::cuda::device;
+#else
 
-namespace arithm
+#include "opencv2/cudev.hpp"
+
+using namespace cv::cudev;
+
+void absDiffScalar(const GpuMat& src, cv::Scalar val, bool, GpuMat& dst, const GpuMat&, double, Stream& stream, int);
+
+namespace
 {
-    template <typename T, typename S> struct AbsDiffScalar : unary_function<T, T>
+    template <typename T, typename S> struct AbsDiffScalarOp : unary_function<T, T>
     {
         S val;
 
-        __host__ explicit AbsDiffScalar(S val_) : val(val_) {}
-
         __device__ __forceinline__ T operator ()(T a) const
         {
             abs_func<S> f;
             return saturate_cast<T>(f(a - val));
         }
     };
-}
 
-namespace cv { namespace cuda { namespace device
-{
-    template <typename T, typename S> struct TransformFunctorTraits< arithm::AbsDiffScalar<T, S> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(T)>
+    template <typename ScalarDepth> struct TransformPolicy : DefaultTransformPolicy
     {
     };
-}}}
-
-namespace arithm
-{
-    template <typename T, typename S>
-    void absDiffScalar(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream)
+    template <> struct TransformPolicy<double> : DefaultTransformPolicy
     {
-        AbsDiffScalar<T, S> op(static_cast<S>(val));
+        enum {
+            shift = 1
+        };
+    };
 
-        device::transform((PtrStepSz<T>) src1, (PtrStepSz<T>) dst, op, WithOutMask(), stream);
+    template <typename SrcType, typename ScalarDepth>
+    void absDiffScalarImpl(const GpuMat& src, double value, GpuMat& dst, Stream& stream)
+    {
+        AbsDiffScalarOp<SrcType, ScalarDepth> op;
+        op.val = static_cast<ScalarDepth>(value);
+        gridTransformUnary_< TransformPolicy<ScalarDepth> >(globPtr<SrcType>(src), globPtr<SrcType>(dst), op, stream);
     }
-
-    template void absDiffScalar<uchar, float>(PtrStepSzb src1, double src2, PtrStepSzb dst, cudaStream_t stream);
-    template void absDiffScalar<schar, float>(PtrStepSzb src1, double src2, PtrStepSzb dst, cudaStream_t stream);
-    template void absDiffScalar<ushort, float>(PtrStepSzb src1, double src2, PtrStepSzb dst, cudaStream_t stream);
-    template void absDiffScalar<short, float>(PtrStepSzb src1, double src2, PtrStepSzb dst, cudaStream_t stream);
-    template void absDiffScalar<int, float>(PtrStepSzb src1, double src2, PtrStepSzb dst, cudaStream_t stream);
-    template void absDiffScalar<float, float>(PtrStepSzb src1, double src2, PtrStepSzb dst, cudaStream_t stream);
-    template void absDiffScalar<double, double>(PtrStepSzb src1, double src2, PtrStepSzb dst, cudaStream_t stream);
 }
 
-#endif // CUDA_DISABLER
+void absDiffScalar(const GpuMat& src, cv::Scalar val, bool, GpuMat& dst, const GpuMat&, double, Stream& stream, int)
+{
+    typedef void (*func_t)(const GpuMat& src, double val, GpuMat& dst, Stream& stream);
+    static const func_t funcs[] =
+    {
+        absDiffScalarImpl<uchar, float>,
+        absDiffScalarImpl<schar, float>,
+        absDiffScalarImpl<ushort, float>,
+        absDiffScalarImpl<short, float>,
+        absDiffScalarImpl<int, float>,
+        absDiffScalarImpl<float, float>,
+        absDiffScalarImpl<double, double>
+    };
+
+    const int depth = src.depth();
+
+    CV_DbgAssert( depth <= CV_64F );
+
+    funcs[depth](src, val[0], dst, stream);
+}
+
+#endif
diff --git a/modules/cudaarithm/src/cuda/add_mat.cu b/modules/cudaarithm/src/cuda/add_mat.cu
index 1270438208..4166cc104e 100644
--- a/modules/cudaarithm/src/cuda/add_mat.cu
+++ b/modules/cudaarithm/src/cuda/add_mat.cu
@@ -40,146 +40,186 @@
 //
 //M*/
 
-#if !defined CUDA_DISABLER
+#include "opencv2/opencv_modules.hpp"
 
-#include "opencv2/core/cuda/common.hpp"
-#include "opencv2/core/cuda/functional.hpp"
-#include "opencv2/core/cuda/transform.hpp"
-#include "opencv2/core/cuda/saturate_cast.hpp"
-#include "opencv2/core/cuda/simd_functions.hpp"
+#ifndef HAVE_OPENCV_CUDEV
 
-#include "arithm_func_traits.hpp"
+#error "opencv_cudev is required"
 
-using namespace cv::cuda;
-using namespace cv::cuda::device;
+#else
 
-namespace arithm
+#include "opencv2/cudev.hpp"
+
+using namespace cv::cudev;
+
+void addMat(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat& mask, double, Stream& _stream, int);
+
+namespace
 {
-    struct VAdd4 : binary_function<uint, uint, uint>
-    {
-        __device__ __forceinline__ uint operator ()(uint a, uint b) const
-        {
-            return vadd4(a, b);
-        }
-
-        __host__ __device__ __forceinline__ VAdd4() {}
-        __host__ __device__ __forceinline__ VAdd4(const VAdd4&) {}
-    };
-
-    struct VAdd2 : binary_function<uint, uint, uint>
-    {
-        __device__ __forceinline__ uint operator ()(uint a, uint b) const
-        {
-            return vadd2(a, b);
-        }
-
-        __host__ __device__ __forceinline__ VAdd2() {}
-        __host__ __device__ __forceinline__ VAdd2(const VAdd2&) {}
-    };
-
-    template <typename T, typename D> struct AddMat : binary_function<T, T, D>
+    template <typename T, typename D> struct AddOp1 : binary_function<T, T, D>
     {
         __device__ __forceinline__ D operator ()(T a, T b) const
         {
             return saturate_cast<D>(a + b);
         }
-
-        __host__ __device__ __forceinline__ AddMat() {}
-        __host__ __device__ __forceinline__ AddMat(const AddMat&) {}
     };
-}
-
-namespace cv { namespace cuda { namespace device
-{
-    template <> struct TransformFunctorTraits< arithm::VAdd4 > : arithm::ArithmFuncTraits<sizeof(uint), sizeof(uint)>
-    {
-    };
-
-    template <> struct TransformFunctorTraits< arithm::VAdd2 > : arithm::ArithmFuncTraits<sizeof(uint), sizeof(uint)>
-    {
-    };
-
-    template <typename T, typename D> struct TransformFunctorTraits< arithm::AddMat<T, D> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(D)>
-    {
-    };
-}}}
-
-namespace arithm
-{
-    void addMat_v4(PtrStepSz<uint> src1, PtrStepSz<uint> src2, PtrStepSz<uint> dst, cudaStream_t stream)
-    {
-        device::transform(src1, src2, dst, VAdd4(), WithOutMask(), stream);
-    }
-
-    void addMat_v2(PtrStepSz<uint> src1, PtrStepSz<uint> src2, PtrStepSz<uint> dst, cudaStream_t stream)
-    {
-        device::transform(src1, src2, dst, VAdd2(), WithOutMask(), stream);
-    }
 
     template <typename T, typename D>
-    void addMat(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream)
+    void addMat_v1(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat& mask, Stream& stream)
     {
         if (mask.data)
-            device::transform((PtrStepSz<T>) src1, (PtrStepSz<T>) src2, (PtrStepSz<D>) dst, AddMat<T, D>(), mask, stream);
+            gridTransformBinary(globPtr<T>(src1), globPtr<T>(src2), globPtr<D>(dst), AddOp1<T, D>(), globPtr<uchar>(mask), stream);
         else
-            device::transform((PtrStepSz<T>) src1, (PtrStepSz<T>) src2, (PtrStepSz<D>) dst, AddMat<T, D>(), WithOutMask(), stream);
+            gridTransformBinary(globPtr<T>(src1), globPtr<T>(src2), globPtr<D>(dst), AddOp1<T, D>(), stream);
     }
 
-    template void addMat<uchar, uchar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addMat<uchar, schar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addMat<uchar, ushort>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addMat<uchar, short>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addMat<uchar, int>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addMat<uchar, float>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addMat<uchar, double>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    struct AddOp2 : binary_function<uint, uint, uint>
+    {
+        __device__ __forceinline__ uint operator ()(uint a, uint b) const
+        {
+            return vadd2(a, b);
+        }
+    };
 
-    template void addMat<schar, uchar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addMat<schar, schar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addMat<schar, ushort>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addMat<schar, short>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addMat<schar, int>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addMat<schar, float>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addMat<schar, double>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    void addMat_v2(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, Stream& stream)
+    {
+        const int vcols = src1.cols >> 1;
 
-    //template void addMat<ushort, uchar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void addMat<ushort, schar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addMat<ushort, ushort>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addMat<ushort, short>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addMat<ushort, int>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addMat<ushort, float>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addMat<ushort, double>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+        GlobPtrSz<uint> src1_ = globPtr((uint*) src1.data, src1.step, src1.rows, vcols);
+        GlobPtrSz<uint> src2_ = globPtr((uint*) src2.data, src2.step, src1.rows, vcols);
+        GlobPtrSz<uint> dst_ = globPtr((uint*) dst.data, dst.step, src1.rows, vcols);
 
-    //template void addMat<short, uchar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void addMat<short, schar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addMat<short, ushort>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addMat<short, short>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addMat<short, int>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addMat<short, float>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addMat<short, double>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+        gridTransformBinary(src1_, src2_, dst_, AddOp2(), stream);
+    }
 
-    //template void addMat<int, uchar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void addMat<int, schar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void addMat<int, ushort>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void addMat<int, short>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addMat<int, int>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addMat<int, float>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addMat<int, double>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    struct AddOp4 : binary_function<uint, uint, uint>
+    {
+        __device__ __forceinline__ uint operator ()(uint a, uint b) const
+        {
+            return vadd4(a, b);
+        }
+    };
 
-    //template void addMat<float, uchar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void addMat<float, schar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void addMat<float, ushort>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void addMat<float, short>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void addMat<float, int>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addMat<float, float>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addMat<float, double>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    void addMat_v4(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, Stream& stream)
+    {
+        const int vcols = src1.cols >> 2;
 
-    //template void addMat<double, uchar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void addMat<double, schar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void addMat<double, ushort>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void addMat<double, short>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void addMat<double, int>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void addMat<double, float>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addMat<double, double>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+        GlobPtrSz<uint> src1_ = globPtr((uint*) src1.data, src1.step, src1.rows, vcols);
+        GlobPtrSz<uint> src2_ = globPtr((uint*) src2.data, src2.step, src1.rows, vcols);
+        GlobPtrSz<uint> dst_ = globPtr((uint*) dst.data, dst.step, src1.rows, vcols);
+
+        gridTransformBinary(src1_, src2_, dst_, AddOp4(), stream);
+    }
 }
 
-#endif // CUDA_DISABLER
+void addMat(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat& mask, double, Stream& stream, int)
+{
+    typedef void (*func_t)(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat& mask, Stream& stream);
+    static const func_t funcs[7][7] =
+    {
+        {
+            addMat_v1<uchar, uchar>,
+            addMat_v1<uchar, schar>,
+            addMat_v1<uchar, ushort>,
+            addMat_v1<uchar, short>,
+            addMat_v1<uchar, int>,
+            addMat_v1<uchar, float>,
+            addMat_v1<uchar, double>
+        },
+        {
+            addMat_v1<schar, uchar>,
+            addMat_v1<schar, schar>,
+            addMat_v1<schar, ushort>,
+            addMat_v1<schar, short>,
+            addMat_v1<schar, int>,
+            addMat_v1<schar, float>,
+            addMat_v1<schar, double>
+        },
+        {
+            0 /*addMat_v1<ushort, uchar>*/,
+            0 /*addMat_v1<ushort, schar>*/,
+            addMat_v1<ushort, ushort>,
+            addMat_v1<ushort, short>,
+            addMat_v1<ushort, int>,
+            addMat_v1<ushort, float>,
+            addMat_v1<ushort, double>
+        },
+        {
+            0 /*addMat_v1<short, uchar>*/,
+            0 /*addMat_v1<short, schar>*/,
+            addMat_v1<short, ushort>,
+            addMat_v1<short, short>,
+            addMat_v1<short, int>,
+            addMat_v1<short, float>,
+            addMat_v1<short, double>
+        },
+        {
+            0 /*addMat_v1<int, uchar>*/,
+            0 /*addMat_v1<int, schar>*/,
+            0 /*addMat_v1<int, ushort>*/,
+            0 /*addMat_v1<int, short>*/,
+            addMat_v1<int, int>,
+            addMat_v1<int, float>,
+            addMat_v1<int, double>
+        },
+        {
+            0 /*addMat_v1<float, uchar>*/,
+            0 /*addMat_v1<float, schar>*/,
+            0 /*addMat_v1<float, ushort>*/,
+            0 /*addMat_v1<float, short>*/,
+            0 /*addMat_v1<float, int>*/,
+            addMat_v1<float, float>,
+            addMat_v1<float, double>
+        },
+        {
+            0 /*addMat_v1<double, uchar>*/,
+            0 /*addMat_v1<double, schar>*/,
+            0 /*addMat_v1<double, ushort>*/,
+            0 /*addMat_v1<double, short>*/,
+            0 /*addMat_v1<double, int>*/,
+            0 /*addMat_v1<double, float>*/,
+            addMat_v1<double, double>
+        }
+    };
+
+    const int sdepth = src1.depth();
+    const int ddepth = dst.depth();
+
+    CV_DbgAssert( sdepth <= CV_64F && ddepth <= CV_64F );
+
+    GpuMat src1_ = src1.reshape(1);
+    GpuMat src2_ = src2.reshape(1);
+    GpuMat dst_ = dst.reshape(1);
+
+    if (mask.empty() && (sdepth == CV_8U || sdepth == CV_16U) && ddepth == sdepth)
+    {
+        const intptr_t src1ptr = reinterpret_cast<intptr_t>(src1_.data);
+        const intptr_t src2ptr = reinterpret_cast<intptr_t>(src2_.data);
+        const intptr_t dstptr = reinterpret_cast<intptr_t>(dst_.data);
+
+        const bool isAllAligned = (src1ptr & 31) == 0 && (src2ptr & 31) == 0 && (dstptr & 31) == 0;
+
+        if (isAllAligned)
+        {
+            if (sdepth == CV_8U && (src1_.cols & 3) == 0)
+            {
+                addMat_v4(src1_, src2_, dst_, stream);
+                return;
+            }
+            else if (sdepth == CV_16U && (src1_.cols & 1) == 0)
+            {
+                addMat_v2(src1_, src2_, dst_, stream);
+                return;
+            }
+        }
+    }
+
+    const func_t func = funcs[sdepth][ddepth];
+
+    if (!func)
+        CV_Error(cv::Error::StsUnsupportedFormat, "Unsupported combination of source and destination types");
+
+    func(src1_, src2_, dst_, mask, stream);
+}
+
+#endif
diff --git a/modules/cudaarithm/src/cuda/add_scalar.cu b/modules/cudaarithm/src/cuda/add_scalar.cu
index 680061be11..92838a2a57 100644
--- a/modules/cudaarithm/src/cuda/add_scalar.cu
+++ b/modules/cudaarithm/src/cuda/add_scalar.cu
@@ -40,109 +40,141 @@
 //
 //M*/
 
-#if !defined CUDA_DISABLER
+#include "opencv2/opencv_modules.hpp"
 
-#include "opencv2/core/cuda/common.hpp"
-#include "opencv2/core/cuda/functional.hpp"
-#include "opencv2/core/cuda/transform.hpp"
-#include "opencv2/core/cuda/saturate_cast.hpp"
-#include "opencv2/core/cuda/simd_functions.hpp"
+#ifndef HAVE_OPENCV_CUDEV
 
-#include "arithm_func_traits.hpp"
+#error "opencv_cudev is required"
 
-using namespace cv::cuda;
-using namespace cv::cuda::device;
+#else
 
-namespace arithm
+#include "opencv2/cudev.hpp"
+
+using namespace cv::cudev;
+
+void addScalar(const GpuMat& src, cv::Scalar val, bool, GpuMat& dst, const GpuMat& mask, double, Stream& stream, int);
+
+namespace
 {
-    template <typename T, typename S, typename D> struct AddScalar : unary_function<T, D>
+    template <typename SrcType, typename ScalarType, typename DstType> struct AddScalarOp : unary_function<SrcType, DstType>
     {
-        S val;
+        ScalarType val;
 
-        __host__ explicit AddScalar(S val_) : val(val_) {}
-
-        __device__ __forceinline__ D operator ()(T a) const
+        __device__ __forceinline__ DstType operator ()(SrcType a) const
         {
-            return saturate_cast<D>(a + val);
+            return saturate_cast<DstType>(saturate_cast<ScalarType>(a) + val);
         }
     };
-}
 
-namespace cv { namespace cuda { namespace device
-{
-    template <typename T, typename S, typename D> struct TransformFunctorTraits< arithm::AddScalar<T, S, D> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(D)>
+    template <typename ScalarDepth> struct TransformPolicy : DefaultTransformPolicy
     {
     };
-}}}
-
-namespace arithm
-{
-    template <typename T, typename S, typename D>
-    void addScalar(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream)
+    template <> struct TransformPolicy<double> : DefaultTransformPolicy
     {
-        AddScalar<T, S, D> op(static_cast<S>(val));
+        enum {
+            shift = 1
+        };
+    };
+
+    template <typename SrcType, typename ScalarDepth, typename DstType>
+    void addScalarImpl(const GpuMat& src, cv::Scalar value, GpuMat& dst, const GpuMat& mask, Stream& stream)
+    {
+        typedef typename MakeVec<ScalarDepth, VecTraits<SrcType>::cn>::type ScalarType;
+
+        cv::Scalar_<ScalarDepth> value_ = value;
+
+        AddScalarOp<SrcType, ScalarType, DstType> op;
+        op.val = VecTraits<ScalarType>::make(value_.val);
 
         if (mask.data)
-            device::transform((PtrStepSz<T>) src1, (PtrStepSz<D>) dst, op, mask, stream);
+            gridTransformUnary_< TransformPolicy<ScalarDepth> >(globPtr<SrcType>(src), globPtr<DstType>(dst), op, globPtr<uchar>(mask), stream);
         else
-            device::transform((PtrStepSz<T>) src1, (PtrStepSz<D>) dst, op, WithOutMask(), stream);
+            gridTransformUnary_< TransformPolicy<ScalarDepth> >(globPtr<SrcType>(src), globPtr<DstType>(dst), op, stream);
     }
-
-    template void addScalar<uchar, float, uchar>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addScalar<uchar, float, schar>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addScalar<uchar, float, ushort>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addScalar<uchar, float, short>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addScalar<uchar, float, int>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addScalar<uchar, float, float>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addScalar<uchar, double, double>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-
-    template void addScalar<schar, float, uchar>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addScalar<schar, float, schar>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addScalar<schar, float, ushort>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addScalar<schar, float, short>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addScalar<schar, float, int>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addScalar<schar, float, float>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addScalar<schar, double, double>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-
-    //template void addScalar<ushort, float, uchar>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void addScalar<ushort, float, schar>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addScalar<ushort, float, ushort>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addScalar<ushort, float, short>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addScalar<ushort, float, int>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addScalar<ushort, float, float>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addScalar<ushort, double, double>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-
-    //template void addScalar<short, float, uchar>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void addScalar<short, float, schar>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addScalar<short, float, ushort>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addScalar<short, float, short>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addScalar<short, float, int>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addScalar<short, float, float>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addScalar<short, double, double>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-
-    //template void addScalar<int, float, uchar>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void addScalar<int, float, schar>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void addScalar<int, float, ushort>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void addScalar<int, float, short>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addScalar<int, float, int>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addScalar<int, float, float>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addScalar<int, double, double>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-
-    //template void addScalar<float, float, uchar>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void addScalar<float, float, schar>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void addScalar<float, float, ushort>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void addScalar<float, float, short>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void addScalar<float, float, int>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addScalar<float, float, float>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addScalar<float, double, double>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-
-    //template void addScalar<double, double, uchar>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void addScalar<double, double, schar>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void addScalar<double, double, ushort>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void addScalar<double, double, short>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void addScalar<double, double, int>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void addScalar<double, double, float>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void addScalar<double, double, double>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
 }
 
-#endif // CUDA_DISABLER
+void addScalar(const GpuMat& src, cv::Scalar val, bool, GpuMat& dst, const GpuMat& mask, double, Stream& stream, int)
+{
+    typedef void (*func_t)(const GpuMat& src, cv::Scalar val, GpuMat& dst, const GpuMat& mask, Stream& stream);
+    static const func_t funcs[7][7][4] =
+    {
+        {
+            {addScalarImpl<uchar, float, uchar>, addScalarImpl<uchar2, float, uchar2>, addScalarImpl<uchar3, float, uchar3>, addScalarImpl<uchar4, float, uchar4>},
+            {addScalarImpl<uchar, float, schar>, addScalarImpl<uchar2, float, char2>, addScalarImpl<uchar3, float, char3>, addScalarImpl<uchar4, float, char4>},
+            {addScalarImpl<uchar, float, ushort>, addScalarImpl<uchar2, float, ushort2>, addScalarImpl<uchar3, float, ushort3>, addScalarImpl<uchar4, float, ushort4>},
+            {addScalarImpl<uchar, float, short>, addScalarImpl<uchar2, float, short2>, addScalarImpl<uchar3, float, short3>, addScalarImpl<uchar4, float, short4>},
+            {addScalarImpl<uchar, float, int>, addScalarImpl<uchar2, float, int2>, addScalarImpl<uchar3, float, int3>, addScalarImpl<uchar4, float, int4>},
+            {addScalarImpl<uchar, float, float>, addScalarImpl<uchar2, float, float2>, addScalarImpl<uchar3, float, float3>, addScalarImpl<uchar4, float, float4>},
+            {addScalarImpl<uchar, double, double>, addScalarImpl<uchar2, double, double2>, addScalarImpl<uchar3, double, double3>, addScalarImpl<uchar4, double, double4>}
+        },
+        {
+            {addScalarImpl<schar, float, uchar>, addScalarImpl<char2, float, uchar2>, addScalarImpl<char3, float, uchar3>, addScalarImpl<char4, float, uchar4>},
+            {addScalarImpl<schar, float, schar>, addScalarImpl<char2, float, char2>, addScalarImpl<char3, float, char3>, addScalarImpl<char4, float, char4>},
+            {addScalarImpl<schar, float, ushort>, addScalarImpl<char2, float, ushort2>, addScalarImpl<char3, float, ushort3>, addScalarImpl<char4, float, ushort4>},
+            {addScalarImpl<schar, float, short>, addScalarImpl<char2, float, short2>, addScalarImpl<char3, float, short3>, addScalarImpl<char4, float, short4>},
+            {addScalarImpl<schar, float, int>, addScalarImpl<char2, float, int2>, addScalarImpl<char3, float, int3>, addScalarImpl<char4, float, int4>},
+            {addScalarImpl<schar, float, float>, addScalarImpl<char2, float, float2>, addScalarImpl<char3, float, float3>, addScalarImpl<char4, float, float4>},
+            {addScalarImpl<schar, double, double>, addScalarImpl<char2, double, double2>, addScalarImpl<char3, double, double3>, addScalarImpl<char4, double, double4>}
+        },
+        {
+            {0 /*addScalarImpl<ushort, float, uchar>*/, 0 /*addScalarImpl<ushort2, float, uchar2>*/, 0 /*addScalarImpl<ushort3, float, uchar3>*/, 0 /*addScalarImpl<ushort4, float, uchar4>*/},
+            {0 /*addScalarImpl<ushort, float, schar>*/, 0 /*addScalarImpl<ushort2, float, char2>*/, 0 /*addScalarImpl<ushort3, float, char3>*/, 0 /*addScalarImpl<ushort4, float, char4>*/},
+            {addScalarImpl<ushort, float, ushort>, addScalarImpl<ushort2, float, ushort2>, addScalarImpl<ushort3, float, ushort3>, addScalarImpl<ushort4, float, ushort4>},
+            {addScalarImpl<ushort, float, short>, addScalarImpl<ushort2, float, short2>, addScalarImpl<ushort3, float, short3>, addScalarImpl<ushort4, float, short4>},
+            {addScalarImpl<ushort, float, int>, addScalarImpl<ushort2, float, int2>, addScalarImpl<ushort3, float, int3>, addScalarImpl<ushort4, float, int4>},
+            {addScalarImpl<ushort, float, float>, addScalarImpl<ushort2, float, float2>, addScalarImpl<ushort3, float, float3>, addScalarImpl<ushort4, float, float4>},
+            {addScalarImpl<ushort, double, double>, addScalarImpl<ushort2, double, double2>, addScalarImpl<ushort3, double, double3>, addScalarImpl<ushort4, double, double4>}
+        },
+        {
+            {0 /*addScalarImpl<short, float, uchar>*/, 0 /*addScalarImpl<short2, float, uchar2>*/, 0 /*addScalarImpl<short3, float, uchar3>*/, 0 /*addScalarImpl<short4, float, uchar4>*/},
+            {0 /*addScalarImpl<short, float, schar>*/, 0 /*addScalarImpl<short2, float, char2>*/, 0 /*addScalarImpl<short3, float, char3>*/, 0 /*addScalarImpl<short4, float, char4>*/},
+            {addScalarImpl<short, float, ushort>, addScalarImpl<short2, float, ushort2>, addScalarImpl<short3, float, ushort3>, addScalarImpl<short4, float, ushort4>},
+            {addScalarImpl<short, float, short>, addScalarImpl<short2, float, short2>, addScalarImpl<short3, float, short3>, addScalarImpl<short4, float, short4>},
+            {addScalarImpl<short, float, int>, addScalarImpl<short2, float, int2>, addScalarImpl<short3, float, int3>, addScalarImpl<short4, float, int4>},
+            {addScalarImpl<short, float, float>, addScalarImpl<short2, float, float2>, addScalarImpl<short3, float, float3>, addScalarImpl<short4, float, float4>},
+            {addScalarImpl<short, double, double>, addScalarImpl<short2, double, double2>, addScalarImpl<short3, double, double3>, addScalarImpl<short4, double, double4>}
+        },
+        {
+            {0 /*addScalarImpl<int, float, uchar>*/, 0 /*addScalarImpl<int2, float, uchar2>*/, 0 /*addScalarImpl<int3, float, uchar3>*/, 0 /*addScalarImpl<int4, float, uchar4>*/},
+            {0 /*addScalarImpl<int, float, schar>*/, 0 /*addScalarImpl<int2, float, char2>*/, 0 /*addScalarImpl<int3, float, char3>*/, 0 /*addScalarImpl<int4, float, char4>*/},
+            {0 /*addScalarImpl<int, float, ushort>*/, 0 /*addScalarImpl<int2, float, ushort2>*/, 0 /*addScalarImpl<int3, float, ushort3>*/, 0 /*addScalarImpl<int4, float, ushort4>*/},
+            {0 /*addScalarImpl<int, float, short>*/, 0 /*addScalarImpl<int2, float, short2>*/, 0 /*addScalarImpl<int3, float, short3>*/, 0 /*addScalarImpl<int4, float, short4>*/},
+            {addScalarImpl<int, float, int>, addScalarImpl<int2, float, int2>, addScalarImpl<int3, float, int3>, addScalarImpl<int4, float, int4>},
+            {addScalarImpl<int, float, float>, addScalarImpl<int2, float, float2>, addScalarImpl<int3, float, float3>, addScalarImpl<int4, float, float4>},
+            {addScalarImpl<int, double, double>, addScalarImpl<int2, double, double2>, addScalarImpl<int3, double, double3>, addScalarImpl<int4, double, double4>}
+        },
+        {
+            {0 /*addScalarImpl<float, float, uchar>*/, 0 /*addScalarImpl<float2, float, uchar2>*/, 0 /*addScalarImpl<float3, float, uchar3>*/, 0 /*addScalarImpl<float4, float, uchar4>*/},
+            {0 /*addScalarImpl<float, float, schar>*/, 0 /*addScalarImpl<float2, float, char2>*/, 0 /*addScalarImpl<float3, float, char3>*/, 0 /*addScalarImpl<float4, float, char4>*/},
+            {0 /*addScalarImpl<float, float, ushort>*/, 0 /*addScalarImpl<float2, float, ushort2>*/, 0 /*addScalarImpl<float3, float, ushort3>*/, 0 /*addScalarImpl<float4, float, ushort4>*/},
+            {0 /*addScalarImpl<float, float, short>*/, 0 /*addScalarImpl<float2, float, short2>*/, 0 /*addScalarImpl<float3, float, short3>*/, 0 /*addScalarImpl<float4, float, short4>*/},
+            {0 /*addScalarImpl<float, float, int>*/, 0 /*addScalarImpl<float2, float, int2>*/, 0 /*addScalarImpl<float3, float, int3>*/, 0 /*addScalarImpl<float4, float, int4>*/},
+            {addScalarImpl<float, float, float>, addScalarImpl<float2, float, float2>, addScalarImpl<float3, float, float3>, addScalarImpl<float4, float, float4>},
+            {addScalarImpl<float, double, double>, addScalarImpl<float2, double, double2>, addScalarImpl<float3, double, double3>, addScalarImpl<float4, double, double4>}
+        },
+        {
+            {0 /*addScalarImpl<double, double, uchar>*/, 0 /*addScalarImpl<double2, double, uchar2>*/, 0 /*addScalarImpl<double3, double, uchar3>*/, 0 /*addScalarImpl<double4, double, uchar4>*/},
+            {0 /*addScalarImpl<double, double, schar>*/, 0 /*addScalarImpl<double2, double, char2>*/, 0 /*addScalarImpl<double3, double, char3>*/, 0 /*addScalarImpl<double4, double, char4>*/},
+            {0 /*addScalarImpl<double, double, ushort>*/, 0 /*addScalarImpl<double2, double, ushort2>*/, 0 /*addScalarImpl<double3, double, ushort3>*/, 0 /*addScalarImpl<double4, double, ushort4>*/},
+            {0 /*addScalarImpl<double, double, short>*/, 0 /*addScalarImpl<double2, double, short2>*/, 0 /*addScalarImpl<double3, double, short3>*/, 0 /*addScalarImpl<double4, double, short4>*/},
+            {0 /*addScalarImpl<double, double, int>*/, 0 /*addScalarImpl<double2, double, int2>*/, 0 /*addScalarImpl<double3, double, int3>*/, 0 /*addScalarImpl<double4, double, int4>*/},
+            {0 /*addScalarImpl<double, double, float>*/, 0 /*addScalarImpl<double2, double, float2>*/, 0 /*addScalarImpl<double3, double, float3>*/, 0 /*addScalarImpl<double4, double, float4>*/},
+            {addScalarImpl<double, double, double>, addScalarImpl<double2, double, double2>, addScalarImpl<double3, double, double3>, addScalarImpl<double4, double, double4>}
+        }
+    };
+
+    const int sdepth = src.depth();
+    const int ddepth = dst.depth();
+    const int cn = src.channels();
+
+    CV_DbgAssert( sdepth <= CV_64F && ddepth <= CV_64F && cn <= 4 );
+
+    const func_t func = funcs[sdepth][ddepth][cn - 1];
+
+    if (!func)
+        CV_Error(cv::Error::StsUnsupportedFormat, "Unsupported combination of source and destination types");
+
+    func(src, val, dst, mask, stream);
+}
+
+#endif
diff --git a/modules/cudaarithm/src/cuda/add_weighted.cu b/modules/cudaarithm/src/cuda/add_weighted.cu
index aa305d993e..d5c00f6072 100644
--- a/modules/cudaarithm/src/cuda/add_weighted.cu
+++ b/modules/cudaarithm/src/cuda/add_weighted.cu
@@ -40,325 +40,553 @@
 //
 //M*/
 
-#if !defined CUDA_DISABLER
+#include "opencv2/opencv_modules.hpp"
 
-#include "opencv2/core/cuda/common.hpp"
-#include "opencv2/core/cuda/functional.hpp"
-#include "opencv2/core/cuda/transform.hpp"
-#include "opencv2/core/cuda/saturate_cast.hpp"
+#ifndef HAVE_OPENCV_CUDEV
 
-#include "arithm_func_traits.hpp"
+#error "opencv_cudev is required"
 
-using namespace cv::cuda;
-using namespace cv::cuda::device;
+#else
 
-namespace arithm
+#include "opencv2/cudaarithm.hpp"
+#include "opencv2/cudev.hpp"
+
+using namespace cv::cudev;
+
+namespace
 {
-    template <typename T> struct UseDouble_
+    template <typename T1, typename T2, typename D, typename S> struct AddWeightedOp : binary_function<T1, T2, D>
     {
-        enum {value = 0};
-    };
-    template <> struct UseDouble_<double>
-    {
-        enum {value = 1};
-    };
-    template <typename T1, typename T2, typename D> struct UseDouble
-    {
-        enum {value = (UseDouble_<T1>::value || UseDouble_<T2>::value || UseDouble_<D>::value)};
-    };
-
-    template <typename T1, typename T2, typename D, bool useDouble> struct AddWeighted_;
-    template <typename T1, typename T2, typename D> struct AddWeighted_<T1, T2, D, false> : binary_function<T1, T2, D>
-    {
-        float alpha;
-        float beta;
-        float gamma;
-
-        __host__ AddWeighted_(double alpha_, double beta_, double gamma_) : alpha(static_cast<float>(alpha_)), beta(static_cast<float>(beta_)), gamma(static_cast<float>(gamma_)) {}
+        S alpha;
+        S beta;
+        S gamma;
 
         __device__ __forceinline__ D operator ()(T1 a, T2 b) const
         {
             return saturate_cast<D>(a * alpha + b * beta + gamma);
         }
     };
-    template <typename T1, typename T2, typename D> struct AddWeighted_<T1, T2, D, true> : binary_function<T1, T2, D>
-    {
-        double alpha;
-        double beta;
-        double gamma;
 
-        __host__ AddWeighted_(double alpha_, double beta_, double gamma_) : alpha(alpha_), beta(beta_), gamma(gamma_) {}
-
-        __device__ __forceinline__ D operator ()(T1 a, T2 b) const
-        {
-            return saturate_cast<D>(a * alpha + b * beta + gamma);
-        }
-    };
-    template <typename T1, typename T2, typename D> struct AddWeighted : AddWeighted_<T1, T2, D, UseDouble<T1, T2, D>::value>
+    template <typename ScalarDepth> struct TransformPolicy : DefaultTransformPolicy
     {
-        AddWeighted(double alpha_, double beta_, double gamma_) : AddWeighted_<T1, T2, D, UseDouble<T1, T2, D>::value>(alpha_, beta_, gamma_) {}
     };
+    template <> struct TransformPolicy<double> : DefaultTransformPolicy
+    {
+        enum {
+            shift = 1
+        };
+    };
+
+    template <typename T1, typename T2, typename D>
+    void addWeightedImpl(const GpuMat& src1, double alpha, const GpuMat& src2, double beta, double gamma, GpuMat& dst, Stream& stream)
+    {
+        typedef typename LargerType<T1, T2>::type larger_type1;
+        typedef typename LargerType<larger_type1, D>::type larger_type2;
+        typedef typename LargerType<larger_type2, float>::type scalar_type;
+
+        AddWeightedOp<T1, T2, D, scalar_type> op;
+        op.alpha = static_cast<scalar_type>(alpha);
+        op.beta = static_cast<scalar_type>(beta);
+        op.gamma = static_cast<scalar_type>(gamma);
+
+        gridTransformBinary_< TransformPolicy<scalar_type> >(globPtr<T1>(src1), globPtr<T2>(src2), globPtr<D>(dst), op, stream);
+    }
 }
 
-namespace cv { namespace cuda { namespace device
+void cv::cuda::addWeighted(InputArray _src1, double alpha, InputArray _src2, double beta, double gamma, OutputArray _dst, int ddepth, Stream& stream)
 {
-    template <typename T1, typename T2, typename D, size_t src1_size, size_t src2_size, size_t dst_size> struct AddWeightedTraits : DefaultTransformFunctorTraits< arithm::AddWeighted<T1, T2, D> >
-    {
-    };
-    template <typename T1, typename T2, typename D, size_t src_size, size_t dst_size> struct AddWeightedTraits<T1, T2, D, src_size, src_size, dst_size> : arithm::ArithmFuncTraits<src_size, dst_size>
+    typedef void (*func_t)(const GpuMat& src1, double alpha, const GpuMat& src2, double beta, double gamma, GpuMat& dst, Stream& stream);
+    static const func_t funcs[7][7][7] =
     {
+        {
+            {
+                addWeightedImpl<uchar, uchar, uchar >,
+                addWeightedImpl<uchar, uchar, schar >,
+                addWeightedImpl<uchar, uchar, ushort>,
+                addWeightedImpl<uchar, uchar, short >,
+                addWeightedImpl<uchar, uchar, int   >,
+                addWeightedImpl<uchar, uchar, float >,
+                addWeightedImpl<uchar, uchar, double>
+            },
+            {
+                addWeightedImpl<uchar, schar, uchar >,
+                addWeightedImpl<uchar, schar, schar >,
+                addWeightedImpl<uchar, schar, ushort>,
+                addWeightedImpl<uchar, schar, short >,
+                addWeightedImpl<uchar, schar, int   >,
+                addWeightedImpl<uchar, schar, float >,
+                addWeightedImpl<uchar, schar, double>
+            },
+            {
+                addWeightedImpl<uchar, ushort, uchar >,
+                addWeightedImpl<uchar, ushort, schar >,
+                addWeightedImpl<uchar, ushort, ushort>,
+                addWeightedImpl<uchar, ushort, short >,
+                addWeightedImpl<uchar, ushort, int   >,
+                addWeightedImpl<uchar, ushort, float >,
+                addWeightedImpl<uchar, ushort, double>
+            },
+            {
+                addWeightedImpl<uchar, short, uchar >,
+                addWeightedImpl<uchar, short, schar >,
+                addWeightedImpl<uchar, short, ushort>,
+                addWeightedImpl<uchar, short, short >,
+                addWeightedImpl<uchar, short, int   >,
+                addWeightedImpl<uchar, short, float >,
+                addWeightedImpl<uchar, short, double>
+            },
+            {
+                addWeightedImpl<uchar, int, uchar >,
+                addWeightedImpl<uchar, int, schar >,
+                addWeightedImpl<uchar, int, ushort>,
+                addWeightedImpl<uchar, int, short >,
+                addWeightedImpl<uchar, int, int   >,
+                addWeightedImpl<uchar, int, float >,
+                addWeightedImpl<uchar, int, double>
+            },
+            {
+                addWeightedImpl<uchar, float, uchar >,
+                addWeightedImpl<uchar, float, schar >,
+                addWeightedImpl<uchar, float, ushort>,
+                addWeightedImpl<uchar, float, short >,
+                addWeightedImpl<uchar, float, int   >,
+                addWeightedImpl<uchar, float, float >,
+                addWeightedImpl<uchar, float, double>
+            },
+            {
+                addWeightedImpl<uchar, double, uchar >,
+                addWeightedImpl<uchar, double, schar >,
+                addWeightedImpl<uchar, double, ushort>,
+                addWeightedImpl<uchar, double, short >,
+                addWeightedImpl<uchar, double, int   >,
+                addWeightedImpl<uchar, double, float >,
+                addWeightedImpl<uchar, double, double>
+            }
+        },
+        {
+            {
+                0/*addWeightedImpl<schar, uchar, uchar >*/,
+                0/*addWeightedImpl<schar, uchar, schar >*/,
+                0/*addWeightedImpl<schar, uchar, ushort>*/,
+                0/*addWeightedImpl<schar, uchar, short >*/,
+                0/*addWeightedImpl<schar, uchar, int   >*/,
+                0/*addWeightedImpl<schar, uchar, float >*/,
+                0/*addWeightedImpl<schar, uchar, double>*/
+            },
+            {
+                addWeightedImpl<schar, schar, uchar >,
+                addWeightedImpl<schar, schar, schar >,
+                addWeightedImpl<schar, schar, ushort>,
+                addWeightedImpl<schar, schar, short >,
+                addWeightedImpl<schar, schar, int   >,
+                addWeightedImpl<schar, schar, float >,
+                addWeightedImpl<schar, schar, double>
+            },
+            {
+                addWeightedImpl<schar, ushort, uchar >,
+                addWeightedImpl<schar, ushort, schar >,
+                addWeightedImpl<schar, ushort, ushort>,
+                addWeightedImpl<schar, ushort, short >,
+                addWeightedImpl<schar, ushort, int   >,
+                addWeightedImpl<schar, ushort, float >,
+                addWeightedImpl<schar, ushort, double>
+            },
+            {
+                addWeightedImpl<schar, short, uchar >,
+                addWeightedImpl<schar, short, schar >,
+                addWeightedImpl<schar, short, ushort>,
+                addWeightedImpl<schar, short, short >,
+                addWeightedImpl<schar, short, int   >,
+                addWeightedImpl<schar, short, float >,
+                addWeightedImpl<schar, short, double>
+            },
+            {
+                addWeightedImpl<schar, int, uchar >,
+                addWeightedImpl<schar, int, schar >,
+                addWeightedImpl<schar, int, ushort>,
+                addWeightedImpl<schar, int, short >,
+                addWeightedImpl<schar, int, int   >,
+                addWeightedImpl<schar, int, float >,
+                addWeightedImpl<schar, int, double>
+            },
+            {
+                addWeightedImpl<schar, float, uchar >,
+                addWeightedImpl<schar, float, schar >,
+                addWeightedImpl<schar, float, ushort>,
+                addWeightedImpl<schar, float, short >,
+                addWeightedImpl<schar, float, int   >,
+                addWeightedImpl<schar, float, float >,
+                addWeightedImpl<schar, float, double>
+            },
+            {
+                addWeightedImpl<schar, double, uchar >,
+                addWeightedImpl<schar, double, schar >,
+                addWeightedImpl<schar, double, ushort>,
+                addWeightedImpl<schar, double, short >,
+                addWeightedImpl<schar, double, int   >,
+                addWeightedImpl<schar, double, float >,
+                addWeightedImpl<schar, double, double>
+            }
+        },
+        {
+            {
+                0/*addWeightedImpl<ushort, uchar, uchar >*/,
+                0/*addWeightedImpl<ushort, uchar, schar >*/,
+                0/*addWeightedImpl<ushort, uchar, ushort>*/,
+                0/*addWeightedImpl<ushort, uchar, short >*/,
+                0/*addWeightedImpl<ushort, uchar, int   >*/,
+                0/*addWeightedImpl<ushort, uchar, float >*/,
+                0/*addWeightedImpl<ushort, uchar, double>*/
+            },
+            {
+                0/*addWeightedImpl<ushort, schar, uchar >*/,
+                0/*addWeightedImpl<ushort, schar, schar >*/,
+                0/*addWeightedImpl<ushort, schar, ushort>*/,
+                0/*addWeightedImpl<ushort, schar, short >*/,
+                0/*addWeightedImpl<ushort, schar, int   >*/,
+                0/*addWeightedImpl<ushort, schar, float >*/,
+                0/*addWeightedImpl<ushort, schar, double>*/
+            },
+            {
+                addWeightedImpl<ushort, ushort, uchar >,
+                addWeightedImpl<ushort, ushort, schar >,
+                addWeightedImpl<ushort, ushort, ushort>,
+                addWeightedImpl<ushort, ushort, short >,
+                addWeightedImpl<ushort, ushort, int   >,
+                addWeightedImpl<ushort, ushort, float >,
+                addWeightedImpl<ushort, ushort, double>
+            },
+            {
+                addWeightedImpl<ushort, short, uchar >,
+                addWeightedImpl<ushort, short, schar >,
+                addWeightedImpl<ushort, short, ushort>,
+                addWeightedImpl<ushort, short, short >,
+                addWeightedImpl<ushort, short, int   >,
+                addWeightedImpl<ushort, short, float >,
+                addWeightedImpl<ushort, short, double>
+            },
+            {
+                addWeightedImpl<ushort, int, uchar >,
+                addWeightedImpl<ushort, int, schar >,
+                addWeightedImpl<ushort, int, ushort>,
+                addWeightedImpl<ushort, int, short >,
+                addWeightedImpl<ushort, int, int   >,
+                addWeightedImpl<ushort, int, float >,
+                addWeightedImpl<ushort, int, double>
+            },
+            {
+                addWeightedImpl<ushort, float, uchar >,
+                addWeightedImpl<ushort, float, schar >,
+                addWeightedImpl<ushort, float, ushort>,
+                addWeightedImpl<ushort, float, short >,
+                addWeightedImpl<ushort, float, int   >,
+                addWeightedImpl<ushort, float, float >,
+                addWeightedImpl<ushort, float, double>
+            },
+            {
+                addWeightedImpl<ushort, double, uchar >,
+                addWeightedImpl<ushort, double, schar >,
+                addWeightedImpl<ushort, double, ushort>,
+                addWeightedImpl<ushort, double, short >,
+                addWeightedImpl<ushort, double, int   >,
+                addWeightedImpl<ushort, double, float >,
+                addWeightedImpl<ushort, double, double>
+            }
+        },
+        {
+            {
+                0/*addWeightedImpl<short, uchar, uchar >*/,
+                0/*addWeightedImpl<short, uchar, schar >*/,
+                0/*addWeightedImpl<short, uchar, ushort>*/,
+                0/*addWeightedImpl<short, uchar, short >*/,
+                0/*addWeightedImpl<short, uchar, int   >*/,
+                0/*addWeightedImpl<short, uchar, float >*/,
+                0/*addWeightedImpl<short, uchar, double>*/
+            },
+            {
+                0/*addWeightedImpl<short, schar, uchar >*/,
+                0/*addWeightedImpl<short, schar, schar >*/,
+                0/*addWeightedImpl<short, schar, ushort>*/,
+                0/*addWeightedImpl<short, schar, short >*/,
+                0/*addWeightedImpl<short, schar, int   >*/,
+                0/*addWeightedImpl<short, schar, float >*/,
+                0/*addWeightedImpl<short, schar, double>*/
+            },
+            {
+                0/*addWeightedImpl<short, ushort, uchar >*/,
+                0/*addWeightedImpl<short, ushort, schar >*/,
+                0/*addWeightedImpl<short, ushort, ushort>*/,
+                0/*addWeightedImpl<short, ushort, short >*/,
+                0/*addWeightedImpl<short, ushort, int   >*/,
+                0/*addWeightedImpl<short, ushort, float >*/,
+                0/*addWeightedImpl<short, ushort, double>*/
+            },
+            {
+                addWeightedImpl<short, short, uchar >,
+                addWeightedImpl<short, short, schar >,
+                addWeightedImpl<short, short, ushort>,
+                addWeightedImpl<short, short, short >,
+                addWeightedImpl<short, short, int   >,
+                addWeightedImpl<short, short, float >,
+                addWeightedImpl<short, short, double>
+            },
+            {
+                addWeightedImpl<short, int, uchar >,
+                addWeightedImpl<short, int, schar >,
+                addWeightedImpl<short, int, ushort>,
+                addWeightedImpl<short, int, short >,
+                addWeightedImpl<short, int, int   >,
+                addWeightedImpl<short, int, float >,
+                addWeightedImpl<short, int, double>
+            },
+            {
+                addWeightedImpl<short, float, uchar >,
+                addWeightedImpl<short, float, schar >,
+                addWeightedImpl<short, float, ushort>,
+                addWeightedImpl<short, float, short >,
+                addWeightedImpl<short, float, int   >,
+                addWeightedImpl<short, float, float >,
+                addWeightedImpl<short, float, double>
+            },
+            {
+                addWeightedImpl<short, double, uchar >,
+                addWeightedImpl<short, double, schar >,
+                addWeightedImpl<short, double, ushort>,
+                addWeightedImpl<short, double, short >,
+                addWeightedImpl<short, double, int   >,
+                addWeightedImpl<short, double, float >,
+                addWeightedImpl<short, double, double>
+            }
+        },
+        {
+            {
+                0/*addWeightedImpl<int, uchar, uchar >*/,
+                0/*addWeightedImpl<int, uchar, schar >*/,
+                0/*addWeightedImpl<int, uchar, ushort>*/,
+                0/*addWeightedImpl<int, uchar, short >*/,
+                0/*addWeightedImpl<int, uchar, int   >*/,
+                0/*addWeightedImpl<int, uchar, float >*/,
+                0/*addWeightedImpl<int, uchar, double>*/
+            },
+            {
+                0/*addWeightedImpl<int, schar, uchar >*/,
+                0/*addWeightedImpl<int, schar, schar >*/,
+                0/*addWeightedImpl<int, schar, ushort>*/,
+                0/*addWeightedImpl<int, schar, short >*/,
+                0/*addWeightedImpl<int, schar, int   >*/,
+                0/*addWeightedImpl<int, schar, float >*/,
+                0/*addWeightedImpl<int, schar, double>*/
+            },
+            {
+                0/*addWeightedImpl<int, ushort, uchar >*/,
+                0/*addWeightedImpl<int, ushort, schar >*/,
+                0/*addWeightedImpl<int, ushort, ushort>*/,
+                0/*addWeightedImpl<int, ushort, short >*/,
+                0/*addWeightedImpl<int, ushort, int   >*/,
+                0/*addWeightedImpl<int, ushort, float >*/,
+                0/*addWeightedImpl<int, ushort, double>*/
+            },
+            {
+                0/*addWeightedImpl<int, short, uchar >*/,
+                0/*addWeightedImpl<int, short, schar >*/,
+                0/*addWeightedImpl<int, short, ushort>*/,
+                0/*addWeightedImpl<int, short, short >*/,
+                0/*addWeightedImpl<int, short, int   >*/,
+                0/*addWeightedImpl<int, short, float >*/,
+                0/*addWeightedImpl<int, short, double>*/
+            },
+            {
+                addWeightedImpl<int, int, uchar >,
+                addWeightedImpl<int, int, schar >,
+                addWeightedImpl<int, int, ushort>,
+                addWeightedImpl<int, int, short >,
+                addWeightedImpl<int, int, int   >,
+                addWeightedImpl<int, int, float >,
+                addWeightedImpl<int, int, double>
+            },
+            {
+                addWeightedImpl<int, float, uchar >,
+                addWeightedImpl<int, float, schar >,
+                addWeightedImpl<int, float, ushort>,
+                addWeightedImpl<int, float, short >,
+                addWeightedImpl<int, float, int   >,
+                addWeightedImpl<int, float, float >,
+                addWeightedImpl<int, float, double>
+            },
+            {
+                addWeightedImpl<int, double, uchar >,
+                addWeightedImpl<int, double, schar >,
+                addWeightedImpl<int, double, ushort>,
+                addWeightedImpl<int, double, short >,
+                addWeightedImpl<int, double, int   >,
+                addWeightedImpl<int, double, float >,
+                addWeightedImpl<int, double, double>
+            }
+        },
+        {
+            {
+                0/*addWeightedImpl<float, uchar, uchar >*/,
+                0/*addWeightedImpl<float, uchar, schar >*/,
+                0/*addWeightedImpl<float, uchar, ushort>*/,
+                0/*addWeightedImpl<float, uchar, short >*/,
+                0/*addWeightedImpl<float, uchar, int   >*/,
+                0/*addWeightedImpl<float, uchar, float >*/,
+                0/*addWeightedImpl<float, uchar, double>*/
+            },
+            {
+                0/*addWeightedImpl<float, schar, uchar >*/,
+                0/*addWeightedImpl<float, schar, schar >*/,
+                0/*addWeightedImpl<float, schar, ushort>*/,
+                0/*addWeightedImpl<float, schar, short >*/,
+                0/*addWeightedImpl<float, schar, int   >*/,
+                0/*addWeightedImpl<float, schar, float >*/,
+                0/*addWeightedImpl<float, schar, double>*/
+            },
+            {
+                0/*addWeightedImpl<float, ushort, uchar >*/,
+                0/*addWeightedImpl<float, ushort, schar >*/,
+                0/*addWeightedImpl<float, ushort, ushort>*/,
+                0/*addWeightedImpl<float, ushort, short >*/,
+                0/*addWeightedImpl<float, ushort, int   >*/,
+                0/*addWeightedImpl<float, ushort, float >*/,
+                0/*addWeightedImpl<float, ushort, double>*/
+            },
+            {
+                0/*addWeightedImpl<float, short, uchar >*/,
+                0/*addWeightedImpl<float, short, schar >*/,
+                0/*addWeightedImpl<float, short, ushort>*/,
+                0/*addWeightedImpl<float, short, short >*/,
+                0/*addWeightedImpl<float, short, int   >*/,
+                0/*addWeightedImpl<float, short, float >*/,
+                0/*addWeightedImpl<float, short, double>*/
+            },
+            {
+                0/*addWeightedImpl<float, int, uchar >*/,
+                0/*addWeightedImpl<float, int, schar >*/,
+                0/*addWeightedImpl<float, int, ushort>*/,
+                0/*addWeightedImpl<float, int, short >*/,
+                0/*addWeightedImpl<float, int, int   >*/,
+                0/*addWeightedImpl<float, int, float >*/,
+                0/*addWeightedImpl<float, int, double>*/
+            },
+            {
+                addWeightedImpl<float, float, uchar >,
+                addWeightedImpl<float, float, schar >,
+                addWeightedImpl<float, float, ushort>,
+                addWeightedImpl<float, float, short >,
+                addWeightedImpl<float, float, int   >,
+                addWeightedImpl<float, float, float >,
+                addWeightedImpl<float, float, double>
+            },
+            {
+                addWeightedImpl<float, double, uchar >,
+                addWeightedImpl<float, double, schar >,
+                addWeightedImpl<float, double, ushort>,
+                addWeightedImpl<float, double, short >,
+                addWeightedImpl<float, double, int   >,
+                addWeightedImpl<float, double, float >,
+                addWeightedImpl<float, double, double>
+            }
+        },
+        {
+            {
+                0/*addWeightedImpl<double, uchar, uchar >*/,
+                0/*addWeightedImpl<double, uchar, schar >*/,
+                0/*addWeightedImpl<double, uchar, ushort>*/,
+                0/*addWeightedImpl<double, uchar, short >*/,
+                0/*addWeightedImpl<double, uchar, int   >*/,
+                0/*addWeightedImpl<double, uchar, float >*/,
+                0/*addWeightedImpl<double, uchar, double>*/
+            },
+            {
+                0/*addWeightedImpl<double, schar, uchar >*/,
+                0/*addWeightedImpl<double, schar, schar >*/,
+                0/*addWeightedImpl<double, schar, ushort>*/,
+                0/*addWeightedImpl<double, schar, short >*/,
+                0/*addWeightedImpl<double, schar, int   >*/,
+                0/*addWeightedImpl<double, schar, float >*/,
+                0/*addWeightedImpl<double, schar, double>*/
+            },
+            {
+                0/*addWeightedImpl<double, ushort, uchar >*/,
+                0/*addWeightedImpl<double, ushort, schar >*/,
+                0/*addWeightedImpl<double, ushort, ushort>*/,
+                0/*addWeightedImpl<double, ushort, short >*/,
+                0/*addWeightedImpl<double, ushort, int   >*/,
+                0/*addWeightedImpl<double, ushort, float >*/,
+                0/*addWeightedImpl<double, ushort, double>*/
+            },
+            {
+                0/*addWeightedImpl<double, short, uchar >*/,
+                0/*addWeightedImpl<double, short, schar >*/,
+                0/*addWeightedImpl<double, short, ushort>*/,
+                0/*addWeightedImpl<double, short, short >*/,
+                0/*addWeightedImpl<double, short, int   >*/,
+                0/*addWeightedImpl<double, short, float >*/,
+                0/*addWeightedImpl<double, short, double>*/
+            },
+            {
+                0/*addWeightedImpl<double, int, uchar >*/,
+                0/*addWeightedImpl<double, int, schar >*/,
+                0/*addWeightedImpl<double, int, ushort>*/,
+                0/*addWeightedImpl<double, int, short >*/,
+                0/*addWeightedImpl<double, int, int   >*/,
+                0/*addWeightedImpl<double, int, float >*/,
+                0/*addWeightedImpl<double, int, double>*/
+            },
+            {
+                0/*addWeightedImpl<double, float, uchar >*/,
+                0/*addWeightedImpl<double, float, schar >*/,
+                0/*addWeightedImpl<double, float, ushort>*/,
+                0/*addWeightedImpl<double, float, short >*/,
+                0/*addWeightedImpl<double, float, int   >*/,
+                0/*addWeightedImpl<double, float, float >*/,
+                0/*addWeightedImpl<double, float, double>*/
+            },
+            {
+                addWeightedImpl<double, double, uchar >,
+                addWeightedImpl<double, double, schar >,
+                addWeightedImpl<double, double, ushort>,
+                addWeightedImpl<double, double, short >,
+                addWeightedImpl<double, double, int   >,
+                addWeightedImpl<double, double, float >,
+                addWeightedImpl<double, double, double>
+            }
+        }
     };
 
-    template <typename T1, typename T2, typename D> struct TransformFunctorTraits< arithm::AddWeighted<T1, T2, D> > : AddWeightedTraits<T1, T2, D, sizeof(T1), sizeof(T2), sizeof(D)>
-    {
-    };
-}}}
+    GpuMat src1 = _src1.getGpuMat();
+    GpuMat src2 = _src2.getGpuMat();
 
-namespace arithm
-{
-    template <typename T1, typename T2, typename D>
-    void addWeighted(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream)
-    {
-        AddWeighted<T1, T2, D> op(alpha, beta, gamma);
+    int sdepth1 = src1.depth();
+    int sdepth2 = src2.depth();
 
-        device::transform((PtrStepSz<T1>) src1, (PtrStepSz<T2>) src2, (PtrStepSz<D>) dst, op, WithOutMask(), stream);
+    ddepth = ddepth >= 0 ? CV_MAT_DEPTH(ddepth) : std::max(sdepth1, sdepth2);
+    const int cn = src1.channels();
+
+    CV_DbgAssert( src2.size() == src1.size() && src2.channels() == cn );
+    CV_DbgAssert( sdepth1 <= CV_64F && sdepth2 <= CV_64F && ddepth <= CV_64F );
+
+    _dst.create(src1.size(), CV_MAKE_TYPE(ddepth, cn));
+    GpuMat dst = _dst.getGpuMat();
+
+    GpuMat src1_ = src1.reshape(1);
+    GpuMat src2_ = src2.reshape(1);
+    GpuMat dst_ = dst.reshape(1);
+
+    if (sdepth1 > sdepth2)
+    {
+        src1_.swap(src2_);
+        std::swap(alpha, beta);
+        std::swap(sdepth1, sdepth2);
     }
 
-    template void addWeighted<uchar, uchar, uchar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<uchar, uchar, schar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<uchar, uchar, ushort>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<uchar, uchar, short>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<uchar, uchar, int>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<uchar, uchar, float>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<uchar, uchar, double>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
+    const func_t func = funcs[sdepth1][sdepth2][ddepth];
 
-    template void addWeighted<uchar, schar, uchar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<uchar, schar, schar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<uchar, schar, ushort>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<uchar, schar, short>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<uchar, schar, int>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<uchar, schar, float>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<uchar, schar, double>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
+    if (!func)
+        CV_Error(cv::Error::StsUnsupportedFormat, "Unsupported combination of source and destination types");
 
-    template void addWeighted<uchar, ushort, uchar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<uchar, ushort, schar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<uchar, ushort, ushort>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<uchar, ushort, short>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<uchar, ushort, int>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<uchar, ushort, float>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<uchar, ushort, double>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-
-    template void addWeighted<uchar, short, uchar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<uchar, short, schar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<uchar, short, ushort>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<uchar, short, short>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<uchar, short, int>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<uchar, short, float>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<uchar, short, double>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-
-    template void addWeighted<uchar, int, uchar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<uchar, int, schar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<uchar, int, ushort>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<uchar, int, short>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<uchar, int, int>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<uchar, int, float>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<uchar, int, double>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-
-    template void addWeighted<uchar, float, uchar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<uchar, float, schar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<uchar, float, ushort>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<uchar, float, short>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<uchar, float, int>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<uchar, float, float>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<uchar, float, double>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-
-    template void addWeighted<uchar, double, uchar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<uchar, double, schar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<uchar, double, ushort>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<uchar, double, short>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<uchar, double, int>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<uchar, double, float>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<uchar, double, double>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-
-
-
-    template void addWeighted<schar, schar, uchar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<schar, schar, schar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<schar, schar, ushort>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<schar, schar, short>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<schar, schar, int>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<schar, schar, float>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<schar, schar, double>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-
-    template void addWeighted<schar, ushort, uchar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<schar, ushort, schar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<schar, ushort, ushort>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<schar, ushort, short>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<schar, ushort, int>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<schar, ushort, float>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<schar, ushort, double>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-
-    template void addWeighted<schar, short, uchar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<schar, short, schar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<schar, short, ushort>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<schar, short, short>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<schar, short, int>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<schar, short, float>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<schar, short, double>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-
-    template void addWeighted<schar, int, uchar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<schar, int, schar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<schar, int, ushort>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<schar, int, short>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<schar, int, int>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<schar, int, float>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<schar, int, double>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-
-    template void addWeighted<schar, float, uchar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<schar, float, schar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<schar, float, ushort>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<schar, float, short>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<schar, float, int>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<schar, float, float>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<schar, float, double>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-
-    template void addWeighted<schar, double, uchar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<schar, double, schar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<schar, double, ushort>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<schar, double, short>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<schar, double, int>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<schar, double, float>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<schar, double, double>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-
-
-
-    template void addWeighted<ushort, ushort, uchar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<ushort, ushort, schar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<ushort, ushort, ushort>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<ushort, ushort, short>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<ushort, ushort, int>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<ushort, ushort, float>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<ushort, ushort, double>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-
-    template void addWeighted<ushort, short, uchar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<ushort, short, schar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<ushort, short, ushort>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<ushort, short, short>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<ushort, short, int>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<ushort, short, float>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<ushort, short, double>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-
-    template void addWeighted<ushort, int, uchar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<ushort, int, schar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<ushort, int, ushort>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<ushort, int, short>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<ushort, int, int>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<ushort, int, float>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<ushort, int, double>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-
-    template void addWeighted<ushort, float, uchar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<ushort, float, schar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<ushort, float, ushort>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<ushort, float, short>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<ushort, float, int>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<ushort, float, float>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<ushort, float, double>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-
-    template void addWeighted<ushort, double, uchar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<ushort, double, schar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<ushort, double, ushort>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<ushort, double, short>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<ushort, double, int>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<ushort, double, float>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<ushort, double, double>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-
-
-
-    template void addWeighted<short, short, uchar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<short, short, schar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<short, short, ushort>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<short, short, short>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<short, short, int>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<short, short, float>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<short, short, double>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-
-    template void addWeighted<short, int, uchar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<short, int, schar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<short, int, ushort>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<short, int, short>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<short, int, int>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<short, int, float>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<short, int, double>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-
-    template void addWeighted<short, float, uchar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<short, float, schar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<short, float, ushort>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<short, float, short>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<short, float, int>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<short, float, float>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<short, float, double>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-
-    template void addWeighted<short, double, uchar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<short, double, schar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<short, double, ushort>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<short, double, short>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<short, double, int>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<short, double, float>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<short, double, double>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-
-
-
-    template void addWeighted<int, int, uchar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<int, int, schar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<int, int, ushort>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<int, int, short>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<int, int, int>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<int, int, float>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<int, int, double>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-
-    template void addWeighted<int, float, uchar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<int, float, schar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<int, float, ushort>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<int, float, short>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<int, float, int>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<int, float, float>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<int, float, double>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-
-    template void addWeighted<int, double, uchar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<int, double, schar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<int, double, ushort>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<int, double, short>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<int, double, int>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<int, double, float>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<int, double, double>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-
-
-
-    template void addWeighted<float, float, uchar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<float, float, schar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<float, float, ushort>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<float, float, short>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<float, float, int>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<float, float, float>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<float, float, double>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-
-    template void addWeighted<float, double, uchar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<float, double, schar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<float, double, ushort>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<float, double, short>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<float, double, int>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<float, double, float>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<float, double, double>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-
-
-
-    template void addWeighted<double, double, uchar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<double, double, schar>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<double, double, ushort>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<double, double, short>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<double, double, int>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<double, double, float>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    template void addWeighted<double, double, double>(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
+    func(src1_, alpha, src2_, beta, gamma, dst_, stream);
 }
 
-#endif /* CUDA_DISABLER */
+#endif
diff --git a/modules/cudaarithm/src/cuda/arithm_func_traits.hpp b/modules/cudaarithm/src/cuda/arithm_func_traits.hpp
deleted file mode 100644
index 051fd9ff08..0000000000
--- a/modules/cudaarithm/src/cuda/arithm_func_traits.hpp
+++ /dev/null
@@ -1,145 +0,0 @@
-/*M///////////////////////////////////////////////////////////////////////////////////////
-//
-//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
-//
-//  By downloading, copying, installing or using the software you agree to this license.
-//  If you do not agree to this license, do not download, install,
-//  copy or use the software.
-//
-//
-//                           License Agreement
-//                For Open Source Computer Vision Library
-//
-// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
-// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
-// Third party copyrights are property of their respective owners.
-//
-// Redistribution and use in source and binary forms, with or without modification,
-// are permitted provided that the following conditions are met:
-//
-//   * Redistribution's of source code must retain the above copyright notice,
-//     this list of conditions and the following disclaimer.
-//
-//   * Redistribution's in binary form must reproduce the above copyright notice,
-//     this list of conditions and the following disclaimer in the documentation
-//     and/or other materials provided with the distribution.
-//
-//   * The name of the copyright holders may not be used to endorse or promote products
-//     derived from this software without specific prior written permission.
-//
-// This software is provided by the copyright holders and contributors "as is" and
-// any express or implied warranties, including, but not limited to, the implied
-// warranties of merchantability and fitness for a particular purpose are disclaimed.
-// In no event shall the Intel Corporation or contributors be liable for any direct,
-// indirect, incidental, special, exemplary, or consequential damages
-// (including, but not limited to, procurement of substitute goods or services;
-// loss of use, data, or profits; or business interruption) however caused
-// and on any theory of liability, whether in contract, strict liability,
-// or tort (including negligence or otherwise) arising in any way out of
-// the use of this software, even if advised of the possibility of such damage.
-//
-//M*/
-
-#ifndef __ARITHM_FUNC_TRAITS_HPP__
-#define __ARITHM_FUNC_TRAITS_HPP__
-
-#include <cstddef>
-
-namespace arithm
-{
-    template <size_t src_size, size_t dst_size> struct ArithmFuncTraits
-    {
-        enum { simple_block_dim_x = 32 };
-        enum { simple_block_dim_y = 8 };
-
-        enum { smart_block_dim_x = 32 };
-        enum { smart_block_dim_y = 8 };
-        enum { smart_shift = 1 };
-    };
-
-    template <> struct ArithmFuncTraits<1, 1>
-    {
-        enum { simple_block_dim_x = 32 };
-        enum { simple_block_dim_y = 8 };
-
-        enum { smart_block_dim_x = 32 };
-        enum { smart_block_dim_y = 8 };
-        enum { smart_shift = 4 };
-    };
-    template <> struct ArithmFuncTraits<1, 2>
-    {
-        enum { simple_block_dim_x = 32 };
-        enum { simple_block_dim_y = 8 };
-
-        enum { smart_block_dim_x = 32 };
-        enum { smart_block_dim_y = 8 };
-        enum { smart_shift = 4 };
-    };
-    template <> struct ArithmFuncTraits<1, 4>
-    {
-        enum { simple_block_dim_x = 32 };
-        enum { simple_block_dim_y = 8 };
-
-        enum { smart_block_dim_x = 32 };
-        enum { smart_block_dim_y = 8 };
-        enum { smart_shift = 4 };
-    };
-
-    template <> struct ArithmFuncTraits<2, 1>
-    {
-        enum { simple_block_dim_x = 32 };
-        enum { simple_block_dim_y = 8 };
-
-        enum { smart_block_dim_x = 32 };
-        enum { smart_block_dim_y = 8 };
-        enum { smart_shift = 4 };
-    };
-    template <> struct ArithmFuncTraits<2, 2>
-    {
-        enum { simple_block_dim_x = 32 };
-        enum { simple_block_dim_y = 8 };
-
-        enum { smart_block_dim_x = 32 };
-        enum { smart_block_dim_y = 8 };
-        enum { smart_shift = 4 };
-    };
-    template <> struct ArithmFuncTraits<2, 4>
-    {
-        enum { simple_block_dim_x = 32 };
-        enum { simple_block_dim_y = 8 };
-
-        enum { smart_block_dim_x = 32 };
-        enum { smart_block_dim_y = 8 };
-        enum { smart_shift = 4 };
-    };
-
-    template <> struct ArithmFuncTraits<4, 1>
-    {
-        enum { simple_block_dim_x = 32 };
-        enum { simple_block_dim_y = 8 };
-
-        enum { smart_block_dim_x = 32 };
-        enum { smart_block_dim_y = 8 };
-        enum { smart_shift = 4 };
-    };
-    template <> struct ArithmFuncTraits<4, 2>
-    {
-        enum { simple_block_dim_x = 32 };
-        enum { simple_block_dim_y = 8 };
-
-        enum { smart_block_dim_x = 32 };
-        enum { smart_block_dim_y = 8 };
-        enum { smart_shift = 4 };
-    };
-    template <> struct ArithmFuncTraits<4, 4>
-    {
-        enum { simple_block_dim_x = 32 };
-        enum { simple_block_dim_y = 8 };
-
-        enum { smart_block_dim_x = 32 };
-        enum { smart_block_dim_y = 8 };
-        enum { smart_shift = 4 };
-    };
-}
-
-#endif // __ARITHM_FUNC_TRAITS_HPP__
diff --git a/modules/cudaarithm/src/cuda/bitwise_mat.cu b/modules/cudaarithm/src/cuda/bitwise_mat.cu
index 7a90cf3fc3..b2bf288be7 100644
--- a/modules/cudaarithm/src/cuda/bitwise_mat.cu
+++ b/modules/cudaarithm/src/cuda/bitwise_mat.cu
@@ -40,87 +40,187 @@
 //
 //M*/
 
-#if !defined CUDA_DISABLER
+#include "opencv2/opencv_modules.hpp"
 
-#include "opencv2/core/cuda/common.hpp"
-#include "opencv2/core/cuda/functional.hpp"
-#include "opencv2/core/cuda/transform.hpp"
-#include "opencv2/core/cuda/saturate_cast.hpp"
-#include "opencv2/core/cuda/simd_functions.hpp"
+#ifndef HAVE_OPENCV_CUDEV
 
-#include "arithm_func_traits.hpp"
+#error "opencv_cudev is required"
 
-using namespace cv::cuda;
-using namespace cv::cuda::device;
+#else
 
-namespace cv { namespace cuda { namespace device
+#include "opencv2/cudaarithm.hpp"
+#include "opencv2/cudev.hpp"
+
+using namespace cv::cudev;
+
+void bitMat(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat& mask, double, Stream& stream, int op);
+
+//////////////////////////////////////////////////////////////////////////////
+/// bitwise_not
+
+void cv::cuda::bitwise_not(InputArray _src, OutputArray _dst, InputArray _mask, Stream& stream)
 {
-    template <typename T> struct TransformFunctorTraits< bit_not<T> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(T)>
-    {
-    };
+    GpuMat src = _src.getGpuMat();
+    GpuMat mask = _mask.getGpuMat();
 
-    template <typename T> struct TransformFunctorTraits< bit_and<T> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(T)>
-    {
-    };
+    const int depth = src.depth();
 
-    template <typename T> struct TransformFunctorTraits< bit_or<T> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(T)>
-    {
-    };
+    CV_DbgAssert( depth <= CV_32F );
+    CV_DbgAssert( mask.empty() || (mask.type() == CV_8UC1 && mask.size() == src.size()) );
 
-    template <typename T> struct TransformFunctorTraits< bit_xor<T> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(T)>
-    {
-    };
-}}}
+    _dst.create(src.size(), src.type());
+    GpuMat dst = _dst.getGpuMat();
 
-namespace arithm
-{
-    template <typename T> void bitMatNot(PtrStepSzb src, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream)
+    if (mask.empty())
     {
-        if (mask.data)
-            device::transform((PtrStepSz<T>) src, (PtrStepSz<T>) dst, bit_not<T>(), mask, stream);
+        const int bcols = (int) (src.cols * src.elemSize());
+
+        if ((bcols & 3) == 0)
+        {
+            const int vcols = bcols >> 2;
+
+            GlobPtrSz<uint> vsrc = globPtr((uint*) src.data, src.step, src.rows, vcols);
+            GlobPtrSz<uint> vdst = globPtr((uint*) dst.data, dst.step, src.rows, vcols);
+
+            gridTransformUnary(vsrc, vdst, bit_not<uint>(), stream);
+        }
+        else if ((bcols & 1) == 0)
+        {
+            const int vcols = bcols >> 1;
+
+            GlobPtrSz<ushort> vsrc = globPtr((ushort*) src.data, src.step, src.rows, vcols);
+            GlobPtrSz<ushort> vdst = globPtr((ushort*) dst.data, dst.step, src.rows, vcols);
+
+            gridTransformUnary(vsrc, vdst, bit_not<ushort>(), stream);
+        }
         else
-            device::transform((PtrStepSz<T>) src, (PtrStepSz<T>) dst, bit_not<T>(), WithOutMask(), stream);
-    }
+        {
+            GlobPtrSz<uchar> vsrc = globPtr((uchar*) src.data, src.step, src.rows, bcols);
+            GlobPtrSz<uchar> vdst = globPtr((uchar*) dst.data, dst.step, src.rows, bcols);
 
-    template <typename T> void bitMatAnd(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream)
+            gridTransformUnary(vsrc, vdst, bit_not<uchar>(), stream);
+        }
+    }
+    else
     {
-        if (mask.data)
-            device::transform((PtrStepSz<T>) src1, (PtrStepSz<T>) src2, (PtrStepSz<T>) dst, bit_and<T>(), mask, stream);
+        if (depth == CV_32F || depth == CV_32S)
+        {
+            GlobPtrSz<uint> vsrc = globPtr((uint*) src.data, src.step, src.rows, src.cols * src.channels());
+            GlobPtrSz<uint> vdst = globPtr((uint*) dst.data, dst.step, src.rows, src.cols * src.channels());
+
+            gridTransformUnary(vsrc, vdst, bit_not<uint>(), singleMaskChannels(globPtr<uchar>(mask), src.channels()), stream);
+        }
+        else if (depth == CV_16S || depth == CV_16U)
+        {
+            GlobPtrSz<ushort> vsrc = globPtr((ushort*) src.data, src.step, src.rows, src.cols * src.channels());
+            GlobPtrSz<ushort> vdst = globPtr((ushort*) dst.data, dst.step, src.rows, src.cols * src.channels());
+
+            gridTransformUnary(vsrc, vdst, bit_not<ushort>(), singleMaskChannels(globPtr<uchar>(mask), src.channels()), stream);
+        }
         else
-            device::transform((PtrStepSz<T>) src1, (PtrStepSz<T>) src2, (PtrStepSz<T>) dst, bit_and<T>(), WithOutMask(), stream);
+        {
+            GlobPtrSz<uchar> vsrc = globPtr((uchar*) src.data, src.step, src.rows, src.cols * src.channels());
+            GlobPtrSz<uchar> vdst = globPtr((uchar*) dst.data, dst.step, src.rows, src.cols * src.channels());
+
+            gridTransformUnary(vsrc, vdst, bit_not<uchar>(), singleMaskChannels(globPtr<uchar>(mask), src.channels()), stream);
+        }
     }
-
-    template <typename T> void bitMatOr(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream)
-    {
-        if (mask.data)
-            device::transform((PtrStepSz<T>) src1, (PtrStepSz<T>) src2, (PtrStepSz<T>) dst, bit_or<T>(), mask, stream);
-        else
-            device::transform((PtrStepSz<T>) src1, (PtrStepSz<T>) src2, (PtrStepSz<T>) dst, bit_or<T>(), WithOutMask(), stream);
-    }
-
-    template <typename T> void bitMatXor(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream)
-    {
-        if (mask.data)
-            device::transform((PtrStepSz<T>) src1, (PtrStepSz<T>) src2, (PtrStepSz<T>) dst, bit_xor<T>(), mask, stream);
-        else
-            device::transform((PtrStepSz<T>) src1, (PtrStepSz<T>) src2, (PtrStepSz<T>) dst, bit_xor<T>(), WithOutMask(), stream);
-    }
-
-    template void bitMatNot<uchar>(PtrStepSzb src, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void bitMatNot<ushort>(PtrStepSzb src, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void bitMatNot<uint>(PtrStepSzb src, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-
-    template void bitMatAnd<uchar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void bitMatAnd<ushort>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void bitMatAnd<uint>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-
-    template void bitMatOr<uchar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void bitMatOr<ushort>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void bitMatOr<uint>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-
-    template void bitMatXor<uchar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void bitMatXor<ushort>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void bitMatXor<uint>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
 }
 
-#endif // CUDA_DISABLER
+//////////////////////////////////////////////////////////////////////////////
+/// Binary bitwise logical operations
+
+namespace
+{
+    template <template <typename> class Op, typename T>
+    void bitMatOp(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat& mask, Stream& stream)
+    {
+        GlobPtrSz<T> vsrc1 = globPtr((T*) src1.data, src1.step, src1.rows, src1.cols * src1.channels());
+        GlobPtrSz<T> vsrc2 = globPtr((T*) src2.data, src2.step, src1.rows, src1.cols * src1.channels());
+        GlobPtrSz<T> vdst = globPtr((T*) dst.data, dst.step, src1.rows, src1.cols * src1.channels());
+
+        if (mask.data)
+            gridTransformBinary(vsrc1, vsrc2, vdst, Op<T>(), singleMaskChannels(globPtr<uchar>(mask), src1.channels()), stream);
+        else
+            gridTransformBinary(vsrc1, vsrc2, vdst, Op<T>(), stream);
+    }
+}
+
+void bitMat(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat& mask, double, Stream& stream, int op)
+{
+    typedef void (*func_t)(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat& mask, Stream& stream);
+    static const func_t funcs32[] =
+    {
+        bitMatOp<bit_and, uint>,
+        bitMatOp<bit_or, uint>,
+        bitMatOp<bit_xor, uint>
+    };
+    static const func_t funcs16[] =
+    {
+        bitMatOp<bit_and, ushort>,
+        bitMatOp<bit_or, ushort>,
+        bitMatOp<bit_xor, ushort>
+    };
+    static const func_t funcs8[] =
+    {
+        bitMatOp<bit_and, uchar>,
+        bitMatOp<bit_or, uchar>,
+        bitMatOp<bit_xor, uchar>
+    };
+
+    const int depth = src1.depth();
+
+    CV_DbgAssert( depth <= CV_32F );
+    CV_DbgAssert( op >= 0 && op < 3 );
+
+    if (mask.empty())
+    {
+        const int bcols = (int) (src1.cols * src1.elemSize());
+
+        if ((bcols & 3) == 0)
+        {
+            const int vcols = bcols >> 2;
+
+            GpuMat vsrc1(src1.rows, vcols, CV_32SC1, src1.data, src1.step);
+            GpuMat vsrc2(src1.rows, vcols, CV_32SC1, src2.data, src2.step);
+            GpuMat vdst(src1.rows, vcols, CV_32SC1, dst.data, dst.step);
+
+            funcs32[op](vsrc1, vsrc2, vdst, GpuMat(), stream);
+        }
+        else if ((bcols & 1) == 0)
+        {
+            const int vcols = bcols >> 1;
+
+            GpuMat vsrc1(src1.rows, vcols, CV_16UC1, src1.data, src1.step);
+            GpuMat vsrc2(src1.rows, vcols, CV_16UC1, src2.data, src2.step);
+            GpuMat vdst(src1.rows, vcols, CV_16UC1, dst.data, dst.step);
+
+            funcs16[op](vsrc1, vsrc2, vdst, GpuMat(), stream);
+        }
+        else
+        {
+            GpuMat vsrc1(src1.rows, bcols, CV_8UC1, src1.data, src1.step);
+            GpuMat vsrc2(src1.rows, bcols, CV_8UC1, src2.data, src2.step);
+            GpuMat vdst(src1.rows, bcols, CV_8UC1, dst.data, dst.step);
+
+            funcs8[op](vsrc1, vsrc2, vdst, GpuMat(), stream);
+        }
+    }
+    else
+    {
+        if (depth == CV_32F || depth == CV_32S)
+        {
+            funcs32[op](src1, src2, dst, mask, stream);
+        }
+        else if (depth == CV_16S || depth == CV_16U)
+        {
+            funcs16[op](src1, src2, dst, mask, stream);
+        }
+        else
+        {
+            funcs8[op](src1, src2, dst, mask, stream);
+        }
+    }
+}
+
+#endif
diff --git a/modules/cudaarithm/src/cuda/bitwise_scalar.cu b/modules/cudaarithm/src/cuda/bitwise_scalar.cu
index 3de413370a..e8e412471f 100644
--- a/modules/cudaarithm/src/cuda/bitwise_scalar.cu
+++ b/modules/cudaarithm/src/cuda/bitwise_scalar.cu
@@ -40,65 +40,132 @@
 //
 //M*/
 
-#if !defined CUDA_DISABLER
+#include "opencv2/opencv_modules.hpp"
 
-#include "opencv2/core/cuda/common.hpp"
-#include "opencv2/core/cuda/functional.hpp"
-#include "opencv2/core/cuda/transform.hpp"
-#include "opencv2/core/cuda/saturate_cast.hpp"
-#include "opencv2/core/cuda/simd_functions.hpp"
+#ifndef HAVE_OPENCV_CUDEV
 
-#include "arithm_func_traits.hpp"
+#error "opencv_cudev is required"
 
-using namespace cv::cuda;
-using namespace cv::cuda::device;
+#else
 
-namespace cv { namespace cuda { namespace device
+#include "opencv2/cudev.hpp"
+#include "opencv2/core/private.cuda.hpp"
+
+using namespace cv::cudev;
+
+void bitScalar(const GpuMat& src, cv::Scalar value, bool, GpuMat& dst, const GpuMat& mask, double, Stream& stream, int op);
+
+namespace
 {
-    template <typename T> struct TransformFunctorTraits< binder2nd< bit_and<T> > > : arithm::ArithmFuncTraits<sizeof(T), sizeof(T)>
+    template <template <typename> class Op, typename T>
+    void bitScalarOp(const GpuMat& src, uint value, GpuMat& dst, Stream& stream)
     {
-    };
-
-    template <typename T> struct TransformFunctorTraits< binder2nd< bit_or<T> > > : arithm::ArithmFuncTraits<sizeof(T), sizeof(T)>
-    {
-    };
-
-    template <typename T> struct TransformFunctorTraits< binder2nd< bit_xor<T> > > : arithm::ArithmFuncTraits<sizeof(T), sizeof(T)>
-    {
-    };
-}}}
-
-namespace arithm
-{
-    template <typename T> void bitScalarAnd(PtrStepSzb src1, uint src2, PtrStepSzb dst, cudaStream_t stream)
-    {
-        device::transform((PtrStepSz<T>) src1, (PtrStepSz<T>) dst, cv::cuda::device::bind2nd(bit_and<T>(), src2), WithOutMask(), stream);
+        gridTransformUnary(globPtr<T>(src), globPtr<T>(dst), bind2nd(Op<T>(), value), stream);
     }
 
-    template <typename T> void bitScalarOr(PtrStepSzb src1, uint src2, PtrStepSzb dst, cudaStream_t stream)
+    typedef void (*bit_scalar_func_t)(const GpuMat& src, uint value, GpuMat& dst, Stream& stream);
+
+    template <typename T, bit_scalar_func_t func> struct BitScalar
     {
-        device::transform((PtrStepSz<T>) src1, (PtrStepSz<T>) dst, cv::cuda::device::bind2nd(bit_or<T>(), src2), WithOutMask(), stream);
-    }
+        static void call(const GpuMat& src, cv::Scalar value, GpuMat& dst, Stream& stream)
+        {
+            func(src, cv::saturate_cast<T>(value[0]), dst, stream);
+        }
+    };
 
-    template <typename T> void bitScalarXor(PtrStepSzb src1, uint src2, PtrStepSzb dst, cudaStream_t stream)
+    template <bit_scalar_func_t func> struct BitScalar4
     {
-        device::transform((PtrStepSz<T>) src1, (PtrStepSz<T>) dst, cv::cuda::device::bind2nd(bit_xor<T>(), src2), WithOutMask(), stream);
-    }
+        static void call(const GpuMat& src, cv::Scalar value, GpuMat& dst, Stream& stream)
+        {
+            uint packedVal = 0;
 
-    template void bitScalarAnd<uchar>(PtrStepSzb src1, uint src2, PtrStepSzb dst, cudaStream_t stream);
-    template void bitScalarAnd<ushort>(PtrStepSzb src1, uint src2, PtrStepSzb dst, cudaStream_t stream);
-    template void bitScalarAnd<int>(PtrStepSzb src1, uint src2, PtrStepSzb dst, cudaStream_t stream);
-    template void bitScalarAnd<unsigned int>(PtrStepSzb src1, uint src2, PtrStepSzb dst, cudaStream_t stream);
+            packedVal |= cv::saturate_cast<uchar>(value[0]);
+            packedVal |= cv::saturate_cast<uchar>(value[1]) << 8;
+            packedVal |= cv::saturate_cast<uchar>(value[2]) << 16;
+            packedVal |= cv::saturate_cast<uchar>(value[3]) << 24;
 
-    template void bitScalarOr<uchar>(PtrStepSzb src1, uint src2, PtrStepSzb dst, cudaStream_t stream);
-    template void bitScalarOr<ushort>(PtrStepSzb src1, uint src2, PtrStepSzb dst, cudaStream_t stream);
-    template void bitScalarOr<int>(PtrStepSzb src1, uint src2, PtrStepSzb dst, cudaStream_t stream);
-    template void bitScalarOr<unsigned int>(PtrStepSzb src1, uint src2, PtrStepSzb dst, cudaStream_t stream);
+            func(src, packedVal, dst, stream);
+        }
+    };
 
-    template void bitScalarXor<uchar>(PtrStepSzb src1, uint src2, PtrStepSzb dst, cudaStream_t stream);
-    template void bitScalarXor<ushort>(PtrStepSzb src1, uint src2, PtrStepSzb dst, cudaStream_t stream);
-    template void bitScalarXor<int>(PtrStepSzb src1, uint src2, PtrStepSzb dst, cudaStream_t stream);
-    template void bitScalarXor<unsigned int>(PtrStepSzb src1, uint src2, PtrStepSzb dst, cudaStream_t stream);
+    template <int DEPTH, int cn> struct NppBitwiseCFunc
+    {
+        typedef typename NPPTypeTraits<DEPTH>::npp_type npp_type;
+
+        typedef NppStatus (*func_t)(const npp_type* pSrc1, int nSrc1Step, const npp_type* pConstants, npp_type* pDst, int nDstStep, NppiSize oSizeROI);
+    };
+
+    template <int DEPTH, int cn, typename NppBitwiseCFunc<DEPTH, cn>::func_t func> struct NppBitwiseC
+    {
+        typedef typename NppBitwiseCFunc<DEPTH, cn>::npp_type npp_type;
+
+        static void call(const GpuMat& src, cv::Scalar value, GpuMat& dst, Stream& _stream)
+        {
+            cudaStream_t stream = StreamAccessor::getStream(_stream);
+            NppStreamHandler h(stream);
+
+            NppiSize oSizeROI;
+            oSizeROI.width = src.cols;
+            oSizeROI.height = src.rows;
+
+            const npp_type pConstants[] =
+            {
+                cv::saturate_cast<npp_type>(value[0]),
+                cv::saturate_cast<npp_type>(value[1]),
+                cv::saturate_cast<npp_type>(value[2]),
+                cv::saturate_cast<npp_type>(value[3])
+            };
+
+            nppSafeCall( func(src.ptr<npp_type>(), static_cast<int>(src.step), pConstants, dst.ptr<npp_type>(), static_cast<int>(dst.step), oSizeROI) );
+
+            if (stream == 0)
+                CV_CUDEV_SAFE_CALL( cudaDeviceSynchronize() );
+        }
+    };
 }
 
-#endif // CUDA_DISABLER
+void bitScalar(const GpuMat& src, cv::Scalar value, bool, GpuMat& dst, const GpuMat& mask, double, Stream& stream, int op)
+{
+    (void) mask;
+
+    typedef void (*func_t)(const GpuMat& src, cv::Scalar value, GpuMat& dst, Stream& stream);
+    static const func_t funcs[3][6][4] =
+    {
+        {
+            {BitScalar<uchar, bitScalarOp<bit_and, uchar> >::call  , 0, NppBitwiseC<CV_8U , 3, nppiAndC_8u_C3R >::call, BitScalar4< bitScalarOp<bit_and, uint> >::call},
+            {BitScalar<uchar, bitScalarOp<bit_and, uchar> >::call  , 0, NppBitwiseC<CV_8U , 3, nppiAndC_8u_C3R >::call, BitScalar4< bitScalarOp<bit_and, uint> >::call},
+            {BitScalar<ushort, bitScalarOp<bit_and, ushort> >::call, 0, NppBitwiseC<CV_16U, 3, nppiAndC_16u_C3R>::call, NppBitwiseC<CV_16U, 4, nppiAndC_16u_C4R>::call},
+            {BitScalar<ushort, bitScalarOp<bit_and, ushort> >::call, 0, NppBitwiseC<CV_16U, 3, nppiAndC_16u_C3R>::call, NppBitwiseC<CV_16U, 4, nppiAndC_16u_C4R>::call},
+            {BitScalar<uint, bitScalarOp<bit_and, uint> >::call    , 0, NppBitwiseC<CV_32S, 3, nppiAndC_32s_C3R>::call, NppBitwiseC<CV_32S, 4, nppiAndC_32s_C4R>::call},
+            {BitScalar<uint, bitScalarOp<bit_and, uint> >::call    , 0, NppBitwiseC<CV_32S, 3, nppiAndC_32s_C3R>::call, NppBitwiseC<CV_32S, 4, nppiAndC_32s_C4R>::call}
+        },
+        {
+            {BitScalar<uchar, bitScalarOp<bit_or, uchar> >::call  , 0, NppBitwiseC<CV_8U , 3, nppiOrC_8u_C3R >::call, BitScalar4< bitScalarOp<bit_or, uint> >::call},
+            {BitScalar<uchar, bitScalarOp<bit_or, uchar> >::call  , 0, NppBitwiseC<CV_8U , 3, nppiOrC_8u_C3R >::call, BitScalar4< bitScalarOp<bit_or, uint> >::call},
+            {BitScalar<ushort, bitScalarOp<bit_or, ushort> >::call, 0, NppBitwiseC<CV_16U, 3, nppiOrC_16u_C3R>::call, NppBitwiseC<CV_16U, 4, nppiOrC_16u_C4R>::call},
+            {BitScalar<ushort, bitScalarOp<bit_or, ushort> >::call, 0, NppBitwiseC<CV_16U, 3, nppiOrC_16u_C3R>::call, NppBitwiseC<CV_16U, 4, nppiOrC_16u_C4R>::call},
+            {BitScalar<uint, bitScalarOp<bit_or, uint> >::call    , 0, NppBitwiseC<CV_32S, 3, nppiOrC_32s_C3R>::call, NppBitwiseC<CV_32S, 4, nppiOrC_32s_C4R>::call},
+            {BitScalar<uint, bitScalarOp<bit_or, uint> >::call    , 0, NppBitwiseC<CV_32S, 3, nppiOrC_32s_C3R>::call, NppBitwiseC<CV_32S, 4, nppiOrC_32s_C4R>::call}
+        },
+        {
+            {BitScalar<uchar, bitScalarOp<bit_xor, uchar> >::call  , 0, NppBitwiseC<CV_8U , 3, nppiXorC_8u_C3R >::call, BitScalar4< bitScalarOp<bit_xor, uint> >::call},
+            {BitScalar<uchar, bitScalarOp<bit_xor, uchar> >::call  , 0, NppBitwiseC<CV_8U , 3, nppiXorC_8u_C3R >::call, BitScalar4< bitScalarOp<bit_xor, uint> >::call},
+            {BitScalar<ushort, bitScalarOp<bit_xor, ushort> >::call, 0, NppBitwiseC<CV_16U, 3, nppiXorC_16u_C3R>::call, NppBitwiseC<CV_16U, 4, nppiXorC_16u_C4R>::call},
+            {BitScalar<ushort, bitScalarOp<bit_xor, ushort> >::call, 0, NppBitwiseC<CV_16U, 3, nppiXorC_16u_C3R>::call, NppBitwiseC<CV_16U, 4, nppiXorC_16u_C4R>::call},
+            {BitScalar<uint, bitScalarOp<bit_xor, uint> >::call    , 0, NppBitwiseC<CV_32S, 3, nppiXorC_32s_C3R>::call, NppBitwiseC<CV_32S, 4, nppiXorC_32s_C4R>::call},
+            {BitScalar<uint, bitScalarOp<bit_xor, uint> >::call    , 0, NppBitwiseC<CV_32S, 3, nppiXorC_32s_C3R>::call, NppBitwiseC<CV_32S, 4, nppiXorC_32s_C4R>::call}
+        }
+    };
+
+    const int depth = src.depth();
+    const int cn = src.channels();
+
+    CV_DbgAssert( depth <= CV_32F );
+    CV_DbgAssert( cn == 1 || cn == 3 || cn == 4 );
+    CV_DbgAssert( mask.empty() );
+    CV_DbgAssert( op >= 0 && op < 3 );
+
+    funcs[op][depth][cn - 1](src, value, dst, stream);
+}
+
+#endif
diff --git a/modules/cudaarithm/src/cuda/cmp_mat.cu b/modules/cudaarithm/src/cuda/cmp_mat.cu
index cdbb9639b4..3693fc2b78 100644
--- a/modules/cudaarithm/src/cuda/cmp_mat.cu
+++ b/modules/cudaarithm/src/cuda/cmp_mat.cu
@@ -40,64 +40,23 @@
 //
 //M*/
 
-#if !defined CUDA_DISABLER
+#include "opencv2/opencv_modules.hpp"
 
-#include "opencv2/core/cuda/common.hpp"
-#include "opencv2/core/cuda/functional.hpp"
-#include "opencv2/core/cuda/transform.hpp"
-#include "opencv2/core/cuda/saturate_cast.hpp"
-#include "opencv2/core/cuda/simd_functions.hpp"
+#ifndef HAVE_OPENCV_CUDEV
 
-#include "arithm_func_traits.hpp"
+#error "opencv_cudev is required"
 
-using namespace cv::cuda;
-using namespace cv::cuda::device;
+#else
 
-namespace arithm
+#include "opencv2/cudev.hpp"
+
+using namespace cv::cudev;
+
+void cmpMat(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat&, double, Stream& stream, int cmpop);
+
+namespace
 {
-    struct VCmpEq4 : binary_function<uint, uint, uint>
-    {
-        __device__ __forceinline__ uint operator ()(uint a, uint b) const
-        {
-            return vcmpeq4(a, b);
-        }
-
-        __host__ __device__ __forceinline__ VCmpEq4() {}
-        __host__ __device__ __forceinline__ VCmpEq4(const VCmpEq4&) {}
-    };
-    struct VCmpNe4 : binary_function<uint, uint, uint>
-    {
-        __device__ __forceinline__ uint operator ()(uint a, uint b) const
-        {
-            return vcmpne4(a, b);
-        }
-
-        __host__ __device__ __forceinline__ VCmpNe4() {}
-        __host__ __device__ __forceinline__ VCmpNe4(const VCmpNe4&) {}
-    };
-    struct VCmpLt4 : binary_function<uint, uint, uint>
-    {
-        __device__ __forceinline__ uint operator ()(uint a, uint b) const
-        {
-            return vcmplt4(a, b);
-        }
-
-        __host__ __device__ __forceinline__ VCmpLt4() {}
-        __host__ __device__ __forceinline__ VCmpLt4(const VCmpLt4&) {}
-    };
-    struct VCmpLe4 : binary_function<uint, uint, uint>
-    {
-        __device__ __forceinline__ uint operator ()(uint a, uint b) const
-        {
-            return vcmple4(a, b);
-        }
-
-        __host__ __device__ __forceinline__ VCmpLe4() {}
-        __host__ __device__ __forceinline__ VCmpLe4(const VCmpLe4&) {}
-    };
-
-    template <class Op, typename T>
-    struct Cmp : binary_function<T, T, uchar>
+    template <class Op, typename T> struct CmpOp : binary_function<T, T, uchar>
     {
         __device__ __forceinline__ uchar operator()(T a, T b) const
         {
@@ -105,102 +64,156 @@ namespace arithm
             return -op(a, b);
         }
     };
-}
 
-namespace cv { namespace cuda { namespace device
-{
-    template <> struct TransformFunctorTraits< arithm::VCmpEq4 > : arithm::ArithmFuncTraits<sizeof(uint), sizeof(uint)>
+    template <typename ScalarDepth> struct TransformPolicy : DefaultTransformPolicy
     {
     };
-    template <> struct TransformFunctorTraits< arithm::VCmpNe4 > : arithm::ArithmFuncTraits<sizeof(uint), sizeof(uint)>
+    template <> struct TransformPolicy<double> : DefaultTransformPolicy
     {
+        enum {
+            shift = 1
+        };
     };
-    template <> struct TransformFunctorTraits< arithm::VCmpLt4 > : arithm::ArithmFuncTraits<sizeof(uint), sizeof(uint)>
-    {
-    };
-    template <> struct TransformFunctorTraits< arithm::VCmpLe4 > : arithm::ArithmFuncTraits<sizeof(uint), sizeof(uint)>
-    {
-    };
-
-    template <class Op, typename T> struct TransformFunctorTraits< arithm::Cmp<Op, T> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(uchar)>
-    {
-    };
-}}}
-
-namespace arithm
-{
-    void cmpMatEq_v4(PtrStepSz<uint> src1, PtrStepSz<uint> src2, PtrStepSz<uint> dst, cudaStream_t stream)
-    {
-        device::transform(src1, src2, dst, VCmpEq4(), WithOutMask(), stream);
-    }
-    void cmpMatNe_v4(PtrStepSz<uint> src1, PtrStepSz<uint> src2, PtrStepSz<uint> dst, cudaStream_t stream)
-    {
-        device::transform(src1, src2, dst, VCmpNe4(), WithOutMask(), stream);
-    }
-    void cmpMatLt_v4(PtrStepSz<uint> src1, PtrStepSz<uint> src2, PtrStepSz<uint> dst, cudaStream_t stream)
-    {
-        device::transform(src1, src2, dst, VCmpLt4(), WithOutMask(), stream);
-    }
-    void cmpMatLe_v4(PtrStepSz<uint> src1, PtrStepSz<uint> src2, PtrStepSz<uint> dst, cudaStream_t stream)
-    {
-        device::transform(src1, src2, dst, VCmpLe4(), WithOutMask(), stream);
-    }
 
     template <template <typename> class Op, typename T>
-    void cmpMat(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream)
+    void cmpMat_v1(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, Stream& stream)
     {
-        Cmp<Op<T>, T> op;
-        device::transform((PtrStepSz<T>) src1, (PtrStepSz<T>) src2, dst, op, WithOutMask(), stream);
+        CmpOp<Op<T>, T> op;
+        gridTransformBinary_< TransformPolicy<T> >(globPtr<T>(src1), globPtr<T>(src2), globPtr<uchar>(dst), op, stream);
     }
 
-    template <typename T> void cmpMatEq(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream)
+    struct VCmpEq4 : binary_function<uint, uint, uint>
     {
-        cmpMat<equal_to, T>(src1, src2, dst, stream);
-    }
-    template <typename T> void cmpMatNe(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream)
+        __device__ __forceinline__ uint operator ()(uint a, uint b) const
+        {
+            return vcmpeq4(a, b);
+        }
+    };
+    struct VCmpNe4 : binary_function<uint, uint, uint>
     {
-        cmpMat<not_equal_to, T>(src1, src2, dst, stream);
-    }
-    template <typename T> void cmpMatLt(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream)
+        __device__ __forceinline__ uint operator ()(uint a, uint b) const
+        {
+            return vcmpne4(a, b);
+        }
+    };
+    struct VCmpLt4 : binary_function<uint, uint, uint>
     {
-        cmpMat<less, T>(src1, src2, dst, stream);
-    }
-    template <typename T> void cmpMatLe(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream)
+        __device__ __forceinline__ uint operator ()(uint a, uint b) const
+        {
+            return vcmplt4(a, b);
+        }
+    };
+    struct VCmpLe4 : binary_function<uint, uint, uint>
     {
-        cmpMat<less_equal, T>(src1, src2, dst, stream);
+        __device__ __forceinline__ uint operator ()(uint a, uint b) const
+        {
+            return vcmple4(a, b);
+        }
+    };
+
+    void cmpMatEq_v4(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, Stream& stream)
+    {
+        const int vcols = src1.cols >> 2;
+
+        GlobPtrSz<uint> src1_ = globPtr((uint*) src1.data, src1.step, src1.rows, vcols);
+        GlobPtrSz<uint> src2_ = globPtr((uint*) src2.data, src2.step, src1.rows, vcols);
+        GlobPtrSz<uint> dst_ = globPtr((uint*) dst.data, dst.step, src1.rows, vcols);
+
+        gridTransformBinary(src1_, src2_, dst_, VCmpEq4(), stream);
     }
+    void cmpMatNe_v4(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, Stream& stream)
+    {
+        const int vcols = src1.cols >> 2;
 
-    template void cmpMatEq<uchar >(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template void cmpMatEq<schar >(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template void cmpMatEq<ushort>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template void cmpMatEq<short >(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template void cmpMatEq<int   >(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template void cmpMatEq<float >(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template void cmpMatEq<double>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
+        GlobPtrSz<uint> src1_ = globPtr((uint*) src1.data, src1.step, src1.rows, vcols);
+        GlobPtrSz<uint> src2_ = globPtr((uint*) src2.data, src2.step, src1.rows, vcols);
+        GlobPtrSz<uint> dst_ = globPtr((uint*) dst.data, dst.step, src1.rows, vcols);
 
-    template void cmpMatNe<uchar >(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template void cmpMatNe<schar >(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template void cmpMatNe<ushort>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template void cmpMatNe<short >(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template void cmpMatNe<int   >(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template void cmpMatNe<float >(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template void cmpMatNe<double>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
+        gridTransformBinary(src1_, src2_, dst_, VCmpNe4(), stream);
+    }
+    void cmpMatLt_v4(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, Stream& stream)
+    {
+        const int vcols = src1.cols >> 2;
 
-    template void cmpMatLt<uchar >(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template void cmpMatLt<schar >(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template void cmpMatLt<ushort>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template void cmpMatLt<short >(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template void cmpMatLt<int   >(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template void cmpMatLt<float >(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template void cmpMatLt<double>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
+        GlobPtrSz<uint> src1_ = globPtr((uint*) src1.data, src1.step, src1.rows, vcols);
+        GlobPtrSz<uint> src2_ = globPtr((uint*) src2.data, src2.step, src1.rows, vcols);
+        GlobPtrSz<uint> dst_ = globPtr((uint*) dst.data, dst.step, src1.rows, vcols);
 
-    template void cmpMatLe<uchar >(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template void cmpMatLe<schar >(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template void cmpMatLe<ushort>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template void cmpMatLe<short >(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template void cmpMatLe<int   >(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template void cmpMatLe<float >(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template void cmpMatLe<double>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
+        gridTransformBinary(src1_, src2_, dst_, VCmpLt4(), stream);
+    }
+    void cmpMatLe_v4(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, Stream& stream)
+    {
+        const int vcols = src1.cols >> 2;
+
+        GlobPtrSz<uint> src1_ = globPtr((uint*) src1.data, src1.step, src1.rows, vcols);
+        GlobPtrSz<uint> src2_ = globPtr((uint*) src2.data, src2.step, src1.rows, vcols);
+        GlobPtrSz<uint> dst_ = globPtr((uint*) dst.data, dst.step, src1.rows, vcols);
+
+        gridTransformBinary(src1_, src2_, dst_, VCmpLe4(), stream);
+    }
 }
 
-#endif // CUDA_DISABLER
+void cmpMat(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat&, double, Stream& stream, int cmpop)
+{
+    typedef void (*func_t)(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, Stream& stream);
+    static const func_t funcs[7][4] =
+    {
+        {cmpMat_v1<equal_to, uchar> , cmpMat_v1<not_equal_to, uchar> , cmpMat_v1<less, uchar> , cmpMat_v1<less_equal, uchar> },
+        {cmpMat_v1<equal_to, schar> , cmpMat_v1<not_equal_to, schar> , cmpMat_v1<less, schar> , cmpMat_v1<less_equal, schar> },
+        {cmpMat_v1<equal_to, ushort>, cmpMat_v1<not_equal_to, ushort>, cmpMat_v1<less, ushort>, cmpMat_v1<less_equal, ushort>},
+        {cmpMat_v1<equal_to, short> , cmpMat_v1<not_equal_to, short> , cmpMat_v1<less, short> , cmpMat_v1<less_equal, short> },
+        {cmpMat_v1<equal_to, int>   , cmpMat_v1<not_equal_to, int>   , cmpMat_v1<less, int>   , cmpMat_v1<less_equal, int>   },
+        {cmpMat_v1<equal_to, float> , cmpMat_v1<not_equal_to, float> , cmpMat_v1<less, float> , cmpMat_v1<less_equal, float> },
+        {cmpMat_v1<equal_to, double>, cmpMat_v1<not_equal_to, double>, cmpMat_v1<less, double>, cmpMat_v1<less_equal, double>}
+    };
+
+    typedef void (*func_v4_t)(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, Stream& stream);
+    static const func_v4_t funcs_v4[] =
+    {
+        cmpMatEq_v4, cmpMatNe_v4, cmpMatLt_v4, cmpMatLe_v4
+    };
+
+    const int depth = src1.depth();
+
+    CV_DbgAssert( depth <= CV_64F );
+
+    static const int codes[] =
+    {
+        0, 2, 3, 2, 3, 1
+    };
+    const GpuMat* psrc1[] =
+    {
+        &src1, &src2, &src2, &src1, &src1, &src1
+    };
+    const GpuMat* psrc2[] =
+    {
+        &src2, &src1, &src1, &src2, &src2, &src2
+    };
+
+    const int code = codes[cmpop];
+
+    GpuMat src1_ = psrc1[cmpop]->reshape(1);
+    GpuMat src2_ = psrc2[cmpop]->reshape(1);
+    GpuMat dst_ = dst.reshape(1);
+
+    if (depth == CV_8U && (src1_.cols & 3) == 0)
+    {
+        const intptr_t src1ptr = reinterpret_cast<intptr_t>(src1_.data);
+        const intptr_t src2ptr = reinterpret_cast<intptr_t>(src2_.data);
+        const intptr_t dstptr = reinterpret_cast<intptr_t>(dst_.data);
+
+        const bool isAllAligned = (src1ptr & 31) == 0 && (src2ptr & 31) == 0 && (dstptr & 31) == 0;
+
+        if (isAllAligned)
+        {
+            funcs_v4[code](src1_, src2_, dst_, stream);
+            return;
+        }
+    }
+
+    const func_t func = funcs[depth][code];
+
+    func(src1_, src2_, dst_, stream);
+}
+
+#endif
diff --git a/modules/cudaarithm/src/cuda/cmp_scalar.cu b/modules/cudaarithm/src/cuda/cmp_scalar.cu
index 12bd876304..df57bc0043 100644
--- a/modules/cudaarithm/src/cuda/cmp_scalar.cu
+++ b/modules/cudaarithm/src/cuda/cmp_scalar.cu
@@ -40,24 +40,23 @@
 //
 //M*/
 
-#if !defined CUDA_DISABLER
+#include "opencv2/opencv_modules.hpp"
 
-#include "opencv2/core/cuda/common.hpp"
-#include "opencv2/core/cuda/functional.hpp"
-#include "opencv2/core/cuda/transform.hpp"
-#include "opencv2/core/cuda/saturate_cast.hpp"
-#include "opencv2/core/cuda/simd_functions.hpp"
-#include "opencv2/core/cuda/vec_math.hpp"
+#ifndef HAVE_OPENCV_CUDEV
 
-#include "arithm_func_traits.hpp"
+#error "opencv_cudev is required"
 
-using namespace cv::cuda;
-using namespace cv::cuda::device;
+#else
 
-namespace arithm
+#include "opencv2/cudev.hpp"
+
+using namespace cv::cudev;
+
+void cmpScalar(const GpuMat& src, cv::Scalar val, bool inv, GpuMat& dst, const GpuMat&, double, Stream& stream, int cmpop);
+
+namespace
 {
-    template <class Op, typename T>
-    struct Cmp : binary_function<T, T, uchar>
+    template <class Op, typename T> struct CmpOp : binary_function<T, T, uchar>
     {
         __device__ __forceinline__ uchar operator()(T a, T b) const
         {
@@ -66,219 +65,161 @@ namespace arithm
         }
     };
 
-#define TYPE_VEC(type, cn) typename TypeVec<type, cn>::vec_type
+#define MAKE_VEC(_type, _cn) typename MakeVec<_type, _cn>::type
+
+    template <class Op, typename T, int cn> struct CmpScalarOp;
 
-    template <class Op, typename T, int cn> struct CmpScalar;
     template <class Op, typename T>
-    struct CmpScalar<Op, T, 1> : unary_function<T, uchar>
+    struct CmpScalarOp<Op, T, 1> : unary_function<T, uchar>
     {
         T val;
 
-        __host__ explicit CmpScalar(T val_) : val(val_) {}
-
         __device__ __forceinline__ uchar operator()(T src) const
         {
-            Cmp<Op, T> op;
+            CmpOp<Op, T> op;
             return op(src, val);
         }
     };
+
     template <class Op, typename T>
-    struct CmpScalar<Op, T, 2> : unary_function<TYPE_VEC(T, 2), TYPE_VEC(uchar, 2)>
+    struct CmpScalarOp<Op, T, 2> : unary_function<MAKE_VEC(T, 2), MAKE_VEC(uchar, 2)>
     {
-        TYPE_VEC(T, 2) val;
+        MAKE_VEC(T, 2) val;
 
-        __host__ explicit CmpScalar(TYPE_VEC(T, 2) val_) : val(val_) {}
-
-        __device__ __forceinline__ TYPE_VEC(uchar, 2) operator()(const TYPE_VEC(T, 2) & src) const
+        __device__ __forceinline__ MAKE_VEC(uchar, 2) operator()(const MAKE_VEC(T, 2) & src) const
         {
-            Cmp<Op, T> op;
-            return VecTraits<TYPE_VEC(uchar, 2)>::make(op(src.x, val.x), op(src.y, val.y));
+            CmpOp<Op, T> op;
+            return VecTraits<MAKE_VEC(uchar, 2)>::make(op(src.x, val.x), op(src.y, val.y));
         }
     };
+
     template <class Op, typename T>
-    struct CmpScalar<Op, T, 3> : unary_function<TYPE_VEC(T, 3), TYPE_VEC(uchar, 3)>
+    struct CmpScalarOp<Op, T, 3> : unary_function<MAKE_VEC(T, 3), MAKE_VEC(uchar, 3)>
     {
-        TYPE_VEC(T, 3) val;
+        MAKE_VEC(T, 3) val;
 
-        __host__ explicit CmpScalar(TYPE_VEC(T, 3) val_) : val(val_) {}
-
-        __device__ __forceinline__ TYPE_VEC(uchar, 3) operator()(const TYPE_VEC(T, 3) & src) const
+        __device__ __forceinline__ MAKE_VEC(uchar, 3) operator()(const MAKE_VEC(T, 3) & src) const
         {
-            Cmp<Op, T> op;
-            return VecTraits<TYPE_VEC(uchar, 3)>::make(op(src.x, val.x), op(src.y, val.y), op(src.z, val.z));
+            CmpOp<Op, T> op;
+            return VecTraits<MAKE_VEC(uchar, 3)>::make(op(src.x, val.x), op(src.y, val.y), op(src.z, val.z));
         }
     };
+
     template <class Op, typename T>
-    struct CmpScalar<Op, T, 4> : unary_function<TYPE_VEC(T, 4), TYPE_VEC(uchar, 4)>
+    struct CmpScalarOp<Op, T, 4> : unary_function<MAKE_VEC(T, 4), MAKE_VEC(uchar, 4)>
     {
-        TYPE_VEC(T, 4) val;
+        MAKE_VEC(T, 4) val;
 
-        __host__ explicit CmpScalar(TYPE_VEC(T, 4) val_) : val(val_) {}
-
-        __device__ __forceinline__ TYPE_VEC(uchar, 4) operator()(const TYPE_VEC(T, 4) & src) const
+        __device__ __forceinline__ MAKE_VEC(uchar, 4) operator()(const MAKE_VEC(T, 4) & src) const
         {
-            Cmp<Op, T> op;
-            return VecTraits<TYPE_VEC(uchar, 4)>::make(op(src.x, val.x), op(src.y, val.y), op(src.z, val.z), op(src.w, val.w));
+            CmpOp<Op, T> op;
+            return VecTraits<MAKE_VEC(uchar, 4)>::make(op(src.x, val.x), op(src.y, val.y), op(src.z, val.z), op(src.w, val.w));
         }
     };
 
 #undef TYPE_VEC
-}
 
-namespace cv { namespace cuda { namespace device
-{
-    template <class Op, typename T> struct TransformFunctorTraits< arithm::CmpScalar<Op, T, 1> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(uchar)>
+    template <typename ScalarDepth> struct TransformPolicy : DefaultTransformPolicy
     {
     };
-}}}
+    template <> struct TransformPolicy<double> : DefaultTransformPolicy
+    {
+        enum {
+            shift = 1
+        };
+    };
 
-namespace arithm
-{
     template <template <typename> class Op, typename T, int cn>
-    void cmpScalar(PtrStepSzb src, double val[4], PtrStepSzb dst, cudaStream_t stream)
+    void cmpScalarImpl(const GpuMat& src, cv::Scalar value, GpuMat& dst, Stream& stream)
     {
-        typedef typename TypeVec<T, cn>::vec_type src_t;
-        typedef typename TypeVec<uchar, cn>::vec_type dst_t;
+        typedef typename MakeVec<T, cn>::type src_type;
+        typedef typename MakeVec<uchar, cn>::type dst_type;
 
-        T sval[] = {static_cast<T>(val[0]), static_cast<T>(val[1]), static_cast<T>(val[2]), static_cast<T>(val[3])};
-        src_t val1 = VecTraits<src_t>::make(sval);
+        cv::Scalar_<T> value_ = value;
 
-        CmpScalar<Op<T>, T, cn> op(val1);
-        device::transform((PtrStepSz<src_t>) src, (PtrStepSz<dst_t>) dst, op, WithOutMask(), stream);
+        CmpScalarOp<Op<T>, T, cn> op;
+        op.val = VecTraits<src_type>::make(value_.val);
+
+        gridTransformUnary_< TransformPolicy<T> >(globPtr<src_type>(src), globPtr<dst_type>(dst), op, stream);
     }
-
-    template <typename T> void cmpScalarEq(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream)
-    {
-        typedef void (*func_t)(PtrStepSzb src, double val[4], PtrStepSzb dst, cudaStream_t stream);
-        static const func_t funcs[] =
-        {
-            0,
-            cmpScalar<equal_to, T, 1>,
-            cmpScalar<equal_to, T, 2>,
-            cmpScalar<equal_to, T, 3>,
-            cmpScalar<equal_to, T, 4>
-        };
-
-        funcs[cn](src, val, dst, stream);
-    }
-    template <typename T> void cmpScalarNe(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream)
-    {
-        typedef void (*func_t)(PtrStepSzb src, double val[4], PtrStepSzb dst, cudaStream_t stream);
-        static const func_t funcs[] =
-        {
-            0,
-            cmpScalar<not_equal_to, T, 1>,
-            cmpScalar<not_equal_to, T, 2>,
-            cmpScalar<not_equal_to, T, 3>,
-            cmpScalar<not_equal_to, T, 4>
-        };
-
-        funcs[cn](src, val, dst, stream);
-    }
-    template <typename T> void cmpScalarLt(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream)
-    {
-        typedef void (*func_t)(PtrStepSzb src, double val[4], PtrStepSzb dst, cudaStream_t stream);
-        static const func_t funcs[] =
-        {
-            0,
-            cmpScalar<less, T, 1>,
-            cmpScalar<less, T, 2>,
-            cmpScalar<less, T, 3>,
-            cmpScalar<less, T, 4>
-        };
-
-        funcs[cn](src, val, dst, stream);
-    }
-    template <typename T> void cmpScalarLe(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream)
-    {
-        typedef void (*func_t)(PtrStepSzb src, double val[4], PtrStepSzb dst, cudaStream_t stream);
-        static const func_t funcs[] =
-        {
-            0,
-            cmpScalar<less_equal, T, 1>,
-            cmpScalar<less_equal, T, 2>,
-            cmpScalar<less_equal, T, 3>,
-            cmpScalar<less_equal, T, 4>
-        };
-
-        funcs[cn](src, val, dst, stream);
-    }
-    template <typename T> void cmpScalarGt(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream)
-    {
-        typedef void (*func_t)(PtrStepSzb src, double val[4], PtrStepSzb dst, cudaStream_t stream);
-        static const func_t funcs[] =
-        {
-            0,
-            cmpScalar<greater, T, 1>,
-            cmpScalar<greater, T, 2>,
-            cmpScalar<greater, T, 3>,
-            cmpScalar<greater, T, 4>
-        };
-
-        funcs[cn](src, val, dst, stream);
-    }
-    template <typename T> void cmpScalarGe(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream)
-    {
-        typedef void (*func_t)(PtrStepSzb src, double val[4], PtrStepSzb dst, cudaStream_t stream);
-        static const func_t funcs[] =
-        {
-            0,
-            cmpScalar<greater_equal, T, 1>,
-            cmpScalar<greater_equal, T, 2>,
-            cmpScalar<greater_equal, T, 3>,
-            cmpScalar<greater_equal, T, 4>
-        };
-
-        funcs[cn](src, val, dst, stream);
-    }
-
-    template void cmpScalarEq<uchar >(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-    template void cmpScalarEq<schar >(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-    template void cmpScalarEq<ushort>(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-    template void cmpScalarEq<short >(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-    template void cmpScalarEq<int   >(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-    template void cmpScalarEq<float >(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-    template void cmpScalarEq<double>(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-
-    template void cmpScalarNe<uchar >(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-    template void cmpScalarNe<schar >(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-    template void cmpScalarNe<ushort>(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-    template void cmpScalarNe<short >(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-    template void cmpScalarNe<int   >(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-    template void cmpScalarNe<float >(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-    template void cmpScalarNe<double>(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-
-    template void cmpScalarLt<uchar >(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-    template void cmpScalarLt<schar >(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-    template void cmpScalarLt<ushort>(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-    template void cmpScalarLt<short >(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-    template void cmpScalarLt<int   >(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-    template void cmpScalarLt<float >(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-    template void cmpScalarLt<double>(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-
-    template void cmpScalarLe<uchar >(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-    template void cmpScalarLe<schar >(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-    template void cmpScalarLe<ushort>(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-    template void cmpScalarLe<short >(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-    template void cmpScalarLe<int   >(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-    template void cmpScalarLe<float >(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-    template void cmpScalarLe<double>(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-
-    template void cmpScalarGt<uchar >(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-    template void cmpScalarGt<schar >(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-    template void cmpScalarGt<ushort>(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-    template void cmpScalarGt<short >(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-    template void cmpScalarGt<int   >(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-    template void cmpScalarGt<float >(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-    template void cmpScalarGt<double>(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-
-    template void cmpScalarGe<uchar >(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-    template void cmpScalarGe<schar >(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-    template void cmpScalarGe<ushort>(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-    template void cmpScalarGe<short >(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-    template void cmpScalarGe<int   >(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-    template void cmpScalarGe<float >(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-    template void cmpScalarGe<double>(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
 }
 
-#endif // CUDA_DISABLER
+void cmpScalar(const GpuMat& src, cv::Scalar val, bool inv, GpuMat& dst, const GpuMat&, double, Stream& stream, int cmpop)
+{
+    typedef void (*func_t)(const GpuMat& src, cv::Scalar value, GpuMat& dst, Stream& stream);
+    static const func_t funcs[7][6][4] =
+    {
+        {
+            {cmpScalarImpl<equal_to,      uchar, 1>, cmpScalarImpl<equal_to,      uchar, 2>, cmpScalarImpl<equal_to,      uchar, 3>, cmpScalarImpl<equal_to,      uchar, 4>},
+            {cmpScalarImpl<greater,       uchar, 1>, cmpScalarImpl<greater,       uchar, 2>, cmpScalarImpl<greater,       uchar, 3>, cmpScalarImpl<greater,       uchar, 4>},
+            {cmpScalarImpl<greater_equal, uchar, 1>, cmpScalarImpl<greater_equal, uchar, 2>, cmpScalarImpl<greater_equal, uchar, 3>, cmpScalarImpl<greater_equal, uchar, 4>},
+            {cmpScalarImpl<less,          uchar, 1>, cmpScalarImpl<less,          uchar, 2>, cmpScalarImpl<less,          uchar, 3>, cmpScalarImpl<less,          uchar, 4>},
+            {cmpScalarImpl<less_equal,    uchar, 1>, cmpScalarImpl<less_equal,    uchar, 2>, cmpScalarImpl<less_equal,    uchar, 3>, cmpScalarImpl<less_equal,    uchar, 4>},
+            {cmpScalarImpl<not_equal_to,  uchar, 1>, cmpScalarImpl<not_equal_to,  uchar, 2>, cmpScalarImpl<not_equal_to,  uchar, 3>, cmpScalarImpl<not_equal_to,  uchar, 4>}
+        },
+        {
+            {cmpScalarImpl<equal_to,      schar, 1>, cmpScalarImpl<equal_to,      schar, 2>, cmpScalarImpl<equal_to,      schar, 3>, cmpScalarImpl<equal_to,      schar, 4>},
+            {cmpScalarImpl<greater,       schar, 1>, cmpScalarImpl<greater,       schar, 2>, cmpScalarImpl<greater,       schar, 3>, cmpScalarImpl<greater,       schar, 4>},
+            {cmpScalarImpl<greater_equal, schar, 1>, cmpScalarImpl<greater_equal, schar, 2>, cmpScalarImpl<greater_equal, schar, 3>, cmpScalarImpl<greater_equal, schar, 4>},
+            {cmpScalarImpl<less,          schar, 1>, cmpScalarImpl<less,          schar, 2>, cmpScalarImpl<less,          schar, 3>, cmpScalarImpl<less,          schar, 4>},
+            {cmpScalarImpl<less_equal,    schar, 1>, cmpScalarImpl<less_equal,    schar, 2>, cmpScalarImpl<less_equal,    schar, 3>, cmpScalarImpl<less_equal,    schar, 4>},
+            {cmpScalarImpl<not_equal_to,  schar, 1>, cmpScalarImpl<not_equal_to,  schar, 2>, cmpScalarImpl<not_equal_to,  schar, 3>, cmpScalarImpl<not_equal_to,  schar, 4>}
+        },
+        {
+            {cmpScalarImpl<equal_to,      ushort, 1>, cmpScalarImpl<equal_to,      ushort, 2>, cmpScalarImpl<equal_to,      ushort, 3>, cmpScalarImpl<equal_to,      ushort, 4>},
+            {cmpScalarImpl<greater,       ushort, 1>, cmpScalarImpl<greater,       ushort, 2>, cmpScalarImpl<greater,       ushort, 3>, cmpScalarImpl<greater,       ushort, 4>},
+            {cmpScalarImpl<greater_equal, ushort, 1>, cmpScalarImpl<greater_equal, ushort, 2>, cmpScalarImpl<greater_equal, ushort, 3>, cmpScalarImpl<greater_equal, ushort, 4>},
+            {cmpScalarImpl<less,          ushort, 1>, cmpScalarImpl<less,          ushort, 2>, cmpScalarImpl<less,          ushort, 3>, cmpScalarImpl<less,          ushort, 4>},
+            {cmpScalarImpl<less_equal,    ushort, 1>, cmpScalarImpl<less_equal,    ushort, 2>, cmpScalarImpl<less_equal,    ushort, 3>, cmpScalarImpl<less_equal,    ushort, 4>},
+            {cmpScalarImpl<not_equal_to,  ushort, 1>, cmpScalarImpl<not_equal_to,  ushort, 2>, cmpScalarImpl<not_equal_to,  ushort, 3>, cmpScalarImpl<not_equal_to,  ushort, 4>}
+        },
+        {
+            {cmpScalarImpl<equal_to,      short, 1>, cmpScalarImpl<equal_to,      short, 2>, cmpScalarImpl<equal_to,      short, 3>, cmpScalarImpl<equal_to,      short, 4>},
+            {cmpScalarImpl<greater,       short, 1>, cmpScalarImpl<greater,       short, 2>, cmpScalarImpl<greater,       short, 3>, cmpScalarImpl<greater,       short, 4>},
+            {cmpScalarImpl<greater_equal, short, 1>, cmpScalarImpl<greater_equal, short, 2>, cmpScalarImpl<greater_equal, short, 3>, cmpScalarImpl<greater_equal, short, 4>},
+            {cmpScalarImpl<less,          short, 1>, cmpScalarImpl<less,          short, 2>, cmpScalarImpl<less,          short, 3>, cmpScalarImpl<less,          short, 4>},
+            {cmpScalarImpl<less_equal,    short, 1>, cmpScalarImpl<less_equal,    short, 2>, cmpScalarImpl<less_equal,    short, 3>, cmpScalarImpl<less_equal,    short, 4>},
+            {cmpScalarImpl<not_equal_to,  short, 1>, cmpScalarImpl<not_equal_to,  short, 2>, cmpScalarImpl<not_equal_to,  short, 3>, cmpScalarImpl<not_equal_to,  short, 4>}
+        },
+        {
+            {cmpScalarImpl<equal_to,      int, 1>, cmpScalarImpl<equal_to,      int, 2>, cmpScalarImpl<equal_to,      int, 3>, cmpScalarImpl<equal_to,      int, 4>},
+            {cmpScalarImpl<greater,       int, 1>, cmpScalarImpl<greater,       int, 2>, cmpScalarImpl<greater,       int, 3>, cmpScalarImpl<greater,       int, 4>},
+            {cmpScalarImpl<greater_equal, int, 1>, cmpScalarImpl<greater_equal, int, 2>, cmpScalarImpl<greater_equal, int, 3>, cmpScalarImpl<greater_equal, int, 4>},
+            {cmpScalarImpl<less,          int, 1>, cmpScalarImpl<less,          int, 2>, cmpScalarImpl<less,          int, 3>, cmpScalarImpl<less,          int, 4>},
+            {cmpScalarImpl<less_equal,    int, 1>, cmpScalarImpl<less_equal,    int, 2>, cmpScalarImpl<less_equal,    int, 3>, cmpScalarImpl<less_equal,    int, 4>},
+            {cmpScalarImpl<not_equal_to,  int, 1>, cmpScalarImpl<not_equal_to,  int, 2>, cmpScalarImpl<not_equal_to,  int, 3>, cmpScalarImpl<not_equal_to,  int, 4>}
+        },
+        {
+            {cmpScalarImpl<equal_to,      float, 1>, cmpScalarImpl<equal_to,      float, 2>, cmpScalarImpl<equal_to,      float, 3>, cmpScalarImpl<equal_to,      float, 4>},
+            {cmpScalarImpl<greater,       float, 1>, cmpScalarImpl<greater,       float, 2>, cmpScalarImpl<greater,       float, 3>, cmpScalarImpl<greater,       float, 4>},
+            {cmpScalarImpl<greater_equal, float, 1>, cmpScalarImpl<greater_equal, float, 2>, cmpScalarImpl<greater_equal, float, 3>, cmpScalarImpl<greater_equal, float, 4>},
+            {cmpScalarImpl<less,          float, 1>, cmpScalarImpl<less,          float, 2>, cmpScalarImpl<less,          float, 3>, cmpScalarImpl<less,          float, 4>},
+            {cmpScalarImpl<less_equal,    float, 1>, cmpScalarImpl<less_equal,    float, 2>, cmpScalarImpl<less_equal,    float, 3>, cmpScalarImpl<less_equal,    float, 4>},
+            {cmpScalarImpl<not_equal_to,  float, 1>, cmpScalarImpl<not_equal_to,  float, 2>, cmpScalarImpl<not_equal_to,  float, 3>, cmpScalarImpl<not_equal_to,  float, 4>}
+        },
+        {
+            {cmpScalarImpl<equal_to,      double, 1>, cmpScalarImpl<equal_to,      double, 2>, cmpScalarImpl<equal_to,      double, 3>, cmpScalarImpl<equal_to,      double, 4>},
+            {cmpScalarImpl<greater,       double, 1>, cmpScalarImpl<greater,       double, 2>, cmpScalarImpl<greater,       double, 3>, cmpScalarImpl<greater,       double, 4>},
+            {cmpScalarImpl<greater_equal, double, 1>, cmpScalarImpl<greater_equal, double, 2>, cmpScalarImpl<greater_equal, double, 3>, cmpScalarImpl<greater_equal, double, 4>},
+            {cmpScalarImpl<less,          double, 1>, cmpScalarImpl<less,          double, 2>, cmpScalarImpl<less,          double, 3>, cmpScalarImpl<less,          double, 4>},
+            {cmpScalarImpl<less_equal,    double, 1>, cmpScalarImpl<less_equal,    double, 2>, cmpScalarImpl<less_equal,    double, 3>, cmpScalarImpl<less_equal,    double, 4>},
+            {cmpScalarImpl<not_equal_to,  double, 1>, cmpScalarImpl<not_equal_to,  double, 2>, cmpScalarImpl<not_equal_to,  double, 3>, cmpScalarImpl<not_equal_to,  double, 4>}
+        }
+    };
+
+    if (inv)
+    {
+        // src1 is a scalar; swap it with src2
+        cmpop = cmpop == cv::CMP_LT ? cv::CMP_GT : cmpop == cv::CMP_LE ? cv::CMP_GE :
+            cmpop == cv::CMP_GE ? cv::CMP_LE : cmpop == cv::CMP_GT ? cv::CMP_LT : cmpop;
+    }
+
+    const int depth = src.depth();
+    const int cn = src.channels();
+
+    CV_DbgAssert( depth <= CV_64F && cn <= 4 );
+
+    funcs[depth][cmpop][cn - 1](src, val, dst, stream);
+}
+
+#endif
diff --git a/modules/cudaarithm/src/cuda/copy_make_border.cu b/modules/cudaarithm/src/cuda/copy_make_border.cu
index 2ec53ce08d..f7dd91f987 100644
--- a/modules/cudaarithm/src/cuda/copy_make_border.cu
+++ b/modules/cudaarithm/src/cuda/copy_make_border.cu
@@ -40,92 +40,116 @@
 //
 //M*/
 
-#if !defined CUDA_DISABLER
+#include "opencv2/opencv_modules.hpp"
 
-#include "opencv2/core/cuda/common.hpp"
-#include "opencv2/core/cuda/border_interpolate.hpp"
+#ifndef HAVE_OPENCV_CUDEV
 
-namespace cv { namespace cuda { namespace device
+#error "opencv_cudev is required"
+
+#else
+
+#include "opencv2/cudaarithm.hpp"
+#include "opencv2/cudev.hpp"
+
+using namespace cv::cudev;
+
+namespace
 {
-    namespace imgproc
+    struct ShiftMap
     {
-        template <typename Ptr2D, typename T> __global__ void copyMakeBorder(const Ptr2D src, PtrStepSz<T> dst, int top, int left)
-        {
-            const int x = blockDim.x * blockIdx.x + threadIdx.x;
-            const int y = blockDim.y * blockIdx.y + threadIdx.y;
+        typedef int2 value_type;
+        typedef int index_type;
 
-            if (x < dst.cols && y < dst.rows)
-                dst.ptr(y)[x] = src(y - top, x - left);
+        int top;
+        int left;
+
+        __device__ __forceinline__ int2 operator ()(int y, int x) const
+        {
+            return make_int2(x - left, y - top);
         }
+    };
 
-        template <template <typename> class B, typename T> struct CopyMakeBorderDispatcher
+    struct ShiftMapSz : ShiftMap
+    {
+        int rows, cols;
+    };
+}
+
+namespace cv { namespace cudev {
+
+template <> struct PtrTraits<ShiftMapSz> : PtrTraitsBase<ShiftMapSz, ShiftMap>
+{
+};
+
+}}
+
+namespace
+{
+    template <typename T, int cn>
+    void copyMakeBorderImpl(const GpuMat& src, GpuMat& dst, int top, int left, int borderMode, cv::Scalar borderValue, Stream& stream)
+    {
+        typedef typename MakeVec<T, cn>::type src_type;
+
+        cv::Scalar_<T> borderValue_ = borderValue;
+        const src_type brdVal = VecTraits<src_type>::make(borderValue_.val);
+
+        ShiftMapSz map;
+        map.top = top;
+        map.left = left;
+        map.rows = dst.rows;
+        map.cols = dst.cols;
+
+        switch (borderMode)
         {
-            static void call(const PtrStepSz<T>& src, const PtrStepSz<T>& dst, int top, int left,
-                const typename VecTraits<T>::elem_type* borderValue, cudaStream_t stream)
-            {
-                dim3 block(32, 8);
-                dim3 grid(divUp(dst.cols, block.x), divUp(dst.rows, block.y));
-
-                B<T> brd(src.rows, src.cols, VecTraits<T>::make(borderValue));
-                BorderReader< PtrStep<T>, B<T> > brdSrc(src, brd);
-
-                copyMakeBorder<<<grid, block, 0, stream>>>(brdSrc, dst, top, left);
-                cudaSafeCall( cudaGetLastError() );
-
-                if (stream == 0)
-                    cudaSafeCall( cudaDeviceSynchronize() );
-            }
+        case cv::BORDER_CONSTANT:
+            gridCopy(remapPtr(brdConstant(globPtr<src_type>(src), brdVal), map), globPtr<src_type>(dst), stream);
+            break;
+        case cv::BORDER_REPLICATE:
+            gridCopy(remapPtr(brdReplicate(globPtr<src_type>(src)), map), globPtr<src_type>(dst), stream);
+            break;
+        case cv::BORDER_REFLECT:
+            gridCopy(remapPtr(brdReflect(globPtr<src_type>(src)), map), globPtr<src_type>(dst), stream);
+            break;
+        case cv::BORDER_WRAP:
+            gridCopy(remapPtr(brdWrap(globPtr<src_type>(src)), map), globPtr<src_type>(dst), stream);
+            break;
+        case cv::BORDER_REFLECT_101:
+            gridCopy(remapPtr(brdReflect101(globPtr<src_type>(src)), map), globPtr<src_type>(dst), stream);
+            break;
         };
+    }
+}
 
-        template <typename T, int cn> void copyMakeBorder_gpu(const PtrStepSzb& src, const PtrStepSzb& dst, int top, int left, int borderMode,
-            const T* borderValue, cudaStream_t stream)
-        {
-            typedef typename TypeVec<T, cn>::vec_type vec_type;
+void cv::cuda::copyMakeBorder(InputArray _src, OutputArray _dst, int top, int bottom, int left, int right, int borderType, Scalar value, Stream& stream)
+{
+    typedef void (*func_t)(const GpuMat& src, GpuMat& dst, int top, int left, int borderMode, cv::Scalar borderValue, Stream& stream);
+    static const func_t funcs[6][4] =
+    {
+        {    copyMakeBorderImpl<uchar , 1>  ,     copyMakeBorderImpl<uchar , 2>  ,     copyMakeBorderImpl<uchar , 3>  ,     copyMakeBorderImpl<uchar , 4>  },
+        {0 /*copyMakeBorderImpl<schar , 1>*/, 0 /*copyMakeBorderImpl<schar , 2>*/, 0 /*copyMakeBorderImpl<schar , 3>*/, 0 /*copyMakeBorderImpl<schar , 4>*/},
+        {    copyMakeBorderImpl<ushort, 1>  , 0 /*copyMakeBorderImpl<ushort, 2>*/,     copyMakeBorderImpl<ushort, 3>  ,     copyMakeBorderImpl<ushort, 4>  },
+        {    copyMakeBorderImpl<short , 1>  , 0 /*copyMakeBorderImpl<short , 2>*/,     copyMakeBorderImpl<short , 3>  ,     copyMakeBorderImpl<short , 4>  },
+        {0 /*copyMakeBorderImpl<int   , 1>*/, 0 /*copyMakeBorderImpl<int   , 2>*/, 0 /*copyMakeBorderImpl<int   , 3>*/, 0 /*copyMakeBorderImpl<int   , 4>*/},
+        {    copyMakeBorderImpl<float , 1>  , 0 /*copyMakeBorderImpl<float , 2>*/,     copyMakeBorderImpl<float , 3>  ,     copyMakeBorderImpl<float  ,4>  }
+    };
 
-            typedef void (*caller_t)(const PtrStepSz<vec_type>& src, const PtrStepSz<vec_type>& dst, int top, int left, const T* borderValue, cudaStream_t stream);
+    GpuMat src = _src.getGpuMat();
 
-            static const caller_t callers[5] =
-            {
-                CopyMakeBorderDispatcher<BrdConstant, vec_type>::call,
-                CopyMakeBorderDispatcher<BrdReplicate, vec_type>::call,
-                CopyMakeBorderDispatcher<BrdReflect, vec_type>::call,
-                CopyMakeBorderDispatcher<BrdWrap, vec_type>::call,
-                CopyMakeBorderDispatcher<BrdReflect101, vec_type>::call
-            };
+    const int depth = src.depth();
+    const int cn = src.channels();
 
-            callers[borderMode](PtrStepSz<vec_type>(src), PtrStepSz<vec_type>(dst), top, left, borderValue, stream);
-        }
+    CV_Assert( depth <= CV_32F && cn <= 4 );
+    CV_Assert( borderType == BORDER_REFLECT_101 || borderType == BORDER_REPLICATE || borderType == BORDER_CONSTANT || borderType == BORDER_REFLECT || borderType == BORDER_WRAP );
 
-        template void copyMakeBorder_gpu<uchar, 1>(const PtrStepSzb& src, const PtrStepSzb& dst, int top, int left, int borderMode, const uchar* borderValue, cudaStream_t stream);
-        template void copyMakeBorder_gpu<uchar, 2>(const PtrStepSzb& src, const PtrStepSzb& dst, int top, int left, int borderMode, const uchar* borderValue, cudaStream_t stream);
-        template void copyMakeBorder_gpu<uchar, 3>(const PtrStepSzb& src, const PtrStepSzb& dst, int top, int left, int borderMode, const uchar* borderValue, cudaStream_t stream);
-        template void copyMakeBorder_gpu<uchar, 4>(const PtrStepSzb& src, const PtrStepSzb& dst, int top, int left, int borderMode, const uchar* borderValue, cudaStream_t stream);
+    _dst.create(src.rows + top + bottom, src.cols + left + right, src.type());
+    GpuMat dst = _dst.getGpuMat();
 
-        //template void copyMakeBorder_gpu<schar, 1>(const PtrStepSzb& src, const PtrStepSzb& dst, int top, int left, int borderMode, const schar* borderValue, cudaStream_t stream);
-        //template void copyMakeBorder_gpu<schar, 2>(const PtrStepSzb& src, const PtrStepSzb& dst, int top, int left, int borderMode, const schar* borderValue, cudaStream_t stream);
-        //template void copyMakeBorder_gpu<schar, 3>(const PtrStepSzb& src, const PtrStepSzb& dst, int top, int left, int borderMode, const schar* borderValue, cudaStream_t stream);
-        //template void copyMakeBorder_gpu<schar, 4>(const PtrStepSzb& src, const PtrStepSzb& dst, int top, int left, int borderMode, const schar* borderValue, cudaStream_t stream);
+    const func_t func = funcs[depth][cn - 1];
 
-        template void copyMakeBorder_gpu<ushort, 1>(const PtrStepSzb& src, const PtrStepSzb& dst, int top, int left, int borderMode, const ushort* borderValue, cudaStream_t stream);
-        //template void copyMakeBorder_gpu<ushort, 2>(const PtrStepSzb& src, const PtrStepSzb& dst, int top, int left, int borderMode, const ushort* borderValue, cudaStream_t stream);
-        template void copyMakeBorder_gpu<ushort, 3>(const PtrStepSzb& src, const PtrStepSzb& dst, int top, int left, int borderMode, const ushort* borderValue, cudaStream_t stream);
-        template void copyMakeBorder_gpu<ushort, 4>(const PtrStepSzb& src, const PtrStepSzb& dst, int top, int left, int borderMode, const ushort* borderValue, cudaStream_t stream);
+    if (!func)
+        CV_Error(cv::Error::StsUnsupportedFormat, "Unsupported combination of source and destination types");
 
-        template void copyMakeBorder_gpu<short, 1>(const PtrStepSzb& src, const PtrStepSzb& dst, int top, int left, int borderMode, const short* borderValue, cudaStream_t stream);
-        //template void copyMakeBorder_gpu<short, 2>(const PtrStepSzb& src, const PtrStepSzb& dst, int top, int left, int borderMode, const short* borderValue, cudaStream_t stream);
-        template void copyMakeBorder_gpu<short, 3>(const PtrStepSzb& src, const PtrStepSzb& dst, int top, int left, int borderMode, const short* borderValue, cudaStream_t stream);
-        template void copyMakeBorder_gpu<short, 4>(const PtrStepSzb& src, const PtrStepSzb& dst, int top, int left, int borderMode, const short* borderValue, cudaStream_t stream);
+    func(src, dst, top, left, borderType, value, stream);
+}
 
-        //template void copyMakeBorder_gpu<int, 1>(const PtrStepSzb& src, const PtrStepSzb& dst, int top, int left, int borderMode, const int* borderValue, cudaStream_t stream);
-        //template void copyMakeBorder_gpu<int, 2>(const PtrStepSzb& src, const PtrStepSzb& dst, int top, int left, int borderMode, const int* borderValue, cudaStream_t stream);
-        //template void copyMakeBorder_gpu<int, 3>(const PtrStepSzb& src, const PtrStepSzb& dst, int top, int left, int borderMode, const int* borderValue, cudaStream_t stream);
-        //template void copyMakeBorder_gpu<int, 4>(const PtrStepSzb& src, const PtrStepSzb& dst, int top, int left, int borderMode, const int* borderValue, cudaStream_t stream);
-
-        template void copyMakeBorder_gpu<float, 1>(const PtrStepSzb& src, const PtrStepSzb& dst, int top, int left, int borderMode, const float* borderValue, cudaStream_t stream);
-        //template void copyMakeBorder_gpu<float, 2>(const PtrStepSzb& src, const PtrStepSzb& dst, int top, int left, int borderMode, const float* borderValue, cudaStream_t stream);
-        template void copyMakeBorder_gpu<float, 3>(const PtrStepSzb& src, const PtrStepSzb& dst, int top, int left, int borderMode, const float* borderValue, cudaStream_t stream);
-        template void copyMakeBorder_gpu<float, 4>(const PtrStepSzb& src, const PtrStepSzb& dst, int top, int left, int borderMode, const float* borderValue, cudaStream_t stream);
-    } // namespace imgproc
-}}} // namespace cv { namespace cuda { namespace cudev
-
-#endif /* CUDA_DISABLER */
+#endif
diff --git a/modules/cudaarithm/src/cuda/countnonzero.cu b/modules/cudaarithm/src/cuda/countnonzero.cu
index 6ab35d9fe9..5de2609093 100644
--- a/modules/cudaarithm/src/cuda/countnonzero.cu
+++ b/modules/cudaarithm/src/cuda/countnonzero.cu
@@ -40,137 +40,57 @@
 //
 //M*/
 
-#if !defined CUDA_DISABLER
+#include "opencv2/opencv_modules.hpp"
 
-#include "opencv2/core/cuda/common.hpp"
-#include "opencv2/core/cuda/vec_traits.hpp"
-#include "opencv2/core/cuda/vec_math.hpp"
-#include "opencv2/core/cuda/functional.hpp"
-#include "opencv2/core/cuda/reduce.hpp"
-#include "opencv2/core/cuda/emulation.hpp"
+#ifndef HAVE_OPENCV_CUDEV
 
-using namespace cv::cuda;
-using namespace cv::cuda::device;
+#error "opencv_cudev is required"
 
-namespace countNonZero
+#else
+
+#include "opencv2/cudaarithm.hpp"
+#include "opencv2/cudev.hpp"
+
+using namespace cv::cudev;
+
+namespace
 {
-    __device__ unsigned int blocks_finished = 0;
-
-    template <int BLOCK_SIZE, typename T>
-    __global__ void kernel(const PtrStepSz<T> src, unsigned int* count, const int twidth, const int theight)
-    {
-        __shared__ unsigned int scount[BLOCK_SIZE];
-
-        const int x0 = blockIdx.x * blockDim.x * twidth + threadIdx.x;
-        const int y0 = blockIdx.y * blockDim.y * theight + threadIdx.y;
-
-        const int tid = threadIdx.y * blockDim.x + threadIdx.x;
-
-        unsigned int mycount = 0;
-
-        for (int i = 0, y = y0; i < theight && y < src.rows; ++i, y += blockDim.y)
-        {
-            const T* ptr = src.ptr(y);
-
-            for (int j = 0, x = x0; j < twidth && x < src.cols; ++j, x += blockDim.x)
-            {
-                const T srcVal = ptr[x];
-
-                mycount += (srcVal != 0);
-            }
-        }
-
-        device::reduce<BLOCK_SIZE>(scount, mycount, tid, plus<unsigned int>());
-
-    #if __CUDA_ARCH__ >= 200
-        if (tid == 0)
-            ::atomicAdd(count, mycount);
-    #else
-        __shared__ bool is_last;
-        const int bid = blockIdx.y * gridDim.x + blockIdx.x;
-
-        if (tid == 0)
-        {
-            count[bid] = mycount;
-
-            __threadfence();
-
-            unsigned int ticket = ::atomicInc(&blocks_finished, gridDim.x * gridDim.y);
-            is_last = (ticket == gridDim.x * gridDim.y - 1);
-        }
-
-        __syncthreads();
-
-        if (is_last)
-        {
-            mycount = tid < gridDim.x * gridDim.y ? count[tid] : 0;
-
-            device::reduce<BLOCK_SIZE>(scount, mycount, tid, plus<unsigned int>());
-
-            if (tid == 0)
-            {
-                count[0] = mycount;
-
-                blocks_finished = 0;
-            }
-        }
-    #endif
-    }
-
-    const int threads_x = 32;
-    const int threads_y = 8;
-
-    void getLaunchCfg(int cols, int rows, dim3& block, dim3& grid)
-    {
-        block = dim3(threads_x, threads_y);
-
-        grid = dim3(divUp(cols, block.x * block.y),
-                    divUp(rows, block.y * block.x));
-
-        grid.x = ::min(grid.x, block.x);
-        grid.y = ::min(grid.y, block.y);
-    }
-
-    void getBufSize(int cols, int rows, int& bufcols, int& bufrows)
-    {
-        dim3 block, grid;
-        getLaunchCfg(cols, rows, block, grid);
-
-        bufcols = grid.x * grid.y * sizeof(int);
-        bufrows = 1;
-    }
-
     template <typename T>
-    int run(const PtrStepSzb src, PtrStep<unsigned int> buf)
+    int countNonZeroImpl(const GpuMat& _src, GpuMat& _buf)
     {
-        dim3 block, grid;
-        getLaunchCfg(src.cols, src.rows, block, grid);
+        const GpuMat_<T>& src = (const GpuMat_<T>&) _src;
+        GpuMat_<int>& buf = (GpuMat_<int>&) _buf;
 
-        const int twidth = divUp(divUp(src.cols, grid.x), block.x);
-        const int theight = divUp(divUp(src.rows, grid.y), block.y);
+        gridCountNonZero(src, buf);
 
-        unsigned int* count_buf = buf.ptr(0);
+        int data;
+        buf.download(cv::Mat(1, 1, buf.type(), &data));
 
-        cudaSafeCall( cudaMemset(count_buf, 0, sizeof(unsigned int)) );
-
-        kernel<threads_x * threads_y><<<grid, block>>>((PtrStepSz<T>) src, count_buf, twidth, theight);
-        cudaSafeCall( cudaGetLastError() );
-
-        cudaSafeCall( cudaDeviceSynchronize() );
-
-        unsigned int count;
-        cudaSafeCall(cudaMemcpy(&count, count_buf, sizeof(unsigned int), cudaMemcpyDeviceToHost));
-
-        return count;
+        return data;
     }
-
-    template int run<uchar >(const PtrStepSzb src, PtrStep<unsigned int> buf);
-    template int run<schar >(const PtrStepSzb src, PtrStep<unsigned int> buf);
-    template int run<ushort>(const PtrStepSzb src, PtrStep<unsigned int> buf);
-    template int run<short >(const PtrStepSzb src, PtrStep<unsigned int> buf);
-    template int run<int   >(const PtrStepSzb src, PtrStep<unsigned int> buf);
-    template int run<float >(const PtrStepSzb src, PtrStep<unsigned int> buf);
-    template int run<double>(const PtrStepSzb src, PtrStep<unsigned int> buf);
 }
 
-#endif // CUDA_DISABLER
+int cv::cuda::countNonZero(InputArray _src, GpuMat& buf)
+{
+    typedef int (*func_t)(const GpuMat& _src, GpuMat& _buf);
+    static const func_t funcs[] =
+    {
+        countNonZeroImpl<uchar>,
+        countNonZeroImpl<schar>,
+        countNonZeroImpl<ushort>,
+        countNonZeroImpl<short>,
+        countNonZeroImpl<int>,
+        countNonZeroImpl<float>,
+        countNonZeroImpl<double>
+    };
+
+    GpuMat src = _src.getGpuMat();
+
+    CV_Assert( src.channels() == 1 );
+
+    const func_t func = funcs[src.depth()];
+
+    return func(src, buf);
+}
+
+#endif
diff --git a/modules/cudaarithm/src/cuda/div_mat.cu b/modules/cudaarithm/src/cuda/div_mat.cu
index 4a62a44ae3..2a2fb9bf51 100644
--- a/modules/cudaarithm/src/cuda/div_mat.cu
+++ b/modules/cudaarithm/src/cuda/div_mat.cu
@@ -40,191 +40,203 @@
 //
 //M*/
 
-#if !defined CUDA_DISABLER
+#include "opencv2/opencv_modules.hpp"
 
-#include "opencv2/core/cuda/common.hpp"
-#include "opencv2/core/cuda/functional.hpp"
-#include "opencv2/core/cuda/transform.hpp"
-#include "opencv2/core/cuda/saturate_cast.hpp"
-#include "opencv2/core/cuda/simd_functions.hpp"
+#ifndef HAVE_OPENCV_CUDEV
 
-#include "arithm_func_traits.hpp"
+#error "opencv_cudev is required"
 
-using namespace cv::cuda;
-using namespace cv::cuda::device;
+#else
 
-namespace arithm
+#include "opencv2/cudev.hpp"
+
+using namespace cv::cudev;
+
+void divMat(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat&, double scale, Stream& stream, int);
+void divMat_8uc4_32f(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, Stream& stream);
+void divMat_16sc4_32f(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, Stream& stream);
+
+namespace
 {
-    struct Div_8uc4_32f : binary_function<uint, float, uint>
-    {
-        __device__ __forceinline__ uint operator ()(uint a, float b) const
-        {
-            uint res = 0;
-
-            if (b != 0)
-            {
-                b = 1.0f / b;
-                res |= (saturate_cast<uchar>((0xffu & (a      )) * b)      );
-                res |= (saturate_cast<uchar>((0xffu & (a >>  8)) * b) <<  8);
-                res |= (saturate_cast<uchar>((0xffu & (a >> 16)) * b) << 16);
-                res |= (saturate_cast<uchar>((0xffu & (a >> 24)) * b) << 24);
-            }
-
-            return res;
-        }
-    };
-
-    struct Div_16sc4_32f : binary_function<short4, float, short4>
-    {
-        __device__ __forceinline__ short4 operator ()(short4 a, float b) const
-        {
-            return b != 0 ? make_short4(saturate_cast<short>(a.x / b), saturate_cast<short>(a.y / b),
-                                        saturate_cast<short>(a.z / b), saturate_cast<short>(a.w / b))
-                          : make_short4(0,0,0,0);
-        }
-    };
-
-    template <typename T, typename D> struct Div : binary_function<T, T, D>
+    template <typename T, typename D> struct DivOp : binary_function<T, T, D>
     {
         __device__ __forceinline__ D operator ()(T a, T b) const
         {
             return b != 0 ? saturate_cast<D>(a / b) : 0;
         }
-
-        __host__ __device__ __forceinline__ Div() {}
-        __host__ __device__ __forceinline__ Div(const Div&) {}
     };
-    template <typename T> struct Div<T, float> : binary_function<T, T, float>
+    template <typename T> struct DivOp<T, float> : binary_function<T, T, float>
     {
         __device__ __forceinline__ float operator ()(T a, T b) const
         {
-            return b != 0 ? static_cast<float>(a) / b : 0;
+            return b != 0 ? static_cast<float>(a) / b : 0.0f;
         }
-
-        __host__ __device__ __forceinline__ Div() {}
-        __host__ __device__ __forceinline__ Div(const Div&) {}
     };
-    template <typename T> struct Div<T, double> : binary_function<T, T, double>
+    template <typename T> struct DivOp<T, double> : binary_function<T, T, double>
     {
         __device__ __forceinline__ double operator ()(T a, T b) const
         {
-            return b != 0 ? static_cast<double>(a) / b : 0;
+            return b != 0 ? static_cast<double>(a) / b : 0.0;
         }
-
-        __host__ __device__ __forceinline__ Div() {}
-        __host__ __device__ __forceinline__ Div(const Div&) {}
     };
 
-    template <typename T, typename S, typename D> struct DivScale : binary_function<T, T, D>
+    template <typename T, typename S, typename D> struct DivScaleOp : binary_function<T, T, D>
     {
         S scale;
 
-        __host__ explicit DivScale(S scale_) : scale(scale_) {}
-
         __device__ __forceinline__ D operator ()(T a, T b) const
         {
             return b != 0 ? saturate_cast<D>(scale * a / b) : 0;
         }
     };
-}
 
-namespace cv { namespace cuda { namespace device
-{
-    template <> struct TransformFunctorTraits<arithm::Div_8uc4_32f> : arithm::ArithmFuncTraits<sizeof(uint), sizeof(uint)>
+    template <typename ScalarDepth> struct TransformPolicy : DefaultTransformPolicy
     {
     };
-
-    template <typename T, typename D> struct TransformFunctorTraits< arithm::Div<T, D> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(D)>
+    template <> struct TransformPolicy<double> : DefaultTransformPolicy
     {
+        enum {
+            shift = 1
+        };
     };
 
-    template <typename T, typename S, typename D> struct TransformFunctorTraits< arithm::DivScale<T, S, D> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(D)>
-    {
-    };
-}}}
-
-namespace arithm
-{
-    void divMat_8uc4_32f(PtrStepSz<uint> src1, PtrStepSzf src2, PtrStepSz<uint> dst, cudaStream_t stream)
-    {
-        device::transform(src1, src2, dst, Div_8uc4_32f(), WithOutMask(), stream);
-    }
-
-    void divMat_16sc4_32f(PtrStepSz<short4> src1, PtrStepSzf src2, PtrStepSz<short4> dst, cudaStream_t stream)
-    {
-        device::transform(src1, src2, dst, Div_16sc4_32f(), WithOutMask(), stream);
-    }
-
     template <typename T, typename S, typename D>
-    void divMat(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream)
+    void divMatImpl(const GpuMat& src1, const GpuMat& src2, const GpuMat& dst, double scale, Stream& stream)
     {
         if (scale == 1)
         {
-            Div<T, D> op;
-            device::transform((PtrStepSz<T>) src1, (PtrStepSz<T>) src2, (PtrStepSz<D>) dst, op, WithOutMask(), stream);
+            DivOp<T, D> op;
+            gridTransformBinary_< TransformPolicy<S> >(globPtr<T>(src1), globPtr<T>(src2), globPtr<D>(dst), op, stream);
         }
         else
         {
-            DivScale<T, S, D> op(static_cast<S>(scale));
-            device::transform((PtrStepSz<T>) src1, (PtrStepSz<T>) src2, (PtrStepSz<D>) dst, op, WithOutMask(), stream);
+            DivScaleOp<T, S, D> op;
+            op.scale = static_cast<S>(scale);
+            gridTransformBinary_< TransformPolicy<S> >(globPtr<T>(src1), globPtr<T>(src2), globPtr<D>(dst), op, stream);
         }
     }
-
-    template void divMat<uchar, float, uchar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void divMat<uchar, float, schar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void divMat<uchar, float, ushort>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void divMat<uchar, float, short>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void divMat<uchar, float, int>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void divMat<uchar, float, float>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void divMat<uchar, double, double>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-
-    template void divMat<schar, float, uchar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void divMat<schar, float, schar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void divMat<schar, float, ushort>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void divMat<schar, float, short>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void divMat<schar, float, int>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void divMat<schar, float, float>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void divMat<schar, double, double>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-
-    //template void divMat<ushort, float, uchar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    //template void divMat<ushort, float, schar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void divMat<ushort, float, ushort>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void divMat<ushort, float, short>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void divMat<ushort, float, int>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void divMat<ushort, float, float>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void divMat<ushort, double, double>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-
-    //template void divMat<short, float, uchar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    //template void divMat<short, float, schar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void divMat<short, float, ushort>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void divMat<short, float, short>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void divMat<short, float, int>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void divMat<short, float, float>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void divMat<short, double, double>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-
-    //template void divMat<int, float, uchar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    //template void divMat<int, float, schar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    //template void divMat<int, float, ushort>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    //template void divMat<int, float, short>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void divMat<int, float, int>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void divMat<int, float, float>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void divMat<int, double, double>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-
-    //template void divMat<float, float, uchar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    //template void divMat<float, float, schar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    //template void divMat<float, float, ushort>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    //template void divMat<float, float, short>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    //template void divMat<float, float, int>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void divMat<float, float, float>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void divMat<float, double, double>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-
-    //template void divMat<double, double, uchar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    //template void divMat<double, double, schar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    //template void divMat<double, double, ushort>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    //template void divMat<double, double, short>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    //template void divMat<double, double, int>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    //template void divMat<double, double, float>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void divMat<double, double, double>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
 }
 
-#endif // CUDA_DISABLER
+void divMat(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat&, double scale, Stream& stream, int)
+{
+    typedef void (*func_t)(const GpuMat& src1, const GpuMat& src2, const GpuMat& dst, double scale, Stream& stream);
+    static const func_t funcs[7][7] =
+    {
+        {
+            divMatImpl<uchar, float, uchar>,
+            divMatImpl<uchar, float, schar>,
+            divMatImpl<uchar, float, ushort>,
+            divMatImpl<uchar, float, short>,
+            divMatImpl<uchar, float, int>,
+            divMatImpl<uchar, float, float>,
+            divMatImpl<uchar, double, double>
+        },
+        {
+            divMatImpl<schar, float, uchar>,
+            divMatImpl<schar, float, schar>,
+            divMatImpl<schar, float, ushort>,
+            divMatImpl<schar, float, short>,
+            divMatImpl<schar, float, int>,
+            divMatImpl<schar, float, float>,
+            divMatImpl<schar, double, double>
+        },
+        {
+            0 /*divMatImpl<ushort, float, uchar>*/,
+            0 /*divMatImpl<ushort, float, schar>*/,
+            divMatImpl<ushort, float, ushort>,
+            divMatImpl<ushort, float, short>,
+            divMatImpl<ushort, float, int>,
+            divMatImpl<ushort, float, float>,
+            divMatImpl<ushort, double, double>
+        },
+        {
+            0 /*divMatImpl<short, float, uchar>*/,
+            0 /*divMatImpl<short, float, schar>*/,
+            divMatImpl<short, float, ushort>,
+            divMatImpl<short, float, short>,
+            divMatImpl<short, float, int>,
+            divMatImpl<short, float, float>,
+            divMatImpl<short, double, double>
+        },
+        {
+            0 /*divMatImpl<int, float, uchar>*/,
+            0 /*divMatImpl<int, float, schar>*/,
+            0 /*divMatImpl<int, float, ushort>*/,
+            0 /*divMatImpl<int, float, short>*/,
+            divMatImpl<int, float, int>,
+            divMatImpl<int, float, float>,
+            divMatImpl<int, double, double>
+        },
+        {
+            0 /*divMatImpl<float, float, uchar>*/,
+            0 /*divMatImpl<float, float, schar>*/,
+            0 /*divMatImpl<float, float, ushort>*/,
+            0 /*divMatImpl<float, float, short>*/,
+            0 /*divMatImpl<float, float, int>*/,
+            divMatImpl<float, float, float>,
+            divMatImpl<float, double, double>
+        },
+        {
+            0 /*divMatImpl<double, double, uchar>*/,
+            0 /*divMatImpl<double, double, schar>*/,
+            0 /*divMatImpl<double, double, ushort>*/,
+            0 /*divMatImpl<double, double, short>*/,
+            0 /*divMatImpl<double, double, int>*/,
+            0 /*divMatImpl<double, double, float>*/,
+            divMatImpl<double, double, double>
+        }
+    };
+
+    const int sdepth = src1.depth();
+    const int ddepth = dst.depth();
+
+    CV_DbgAssert( sdepth <= CV_64F && ddepth <= CV_64F );
+
+    GpuMat src1_ = src1.reshape(1);
+    GpuMat src2_ = src2.reshape(1);
+    GpuMat dst_ = dst.reshape(1);
+
+    const func_t func = funcs[sdepth][ddepth];
+
+    if (!func)
+        CV_Error(cv::Error::StsUnsupportedFormat, "Unsupported combination of source and destination types");
+
+    func(src1_, src2_, dst_, scale, stream);
+}
+
+namespace
+{
+    template <typename T>
+    struct DivOpSpecial : binary_function<T, float, T>
+    {
+        __device__ __forceinline__ T operator ()(const T& a, float b) const
+        {
+            typedef typename VecTraits<T>::elem_type elem_type;
+
+            T res = VecTraits<T>::all(0);
+
+            if (b != 0)
+            {
+                b = 1.0f / b;
+                res.x = saturate_cast<elem_type>(a.x * b);
+                res.y = saturate_cast<elem_type>(a.y * b);
+                res.z = saturate_cast<elem_type>(a.z * b);
+                res.w = saturate_cast<elem_type>(a.w * b);
+            }
+
+            return res;
+        }
+    };
+}
+
+void divMat_8uc4_32f(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, Stream& stream)
+{
+    gridTransformBinary(globPtr<uchar4>(src1), globPtr<float>(src2), globPtr<uchar4>(dst), DivOpSpecial<uchar4>(), stream);
+}
+
+void divMat_16sc4_32f(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, Stream& stream)
+{
+    gridTransformBinary(globPtr<short4>(src1), globPtr<float>(src2), globPtr<short4>(dst), DivOpSpecial<short4>(), stream);
+}
+
+#endif
diff --git a/modules/cudaarithm/src/cuda/div_scalar.cu b/modules/cudaarithm/src/cuda/div_scalar.cu
index 9ee1727a75..97ada83410 100644
--- a/modules/cudaarithm/src/cuda/div_scalar.cu
+++ b/modules/cudaarithm/src/cuda/div_scalar.cu
@@ -40,129 +40,221 @@
 //
 //M*/
 
-#if !defined CUDA_DISABLER
+#include "opencv2/opencv_modules.hpp"
 
-#include "opencv2/core/cuda/common.hpp"
-#include "opencv2/core/cuda/functional.hpp"
-#include "opencv2/core/cuda/transform.hpp"
-#include "opencv2/core/cuda/saturate_cast.hpp"
-#include "opencv2/core/cuda/simd_functions.hpp"
+#ifndef HAVE_OPENCV_CUDEV
 
-#include "arithm_func_traits.hpp"
+#error "opencv_cudev is required"
 
-using namespace cv::cuda;
-using namespace cv::cuda::device;
+#else
 
-namespace arithm
+#include "opencv2/cudev.hpp"
+
+using namespace cv::cudev;
+
+void divScalar(const GpuMat& src, cv::Scalar val, bool inv, GpuMat& dst, const GpuMat& mask, double scale, Stream& stream, int);
+
+namespace
 {
-    template <typename T, typename S, typename D> struct DivScalar : unary_function<T, D>
+    template <typename T, int cn> struct SafeDiv;
+    template <typename T> struct SafeDiv<T, 1>
     {
-        S val;
-
-        __host__ explicit DivScalar(S val_) : val(val_) {}
-
-        __device__ __forceinline__ D operator ()(T a) const
+        __device__ __forceinline__ static T op(T a, T b)
         {
-            return saturate_cast<D>(a / val);
+            return b != 0 ? a / b : 0;
+        }
+    };
+    template <typename T> struct SafeDiv<T, 2>
+    {
+        __device__ __forceinline__ static T op(const T& a, const T& b)
+        {
+            T res;
+
+            res.x = b.x != 0 ? a.x / b.x : 0;
+            res.y = b.y != 0 ? a.y / b.y : 0;
+
+            return res;
+        }
+    };
+    template <typename T> struct SafeDiv<T, 3>
+    {
+        __device__ __forceinline__ static T op(const T& a, const T& b)
+        {
+            T res;
+
+            res.x = b.x != 0 ? a.x / b.x : 0;
+            res.y = b.y != 0 ? a.y / b.y : 0;
+            res.z = b.z != 0 ? a.z / b.z : 0;
+
+            return res;
+        }
+    };
+    template <typename T> struct SafeDiv<T, 4>
+    {
+        __device__ __forceinline__ static T op(const T& a, const T& b)
+        {
+            T res;
+
+            res.x = b.x != 0 ? a.x / b.x : 0;
+            res.y = b.y != 0 ? a.y / b.y : 0;
+            res.z = b.z != 0 ? a.z / b.z : 0;
+            res.w = b.w != 0 ? a.w / b.w : 0;
+
+            return res;
         }
     };
 
-    template <typename T, typename S, typename D> struct DivScalarInv : unary_function<T, D>
+    template <typename SrcType, typename ScalarType, typename DstType> struct DivScalarOp : unary_function<SrcType, DstType>
     {
-        S val;
+        ScalarType val;
 
-        explicit DivScalarInv(S val_) : val(val_) {}
-
-        __device__ __forceinline__ D operator ()(T a) const
+        __device__ __forceinline__ DstType operator ()(SrcType a) const
         {
-            return a != 0 ? saturate_cast<D>(val / a) : 0;
+            return saturate_cast<DstType>(SafeDiv<ScalarType, VecTraits<ScalarType>::cn>::op(saturate_cast<ScalarType>(a), val));
         }
     };
-}
 
-namespace cv { namespace cuda { namespace device
-{
-    template <typename T, typename S, typename D> struct TransformFunctorTraits< arithm::DivScalar<T, S, D> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(D)>
+    template <typename SrcType, typename ScalarType, typename DstType> struct DivScalarOpInv : unary_function<SrcType, DstType>
     {
+        ScalarType val;
+
+        __device__ __forceinline__ DstType operator ()(SrcType a) const
+        {
+            return saturate_cast<DstType>(SafeDiv<ScalarType, VecTraits<ScalarType>::cn>::op(val, saturate_cast<ScalarType>(a)));
+        }
     };
 
-    template <typename T, typename S, typename D> struct TransformFunctorTraits< arithm::DivScalarInv<T, S, D> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(D)>
+    template <typename ScalarDepth> struct TransformPolicy : DefaultTransformPolicy
     {
     };
-}}}
-
-namespace arithm
-{
-    template <typename T, typename S, typename D>
-    void divScalar(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream)
+    template <> struct TransformPolicy<double> : DefaultTransformPolicy
     {
+        enum {
+            shift = 1
+        };
+    };
+
+    template <typename SrcType, typename ScalarDepth, typename DstType>
+    void divScalarImpl(const GpuMat& src, cv::Scalar value, bool inv, GpuMat& dst, Stream& stream)
+    {
+        typedef typename MakeVec<ScalarDepth, VecTraits<SrcType>::cn>::type ScalarType;
+
+        cv::Scalar_<ScalarDepth> value_ = value;
+
         if (inv)
         {
-            DivScalarInv<T, S, D> op(static_cast<S>(val));
-            device::transform((PtrStepSz<T>) src1, (PtrStepSz<D>) dst, op, WithOutMask(), stream);
+            DivScalarOpInv<SrcType, ScalarType, DstType> op;
+            op.val = VecTraits<ScalarType>::make(value_.val);
+
+            gridTransformUnary_< TransformPolicy<ScalarDepth> >(globPtr<SrcType>(src), globPtr<DstType>(dst), op, stream);
         }
         else
         {
-            DivScalar<T, S, D> op(static_cast<S>(val));
-            device::transform((PtrStepSz<T>) src1, (PtrStepSz<D>) dst, op, WithOutMask(), stream);
+            DivScalarOp<SrcType, ScalarType, DstType> op;
+            op.val = VecTraits<ScalarType>::make(value_.val);
+
+            gridTransformUnary_< TransformPolicy<ScalarDepth> >(globPtr<SrcType>(src), globPtr<DstType>(dst), op, stream);
         }
     }
-
-    template void divScalar<uchar, float, uchar>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-    template void divScalar<uchar, float, schar>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-    template void divScalar<uchar, float, ushort>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-    template void divScalar<uchar, float, short>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-    template void divScalar<uchar, float, int>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-    template void divScalar<uchar, float, float>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-    template void divScalar<uchar, double, double>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-
-    template void divScalar<schar, float, uchar>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-    template void divScalar<schar, float, schar>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-    template void divScalar<schar, float, ushort>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-    template void divScalar<schar, float, short>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-    template void divScalar<schar, float, int>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-    template void divScalar<schar, float, float>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-    template void divScalar<schar, double, double>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-
-    //template void divScalar<ushort, float, uchar>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-    //template void divScalar<ushort, float, schar>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-    template void divScalar<ushort, float, ushort>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-    template void divScalar<ushort, float, short>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-    template void divScalar<ushort, float, int>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-    template void divScalar<ushort, float, float>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-    template void divScalar<ushort, double, double>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-
-    //template void divScalar<short, float, uchar>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-    //template void divScalar<short, float, schar>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-    template void divScalar<short, float, ushort>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-    template void divScalar<short, float, short>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-    template void divScalar<short, float, int>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-    template void divScalar<short, float, float>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-    template void divScalar<short, double, double>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-
-    //template void divScalar<int, float, uchar>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-    //template void divScalar<int, float, schar>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-    //template void divScalar<int, float, ushort>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-    //template void divScalar<int, float, short>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-    template void divScalar<int, float, int>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-    template void divScalar<int, float, float>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-    template void divScalar<int, double, double>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-
-    //template void divScalar<float, float, uchar>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-    //template void divScalar<float, float, schar>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-    //template void divScalar<float, float, ushort>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-    //template void divScalar<float, float, short>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-    //template void divScalar<float, float, int>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-    template void divScalar<float, float, float>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-    template void divScalar<float, double, double>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-
-    //template void divScalar<double, double, uchar>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-    //template void divScalar<double, double, schar>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-    //template void divScalar<double, double, ushort>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-    //template void divScalar<double, double, short>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-    //template void divScalar<double, double, int>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-    //template void divScalar<double, double, float>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-    template void divScalar<double, double, double>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
 }
 
-#endif // CUDA_DISABLER
+void divScalar(const GpuMat& src, cv::Scalar val, bool inv, GpuMat& dst, const GpuMat&, double scale, Stream& stream, int)
+{
+    typedef void (*func_t)(const GpuMat& src, cv::Scalar val, bool inv, GpuMat& dst, Stream& stream);
+    static const func_t funcs[7][7][4] =
+    {
+        {
+            {divScalarImpl<uchar, float, uchar>, divScalarImpl<uchar2, float, uchar2>, divScalarImpl<uchar3, float, uchar3>, divScalarImpl<uchar4, float, uchar4>},
+            {divScalarImpl<uchar, float, schar>, divScalarImpl<uchar2, float, char2>, divScalarImpl<uchar3, float, char3>, divScalarImpl<uchar4, float, char4>},
+            {divScalarImpl<uchar, float, ushort>, divScalarImpl<uchar2, float, ushort2>, divScalarImpl<uchar3, float, ushort3>, divScalarImpl<uchar4, float, ushort4>},
+            {divScalarImpl<uchar, float, short>, divScalarImpl<uchar2, float, short2>, divScalarImpl<uchar3, float, short3>, divScalarImpl<uchar4, float, short4>},
+            {divScalarImpl<uchar, float, int>, divScalarImpl<uchar2, float, int2>, divScalarImpl<uchar3, float, int3>, divScalarImpl<uchar4, float, int4>},
+            {divScalarImpl<uchar, float, float>, divScalarImpl<uchar2, float, float2>, divScalarImpl<uchar3, float, float3>, divScalarImpl<uchar4, float, float4>},
+            {divScalarImpl<uchar, double, double>, divScalarImpl<uchar2, double, double2>, divScalarImpl<uchar3, double, double3>, divScalarImpl<uchar4, double, double4>}
+        },
+        {
+            {divScalarImpl<schar, float, uchar>, divScalarImpl<char2, float, uchar2>, divScalarImpl<char3, float, uchar3>, divScalarImpl<char4, float, uchar4>},
+            {divScalarImpl<schar, float, schar>, divScalarImpl<char2, float, char2>, divScalarImpl<char3, float, char3>, divScalarImpl<char4, float, char4>},
+            {divScalarImpl<schar, float, ushort>, divScalarImpl<char2, float, ushort2>, divScalarImpl<char3, float, ushort3>, divScalarImpl<char4, float, ushort4>},
+            {divScalarImpl<schar, float, short>, divScalarImpl<char2, float, short2>, divScalarImpl<char3, float, short3>, divScalarImpl<char4, float, short4>},
+            {divScalarImpl<schar, float, int>, divScalarImpl<char2, float, int2>, divScalarImpl<char3, float, int3>, divScalarImpl<char4, float, int4>},
+            {divScalarImpl<schar, float, float>, divScalarImpl<char2, float, float2>, divScalarImpl<char3, float, float3>, divScalarImpl<char4, float, float4>},
+            {divScalarImpl<schar, double, double>, divScalarImpl<char2, double, double2>, divScalarImpl<char3, double, double3>, divScalarImpl<char4, double, double4>}
+        },
+        {
+            {0 /*divScalarImpl<ushort, float, uchar>*/, 0 /*divScalarImpl<ushort2, float, uchar2>*/, 0 /*divScalarImpl<ushort3, float, uchar3>*/, 0 /*divScalarImpl<ushort4, float, uchar4>*/},
+            {0 /*divScalarImpl<ushort, float, schar>*/, 0 /*divScalarImpl<ushort2, float, char2>*/, 0 /*divScalarImpl<ushort3, float, char3>*/, 0 /*divScalarImpl<ushort4, float, char4>*/},
+            {divScalarImpl<ushort, float, ushort>, divScalarImpl<ushort2, float, ushort2>, divScalarImpl<ushort3, float, ushort3>, divScalarImpl<ushort4, float, ushort4>},
+            {divScalarImpl<ushort, float, short>, divScalarImpl<ushort2, float, short2>, divScalarImpl<ushort3, float, short3>, divScalarImpl<ushort4, float, short4>},
+            {divScalarImpl<ushort, float, int>, divScalarImpl<ushort2, float, int2>, divScalarImpl<ushort3, float, int3>, divScalarImpl<ushort4, float, int4>},
+            {divScalarImpl<ushort, float, float>, divScalarImpl<ushort2, float, float2>, divScalarImpl<ushort3, float, float3>, divScalarImpl<ushort4, float, float4>},
+            {divScalarImpl<ushort, double, double>, divScalarImpl<ushort2, double, double2>, divScalarImpl<ushort3, double, double3>, divScalarImpl<ushort4, double, double4>}
+        },
+        {
+            {0 /*divScalarImpl<short, float, uchar>*/, 0 /*divScalarImpl<short2, float, uchar2>*/, 0 /*divScalarImpl<short3, float, uchar3>*/, 0 /*divScalarImpl<short4, float, uchar4>*/},
+            {0 /*divScalarImpl<short, float, schar>*/, 0 /*divScalarImpl<short2, float, char2>*/, 0 /*divScalarImpl<short3, float, char3>*/, 0 /*divScalarImpl<short4, float, char4>*/},
+            {divScalarImpl<short, float, ushort>, divScalarImpl<short2, float, ushort2>, divScalarImpl<short3, float, ushort3>, divScalarImpl<short4, float, ushort4>},
+            {divScalarImpl<short, float, short>, divScalarImpl<short2, float, short2>, divScalarImpl<short3, float, short3>, divScalarImpl<short4, float, short4>},
+            {divScalarImpl<short, float, int>, divScalarImpl<short2, float, int2>, divScalarImpl<short3, float, int3>, divScalarImpl<short4, float, int4>},
+            {divScalarImpl<short, float, float>, divScalarImpl<short2, float, float2>, divScalarImpl<short3, float, float3>, divScalarImpl<short4, float, float4>},
+            {divScalarImpl<short, double, double>, divScalarImpl<short2, double, double2>, divScalarImpl<short3, double, double3>, divScalarImpl<short4, double, double4>}
+        },
+        {
+            {0 /*divScalarImpl<int, float, uchar>*/, 0 /*divScalarImpl<int2, float, uchar2>*/, 0 /*divScalarImpl<int3, float, uchar3>*/, 0 /*divScalarImpl<int4, float, uchar4>*/},
+            {0 /*divScalarImpl<int, float, schar>*/, 0 /*divScalarImpl<int2, float, char2>*/, 0 /*divScalarImpl<int3, float, char3>*/, 0 /*divScalarImpl<int4, float, char4>*/},
+            {0 /*divScalarImpl<int, float, ushort>*/, 0 /*divScalarImpl<int2, float, ushort2>*/, 0 /*divScalarImpl<int3, float, ushort3>*/, 0 /*divScalarImpl<int4, float, ushort4>*/},
+            {0 /*divScalarImpl<int, float, short>*/, 0 /*divScalarImpl<int2, float, short2>*/, 0 /*divScalarImpl<int3, float, short3>*/, 0 /*divScalarImpl<int4, float, short4>*/},
+            {divScalarImpl<int, float, int>, divScalarImpl<int2, float, int2>, divScalarImpl<int3, float, int3>, divScalarImpl<int4, float, int4>},
+            {divScalarImpl<int, float, float>, divScalarImpl<int2, float, float2>, divScalarImpl<int3, float, float3>, divScalarImpl<int4, float, float4>},
+            {divScalarImpl<int, double, double>, divScalarImpl<int2, double, double2>, divScalarImpl<int3, double, double3>, divScalarImpl<int4, double, double4>}
+        },
+        {
+            {0 /*divScalarImpl<float, float, uchar>*/, 0 /*divScalarImpl<float2, float, uchar2>*/, 0 /*divScalarImpl<float3, float, uchar3>*/, 0 /*divScalarImpl<float4, float, uchar4>*/},
+            {0 /*divScalarImpl<float, float, schar>*/, 0 /*divScalarImpl<float2, float, char2>*/, 0 /*divScalarImpl<float3, float, char3>*/, 0 /*divScalarImpl<float4, float, char4>*/},
+            {0 /*divScalarImpl<float, float, ushort>*/, 0 /*divScalarImpl<float2, float, ushort2>*/, 0 /*divScalarImpl<float3, float, ushort3>*/, 0 /*divScalarImpl<float4, float, ushort4>*/},
+            {0 /*divScalarImpl<float, float, short>*/, 0 /*divScalarImpl<float2, float, short2>*/, 0 /*divScalarImpl<float3, float, short3>*/, 0 /*divScalarImpl<float4, float, short4>*/},
+            {0 /*divScalarImpl<float, float, int>*/, 0 /*divScalarImpl<float2, float, int2>*/, 0 /*divScalarImpl<float3, float, int3>*/, 0 /*divScalarImpl<float4, float, int4>*/},
+            {divScalarImpl<float, float, float>, divScalarImpl<float2, float, float2>, divScalarImpl<float3, float, float3>, divScalarImpl<float4, float, float4>},
+            {divScalarImpl<float, double, double>, divScalarImpl<float2, double, double2>, divScalarImpl<float3, double, double3>, divScalarImpl<float4, double, double4>}
+        },
+        {
+            {0 /*divScalarImpl<double, double, uchar>*/, 0 /*divScalarImpl<double2, double, uchar2>*/, 0 /*divScalarImpl<double3, double, uchar3>*/, 0 /*divScalarImpl<double4, double, uchar4>*/},
+            {0 /*divScalarImpl<double, double, schar>*/, 0 /*divScalarImpl<double2, double, char2>*/, 0 /*divScalarImpl<double3, double, char3>*/, 0 /*divScalarImpl<double4, double, char4>*/},
+            {0 /*divScalarImpl<double, double, ushort>*/, 0 /*divScalarImpl<double2, double, ushort2>*/, 0 /*divScalarImpl<double3, double, ushort3>*/, 0 /*divScalarImpl<double4, double, ushort4>*/},
+            {0 /*divScalarImpl<double, double, short>*/, 0 /*divScalarImpl<double2, double, short2>*/, 0 /*divScalarImpl<double3, double, short3>*/, 0 /*divScalarImpl<double4, double, short4>*/},
+            {0 /*divScalarImpl<double, double, int>*/, 0 /*divScalarImpl<double2, double, int2>*/, 0 /*divScalarImpl<double3, double, int3>*/, 0 /*divScalarImpl<double4, double, int4>*/},
+            {0 /*divScalarImpl<double, double, float>*/, 0 /*divScalarImpl<double2, double, float2>*/, 0 /*divScalarImpl<double3, double, float3>*/, 0 /*divScalarImpl<double4, double, float4>*/},
+            {divScalarImpl<double, double, double>, divScalarImpl<double2, double, double2>, divScalarImpl<double3, double, double3>, divScalarImpl<double4, double, double4>}
+        }
+    };
+
+    const int sdepth = src.depth();
+    const int ddepth = dst.depth();
+    const int cn = src.channels();
+
+    CV_DbgAssert( sdepth <= CV_64F && ddepth <= CV_64F && cn <= 4 );
+
+    if (inv)
+    {
+        val[0] *= scale;
+        val[1] *= scale;
+        val[2] *= scale;
+        val[3] *= scale;
+    }
+    else
+    {
+        val[0] /= scale;
+        val[1] /= scale;
+        val[2] /= scale;
+        val[3] /= scale;
+    }
+
+    const func_t func = funcs[sdepth][ddepth][cn - 1];
+
+    if (!func)
+        CV_Error(cv::Error::StsUnsupportedFormat, "Unsupported combination of source and destination types");
+
+    func(src, val, inv, dst, stream);
+}
+
+#endif
diff --git a/modules/cudaarithm/src/cuda/integral.cu b/modules/cudaarithm/src/cuda/integral.cu
index ef49f1814d..db554eb301 100644
--- a/modules/cudaarithm/src/cuda/integral.cu
+++ b/modules/cudaarithm/src/cuda/integral.cu
@@ -40,433 +40,61 @@
 //
 //M*/
 
-#if !defined CUDA_DISABLER
+#include "opencv2/opencv_modules.hpp"
 
-#include "opencv2/core/cuda/common.hpp"
+#ifndef HAVE_OPENCV_CUDEV
 
-namespace cv { namespace cuda { namespace device
+#error "opencv_cudev is required"
+
+#else
+
+#include "opencv2/cudaarithm.hpp"
+#include "opencv2/cudev.hpp"
+
+using namespace cv::cudev;
+
+////////////////////////////////////////////////////////////////////////
+// integral
+
+void cv::cuda::integral(InputArray _src, OutputArray _dst, GpuMat& buffer, Stream& stream)
 {
-    namespace imgproc
-    {
-        // Utility function to extract unsigned chars from an unsigned integer
-        __device__ uchar4 int_to_uchar4(unsigned int in)
-        {
-            uchar4 bytes;
-            bytes.x = (in & 0x000000ff) >>  0;
-            bytes.y = (in & 0x0000ff00) >>  8;
-            bytes.z = (in & 0x00ff0000) >> 16;
-            bytes.w = (in & 0xff000000) >> 24;
-            return bytes;
-        }
+    GpuMat src = _src.getGpuMat();
 
-        __global__ void shfl_integral_horizontal(const PtrStep<uint4> img, PtrStep<uint4> integral)
-        {
-        #if defined(__CUDA_ARCH__) && (__CUDA_ARCH__ >= 300)
-            __shared__ int sums[128];
+    CV_Assert( src.type() == CV_8UC1 );
 
-            const int id = threadIdx.x;
-            const int lane_id = id % warpSize;
-            const int warp_id = id / warpSize;
+    GpuMat_<int>& res = (GpuMat_<int>&) buffer;
 
-            const uint4 data = img(blockIdx.x, id);
+    gridIntegral(globPtr<uchar>(src), res, stream);
 
-            const uchar4 a = int_to_uchar4(data.x);
-            const uchar4 b = int_to_uchar4(data.y);
-            const uchar4 c = int_to_uchar4(data.z);
-            const uchar4 d = int_to_uchar4(data.w);
+    _dst.create(src.rows + 1, src.cols + 1, CV_32SC1);
+    GpuMat dst = _dst.getGpuMat();
 
-            int result[16];
+    dst.setTo(Scalar::all(0), stream);
 
-            result[0]  =              a.x;
-            result[1]  = result[0]  + a.y;
-            result[2]  = result[1]  + a.z;
-            result[3]  = result[2]  + a.w;
+    GpuMat inner = dst(Rect(1, 1, src.cols, src.rows));
+    res.copyTo(inner, stream);
+}
 
-            result[4]  = result[3]  + b.x;
-            result[5]  = result[4]  + b.y;
-            result[6]  = result[5]  + b.z;
-            result[7]  = result[6]  + b.w;
+//////////////////////////////////////////////////////////////////////////////
+// sqrIntegral
 
-            result[8]  = result[7]  + c.x;
-            result[9]  = result[8]  + c.y;
-            result[10] = result[9]  + c.z;
-            result[11] = result[10] + c.w;
+void cv::cuda::sqrIntegral(InputArray _src, OutputArray _dst, GpuMat& buf, Stream& stream)
+{
+    GpuMat src = _src.getGpuMat();
 
-            result[12] = result[11] + d.x;
-            result[13] = result[12] + d.y;
-            result[14] = result[13] + d.z;
-            result[15] = result[14] + d.w;
+    CV_Assert( src.type() == CV_8UC1 );
 
-            int sum = result[15];
+    GpuMat_<double>& res = (GpuMat_<double>&) buf;
 
-            // the prefix sum for each thread's 16 value is computed,
-            // now the final sums (result[15]) need to be shared
-            // with the other threads and add.  To do this,
-            // the __shfl_up() instruction is used and a shuffle scan
-            // operation is performed to distribute the sums to the correct
-            // threads
-            #pragma unroll
-            for (int i = 1; i < 32; i *= 2)
-            {
-                const int n = __shfl_up(sum, i, 32);
+    gridIntegral(sqr_(cvt_<int>(globPtr<uchar>(src))), res, stream);
 
-                if (lane_id >= i)
-                {
-                    #pragma unroll
-                    for (int i = 0; i < 16; ++i)
-                        result[i] += n;
+    _dst.create(src.rows + 1, src.cols + 1, CV_64FC1);
+    GpuMat dst = _dst.getGpuMat();
 
-                    sum += n;
-                }
-            }
+    dst.setTo(Scalar::all(0), stream);
 
-            // Now the final sum for the warp must be shared
-            // between warps.  This is done by each warp
-            // having a thread store to shared memory, then
-            // having some other warp load the values and
-            // compute a prefix sum, again by using __shfl_up.
-            // The results are uniformly added back to the warps.
-            // last thread in the warp holding sum of the warp
-            // places that in shared
-            if (threadIdx.x % warpSize == warpSize - 1)
-                sums[warp_id] = result[15];
+    GpuMat inner = dst(Rect(1, 1, src.cols, src.rows));
+    res.copyTo(inner, stream);
+}
 
-            __syncthreads();
-
-            if (warp_id == 0)
-            {
-                int warp_sum = sums[lane_id];
-
-                #pragma unroll
-                for (int i = 1; i <= 32; i *= 2)
-                {
-                    const int n = __shfl_up(warp_sum, i, 32);
-
-                    if (lane_id >= i)
-                        warp_sum += n;
-                }
-
-                sums[lane_id] = warp_sum;
-            }
-
-            __syncthreads();
-
-            int blockSum = 0;
-
-            // fold in unused warp
-            if (warp_id > 0)
-            {
-                blockSum = sums[warp_id - 1];
-
-                #pragma unroll
-                for (int i = 0; i < 16; ++i)
-                    result[i] += blockSum;
-            }
-
-            // assemble result
-            // Each thread has 16 values to write, which are
-            // now integer data (to avoid overflow).  Instead of
-            // each thread writing consecutive uint4s, the
-            // approach shown here experiments using
-            // the shuffle command to reformat the data
-            // inside the registers so that each thread holds
-            // consecutive data to be written so larger contiguous
-            // segments can be assembled for writing.
-
-            /*
-                For example data that needs to be written as
-
-                GMEM[16] <- x0 x1 x2 x3 y0 y1 y2 y3 z0 z1 z2 z3 w0 w1 w2 w3
-                but is stored in registers (r0..r3), in four threads (0..3) as:
-
-                threadId   0  1  2  3
-                  r0      x0 y0 z0 w0
-                  r1      x1 y1 z1 w1
-                  r2      x2 y2 z2 w2
-                  r3      x3 y3 z3 w3
-
-                  after apply __shfl_xor operations to move data between registers r1..r3:
-
-                threadId  00 01 10 11
-                          x0 y0 z0 w0
-                 xor(01)->y1 x1 w1 z1
-                 xor(10)->z2 w2 x2 y2
-                 xor(11)->w3 z3 y3 x3
-
-                 and now x0..x3, and z0..z3 can be written out in order by all threads.
-
-                 In the current code, each register above is actually representing
-                 four integers to be written as uint4's to GMEM.
-            */
-
-            result[4]  = __shfl_xor(result[4] , 1, 32);
-            result[5]  = __shfl_xor(result[5] , 1, 32);
-            result[6]  = __shfl_xor(result[6] , 1, 32);
-            result[7]  = __shfl_xor(result[7] , 1, 32);
-
-            result[8]  = __shfl_xor(result[8] , 2, 32);
-            result[9]  = __shfl_xor(result[9] , 2, 32);
-            result[10] = __shfl_xor(result[10], 2, 32);
-            result[11] = __shfl_xor(result[11], 2, 32);
-
-            result[12] = __shfl_xor(result[12], 3, 32);
-            result[13] = __shfl_xor(result[13], 3, 32);
-            result[14] = __shfl_xor(result[14], 3, 32);
-            result[15] = __shfl_xor(result[15], 3, 32);
-
-            uint4* integral_row = integral.ptr(blockIdx.x);
-            uint4 output;
-
-            ///////
-
-            if (threadIdx.x % 4 == 0)
-                output = make_uint4(result[0], result[1], result[2], result[3]);
-
-            if (threadIdx.x % 4 == 1)
-                output = make_uint4(result[4], result[5], result[6], result[7]);
-
-            if (threadIdx.x % 4 == 2)
-                output = make_uint4(result[8], result[9], result[10], result[11]);
-
-            if (threadIdx.x % 4 == 3)
-                output = make_uint4(result[12], result[13], result[14], result[15]);
-
-            integral_row[threadIdx.x % 4 + (threadIdx.x / 4) * 16] = output;
-
-            ///////
-
-            if (threadIdx.x % 4 == 2)
-                output = make_uint4(result[0], result[1], result[2], result[3]);
-
-            if (threadIdx.x % 4 == 3)
-                output = make_uint4(result[4], result[5], result[6], result[7]);
-
-            if (threadIdx.x % 4 == 0)
-                output = make_uint4(result[8], result[9], result[10], result[11]);
-
-            if (threadIdx.x % 4 == 1)
-                output = make_uint4(result[12], result[13], result[14], result[15]);
-
-            integral_row[(threadIdx.x + 2) % 4 + (threadIdx.x / 4) * 16 + 8] = output;
-
-            // continuning from the above example,
-            // this use of __shfl_xor() places the y0..y3 and w0..w3 data
-            // in order.
-
-            #pragma unroll
-            for (int i = 0; i < 16; ++i)
-                result[i] = __shfl_xor(result[i], 1, 32);
-
-            if (threadIdx.x % 4 == 0)
-                output = make_uint4(result[0], result[1], result[2], result[3]);
-
-            if (threadIdx.x % 4 == 1)
-                output = make_uint4(result[4], result[5], result[6], result[7]);
-
-            if (threadIdx.x % 4 == 2)
-                output = make_uint4(result[8], result[9], result[10], result[11]);
-
-            if (threadIdx.x % 4 == 3)
-                output = make_uint4(result[12], result[13], result[14], result[15]);
-
-            integral_row[threadIdx.x % 4 + (threadIdx.x / 4) * 16 + 4] = output;
-
-            ///////
-
-            if (threadIdx.x % 4 == 2)
-                output = make_uint4(result[0], result[1], result[2], result[3]);
-
-            if (threadIdx.x % 4 == 3)
-                output = make_uint4(result[4], result[5], result[6], result[7]);
-
-            if (threadIdx.x % 4 == 0)
-                output = make_uint4(result[8], result[9], result[10], result[11]);
-
-            if (threadIdx.x % 4 == 1)
-                output = make_uint4(result[12], result[13], result[14], result[15]);
-
-            integral_row[(threadIdx.x + 2) % 4 + (threadIdx.x / 4) * 16 + 12] = output;
-        #endif
-        }
-
-        // This kernel computes columnwise prefix sums.  When the data input is
-        // the row sums from above, this completes the integral image.
-        // The approach here is to have each block compute a local set of sums.
-        // First , the data covered by the block is loaded into shared memory,
-        // then instead of performing a sum in shared memory using __syncthreads
-        // between stages, the data is reformatted so that the necessary sums
-        // occur inside warps and the shuffle scan operation is used.
-        // The final set of sums from the block is then propgated, with the block
-        // computing "down" the image and adding the running sum to the local
-        // block sums.
-        __global__ void shfl_integral_vertical(PtrStepSz<unsigned int> integral)
-        {
-        #if defined(__CUDA_ARCH__) && (__CUDA_ARCH__ >= 300)
-            __shared__ unsigned int sums[32][9];
-
-            const int tidx = blockIdx.x * blockDim.x + threadIdx.x;
-            const int lane_id = tidx % 8;
-
-            if (tidx >= integral.cols)
-                return;
-
-            sums[threadIdx.x][threadIdx.y] = 0;
-            __syncthreads();
-
-            unsigned int stepSum = 0;
-
-            for (int y = threadIdx.y; y < integral.rows; y += blockDim.y)
-            {
-                unsigned int* p = integral.ptr(y) + tidx;
-
-                unsigned int sum = *p;
-
-                sums[threadIdx.x][threadIdx.y] = sum;
-                __syncthreads();
-
-                // place into SMEM
-                // shfl scan reduce the SMEM, reformating so the column
-                // sums are computed in a warp
-                // then read out properly
-                const int j = threadIdx.x % 8;
-                const int k = threadIdx.x / 8 + threadIdx.y * 4;
-
-                int partial_sum = sums[k][j];
-
-                for (int i = 1; i <= 8; i *= 2)
-                {
-                    int n = __shfl_up(partial_sum, i, 32);
-
-                    if (lane_id >= i)
-                        partial_sum += n;
-                }
-
-                sums[k][j] = partial_sum;
-                __syncthreads();
-
-                if (threadIdx.y > 0)
-                    sum += sums[threadIdx.x][threadIdx.y - 1];
-
-                sum += stepSum;
-                stepSum += sums[threadIdx.x][blockDim.y - 1];
-
-                __syncthreads();
-
-                *p = sum;
-            }
-        #endif
-        }
-
-        void shfl_integral_gpu(const PtrStepSzb& img, PtrStepSz<unsigned int> integral, cudaStream_t stream)
-        {
-            {
-                // each thread handles 16 values, use 1 block/row
-                // save, becouse step is actually can't be less 512 bytes
-                int block = integral.cols / 16;
-
-                // launch 1 block / row
-                const int grid = img.rows;
-
-                cudaSafeCall( cudaFuncSetCacheConfig(shfl_integral_horizontal, cudaFuncCachePreferL1) );
-
-                shfl_integral_horizontal<<<grid, block, 0, stream>>>((const PtrStepSz<uint4>) img, (PtrStepSz<uint4>) integral);
-                cudaSafeCall( cudaGetLastError() );
-            }
-
-            {
-                const dim3 block(32, 8);
-                const dim3 grid(divUp(integral.cols, block.x), 1);
-
-                shfl_integral_vertical<<<grid, block, 0, stream>>>(integral);
-                cudaSafeCall( cudaGetLastError() );
-            }
-
-            if (stream == 0)
-                cudaSafeCall( cudaDeviceSynchronize() );
-        }
-
-        __global__ void shfl_integral_vertical(PtrStepSz<unsigned int> buffer, PtrStepSz<unsigned int> integral)
-        {
-        #if defined(__CUDA_ARCH__) && (__CUDA_ARCH__ >= 300)
-            __shared__ unsigned int sums[32][9];
-
-            const int tidx = blockIdx.x * blockDim.x + threadIdx.x;
-            const int lane_id = tidx % 8;
-
-            if (tidx >= integral.cols)
-                return;
-
-            sums[threadIdx.x][threadIdx.y] = 0;
-            __syncthreads();
-
-            unsigned int stepSum = 0;
-
-            for (int y = threadIdx.y; y < integral.rows; y += blockDim.y)
-            {
-                unsigned int* p = buffer.ptr(y) + tidx;
-                unsigned int* dst = integral.ptr(y + 1) + tidx + 1;
-
-                unsigned int sum = *p;
-
-                sums[threadIdx.x][threadIdx.y] = sum;
-                __syncthreads();
-
-                // place into SMEM
-                // shfl scan reduce the SMEM, reformating so the column
-                // sums are computed in a warp
-                // then read out properly
-                const int j = threadIdx.x % 8;
-                const int k = threadIdx.x / 8 + threadIdx.y * 4;
-
-                int partial_sum = sums[k][j];
-
-                for (int i = 1; i <= 8; i *= 2)
-                {
-                    int n = __shfl_up(partial_sum, i, 32);
-
-                    if (lane_id >= i)
-                        partial_sum += n;
-                }
-
-                sums[k][j] = partial_sum;
-                __syncthreads();
-
-                if (threadIdx.y > 0)
-                    sum += sums[threadIdx.x][threadIdx.y - 1];
-
-                sum += stepSum;
-                stepSum += sums[threadIdx.x][blockDim.y - 1];
-
-                __syncthreads();
-
-                *dst = sum;
-            }
-        #endif
-        }
-
-        // used for frame preprocessing before Soft Cascade evaluation: no synchronization needed
-        void shfl_integral_gpu_buffered(PtrStepSzb img, PtrStepSz<uint4> buffer, PtrStepSz<unsigned int> integral,
-            int blockStep, cudaStream_t stream)
-        {
-            {
-                const int block = blockStep;
-                const int grid = img.rows;
-
-                cudaSafeCall( cudaFuncSetCacheConfig(shfl_integral_horizontal, cudaFuncCachePreferL1) );
-
-                shfl_integral_horizontal<<<grid, block, 0, stream>>>((PtrStepSz<uint4>) img, buffer);
-                cudaSafeCall( cudaGetLastError() );
-            }
-
-            {
-                const dim3 block(32, 8);
-                const dim3 grid(divUp(integral.cols, block.x), 1);
-
-                shfl_integral_vertical<<<grid, block, 0, stream>>>((PtrStepSz<uint>)buffer, integral);
-                cudaSafeCall( cudaGetLastError() );
-            }
-        }
-    }
-}}}
-
-#endif /* CUDA_DISABLER */
+#endif
diff --git a/modules/cudaarithm/src/cuda/lut.cu b/modules/cudaarithm/src/cuda/lut.cu
new file mode 100644
index 0000000000..a8d5bc5b06
--- /dev/null
+++ b/modules/cudaarithm/src/cuda/lut.cu
@@ -0,0 +1,207 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                           License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#include "opencv2/opencv_modules.hpp"
+
+#ifndef HAVE_OPENCV_CUDEV
+
+#error "opencv_cudev is required"
+
+#else
+
+#include "opencv2/cudaarithm.hpp"
+#include "opencv2/cudev.hpp"
+
+using namespace cv;
+using namespace cv::cudev;
+
+namespace
+{
+    texture<uchar, cudaTextureType1D, cudaReadModeElementType> texLutTable;
+
+    class LookUpTableImpl : public LookUpTable
+    {
+    public:
+        LookUpTableImpl(InputArray lut);
+        ~LookUpTableImpl();
+
+        void transform(InputArray src, OutputArray dst, Stream& stream = Stream::Null());
+
+    private:
+        GpuMat d_lut;
+        cudaTextureObject_t texLutTableObj;
+        bool cc30;
+    };
+
+    LookUpTableImpl::LookUpTableImpl(InputArray _lut)
+    {
+        if (_lut.kind() == _InputArray::GPU_MAT)
+        {
+            d_lut = _lut.getGpuMat();
+        }
+        else
+        {
+            Mat h_lut = _lut.getMat();
+            d_lut.upload(Mat(1, 256, h_lut.type(), h_lut.data));
+        }
+
+        CV_Assert( d_lut.depth() == CV_8U );
+        CV_Assert( d_lut.rows == 1 && d_lut.cols == 256 );
+
+        cc30 = deviceSupports(FEATURE_SET_COMPUTE_30);
+
+        if (cc30)
+        {
+            // Use the texture object
+            cudaResourceDesc texRes;
+            std::memset(&texRes, 0, sizeof(texRes));
+            texRes.resType = cudaResourceTypeLinear;
+            texRes.res.linear.devPtr = d_lut.data;
+            texRes.res.linear.desc = cudaCreateChannelDesc<uchar>();
+            texRes.res.linear.sizeInBytes = 256 * d_lut.channels() * sizeof(uchar);
+
+            cudaTextureDesc texDescr;
+            std::memset(&texDescr, 0, sizeof(texDescr));
+
+            CV_CUDEV_SAFE_CALL( cudaCreateTextureObject(&texLutTableObj, &texRes, &texDescr, 0) );
+        }
+        else
+        {
+            // Use the texture reference
+            cudaChannelFormatDesc desc = cudaCreateChannelDesc<uchar>();
+            CV_CUDEV_SAFE_CALL( cudaBindTexture(0, &texLutTable, d_lut.data, &desc) );
+        }
+    }
+
+    LookUpTableImpl::~LookUpTableImpl()
+    {
+        if (cc30)
+        {
+            // Use the texture object
+            cudaDestroyTextureObject(texLutTableObj);
+        }
+        else
+        {
+            // Use the texture reference
+            cudaUnbindTexture(texLutTable);
+        }
+    }
+
+    struct LutTablePtrC1
+    {
+        typedef uchar value_type;
+        typedef uchar index_type;
+
+        cudaTextureObject_t texLutTableObj;
+
+        __device__ __forceinline__ uchar operator ()(uchar, uchar x) const
+        {
+        #if CV_CUDEV_ARCH < 300
+            // Use the texture reference
+            return tex1Dfetch(texLutTable, x);
+        #else
+            // Use the texture object
+            return tex1Dfetch<uchar>(texLutTableObj, x);
+        #endif
+        }
+    };
+    struct LutTablePtrC3
+    {
+        typedef uchar3 value_type;
+        typedef uchar3 index_type;
+
+        cudaTextureObject_t texLutTableObj;
+
+        __device__ __forceinline__ uchar3 operator ()(const uchar3&, const uchar3& x) const
+        {
+        #if CV_CUDEV_ARCH < 300
+            // Use the texture reference
+            return make_uchar3(tex1Dfetch(texLutTable, x.x * 3), tex1Dfetch(texLutTable, x.y * 3 + 1), tex1Dfetch(texLutTable, x.z * 3 + 2));
+        #else
+            // Use the texture object
+            return make_uchar3(tex1Dfetch<uchar>(texLutTableObj, x.x * 3), tex1Dfetch<uchar>(texLutTableObj, x.y * 3 + 1), tex1Dfetch<uchar>(texLutTableObj, x.z * 3 + 2));
+        #endif
+        }
+    };
+
+    void LookUpTableImpl::transform(InputArray _src, OutputArray _dst, Stream& stream)
+    {
+        GpuMat src = _src.getGpuMat();
+
+        const int cn = src.channels();
+        const int lut_cn = d_lut.channels();
+
+        CV_Assert( src.type() == CV_8UC1 || src.type() == CV_8UC3 );
+        CV_Assert( lut_cn == 1 || lut_cn == cn );
+
+        _dst.create(src.size(), src.type());
+        GpuMat dst = _dst.getGpuMat();
+
+        if (lut_cn == 1)
+        {
+            GpuMat_<uchar> src1(src.reshape(1));
+            GpuMat_<uchar> dst1(dst.reshape(1));
+
+            LutTablePtrC1 tbl;
+            tbl.texLutTableObj = texLutTableObj;
+
+            dst1.assign(lut_(src1, tbl), stream);
+        }
+        else if (lut_cn == 3)
+        {
+            GpuMat_<uchar3>& src3 = (GpuMat_<uchar3>&) src;
+            GpuMat_<uchar3>& dst3 = (GpuMat_<uchar3>&) dst;
+
+            LutTablePtrC3 tbl;
+            tbl.texLutTableObj = texLutTableObj;
+
+            dst3.assign(lut_(src3, tbl), stream);
+        }
+    }
+}
+
+Ptr<LookUpTable> cv::cuda::createLookUpTable(InputArray lut)
+{
+    return makePtr<LookUpTableImpl>(lut);
+}
+
+#endif
diff --git a/modules/cudaarithm/src/cuda/math.cu b/modules/cudaarithm/src/cuda/math.cu
index ecd9a8cf90..39f822081d 100644
--- a/modules/cudaarithm/src/cuda/math.cu
+++ b/modules/cudaarithm/src/cuda/math.cu
@@ -40,196 +40,248 @@
 //
 //M*/
 
-#if !defined CUDA_DISABLER
+#include "opencv2/opencv_modules.hpp"
 
-#include "opencv2/core/cuda/common.hpp"
-#include "opencv2/core/cuda/functional.hpp"
-#include "opencv2/core/cuda/transform.hpp"
-#include "opencv2/core/cuda/saturate_cast.hpp"
-#include "opencv2/core/cuda/simd_functions.hpp"
-#include "opencv2/core/cuda/limits.hpp"
-#include "opencv2/core/cuda/type_traits.hpp"
+#ifndef HAVE_OPENCV_CUDEV
 
-#include "arithm_func_traits.hpp"
+#error "opencv_cudev is required"
 
-using namespace cv::cuda;
-using namespace cv::cuda::device;
+#else
 
-//////////////////////////////////////////////////////////////////////////
-// absMat
+#include "opencv2/cudaarithm.hpp"
+#include "opencv2/cudev.hpp"
 
-namespace cv { namespace cuda { namespace device
+using namespace cv::cudev;
+
+namespace
 {
-    template <typename T> struct TransformFunctorTraits< abs_func<T> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(T)>
+    template <typename ScalarDepth> struct TransformPolicy : DefaultTransformPolicy
     {
     };
-}}}
-
-namespace arithm
-{
-    template <typename T>
-    void absMat(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream)
+    template <> struct TransformPolicy<double> : DefaultTransformPolicy
     {
-        device::transform((PtrStepSz<T>) src, (PtrStepSz<T>) dst, abs_func<T>(), WithOutMask(), stream);
-    }
-
-    template void absMat<uchar>(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
-    template void absMat<schar>(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
-    template void absMat<ushort>(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
-    template void absMat<short>(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
-    template void absMat<int>(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
-    template void absMat<float>(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
-    template void absMat<double>(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
+        enum {
+            shift = 1
+        };
+    };
 }
 
-//////////////////////////////////////////////////////////////////////////
-// sqrMat
+//////////////////////////////////////////////////////////////////////////////
+/// abs
 
-namespace arithm
+namespace
 {
-    template <typename T> struct Sqr : unary_function<T, T>
+    template <typename T>
+    void absMat(const GpuMat& src, const GpuMat& dst, Stream& stream)
+    {
+        gridTransformUnary_< TransformPolicy<T> >(globPtr<T>(src), globPtr<T>(dst), abs_func<T>(), stream);
+    }
+}
+
+void cv::cuda::abs(InputArray _src, OutputArray _dst, Stream& stream)
+{
+    typedef void (*func_t)(const GpuMat& src, const GpuMat& dst, Stream& stream);
+    static const func_t funcs[] =
+    {
+        absMat<uchar>,
+        absMat<schar>,
+        absMat<ushort>,
+        absMat<short>,
+        absMat<int>,
+        absMat<float>,
+        absMat<double>
+    };
+
+    GpuMat src = _src.getGpuMat();
+
+    const int depth = src.depth();
+
+    CV_DbgAssert( depth <= CV_64F );
+
+    _dst.create(src.size(), src.type());
+    GpuMat dst = _dst.getGpuMat();
+
+    funcs[depth](src.reshape(1), dst.reshape(1), stream);
+}
+
+//////////////////////////////////////////////////////////////////////////////
+/// sqr
+
+namespace
+{
+    template <typename T> struct SqrOp : unary_function<T, T>
     {
         __device__ __forceinline__ T operator ()(T x) const
         {
             return saturate_cast<T>(x * x);
         }
-
-        __host__ __device__ __forceinline__ Sqr() {}
-        __host__ __device__ __forceinline__ Sqr(const Sqr&) {}
     };
+
+    template <typename T>
+    void sqrMat(const GpuMat& src, const GpuMat& dst, Stream& stream)
+    {
+        gridTransformUnary_< TransformPolicy<T> >(globPtr<T>(src), globPtr<T>(dst), SqrOp<T>(), stream);
+    }
 }
 
-namespace cv { namespace cuda { namespace device
+void cv::cuda::sqr(InputArray _src, OutputArray _dst, Stream& stream)
 {
-    template <typename T> struct TransformFunctorTraits< arithm::Sqr<T> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(T)>
+    typedef void (*func_t)(const GpuMat& src, const GpuMat& dst, Stream& stream);
+    static const func_t funcs[] =
     {
+        sqrMat<uchar>,
+        sqrMat<schar>,
+        sqrMat<ushort>,
+        sqrMat<short>,
+        sqrMat<int>,
+        sqrMat<float>,
+        sqrMat<double>
     };
-}}}
 
-namespace arithm
+    GpuMat src = _src.getGpuMat();
+
+    const int depth = src.depth();
+
+    CV_DbgAssert( depth <= CV_64F );
+
+    _dst.create(src.size(), src.type());
+    GpuMat dst = _dst.getGpuMat();
+
+    funcs[depth](src.reshape(1), dst.reshape(1), stream);
+}
+
+//////////////////////////////////////////////////////////////////////////////
+/// sqrt
+
+namespace
 {
     template <typename T>
-    void sqrMat(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream)
+    void sqrtMat(const GpuMat& src, const GpuMat& dst, Stream& stream)
     {
-        device::transform((PtrStepSz<T>) src, (PtrStepSz<T>) dst, Sqr<T>(), WithOutMask(), stream);
+        gridTransformUnary_< TransformPolicy<T> >(globPtr<T>(src), globPtr<T>(dst), sqrt_func<T>(), stream);
     }
-
-    template void sqrMat<uchar>(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
-    template void sqrMat<schar>(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
-    template void sqrMat<ushort>(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
-    template void sqrMat<short>(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
-    template void sqrMat<int>(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
-    template void sqrMat<float>(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
-    template void sqrMat<double>(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
 }
 
-//////////////////////////////////////////////////////////////////////////
-// sqrtMat
-
-namespace cv { namespace cuda { namespace device
+void cv::cuda::sqrt(InputArray _src, OutputArray _dst, Stream& stream)
 {
-    template <typename T> struct TransformFunctorTraits< sqrt_func<T> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(T)>
+    typedef void (*func_t)(const GpuMat& src, const GpuMat& dst, Stream& stream);
+    static const func_t funcs[] =
     {
+        sqrtMat<uchar>,
+        sqrtMat<schar>,
+        sqrtMat<ushort>,
+        sqrtMat<short>,
+        sqrtMat<int>,
+        sqrtMat<float>,
+        sqrtMat<double>
     };
-}}}
 
-namespace arithm
-{
-    template <typename T>
-    void sqrtMat(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream)
-    {
-        device::transform((PtrStepSz<T>) src, (PtrStepSz<T>) dst, sqrt_func<T>(), WithOutMask(), stream);
-    }
+    GpuMat src = _src.getGpuMat();
 
-    template void sqrtMat<uchar>(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
-    template void sqrtMat<schar>(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
-    template void sqrtMat<ushort>(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
-    template void sqrtMat<short>(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
-    template void sqrtMat<int>(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
-    template void sqrtMat<float>(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
-    template void sqrtMat<double>(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
+    const int depth = src.depth();
+
+    CV_DbgAssert( depth <= CV_64F );
+
+    _dst.create(src.size(), src.type());
+    GpuMat dst = _dst.getGpuMat();
+
+    funcs[depth](src.reshape(1), dst.reshape(1), stream);
 }
 
-//////////////////////////////////////////////////////////////////////////
-// logMat
+////////////////////////////////////////////////////////////////////////
+/// exp
 
-namespace cv { namespace cuda { namespace device
+namespace
 {
-    template <typename T> struct TransformFunctorTraits< log_func<T> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(T)>
-    {
-    };
-}}}
-
-namespace arithm
-{
-    template <typename T>
-    void logMat(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream)
-    {
-        device::transform((PtrStepSz<T>) src, (PtrStepSz<T>) dst, log_func<T>(), WithOutMask(), stream);
-    }
-
-    template void logMat<uchar>(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
-    template void logMat<schar>(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
-    template void logMat<ushort>(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
-    template void logMat<short>(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
-    template void logMat<int>(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
-    template void logMat<float>(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
-    template void logMat<double>(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
-}
-
-//////////////////////////////////////////////////////////////////////////
-// expMat
-
-namespace arithm
-{
-    template <typename T> struct Exp : unary_function<T, T>
+    template <typename T> struct ExpOp : unary_function<T, T>
     {
         __device__ __forceinline__ T operator ()(T x) const
         {
             exp_func<T> f;
             return saturate_cast<T>(f(x));
         }
-
-        __host__ __device__ __forceinline__ Exp() {}
-        __host__ __device__ __forceinline__ Exp(const Exp&) {}
     };
+
+    template <typename T>
+    void expMat(const GpuMat& src, const GpuMat& dst, Stream& stream)
+    {
+        gridTransformUnary_< TransformPolicy<T> >(globPtr<T>(src), globPtr<T>(dst), ExpOp<T>(), stream);
+    }
 }
 
-namespace cv { namespace cuda { namespace device
+void cv::cuda::exp(InputArray _src, OutputArray _dst, Stream& stream)
 {
-    template <typename T> struct TransformFunctorTraits< arithm::Exp<T> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(T)>
+    typedef void (*func_t)(const GpuMat& src, const GpuMat& dst, Stream& stream);
+    static const func_t funcs[] =
     {
+        expMat<uchar>,
+        expMat<schar>,
+        expMat<ushort>,
+        expMat<short>,
+        expMat<int>,
+        expMat<float>,
+        expMat<double>
     };
-}}}
 
-namespace arithm
+    GpuMat src = _src.getGpuMat();
+
+    const int depth = src.depth();
+
+    CV_DbgAssert( depth <= CV_64F );
+
+    _dst.create(src.size(), src.type());
+    GpuMat dst = _dst.getGpuMat();
+
+    funcs[depth](src.reshape(1), dst.reshape(1), stream);
+}
+
+////////////////////////////////////////////////////////////////////////
+// log
+
+namespace
 {
     template <typename T>
-    void expMat(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream)
+    void logMat(const GpuMat& src, const GpuMat& dst, Stream& stream)
     {
-        device::transform((PtrStepSz<T>) src, (PtrStepSz<T>) dst, Exp<T>(), WithOutMask(), stream);
+        gridTransformUnary_< TransformPolicy<T> >(globPtr<T>(src), globPtr<T>(dst), log_func<T>(), stream);
     }
-
-    template void expMat<uchar>(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
-    template void expMat<schar>(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
-    template void expMat<ushort>(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
-    template void expMat<short>(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
-    template void expMat<int>(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
-    template void expMat<float>(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
-    template void expMat<double>(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
 }
 
-//////////////////////////////////////////////////////////////////////////
+void cv::cuda::log(InputArray _src, OutputArray _dst, Stream& stream)
+{
+    typedef void (*func_t)(const GpuMat& src, const GpuMat& dst, Stream& stream);
+    static const func_t funcs[] =
+    {
+        logMat<uchar>,
+        logMat<schar>,
+        logMat<ushort>,
+        logMat<short>,
+        logMat<int>,
+        logMat<float>,
+        logMat<double>
+    };
+
+    GpuMat src = _src.getGpuMat();
+
+    const int depth = src.depth();
+
+    CV_DbgAssert( depth <= CV_64F );
+
+    _dst.create(src.size(), src.type());
+    GpuMat dst = _dst.getGpuMat();
+
+    funcs[depth](src.reshape(1), dst.reshape(1), stream);
+}
+
+////////////////////////////////////////////////////////////////////////
 // pow
 
-namespace arithm
+namespace
 {
     template<typename T, bool Signed = numeric_limits<T>::is_signed> struct PowOp : unary_function<T, T>
     {
         float power;
 
-        __host__ explicit PowOp(double power_) : power(static_cast<float>(power_)) {}
-
         __device__ __forceinline__ T operator()(T e) const
         {
             return saturate_cast<T>(__powf((float)e, power));
@@ -239,8 +291,6 @@ namespace arithm
     {
         float power;
 
-        __host__ explicit PowOp(double power_) : power(static_cast<float>(power_)) {}
-
         __device__ __forceinline__ T operator()(T e) const
         {
             T res = saturate_cast<T>(__powf((float)e, power));
@@ -255,8 +305,6 @@ namespace arithm
     {
         float power;
 
-        __host__ explicit PowOp(double power_) : power(static_cast<float>(power_)) {}
-
         __device__ __forceinline__ float operator()(float e) const
         {
             return __powf(::fabs(e), power);
@@ -266,37 +314,46 @@ namespace arithm
     {
         double power;
 
-        __host__ explicit PowOp(double power_) : power(power_) {}
-
         __device__ __forceinline__ double operator()(double e) const
         {
             return ::pow(::fabs(e), power);
         }
     };
-}
 
-namespace cv { namespace cuda { namespace device
-{
-    template <typename T> struct TransformFunctorTraits< arithm::PowOp<T> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(T)>
-    {
-    };
-}}}
-
-namespace arithm
-{
     template<typename T>
-    void pow(PtrStepSzb src, double power, PtrStepSzb dst, cudaStream_t stream)
+    void powMat(const GpuMat& src, double power, const GpuMat& dst, Stream& stream)
     {
-        device::transform((PtrStepSz<T>) src, (PtrStepSz<T>) dst, PowOp<T>(power), WithOutMask(), stream);
-    }
+        PowOp<T> op;
+        op.power = static_cast<typename LargerType<T, float>::type>(power);
 
-    template void pow<uchar>(PtrStepSzb src, double power, PtrStepSzb dst, cudaStream_t stream);
-    template void pow<schar>(PtrStepSzb src, double power, PtrStepSzb dst, cudaStream_t stream);
-    template void pow<short>(PtrStepSzb src, double power, PtrStepSzb dst, cudaStream_t stream);
-    template void pow<ushort>(PtrStepSzb src, double power, PtrStepSzb dst, cudaStream_t stream);
-    template void pow<int>(PtrStepSzb src, double power, PtrStepSzb dst, cudaStream_t stream);
-    template void pow<float>(PtrStepSzb src, double power, PtrStepSzb dst, cudaStream_t stream);
-    template void pow<double>(PtrStepSzb src, double power, PtrStepSzb dst, cudaStream_t stream);
+        gridTransformUnary_< TransformPolicy<T> >(globPtr<T>(src), globPtr<T>(dst), op, stream);
+    }
 }
 
-#endif // CUDA_DISABLER
+void cv::cuda::pow(InputArray _src, double power, OutputArray _dst, Stream& stream)
+{
+    typedef void (*func_t)(const GpuMat& src, double power, const GpuMat& dst, Stream& stream);
+    static const func_t funcs[] =
+    {
+        powMat<uchar>,
+        powMat<schar>,
+        powMat<ushort>,
+        powMat<short>,
+        powMat<int>,
+        powMat<float>,
+        powMat<double>
+    };
+
+    GpuMat src = _src.getGpuMat();
+
+    const int depth = src.depth();
+
+    CV_DbgAssert(depth <= CV_64F);
+
+    _dst.create(src.size(), src.type());
+    GpuMat dst = _dst.getGpuMat();
+
+    funcs[depth](src.reshape(1), power, dst.reshape(1), stream);
+}
+
+#endif
diff --git a/modules/cudaarithm/src/cuda/minmax.cu b/modules/cudaarithm/src/cuda/minmax.cu
index c92de44ad3..084bed8706 100644
--- a/modules/cudaarithm/src/cuda/minmax.cu
+++ b/modules/cudaarithm/src/cuda/minmax.cu
@@ -40,208 +40,72 @@
 //
 //M*/
 
-#if !defined CUDA_DISABLER
+#include "opencv2/opencv_modules.hpp"
 
-#include "opencv2/core/cuda/common.hpp"
-#include "opencv2/core/cuda/vec_traits.hpp"
-#include "opencv2/core/cuda/vec_math.hpp"
-#include "opencv2/core/cuda/functional.hpp"
-#include "opencv2/core/cuda/reduce.hpp"
-#include "opencv2/core/cuda/emulation.hpp"
-#include "opencv2/core/cuda/limits.hpp"
-#include "opencv2/core/cuda/utility.hpp"
+#ifndef HAVE_OPENCV_CUDEV
 
-using namespace cv::cuda;
-using namespace cv::cuda::device;
+#error "opencv_cudev is required"
 
-namespace minMax
+#else
+
+#include "opencv2/cudaarithm.hpp"
+#include "opencv2/cudev.hpp"
+
+using namespace cv::cudev;
+
+namespace
 {
-    __device__ unsigned int blocks_finished = 0;
-
-    // To avoid shared bank conflicts we convert each value into value of
-    // appropriate type (32 bits minimum)
-    template <typename T> struct MinMaxTypeTraits;
-    template <> struct MinMaxTypeTraits<uchar> { typedef int best_type; };
-    template <> struct MinMaxTypeTraits<schar> { typedef int best_type; };
-    template <> struct MinMaxTypeTraits<ushort> { typedef int best_type; };
-    template <> struct MinMaxTypeTraits<short> { typedef int best_type; };
-    template <> struct MinMaxTypeTraits<int> { typedef int best_type; };
-    template <> struct MinMaxTypeTraits<float> { typedef float best_type; };
-    template <> struct MinMaxTypeTraits<double> { typedef double best_type; };
-
-    template <int BLOCK_SIZE, typename R>
-    struct GlobalReduce
-    {
-        static __device__ void run(R& mymin, R& mymax, R* minval, R* maxval, int tid, int bid, R* sminval, R* smaxval)
-        {
-        #if __CUDA_ARCH__ >= 200
-            if (tid == 0)
-            {
-                Emulation::glob::atomicMin(minval, mymin);
-                Emulation::glob::atomicMax(maxval, mymax);
-            }
-        #else
-            __shared__ bool is_last;
-
-            if (tid == 0)
-            {
-                minval[bid] = mymin;
-                maxval[bid] = mymax;
-
-                __threadfence();
-
-                unsigned int ticket = ::atomicAdd(&blocks_finished, 1);
-                is_last = (ticket == gridDim.x * gridDim.y - 1);
-            }
-
-            __syncthreads();
-
-            if (is_last)
-            {
-                int idx = ::min(tid, gridDim.x * gridDim.y - 1);
-
-                mymin = minval[idx];
-                mymax = maxval[idx];
-
-                const minimum<R> minOp;
-                const maximum<R> maxOp;
-                device::reduce<BLOCK_SIZE>(smem_tuple(sminval, smaxval), thrust::tie(mymin, mymax), tid, thrust::make_tuple(minOp, maxOp));
-
-                if (tid == 0)
-                {
-                    minval[0] = mymin;
-                    maxval[0] = mymax;
-
-                    blocks_finished = 0;
-                }
-            }
-        #endif
-        }
-    };
-
-    template <int BLOCK_SIZE, typename T, typename R, class Mask>
-    __global__ void kernel(const PtrStepSz<T> src, const Mask mask, R* minval, R* maxval, const int twidth, const int theight)
-    {
-        __shared__ R sminval[BLOCK_SIZE];
-        __shared__ R smaxval[BLOCK_SIZE];
-
-        const int x0 = blockIdx.x * blockDim.x * twidth + threadIdx.x;
-        const int y0 = blockIdx.y * blockDim.y * theight + threadIdx.y;
-
-        const int tid = threadIdx.y * blockDim.x + threadIdx.x;
-        const int bid = blockIdx.y * gridDim.x + blockIdx.x;
-
-        R mymin = numeric_limits<R>::max();
-        R mymax = -numeric_limits<R>::max();
-
-        const minimum<R> minOp;
-        const maximum<R> maxOp;
-
-        for (int i = 0, y = y0; i < theight && y < src.rows; ++i, y += blockDim.y)
-        {
-            const T* ptr = src.ptr(y);
-
-            for (int j = 0, x = x0; j < twidth && x < src.cols; ++j, x += blockDim.x)
-            {
-                if (mask(y, x))
-                {
-                    const R srcVal = ptr[x];
-
-                    mymin = minOp(mymin, srcVal);
-                    mymax = maxOp(mymax, srcVal);
-                }
-            }
-        }
-
-        device::reduce<BLOCK_SIZE>(smem_tuple(sminval, smaxval), thrust::tie(mymin, mymax), tid, thrust::make_tuple(minOp, maxOp));
-
-        GlobalReduce<BLOCK_SIZE, R>::run(mymin, mymax, minval, maxval, tid, bid, sminval, smaxval);
-    }
-
-    const int threads_x = 32;
-    const int threads_y = 8;
-
-    void getLaunchCfg(int cols, int rows, dim3& block, dim3& grid)
-    {
-        block = dim3(threads_x, threads_y);
-
-        grid = dim3(divUp(cols, block.x * block.y),
-                    divUp(rows, block.y * block.x));
-
-        grid.x = ::min(grid.x, block.x);
-        grid.y = ::min(grid.y, block.y);
-    }
-
-    void getBufSize(int cols, int rows, int& bufcols, int& bufrows)
-    {
-        dim3 block, grid;
-        getLaunchCfg(cols, rows, block, grid);
-
-        bufcols = grid.x * grid.y * sizeof(double);
-        bufrows = 2;
-    }
-
-    __global__ void setDefaultKernel(int* minval_buf, int* maxval_buf)
-    {
-        *minval_buf = numeric_limits<int>::max();
-        *maxval_buf = numeric_limits<int>::min();
-    }
-    __global__ void setDefaultKernel(float* minval_buf, float* maxval_buf)
-    {
-        *minval_buf = numeric_limits<float>::max();
-        *maxval_buf = -numeric_limits<float>::max();
-    }
-    __global__ void setDefaultKernel(double* minval_buf, double* maxval_buf)
-    {
-        *minval_buf = numeric_limits<double>::max();
-        *maxval_buf = -numeric_limits<double>::max();
-    }
-
-    template <typename R>
-    void setDefault(R* minval_buf, R* maxval_buf)
-    {
-        setDefaultKernel<<<1, 1>>>(minval_buf, maxval_buf);
-    }
-
     template <typename T>
-    void run(const PtrStepSzb src, const PtrStepb mask, double* minval, double* maxval, PtrStepb buf)
+    void minMaxImpl(const GpuMat& _src, const GpuMat& mask, GpuMat& _buf, double* minVal, double* maxVal)
     {
-        typedef typename MinMaxTypeTraits<T>::best_type R;
+        typedef typename SelectIf<
+                TypesEquals<T, double>::value,
+                double,
+                typename SelectIf<TypesEquals<T, float>::value, float, int>::type
+                >::type work_type;
 
-        dim3 block, grid;
-        getLaunchCfg(src.cols, src.rows, block, grid);
+        const GpuMat_<T>& src = (const GpuMat_<T>&) _src;
+        GpuMat_<work_type>& buf = (GpuMat_<work_type>&) _buf;
 
-        const int twidth = divUp(divUp(src.cols, grid.x), block.x);
-        const int theight = divUp(divUp(src.rows, grid.y), block.y);
-
-        R* minval_buf = (R*) buf.ptr(0);
-        R* maxval_buf = (R*) buf.ptr(1);
-
-        setDefault(minval_buf, maxval_buf);
-
-        if (mask.data)
-            kernel<threads_x * threads_y><<<grid, block>>>((PtrStepSz<T>) src, SingleMask(mask), minval_buf, maxval_buf, twidth, theight);
+        if (mask.empty())
+            gridFindMinMaxVal(src, buf);
         else
-            kernel<threads_x * threads_y><<<grid, block>>>((PtrStepSz<T>) src, WithOutMask(), minval_buf, maxval_buf, twidth, theight);
+            gridFindMinMaxVal(src, buf, globPtr<uchar>(mask));
 
-        cudaSafeCall( cudaGetLastError() );
+        work_type data[2];
+        buf.download(cv::Mat(1, 2, buf.type(), data));
 
-        cudaSafeCall( cudaDeviceSynchronize() );
+        if (minVal)
+            *minVal = data[0];
 
-        R minval_, maxval_;
-        cudaSafeCall( cudaMemcpy(&minval_, minval_buf, sizeof(R), cudaMemcpyDeviceToHost) );
-        cudaSafeCall( cudaMemcpy(&maxval_, maxval_buf, sizeof(R), cudaMemcpyDeviceToHost) );
-        *minval = minval_;
-        *maxval = maxval_;
+        if (maxVal)
+            *maxVal = data[1];
     }
-
-    template void run<uchar >(const PtrStepSzb src, const PtrStepb mask, double* minval, double* maxval, PtrStepb buf);
-    template void run<schar >(const PtrStepSzb src, const PtrStepb mask, double* minval, double* maxval, PtrStepb buf);
-    template void run<ushort>(const PtrStepSzb src, const PtrStepb mask, double* minval, double* maxval, PtrStepb buf);
-    template void run<short >(const PtrStepSzb src, const PtrStepb mask, double* minval, double* maxval, PtrStepb buf);
-    template void run<int   >(const PtrStepSzb src, const PtrStepb mask, double* minval, double* maxval, PtrStepb buf);
-    template void run<float >(const PtrStepSzb src, const PtrStepb mask, double* minval, double* maxval, PtrStepb buf);
-    template void run<double>(const PtrStepSzb src, const PtrStepb mask, double* minval, double* maxval, PtrStepb buf);
 }
 
-#endif // CUDA_DISABLER
+void cv::cuda::minMax(InputArray _src, double* minVal, double* maxVal, InputArray _mask, GpuMat& buf)
+{
+    typedef void (*func_t)(const GpuMat& _src, const GpuMat& mask, GpuMat& _buf, double* minVal, double* maxVal);
+    static const func_t funcs[] =
+    {
+        minMaxImpl<uchar>,
+        minMaxImpl<schar>,
+        minMaxImpl<ushort>,
+        minMaxImpl<short>,
+        minMaxImpl<int>,
+        minMaxImpl<float>,
+        minMaxImpl<double>
+    };
+
+    GpuMat src = _src.getGpuMat();
+    GpuMat mask = _mask.getGpuMat();
+
+    CV_Assert( src.channels() == 1 );
+    CV_DbgAssert( mask.empty() || (mask.size() == src.size() && mask.type() == CV_8U) );
+
+    const func_t func = funcs[src.depth()];
+
+    func(src, mask, buf, minVal, maxVal);
+}
+
+#endif
diff --git a/modules/cudaarithm/src/cuda/minmax_mat.cu b/modules/cudaarithm/src/cuda/minmax_mat.cu
index 9af5c81e39..405b230868 100644
--- a/modules/cudaarithm/src/cuda/minmax_mat.cu
+++ b/modules/cudaarithm/src/cuda/minmax_mat.cu
@@ -40,189 +40,204 @@
 //
 //M*/
 
-#if !defined CUDA_DISABLER
+#include "opencv2/opencv_modules.hpp"
 
-#include "opencv2/core/cuda/common.hpp"
-#include "opencv2/core/cuda/functional.hpp"
-#include "opencv2/core/cuda/transform.hpp"
-#include "opencv2/core/cuda/saturate_cast.hpp"
-#include "opencv2/core/cuda/simd_functions.hpp"
+#ifndef HAVE_OPENCV_CUDEV
 
-#include "arithm_func_traits.hpp"
+#error "opencv_cudev is required"
 
-using namespace cv::cuda;
-using namespace cv::cuda::device;
+#else
 
-//////////////////////////////////////////////////////////////////////////
-// min
+#include "opencv2/cudev.hpp"
 
-namespace arithm
+using namespace cv::cudev;
+
+void minMaxMat(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat&, double, Stream& stream, int op);
+
+void minMaxScalar(const GpuMat& src, cv::Scalar value, bool, GpuMat& dst, const GpuMat&, double, Stream& stream, int op);
+
+///////////////////////////////////////////////////////////////////////
+/// minMaxMat
+
+namespace
 {
-    struct VMin4 : binary_function<uint, uint, uint>
+    template <template <typename> class Op, typename T>
+    void minMaxMat_v1(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, Stream& stream)
     {
-        __device__ __forceinline__ uint operator ()(uint a, uint b) const
-        {
-            return vmin4(a, b);
-        }
+        gridTransformBinary(globPtr<T>(src1), globPtr<T>(src2), globPtr<T>(dst), Op<T>(), stream);
+    }
 
-        __host__ __device__ __forceinline__ VMin4() {}
-        __host__ __device__ __forceinline__ VMin4(const VMin4&) {}
-    };
-
-    struct VMin2 : binary_function<uint, uint, uint>
+    struct MinOp2 : binary_function<uint, uint, uint>
     {
         __device__ __forceinline__ uint operator ()(uint a, uint b) const
         {
             return vmin2(a, b);
         }
-
-        __host__ __device__ __forceinline__ VMin2() {}
-        __host__ __device__ __forceinline__ VMin2(const VMin2&) {}
-    };
-}
-
-namespace cv { namespace cuda { namespace device
-{
-    template <> struct TransformFunctorTraits< arithm::VMin4 > : arithm::ArithmFuncTraits<sizeof(uint), sizeof(uint)>
-    {
     };
 
-    template <> struct TransformFunctorTraits< arithm::VMin2 > : arithm::ArithmFuncTraits<sizeof(uint), sizeof(uint)>
-    {
-    };
-
-    template <typename T> struct TransformFunctorTraits< minimum<T> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(T)>
-    {
-    };
-
-    template <typename T> struct TransformFunctorTraits< binder2nd< minimum<T> > > : arithm::ArithmFuncTraits<sizeof(T), sizeof(T)>
-    {
-    };
-}}}
-
-namespace arithm
-{
-    void minMat_v4(PtrStepSz<uint> src1, PtrStepSz<uint> src2, PtrStepSz<uint> dst, cudaStream_t stream)
-    {
-        device::transform(src1, src2, dst, VMin4(), WithOutMask(), stream);
-    }
-
-    void minMat_v2(PtrStepSz<uint> src1, PtrStepSz<uint> src2, PtrStepSz<uint> dst, cudaStream_t stream)
-    {
-        device::transform(src1, src2, dst, VMin2(), WithOutMask(), stream);
-    }
-
-    template <typename T> void minMat(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream)
-    {
-        device::transform((PtrStepSz<T>) src1, (PtrStepSz<T>) src2, (PtrStepSz<T>) dst, minimum<T>(), WithOutMask(), stream);
-    }
-
-    template void minMat<uchar >(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template void minMat<schar >(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template void minMat<ushort>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template void minMat<short >(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template void minMat<int   >(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template void minMat<float >(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template void minMat<double>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-
-    template <typename T> void minScalar(PtrStepSzb src1, double src2, PtrStepSzb dst, cudaStream_t stream)
-    {
-        device::transform((PtrStepSz<T>) src1, (PtrStepSz<T>) dst, cv::cuda::device::bind2nd(minimum<T>(), src2), WithOutMask(), stream);
-    }
-
-    template void minScalar<uchar >(PtrStepSzb src1, double src2, PtrStepSzb dst, cudaStream_t stream);
-    template void minScalar<schar >(PtrStepSzb src1, double src2, PtrStepSzb dst, cudaStream_t stream);
-    template void minScalar<ushort>(PtrStepSzb src1, double src2, PtrStepSzb dst, cudaStream_t stream);
-    template void minScalar<short >(PtrStepSzb src1, double src2, PtrStepSzb dst, cudaStream_t stream);
-    template void minScalar<int   >(PtrStepSzb src1, double src2, PtrStepSzb dst, cudaStream_t stream);
-    template void minScalar<float >(PtrStepSzb src1, double src2, PtrStepSzb dst, cudaStream_t stream);
-    template void minScalar<double>(PtrStepSzb src1, double src2, PtrStepSzb dst, cudaStream_t stream);
-}
-
-//////////////////////////////////////////////////////////////////////////
-// max
-
-namespace arithm
-{
-    struct VMax4 : binary_function<uint, uint, uint>
-    {
-        __device__ __forceinline__ uint operator ()(uint a, uint b) const
-        {
-            return vmax4(a, b);
-        }
-
-        __host__ __device__ __forceinline__ VMax4() {}
-        __host__ __device__ __forceinline__ VMax4(const VMax4&) {}
-    };
-
-    struct VMax2 : binary_function<uint, uint, uint>
+    struct MaxOp2 : binary_function<uint, uint, uint>
     {
         __device__ __forceinline__ uint operator ()(uint a, uint b) const
         {
             return vmax2(a, b);
         }
-
-        __host__ __device__ __forceinline__ VMax2() {}
-        __host__ __device__ __forceinline__ VMax2(const VMax2&) {}
     };
+
+    template <class Op2>
+    void minMaxMat_v2(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, Stream& stream)
+    {
+        const int vcols = src1.cols >> 1;
+
+        GlobPtrSz<uint> src1_ = globPtr((uint*) src1.data, src1.step, src1.rows, vcols);
+        GlobPtrSz<uint> src2_ = globPtr((uint*) src2.data, src2.step, src1.rows, vcols);
+        GlobPtrSz<uint> dst_ = globPtr((uint*) dst.data, dst.step, src1.rows, vcols);
+
+        gridTransformBinary(src1_, src2_, dst_, Op2(), stream);
+    }
+
+    struct MinOp4 : binary_function<uint, uint, uint>
+    {
+        __device__ __forceinline__ uint operator ()(uint a, uint b) const
+        {
+            return vmin4(a, b);
+        }
+    };
+
+    struct MaxOp4 : binary_function<uint, uint, uint>
+    {
+        __device__ __forceinline__ uint operator ()(uint a, uint b) const
+        {
+            return vmax4(a, b);
+        }
+    };
+
+    template <class Op4>
+    void minMaxMat_v4(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, Stream& stream)
+    {
+        const int vcols = src1.cols >> 2;
+
+        GlobPtrSz<uint> src1_ = globPtr((uint*) src1.data, src1.step, src1.rows, vcols);
+        GlobPtrSz<uint> src2_ = globPtr((uint*) src2.data, src2.step, src1.rows, vcols);
+        GlobPtrSz<uint> dst_ = globPtr((uint*) dst.data, dst.step, src1.rows, vcols);
+
+        gridTransformBinary(src1_, src2_, dst_, Op4(), stream);
+    }
 }
 
-namespace cv { namespace cuda { namespace device
+void minMaxMat(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat&, double, Stream& stream, int op)
 {
-    template <> struct TransformFunctorTraits< arithm::VMax4 > : arithm::ArithmFuncTraits<sizeof(uint), sizeof(uint)>
+    typedef void (*func_t)(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, Stream& stream);
+    static const func_t funcs_v1[2][7] =
     {
+        {
+            minMaxMat_v1<minimum, uchar>,
+            minMaxMat_v1<minimum, schar>,
+            minMaxMat_v1<minimum, ushort>,
+            minMaxMat_v1<minimum, short>,
+            minMaxMat_v1<minimum, int>,
+            minMaxMat_v1<minimum, float>,
+            minMaxMat_v1<minimum, double>
+        },
+        {
+            minMaxMat_v1<maximum, uchar>,
+            minMaxMat_v1<maximum, schar>,
+            minMaxMat_v1<maximum, ushort>,
+            minMaxMat_v1<maximum, short>,
+            minMaxMat_v1<maximum, int>,
+            minMaxMat_v1<maximum, float>,
+            minMaxMat_v1<maximum, double>
+        }
     };
 
-    template <> struct TransformFunctorTraits< arithm::VMax2 > : arithm::ArithmFuncTraits<sizeof(uint), sizeof(uint)>
+    static const func_t funcs_v2[2] =
     {
+        minMaxMat_v2<MinOp2>, minMaxMat_v2<MaxOp2>
     };
 
-    template <typename T> struct TransformFunctorTraits< maximum<T> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(T)>
+    static const func_t funcs_v4[2] =
     {
+        minMaxMat_v4<MinOp4>, minMaxMat_v4<MaxOp4>
     };
 
-    template <typename T> struct TransformFunctorTraits< binder2nd< maximum<T> > > : arithm::ArithmFuncTraits<sizeof(T), sizeof(T)>
-    {
-    };
-}}}
+    const int depth = src1.depth();
 
-namespace arithm
-{
-    void maxMat_v4(PtrStepSz<uint> src1, PtrStepSz<uint> src2, PtrStepSz<uint> dst, cudaStream_t stream)
+    CV_DbgAssert( depth <= CV_64F );
+
+    GpuMat src1_ = src1.reshape(1);
+    GpuMat src2_ = src2.reshape(1);
+    GpuMat dst_ = dst.reshape(1);
+
+    if (depth == CV_8U || depth == CV_16U)
     {
-        device::transform(src1, src2, dst, VMax4(), WithOutMask(), stream);
+        const intptr_t src1ptr = reinterpret_cast<intptr_t>(src1_.data);
+        const intptr_t src2ptr = reinterpret_cast<intptr_t>(src2_.data);
+        const intptr_t dstptr = reinterpret_cast<intptr_t>(dst_.data);
+
+        const bool isAllAligned = (src1ptr & 31) == 0 && (src2ptr & 31) == 0 && (dstptr & 31) == 0;
+
+        if (isAllAligned)
+        {
+            if (depth == CV_8U && (src1_.cols & 3) == 0)
+            {
+                funcs_v4[op](src1_, src2_, dst_, stream);
+                return;
+            }
+            else if (depth == CV_16U && (src1_.cols & 1) == 0)
+            {
+                funcs_v2[op](src1_, src2_, dst_, stream);
+                return;
+            }
+        }
     }
 
-    void maxMat_v2(PtrStepSz<uint> src1, PtrStepSz<uint> src2, PtrStepSz<uint> dst, cudaStream_t stream)
-    {
-        device::transform(src1, src2, dst, VMax2(), WithOutMask(), stream);
-    }
+    const func_t func = funcs_v1[op][depth];
 
-    template <typename T> void maxMat(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream)
-    {
-        device::transform((PtrStepSz<T>) src1, (PtrStepSz<T>) src2, (PtrStepSz<T>) dst, maximum<T>(), WithOutMask(), stream);
-    }
-
-    template void maxMat<uchar >(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template void maxMat<schar >(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template void maxMat<ushort>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template void maxMat<short >(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template void maxMat<int   >(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template void maxMat<float >(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template void maxMat<double>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-
-    template <typename T> void maxScalar(PtrStepSzb src1, double src2, PtrStepSzb dst, cudaStream_t stream)
-    {
-        device::transform((PtrStepSz<T>) src1, (PtrStepSz<T>) dst, cv::cuda::device::bind2nd(maximum<T>(), src2), WithOutMask(), stream);
-    }
-
-    template void maxScalar<uchar >(PtrStepSzb src1, double src2, PtrStepSzb dst, cudaStream_t stream);
-    template void maxScalar<schar >(PtrStepSzb src1, double src2, PtrStepSzb dst, cudaStream_t stream);
-    template void maxScalar<ushort>(PtrStepSzb src1, double src2, PtrStepSzb dst, cudaStream_t stream);
-    template void maxScalar<short >(PtrStepSzb src1, double src2, PtrStepSzb dst, cudaStream_t stream);
-    template void maxScalar<int   >(PtrStepSzb src1, double src2, PtrStepSzb dst, cudaStream_t stream);
-    template void maxScalar<float >(PtrStepSzb src1, double src2, PtrStepSzb dst, cudaStream_t stream);
-    template void maxScalar<double>(PtrStepSzb src1, double src2, PtrStepSzb dst, cudaStream_t stream);
+    func(src1_, src2_, dst_, stream);
 }
 
-#endif // CUDA_DISABLER
+///////////////////////////////////////////////////////////////////////
+/// minMaxScalar
+
+namespace
+{
+    template <template <typename> class Op, typename T>
+    void minMaxScalar(const GpuMat& src, double value, GpuMat& dst, Stream& stream)
+    {
+        gridTransformUnary(globPtr<T>(src), globPtr<T>(dst), bind2nd(Op<T>(), cv::saturate_cast<T>(value)), stream);
+    }
+}
+
+void minMaxScalar(const GpuMat& src, cv::Scalar value, bool, GpuMat& dst, const GpuMat&, double, Stream& stream, int op)
+{
+    typedef void (*func_t)(const GpuMat& src, double value, GpuMat& dst, Stream& stream);
+    static const func_t funcs[2][7] =
+    {
+        {
+            minMaxScalar<minimum, uchar>,
+            minMaxScalar<minimum, schar>,
+            minMaxScalar<minimum, ushort>,
+            minMaxScalar<minimum, short>,
+            minMaxScalar<minimum, int>,
+            minMaxScalar<minimum, float>,
+            minMaxScalar<minimum, double>
+        },
+        {
+            minMaxScalar<maximum, uchar>,
+            minMaxScalar<maximum, schar>,
+            minMaxScalar<maximum, ushort>,
+            minMaxScalar<maximum, short>,
+            minMaxScalar<maximum, int>,
+            minMaxScalar<maximum, float>,
+            minMaxScalar<maximum, double>
+        }
+    };
+
+    const int depth = src.depth();
+
+    CV_DbgAssert( depth <= CV_64F );
+    CV_DbgAssert( src.channels() == 1 );
+
+    funcs[op][depth](src, value[0], dst, stream);
+}
+
+#endif
diff --git a/modules/cudaarithm/src/cuda/minmaxloc.cu b/modules/cudaarithm/src/cuda/minmaxloc.cu
index 2374504a88..6f8cc53d6b 100644
--- a/modules/cudaarithm/src/cuda/minmaxloc.cu
+++ b/modules/cudaarithm/src/cuda/minmaxloc.cu
@@ -40,197 +40,88 @@
 //
 //M*/
 
-#if !defined CUDA_DISABLER
+#include "opencv2/opencv_modules.hpp"
 
-#include "opencv2/core/cuda/common.hpp"
-#include "opencv2/core/cuda/vec_traits.hpp"
-#include "opencv2/core/cuda/vec_math.hpp"
-#include "opencv2/core/cuda/functional.hpp"
-#include "opencv2/core/cuda/reduce.hpp"
-#include "opencv2/core/cuda/emulation.hpp"
-#include "opencv2/core/cuda/limits.hpp"
-#include "opencv2/core/cuda/utility.hpp"
+#ifndef HAVE_OPENCV_CUDEV
 
-using namespace cv::cuda;
-using namespace cv::cuda::device;
+#error "opencv_cudev is required"
 
-namespace minMaxLoc
+#else
+
+#include "opencv2/cudaarithm.hpp"
+#include "opencv2/cudev.hpp"
+
+using namespace cv::cudev;
+
+namespace
 {
-    // To avoid shared bank conflicts we convert each value into value of
-    // appropriate type (32 bits minimum)
-    template <typename T> struct MinMaxTypeTraits;
-    template <> struct MinMaxTypeTraits<unsigned char> { typedef int best_type; };
-    template <> struct MinMaxTypeTraits<signed char> { typedef int best_type; };
-    template <> struct MinMaxTypeTraits<unsigned short> { typedef int best_type; };
-    template <> struct MinMaxTypeTraits<short> { typedef int best_type; };
-    template <> struct MinMaxTypeTraits<int> { typedef int best_type; };
-    template <> struct MinMaxTypeTraits<float> { typedef float best_type; };
-    template <> struct MinMaxTypeTraits<double> { typedef double best_type; };
-
-    template <int BLOCK_SIZE, typename T, class Mask>
-    __global__ void kernel_pass_1(const PtrStepSz<T> src, const Mask mask, T* minval, T* maxval, unsigned int* minloc, unsigned int* maxloc, const int twidth, const int theight)
-    {
-        typedef typename MinMaxTypeTraits<T>::best_type work_type;
-
-        __shared__ work_type sminval[BLOCK_SIZE];
-        __shared__ work_type smaxval[BLOCK_SIZE];
-        __shared__ unsigned int sminloc[BLOCK_SIZE];
-        __shared__ unsigned int smaxloc[BLOCK_SIZE];
-
-        const int x0 = blockIdx.x * blockDim.x * twidth + threadIdx.x;
-        const int y0 = blockIdx.y * blockDim.y * theight + threadIdx.y;
-
-        const int tid = threadIdx.y * blockDim.x + threadIdx.x;
-        const int bid = blockIdx.y * gridDim.x + blockIdx.x;
-
-        work_type mymin = numeric_limits<work_type>::max();
-        work_type mymax = -numeric_limits<work_type>::max();
-        unsigned int myminloc = 0;
-        unsigned int mymaxloc = 0;
-
-        for (int i = 0, y = y0; i < theight && y < src.rows; ++i, y += blockDim.y)
-        {
-            const T* ptr = src.ptr(y);
-
-            for (int j = 0, x = x0; j < twidth && x < src.cols; ++j, x += blockDim.x)
-            {
-                if (mask(y, x))
-                {
-                    const work_type srcVal = ptr[x];
-
-                    if (srcVal < mymin)
-                    {
-                        mymin = srcVal;
-                        myminloc = y * src.cols + x;
-                    }
-
-                    if (srcVal > mymax)
-                    {
-                        mymax = srcVal;
-                        mymaxloc = y * src.cols + x;
-                    }
-                }
-            }
-        }
-
-        reduceKeyVal<BLOCK_SIZE>(smem_tuple(sminval, smaxval), thrust::tie(mymin, mymax),
-                                 smem_tuple(sminloc, smaxloc), thrust::tie(myminloc, mymaxloc),
-                                 tid,
-                                 thrust::make_tuple(less<work_type>(), greater<work_type>()));
-
-        if (tid == 0)
-        {
-            minval[bid] = (T) mymin;
-            maxval[bid] = (T) mymax;
-            minloc[bid] = myminloc;
-            maxloc[bid] = mymaxloc;
-        }
-    }
-    template <int BLOCK_SIZE, typename T>
-    __global__ void kernel_pass_2(T* minval, T* maxval, unsigned int* minloc, unsigned int* maxloc, int count)
-    {
-        typedef typename MinMaxTypeTraits<T>::best_type work_type;
-
-        __shared__ work_type sminval[BLOCK_SIZE];
-        __shared__ work_type smaxval[BLOCK_SIZE];
-        __shared__ unsigned int sminloc[BLOCK_SIZE];
-        __shared__ unsigned int smaxloc[BLOCK_SIZE];
-
-        unsigned int idx = ::min(threadIdx.x, count - 1);
-
-        work_type mymin = minval[idx];
-        work_type mymax = maxval[idx];
-        unsigned int myminloc = minloc[idx];
-        unsigned int mymaxloc = maxloc[idx];
-
-        reduceKeyVal<BLOCK_SIZE>(smem_tuple(sminval, smaxval), thrust::tie(mymin, mymax),
-                                 smem_tuple(sminloc, smaxloc), thrust::tie(myminloc, mymaxloc),
-                                 threadIdx.x,
-                                 thrust::make_tuple(less<work_type>(), greater<work_type>()));
-
-        if (threadIdx.x == 0)
-        {
-            minval[0] = (T) mymin;
-            maxval[0] = (T) mymax;
-            minloc[0] = myminloc;
-            maxloc[0] = mymaxloc;
-        }
-    }
-
-    const int threads_x = 32;
-    const int threads_y = 8;
-
-    void getLaunchCfg(int cols, int rows, dim3& block, dim3& grid)
-    {
-        block = dim3(threads_x, threads_y);
-
-        grid = dim3(divUp(cols, block.x * block.y),
-                    divUp(rows, block.y * block.x));
-
-        grid.x = ::min(grid.x, block.x);
-        grid.y = ::min(grid.y, block.y);
-    }
-
-    void getBufSize(int cols, int rows, size_t elem_size, int& b1cols, int& b1rows, int& b2cols, int& b2rows)
-    {
-        dim3 block, grid;
-        getLaunchCfg(cols, rows, block, grid);
-
-        // For values
-        b1cols = (int)(grid.x * grid.y * elem_size);
-        b1rows = 2;
-
-        // For locations
-        b2cols = grid.x * grid.y * sizeof(int);
-        b2rows = 2;
-    }
-
     template <typename T>
-    void run(const PtrStepSzb src, const PtrStepb mask, double* minval, double* maxval, int* minloc, int* maxloc, PtrStepb valbuf, PtrStep<unsigned int> locbuf)
+    void minMaxLocImpl(const GpuMat& _src, const GpuMat& mask, GpuMat& _valBuf, GpuMat& _locBuf, double* minVal, double* maxVal, cv::Point* minLoc, cv::Point* maxLoc)
     {
-        dim3 block, grid;
-        getLaunchCfg(src.cols, src.rows, block, grid);
+        typedef typename SelectIf<
+                TypesEquals<T, double>::value,
+                double,
+                typename SelectIf<TypesEquals<T, float>::value, float, int>::type
+                >::type work_type;
 
-        const int twidth = divUp(divUp(src.cols, grid.x), block.x);
-        const int theight = divUp(divUp(src.rows, grid.y), block.y);
+        const GpuMat_<T>& src = (const GpuMat_<T>&) _src;
+        GpuMat_<work_type>& valBuf = (GpuMat_<work_type>&) _valBuf;
+        GpuMat_<int>& locBuf = (GpuMat_<int>&) _locBuf;
 
-        T* minval_buf = (T*) valbuf.ptr(0);
-        T* maxval_buf = (T*) valbuf.ptr(1);
-        unsigned int* minloc_buf = locbuf.ptr(0);
-        unsigned int* maxloc_buf = locbuf.ptr(1);
-
-        if (mask.data)
-            kernel_pass_1<threads_x * threads_y><<<grid, block>>>((PtrStepSz<T>) src, SingleMask(mask), minval_buf, maxval_buf, minloc_buf, maxloc_buf, twidth, theight);
+        if (mask.empty())
+            gridMinMaxLoc(src, valBuf, locBuf);
         else
-            kernel_pass_1<threads_x * threads_y><<<grid, block>>>((PtrStepSz<T>) src, WithOutMask(), minval_buf, maxval_buf, minloc_buf, maxloc_buf, twidth, theight);
+            gridMinMaxLoc(src, valBuf, locBuf, globPtr<uchar>(mask));
 
-        cudaSafeCall( cudaGetLastError() );
+        cv::Mat_<work_type> h_valBuf;
+        cv::Mat_<int> h_locBuf;
 
-        kernel_pass_2<threads_x * threads_y><<<1, threads_x * threads_y>>>(minval_buf, maxval_buf, minloc_buf, maxloc_buf, grid.x * grid.y);
-        cudaSafeCall( cudaGetLastError() );
+        valBuf.download(h_valBuf);
+        locBuf.download(h_locBuf);
 
-        cudaSafeCall( cudaDeviceSynchronize() );
+        if (minVal)
+            *minVal = h_valBuf(0, 0);
 
-        T minval_, maxval_;
-        cudaSafeCall( cudaMemcpy(&minval_, minval_buf, sizeof(T), cudaMemcpyDeviceToHost) );
-        cudaSafeCall( cudaMemcpy(&maxval_, maxval_buf, sizeof(T), cudaMemcpyDeviceToHost) );
-        *minval = minval_;
-        *maxval = maxval_;
+        if (maxVal)
+            *maxVal = h_valBuf(1, 0);
 
-        unsigned int minloc_, maxloc_;
-        cudaSafeCall( cudaMemcpy(&minloc_, minloc_buf, sizeof(unsigned int), cudaMemcpyDeviceToHost) );
-        cudaSafeCall( cudaMemcpy(&maxloc_, maxloc_buf, sizeof(unsigned int), cudaMemcpyDeviceToHost) );
-        minloc[1] = minloc_ / src.cols; minloc[0] = minloc_ - minloc[1] * src.cols;
-        maxloc[1] = maxloc_ / src.cols; maxloc[0] = maxloc_ - maxloc[1] * src.cols;
+        if (minLoc)
+        {
+            const int idx = h_locBuf(0, 0);
+            *minLoc = cv::Point(idx % src.cols, idx / src.cols);
+        }
+
+        if (maxLoc)
+        {
+            const int idx = h_locBuf(1, 0);
+            *maxLoc = cv::Point(idx % src.cols, idx / src.cols);
+        }
     }
-
-    template void run<unsigned char >(const PtrStepSzb src, const PtrStepb mask, double* minval, double* maxval, int* minloc, int* maxloc, PtrStepb valbuf, PtrStep<unsigned int> locbuf);
-    template void run<signed char >(const PtrStepSzb src, const PtrStepb mask, double* minval, double* maxval, int* minloc, int* maxloc, PtrStepb valbuf, PtrStep<unsigned int> locbuf);
-    template void run<unsigned short>(const PtrStepSzb src, const PtrStepb mask, double* minval, double* maxval, int* minloc, int* maxloc, PtrStepb valbuf, PtrStep<unsigned int> locbuf);
-    template void run<short >(const PtrStepSzb src, const PtrStepb mask, double* minval, double* maxval, int* minloc, int* maxloc, PtrStepb valbuf, PtrStep<unsigned int> locbuf);
-    template void run<int   >(const PtrStepSzb src, const PtrStepb mask, double* minval, double* maxval, int* minloc, int* maxloc, PtrStepb valbuf, PtrStep<unsigned int> locbuf);
-    template void run<float >(const PtrStepSzb src, const PtrStepb mask, double* minval, double* maxval, int* minloc, int* maxloc, PtrStepb valbuf, PtrStep<unsigned int> locbuf);
-    template void run<double>(const PtrStepSzb src, const PtrStepb mask, double* minval, double* maxval, int* minloc, int* maxloc, PtrStepb valbuf, PtrStep<unsigned int> locbuf);
 }
 
-#endif // CUDA_DISABLER
+void cv::cuda::minMaxLoc(InputArray _src, double* minVal, double* maxVal, Point* minLoc, Point* maxLoc, InputArray _mask, GpuMat& valBuf, GpuMat& locBuf)
+{
+    typedef void (*func_t)(const GpuMat& _src, const GpuMat& mask, GpuMat& _valBuf, GpuMat& _locBuf, double* minVal, double* maxVal, cv::Point* minLoc, cv::Point* maxLoc);
+    static const func_t funcs[] =
+    {
+        minMaxLocImpl<uchar>,
+        minMaxLocImpl<schar>,
+        minMaxLocImpl<ushort>,
+        minMaxLocImpl<short>,
+        minMaxLocImpl<int>,
+        minMaxLocImpl<float>,
+        minMaxLocImpl<double>
+    };
+
+    GpuMat src = _src.getGpuMat();
+    GpuMat mask = _mask.getGpuMat();
+
+    CV_Assert( src.channels() == 1 );
+    CV_DbgAssert( mask.empty() || (mask.size() == src.size() && mask.type() == CV_8U) );
+
+    const func_t func = funcs[src.depth()];
+
+    func(src, mask, valBuf, locBuf, minVal, maxVal, minLoc, maxLoc);
+}
+
+#endif
diff --git a/modules/cudaarithm/src/cuda/mul_mat.cu b/modules/cudaarithm/src/cuda/mul_mat.cu
index dda596341a..6ea7065573 100644
--- a/modules/cudaarithm/src/cuda/mul_mat.cu
+++ b/modules/cudaarithm/src/cuda/mul_mat.cu
@@ -40,172 +40,185 @@
 //
 //M*/
 
-#if !defined CUDA_DISABLER
+#include "opencv2/opencv_modules.hpp"
 
-#include "opencv2/core/cuda/common.hpp"
-#include "opencv2/core/cuda/functional.hpp"
-#include "opencv2/core/cuda/transform.hpp"
-#include "opencv2/core/cuda/saturate_cast.hpp"
-#include "opencv2/core/cuda/simd_functions.hpp"
+#ifndef HAVE_OPENCV_CUDEV
 
-#include "arithm_func_traits.hpp"
+#error "opencv_cudev is required"
 
-using namespace cv::cuda;
-using namespace cv::cuda::device;
+#else
 
-namespace arithm
+#include "opencv2/cudev.hpp"
+
+using namespace cv::cudev;
+
+void mulMat(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat&, double scale, Stream& stream, int);
+void mulMat_8uc4_32f(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, Stream& stream);
+void mulMat_16sc4_32f(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, Stream& stream);
+
+namespace
 {
-    struct Mul_8uc4_32f : binary_function<uint, float, uint>
-    {
-        __device__ __forceinline__ uint operator ()(uint a, float b) const
-        {
-            uint res = 0;
-
-            res |= (saturate_cast<uchar>((0xffu & (a      )) * b)      );
-            res |= (saturate_cast<uchar>((0xffu & (a >>  8)) * b) <<  8);
-            res |= (saturate_cast<uchar>((0xffu & (a >> 16)) * b) << 16);
-            res |= (saturate_cast<uchar>((0xffu & (a >> 24)) * b) << 24);
-
-            return res;
-        }
-
-        __host__ __device__ __forceinline__ Mul_8uc4_32f() {}
-        __host__ __device__ __forceinline__ Mul_8uc4_32f(const Mul_8uc4_32f&) {}
-    };
-
-    struct Mul_16sc4_32f : binary_function<short4, float, short4>
-    {
-        __device__ __forceinline__ short4 operator ()(short4 a, float b) const
-        {
-            return make_short4(saturate_cast<short>(a.x * b), saturate_cast<short>(a.y * b),
-                               saturate_cast<short>(a.z * b), saturate_cast<short>(a.w * b));
-        }
-
-        __host__ __device__ __forceinline__ Mul_16sc4_32f() {}
-        __host__ __device__ __forceinline__ Mul_16sc4_32f(const Mul_16sc4_32f&) {}
-    };
-
-    template <typename T, typename D> struct Mul : binary_function<T, T, D>
+    template <typename T, typename D> struct MulOp : binary_function<T, T, D>
     {
         __device__ __forceinline__ D operator ()(T a, T b) const
         {
             return saturate_cast<D>(a * b);
         }
-
-        __host__ __device__ __forceinline__ Mul() {}
-        __host__ __device__ __forceinline__ Mul(const Mul&) {}
     };
 
-    template <typename T, typename S, typename D> struct MulScale : binary_function<T, T, D>
+    template <typename T, typename S, typename D> struct MulScaleOp : binary_function<T, T, D>
     {
         S scale;
 
-        __host__ explicit MulScale(S scale_) : scale(scale_) {}
-
         __device__ __forceinline__ D operator ()(T a, T b) const
         {
             return saturate_cast<D>(scale * a * b);
         }
     };
-}
 
-namespace cv { namespace cuda { namespace device
-{
-    template <> struct TransformFunctorTraits<arithm::Mul_8uc4_32f> : arithm::ArithmFuncTraits<sizeof(uint), sizeof(uint)>
+    template <typename ScalarDepth> struct TransformPolicy : DefaultTransformPolicy
     {
     };
-
-    template <typename T, typename D> struct TransformFunctorTraits< arithm::Mul<T, D> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(D)>
+    template <> struct TransformPolicy<double> : DefaultTransformPolicy
     {
+        enum {
+            shift = 1
+        };
     };
 
-    template <typename T, typename S, typename D> struct TransformFunctorTraits< arithm::MulScale<T, S, D> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(D)>
-    {
-    };
-}}}
-
-namespace arithm
-{
-    void mulMat_8uc4_32f(PtrStepSz<uint> src1, PtrStepSzf src2, PtrStepSz<uint> dst, cudaStream_t stream)
-    {
-        device::transform(src1, src2, dst, Mul_8uc4_32f(), WithOutMask(), stream);
-    }
-
-    void mulMat_16sc4_32f(PtrStepSz<short4> src1, PtrStepSzf src2, PtrStepSz<short4> dst, cudaStream_t stream)
-    {
-        device::transform(src1, src2, dst, Mul_16sc4_32f(), WithOutMask(), stream);
-    }
-
     template <typename T, typename S, typename D>
-    void mulMat(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream)
+    void mulMatImpl(const GpuMat& src1, const GpuMat& src2, const GpuMat& dst, double scale, Stream& stream)
     {
         if (scale == 1)
         {
-            Mul<T, D> op;
-            device::transform((PtrStepSz<T>) src1, (PtrStepSz<T>) src2, (PtrStepSz<D>) dst, op, WithOutMask(), stream);
+            MulOp<T, D> op;
+            gridTransformBinary_< TransformPolicy<S> >(globPtr<T>(src1), globPtr<T>(src2), globPtr<D>(dst), op, stream);
         }
         else
         {
-            MulScale<T, S, D> op(static_cast<S>(scale));
-            device::transform((PtrStepSz<T>) src1, (PtrStepSz<T>) src2, (PtrStepSz<D>) dst, op, WithOutMask(), stream);
+            MulScaleOp<T, S, D> op;
+            op.scale = static_cast<S>(scale);
+            gridTransformBinary_< TransformPolicy<S> >(globPtr<T>(src1), globPtr<T>(src2), globPtr<D>(dst), op, stream);
         }
     }
-
-    template void mulMat<uchar, float, uchar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void mulMat<uchar, float, schar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void mulMat<uchar, float, ushort>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void mulMat<uchar, float, short>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void mulMat<uchar, float, int>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void mulMat<uchar, float, float>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void mulMat<uchar, double, double>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-
-    template void mulMat<schar, float, uchar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void mulMat<schar, float, schar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void mulMat<schar, float, ushort>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void mulMat<schar, float, short>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void mulMat<schar, float, int>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void mulMat<schar, float, float>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void mulMat<schar, double, double>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-
-    //template void mulMat<ushort, float, uchar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    //template void mulMat<ushort, float, schar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void mulMat<ushort, float, ushort>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void mulMat<ushort, float, short>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void mulMat<ushort, float, int>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void mulMat<ushort, float, float>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void mulMat<ushort, double, double>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-
-    //template void mulMat<short, float, uchar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    //template void mulMat<short, float, schar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void mulMat<short, float, ushort>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void mulMat<short, float, short>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void mulMat<short, float, int>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void mulMat<short, float, float>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void mulMat<short, double, double>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-
-    //template void mulMat<int, float, uchar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    //template void mulMat<int, float, schar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    //template void mulMat<int, float, ushort>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    //template void mulMat<int, float, short>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void mulMat<int, float, int>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void mulMat<int, float, float>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void mulMat<int, double, double>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-
-    //template void mulMat<float, float, uchar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    //template void mulMat<float, float, schar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    //template void mulMat<float, float, ushort>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    //template void mulMat<float, float, short>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    //template void mulMat<float, float, int>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void mulMat<float, float, float>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void mulMat<float, double, double>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-
-    //template void mulMat<double, double, uchar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    //template void mulMat<double, double, schar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    //template void mulMat<double, double, ushort>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    //template void mulMat<double, double, short>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    //template void mulMat<double, double, int>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    //template void mulMat<double, double, float>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    template void mulMat<double, double, double>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
 }
 
-#endif // CUDA_DISABLER
+void mulMat(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat&, double scale, Stream& stream, int)
+{
+    typedef void (*func_t)(const GpuMat& src1, const GpuMat& src2, const GpuMat& dst, double scale, Stream& stream);
+    static const func_t funcs[7][7] =
+    {
+        {
+            mulMatImpl<uchar, float, uchar>,
+            mulMatImpl<uchar, float, schar>,
+            mulMatImpl<uchar, float, ushort>,
+            mulMatImpl<uchar, float, short>,
+            mulMatImpl<uchar, float, int>,
+            mulMatImpl<uchar, float, float>,
+            mulMatImpl<uchar, double, double>
+        },
+        {
+            mulMatImpl<schar, float, uchar>,
+            mulMatImpl<schar, float, schar>,
+            mulMatImpl<schar, float, ushort>,
+            mulMatImpl<schar, float, short>,
+            mulMatImpl<schar, float, int>,
+            mulMatImpl<schar, float, float>,
+            mulMatImpl<schar, double, double>
+        },
+        {
+            0 /*mulMatImpl<ushort, float, uchar>*/,
+            0 /*mulMatImpl<ushort, float, schar>*/,
+            mulMatImpl<ushort, float, ushort>,
+            mulMatImpl<ushort, float, short>,
+            mulMatImpl<ushort, float, int>,
+            mulMatImpl<ushort, float, float>,
+            mulMatImpl<ushort, double, double>
+        },
+        {
+            0 /*mulMatImpl<short, float, uchar>*/,
+            0 /*mulMatImpl<short, float, schar>*/,
+            mulMatImpl<short, float, ushort>,
+            mulMatImpl<short, float, short>,
+            mulMatImpl<short, float, int>,
+            mulMatImpl<short, float, float>,
+            mulMatImpl<short, double, double>
+        },
+        {
+            0 /*mulMatImpl<int, float, uchar>*/,
+            0 /*mulMatImpl<int, float, schar>*/,
+            0 /*mulMatImpl<int, float, ushort>*/,
+            0 /*mulMatImpl<int, float, short>*/,
+            mulMatImpl<int, float, int>,
+            mulMatImpl<int, float, float>,
+            mulMatImpl<int, double, double>
+        },
+        {
+            0 /*mulMatImpl<float, float, uchar>*/,
+            0 /*mulMatImpl<float, float, schar>*/,
+            0 /*mulMatImpl<float, float, ushort>*/,
+            0 /*mulMatImpl<float, float, short>*/,
+            0 /*mulMatImpl<float, float, int>*/,
+            mulMatImpl<float, float, float>,
+            mulMatImpl<float, double, double>
+        },
+        {
+            0 /*mulMatImpl<double, double, uchar>*/,
+            0 /*mulMatImpl<double, double, schar>*/,
+            0 /*mulMatImpl<double, double, ushort>*/,
+            0 /*mulMatImpl<double, double, short>*/,
+            0 /*mulMatImpl<double, double, int>*/,
+            0 /*mulMatImpl<double, double, float>*/,
+            mulMatImpl<double, double, double>
+        }
+    };
+
+    const int sdepth = src1.depth();
+    const int ddepth = dst.depth();
+
+    CV_DbgAssert( sdepth <= CV_64F && ddepth <= CV_64F );
+
+    GpuMat src1_ = src1.reshape(1);
+    GpuMat src2_ = src2.reshape(1);
+    GpuMat dst_ = dst.reshape(1);
+
+    const func_t func = funcs[sdepth][ddepth];
+
+    if (!func)
+        CV_Error(cv::Error::StsUnsupportedFormat, "Unsupported combination of source and destination types");
+
+    func(src1_, src2_, dst_, scale, stream);
+}
+
+namespace
+{
+    template <typename T>
+    struct MulOpSpecial : binary_function<T, float, T>
+    {
+        __device__ __forceinline__ T operator ()(const T& a, float b) const
+        {
+            typedef typename VecTraits<T>::elem_type elem_type;
+
+            T res;
+
+            res.x = saturate_cast<elem_type>(a.x * b);
+            res.y = saturate_cast<elem_type>(a.y * b);
+            res.z = saturate_cast<elem_type>(a.z * b);
+            res.w = saturate_cast<elem_type>(a.w * b);
+
+            return res;
+        }
+    };
+}
+
+void mulMat_8uc4_32f(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, Stream& stream)
+{
+    gridTransformBinary(globPtr<uchar4>(src1), globPtr<float>(src2), globPtr<uchar4>(dst), MulOpSpecial<uchar4>(), stream);
+}
+
+void mulMat_16sc4_32f(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, Stream& stream)
+{
+    gridTransformBinary(globPtr<short4>(src1), globPtr<float>(src2), globPtr<short4>(dst), MulOpSpecial<short4>(), stream);
+}
+
+#endif
diff --git a/modules/cudaarithm/src/cuda/mul_scalar.cu b/modules/cudaarithm/src/cuda/mul_scalar.cu
index 341cd9bc44..f27ef26ddd 100644
--- a/modules/cudaarithm/src/cuda/mul_scalar.cu
+++ b/modules/cudaarithm/src/cuda/mul_scalar.cu
@@ -40,105 +40,143 @@
 //
 //M*/
 
-#if !defined CUDA_DISABLER
+#include "opencv2/opencv_modules.hpp"
 
-#include "opencv2/core/cuda/common.hpp"
-#include "opencv2/core/cuda/functional.hpp"
-#include "opencv2/core/cuda/transform.hpp"
-#include "opencv2/core/cuda/saturate_cast.hpp"
-#include "opencv2/core/cuda/simd_functions.hpp"
+#ifndef HAVE_OPENCV_CUDEV
 
-#include "arithm_func_traits.hpp"
+#error "opencv_cudev is required"
 
-using namespace cv::cuda;
-using namespace cv::cuda::device;
+#else
 
-namespace arithm
+#include "opencv2/cudev.hpp"
+
+using namespace cv::cudev;
+
+void mulScalar(const GpuMat& src, cv::Scalar val, bool, GpuMat& dst, const GpuMat& mask, double scale, Stream& stream, int);
+
+namespace
 {
-    template <typename T, typename S, typename D> struct MulScalar : unary_function<T, D>
+    template <typename SrcType, typename ScalarType, typename DstType> struct MulScalarOp : unary_function<SrcType, DstType>
     {
-        S val;
+        ScalarType val;
 
-        __host__ explicit MulScalar(S val_) : val(val_) {}
-
-        __device__ __forceinline__ D operator ()(T a) const
+        __device__ __forceinline__ DstType operator ()(SrcType a) const
         {
-            return saturate_cast<D>(a * val);
+            return saturate_cast<DstType>(saturate_cast<ScalarType>(a) * val);
         }
     };
-}
 
-namespace cv { namespace cuda { namespace device
-{
-    template <typename T, typename S, typename D> struct TransformFunctorTraits< arithm::MulScalar<T, S, D> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(D)>
+    template <typename ScalarDepth> struct TransformPolicy : DefaultTransformPolicy
     {
     };
-}}}
-
-namespace arithm
-{
-    template <typename T, typename S, typename D>
-    void mulScalar(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream)
+    template <> struct TransformPolicy<double> : DefaultTransformPolicy
     {
-        MulScalar<T, S, D> op(static_cast<S>(val));
-        device::transform((PtrStepSz<T>) src1, (PtrStepSz<D>) dst, op, WithOutMask(), stream);
+        enum {
+            shift = 1
+        };
+    };
+
+    template <typename SrcType, typename ScalarDepth, typename DstType>
+    void mulScalarImpl(const GpuMat& src, cv::Scalar value, GpuMat& dst, Stream& stream)
+    {
+        typedef typename MakeVec<ScalarDepth, VecTraits<SrcType>::cn>::type ScalarType;
+
+        cv::Scalar_<ScalarDepth> value_ = value;
+
+        MulScalarOp<SrcType, ScalarType, DstType> op;
+        op.val = VecTraits<ScalarType>::make(value_.val);
+
+        gridTransformUnary_< TransformPolicy<ScalarDepth> >(globPtr<SrcType>(src), globPtr<DstType>(dst), op, stream);
     }
-
-    template void mulScalar<uchar, float, uchar>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-    template void mulScalar<uchar, float, schar>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-    template void mulScalar<uchar, float, ushort>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-    template void mulScalar<uchar, float, short>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-    template void mulScalar<uchar, float, int>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-    template void mulScalar<uchar, float, float>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-    template void mulScalar<uchar, double, double>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-
-    template void mulScalar<schar, float, uchar>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-    template void mulScalar<schar, float, schar>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-    template void mulScalar<schar, float, ushort>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-    template void mulScalar<schar, float, short>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-    template void mulScalar<schar, float, int>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-    template void mulScalar<schar, float, float>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-    template void mulScalar<schar, double, double>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-
-    //template void mulScalar<ushort, float, uchar>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-    //template void mulScalar<ushort, float, schar>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-    template void mulScalar<ushort, float, ushort>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-    template void mulScalar<ushort, float, short>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-    template void mulScalar<ushort, float, int>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-    template void mulScalar<ushort, float, float>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-    template void mulScalar<ushort, double, double>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-
-    //template void mulScalar<short, float, uchar>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-    //template void mulScalar<short, float, schar>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-    template void mulScalar<short, float, ushort>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-    template void mulScalar<short, float, short>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-    template void mulScalar<short, float, int>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-    template void mulScalar<short, float, float>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-    template void mulScalar<short, double, double>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-
-    //template void mulScalar<int, float, uchar>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-    //template void mulScalar<int, float, schar>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-    //template void mulScalar<int, float, ushort>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-    //template void mulScalar<int, float, short>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-    template void mulScalar<int, float, int>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-    template void mulScalar<int, float, float>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-    template void mulScalar<int, double, double>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-
-    //template void mulScalar<float, float, uchar>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-    //template void mulScalar<float, float, schar>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-    //template void mulScalar<float, float, ushort>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-    //template void mulScalar<float, float, short>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-    //template void mulScalar<float, float, int>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-    template void mulScalar<float, float, float>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-    template void mulScalar<float, double, double>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-
-    //template void mulScalar<double, double, uchar>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-    //template void mulScalar<double, double, schar>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-    //template void mulScalar<double, double, ushort>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-    //template void mulScalar<double, double, short>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-    //template void mulScalar<double, double, int>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-    //template void mulScalar<double, double, float>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-    template void mulScalar<double, double, double>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
 }
 
-#endif // CUDA_DISABLER
+void mulScalar(const GpuMat& src, cv::Scalar val, bool, GpuMat& dst, const GpuMat&, double scale, Stream& stream, int)
+{
+    typedef void (*func_t)(const GpuMat& src, cv::Scalar val, GpuMat& dst, Stream& stream);
+    static const func_t funcs[7][7][4] =
+    {
+        {
+            {mulScalarImpl<uchar, float, uchar>, mulScalarImpl<uchar2, float, uchar2>, mulScalarImpl<uchar3, float, uchar3>, mulScalarImpl<uchar4, float, uchar4>},
+            {mulScalarImpl<uchar, float, schar>, mulScalarImpl<uchar2, float, char2>, mulScalarImpl<uchar3, float, char3>, mulScalarImpl<uchar4, float, char4>},
+            {mulScalarImpl<uchar, float, ushort>, mulScalarImpl<uchar2, float, ushort2>, mulScalarImpl<uchar3, float, ushort3>, mulScalarImpl<uchar4, float, ushort4>},
+            {mulScalarImpl<uchar, float, short>, mulScalarImpl<uchar2, float, short2>, mulScalarImpl<uchar3, float, short3>, mulScalarImpl<uchar4, float, short4>},
+            {mulScalarImpl<uchar, float, int>, mulScalarImpl<uchar2, float, int2>, mulScalarImpl<uchar3, float, int3>, mulScalarImpl<uchar4, float, int4>},
+            {mulScalarImpl<uchar, float, float>, mulScalarImpl<uchar2, float, float2>, mulScalarImpl<uchar3, float, float3>, mulScalarImpl<uchar4, float, float4>},
+            {mulScalarImpl<uchar, double, double>, mulScalarImpl<uchar2, double, double2>, mulScalarImpl<uchar3, double, double3>, mulScalarImpl<uchar4, double, double4>}
+        },
+        {
+            {mulScalarImpl<schar, float, uchar>, mulScalarImpl<char2, float, uchar2>, mulScalarImpl<char3, float, uchar3>, mulScalarImpl<char4, float, uchar4>},
+            {mulScalarImpl<schar, float, schar>, mulScalarImpl<char2, float, char2>, mulScalarImpl<char3, float, char3>, mulScalarImpl<char4, float, char4>},
+            {mulScalarImpl<schar, float, ushort>, mulScalarImpl<char2, float, ushort2>, mulScalarImpl<char3, float, ushort3>, mulScalarImpl<char4, float, ushort4>},
+            {mulScalarImpl<schar, float, short>, mulScalarImpl<char2, float, short2>, mulScalarImpl<char3, float, short3>, mulScalarImpl<char4, float, short4>},
+            {mulScalarImpl<schar, float, int>, mulScalarImpl<char2, float, int2>, mulScalarImpl<char3, float, int3>, mulScalarImpl<char4, float, int4>},
+            {mulScalarImpl<schar, float, float>, mulScalarImpl<char2, float, float2>, mulScalarImpl<char3, float, float3>, mulScalarImpl<char4, float, float4>},
+            {mulScalarImpl<schar, double, double>, mulScalarImpl<char2, double, double2>, mulScalarImpl<char3, double, double3>, mulScalarImpl<char4, double, double4>}
+        },
+        {
+            {0 /*mulScalarImpl<ushort, float, uchar>*/, 0 /*mulScalarImpl<ushort2, float, uchar2>*/, 0 /*mulScalarImpl<ushort3, float, uchar3>*/, 0 /*mulScalarImpl<ushort4, float, uchar4>*/},
+            {0 /*mulScalarImpl<ushort, float, schar>*/, 0 /*mulScalarImpl<ushort2, float, char2>*/, 0 /*mulScalarImpl<ushort3, float, char3>*/, 0 /*mulScalarImpl<ushort4, float, char4>*/},
+            {mulScalarImpl<ushort, float, ushort>, mulScalarImpl<ushort2, float, ushort2>, mulScalarImpl<ushort3, float, ushort3>, mulScalarImpl<ushort4, float, ushort4>},
+            {mulScalarImpl<ushort, float, short>, mulScalarImpl<ushort2, float, short2>, mulScalarImpl<ushort3, float, short3>, mulScalarImpl<ushort4, float, short4>},
+            {mulScalarImpl<ushort, float, int>, mulScalarImpl<ushort2, float, int2>, mulScalarImpl<ushort3, float, int3>, mulScalarImpl<ushort4, float, int4>},
+            {mulScalarImpl<ushort, float, float>, mulScalarImpl<ushort2, float, float2>, mulScalarImpl<ushort3, float, float3>, mulScalarImpl<ushort4, float, float4>},
+            {mulScalarImpl<ushort, double, double>, mulScalarImpl<ushort2, double, double2>, mulScalarImpl<ushort3, double, double3>, mulScalarImpl<ushort4, double, double4>}
+        },
+        {
+            {0 /*mulScalarImpl<short, float, uchar>*/, 0 /*mulScalarImpl<short2, float, uchar2>*/, 0 /*mulScalarImpl<short3, float, uchar3>*/, 0 /*mulScalarImpl<short4, float, uchar4>*/},
+            {0 /*mulScalarImpl<short, float, schar>*/, 0 /*mulScalarImpl<short2, float, char2>*/, 0 /*mulScalarImpl<short3, float, char3>*/, 0 /*mulScalarImpl<short4, float, char4>*/},
+            {mulScalarImpl<short, float, ushort>, mulScalarImpl<short2, float, ushort2>, mulScalarImpl<short3, float, ushort3>, mulScalarImpl<short4, float, ushort4>},
+            {mulScalarImpl<short, float, short>, mulScalarImpl<short2, float, short2>, mulScalarImpl<short3, float, short3>, mulScalarImpl<short4, float, short4>},
+            {mulScalarImpl<short, float, int>, mulScalarImpl<short2, float, int2>, mulScalarImpl<short3, float, int3>, mulScalarImpl<short4, float, int4>},
+            {mulScalarImpl<short, float, float>, mulScalarImpl<short2, float, float2>, mulScalarImpl<short3, float, float3>, mulScalarImpl<short4, float, float4>},
+            {mulScalarImpl<short, double, double>, mulScalarImpl<short2, double, double2>, mulScalarImpl<short3, double, double3>, mulScalarImpl<short4, double, double4>}
+        },
+        {
+            {0 /*mulScalarImpl<int, float, uchar>*/, 0 /*mulScalarImpl<int2, float, uchar2>*/, 0 /*mulScalarImpl<int3, float, uchar3>*/, 0 /*mulScalarImpl<int4, float, uchar4>*/},
+            {0 /*mulScalarImpl<int, float, schar>*/, 0 /*mulScalarImpl<int2, float, char2>*/, 0 /*mulScalarImpl<int3, float, char3>*/, 0 /*mulScalarImpl<int4, float, char4>*/},
+            {0 /*mulScalarImpl<int, float, ushort>*/, 0 /*mulScalarImpl<int2, float, ushort2>*/, 0 /*mulScalarImpl<int3, float, ushort3>*/, 0 /*mulScalarImpl<int4, float, ushort4>*/},
+            {0 /*mulScalarImpl<int, float, short>*/, 0 /*mulScalarImpl<int2, float, short2>*/, 0 /*mulScalarImpl<int3, float, short3>*/, 0 /*mulScalarImpl<int4, float, short4>*/},
+            {mulScalarImpl<int, float, int>, mulScalarImpl<int2, float, int2>, mulScalarImpl<int3, float, int3>, mulScalarImpl<int4, float, int4>},
+            {mulScalarImpl<int, float, float>, mulScalarImpl<int2, float, float2>, mulScalarImpl<int3, float, float3>, mulScalarImpl<int4, float, float4>},
+            {mulScalarImpl<int, double, double>, mulScalarImpl<int2, double, double2>, mulScalarImpl<int3, double, double3>, mulScalarImpl<int4, double, double4>}
+        },
+        {
+            {0 /*mulScalarImpl<float, float, uchar>*/, 0 /*mulScalarImpl<float2, float, uchar2>*/, 0 /*mulScalarImpl<float3, float, uchar3>*/, 0 /*mulScalarImpl<float4, float, uchar4>*/},
+            {0 /*mulScalarImpl<float, float, schar>*/, 0 /*mulScalarImpl<float2, float, char2>*/, 0 /*mulScalarImpl<float3, float, char3>*/, 0 /*mulScalarImpl<float4, float, char4>*/},
+            {0 /*mulScalarImpl<float, float, ushort>*/, 0 /*mulScalarImpl<float2, float, ushort2>*/, 0 /*mulScalarImpl<float3, float, ushort3>*/, 0 /*mulScalarImpl<float4, float, ushort4>*/},
+            {0 /*mulScalarImpl<float, float, short>*/, 0 /*mulScalarImpl<float2, float, short2>*/, 0 /*mulScalarImpl<float3, float, short3>*/, 0 /*mulScalarImpl<float4, float, short4>*/},
+            {0 /*mulScalarImpl<float, float, int>*/, 0 /*mulScalarImpl<float2, float, int2>*/, 0 /*mulScalarImpl<float3, float, int3>*/, 0 /*mulScalarImpl<float4, float, int4>*/},
+            {mulScalarImpl<float, float, float>, mulScalarImpl<float2, float, float2>, mulScalarImpl<float3, float, float3>, mulScalarImpl<float4, float, float4>},
+            {mulScalarImpl<float, double, double>, mulScalarImpl<float2, double, double2>, mulScalarImpl<float3, double, double3>, mulScalarImpl<float4, double, double4>}
+        },
+        {
+            {0 /*mulScalarImpl<double, double, uchar>*/, 0 /*mulScalarImpl<double2, double, uchar2>*/, 0 /*mulScalarImpl<double3, double, uchar3>*/, 0 /*mulScalarImpl<double4, double, uchar4>*/},
+            {0 /*mulScalarImpl<double, double, schar>*/, 0 /*mulScalarImpl<double2, double, char2>*/, 0 /*mulScalarImpl<double3, double, char3>*/, 0 /*mulScalarImpl<double4, double, char4>*/},
+            {0 /*mulScalarImpl<double, double, ushort>*/, 0 /*mulScalarImpl<double2, double, ushort2>*/, 0 /*mulScalarImpl<double3, double, ushort3>*/, 0 /*mulScalarImpl<double4, double, ushort4>*/},
+            {0 /*mulScalarImpl<double, double, short>*/, 0 /*mulScalarImpl<double2, double, short2>*/, 0 /*mulScalarImpl<double3, double, short3>*/, 0 /*mulScalarImpl<double4, double, short4>*/},
+            {0 /*mulScalarImpl<double, double, int>*/, 0 /*mulScalarImpl<double2, double, int2>*/, 0 /*mulScalarImpl<double3, double, int3>*/, 0 /*mulScalarImpl<double4, double, int4>*/},
+            {0 /*mulScalarImpl<double, double, float>*/, 0 /*mulScalarImpl<double2, double, float2>*/, 0 /*mulScalarImpl<double3, double, float3>*/, 0 /*mulScalarImpl<double4, double, float4>*/},
+            {mulScalarImpl<double, double, double>, mulScalarImpl<double2, double, double2>, mulScalarImpl<double3, double, double3>, mulScalarImpl<double4, double, double4>}
+        }
+    };
+
+    const int sdepth = src.depth();
+    const int ddepth = dst.depth();
+    const int cn = src.channels();
+
+    CV_DbgAssert( sdepth <= CV_64F && ddepth <= CV_64F && cn <= 4 );
+
+    val[0] *= scale;
+    val[1] *= scale;
+    val[2] *= scale;
+    val[3] *= scale;
+
+    const func_t func = funcs[sdepth][ddepth][cn - 1];
+
+    if (!func)
+        CV_Error(cv::Error::StsUnsupportedFormat, "Unsupported combination of source and destination types");
+
+    func(src, val, dst, stream);
+}
+
+#endif
diff --git a/modules/cudaarithm/src/cuda/mul_spectrums.cu b/modules/cudaarithm/src/cuda/mul_spectrums.cu
index 42b766dc00..b060904816 100644
--- a/modules/cudaarithm/src/cuda/mul_spectrums.cu
+++ b/modules/cudaarithm/src/cuda/mul_spectrums.cu
@@ -40,132 +40,126 @@
 //
 //M*/
 
-#if !defined CUDA_DISABLER
+#include "opencv2/opencv_modules.hpp"
 
-#include "cvconfig.h"
+#ifndef HAVE_OPENCV_CUDEV
 
-#ifdef HAVE_CUFFT
+#error "opencv_cudev is required"
 
-#include <cufft.h>
+#else
 
-#include "opencv2/core/cuda/common.hpp"
+#include "opencv2/cudaarithm.hpp"
+#include "opencv2/cudev.hpp"
 
-namespace cv { namespace cuda { namespace device
+using namespace cv::cudev;
+
+//////////////////////////////////////////////////////////////////////////////
+// mulSpectrums
+
+namespace
 {
-    //////////////////////////////////////////////////////////////////////////
-    // mulSpectrums
-
-    __global__ void mulSpectrumsKernel(const PtrStep<cufftComplex> a, const PtrStep<cufftComplex> b, PtrStepSz<cufftComplex> c)
+    __device__ __forceinline__ float real(const float2& val)
     {
-        const int x = blockIdx.x * blockDim.x + threadIdx.x;
-        const int y = blockIdx.y * blockDim.y + threadIdx.y;
+        return val.x;
+    }
 
-        if (x < c.cols && y < c.rows)
+    __device__ __forceinline__ float imag(const float2& val)
+    {
+        return val.y;
+    }
+
+    __device__ __forceinline__ float2 cmul(const float2& a, const float2& b)
+    {
+        return make_float2((real(a) * real(b)) - (imag(a) * imag(b)),
+                           (real(a) * imag(b)) + (imag(a) * real(b)));
+    }
+
+    __device__ __forceinline__ float2 conj(const float2& a)
+    {
+        return make_float2(real(a), -imag(a));
+    }
+
+    struct comlex_mul : binary_function<float2, float2, float2>
+    {
+        __device__ __forceinline__ float2 operator ()(const float2& a, const float2& b) const
         {
-            c.ptr(y)[x] = cuCmulf(a.ptr(y)[x], b.ptr(y)[x]);
+            return cmul(a, b);
         }
-    }
+    };
 
-
-    void mulSpectrums(const PtrStep<cufftComplex> a, const PtrStep<cufftComplex> b, PtrStepSz<cufftComplex> c, cudaStream_t stream)
+    struct comlex_mul_conj : binary_function<float2, float2, float2>
     {
-        dim3 threads(256);
-        dim3 grid(divUp(c.cols, threads.x), divUp(c.rows, threads.y));
-
-        mulSpectrumsKernel<<<grid, threads, 0, stream>>>(a, b, c);
-        cudaSafeCall( cudaGetLastError() );
-
-        if (stream == 0)
-            cudaSafeCall( cudaDeviceSynchronize() );
-    }
-
-
-    //////////////////////////////////////////////////////////////////////////
-    // mulSpectrums_CONJ
-
-    __global__ void mulSpectrumsKernel_CONJ(const PtrStep<cufftComplex> a, const PtrStep<cufftComplex> b, PtrStepSz<cufftComplex> c)
-    {
-        const int x = blockIdx.x * blockDim.x + threadIdx.x;
-        const int y = blockIdx.y * blockDim.y + threadIdx.y;
-
-        if (x < c.cols && y < c.rows)
+        __device__ __forceinline__ float2 operator ()(const float2& a, const float2& b) const
         {
-            c.ptr(y)[x] = cuCmulf(a.ptr(y)[x], cuConjf(b.ptr(y)[x]));
+            return cmul(a, conj(b));
         }
-    }
+    };
 
-
-    void mulSpectrums_CONJ(const PtrStep<cufftComplex> a, const PtrStep<cufftComplex> b, PtrStepSz<cufftComplex> c, cudaStream_t stream)
+    struct comlex_mul_scale : binary_function<float2, float2, float2>
     {
-        dim3 threads(256);
-        dim3 grid(divUp(c.cols, threads.x), divUp(c.rows, threads.y));
+        float scale;
 
-        mulSpectrumsKernel_CONJ<<<grid, threads, 0, stream>>>(a, b, c);
-        cudaSafeCall( cudaGetLastError() );
-
-        if (stream == 0)
-            cudaSafeCall( cudaDeviceSynchronize() );
-    }
-
-
-    //////////////////////////////////////////////////////////////////////////
-    // mulAndScaleSpectrums
-
-    __global__ void mulAndScaleSpectrumsKernel(const PtrStep<cufftComplex> a, const PtrStep<cufftComplex> b, float scale, PtrStepSz<cufftComplex> c)
-    {
-        const int x = blockIdx.x * blockDim.x + threadIdx.x;
-        const int y = blockIdx.y * blockDim.y + threadIdx.y;
-
-        if (x < c.cols && y < c.rows)
+        __device__ __forceinline__ float2 operator ()(const float2& a, const float2& b) const
         {
-            cufftComplex v = cuCmulf(a.ptr(y)[x], b.ptr(y)[x]);
-            c.ptr(y)[x] = make_cuFloatComplex(cuCrealf(v) * scale, cuCimagf(v) * scale);
+            return scale * cmul(a, b);
         }
-    }
+    };
 
-
-    void mulAndScaleSpectrums(const PtrStep<cufftComplex> a, const PtrStep<cufftComplex> b, float scale, PtrStepSz<cufftComplex> c, cudaStream_t stream)
+    struct comlex_mul_conj_scale : binary_function<float2, float2, float2>
     {
-        dim3 threads(256);
-        dim3 grid(divUp(c.cols, threads.x), divUp(c.rows, threads.y));
+        float scale;
 
-        mulAndScaleSpectrumsKernel<<<grid, threads, 0, stream>>>(a, b, scale, c);
-        cudaSafeCall( cudaGetLastError() );
-
-        if (stream)
-            cudaSafeCall( cudaDeviceSynchronize() );
-    }
-
-
-    //////////////////////////////////////////////////////////////////////////
-    // mulAndScaleSpectrums_CONJ
-
-    __global__ void mulAndScaleSpectrumsKernel_CONJ(const PtrStep<cufftComplex> a, const PtrStep<cufftComplex> b, float scale, PtrStepSz<cufftComplex> c)
-    {
-        const int x = blockIdx.x * blockDim.x + threadIdx.x;
-        const int y = blockIdx.y * blockDim.y + threadIdx.y;
-
-        if (x < c.cols && y < c.rows)
+        __device__ __forceinline__ float2 operator ()(const float2& a, const float2& b) const
         {
-            cufftComplex v = cuCmulf(a.ptr(y)[x], cuConjf(b.ptr(y)[x]));
-            c.ptr(y)[x] = make_cuFloatComplex(cuCrealf(v) * scale, cuCimagf(v) * scale);
+            return scale * cmul(a, conj(b));
         }
-    }
+    };
+}
 
+void cv::cuda::mulSpectrums(InputArray _src1, InputArray _src2, OutputArray _dst, int flags, bool conjB, Stream& stream)
+{
+    (void) flags;
 
-    void mulAndScaleSpectrums_CONJ(const PtrStep<cufftComplex> a, const PtrStep<cufftComplex> b, float scale, PtrStepSz<cufftComplex> c, cudaStream_t stream)
+    GpuMat src1 = _src1.getGpuMat();
+    GpuMat src2 = _src2.getGpuMat();
+
+    CV_Assert( src1.type() == src2.type() && src1.type() == CV_32FC2 );
+    CV_Assert( src1.size() == src2.size() );
+
+    _dst.create(src1.size(), CV_32FC2);
+    GpuMat dst = _dst.getGpuMat();
+
+    if (conjB)
+        gridTransformBinary(globPtr<float2>(src1), globPtr<float2>(src2), globPtr<float2>(dst), comlex_mul_conj(), stream);
+    else
+        gridTransformBinary(globPtr<float2>(src1), globPtr<float2>(src2), globPtr<float2>(dst), comlex_mul(), stream);
+}
+
+void cv::cuda::mulAndScaleSpectrums(InputArray _src1, InputArray _src2, OutputArray _dst, int flags, float scale, bool conjB, Stream& stream)
+{
+    (void) flags;
+
+    GpuMat src1 = _src1.getGpuMat();
+    GpuMat src2 = _src2.getGpuMat();
+
+    CV_Assert( src1.type() == src2.type() && src1.type() == CV_32FC2);
+    CV_Assert( src1.size() == src2.size() );
+
+    _dst.create(src1.size(), CV_32FC2);
+    GpuMat dst = _dst.getGpuMat();
+
+    if (conjB)
     {
-        dim3 threads(256);
-        dim3 grid(divUp(c.cols, threads.x), divUp(c.rows, threads.y));
-
-        mulAndScaleSpectrumsKernel_CONJ<<<grid, threads, 0, stream>>>(a, b, scale, c);
-        cudaSafeCall( cudaGetLastError() );
-
-        if (stream == 0)
-            cudaSafeCall( cudaDeviceSynchronize() );
+        comlex_mul_conj_scale op;
+        op.scale = scale;
+        gridTransformBinary(globPtr<float2>(src1), globPtr<float2>(src2), globPtr<float2>(dst), op, stream);
     }
-}}} // namespace cv { namespace cuda { namespace cudev
+    else
+    {
+        comlex_mul_scale op;
+        op.scale = scale;
+        gridTransformBinary(globPtr<float2>(src1), globPtr<float2>(src2), globPtr<float2>(dst), op, stream);
+    }
+}
 
-#endif // HAVE_CUFFT
-
-#endif /* CUDA_DISABLER */
+#endif
diff --git a/modules/cudaarithm/src/cuda/norm.cu b/modules/cudaarithm/src/cuda/norm.cu
new file mode 100644
index 0000000000..bda6b45815
--- /dev/null
+++ b/modules/cudaarithm/src/cuda/norm.cu
@@ -0,0 +1,119 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                           License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#include "opencv2/opencv_modules.hpp"
+
+#ifndef HAVE_OPENCV_CUDEV
+
+#error "opencv_cudev is required"
+
+#else
+
+#include "opencv2/cudaarithm.hpp"
+#include "opencv2/cudev.hpp"
+
+using namespace cv::cudev;
+
+namespace
+{
+    double normDiffInf(const GpuMat& _src1, const GpuMat& _src2, GpuMat& _buf)
+    {
+        const GpuMat_<uchar>& src1 = (const GpuMat_<uchar>&) _src1;
+        const GpuMat_<uchar>& src2 = (const GpuMat_<uchar>&) _src2;
+        GpuMat_<int>& buf = (GpuMat_<int>&) _buf;
+
+        gridFindMinMaxVal(abs_(cvt_<int>(src1) - cvt_<int>(src2)), buf);
+
+        int data[2];
+        buf.download(cv::Mat(1, 2, buf.type(), data));
+
+        return data[1];
+    }
+
+    double normDiffL1(const GpuMat& _src1, const GpuMat& _src2, GpuMat& _buf)
+    {
+        const GpuMat_<uchar>& src1 = (const GpuMat_<uchar>&) _src1;
+        const GpuMat_<uchar>& src2 = (const GpuMat_<uchar>&) _src2;
+        GpuMat_<int>& buf = (GpuMat_<int>&) _buf;
+
+        gridCalcSum(abs_(cvt_<int>(src1) - cvt_<int>(src2)), buf);
+
+        int data;
+        buf.download(cv::Mat(1, 1, buf.type(), &data));
+
+        return data;
+    }
+
+    double normDiffL2(const GpuMat& _src1, const GpuMat& _src2, GpuMat& _buf)
+    {
+        const GpuMat_<uchar>& src1 = (const GpuMat_<uchar>&) _src1;
+        const GpuMat_<uchar>& src2 = (const GpuMat_<uchar>&) _src2;
+        GpuMat_<double>& buf = (GpuMat_<double>&) _buf;
+
+        gridCalcSum(sqr_(cvt_<double>(src1) - cvt_<double>(src2)), buf);
+
+        double data;
+        buf.download(cv::Mat(1, 1, buf.type(), &data));
+
+        return std::sqrt(data);
+    }
+}
+
+double cv::cuda::norm(InputArray _src1, InputArray _src2, GpuMat& buf, int normType)
+{
+    typedef double (*func_t)(const GpuMat& _src1, const GpuMat& _src2, GpuMat& _buf);
+    static const func_t funcs[] =
+    {
+        0, normDiffInf, normDiffL1, 0, normDiffL2
+    };
+
+    GpuMat src1 = _src1.getGpuMat();
+    GpuMat src2 = _src2.getGpuMat();
+
+    CV_Assert( src1.type() == CV_8UC1 );
+    CV_Assert( src1.size() == src2.size() && src1.type() == src2.type() );
+    CV_Assert( normType == NORM_INF || normType == NORM_L1 || normType == NORM_L2 );
+
+    return funcs[normType](src1, src2, buf);
+}
+
+#endif
diff --git a/modules/cudaarithm/src/cuda/polar_cart.cu b/modules/cudaarithm/src/cuda/polar_cart.cu
index ce2143b253..200b79c055 100644
--- a/modules/cudaarithm/src/cuda/polar_cart.cu
+++ b/modules/cudaarithm/src/cuda/polar_cart.cu
@@ -40,178 +40,172 @@
 //
 //M*/
 
-#if !defined CUDA_DISABLER
+#include "opencv2/opencv_modules.hpp"
 
-#include "opencv2/core/cuda/common.hpp"
+#ifndef HAVE_OPENCV_CUDEV
 
-namespace cv { namespace cuda { namespace device
+#error "opencv_cudev is required"
+
+#else
+
+#include "opencv2/cudaarithm.hpp"
+#include "opencv2/cudev.hpp"
+
+using namespace cv::cudev;
+
+void cv::cuda::magnitude(InputArray _x, InputArray _y, OutputArray _dst, Stream& stream)
 {
-    namespace mathfunc
+    GpuMat x = _x.getGpuMat();
+    GpuMat y = _y.getGpuMat();
+
+    CV_DbgAssert( x.depth() == CV_32F );
+    CV_DbgAssert( y.type() == x.type() && y.size() == x.size() );
+
+    _dst.create(x.size(), CV_32FC1);
+    GpuMat dst = _dst.getGpuMat();
+
+    GpuMat_<float> xc(x.reshape(1));
+    GpuMat_<float> yc(y.reshape(1));
+    GpuMat_<float> magc(dst.reshape(1));
+
+    gridTransformBinary(xc, yc, magc, magnitude_func<float>(), stream);
+}
+
+void cv::cuda::magnitudeSqr(InputArray _x, InputArray _y, OutputArray _dst, Stream& stream)
+{
+    GpuMat x = _x.getGpuMat();
+    GpuMat y = _y.getGpuMat();
+
+    CV_DbgAssert( x.depth() == CV_32F );
+    CV_DbgAssert( y.type() == x.type() && y.size() == x.size() );
+
+    _dst.create(x.size(), CV_32FC1);
+    GpuMat dst = _dst.getGpuMat();
+
+    GpuMat_<float> xc(x.reshape(1));
+    GpuMat_<float> yc(y.reshape(1));
+    GpuMat_<float> magc(dst.reshape(1));
+
+    gridTransformBinary(xc, yc, magc, magnitude_sqr_func<float>(), stream);
+}
+
+void cv::cuda::phase(InputArray _x, InputArray _y, OutputArray _dst, bool angleInDegrees, Stream& stream)
+{
+    GpuMat x = _x.getGpuMat();
+    GpuMat y = _y.getGpuMat();
+
+    CV_DbgAssert( x.depth() == CV_32F );
+    CV_DbgAssert( y.type() == x.type() && y.size() == x.size() );
+
+    _dst.create(x.size(), CV_32FC1);
+    GpuMat dst = _dst.getGpuMat();
+
+    GpuMat_<float> xc(x.reshape(1));
+    GpuMat_<float> yc(y.reshape(1));
+    GpuMat_<float> anglec(dst.reshape(1));
+
+    if (angleInDegrees)
+        gridTransformBinary(xc, yc, anglec, direction_func<float, true>(), stream);
+    else
+        gridTransformBinary(xc, yc, anglec, direction_func<float, false>(), stream);
+}
+
+void cv::cuda::cartToPolar(InputArray _x, InputArray _y, OutputArray _mag, OutputArray _angle, bool angleInDegrees, Stream& stream)
+{
+    GpuMat x = _x.getGpuMat();
+    GpuMat y = _y.getGpuMat();
+
+    CV_DbgAssert( x.depth() == CV_32F );
+    CV_DbgAssert( y.type() == x.type() && y.size() == x.size() );
+
+    _mag.create(x.size(), CV_32FC1);
+    GpuMat mag = _mag.getGpuMat();
+
+    _angle.create(x.size(), CV_32FC1);
+    GpuMat angle = _angle.getGpuMat();
+
+    GpuMat_<float> xc(x.reshape(1));
+    GpuMat_<float> yc(y.reshape(1));
+    GpuMat_<float> magc(mag.reshape(1));
+    GpuMat_<float> anglec(angle.reshape(1));
+
+    if (angleInDegrees)
     {
-        //////////////////////////////////////////////////////////////////////////////////////
-        // Cart <-> Polar
+        gridTransformTuple(zipPtr(xc, yc),
+                           tie(magc, anglec),
+                           make_tuple(
+                               binaryTupleAdapter<0, 1>(magnitude_func<float>()),
+                               binaryTupleAdapter<0, 1>(direction_func<float, true>())),
+                           stream);
+    }
+    else
+    {
+        gridTransformTuple(zipPtr(xc, yc),
+                           tie(magc, anglec),
+                           make_tuple(
+                               binaryTupleAdapter<0, 1>(magnitude_func<float>()),
+                               binaryTupleAdapter<0, 1>(direction_func<float, false>())),
+                           stream);
+    }
+}
 
-        struct Nothing
-        {
-            static __device__ __forceinline__ void calc(int, int, float, float, float*, size_t, float)
-            {
-            }
-        };
-        struct Magnitude
-        {
-            static __device__ __forceinline__ void calc(int x, int y, float x_data, float y_data, float* dst, size_t dst_step, float)
-            {
-                dst[y * dst_step + x] = ::sqrtf(x_data * x_data + y_data * y_data);
-            }
-        };
-        struct MagnitudeSqr
-        {
-            static __device__ __forceinline__ void calc(int x, int y, float x_data, float y_data, float* dst, size_t dst_step, float)
-            {
-                dst[y * dst_step + x] = x_data * x_data + y_data * y_data;
-            }
-        };
-        struct Atan2
-        {
-            static __device__ __forceinline__ void calc(int x, int y, float x_data, float y_data, float* dst, size_t dst_step, float scale)
-            {
-                float angle = ::atan2f(y_data, x_data);
-                angle += (angle < 0) * 2.0f * CV_PI_F;
-                dst[y * dst_step + x] = scale * angle;
-            }
-        };
-        template <typename Mag, typename Angle>
-        __global__ void cartToPolar(const float* xptr, size_t x_step, const float* yptr, size_t y_step,
-                                    float* mag, size_t mag_step, float* angle, size_t angle_step, float scale, int width, int height)
-        {
-            const int x = blockDim.x * blockIdx.x + threadIdx.x;
-            const int y = blockDim.y * blockIdx.y + threadIdx.y;
+namespace
+{
+    template <bool useMag>
+    __global__ void polarToCartImpl(const GlobPtr<float> mag, const GlobPtr<float> angle, GlobPtr<float> xmat, GlobPtr<float> ymat, const float scale, const int rows, const int cols)
+    {
+        const int x = blockDim.x * blockIdx.x + threadIdx.x;
+        const int y = blockDim.y * blockIdx.y + threadIdx.y;
 
-            if (x < width && y < height)
-            {
-                float x_data = xptr[y * x_step + x];
-                float y_data = yptr[y * y_step + x];
+        if (x >= cols || y >= rows)
+            return;
 
-                Mag::calc(x, y, x_data, y_data, mag, mag_step, scale);
-                Angle::calc(x, y, x_data, y_data, angle, angle_step, scale);
-            }
-        }
+        const float mag_val = useMag ? mag(y, x) : 1.0f;
+        const float angle_val = angle(y, x);
 
-        struct NonEmptyMag
-        {
-            static __device__ __forceinline__ float get(const float* mag, size_t mag_step, int x, int y)
-            {
-                return mag[y * mag_step + x];
-            }
-        };
-        struct EmptyMag
-        {
-            static __device__ __forceinline__ float get(const float*, size_t, int, int)
-            {
-                return 1.0f;
-            }
-        };
-        template <typename Mag>
-        __global__ void polarToCart(const float* mag, size_t mag_step, const float* angle, size_t angle_step, float scale,
-            float* xptr, size_t x_step, float* yptr, size_t y_step, int width, int height)
-        {
-            const int x = blockDim.x * blockIdx.x + threadIdx.x;
-            const int y = blockDim.y * blockIdx.y + threadIdx.y;
+        float sin_a, cos_a;
+        ::sincosf(scale * angle_val, &sin_a, &cos_a);
 
-            if (x < width && y < height)
-            {
-                float mag_data = Mag::get(mag, mag_step, x, y);
-                float angle_data = angle[y * angle_step + x];
-                float sin_a, cos_a;
+        xmat(y, x) = mag_val * cos_a;
+        ymat(y, x) = mag_val * sin_a;
+    }
+}
 
-                ::sincosf(scale * angle_data, &sin_a, &cos_a);
+void cv::cuda::polarToCart(InputArray _mag, InputArray _angle, OutputArray _x, OutputArray _y, bool angleInDegrees, Stream& _stream)
+{
+    GpuMat mag = _mag.getGpuMat();
+    GpuMat angle = _angle.getGpuMat();
 
-                xptr[y * x_step + x] = mag_data * cos_a;
-                yptr[y * y_step + x] = mag_data * sin_a;
-            }
-        }
+    CV_DbgAssert( angle.depth() == CV_32F );
+    CV_DbgAssert( mag.empty() || (mag.type() == angle.type() && mag.size() == angle.size()) );
 
-        template <typename Mag, typename Angle>
-        void cartToPolar_caller(PtrStepSzf x, PtrStepSzf y, PtrStepSzf mag, PtrStepSzf angle, bool angleInDegrees, cudaStream_t stream)
-        {
-            dim3 threads(32, 8, 1);
-            dim3 grid(1, 1, 1);
+    _x.create(angle.size(), CV_32FC1);
+    GpuMat x = _x.getGpuMat();
 
-            grid.x = divUp(x.cols, threads.x);
-            grid.y = divUp(x.rows, threads.y);
+    _y.create(angle.size(), CV_32FC1);
+    GpuMat y = _y.getGpuMat();
 
-            const float scale = angleInDegrees ? (180.0f / CV_PI_F) : 1.f;
+    GpuMat_<float> xc(x.reshape(1));
+    GpuMat_<float> yc(y.reshape(1));
+    GpuMat_<float> magc(mag.reshape(1));
+    GpuMat_<float> anglec(angle.reshape(1));
 
-            cartToPolar<Mag, Angle><<<grid, threads, 0, stream>>>(
-                x.data, x.step/x.elemSize(), y.data, y.step/y.elemSize(),
-                mag.data, mag.step/mag.elemSize(), angle.data, angle.step/angle.elemSize(), scale, x.cols, x.rows);
-            cudaSafeCall( cudaGetLastError() );
+    const dim3 block(32, 8);
+    const dim3 grid(divUp(anglec.cols, block.x), divUp(anglec.rows, block.y));
 
-            if (stream == 0)
-                cudaSafeCall( cudaDeviceSynchronize() );
-        }
+    const float scale = angleInDegrees ? (CV_PI_F / 180.0f) : 1.0f;
 
-        void cartToPolar_gpu(PtrStepSzf x, PtrStepSzf y, PtrStepSzf mag, bool magSqr, PtrStepSzf angle, bool angleInDegrees, cudaStream_t stream)
-        {
-            typedef void (*caller_t)(PtrStepSzf x, PtrStepSzf y, PtrStepSzf mag, PtrStepSzf angle, bool angleInDegrees, cudaStream_t stream);
-            static const caller_t callers[2][2][2] =
-            {
-                {
-                    {
-                        cartToPolar_caller<Magnitude, Atan2>,
-                        cartToPolar_caller<Magnitude, Nothing>
-                    },
-                    {
-                        cartToPolar_caller<MagnitudeSqr, Atan2>,
-                        cartToPolar_caller<MagnitudeSqr, Nothing>,
-                    }
-                },
-                {
-                    {
-                        cartToPolar_caller<Nothing, Atan2>,
-                        cartToPolar_caller<Nothing, Nothing>
-                    },
-                    {
-                        cartToPolar_caller<Nothing, Atan2>,
-                        cartToPolar_caller<Nothing, Nothing>,
-                    }
-                }
-            };
+    cudaStream_t stream = StreamAccessor::getStream(_stream);
 
-            callers[mag.data == 0][magSqr][angle.data == 0](x, y, mag, angle, angleInDegrees, stream);
-        }
+    if (magc.empty())
+        polarToCartImpl<false><<<grid, block, 0, stream>>>(shrinkPtr(magc), shrinkPtr(anglec), shrinkPtr(xc), shrinkPtr(yc), scale, anglec.rows, anglec.cols);
+    else
+        polarToCartImpl<true><<<grid, block, 0, stream>>>(shrinkPtr(magc), shrinkPtr(anglec), shrinkPtr(xc), shrinkPtr(yc), scale, anglec.rows, anglec.cols);
 
-        template <typename Mag>
-        void polarToCart_caller(PtrStepSzf mag, PtrStepSzf angle, PtrStepSzf x, PtrStepSzf y, bool angleInDegrees, cudaStream_t stream)
-        {
-            dim3 threads(32, 8, 1);
-            dim3 grid(1, 1, 1);
+    CV_CUDEV_SAFE_CALL( cudaGetLastError() );
 
-            grid.x = divUp(mag.cols, threads.x);
-            grid.y = divUp(mag.rows, threads.y);
+    if (stream == 0)
+        CV_CUDEV_SAFE_CALL( cudaDeviceSynchronize() );
+}
 
-            const float scale = angleInDegrees ? (CV_PI_F / 180.0f) : 1.0f;
-
-            polarToCart<Mag><<<grid, threads, 0, stream>>>(mag.data, mag.step/mag.elemSize(),
-                angle.data, angle.step/angle.elemSize(), scale, x.data, x.step/x.elemSize(), y.data, y.step/y.elemSize(), mag.cols, mag.rows);
-            cudaSafeCall( cudaGetLastError() );
-
-            if (stream == 0)
-                cudaSafeCall( cudaDeviceSynchronize() );
-        }
-
-        void polarToCart_gpu(PtrStepSzf mag, PtrStepSzf angle, PtrStepSzf x, PtrStepSzf y, bool angleInDegrees, cudaStream_t stream)
-        {
-            typedef void (*caller_t)(PtrStepSzf mag, PtrStepSzf angle, PtrStepSzf x, PtrStepSzf y, bool angleInDegrees, cudaStream_t stream);
-            static const caller_t callers[2] =
-            {
-                polarToCart_caller<NonEmptyMag>,
-                polarToCart_caller<EmptyMag>
-            };
-
-            callers[mag.data == 0](mag, angle, x, y, angleInDegrees, stream);
-        }
-    } // namespace mathfunc
-}}} // namespace cv { namespace cuda { namespace cudev
-
-#endif /* CUDA_DISABLER */
+#endif
diff --git a/modules/cudaarithm/src/cuda/reduce.cu b/modules/cudaarithm/src/cuda/reduce.cu
index 2cc4a5b95e..2cb2dacc73 100644
--- a/modules/cudaarithm/src/cuda/reduce.cu
+++ b/modules/cudaarithm/src/cuda/reduce.cu
@@ -40,301 +40,258 @@
 //
 //M*/
 
-#if !defined CUDA_DISABLER
+#include "opencv2/opencv_modules.hpp"
 
-#include "opencv2/core/cuda/common.hpp"
-#include "opencv2/core/cuda/saturate_cast.hpp"
-#include "opencv2/core/cuda/vec_traits.hpp"
-#include "opencv2/core/cuda/vec_math.hpp"
-#include "opencv2/core/cuda/functional.hpp"
-#include "opencv2/core/cuda/reduce.hpp"
-#include "opencv2/core/cuda/limits.hpp"
+#ifndef HAVE_OPENCV_CUDEV
 
-#include "unroll_detail.hpp"
+#error "opencv_cudev is required"
 
-using namespace cv::cuda;
-using namespace cv::cuda::device;
+#else
 
-namespace reduce
+#include "opencv2/cudaarithm.hpp"
+#include "opencv2/cudev.hpp"
+
+using namespace cv::cudev;
+
+namespace
 {
-    struct Sum
+    template <typename T, typename S, typename D>
+    void reduceToRowImpl(const GpuMat& _src, GpuMat& _dst, int reduceOp, Stream& stream)
     {
-        template <typename T>
-        __device__ __forceinline__ T startValue() const
+        const GpuMat_<T>& src = (const GpuMat_<T>&) _src;
+        GpuMat_<D>& dst = (GpuMat_<D>&) _dst;
+
+        switch (reduceOp)
         {
-            return VecTraits<T>::all(0);
-        }
+        case cv::REDUCE_SUM:
+            gridReduceToRow< Sum<S> >(src, dst, stream);
+            break;
 
-        template <typename T>
-        __device__ __forceinline__ T operator ()(T a, T b) const
-        {
-            return a + b;
-        }
+        case cv::REDUCE_AVG:
+            gridReduceToRow< Avg<S> >(src, dst, stream);
+            break;
 
-        template <typename T>
-        __device__ __forceinline__ T result(T r, int) const
-        {
-            return r;
-        }
+        case cv::REDUCE_MIN:
+            gridReduceToRow< Min<S> >(src, dst, stream);
+            break;
 
-        __host__ __device__ __forceinline__ Sum() {}
-        __host__ __device__ __forceinline__ Sum(const Sum&) {}
-    };
-
-    template <typename T> struct OutputType
-    {
-        typedef float type;
-    };
-    template <> struct OutputType<double>
-    {
-        typedef double type;
-    };
-
-    struct Avg
-    {
-        template <typename T>
-        __device__ __forceinline__ T startValue() const
-        {
-            return VecTraits<T>::all(0);
-        }
-
-        template <typename T>
-        __device__ __forceinline__ T operator ()(T a, T b) const
-        {
-            return a + b;
-        }
-
-        template <typename T>
-        __device__ __forceinline__ typename TypeVec<typename OutputType<typename VecTraits<T>::elem_type>::type, VecTraits<T>::cn>::vec_type result(T r, float sz) const
-        {
-            return r / sz;
-        }
-
-        __host__ __device__ __forceinline__ Avg() {}
-        __host__ __device__ __forceinline__ Avg(const Avg&) {}
-    };
-
-    struct Min
-    {
-        template <typename T>
-        __device__ __forceinline__ T startValue() const
-        {
-            return VecTraits<T>::all(numeric_limits<typename VecTraits<T>::elem_type>::max());
-        }
-
-        template <typename T>
-        __device__ __forceinline__ T operator ()(T a, T b) const
-        {
-            minimum<T> minOp;
-            return minOp(a, b);
-        }
-
-        template <typename T>
-        __device__ __forceinline__ T result(T r, int) const
-        {
-            return r;
-        }
-
-        __host__ __device__ __forceinline__ Min() {}
-        __host__ __device__ __forceinline__ Min(const Min&) {}
-    };
-
-    struct Max
-    {
-        template <typename T>
-        __device__ __forceinline__ T startValue() const
-        {
-            return VecTraits<T>::all(-numeric_limits<typename VecTraits<T>::elem_type>::max());
-        }
-
-        template <typename T>
-        __device__ __forceinline__ T operator ()(T a, T b) const
-        {
-            maximum<T> maxOp;
-            return maxOp(a, b);
-        }
-
-        template <typename T>
-        __device__ __forceinline__ T result(T r, int) const
-        {
-            return r;
-        }
-
-        __host__ __device__ __forceinline__ Max() {}
-        __host__ __device__ __forceinline__ Max(const Max&) {}
-    };
-
-    ///////////////////////////////////////////////////////////
-
-    template <typename T, typename S, typename D, class Op>
-    __global__ void rowsKernel(const PtrStepSz<T> src, D* dst, const Op op)
-    {
-        __shared__ S smem[16 * 16];
-
-        const int x = blockIdx.x * 16 + threadIdx.x;
-
-        S myVal = op.template startValue<S>();
-
-        if (x < src.cols)
-        {
-            for (int y = threadIdx.y; y < src.rows; y += 16)
-            {
-                S srcVal = src(y, x);
-                myVal = op(myVal, srcVal);
-            }
-        }
-
-        smem[threadIdx.x * 16 + threadIdx.y] = myVal;
-
-        __syncthreads();
-
-        volatile S* srow = smem + threadIdx.y * 16;
-
-        myVal = srow[threadIdx.x];
-        device::reduce<16>(srow, myVal, threadIdx.x, op);
-
-        if (threadIdx.x == 0)
-            srow[0] = myVal;
-
-        __syncthreads();
-
-        if (threadIdx.y == 0 && x < src.cols)
-            dst[x] = (D) op.result(smem[threadIdx.x * 16], src.rows);
-    }
-
-    template <typename T, typename S, typename D, class Op>
-    void rowsCaller(PtrStepSz<T> src, D* dst, cudaStream_t stream)
-    {
-        const dim3 block(16, 16);
-        const dim3 grid(divUp(src.cols, block.x));
-
-        Op op;
-        rowsKernel<T, S, D, Op><<<grid, block, 0, stream>>>(src, dst, op);
-        cudaSafeCall( cudaGetLastError() );
-
-        if (stream == 0)
-            cudaSafeCall( cudaDeviceSynchronize() );
+        case cv::REDUCE_MAX:
+            gridReduceToRow< Max<S> >(src, dst, stream);
+            break;
+        };
     }
 
     template <typename T, typename S, typename D>
-    void rows(PtrStepSzb src, void* dst, int op, cudaStream_t stream)
+    void reduceToColumnImpl_(const GpuMat& _src, GpuMat& _dst, int reduceOp, Stream& stream)
     {
-        typedef void (*func_t)(PtrStepSz<T> src, D* dst, cudaStream_t stream);
-        static const func_t funcs[] =
+        const GpuMat_<T>& src = (const GpuMat_<T>&) _src;
+        GpuMat_<D>& dst = (GpuMat_<D>&) _dst;
+
+        switch (reduceOp)
         {
-            rowsCaller<T, S, D, Sum>,
-            rowsCaller<T, S, D, Avg>,
-            rowsCaller<T, S, D, Max>,
-            rowsCaller<T, S, D, Min>
+        case cv::REDUCE_SUM:
+            gridReduceToColumn< Sum<S> >(src, dst, stream);
+            break;
+
+        case cv::REDUCE_AVG:
+            gridReduceToColumn< Avg<S> >(src, dst, stream);
+            break;
+
+        case cv::REDUCE_MIN:
+            gridReduceToColumn< Min<S> >(src, dst, stream);
+            break;
+
+        case cv::REDUCE_MAX:
+            gridReduceToColumn< Max<S> >(src, dst, stream);
+            break;
+        };
+    }
+
+    template <typename T, typename S, typename D>
+    void reduceToColumnImpl(const GpuMat& src, GpuMat& dst, int reduceOp, Stream& stream)
+    {
+        typedef void (*func_t)(const GpuMat& src, GpuMat& dst, int reduceOp, Stream& stream);
+        static const func_t funcs[4] =
+        {
+            reduceToColumnImpl_<T, S, D>,
+            reduceToColumnImpl_<typename MakeVec<T, 2>::type, typename MakeVec<S, 2>::type, typename MakeVec<D, 2>::type>,
+            reduceToColumnImpl_<typename MakeVec<T, 3>::type, typename MakeVec<S, 3>::type, typename MakeVec<D, 3>::type>,
+            reduceToColumnImpl_<typename MakeVec<T, 4>::type, typename MakeVec<S, 4>::type, typename MakeVec<D, 4>::type>
         };
 
-        funcs[op]((PtrStepSz<T>) src, (D*) dst, stream);
+        funcs[src.channels() - 1](src, dst, reduceOp, stream);
     }
-
-    template void rows<unsigned char, int, unsigned char>(PtrStepSzb src, void* dst, int op, cudaStream_t stream);
-    template void rows<unsigned char, int, int>(PtrStepSzb src, void* dst, int op, cudaStream_t stream);
-    template void rows<unsigned char, float, float>(PtrStepSzb src, void* dst, int op, cudaStream_t stream);
-    template void rows<unsigned char, double, double>(PtrStepSzb src, void* dst, int op, cudaStream_t stream);
-
-    template void rows<unsigned short, int, unsigned short>(PtrStepSzb src, void* dst, int op, cudaStream_t stream);
-    template void rows<unsigned short, int, int>(PtrStepSzb src, void* dst, int op, cudaStream_t stream);
-    template void rows<unsigned short, float, float>(PtrStepSzb src, void* dst, int op, cudaStream_t stream);
-    template void rows<unsigned short, double, double>(PtrStepSzb src, void* dst, int op, cudaStream_t stream);
-
-    template void rows<short, int, short>(PtrStepSzb src, void* dst, int op, cudaStream_t stream);
-    template void rows<short, int, int>(PtrStepSzb src, void* dst, int op, cudaStream_t stream);
-    template void rows<short, float, float>(PtrStepSzb src, void* dst, int op, cudaStream_t stream);
-    template void rows<short, double, double>(PtrStepSzb src, void* dst, int op, cudaStream_t stream);
-
-    template void rows<int, int, int>(PtrStepSzb src, void* dst, int op, cudaStream_t stream);
-    template void rows<int, float, float>(PtrStepSzb src, void* dst, int op, cudaStream_t stream);
-    template void rows<int, double, double>(PtrStepSzb src, void* dst, int op, cudaStream_t stream);
-
-    template void rows<float, float, float>(PtrStepSzb src, void* dst, int op, cudaStream_t stream);
-    template void rows<float, double, double>(PtrStepSzb src, void* dst, int op, cudaStream_t stream);
-
-    template void rows<double, double, double>(PtrStepSzb src, void* dst, int op, cudaStream_t stream);
-
-    ///////////////////////////////////////////////////////////
-
-    template <int BLOCK_SIZE, typename T, typename S, typename D, int cn, class Op>
-    __global__ void colsKernel(const PtrStepSz<typename TypeVec<T, cn>::vec_type> src, typename TypeVec<D, cn>::vec_type* dst, const Op op)
-    {
-        typedef typename TypeVec<T, cn>::vec_type src_type;
-        typedef typename TypeVec<S, cn>::vec_type work_type;
-        typedef typename TypeVec<D, cn>::vec_type dst_type;
-
-        __shared__ S smem[BLOCK_SIZE * cn];
-
-        const int y = blockIdx.x;
-
-        const src_type* srcRow = src.ptr(y);
-
-        work_type myVal = op.template startValue<work_type>();
-
-        for (int x = threadIdx.x; x < src.cols; x += BLOCK_SIZE)
-            myVal = op(myVal, saturate_cast<work_type>(srcRow[x]));
-
-        device::reduce<BLOCK_SIZE>(detail::Unroll<cn>::template smem_tuple<BLOCK_SIZE>(smem), detail::Unroll<cn>::tie(myVal), threadIdx.x, detail::Unroll<cn>::op(op));
-
-        if (threadIdx.x == 0)
-            dst[y] = saturate_cast<dst_type>(op.result(myVal, src.cols));
-    }
-
-    template <typename T, typename S, typename D, int cn, class Op> void colsCaller(PtrStepSzb src, void* dst, cudaStream_t stream)
-    {
-        const int BLOCK_SIZE = 256;
-
-        const dim3 block(BLOCK_SIZE);
-        const dim3 grid(src.rows);
-
-        Op op;
-        colsKernel<BLOCK_SIZE, T, S, D, cn, Op><<<grid, block, 0, stream>>>((PtrStepSz<typename TypeVec<T, cn>::vec_type>) src, (typename TypeVec<D, cn>::vec_type*) dst, op);
-        cudaSafeCall( cudaGetLastError() );
-
-        if (stream == 0)
-            cudaSafeCall( cudaDeviceSynchronize() );
-
-    }
-
-    template <typename T, typename S, typename D> void cols(PtrStepSzb src, void* dst, int cn, int op, cudaStream_t stream)
-    {
-        typedef void (*func_t)(PtrStepSzb src, void* dst, cudaStream_t stream);
-        static const func_t funcs[5][4] =
-        {
-            {0,0,0,0},
-            {colsCaller<T, S, D, 1, Sum>, colsCaller<T, S, D, 1, Avg>, colsCaller<T, S, D, 1, Max>, colsCaller<T, S, D, 1, Min>},
-            {colsCaller<T, S, D, 2, Sum>, colsCaller<T, S, D, 2, Avg>, colsCaller<T, S, D, 2, Max>, colsCaller<T, S, D, 2, Min>},
-            {colsCaller<T, S, D, 3, Sum>, colsCaller<T, S, D, 3, Avg>, colsCaller<T, S, D, 3, Max>, colsCaller<T, S, D, 3, Min>},
-            {colsCaller<T, S, D, 4, Sum>, colsCaller<T, S, D, 4, Avg>, colsCaller<T, S, D, 4, Max>, colsCaller<T, S, D, 4, Min>},
-        };
-
-        funcs[cn][op](src, dst, stream);
-    }
-
-    template void cols<unsigned char, int, unsigned char>(PtrStepSzb src, void* dst, int cn, int op, cudaStream_t stream);
-    template void cols<unsigned char, int, int>(PtrStepSzb src, void* dst, int cn, int op, cudaStream_t stream);
-    template void cols<unsigned char, float, float>(PtrStepSzb src, void* dst, int cn, int op, cudaStream_t stream);
-    template void cols<unsigned char, double, double>(PtrStepSzb src, void* dst, int cn, int op, cudaStream_t stream);
-
-    template void cols<unsigned short, int, unsigned short>(PtrStepSzb src, void* dst, int cn, int op, cudaStream_t stream);
-    template void cols<unsigned short, int, int>(PtrStepSzb src, void* dst, int cn, int op, cudaStream_t stream);
-    template void cols<unsigned short, float, float>(PtrStepSzb src, void* dst, int cn, int op, cudaStream_t stream);
-    template void cols<unsigned short, double, double>(PtrStepSzb src, void* dst, int cn, int op, cudaStream_t stream);
-
-    template void cols<short, int, short>(PtrStepSzb src, void* dst, int cn, int op, cudaStream_t stream);
-    template void cols<short, int, int>(PtrStepSzb src, void* dst, int cn, int op, cudaStream_t stream);
-    template void cols<short, float, float>(PtrStepSzb src, void* dst, int cn, int op, cudaStream_t stream);
-    template void cols<short, double, double>(PtrStepSzb src, void* dst, int cn, int op, cudaStream_t stream);
-
-    template void cols<int, int, int>(PtrStepSzb src, void* dst, int cn, int op, cudaStream_t stream);
-    template void cols<int, float, float>(PtrStepSzb src, void* dst, int cn, int op, cudaStream_t stream);
-    template void cols<int, double, double>(PtrStepSzb src, void* dst, int cn, int op, cudaStream_t stream);
-
-    template void cols<float, float, float>(PtrStepSzb src, void* dst, int cn, int op, cudaStream_t stream);
-    template void cols<float, double, double>(PtrStepSzb src, void* dst, int cn, int op, cudaStream_t stream);
-
-    template void cols<double, double, double>(PtrStepSzb src, void* dst, int cn, int op, cudaStream_t stream);
 }
 
-#endif /* CUDA_DISABLER */
+void cv::cuda::reduce(InputArray _src, OutputArray _dst, int dim, int reduceOp, int dtype, Stream& stream)
+{
+    GpuMat src = _src.getGpuMat();
+
+    CV_Assert( src.channels() <= 4 );
+    CV_Assert( dim == 0 || dim == 1 );
+    CV_Assert( reduceOp == REDUCE_SUM || reduceOp == REDUCE_AVG || reduceOp == REDUCE_MAX || reduceOp == REDUCE_MIN );
+
+    if (dtype < 0)
+        dtype = src.depth();
+
+    _dst.create(1, dim == 0 ? src.cols : src.rows, CV_MAKE_TYPE(CV_MAT_DEPTH(dtype), src.channels()));
+    GpuMat dst = _dst.getGpuMat();
+
+    if (dim == 0)
+    {
+        typedef void (*func_t)(const GpuMat& _src, GpuMat& _dst, int reduceOp, Stream& stream);
+        static const func_t funcs[7][7] =
+        {
+            {
+                reduceToRowImpl<uchar, int, uchar>,
+                0 /*reduceToRowImpl<uchar, int, schar>*/,
+                0 /*reduceToRowImpl<uchar, int, ushort>*/,
+                0 /*reduceToRowImpl<uchar, int, short>*/,
+                reduceToRowImpl<uchar, int, int>,
+                reduceToRowImpl<uchar, float, float>,
+                reduceToRowImpl<uchar, double, double>
+            },
+            {
+                0 /*reduceToRowImpl<schar, int, uchar>*/,
+                0 /*reduceToRowImpl<schar, int, schar>*/,
+                0 /*reduceToRowImpl<schar, int, ushort>*/,
+                0 /*reduceToRowImpl<schar, int, short>*/,
+                0 /*reduceToRowImpl<schar, int, int>*/,
+                0 /*reduceToRowImpl<schar, float, float>*/,
+                0 /*reduceToRowImpl<schar, double, double>*/
+            },
+            {
+                0 /*reduceToRowImpl<ushort, int, uchar>*/,
+                0 /*reduceToRowImpl<ushort, int, schar>*/,
+                reduceToRowImpl<ushort, int, ushort>,
+                0 /*reduceToRowImpl<ushort, int, short>*/,
+                reduceToRowImpl<ushort, int, int>,
+                reduceToRowImpl<ushort, float, float>,
+                reduceToRowImpl<ushort, double, double>
+            },
+            {
+                0 /*reduceToRowImpl<short, int, uchar>*/,
+                0 /*reduceToRowImpl<short, int, schar>*/,
+                0 /*reduceToRowImpl<short, int, ushort>*/,
+                reduceToRowImpl<short, int, short>,
+                reduceToRowImpl<short, int, int>,
+                reduceToRowImpl<short, float, float>,
+                reduceToRowImpl<short, double, double>
+            },
+            {
+                0 /*reduceToRowImpl<int, int, uchar>*/,
+                0 /*reduceToRowImpl<int, int, schar>*/,
+                0 /*reduceToRowImpl<int, int, ushort>*/,
+                0 /*reduceToRowImpl<int, int, short>*/,
+                reduceToRowImpl<int, int, int>,
+                reduceToRowImpl<int, float, float>,
+                reduceToRowImpl<int, double, double>
+            },
+            {
+                0 /*reduceToRowImpl<float, float, uchar>*/,
+                0 /*reduceToRowImpl<float, float, schar>*/,
+                0 /*reduceToRowImpl<float, float, ushort>*/,
+                0 /*reduceToRowImpl<float, float, short>*/,
+                0 /*reduceToRowImpl<float, float, int>*/,
+                reduceToRowImpl<float, float, float>,
+                reduceToRowImpl<float, double, double>
+            },
+            {
+                0 /*reduceToRowImpl<double, double, uchar>*/,
+                0 /*reduceToRowImpl<double, double, schar>*/,
+                0 /*reduceToRowImpl<double, double, ushort>*/,
+                0 /*reduceToRowImpl<double, double, short>*/,
+                0 /*reduceToRowImpl<double, double, int>*/,
+                0 /*reduceToRowImpl<double, double, float>*/,
+                reduceToRowImpl<double, double, double>
+            }
+        };
+
+        const func_t func = funcs[src.depth()][dst.depth()];
+
+        if (!func)
+            CV_Error(cv::Error::StsUnsupportedFormat, "Unsupported combination of input and output array formats");
+
+        GpuMat dst_cont = dst.reshape(1);
+        func(src.reshape(1), dst_cont, reduceOp, stream);
+    }
+    else
+    {
+        typedef void (*func_t)(const GpuMat& _src, GpuMat& _dst, int reduceOp, Stream& stream);
+        static const func_t funcs[7][7] =
+        {
+            {
+                reduceToColumnImpl<uchar, int, uchar>,
+                0 /*reduceToColumnImpl<uchar, int, schar>*/,
+                0 /*reduceToColumnImpl<uchar, int, ushort>*/,
+                0 /*reduceToColumnImpl<uchar, int, short>*/,
+                reduceToColumnImpl<uchar, int, int>,
+                reduceToColumnImpl<uchar, float, float>,
+                reduceToColumnImpl<uchar, double, double>
+            },
+            {
+                0 /*reduceToColumnImpl<schar, int, uchar>*/,
+                0 /*reduceToColumnImpl<schar, int, schar>*/,
+                0 /*reduceToColumnImpl<schar, int, ushort>*/,
+                0 /*reduceToColumnImpl<schar, int, short>*/,
+                0 /*reduceToColumnImpl<schar, int, int>*/,
+                0 /*reduceToColumnImpl<schar, float, float>*/,
+                0 /*reduceToColumnImpl<schar, double, double>*/
+            },
+            {
+                0 /*reduceToColumnImpl<ushort, int, uchar>*/,
+                0 /*reduceToColumnImpl<ushort, int, schar>*/,
+                reduceToColumnImpl<ushort, int, ushort>,
+                0 /*reduceToColumnImpl<ushort, int, short>*/,
+                reduceToColumnImpl<ushort, int, int>,
+                reduceToColumnImpl<ushort, float, float>,
+                reduceToColumnImpl<ushort, double, double>
+            },
+            {
+                0 /*reduceToColumnImpl<short, int, uchar>*/,
+                0 /*reduceToColumnImpl<short, int, schar>*/,
+                0 /*reduceToColumnImpl<short, int, ushort>*/,
+                reduceToColumnImpl<short, int, short>,
+                reduceToColumnImpl<short, int, int>,
+                reduceToColumnImpl<short, float, float>,
+                reduceToColumnImpl<short, double, double>
+            },
+            {
+                0 /*reduceToColumnImpl<int, int, uchar>*/,
+                0 /*reduceToColumnImpl<int, int, schar>*/,
+                0 /*reduceToColumnImpl<int, int, ushort>*/,
+                0 /*reduceToColumnImpl<int, int, short>*/,
+                reduceToColumnImpl<int, int, int>,
+                reduceToColumnImpl<int, float, float>,
+                reduceToColumnImpl<int, double, double>
+            },
+            {
+                0 /*reduceToColumnImpl<float, float, uchar>*/,
+                0 /*reduceToColumnImpl<float, float, schar>*/,
+                0 /*reduceToColumnImpl<float, float, ushort>*/,
+                0 /*reduceToColumnImpl<float, float, short>*/,
+                0 /*reduceToColumnImpl<float, float, int>*/,
+                reduceToColumnImpl<float, float, float>,
+                reduceToColumnImpl<float, double, double>
+            },
+            {
+                0 /*reduceToColumnImpl<double, double, uchar>*/,
+                0 /*reduceToColumnImpl<double, double, schar>*/,
+                0 /*reduceToColumnImpl<double, double, ushort>*/,
+                0 /*reduceToColumnImpl<double, double, short>*/,
+                0 /*reduceToColumnImpl<double, double, int>*/,
+                0 /*reduceToColumnImpl<double, double, float>*/,
+                reduceToColumnImpl<double, double, double>
+            }
+        };
+
+        const func_t func = funcs[src.depth()][dst.depth()];
+
+        if (!func)
+            CV_Error(cv::Error::StsUnsupportedFormat, "Unsupported combination of input and output array formats");
+
+        func(src, dst, reduceOp, stream);
+    }
+}
+
+#endif
diff --git a/modules/cudaarithm/src/cuda/split_merge.cu b/modules/cudaarithm/src/cuda/split_merge.cu
index 44853fc731..13d6a349fb 100644
--- a/modules/cudaarithm/src/cuda/split_merge.cu
+++ b/modules/cudaarithm/src/cuda/split_merge.cu
@@ -40,472 +40,209 @@
 //
 //M*/
 
-#if !defined CUDA_DISABLER
+#include "opencv2/opencv_modules.hpp"
 
-#include "opencv2/core/cuda/common.hpp"
+#ifndef HAVE_OPENCV_CUDEV
 
-namespace cv { namespace cuda { namespace device
+#error "opencv_cudev is required"
+
+#else
+
+#include "opencv2/cudaarithm.hpp"
+#include "opencv2/cudev.hpp"
+
+using namespace cv::cudev;
+
+////////////////////////////////////////////////////////////////////////
+/// merge
+
+namespace
 {
-    namespace split_merge
+    template <int cn, typename T> struct MergeFunc;
+
+    template <typename T> struct MergeFunc<2, T>
     {
-        template <typename T, size_t elem_size = sizeof(T)>
-        struct TypeTraits
+        static void call(const GpuMat* src, GpuMat& dst, Stream& stream)
         {
-            typedef T type;
-            typedef T type2;
-            typedef T type3;
-            typedef T type4;
-        };
+            gridMerge(zipPtr(globPtr<T>(src[0]), globPtr<T>(src[1])),
+                    globPtr<typename MakeVec<T, 2>::type>(dst),
+                    stream);
+        }
+    };
 
-        template <typename T>
-        struct TypeTraits<T, 1>
+    template <typename T> struct MergeFunc<3, T>
+    {
+        static void call(const GpuMat* src, GpuMat& dst, Stream& stream)
         {
-            typedef char type;
-            typedef char2 type2;
-            typedef char3 type3;
-            typedef char4 type4;
-        };
+            gridMerge(zipPtr(globPtr<T>(src[0]), globPtr<T>(src[1]), globPtr<T>(src[2])),
+                    globPtr<typename MakeVec<T, 3>::type>(dst),
+                    stream);
+        }
+    };
 
-        template <typename T>
-        struct TypeTraits<T, 2>
+    template <typename T> struct MergeFunc<4, T>
+    {
+        static void call(const GpuMat* src, GpuMat& dst, Stream& stream)
         {
-            typedef short type;
-            typedef short2 type2;
-            typedef short3 type3;
-            typedef short4 type4;
-        };
+            gridMerge(zipPtr(globPtr<T>(src[0]), globPtr<T>(src[1]), globPtr<T>(src[2]), globPtr<T>(src[3])),
+                    globPtr<typename MakeVec<T, 4>::type>(dst),
+                    stream);
+        }
+    };
 
-        template <typename T>
-        struct TypeTraits<T, 4>
+    void mergeImpl(const GpuMat* src, size_t n, cv::OutputArray _dst, Stream& stream)
+    {
+        CV_DbgAssert( src != 0 );
+        CV_DbgAssert( n > 0 && n <= 4 );
+
+        const int depth = src[0].depth();
+        const cv::Size size = src[0].size();
+
+#ifdef _DEBUG
+        for (size_t i = 0; i < n; ++i)
         {
-            typedef int type;
-            typedef int2 type2;
-            typedef int3 type3;
-            typedef int4 type4;
-        };
+            CV_Assert( src[i].size() == size );
+            CV_Assert( src[i].depth() == depth );
+            CV_Assert( src[i].channels() == 1 );
+        }
+#endif
 
-        template <typename T>
-        struct TypeTraits<T, 8>
+        if (n == 1)
         {
-            typedef double type;
-            typedef double2 type2;
-            //typedef double3 type3;
-            //typedef double4 type3;
-        };
-
-        typedef void (*MergeFunction)(const PtrStepSzb* src, PtrStepSzb& dst, const cudaStream_t& stream);
-        typedef void (*SplitFunction)(const PtrStepSzb& src, PtrStepSzb* dst, const cudaStream_t& stream);
-
-        //------------------------------------------------------------
-        // Merge
-
-        template <typename T>
-        __global__ void mergeC2_(const uchar* src0, size_t src0_step,
-                                 const uchar* src1, size_t src1_step,
-                                 int rows, int cols, uchar* dst, size_t dst_step)
+            src[0].copyTo(_dst, stream);
+        }
+        else
         {
-            typedef typename TypeTraits<T>::type2 dst_type;
-
-            const int x = blockIdx.x * blockDim.x + threadIdx.x;
-            const int y = blockIdx.y * blockDim.y + threadIdx.y;
-
-            const T* src0_y = (const T*)(src0 + y * src0_step);
-            const T* src1_y = (const T*)(src1 + y * src1_step);
-            dst_type* dst_y = (dst_type*)(dst + y * dst_step);
-
-            if (x < cols && y < rows)
+            typedef void (*func_t)(const GpuMat* src, GpuMat& dst, Stream& stream);
+            static const func_t funcs[3][5] =
             {
-                dst_type dst_elem;
-                dst_elem.x = src0_y[x];
-                dst_elem.y = src1_y[x];
-                dst_y[x] = dst_elem;
-            }
-        }
-
-
-        template <typename T>
-        __global__ void mergeC3_(const uchar* src0, size_t src0_step,
-                                 const uchar* src1, size_t src1_step,
-                                 const uchar* src2, size_t src2_step,
-                                 int rows, int cols, uchar* dst, size_t dst_step)
-        {
-            typedef typename TypeTraits<T>::type3 dst_type;
-
-            const int x = blockIdx.x * blockDim.x + threadIdx.x;
-            const int y = blockIdx.y * blockDim.y + threadIdx.y;
-
-            const T* src0_y = (const T*)(src0 + y * src0_step);
-            const T* src1_y = (const T*)(src1 + y * src1_step);
-            const T* src2_y = (const T*)(src2 + y * src2_step);
-            dst_type* dst_y = (dst_type*)(dst + y * dst_step);
-
-            if (x < cols && y < rows)
-            {
-                dst_type dst_elem;
-                dst_elem.x = src0_y[x];
-                dst_elem.y = src1_y[x];
-                dst_elem.z = src2_y[x];
-                dst_y[x] = dst_elem;
-            }
-        }
-
-
-        template <>
-        __global__ void mergeC3_<double>(const uchar* src0, size_t src0_step,
-                                 const uchar* src1, size_t src1_step,
-                                 const uchar* src2, size_t src2_step,
-                                 int rows, int cols, uchar* dst, size_t dst_step)
-        {
-            const int x = blockIdx.x * blockDim.x + threadIdx.x;
-            const int y = blockIdx.y * blockDim.y + threadIdx.y;
-
-            const double* src0_y = (const double*)(src0 + y * src0_step);
-            const double* src1_y = (const double*)(src1 + y * src1_step);
-            const double* src2_y = (const double*)(src2 + y * src2_step);
-            double* dst_y = (double*)(dst + y * dst_step);
-
-            if (x < cols && y < rows)
-            {
-                dst_y[3 * x] = src0_y[x];
-                dst_y[3 * x + 1] = src1_y[x];
-                dst_y[3 * x + 2] = src2_y[x];
-            }
-        }
-
-
-        template <typename T>
-        __global__ void mergeC4_(const uchar* src0, size_t src0_step,
-                                 const uchar* src1, size_t src1_step,
-                                 const uchar* src2, size_t src2_step,
-                                 const uchar* src3, size_t src3_step,
-                                 int rows, int cols, uchar* dst, size_t dst_step)
-        {
-            typedef typename TypeTraits<T>::type4 dst_type;
-
-            const int x = blockIdx.x * blockDim.x + threadIdx.x;
-            const int y = blockIdx.y * blockDim.y + threadIdx.y;
-
-            const T* src0_y = (const T*)(src0 + y * src0_step);
-            const T* src1_y = (const T*)(src1 + y * src1_step);
-            const T* src2_y = (const T*)(src2 + y * src2_step);
-            const T* src3_y = (const T*)(src3 + y * src3_step);
-            dst_type* dst_y = (dst_type*)(dst + y * dst_step);
-
-            if (x < cols && y < rows)
-            {
-                dst_type dst_elem;
-                dst_elem.x = src0_y[x];
-                dst_elem.y = src1_y[x];
-                dst_elem.z = src2_y[x];
-                dst_elem.w = src3_y[x];
-                dst_y[x] = dst_elem;
-            }
-        }
-
-
-        template <>
-        __global__ void mergeC4_<double>(const uchar* src0, size_t src0_step,
-                                 const uchar* src1, size_t src1_step,
-                                 const uchar* src2, size_t src2_step,
-                                 const uchar* src3, size_t src3_step,
-                                 int rows, int cols, uchar* dst, size_t dst_step)
-        {
-            const int x = blockIdx.x * blockDim.x + threadIdx.x;
-            const int y = blockIdx.y * blockDim.y + threadIdx.y;
-
-            const double* src0_y = (const double*)(src0 + y * src0_step);
-            const double* src1_y = (const double*)(src1 + y * src1_step);
-            const double* src2_y = (const double*)(src2 + y * src2_step);
-            const double* src3_y = (const double*)(src3 + y * src3_step);
-            double2* dst_y = (double2*)(dst + y * dst_step);
-
-            if (x < cols && y < rows)
-            {
-                dst_y[2 * x] = make_double2(src0_y[x], src1_y[x]);
-                dst_y[2 * x + 1] = make_double2(src2_y[x], src3_y[x]);
-            }
-        }
-
-
-        template <typename T>
-        static void mergeC2_(const PtrStepSzb* src, PtrStepSzb& dst, const cudaStream_t& stream)
-        {
-            dim3 block(32, 8);
-            dim3 grid(divUp(dst.cols, block.x), divUp(dst.rows, block.y));
-            mergeC2_<T><<<grid, block, 0, stream>>>(
-                    src[0].data, src[0].step,
-                    src[1].data, src[1].step,
-                    dst.rows, dst.cols, dst.data, dst.step);
-            cudaSafeCall( cudaGetLastError() );
-
-            if (stream == 0)
-                cudaSafeCall(cudaDeviceSynchronize());
-        }
-
-
-        template <typename T>
-        static void mergeC3_(const PtrStepSzb* src, PtrStepSzb& dst, const cudaStream_t& stream)
-        {
-            dim3 block(32, 8);
-            dim3 grid(divUp(dst.cols, block.x), divUp(dst.rows, block.y));
-            mergeC3_<T><<<grid, block, 0, stream>>>(
-                    src[0].data, src[0].step,
-                    src[1].data, src[1].step,
-                    src[2].data, src[2].step,
-                    dst.rows, dst.cols, dst.data, dst.step);
-            cudaSafeCall( cudaGetLastError() );
-
-            if (stream == 0)
-                cudaSafeCall(cudaDeviceSynchronize());
-        }
-
-
-        template <typename T>
-        static void mergeC4_(const PtrStepSzb* src, PtrStepSzb& dst, const cudaStream_t& stream)
-        {
-            dim3 block(32, 8);
-            dim3 grid(divUp(dst.cols, block.x), divUp(dst.rows, block.y));
-            mergeC4_<T><<<grid, block, 0, stream>>>(
-                    src[0].data, src[0].step,
-                    src[1].data, src[1].step,
-                    src[2].data, src[2].step,
-                    src[3].data, src[3].step,
-                    dst.rows, dst.cols, dst.data, dst.step);
-            cudaSafeCall( cudaGetLastError() );
-
-            if (stream == 0)
-                cudaSafeCall(cudaDeviceSynchronize());
-        }
-
-
-        void merge(const PtrStepSzb* src, PtrStepSzb& dst,
-                                     int total_channels, size_t elem_size,
-                                     const cudaStream_t& stream)
-        {
-            static MergeFunction merge_func_tbl[] =
-            {
-                mergeC2_<char>, mergeC2_<short>, mergeC2_<int>, 0, mergeC2_<double>,
-                mergeC3_<char>, mergeC3_<short>, mergeC3_<int>, 0, mergeC3_<double>,
-                mergeC4_<char>, mergeC4_<short>, mergeC4_<int>, 0, mergeC4_<double>,
+                {MergeFunc<2, uchar>::call, MergeFunc<2, ushort>::call, MergeFunc<2, int>::call, 0, MergeFunc<2, double>::call},
+                {MergeFunc<3, uchar>::call, MergeFunc<3, ushort>::call, MergeFunc<3, int>::call, 0, MergeFunc<3, double>::call},
+                {MergeFunc<4, uchar>::call, MergeFunc<4, ushort>::call, MergeFunc<4, int>::call, 0, MergeFunc<4, double>::call}
             };
 
-            size_t merge_func_id = (total_channels - 2) * 5 + (elem_size >> 1);
-            MergeFunction merge_func = merge_func_tbl[merge_func_id];
+            const int channels = static_cast<int>(n);
 
-            if (merge_func == 0)
+            _dst.create(size, CV_MAKE_TYPE(depth, channels));
+            GpuMat dst = _dst.getGpuMat();
+
+            const func_t func = funcs[channels - 2][CV_ELEM_SIZE(depth) / 2];
+
+            if (func == 0)
                 CV_Error(cv::Error::StsUnsupportedFormat, "Unsupported channel count or data type");
 
-            merge_func(src, dst, stream);
+            func(src, dst, stream);
         }
+    }
+}
+
+void cv::cuda::merge(const GpuMat* src, size_t n, OutputArray dst, Stream& stream)
+{
+    mergeImpl(src, n, dst, stream);
+}
 
 
+void cv::cuda::merge(const std::vector<GpuMat>& src, OutputArray dst, Stream& stream)
+{
+    mergeImpl(&src[0], src.size(), dst, stream);
+}
 
-        //------------------------------------------------------------
-        // Split
+////////////////////////////////////////////////////////////////////////
+/// split
 
+namespace
+{
+    template <int cn, typename T> struct SplitFunc;
 
-        template <typename T>
-        __global__ void splitC2_(const uchar* src, size_t src_step,
-                                int rows, int cols,
-                                uchar* dst0, size_t dst0_step,
-                                uchar* dst1, size_t dst1_step)
+    template <typename T> struct SplitFunc<2, T>
+    {
+        static void call(const GpuMat& src, GpuMat* dst, Stream& stream)
         {
-            typedef typename TypeTraits<T>::type2 src_type;
-
-            const int x = blockIdx.x * blockDim.x + threadIdx.x;
-            const int y = blockIdx.y * blockDim.y + threadIdx.y;
-
-            const src_type* src_y = (const src_type*)(src + y * src_step);
-            T* dst0_y = (T*)(dst0 + y * dst0_step);
-            T* dst1_y = (T*)(dst1 + y * dst1_step);
-
-            if (x < cols && y < rows)
+            GlobPtrSz<T> dstarr[2] =
             {
-                src_type src_elem = src_y[x];
-                dst0_y[x] = src_elem.x;
-                dst1_y[x] = src_elem.y;
-            }
-        }
-
-
-        template <typename T>
-        __global__ void splitC3_(const uchar* src, size_t src_step,
-                                int rows, int cols,
-                                uchar* dst0, size_t dst0_step,
-                                uchar* dst1, size_t dst1_step,
-                                uchar* dst2, size_t dst2_step)
-        {
-            typedef typename TypeTraits<T>::type3 src_type;
-
-            const int x = blockIdx.x * blockDim.x + threadIdx.x;
-            const int y = blockIdx.y * blockDim.y + threadIdx.y;
-
-            const src_type* src_y = (const src_type*)(src + y * src_step);
-            T* dst0_y = (T*)(dst0 + y * dst0_step);
-            T* dst1_y = (T*)(dst1 + y * dst1_step);
-            T* dst2_y = (T*)(dst2 + y * dst2_step);
-
-            if (x < cols && y < rows)
-            {
-                src_type src_elem = src_y[x];
-                dst0_y[x] = src_elem.x;
-                dst1_y[x] = src_elem.y;
-                dst2_y[x] = src_elem.z;
-            }
-        }
-
-
-        template <>
-        __global__ void splitC3_<double>(
-                const uchar* src, size_t src_step, int rows, int cols,
-                uchar* dst0, size_t dst0_step,
-                uchar* dst1, size_t dst1_step,
-                uchar* dst2, size_t dst2_step)
-        {
-            const int x = blockIdx.x * blockDim.x + threadIdx.x;
-            const int y = blockIdx.y * blockDim.y + threadIdx.y;
-
-            const double* src_y = (const double*)(src + y * src_step);
-            double* dst0_y = (double*)(dst0 + y * dst0_step);
-            double* dst1_y = (double*)(dst1 + y * dst1_step);
-            double* dst2_y = (double*)(dst2 + y * dst2_step);
-
-            if (x < cols && y < rows)
-            {
-                dst0_y[x] = src_y[3 * x];
-                dst1_y[x] = src_y[3 * x + 1];
-                dst2_y[x] = src_y[3 * x + 2];
-            }
-        }
-
-
-        template <typename T>
-        __global__ void splitC4_(const uchar* src, size_t src_step, int rows, int cols,
-                                uchar* dst0, size_t dst0_step,
-                                uchar* dst1, size_t dst1_step,
-                                uchar* dst2, size_t dst2_step,
-                                uchar* dst3, size_t dst3_step)
-        {
-            typedef typename TypeTraits<T>::type4 src_type;
-
-            const int x = blockIdx.x * blockDim.x + threadIdx.x;
-            const int y = blockIdx.y * blockDim.y + threadIdx.y;
-
-            const src_type* src_y = (const src_type*)(src + y * src_step);
-            T* dst0_y = (T*)(dst0 + y * dst0_step);
-            T* dst1_y = (T*)(dst1 + y * dst1_step);
-            T* dst2_y = (T*)(dst2 + y * dst2_step);
-            T* dst3_y = (T*)(dst3 + y * dst3_step);
-
-            if (x < cols && y < rows)
-            {
-                src_type src_elem = src_y[x];
-                dst0_y[x] = src_elem.x;
-                dst1_y[x] = src_elem.y;
-                dst2_y[x] = src_elem.z;
-                dst3_y[x] = src_elem.w;
-            }
-        }
-
-
-        template <>
-        __global__ void splitC4_<double>(
-                const uchar* src, size_t src_step, int rows, int cols,
-                uchar* dst0, size_t dst0_step,
-                uchar* dst1, size_t dst1_step,
-                uchar* dst2, size_t dst2_step,
-                uchar* dst3, size_t dst3_step)
-        {
-            const int x = blockIdx.x * blockDim.x + threadIdx.x;
-            const int y = blockIdx.y * blockDim.y + threadIdx.y;
-
-            const double2* src_y = (const double2*)(src + y * src_step);
-            double* dst0_y = (double*)(dst0 + y * dst0_step);
-            double* dst1_y = (double*)(dst1 + y * dst1_step);
-            double* dst2_y = (double*)(dst2 + y * dst2_step);
-            double* dst3_y = (double*)(dst3 + y * dst3_step);
-
-            if (x < cols && y < rows)
-            {
-                double2 src_elem1 = src_y[2 * x];
-                double2 src_elem2 = src_y[2 * x + 1];
-                dst0_y[x] = src_elem1.x;
-                dst1_y[x] = src_elem1.y;
-                dst2_y[x] = src_elem2.x;
-                dst3_y[x] = src_elem2.y;
-            }
-        }
-
-        template <typename T>
-        static void splitC2_(const PtrStepSzb& src, PtrStepSzb* dst, const cudaStream_t& stream)
-        {
-            dim3 block(32, 8);
-            dim3 grid(divUp(src.cols, block.x), divUp(src.rows, block.y));
-            splitC2_<T><<<grid, block, 0, stream>>>(
-                    src.data, src.step, src.rows, src.cols,
-                    dst[0].data, dst[0].step,
-                    dst[1].data, dst[1].step);
-            cudaSafeCall( cudaGetLastError() );
-
-            if (stream == 0)
-                cudaSafeCall(cudaDeviceSynchronize());
-        }
-
-
-        template <typename T>
-        static void splitC3_(const PtrStepSzb& src, PtrStepSzb* dst, const cudaStream_t& stream)
-        {
-            dim3 block(32, 8);
-            dim3 grid(divUp(src.cols, block.x), divUp(src.rows, block.y));
-            splitC3_<T><<<grid, block, 0, stream>>>(
-                    src.data, src.step, src.rows, src.cols,
-                    dst[0].data, dst[0].step,
-                    dst[1].data, dst[1].step,
-                    dst[2].data, dst[2].step);
-            cudaSafeCall( cudaGetLastError() );
-
-            if (stream == 0)
-                cudaSafeCall(cudaDeviceSynchronize());
-        }
-
-
-        template <typename T>
-        static void splitC4_(const PtrStepSzb& src, PtrStepSzb* dst, const cudaStream_t& stream)
-        {
-            dim3 block(32, 8);
-            dim3 grid(divUp(src.cols, block.x), divUp(src.rows, block.y));
-            splitC4_<T><<<grid, block, 0, stream>>>(
-                     src.data, src.step, src.rows, src.cols,
-                     dst[0].data, dst[0].step,
-                     dst[1].data, dst[1].step,
-                     dst[2].data, dst[2].step,
-                     dst[3].data, dst[3].step);
-            cudaSafeCall( cudaGetLastError() );
-
-            if (stream == 0)
-                cudaSafeCall(cudaDeviceSynchronize());
-        }
-
-
-        void split(const PtrStepSzb& src, PtrStepSzb* dst, int num_channels, size_t elem_size1, const cudaStream_t& stream)
-        {
-            static SplitFunction split_func_tbl[] =
-            {
-                splitC2_<char>, splitC2_<short>, splitC2_<int>, 0, splitC2_<double>,
-                splitC3_<char>, splitC3_<short>, splitC3_<int>, 0, splitC3_<double>,
-                splitC4_<char>, splitC4_<short>, splitC4_<int>, 0, splitC4_<double>,
+                globPtr<T>(dst[0]), globPtr<T>(dst[1])
             };
 
-            size_t split_func_id = (num_channels - 2) * 5 + (elem_size1 >> 1);
-            SplitFunction split_func = split_func_tbl[split_func_id];
-
-            if (split_func == 0)
-                CV_Error(cv::Error::StsUnsupportedFormat, "Unsupported channel count or data type");
-
-            split_func(src, dst, stream);
+            gridSplit(globPtr<typename MakeVec<T, 2>::type>(src), dstarr, stream);
         }
-    } // namespace split_merge
-}}} // namespace cv { namespace cuda { namespace cudev
+    };
 
+    template <typename T> struct SplitFunc<3, T>
+    {
+        static void call(const GpuMat& src, GpuMat* dst, Stream& stream)
+        {
+            GlobPtrSz<T> dstarr[3] =
+            {
+                globPtr<T>(dst[0]), globPtr<T>(dst[1]), globPtr<T>(dst[2])
+            };
 
-#endif /* CUDA_DISABLER */
+            gridSplit(globPtr<typename MakeVec<T, 3>::type>(src), dstarr, stream);
+        }
+    };
+
+    template <typename T> struct SplitFunc<4, T>
+    {
+        static void call(const GpuMat& src, GpuMat* dst, Stream& stream)
+        {
+            GlobPtrSz<T> dstarr[4] =
+            {
+                globPtr<T>(dst[0]), globPtr<T>(dst[1]), globPtr<T>(dst[2]), globPtr<T>(dst[3])
+            };
+
+            gridSplit(globPtr<typename MakeVec<T, 4>::type>(src), dstarr, stream);
+        }
+    };
+
+    void splitImpl(const GpuMat& src, GpuMat* dst, Stream& stream)
+    {
+        typedef void (*func_t)(const GpuMat& src, GpuMat* dst, Stream& stream);
+        static const func_t funcs[3][5] =
+        {
+            {SplitFunc<2, uchar>::call, SplitFunc<2, ushort>::call, SplitFunc<2, int>::call, 0, SplitFunc<2, double>::call},
+            {SplitFunc<3, uchar>::call, SplitFunc<3, ushort>::call, SplitFunc<3, int>::call, 0, SplitFunc<3, double>::call},
+            {SplitFunc<4, uchar>::call, SplitFunc<4, ushort>::call, SplitFunc<4, int>::call, 0, SplitFunc<4, double>::call}
+        };
+
+        CV_DbgAssert( dst != 0 );
+
+        const int depth = src.depth();
+        const int channels = src.channels();
+
+        CV_DbgAssert( channels <= 4 );
+
+        if (channels == 0)
+            return;
+
+        if (channels == 1)
+        {
+            src.copyTo(dst[0], stream);
+            return;
+        }
+
+        for (int i = 0; i < channels; ++i)
+            dst[i].create(src.size(), depth);
+
+        const func_t func = funcs[channels - 2][CV_ELEM_SIZE(depth) / 2];
+
+        if (func == 0)
+            CV_Error(cv::Error::StsUnsupportedFormat, "Unsupported channel count or data type");
+
+        func(src, dst, stream);
+    }
+}
+
+void cv::cuda::split(InputArray _src, GpuMat* dst, Stream& stream)
+{
+    GpuMat src = _src.getGpuMat();
+    splitImpl(src, dst, stream);
+}
+
+void cv::cuda::split(InputArray _src, std::vector<GpuMat>& dst, Stream& stream)
+{
+    GpuMat src = _src.getGpuMat();
+    dst.resize(src.channels());
+    if (src.channels() > 0)
+        splitImpl(src, &dst[0], stream);
+}
+
+#endif
diff --git a/modules/cudaarithm/src/cuda/sub_mat.cu b/modules/cudaarithm/src/cuda/sub_mat.cu
index 873b73c46a..6468692aee 100644
--- a/modules/cudaarithm/src/cuda/sub_mat.cu
+++ b/modules/cudaarithm/src/cuda/sub_mat.cu
@@ -40,146 +40,186 @@
 //
 //M*/
 
-#if !defined CUDA_DISABLER
+#include "opencv2/opencv_modules.hpp"
 
-#include "opencv2/core/cuda/common.hpp"
-#include "opencv2/core/cuda/functional.hpp"
-#include "opencv2/core/cuda/transform.hpp"
-#include "opencv2/core/cuda/saturate_cast.hpp"
-#include "opencv2/core/cuda/simd_functions.hpp"
+#ifndef HAVE_OPENCV_CUDEV
 
-#include "arithm_func_traits.hpp"
+#error "opencv_cudev is required"
 
-using namespace cv::cuda;
-using namespace cv::cuda::device;
+#else
 
-namespace arithm
+#include "opencv2/cudev.hpp"
+
+using namespace cv::cudev;
+
+void subMat(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat& mask, double, Stream& _stream, int);
+
+namespace
 {
-    struct VSub4 : binary_function<uint, uint, uint>
-    {
-        __device__ __forceinline__ uint operator ()(uint a, uint b) const
-        {
-            return vsub4(a, b);
-        }
-
-        __host__ __device__ __forceinline__ VSub4() {}
-        __host__ __device__ __forceinline__ VSub4(const VSub4&) {}
-    };
-
-    struct VSub2 : binary_function<uint, uint, uint>
-    {
-        __device__ __forceinline__ uint operator ()(uint a, uint b) const
-        {
-            return vsub2(a, b);
-        }
-
-        __host__ __device__ __forceinline__ VSub2() {}
-        __host__ __device__ __forceinline__ VSub2(const VSub2&) {}
-    };
-
-    template <typename T, typename D> struct SubMat : binary_function<T, T, D>
+    template <typename T, typename D> struct SubOp1 : binary_function<T, T, D>
     {
         __device__ __forceinline__ D operator ()(T a, T b) const
         {
             return saturate_cast<D>(a - b);
         }
-
-        __host__ __device__ __forceinline__ SubMat() {}
-        __host__ __device__ __forceinline__ SubMat(const SubMat&) {}
     };
-}
-
-namespace cv { namespace cuda { namespace device
-{
-    template <> struct TransformFunctorTraits< arithm::VSub4 > : arithm::ArithmFuncTraits<sizeof(uint), sizeof(uint)>
-    {
-    };
-
-    template <> struct TransformFunctorTraits< arithm::VSub2 > : arithm::ArithmFuncTraits<sizeof(uint), sizeof(uint)>
-    {
-    };
-
-    template <typename T, typename D> struct TransformFunctorTraits< arithm::SubMat<T, D> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(D)>
-    {
-    };
-}}}
-
-namespace arithm
-{
-    void subMat_v4(PtrStepSz<uint> src1, PtrStepSz<uint> src2, PtrStepSz<uint> dst, cudaStream_t stream)
-    {
-        device::transform(src1, src2, dst, VSub4(), WithOutMask(), stream);
-    }
-
-    void subMat_v2(PtrStepSz<uint> src1, PtrStepSz<uint> src2, PtrStepSz<uint> dst, cudaStream_t stream)
-    {
-        device::transform(src1, src2, dst, VSub2(), WithOutMask(), stream);
-    }
 
     template <typename T, typename D>
-    void subMat(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream)
+    void subMat_v1(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat& mask, Stream& stream)
     {
         if (mask.data)
-            device::transform((PtrStepSz<T>) src1, (PtrStepSz<T>) src2, (PtrStepSz<D>) dst, SubMat<T, D>(), mask, stream);
+            gridTransformBinary(globPtr<T>(src1), globPtr<T>(src2), globPtr<D>(dst), SubOp1<T, D>(), globPtr<uchar>(mask), stream);
         else
-            device::transform((PtrStepSz<T>) src1, (PtrStepSz<T>) src2, (PtrStepSz<D>) dst, SubMat<T, D>(), WithOutMask(), stream);
+            gridTransformBinary(globPtr<T>(src1), globPtr<T>(src2), globPtr<D>(dst), SubOp1<T, D>(), stream);
     }
 
-    template void subMat<uchar, uchar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subMat<uchar, schar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subMat<uchar, ushort>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subMat<uchar, short>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subMat<uchar, int>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subMat<uchar, float>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subMat<uchar, double>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    struct SubOp2 : binary_function<uint, uint, uint>
+    {
+        __device__ __forceinline__ uint operator ()(uint a, uint b) const
+        {
+            return vsub2(a, b);
+        }
+    };
 
-    template void subMat<schar, uchar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subMat<schar, schar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subMat<schar, ushort>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subMat<schar, short>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subMat<schar, int>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subMat<schar, float>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subMat<schar, double>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    void subMat_v2(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, Stream& stream)
+    {
+        const int vcols = src1.cols >> 1;
 
-    //template void subMat<ushort, uchar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void subMat<ushort, schar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subMat<ushort, ushort>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subMat<ushort, short>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subMat<ushort, int>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subMat<ushort, float>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subMat<ushort, double>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+        GlobPtrSz<uint> src1_ = globPtr((uint*) src1.data, src1.step, src1.rows, vcols);
+        GlobPtrSz<uint> src2_ = globPtr((uint*) src2.data, src2.step, src1.rows, vcols);
+        GlobPtrSz<uint> dst_ = globPtr((uint*) dst.data, dst.step, src1.rows, vcols);
 
-    //template void subMat<short, uchar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void subMat<short, schar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subMat<short, ushort>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subMat<short, short>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subMat<short, int>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subMat<short, float>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subMat<short, double>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+        gridTransformBinary(src1_, src2_, dst_, SubOp2(), stream);
+    }
 
-    //template void subMat<int, uchar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void subMat<int, schar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void subMat<int, ushort>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void subMat<int, short>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subMat<int, int>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subMat<int, float>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subMat<int, double>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    struct SubOp4 : binary_function<uint, uint, uint>
+    {
+        __device__ __forceinline__ uint operator ()(uint a, uint b) const
+        {
+            return vsub4(a, b);
+        }
+    };
 
-    //template void subMat<float, uchar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void subMat<float, schar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void subMat<float, ushort>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void subMat<float, short>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void subMat<float, int>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subMat<float, float>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subMat<float, double>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    void subMat_v4(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, Stream& stream)
+    {
+        const int vcols = src1.cols >> 2;
 
-    //template void subMat<double, uchar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void subMat<double, schar>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void subMat<double, ushort>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void subMat<double, short>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void subMat<double, int>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void subMat<double, float>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subMat<double, double>(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+        GlobPtrSz<uint> src1_ = globPtr((uint*) src1.data, src1.step, src1.rows, vcols);
+        GlobPtrSz<uint> src2_ = globPtr((uint*) src2.data, src2.step, src1.rows, vcols);
+        GlobPtrSz<uint> dst_ = globPtr((uint*) dst.data, dst.step, src1.rows, vcols);
+
+        gridTransformBinary(src1_, src2_, dst_, SubOp4(), stream);
+    }
 }
 
-#endif // CUDA_DISABLER
+void subMat(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat& mask, double, Stream& stream, int)
+{
+    typedef void (*func_t)(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat& mask, Stream& stream);
+    static const func_t funcs[7][7] =
+    {
+        {
+            subMat_v1<uchar, uchar>,
+            subMat_v1<uchar, schar>,
+            subMat_v1<uchar, ushort>,
+            subMat_v1<uchar, short>,
+            subMat_v1<uchar, int>,
+            subMat_v1<uchar, float>,
+            subMat_v1<uchar, double>
+        },
+        {
+            subMat_v1<schar, uchar>,
+            subMat_v1<schar, schar>,
+            subMat_v1<schar, ushort>,
+            subMat_v1<schar, short>,
+            subMat_v1<schar, int>,
+            subMat_v1<schar, float>,
+            subMat_v1<schar, double>
+        },
+        {
+            0 /*subMat_v1<ushort, uchar>*/,
+            0 /*subMat_v1<ushort, schar>*/,
+            subMat_v1<ushort, ushort>,
+            subMat_v1<ushort, short>,
+            subMat_v1<ushort, int>,
+            subMat_v1<ushort, float>,
+            subMat_v1<ushort, double>
+        },
+        {
+            0 /*subMat_v1<short, uchar>*/,
+            0 /*subMat_v1<short, schar>*/,
+            subMat_v1<short, ushort>,
+            subMat_v1<short, short>,
+            subMat_v1<short, int>,
+            subMat_v1<short, float>,
+            subMat_v1<short, double>
+        },
+        {
+            0 /*subMat_v1<int, uchar>*/,
+            0 /*subMat_v1<int, schar>*/,
+            0 /*subMat_v1<int, ushort>*/,
+            0 /*subMat_v1<int, short>*/,
+            subMat_v1<int, int>,
+            subMat_v1<int, float>,
+            subMat_v1<int, double>
+        },
+        {
+            0 /*subMat_v1<float, uchar>*/,
+            0 /*subMat_v1<float, schar>*/,
+            0 /*subMat_v1<float, ushort>*/,
+            0 /*subMat_v1<float, short>*/,
+            0 /*subMat_v1<float, int>*/,
+            subMat_v1<float, float>,
+            subMat_v1<float, double>
+        },
+        {
+            0 /*subMat_v1<double, uchar>*/,
+            0 /*subMat_v1<double, schar>*/,
+            0 /*subMat_v1<double, ushort>*/,
+            0 /*subMat_v1<double, short>*/,
+            0 /*subMat_v1<double, int>*/,
+            0 /*subMat_v1<double, float>*/,
+            subMat_v1<double, double>
+        }
+    };
+
+    const int sdepth = src1.depth();
+    const int ddepth = dst.depth();
+
+    CV_DbgAssert( sdepth <= CV_64F && ddepth <= CV_64F );
+
+    GpuMat src1_ = src1.reshape(1);
+    GpuMat src2_ = src2.reshape(1);
+    GpuMat dst_ = dst.reshape(1);
+
+    if (mask.empty() && (sdepth == CV_8U || sdepth == CV_16U) && ddepth == sdepth)
+    {
+        const intptr_t src1ptr = reinterpret_cast<intptr_t>(src1_.data);
+        const intptr_t src2ptr = reinterpret_cast<intptr_t>(src2_.data);
+        const intptr_t dstptr = reinterpret_cast<intptr_t>(dst_.data);
+
+        const bool isAllAligned = (src1ptr & 31) == 0 && (src2ptr & 31) == 0 && (dstptr & 31) == 0;
+
+        if (isAllAligned)
+        {
+            if (sdepth == CV_8U && (src1_.cols & 3) == 0)
+            {
+                subMat_v4(src1_, src2_, dst_, stream);
+                return;
+            }
+            else if (sdepth == CV_16U && (src1_.cols & 1) == 0)
+            {
+                subMat_v2(src1_, src2_, dst_, stream);
+                return;
+            }
+        }
+    }
+
+    const func_t func = funcs[sdepth][ddepth];
+
+    if (!func)
+        CV_Error(cv::Error::StsUnsupportedFormat, "Unsupported combination of source and destination types");
+
+    func(src1_, src2_, dst_, mask, stream);
+}
+
+#endif
diff --git a/modules/cudaarithm/src/cuda/sub_scalar.cu b/modules/cudaarithm/src/cuda/sub_scalar.cu
index 5f4ef66a9c..c4eeec0148 100644
--- a/modules/cudaarithm/src/cuda/sub_scalar.cu
+++ b/modules/cudaarithm/src/cuda/sub_scalar.cu
@@ -40,110 +40,164 @@
 //
 //M*/
 
-#if !defined CUDA_DISABLER
+#include "opencv2/opencv_modules.hpp"
 
-#include "opencv2/core/cuda/common.hpp"
-#include "opencv2/core/cuda/functional.hpp"
-#include "opencv2/core/cuda/transform.hpp"
-#include "opencv2/core/cuda/saturate_cast.hpp"
-#include "opencv2/core/cuda/simd_functions.hpp"
+#ifndef HAVE_OPENCV_CUDEV
 
-#include "arithm_func_traits.hpp"
+#error "opencv_cudev is required"
 
-using namespace cv::cuda;
-using namespace cv::cuda::device;
+#else
 
-namespace arithm
+#include "opencv2/cudev.hpp"
+
+using namespace cv::cudev;
+
+void subScalar(const GpuMat& src, cv::Scalar val, bool inv, GpuMat& dst, const GpuMat& mask, double, Stream& stream, int);
+
+namespace
 {
-    template <typename T, typename S, typename D> struct SubScalar : unary_function<T, D>
+    template <typename SrcType, typename ScalarType, typename DstType> struct SubScalarOp : unary_function<SrcType, DstType>
     {
-        S val;
-        int scale;
+        ScalarType val;
 
-        __host__ SubScalar(S val_, int scale_) : val(val_), scale(scale_) {}
-
-        __device__ __forceinline__ D operator ()(T a) const
+        __device__ __forceinline__ DstType operator ()(SrcType a) const
         {
-            return saturate_cast<D>(scale * (a - val));
+            return saturate_cast<DstType>(saturate_cast<ScalarType>(a) - val);
         }
     };
-}
 
-namespace cv { namespace cuda { namespace device
-{
-    template <typename T, typename S, typename D> struct TransformFunctorTraits< arithm::SubScalar<T, S, D> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(D)>
+    template <typename SrcType, typename ScalarType, typename DstType> struct SubScalarOpInv : unary_function<SrcType, DstType>
+    {
+        ScalarType val;
+
+        __device__ __forceinline__ DstType operator ()(SrcType a) const
+        {
+            return saturate_cast<DstType>(val - saturate_cast<ScalarType>(a));
+        }
+    };
+
+    template <typename ScalarDepth> struct TransformPolicy : DefaultTransformPolicy
     {
     };
-}}}
-
-namespace arithm
-{
-    template <typename T, typename S, typename D>
-    void subScalar(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream)
+    template <> struct TransformPolicy<double> : DefaultTransformPolicy
     {
-        SubScalar<T, S, D> op(static_cast<S>(val), inv ? -1 : 1);
+        enum {
+            shift = 1
+        };
+    };
 
-        if (mask.data)
-            device::transform((PtrStepSz<T>) src1, (PtrStepSz<D>) dst, op, mask, stream);
+    template <typename SrcType, typename ScalarDepth, typename DstType>
+    void subScalarImpl(const GpuMat& src, cv::Scalar value, bool inv, GpuMat& dst, const GpuMat& mask, Stream& stream)
+    {
+        typedef typename MakeVec<ScalarDepth, VecTraits<SrcType>::cn>::type ScalarType;
+
+        cv::Scalar_<ScalarDepth> value_ = value;
+
+        if (inv)
+        {
+            SubScalarOpInv<SrcType, ScalarType, DstType> op;
+            op.val = VecTraits<ScalarType>::make(value_.val);
+
+            if (mask.data)
+                gridTransformUnary_< TransformPolicy<ScalarDepth> >(globPtr<SrcType>(src), globPtr<DstType>(dst), op, globPtr<uchar>(mask), stream);
+            else
+                gridTransformUnary_< TransformPolicy<ScalarDepth> >(globPtr<SrcType>(src), globPtr<DstType>(dst), op, stream);
+        }
         else
-            device::transform((PtrStepSz<T>) src1, (PtrStepSz<D>) dst, op, WithOutMask(), stream);
+        {
+            SubScalarOp<SrcType, ScalarType, DstType> op;
+            op.val = VecTraits<ScalarType>::make(value_.val);
+
+            if (mask.data)
+                gridTransformUnary_< TransformPolicy<ScalarDepth> >(globPtr<SrcType>(src), globPtr<DstType>(dst), op, globPtr<uchar>(mask), stream);
+            else
+                gridTransformUnary_< TransformPolicy<ScalarDepth> >(globPtr<SrcType>(src), globPtr<DstType>(dst), op, stream);
+        }
     }
-
-    template void subScalar<uchar, float, uchar>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subScalar<uchar, float, schar>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subScalar<uchar, float, ushort>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subScalar<uchar, float, short>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subScalar<uchar, float, int>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subScalar<uchar, float, float>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subScalar<uchar, double, double>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-
-    template void subScalar<schar, float, uchar>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subScalar<schar, float, schar>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subScalar<schar, float, ushort>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subScalar<schar, float, short>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subScalar<schar, float, int>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subScalar<schar, float, float>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subScalar<schar, double, double>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-
-    //template void subScalar<ushort, float, uchar>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void subScalar<ushort, float, schar>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subScalar<ushort, float, ushort>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subScalar<ushort, float, short>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subScalar<ushort, float, int>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subScalar<ushort, float, float>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subScalar<ushort, double, double>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-
-    //template void subScalar<short, float, uchar>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void subScalar<short, float, schar>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subScalar<short, float, ushort>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subScalar<short, float, short>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subScalar<short, float, int>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subScalar<short, float, float>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subScalar<short, double, double>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-
-    //template void subScalar<int, float, uchar>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void subScalar<int, float, schar>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void subScalar<int, float, ushort>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void subScalar<int, float, short>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subScalar<int, float, int>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subScalar<int, float, float>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subScalar<int, double, double>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-
-    //template void subScalar<float, float, uchar>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void subScalar<float, float, schar>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void subScalar<float, float, ushort>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void subScalar<float, float, short>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void subScalar<float, float, int>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subScalar<float, float, float>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subScalar<float, double, double>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-
-    //template void subScalar<double, double, uchar>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void subScalar<double, double, schar>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void subScalar<double, double, ushort>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void subScalar<double, double, short>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void subScalar<double, double, int>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    //template void subScalar<double, double, float>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template void subScalar<double, double, double>(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
 }
 
-#endif // CUDA_DISABLER
+void subScalar(const GpuMat& src, cv::Scalar val, bool inv, GpuMat& dst, const GpuMat& mask, double, Stream& stream, int)
+{
+    typedef void (*func_t)(const GpuMat& src, cv::Scalar val, bool inv, GpuMat& dst, const GpuMat& mask, Stream& stream);
+    static const func_t funcs[7][7][4] =
+    {
+        {
+            {subScalarImpl<uchar, float, uchar>, subScalarImpl<uchar2, float, uchar2>, subScalarImpl<uchar3, float, uchar3>, subScalarImpl<uchar4, float, uchar4>},
+            {subScalarImpl<uchar, float, schar>, subScalarImpl<uchar2, float, char2>, subScalarImpl<uchar3, float, char3>, subScalarImpl<uchar4, float, char4>},
+            {subScalarImpl<uchar, float, ushort>, subScalarImpl<uchar2, float, ushort2>, subScalarImpl<uchar3, float, ushort3>, subScalarImpl<uchar4, float, ushort4>},
+            {subScalarImpl<uchar, float, short>, subScalarImpl<uchar2, float, short2>, subScalarImpl<uchar3, float, short3>, subScalarImpl<uchar4, float, short4>},
+            {subScalarImpl<uchar, float, int>, subScalarImpl<uchar2, float, int2>, subScalarImpl<uchar3, float, int3>, subScalarImpl<uchar4, float, int4>},
+            {subScalarImpl<uchar, float, float>, subScalarImpl<uchar2, float, float2>, subScalarImpl<uchar3, float, float3>, subScalarImpl<uchar4, float, float4>},
+            {subScalarImpl<uchar, double, double>, subScalarImpl<uchar2, double, double2>, subScalarImpl<uchar3, double, double3>, subScalarImpl<uchar4, double, double4>}
+        },
+        {
+            {subScalarImpl<schar, float, uchar>, subScalarImpl<char2, float, uchar2>, subScalarImpl<char3, float, uchar3>, subScalarImpl<char4, float, uchar4>},
+            {subScalarImpl<schar, float, schar>, subScalarImpl<char2, float, char2>, subScalarImpl<char3, float, char3>, subScalarImpl<char4, float, char4>},
+            {subScalarImpl<schar, float, ushort>, subScalarImpl<char2, float, ushort2>, subScalarImpl<char3, float, ushort3>, subScalarImpl<char4, float, ushort4>},
+            {subScalarImpl<schar, float, short>, subScalarImpl<char2, float, short2>, subScalarImpl<char3, float, short3>, subScalarImpl<char4, float, short4>},
+            {subScalarImpl<schar, float, int>, subScalarImpl<char2, float, int2>, subScalarImpl<char3, float, int3>, subScalarImpl<char4, float, int4>},
+            {subScalarImpl<schar, float, float>, subScalarImpl<char2, float, float2>, subScalarImpl<char3, float, float3>, subScalarImpl<char4, float, float4>},
+            {subScalarImpl<schar, double, double>, subScalarImpl<char2, double, double2>, subScalarImpl<char3, double, double3>, subScalarImpl<char4, double, double4>}
+        },
+        {
+            {0 /*subScalarImpl<ushort, float, uchar>*/, 0 /*subScalarImpl<ushort2, float, uchar2>*/, 0 /*subScalarImpl<ushort3, float, uchar3>*/, 0 /*subScalarImpl<ushort4, float, uchar4>*/},
+            {0 /*subScalarImpl<ushort, float, schar>*/, 0 /*subScalarImpl<ushort2, float, char2>*/, 0 /*subScalarImpl<ushort3, float, char3>*/, 0 /*subScalarImpl<ushort4, float, char4>*/},
+            {subScalarImpl<ushort, float, ushort>, subScalarImpl<ushort2, float, ushort2>, subScalarImpl<ushort3, float, ushort3>, subScalarImpl<ushort4, float, ushort4>},
+            {subScalarImpl<ushort, float, short>, subScalarImpl<ushort2, float, short2>, subScalarImpl<ushort3, float, short3>, subScalarImpl<ushort4, float, short4>},
+            {subScalarImpl<ushort, float, int>, subScalarImpl<ushort2, float, int2>, subScalarImpl<ushort3, float, int3>, subScalarImpl<ushort4, float, int4>},
+            {subScalarImpl<ushort, float, float>, subScalarImpl<ushort2, float, float2>, subScalarImpl<ushort3, float, float3>, subScalarImpl<ushort4, float, float4>},
+            {subScalarImpl<ushort, double, double>, subScalarImpl<ushort2, double, double2>, subScalarImpl<ushort3, double, double3>, subScalarImpl<ushort4, double, double4>}
+        },
+        {
+            {0 /*subScalarImpl<short, float, uchar>*/, 0 /*subScalarImpl<short2, float, uchar2>*/, 0 /*subScalarImpl<short3, float, uchar3>*/, 0 /*subScalarImpl<short4, float, uchar4>*/},
+            {0 /*subScalarImpl<short, float, schar>*/, 0 /*subScalarImpl<short2, float, char2>*/, 0 /*subScalarImpl<short3, float, char3>*/, 0 /*subScalarImpl<short4, float, char4>*/},
+            {subScalarImpl<short, float, ushort>, subScalarImpl<short2, float, ushort2>, subScalarImpl<short3, float, ushort3>, subScalarImpl<short4, float, ushort4>},
+            {subScalarImpl<short, float, short>, subScalarImpl<short2, float, short2>, subScalarImpl<short3, float, short3>, subScalarImpl<short4, float, short4>},
+            {subScalarImpl<short, float, int>, subScalarImpl<short2, float, int2>, subScalarImpl<short3, float, int3>, subScalarImpl<short4, float, int4>},
+            {subScalarImpl<short, float, float>, subScalarImpl<short2, float, float2>, subScalarImpl<short3, float, float3>, subScalarImpl<short4, float, float4>},
+            {subScalarImpl<short, double, double>, subScalarImpl<short2, double, double2>, subScalarImpl<short3, double, double3>, subScalarImpl<short4, double, double4>}
+        },
+        {
+            {0 /*subScalarImpl<int, float, uchar>*/, 0 /*subScalarImpl<int2, float, uchar2>*/, 0 /*subScalarImpl<int3, float, uchar3>*/, 0 /*subScalarImpl<int4, float, uchar4>*/},
+            {0 /*subScalarImpl<int, float, schar>*/, 0 /*subScalarImpl<int2, float, char2>*/, 0 /*subScalarImpl<int3, float, char3>*/, 0 /*subScalarImpl<int4, float, char4>*/},
+            {0 /*subScalarImpl<int, float, ushort>*/, 0 /*subScalarImpl<int2, float, ushort2>*/, 0 /*subScalarImpl<int3, float, ushort3>*/, 0 /*subScalarImpl<int4, float, ushort4>*/},
+            {0 /*subScalarImpl<int, float, short>*/, 0 /*subScalarImpl<int2, float, short2>*/, 0 /*subScalarImpl<int3, float, short3>*/, 0 /*subScalarImpl<int4, float, short4>*/},
+            {subScalarImpl<int, float, int>, subScalarImpl<int2, float, int2>, subScalarImpl<int3, float, int3>, subScalarImpl<int4, float, int4>},
+            {subScalarImpl<int, float, float>, subScalarImpl<int2, float, float2>, subScalarImpl<int3, float, float3>, subScalarImpl<int4, float, float4>},
+            {subScalarImpl<int, double, double>, subScalarImpl<int2, double, double2>, subScalarImpl<int3, double, double3>, subScalarImpl<int4, double, double4>}
+        },
+        {
+            {0 /*subScalarImpl<float, float, uchar>*/, 0 /*subScalarImpl<float2, float, uchar2>*/, 0 /*subScalarImpl<float3, float, uchar3>*/, 0 /*subScalarImpl<float4, float, uchar4>*/},
+            {0 /*subScalarImpl<float, float, schar>*/, 0 /*subScalarImpl<float2, float, char2>*/, 0 /*subScalarImpl<float3, float, char3>*/, 0 /*subScalarImpl<float4, float, char4>*/},
+            {0 /*subScalarImpl<float, float, ushort>*/, 0 /*subScalarImpl<float2, float, ushort2>*/, 0 /*subScalarImpl<float3, float, ushort3>*/, 0 /*subScalarImpl<float4, float, ushort4>*/},
+            {0 /*subScalarImpl<float, float, short>*/, 0 /*subScalarImpl<float2, float, short2>*/, 0 /*subScalarImpl<float3, float, short3>*/, 0 /*subScalarImpl<float4, float, short4>*/},
+            {0 /*subScalarImpl<float, float, int>*/, 0 /*subScalarImpl<float2, float, int2>*/, 0 /*subScalarImpl<float3, float, int3>*/, 0 /*subScalarImpl<float4, float, int4>*/},
+            {subScalarImpl<float, float, float>, subScalarImpl<float2, float, float2>, subScalarImpl<float3, float, float3>, subScalarImpl<float4, float, float4>},
+            {subScalarImpl<float, double, double>, subScalarImpl<float2, double, double2>, subScalarImpl<float3, double, double3>, subScalarImpl<float4, double, double4>}
+        },
+        {
+            {0 /*subScalarImpl<double, double, uchar>*/, 0 /*subScalarImpl<double2, double, uchar2>*/, 0 /*subScalarImpl<double3, double, uchar3>*/, 0 /*subScalarImpl<double4, double, uchar4>*/},
+            {0 /*subScalarImpl<double, double, schar>*/, 0 /*subScalarImpl<double2, double, char2>*/, 0 /*subScalarImpl<double3, double, char3>*/, 0 /*subScalarImpl<double4, double, char4>*/},
+            {0 /*subScalarImpl<double, double, ushort>*/, 0 /*subScalarImpl<double2, double, ushort2>*/, 0 /*subScalarImpl<double3, double, ushort3>*/, 0 /*subScalarImpl<double4, double, ushort4>*/},
+            {0 /*subScalarImpl<double, double, short>*/, 0 /*subScalarImpl<double2, double, short2>*/, 0 /*subScalarImpl<double3, double, short3>*/, 0 /*subScalarImpl<double4, double, short4>*/},
+            {0 /*subScalarImpl<double, double, int>*/, 0 /*subScalarImpl<double2, double, int2>*/, 0 /*subScalarImpl<double3, double, int3>*/, 0 /*subScalarImpl<double4, double, int4>*/},
+            {0 /*subScalarImpl<double, double, float>*/, 0 /*subScalarImpl<double2, double, float2>*/, 0 /*subScalarImpl<double3, double, float3>*/, 0 /*subScalarImpl<double4, double, float4>*/},
+            {subScalarImpl<double, double, double>, subScalarImpl<double2, double, double2>, subScalarImpl<double3, double, double3>, subScalarImpl<double4, double, double4>}
+        }
+    };
+
+    const int sdepth = src.depth();
+    const int ddepth = dst.depth();
+    const int cn = src.channels();
+
+    CV_DbgAssert( sdepth <= CV_64F && ddepth <= CV_64F && cn <= 4 );
+
+    const func_t func = funcs[sdepth][ddepth][cn - 1];
+
+    if (!func)
+        CV_Error(cv::Error::StsUnsupportedFormat, "Unsupported combination of source and destination types");
+
+    func(src, val, inv, dst, mask, stream);
+}
+
+#endif
diff --git a/modules/cudaarithm/src/cuda/sum.cu b/modules/cudaarithm/src/cuda/sum.cu
index cd8c3a1057..cced9c56e8 100644
--- a/modules/cudaarithm/src/cuda/sum.cu
+++ b/modules/cudaarithm/src/cuda/sum.cu
@@ -40,342 +40,155 @@
 //
 //M*/
 
-#if !defined CUDA_DISABLER
+#include "opencv2/opencv_modules.hpp"
 
-#include "opencv2/core/cuda/common.hpp"
-#include "opencv2/core/cuda/vec_traits.hpp"
-#include "opencv2/core/cuda/vec_math.hpp"
-#include "opencv2/core/cuda/functional.hpp"
-#include "opencv2/core/cuda/reduce.hpp"
-#include "opencv2/core/cuda/emulation.hpp"
-#include "opencv2/core/cuda/utility.hpp"
+#ifndef HAVE_OPENCV_CUDEV
 
-#include "unroll_detail.hpp"
+#error "opencv_cudev is required"
 
-using namespace cv::cuda;
-using namespace cv::cuda::device;
+#else
 
-namespace sum
+#include "opencv2/cudaarithm.hpp"
+#include "opencv2/cudev.hpp"
+
+using namespace cv::cudev;
+
+namespace
 {
-    __device__ unsigned int blocks_finished = 0;
-
-    template <typename R, int cn> struct AtomicAdd;
-    template <typename R> struct AtomicAdd<R, 1>
+    template <typename T, typename R, int cn>
+    cv::Scalar sumImpl(const GpuMat& _src, const GpuMat& mask, GpuMat& _buf)
     {
-        static __device__ void run(R* ptr, R val)
-        {
-            Emulation::glob::atomicAdd(ptr, val);
-        }
-    };
-    template <typename R> struct AtomicAdd<R, 2>
-    {
-        typedef typename TypeVec<R, 2>::vec_type val_type;
+        typedef typename MakeVec<T, cn>::type src_type;
+        typedef typename MakeVec<R, cn>::type res_type;
 
-        static __device__ void run(R* ptr, val_type val)
-        {
-            Emulation::glob::atomicAdd(ptr, val.x);
-            Emulation::glob::atomicAdd(ptr + 1, val.y);
-        }
-    };
-    template <typename R> struct AtomicAdd<R, 3>
-    {
-        typedef typename TypeVec<R, 3>::vec_type val_type;
+        const GpuMat_<src_type>& src = (const GpuMat_<src_type>&) _src;
+        GpuMat_<res_type>& buf = (GpuMat_<res_type>&) _buf;
 
-        static __device__ void run(R* ptr, val_type val)
-        {
-            Emulation::glob::atomicAdd(ptr, val.x);
-            Emulation::glob::atomicAdd(ptr + 1, val.y);
-            Emulation::glob::atomicAdd(ptr + 2, val.z);
-        }
-    };
-    template <typename R> struct AtomicAdd<R, 4>
-    {
-        typedef typename TypeVec<R, 4>::vec_type val_type;
-
-        static __device__ void run(R* ptr, val_type val)
-        {
-            Emulation::glob::atomicAdd(ptr, val.x);
-            Emulation::glob::atomicAdd(ptr + 1, val.y);
-            Emulation::glob::atomicAdd(ptr + 2, val.z);
-            Emulation::glob::atomicAdd(ptr + 3, val.w);
-        }
-    };
-
-    template <int BLOCK_SIZE, typename R, int cn>
-    struct GlobalReduce
-    {
-        typedef typename TypeVec<R, cn>::vec_type result_type;
-
-        static __device__ void run(result_type& sum, result_type* result, int tid, int bid, R* smem)
-        {
-        #if __CUDA_ARCH__ >= 200
-            if (tid == 0)
-                AtomicAdd<R, cn>::run((R*) result, sum);
-        #else
-            __shared__ bool is_last;
-
-            if (tid == 0)
-            {
-                result[bid] = sum;
-
-                __threadfence();
-
-                unsigned int ticket = ::atomicAdd(&blocks_finished, 1);
-                is_last = (ticket == gridDim.x * gridDim.y - 1);
-            }
-
-            __syncthreads();
-
-            if (is_last)
-            {
-                sum = tid < gridDim.x * gridDim.y ? result[tid] : VecTraits<result_type>::all(0);
-
-                device::reduce<BLOCK_SIZE>(detail::Unroll<cn>::template smem_tuple<BLOCK_SIZE>(smem), detail::Unroll<cn>::tie(sum), tid, detail::Unroll<cn>::op(plus<R>()));
-
-                if (tid == 0)
-                {
-                    result[0] = sum;
-                    blocks_finished = 0;
-                }
-            }
-        #endif
-        }
-    };
-
-    template <int BLOCK_SIZE, typename src_type, typename result_type, class Mask, class Op>
-    __global__ void kernel(const PtrStepSz<src_type> src, result_type* result, const Mask mask, const Op op, const int twidth, const int theight)
-    {
-        typedef typename VecTraits<src_type>::elem_type T;
-        typedef typename VecTraits<result_type>::elem_type R;
-        const int cn = VecTraits<src_type>::cn;
-
-        __shared__ R smem[BLOCK_SIZE * cn];
-
-        const int x0 = blockIdx.x * blockDim.x * twidth + threadIdx.x;
-        const int y0 = blockIdx.y * blockDim.y * theight + threadIdx.y;
-
-        const int tid = threadIdx.y * blockDim.x + threadIdx.x;
-        const int bid = blockIdx.y * gridDim.x + blockIdx.x;
-
-        result_type sum = VecTraits<result_type>::all(0);
-
-        for (int i = 0, y = y0; i < theight && y < src.rows; ++i, y += blockDim.y)
-        {
-            const src_type* ptr = src.ptr(y);
-
-            for (int j = 0, x = x0; j < twidth && x < src.cols; ++j, x += blockDim.x)
-            {
-                if (mask(y, x))
-                {
-                    const src_type srcVal = ptr[x];
-                    sum = sum + op(saturate_cast<result_type>(srcVal));
-                }
-            }
-        }
-
-        device::reduce<BLOCK_SIZE>(detail::Unroll<cn>::template smem_tuple<BLOCK_SIZE>(smem), detail::Unroll<cn>::tie(sum), tid, detail::Unroll<cn>::op(plus<R>()));
-
-        GlobalReduce<BLOCK_SIZE, R, cn>::run(sum, result, tid, bid, smem);
-    }
-
-    const int threads_x = 32;
-    const int threads_y = 8;
-
-    void getLaunchCfg(int cols, int rows, dim3& block, dim3& grid)
-    {
-        block = dim3(threads_x, threads_y);
-
-        grid = dim3(divUp(cols, block.x * block.y),
-                    divUp(rows, block.y * block.x));
-
-        grid.x = ::min(grid.x, block.x);
-        grid.y = ::min(grid.y, block.y);
-    }
-
-    void getBufSize(int cols, int rows, int cn, int& bufcols, int& bufrows)
-    {
-        dim3 block, grid;
-        getLaunchCfg(cols, rows, block, grid);
-
-        bufcols = grid.x * grid.y * sizeof(double) * cn;
-        bufrows = 1;
-    }
-
-    template <typename T, typename R, int cn, template <typename> class Op>
-    void caller(PtrStepSzb src_, void* buf_, double* out, PtrStepSzb mask)
-    {
-        typedef typename TypeVec<T, cn>::vec_type src_type;
-        typedef typename TypeVec<R, cn>::vec_type result_type;
-
-        PtrStepSz<src_type> src(src_);
-        result_type* buf = (result_type*) buf_;
-
-        dim3 block, grid;
-        getLaunchCfg(src.cols, src.rows, block, grid);
-
-        const int twidth = divUp(divUp(src.cols, grid.x), block.x);
-        const int theight = divUp(divUp(src.rows, grid.y), block.y);
-
-        Op<result_type> op;
-
-        if (mask.data)
-            kernel<threads_x * threads_y><<<grid, block>>>(src, buf, SingleMask(mask), op, twidth, theight);
+        if (mask.empty())
+            gridCalcSum(src, buf);
         else
-            kernel<threads_x * threads_y><<<grid, block>>>(src, buf, WithOutMask(), op, twidth, theight);
-        cudaSafeCall( cudaGetLastError() );
+            gridCalcSum(src, buf, globPtr<uchar>(mask));
 
-        cudaSafeCall( cudaDeviceSynchronize() );
+        cv::Scalar_<R> res;
+        cv::Mat res_mat(buf.size(), buf.type(), res.val);
+        buf.download(res_mat);
 
-        R result[4] = {0, 0, 0, 0};
-        cudaSafeCall( cudaMemcpy(&result, buf, sizeof(result_type), cudaMemcpyDeviceToHost) );
-
-        out[0] = result[0];
-        out[1] = result[1];
-        out[2] = result[2];
-        out[3] = result[3];
+        return res;
     }
 
-    template <typename T> struct SumType;
-    template <> struct SumType<uchar> { typedef unsigned int R; };
-    template <> struct SumType<schar> { typedef int R; };
-    template <> struct SumType<ushort> { typedef unsigned int R; };
-    template <> struct SumType<short> { typedef int R; };
-    template <> struct SumType<int> { typedef int R; };
-    template <> struct SumType<float> { typedef float R; };
-    template <> struct SumType<double> { typedef double R; };
-
-    template <typename T, int cn>
-    void run(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask)
+    template <typename T, typename R, int cn>
+    cv::Scalar sumAbsImpl(const GpuMat& _src, const GpuMat& mask, GpuMat& _buf)
     {
-        typedef typename SumType<T>::R R;
-        caller<T, R, cn, identity>(src, buf, out, mask);
+        typedef typename MakeVec<T, cn>::type src_type;
+        typedef typename MakeVec<R, cn>::type res_type;
+
+        const GpuMat_<src_type>& src = (const GpuMat_<src_type>&) _src;
+        GpuMat_<res_type>& buf = (GpuMat_<res_type>&) _buf;
+
+        if (mask.empty())
+            gridCalcSum(abs_(cvt_<res_type>(src)), buf);
+        else
+            gridCalcSum(abs_(cvt_<res_type>(src)), buf, globPtr<uchar>(mask));
+
+        cv::Scalar_<R> res;
+        cv::Mat res_mat(buf.size(), buf.type(), res.val);
+        buf.download(res_mat);
+
+        return res;
     }
 
-    template void run<uchar, 1>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void run<uchar, 2>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void run<uchar, 3>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void run<uchar, 4>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-
-    template void run<schar, 1>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void run<schar, 2>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void run<schar, 3>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void run<schar, 4>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-
-    template void run<ushort, 1>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void run<ushort, 2>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void run<ushort, 3>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void run<ushort, 4>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-
-    template void run<short, 1>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void run<short, 2>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void run<short, 3>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void run<short, 4>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-
-    template void run<int, 1>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void run<int, 2>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void run<int, 3>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void run<int, 4>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-
-    template void run<float, 1>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void run<float, 2>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void run<float, 3>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void run<float, 4>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-
-    template void run<double, 1>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void run<double, 2>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void run<double, 3>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void run<double, 4>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-
-    template <typename T, int cn>
-    void runAbs(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask)
+    template <typename T, typename R, int cn>
+    cv::Scalar sumSqrImpl(const GpuMat& _src, const GpuMat& mask, GpuMat& _buf)
     {
-        typedef typename SumType<T>::R R;
-        caller<T, R, cn, abs_func>(src, buf, out, mask);
+        typedef typename MakeVec<T, cn>::type src_type;
+        typedef typename MakeVec<R, cn>::type res_type;
+
+        const GpuMat_<src_type>& src = (const GpuMat_<src_type>&) _src;
+        GpuMat_<res_type>& buf = (GpuMat_<res_type>&) _buf;
+
+        if (mask.empty())
+            gridCalcSum(sqr_(cvt_<res_type>(src)), buf);
+        else
+            gridCalcSum(sqr_(cvt_<res_type>(src)), buf, globPtr<uchar>(mask));
+
+        cv::Scalar_<R> res;
+        cv::Mat res_mat(buf.size(), buf.type(), res.val);
+        buf.download(res_mat);
+
+        return res;
     }
-
-    template void runAbs<uchar, 1>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void runAbs<uchar, 2>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void runAbs<uchar, 3>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void runAbs<uchar, 4>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-
-    template void runAbs<schar, 1>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void runAbs<schar, 2>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void runAbs<schar, 3>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void runAbs<schar, 4>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-
-    template void runAbs<ushort, 1>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void runAbs<ushort, 2>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void runAbs<ushort, 3>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void runAbs<ushort, 4>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-
-    template void runAbs<short, 1>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void runAbs<short, 2>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void runAbs<short, 3>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void runAbs<short, 4>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-
-    template void runAbs<int, 1>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void runAbs<int, 2>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void runAbs<int, 3>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void runAbs<int, 4>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-
-    template void runAbs<float, 1>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void runAbs<float, 2>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void runAbs<float, 3>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void runAbs<float, 4>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-
-    template void runAbs<double, 1>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void runAbs<double, 2>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void runAbs<double, 3>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void runAbs<double, 4>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-
-    template <typename T> struct Sqr : unary_function<T, T>
-    {
-        __device__ __forceinline__ T operator ()(T x) const
-        {
-            return x * x;
-        }
-    };
-
-    template <typename T, int cn>
-    void runSqr(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask)
-    {
-        caller<T, double, cn, Sqr>(src, buf, out, mask);
-    }
-
-    template void runSqr<uchar, 1>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void runSqr<uchar, 2>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void runSqr<uchar, 3>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void runSqr<uchar, 4>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-
-    template void runSqr<schar, 1>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void runSqr<schar, 2>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void runSqr<schar, 3>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void runSqr<schar, 4>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-
-    template void runSqr<ushort, 1>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void runSqr<ushort, 2>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void runSqr<ushort, 3>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void runSqr<ushort, 4>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-
-    template void runSqr<short, 1>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void runSqr<short, 2>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void runSqr<short, 3>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void runSqr<short, 4>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-
-    template void runSqr<int, 1>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void runSqr<int, 2>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void runSqr<int, 3>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void runSqr<int, 4>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-
-    template void runSqr<float, 1>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void runSqr<float, 2>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void runSqr<float, 3>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void runSqr<float, 4>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-
-    template void runSqr<double, 1>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void runSqr<double, 2>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void runSqr<double, 3>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
-    template void runSqr<double, 4>(PtrStepSzb src, void* buf, double* out, PtrStepSzb mask);
 }
 
-#endif // CUDA_DISABLER
+cv::Scalar cv::cuda::sum(InputArray _src, InputArray _mask, GpuMat& buf)
+{
+    typedef cv::Scalar (*func_t)(const GpuMat& _src, const GpuMat& mask, GpuMat& _buf);
+    static const func_t funcs[7][4] =
+    {
+        {sumImpl<uchar , uint  , 1>, sumImpl<uchar , uint  , 2>, sumImpl<uchar , uint  , 3>, sumImpl<uchar , uint  , 4>},
+        {sumImpl<schar , int   , 1>, sumImpl<schar , int   , 2>, sumImpl<schar , int   , 3>, sumImpl<schar , int   , 4>},
+        {sumImpl<ushort, uint  , 1>, sumImpl<ushort, uint  , 2>, sumImpl<ushort, uint  , 3>, sumImpl<ushort, uint  , 4>},
+        {sumImpl<short , int   , 1>, sumImpl<short , int   , 2>, sumImpl<short , int   , 3>, sumImpl<short , int   , 4>},
+        {sumImpl<int   , int   , 1>, sumImpl<int   , int   , 2>, sumImpl<int   , int   , 3>, sumImpl<int   , int   , 4>},
+        {sumImpl<float , float , 1>, sumImpl<float , float , 2>, sumImpl<float , float , 3>, sumImpl<float , float , 4>},
+        {sumImpl<double, double, 1>, sumImpl<double, double, 2>, sumImpl<double, double, 3>, sumImpl<double, double, 4>}
+    };
+
+    GpuMat src = _src.getGpuMat();
+    GpuMat mask = _mask.getGpuMat();
+
+    CV_DbgAssert( mask.empty() || (mask.type() == CV_8UC1 && mask.size() == src.size()) );
+
+    const func_t func = funcs[src.depth()][src.channels() - 1];
+
+    return func(src, mask, buf);
+}
+
+cv::Scalar cv::cuda::absSum(InputArray _src, InputArray _mask, GpuMat& buf)
+{
+    typedef cv::Scalar (*func_t)(const GpuMat& _src, const GpuMat& mask, GpuMat& _buf);
+    static const func_t funcs[7][4] =
+    {
+        {sumAbsImpl<uchar , uint  , 1>, sumAbsImpl<uchar , uint  , 2>, sumAbsImpl<uchar , uint  , 3>, sumAbsImpl<uchar , uint  , 4>},
+        {sumAbsImpl<schar , int   , 1>, sumAbsImpl<schar , int   , 2>, sumAbsImpl<schar , int   , 3>, sumAbsImpl<schar , int   , 4>},
+        {sumAbsImpl<ushort, uint  , 1>, sumAbsImpl<ushort, uint  , 2>, sumAbsImpl<ushort, uint  , 3>, sumAbsImpl<ushort, uint  , 4>},
+        {sumAbsImpl<short , int   , 1>, sumAbsImpl<short , int   , 2>, sumAbsImpl<short , int   , 3>, sumAbsImpl<short , int   , 4>},
+        {sumAbsImpl<int   , int   , 1>, sumAbsImpl<int   , int   , 2>, sumAbsImpl<int   , int   , 3>, sumAbsImpl<int   , int   , 4>},
+        {sumAbsImpl<float , float , 1>, sumAbsImpl<float , float , 2>, sumAbsImpl<float , float , 3>, sumAbsImpl<float , float , 4>},
+        {sumAbsImpl<double, double, 1>, sumAbsImpl<double, double, 2>, sumAbsImpl<double, double, 3>, sumAbsImpl<double, double, 4>}
+    };
+
+    GpuMat src = _src.getGpuMat();
+    GpuMat mask = _mask.getGpuMat();
+
+    CV_DbgAssert( mask.empty() || (mask.type() == CV_8UC1 && mask.size() == src.size()) );
+
+    const func_t func = funcs[src.depth()][src.channels() - 1];
+
+    return func(src, mask, buf);
+}
+
+cv::Scalar cv::cuda::sqrSum(InputArray _src, InputArray _mask, GpuMat& buf)
+{
+    typedef cv::Scalar (*func_t)(const GpuMat& _src, const GpuMat& mask, GpuMat& _buf);
+    static const func_t funcs[7][4] =
+    {
+        {sumSqrImpl<uchar , double, 1>, sumSqrImpl<uchar , double, 2>, sumSqrImpl<uchar , double, 3>, sumSqrImpl<uchar , double, 4>},
+        {sumSqrImpl<schar , double, 1>, sumSqrImpl<schar , double, 2>, sumSqrImpl<schar , double, 3>, sumSqrImpl<schar , double, 4>},
+        {sumSqrImpl<ushort, double, 1>, sumSqrImpl<ushort, double, 2>, sumSqrImpl<ushort, double, 3>, sumSqrImpl<ushort, double, 4>},
+        {sumSqrImpl<short , double, 1>, sumSqrImpl<short , double, 2>, sumSqrImpl<short , double, 3>, sumSqrImpl<short , double, 4>},
+        {sumSqrImpl<int   , double, 1>, sumSqrImpl<int   , double, 2>, sumSqrImpl<int   , double, 3>, sumSqrImpl<int   , double, 4>},
+        {sumSqrImpl<float , double, 1>, sumSqrImpl<float , double, 2>, sumSqrImpl<float , double, 3>, sumSqrImpl<float , double, 4>},
+        {sumSqrImpl<double, double, 1>, sumSqrImpl<double, double, 2>, sumSqrImpl<double, double, 3>, sumSqrImpl<double, double, 4>}
+    };
+
+    GpuMat src = _src.getGpuMat();
+    GpuMat mask = _mask.getGpuMat();
+
+    CV_DbgAssert( mask.empty() || (mask.type() == CV_8UC1 && mask.size() == src.size()) );
+
+    const func_t func = funcs[src.depth()][src.channels() - 1];
+
+    return func(src, mask, buf);
+}
+
+#endif
diff --git a/modules/cudaarithm/src/cuda/threshold.cu b/modules/cudaarithm/src/cuda/threshold.cu
index 7bb20bf7ee..21665cbe73 100644
--- a/modules/cudaarithm/src/cuda/threshold.cu
+++ b/modules/cudaarithm/src/cuda/threshold.cu
@@ -40,75 +40,109 @@
 //
 //M*/
 
-#if !defined CUDA_DISABLER
+#include "opencv2/opencv_modules.hpp"
 
-#include "opencv2/core/cuda/common.hpp"
-#include "opencv2/core/cuda/functional.hpp"
-#include "opencv2/core/cuda/transform.hpp"
-#include "opencv2/core/cuda/saturate_cast.hpp"
-#include "opencv2/core/cuda/simd_functions.hpp"
+#ifndef HAVE_OPENCV_CUDEV
 
-#include "arithm_func_traits.hpp"
+#error "opencv_cudev is required"
 
-using namespace cv::cuda;
-using namespace cv::cuda::device;
+#else
 
-namespace cv { namespace cuda { namespace device
+#include "opencv2/cudaarithm.hpp"
+#include "opencv2/cudev.hpp"
+#include "opencv2/core/private.cuda.hpp"
+
+using namespace cv::cudev;
+
+namespace
 {
-    template <typename T> struct TransformFunctorTraits< thresh_binary_func<T> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(T)>
+    template <typename ScalarDepth> struct TransformPolicy : DefaultTransformPolicy
     {
     };
-
-    template <typename T> struct TransformFunctorTraits< thresh_binary_inv_func<T> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(T)>
+    template <> struct TransformPolicy<double> : DefaultTransformPolicy
     {
+        enum {
+            shift = 1
+        };
     };
 
-    template <typename T> struct TransformFunctorTraits< thresh_trunc_func<T> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(T)>
-    {
-    };
-
-    template <typename T> struct TransformFunctorTraits< thresh_to_zero_func<T> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(T)>
-    {
-    };
-
-    template <typename T> struct TransformFunctorTraits< thresh_to_zero_inv_func<T> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(T)>
-    {
-    };
-}}}
-
-namespace arithm
-{
-    template <template <typename> class Op, typename T>
-    void threshold_caller(PtrStepSz<T> src, PtrStepSz<T> dst, T thresh, T maxVal, cudaStream_t stream)
-    {
-        Op<T> op(thresh, maxVal);
-        device::transform(src, dst, op, WithOutMask(), stream);
-    }
-
     template <typename T>
-    void threshold(PtrStepSzb src, PtrStepSzb dst, double thresh, double maxVal, int type, cudaStream_t stream)
+    void thresholdImpl(const GpuMat& src, GpuMat& dst, double thresh, double maxVal, int type, Stream& stream)
     {
-        typedef void (*caller_t)(PtrStepSz<T> src, PtrStepSz<T> dst, T thresh, T maxVal, cudaStream_t stream);
+        const T thresh_ = static_cast<T>(thresh);
+        const T maxVal_ = static_cast<T>(maxVal);
 
-        static const caller_t callers[] =
+        switch (type)
         {
-            threshold_caller<thresh_binary_func, T>,
-            threshold_caller<thresh_binary_inv_func, T>,
-            threshold_caller<thresh_trunc_func, T>,
-            threshold_caller<thresh_to_zero_func, T>,
-            threshold_caller<thresh_to_zero_inv_func, T>
+        case 0:
+            gridTransformUnary_< TransformPolicy<T> >(globPtr<T>(src), globPtr<T>(dst), thresh_binary_func(thresh_, maxVal_), stream);
+            break;
+        case 1:
+            gridTransformUnary_< TransformPolicy<T> >(globPtr<T>(src), globPtr<T>(dst), thresh_binary_inv_func(thresh_, maxVal_), stream);
+            break;
+        case 2:
+            gridTransformUnary_< TransformPolicy<T> >(globPtr<T>(src), globPtr<T>(dst), thresh_trunc_func(thresh_), stream);
+            break;
+        case 3:
+            gridTransformUnary_< TransformPolicy<T> >(globPtr<T>(src), globPtr<T>(dst), thresh_to_zero_func(thresh_), stream);
+            break;
+        case 4:
+            gridTransformUnary_< TransformPolicy<T> >(globPtr<T>(src), globPtr<T>(dst), thresh_to_zero_inv_func(thresh_), stream);
+            break;
         };
-
-        callers[type]((PtrStepSz<T>) src, (PtrStepSz<T>) dst, static_cast<T>(thresh), static_cast<T>(maxVal), stream);
     }
-
-    template void threshold<uchar>(PtrStepSzb src, PtrStepSzb dst, double thresh, double maxVal, int type, cudaStream_t stream);
-    template void threshold<schar>(PtrStepSzb src, PtrStepSzb dst, double thresh, double maxVal, int type, cudaStream_t stream);
-    template void threshold<ushort>(PtrStepSzb src, PtrStepSzb dst, double thresh, double maxVal, int type, cudaStream_t stream);
-    template void threshold<short>(PtrStepSzb src, PtrStepSzb dst, double thresh, double maxVal, int type, cudaStream_t stream);
-    template void threshold<int>(PtrStepSzb src, PtrStepSzb dst, double thresh, double maxVal, int type, cudaStream_t stream);
-    template void threshold<float>(PtrStepSzb src, PtrStepSzb dst, double thresh, double maxVal, int type, cudaStream_t stream);
-    template void threshold<double>(PtrStepSzb src, PtrStepSzb dst, double thresh, double maxVal, int type, cudaStream_t stream);
 }
 
-#endif // CUDA_DISABLER
+double cv::cuda::threshold(InputArray _src, OutputArray _dst, double thresh, double maxVal, int type, Stream& stream)
+{
+    GpuMat src = _src.getGpuMat();
+
+    const int depth = src.depth();
+
+    CV_DbgAssert( src.channels() == 1 && depth <= CV_64F );
+    CV_DbgAssert( type <= 4 /*THRESH_TOZERO_INV*/ );
+
+    _dst.create(src.size(), src.type());
+    GpuMat dst = _dst.getGpuMat();
+
+    if (depth == CV_32F && type == 2 /*THRESH_TRUNC*/)
+    {
+        NppStreamHandler h(StreamAccessor::getStream(stream));
+
+        NppiSize sz;
+        sz.width  = src.cols;
+        sz.height = src.rows;
+
+        nppSafeCall( nppiThreshold_32f_C1R(src.ptr<Npp32f>(), static_cast<int>(src.step),
+            dst.ptr<Npp32f>(), static_cast<int>(dst.step), sz, static_cast<Npp32f>(thresh), NPP_CMP_GREATER) );
+
+        if (!stream)
+            CV_CUDEV_SAFE_CALL( cudaDeviceSynchronize() );
+    }
+    else
+    {
+        typedef void (*func_t)(const GpuMat& src, GpuMat& dst, double thresh, double maxVal, int type, Stream& stream);
+        static const func_t funcs[] =
+        {
+            thresholdImpl<uchar>,
+            thresholdImpl<schar>,
+            thresholdImpl<ushort>,
+            thresholdImpl<short>,
+            thresholdImpl<int>,
+            thresholdImpl<float>,
+            thresholdImpl<double>
+        };
+
+        if (depth != CV_32F && depth != CV_64F)
+        {
+            thresh = cvFloor(thresh);
+            maxVal = cvRound(maxVal);
+        }
+
+        funcs[depth](src, dst, thresh, maxVal, type, stream);
+    }
+
+    return thresh;
+}
+
+#endif
diff --git a/modules/cudaarithm/src/cuda/transpose.cu b/modules/cudaarithm/src/cuda/transpose.cu
index 6111b4b1a9..aa85004425 100644
--- a/modules/cudaarithm/src/cuda/transpose.cu
+++ b/modules/cudaarithm/src/cuda/transpose.cu
@@ -40,83 +40,53 @@
 //
 //M*/
 
-#if !defined CUDA_DISABLER
+#include "opencv2/opencv_modules.hpp"
 
-#include "opencv2/core/cuda/common.hpp"
+#ifndef HAVE_OPENCV_CUDEV
 
-using namespace cv::cuda;
-using namespace cv::cuda::device;
+#error "opencv_cudev is required"
 
-namespace arithm
+#else
+
+#include "opencv2/cudaarithm.hpp"
+#include "opencv2/cudev.hpp"
+#include "opencv2/core/private.cuda.hpp"
+
+using namespace cv::cudev;
+
+void cv::cuda::transpose(InputArray _src, OutputArray _dst, Stream& stream)
 {
-    const int TRANSPOSE_TILE_DIM   = 16;
-    const int TRANSPOSE_BLOCK_ROWS = 16;
+    GpuMat src = _src.getGpuMat();
 
-    template <typename T>
-    __global__ void transposeKernel(const PtrStepSz<T> src, PtrStep<T> dst)
+    const size_t elemSize = src.elemSize();
+
+    CV_Assert( elemSize == 1 || elemSize == 4 || elemSize == 8 );
+
+    _dst.create( src.cols, src.rows, src.type() );
+    GpuMat dst = _dst.getGpuMat();
+
+    if (elemSize == 1)
     {
-        __shared__ T tile[TRANSPOSE_TILE_DIM][TRANSPOSE_TILE_DIM + 1];
+        NppStreamHandler h(StreamAccessor::getStream(stream));
 
-        int blockIdx_x, blockIdx_y;
+        NppiSize sz;
+        sz.width  = src.cols;
+        sz.height = src.rows;
 
-        // do diagonal reordering
-        if (gridDim.x == gridDim.y)
-        {
-            blockIdx_y = blockIdx.x;
-            blockIdx_x = (blockIdx.x + blockIdx.y) % gridDim.x;
-        }
-        else
-        {
-            int bid = blockIdx.x + gridDim.x * blockIdx.y;
-            blockIdx_y = bid % gridDim.y;
-            blockIdx_x = ((bid / gridDim.y) + blockIdx_y) % gridDim.x;
-        }
+        nppSafeCall( nppiTranspose_8u_C1R(src.ptr<Npp8u>(), static_cast<int>(src.step),
+            dst.ptr<Npp8u>(), static_cast<int>(dst.step), sz) );
 
-        int xIndex = blockIdx_x * TRANSPOSE_TILE_DIM + threadIdx.x;
-        int yIndex = blockIdx_y * TRANSPOSE_TILE_DIM + threadIdx.y;
-
-        if (xIndex < src.cols)
-        {
-            for (int i = 0; i < TRANSPOSE_TILE_DIM; i += TRANSPOSE_BLOCK_ROWS)
-            {
-                if (yIndex + i < src.rows)
-                {
-                    tile[threadIdx.y + i][threadIdx.x] = src(yIndex + i, xIndex);
-                }
-            }
-        }
-
-        __syncthreads();
-
-        xIndex = blockIdx_y * TRANSPOSE_TILE_DIM + threadIdx.x;
-        yIndex = blockIdx_x * TRANSPOSE_TILE_DIM + threadIdx.y;
-
-        if (xIndex < src.rows)
-        {
-            for (int i = 0; i < TRANSPOSE_TILE_DIM; i += TRANSPOSE_BLOCK_ROWS)
-            {
-                if (yIndex + i < src.cols)
-                {
-                    dst(yIndex + i, xIndex) = tile[threadIdx.x][threadIdx.y + i];
-                }
-            }
-        }
+        if (!stream)
+            CV_CUDEV_SAFE_CALL( cudaDeviceSynchronize() );
     }
-
-    template <typename T> void transpose(PtrStepSz<T> src, PtrStepSz<T> dst, cudaStream_t stream)
+    else if (elemSize == 4)
     {
-        const dim3 block(TRANSPOSE_TILE_DIM, TRANSPOSE_TILE_DIM);
-        const dim3 grid(divUp(src.cols, block.x), divUp(src.rows, block.y));
-
-        transposeKernel<<<grid, block, 0, stream>>>(src, dst);
-        cudaSafeCall( cudaGetLastError() );
-
-        if (stream == 0)
-            cudaSafeCall( cudaDeviceSynchronize() );
+        gridTranspose(globPtr<int>(src), globPtr<int>(dst), stream);
+    }
+    else // if (elemSize == 8)
+    {
+        gridTranspose(globPtr<double>(src), globPtr<double>(dst), stream);
     }
-
-    template void transpose<int>(PtrStepSz<int> src, PtrStepSz<int> dst, cudaStream_t stream);
-    template void transpose<double>(PtrStepSz<double> src, PtrStepSz<double> dst, cudaStream_t stream);
 }
 
-#endif // CUDA_DISABLER
+#endif
diff --git a/modules/cudaarithm/src/cuda/unroll_detail.hpp b/modules/cudaarithm/src/cuda/unroll_detail.hpp
deleted file mode 100644
index 544d838b27..0000000000
--- a/modules/cudaarithm/src/cuda/unroll_detail.hpp
+++ /dev/null
@@ -1,135 +0,0 @@
-/*M///////////////////////////////////////////////////////////////////////////////////////
-//
-//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
-//
-//  By downloading, copying, installing or using the software you agree to this license.
-//  If you do not agree to this license, do not download, install,
-//  copy or use the software.
-//
-//
-//                           License Agreement
-//                For Open Source Computer Vision Library
-//
-// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
-// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
-// Third party copyrights are property of their respective owners.
-//
-// Redistribution and use in source and binary forms, with or without modification,
-// are permitted provided that the following conditions are met:
-//
-//   * Redistribution's of source code must retain the above copyright notice,
-//     this list of conditions and the following disclaimer.
-//
-//   * Redistribution's in binary form must reproduce the above copyright notice,
-//     this list of conditions and the following disclaimer in the documentation
-//     and/or other materials provided with the distribution.
-//
-//   * The name of the copyright holders may not be used to endorse or promote products
-//     derived from this software without specific prior written permission.
-//
-// This software is provided by the copyright holders and contributors "as is" and
-// any express or implied warranties, including, but not limited to, the implied
-// warranties of merchantability and fitness for a particular purpose are disclaimed.
-// In no event shall the Intel Corporation or contributors be liable for any direct,
-// indirect, incidental, special, exemplary, or consequential damages
-// (including, but not limited to, procurement of substitute goods or services;
-// loss of use, data, or profits; or business interruption) however caused
-// and on any theory of liability, whether in contract, strict liability,
-// or tort (including negligence or otherwise) arising in any way out of
-// the use of this software, even if advised of the possibility of such damage.
-//
-//M*/
-
-#ifndef __UNROLL_DETAIL_HPP__
-#define __UNROLL_DETAIL_HPP__
-
-#include <thrust/tuple.h>
-#include "opencv2/core/cuda/common.hpp"
-#include "opencv2/core/cuda/vec_traits.hpp"
-
-namespace detail
-{
-    template <int cn> struct Unroll;
-    template <> struct Unroll<1>
-    {
-        template <int BLOCK_SIZE, typename R>
-        static __device__ __forceinline__ volatile R* smem_tuple(R* smem)
-        {
-            return smem;
-        }
-
-        template <typename R>
-        static __device__ __forceinline__ R& tie(R& val)
-        {
-            return val;
-        }
-
-        template <class Op>
-        static __device__ __forceinline__ const Op& op(const Op& op)
-        {
-            return op;
-        }
-    };
-    template <> struct Unroll<2>
-    {
-        template <int BLOCK_SIZE, typename R>
-        static __device__ __forceinline__ thrust::tuple<volatile R*, volatile R*> smem_tuple(R* smem)
-        {
-            return cv::cuda::device::smem_tuple(smem, smem + BLOCK_SIZE);
-        }
-
-        template <typename R>
-        static __device__ __forceinline__ thrust::tuple<typename cv::cuda::device::VecTraits<R>::elem_type&, typename cv::cuda::device::VecTraits<R>::elem_type&> tie(R& val)
-        {
-            return thrust::tie(val.x, val.y);
-        }
-
-        template <class Op>
-        static __device__ __forceinline__ const thrust::tuple<Op, Op> op(const Op& op)
-        {
-            return thrust::make_tuple(op, op);
-        }
-    };
-    template <> struct Unroll<3>
-    {
-        template <int BLOCK_SIZE, typename R>
-        static __device__ __forceinline__ thrust::tuple<volatile R*, volatile R*, volatile R*> smem_tuple(R* smem)
-        {
-            return cv::cuda::device::smem_tuple(smem, smem + BLOCK_SIZE, smem + 2 * BLOCK_SIZE);
-        }
-
-        template <typename R>
-        static __device__ __forceinline__ thrust::tuple<typename cv::cuda::device::VecTraits<R>::elem_type&, typename cv::cuda::device::VecTraits<R>::elem_type&, typename cv::cuda::device::VecTraits<R>::elem_type&> tie(R& val)
-        {
-            return thrust::tie(val.x, val.y, val.z);
-        }
-
-        template <class Op>
-        static __device__ __forceinline__ const thrust::tuple<Op, Op, Op> op(const Op& op)
-        {
-            return thrust::make_tuple(op, op, op);
-        }
-    };
-    template <> struct Unroll<4>
-    {
-        template <int BLOCK_SIZE, typename R>
-        static __device__ __forceinline__ thrust::tuple<volatile R*, volatile R*, volatile R*, volatile R*> smem_tuple(R* smem)
-        {
-            return cv::cuda::device::smem_tuple(smem, smem + BLOCK_SIZE, smem + 2 * BLOCK_SIZE, smem + 3 * BLOCK_SIZE);
-        }
-
-        template <typename R>
-        static __device__ __forceinline__ thrust::tuple<typename cv::cuda::device::VecTraits<R>::elem_type&, typename cv::cuda::device::VecTraits<R>::elem_type&, typename cv::cuda::device::VecTraits<R>::elem_type&, typename cv::cuda::device::VecTraits<R>::elem_type&> tie(R& val)
-        {
-            return thrust::tie(val.x, val.y, val.z, val.w);
-        }
-
-        template <class Op>
-        static __device__ __forceinline__ const thrust::tuple<Op, Op, Op, Op> op(const Op& op)
-        {
-            return thrust::make_tuple(op, op, op, op);
-        }
-    };
-}
-
-#endif // __UNROLL_DETAIL_HPP__
diff --git a/modules/cudaarithm/src/element_operations.cpp b/modules/cudaarithm/src/element_operations.cpp
index 715771b671..795d7ffaa7 100644
--- a/modules/cudaarithm/src/element_operations.cpp
+++ b/modules/cudaarithm/src/element_operations.cpp
@@ -159,425 +159,12 @@ namespace
     }
 }
 
-
-////////////////////////////////////////////////////////////////////////
-// Basic arithmetical operations (add subtract multiply divide)
-
-namespace
-{
-    template<int DEPTH> struct NppTypeTraits;
-    template<> struct NppTypeTraits<CV_8U>  { typedef Npp8u npp_t; };
-    template<> struct NppTypeTraits<CV_8S>  { typedef Npp8s npp_t; };
-    template<> struct NppTypeTraits<CV_16U> { typedef Npp16u npp_t; };
-    template<> struct NppTypeTraits<CV_16S> { typedef Npp16s npp_t; typedef Npp16sc npp_complex_type; };
-    template<> struct NppTypeTraits<CV_32S> { typedef Npp32s npp_t; typedef Npp32sc npp_complex_type; };
-    template<> struct NppTypeTraits<CV_32F> { typedef Npp32f npp_t; typedef Npp32fc npp_complex_type; };
-    template<> struct NppTypeTraits<CV_64F> { typedef Npp64f npp_t; typedef Npp64fc npp_complex_type; };
-
-    template<int DEPTH, int cn> struct NppArithmScalarFunc
-    {
-        typedef typename NppTypeTraits<DEPTH>::npp_t npp_t;
-
-        typedef NppStatus (*func_ptr)(const npp_t* pSrc1, int nSrc1Step, const npp_t* pConstants,
-            npp_t* pDst, int nDstStep, NppiSize oSizeROI, int nScaleFactor);
-    };
-    template<int DEPTH> struct NppArithmScalarFunc<DEPTH, 1>
-    {
-        typedef typename NppTypeTraits<DEPTH>::npp_t npp_t;
-
-        typedef NppStatus (*func_ptr)(const npp_t* pSrc1, int nSrc1Step, const npp_t pConstants,
-            npp_t* pDst, int nDstStep, NppiSize oSizeROI, int nScaleFactor);
-    };
-    template<int DEPTH> struct NppArithmScalarFunc<DEPTH, 2>
-    {
-        typedef typename NppTypeTraits<DEPTH>::npp_complex_type npp_complex_type;
-
-        typedef NppStatus (*func_ptr)(const npp_complex_type* pSrc1, int nSrc1Step, const npp_complex_type pConstants,
-            npp_complex_type* pDst, int nDstStep, NppiSize oSizeROI, int nScaleFactor);
-    };
-    template<int cn> struct NppArithmScalarFunc<CV_32F, cn>
-    {
-        typedef NppStatus (*func_ptr)(const Npp32f* pSrc1, int nSrc1Step, const Npp32f* pConstants, Npp32f* pDst, int nDstStep, NppiSize oSizeROI);
-    };
-    template<> struct NppArithmScalarFunc<CV_32F, 1>
-    {
-        typedef NppStatus (*func_ptr)(const Npp32f* pSrc1, int nSrc1Step, const Npp32f pConstants, Npp32f* pDst, int nDstStep, NppiSize oSizeROI);
-    };
-    template<> struct NppArithmScalarFunc<CV_32F, 2>
-    {
-        typedef NppStatus (*func_ptr)(const Npp32fc* pSrc1, int nSrc1Step, const Npp32fc pConstants, Npp32fc* pDst, int nDstStep, NppiSize oSizeROI);
-    };
-
-    template<int DEPTH, int cn, typename NppArithmScalarFunc<DEPTH, cn>::func_ptr func> struct NppArithmScalar
-    {
-        typedef typename NppTypeTraits<DEPTH>::npp_t npp_t;
-
-        static void call(const PtrStepSzb src, Scalar sc, PtrStepb dst, cudaStream_t stream)
-        {
-            NppStreamHandler h(stream);
-
-            NppiSize sz;
-            sz.width = src.cols;
-            sz.height = src.rows;
-
-            const npp_t pConstants[] = { saturate_cast<npp_t>(sc.val[0]), saturate_cast<npp_t>(sc.val[1]), saturate_cast<npp_t>(sc.val[2]), saturate_cast<npp_t>(sc.val[3]) };
-
-            nppSafeCall( func((const npp_t*)src.data, static_cast<int>(src.step), pConstants, (npp_t*)dst.data, static_cast<int>(dst.step), sz, 0) );
-
-            if (stream == 0)
-                cudaSafeCall( cudaDeviceSynchronize() );
-        }
-    };
-    template<int DEPTH, typename NppArithmScalarFunc<DEPTH, 1>::func_ptr func> struct NppArithmScalar<DEPTH, 1, func>
-    {
-        typedef typename NppTypeTraits<DEPTH>::npp_t npp_t;
-
-        static void call(const PtrStepSzb src, Scalar sc, PtrStepb dst, cudaStream_t stream)
-        {
-            NppStreamHandler h(stream);
-
-            NppiSize sz;
-            sz.width = src.cols;
-            sz.height = src.rows;
-
-            nppSafeCall( func((const npp_t*)src.data, static_cast<int>(src.step), saturate_cast<npp_t>(sc.val[0]), (npp_t*)dst.data, static_cast<int>(dst.step), sz, 0) );
-
-            if (stream == 0)
-                cudaSafeCall( cudaDeviceSynchronize() );
-        }
-    };
-    template<int DEPTH, typename NppArithmScalarFunc<DEPTH, 2>::func_ptr func> struct NppArithmScalar<DEPTH, 2, func>
-    {
-        typedef typename NppTypeTraits<DEPTH>::npp_t npp_t;
-        typedef typename NppTypeTraits<DEPTH>::npp_complex_type npp_complex_type;
-
-        static void call(const PtrStepSzb src, Scalar sc, PtrStepb dst, cudaStream_t stream)
-        {
-            NppStreamHandler h(stream);
-
-            NppiSize sz;
-            sz.width = src.cols;
-            sz.height = src.rows;
-
-            npp_complex_type nConstant;
-            nConstant.re = saturate_cast<npp_t>(sc.val[0]);
-            nConstant.im = saturate_cast<npp_t>(sc.val[1]);
-
-            nppSafeCall( func((const npp_complex_type*)src.data, static_cast<int>(src.step), nConstant,
-                              (npp_complex_type*)dst.data, static_cast<int>(dst.step), sz, 0) );
-
-            if (stream == 0)
-                cudaSafeCall( cudaDeviceSynchronize() );
-        }
-    };
-    template<int cn, typename NppArithmScalarFunc<CV_32F, cn>::func_ptr func> struct NppArithmScalar<CV_32F, cn, func>
-    {
-        typedef typename NppTypeTraits<CV_32F>::npp_t npp_t;
-
-        static void call(const PtrStepSzb src, Scalar sc, PtrStepb dst, cudaStream_t stream)
-        {
-            NppStreamHandler h(stream);
-
-            NppiSize sz;
-            sz.width = src.cols;
-            sz.height = src.rows;
-
-            const Npp32f pConstants[] = { saturate_cast<Npp32f>(sc.val[0]), saturate_cast<Npp32f>(sc.val[1]), saturate_cast<Npp32f>(sc.val[2]), saturate_cast<Npp32f>(sc.val[3]) };
-
-            nppSafeCall( func((const npp_t*)src.data, static_cast<int>(src.step), pConstants, (npp_t*)dst.data, static_cast<int>(dst.step), sz) );
-
-            if (stream == 0)
-                cudaSafeCall( cudaDeviceSynchronize() );
-        }
-    };
-    template<typename NppArithmScalarFunc<CV_32F, 1>::func_ptr func> struct NppArithmScalar<CV_32F, 1, func>
-    {
-        typedef typename NppTypeTraits<CV_32F>::npp_t npp_t;
-
-        static void call(const PtrStepSzb src, Scalar sc, PtrStepb dst, cudaStream_t stream)
-        {
-            NppStreamHandler h(stream);
-
-            NppiSize sz;
-            sz.width = src.cols;
-            sz.height = src.rows;
-
-            nppSafeCall( func((const npp_t*)src.data, static_cast<int>(src.step), saturate_cast<Npp32f>(sc.val[0]), (npp_t*)dst.data, static_cast<int>(dst.step), sz) );
-
-            if (stream == 0)
-                cudaSafeCall( cudaDeviceSynchronize() );
-        }
-    };
-    template<typename NppArithmScalarFunc<CV_32F, 2>::func_ptr func> struct NppArithmScalar<CV_32F, 2, func>
-    {
-        typedef typename NppTypeTraits<CV_32F>::npp_t npp_t;
-        typedef typename NppTypeTraits<CV_32F>::npp_complex_type npp_complex_type;
-
-        static void call(const PtrStepSzb src, Scalar sc, PtrStepb dst, cudaStream_t stream)
-        {
-            NppStreamHandler h(stream);
-
-            NppiSize sz;
-            sz.width = src.cols;
-            sz.height = src.rows;
-
-            Npp32fc nConstant;
-            nConstant.re = saturate_cast<Npp32f>(sc.val[0]);
-            nConstant.im = saturate_cast<Npp32f>(sc.val[1]);
-
-            nppSafeCall( func((const npp_complex_type*)src.data, static_cast<int>(src.step), nConstant, (npp_complex_type*)dst.data, static_cast<int>(dst.step), sz) );
-
-            if (stream == 0)
-                cudaSafeCall( cudaDeviceSynchronize() );
-        }
-    };
-}
-
 ////////////////////////////////////////////////////////////////////////
 // add
 
-namespace arithm
-{
-    void addMat_v4(PtrStepSz<unsigned int> src1, PtrStepSz<unsigned int> src2, PtrStepSz<unsigned int> dst, cudaStream_t stream);
-    void addMat_v2(PtrStepSz<unsigned int> src1, PtrStepSz<unsigned int> src2, PtrStepSz<unsigned int> dst, cudaStream_t stream);
+void addMat(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat& mask, double, Stream& _stream, int);
 
-    template <typename T, typename D>
-    void addMat(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-}
-
-static void addMat(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat& mask, double, Stream& _stream, int)
-{
-    typedef void (*func_t)(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    static const func_t funcs[7][7] =
-    {
-        {
-            arithm::addMat<unsigned char, unsigned char>,
-            arithm::addMat<unsigned char, signed char>,
-            arithm::addMat<unsigned char, unsigned short>,
-            arithm::addMat<unsigned char, short>,
-            arithm::addMat<unsigned char, int>,
-            arithm::addMat<unsigned char, float>,
-            arithm::addMat<unsigned char, double>
-        },
-        {
-            arithm::addMat<signed char, unsigned char>,
-            arithm::addMat<signed char, signed char>,
-            arithm::addMat<signed char, unsigned short>,
-            arithm::addMat<signed char, short>,
-            arithm::addMat<signed char, int>,
-            arithm::addMat<signed char, float>,
-            arithm::addMat<signed char, double>
-        },
-        {
-            0 /*arithm::addMat<unsigned short, unsigned char>*/,
-            0 /*arithm::addMat<unsigned short, signed char>*/,
-            arithm::addMat<unsigned short, unsigned short>,
-            arithm::addMat<unsigned short, short>,
-            arithm::addMat<unsigned short, int>,
-            arithm::addMat<unsigned short, float>,
-            arithm::addMat<unsigned short, double>
-        },
-        {
-            0 /*arithm::addMat<short, unsigned char>*/,
-            0 /*arithm::addMat<short, signed char>*/,
-            arithm::addMat<short, unsigned short>,
-            arithm::addMat<short, short>,
-            arithm::addMat<short, int>,
-            arithm::addMat<short, float>,
-            arithm::addMat<short, double>
-        },
-        {
-            0 /*arithm::addMat<int, unsigned char>*/,
-            0 /*arithm::addMat<int, signed char>*/,
-            0 /*arithm::addMat<int, unsigned short>*/,
-            0 /*arithm::addMat<int, short>*/,
-            arithm::addMat<int, int>,
-            arithm::addMat<int, float>,
-            arithm::addMat<int, double>
-        },
-        {
-            0 /*arithm::addMat<float, unsigned char>*/,
-            0 /*arithm::addMat<float, signed char>*/,
-            0 /*arithm::addMat<float, unsigned short>*/,
-            0 /*arithm::addMat<float, short>*/,
-            0 /*arithm::addMat<float, int>*/,
-            arithm::addMat<float, float>,
-            arithm::addMat<float, double>
-        },
-        {
-            0 /*arithm::addMat<double, unsigned char>*/,
-            0 /*arithm::addMat<double, signed char>*/,
-            0 /*arithm::addMat<double, unsigned short>*/,
-            0 /*arithm::addMat<double, short>*/,
-            0 /*arithm::addMat<double, int>*/,
-            0 /*arithm::addMat<double, float>*/,
-            arithm::addMat<double, double>
-        }
-    };
-
-    const int sdepth = src1.depth();
-    const int ddepth = dst.depth();
-    const int cn = src1.channels();
-
-    cudaStream_t stream = StreamAccessor::getStream(_stream);
-
-    PtrStepSzb src1_(src1.rows, src1.cols * cn, src1.data, src1.step);
-    PtrStepSzb src2_(src1.rows, src1.cols * cn, src2.data, src2.step);
-    PtrStepSzb dst_(src1.rows, src1.cols * cn, dst.data, dst.step);
-
-    if (mask.empty() && (sdepth == CV_8U || sdepth == CV_16U) && ddepth == sdepth)
-    {
-        const intptr_t src1ptr = reinterpret_cast<intptr_t>(src1_.data);
-        const intptr_t src2ptr = reinterpret_cast<intptr_t>(src2_.data);
-        const intptr_t dstptr = reinterpret_cast<intptr_t>(dst_.data);
-
-        const bool isAllAligned = (src1ptr & 31) == 0 && (src2ptr & 31) == 0 && (dstptr & 31) == 0;
-
-        if (isAllAligned)
-        {
-            if (sdepth == CV_8U && (src1_.cols & 3) == 0)
-            {
-                const int vcols = src1_.cols >> 2;
-
-                arithm::addMat_v4(PtrStepSz<unsigned int>(src1_.rows, vcols, (unsigned int*) src1_.data, src1_.step),
-                                  PtrStepSz<unsigned int>(src1_.rows, vcols, (unsigned int*) src2_.data, src2_.step),
-                                  PtrStepSz<unsigned int>(src1_.rows, vcols, (unsigned int*) dst_.data, dst_.step),
-                                  stream);
-
-                return;
-            }
-            else if (sdepth == CV_16U && (src1_.cols & 1) == 0)
-            {
-                const int vcols = src1_.cols >> 1;
-
-                arithm::addMat_v2(PtrStepSz<unsigned int>(src1_.rows, vcols, (unsigned int*) src1_.data, src1_.step),
-                                  PtrStepSz<unsigned int>(src1_.rows, vcols, (unsigned int*) src2_.data, src2_.step),
-                                  PtrStepSz<unsigned int>(src1_.rows, vcols, (unsigned int*) dst_.data, dst_.step),
-                                  stream);
-
-                return;
-            }
-        }
-    }
-
-    const func_t func = funcs[sdepth][ddepth];
-
-    if (!func)
-        CV_Error(cv::Error::StsUnsupportedFormat, "Unsupported combination of source and destination types");
-
-    func(src1_, src2_, dst_, mask, stream);
-}
-
-namespace arithm
-{
-    template <typename T, typename S, typename D>
-    void addScalar(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-}
-
-static void addScalar(const GpuMat& src, Scalar val, bool, GpuMat& dst, const GpuMat& mask, double, Stream& _stream, int)
-{
-    typedef void (*func_t)(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    static const func_t funcs[7][7] =
-    {
-        {
-            arithm::addScalar<unsigned char, float, unsigned char>,
-            arithm::addScalar<unsigned char, float, signed char>,
-            arithm::addScalar<unsigned char, float, unsigned short>,
-            arithm::addScalar<unsigned char, float, short>,
-            arithm::addScalar<unsigned char, float, int>,
-            arithm::addScalar<unsigned char, float, float>,
-            arithm::addScalar<unsigned char, double, double>
-        },
-        {
-            arithm::addScalar<signed char, float, unsigned char>,
-            arithm::addScalar<signed char, float, signed char>,
-            arithm::addScalar<signed char, float, unsigned short>,
-            arithm::addScalar<signed char, float, short>,
-            arithm::addScalar<signed char, float, int>,
-            arithm::addScalar<signed char, float, float>,
-            arithm::addScalar<signed char, double, double>
-        },
-        {
-            0 /*arithm::addScalar<unsigned short, float, unsigned char>*/,
-            0 /*arithm::addScalar<unsigned short, float, signed char>*/,
-            arithm::addScalar<unsigned short, float, unsigned short>,
-            arithm::addScalar<unsigned short, float, short>,
-            arithm::addScalar<unsigned short, float, int>,
-            arithm::addScalar<unsigned short, float, float>,
-            arithm::addScalar<unsigned short, double, double>
-        },
-        {
-            0 /*arithm::addScalar<short, float, unsigned char>*/,
-            0 /*arithm::addScalar<short, float, signed char>*/,
-            arithm::addScalar<short, float, unsigned short>,
-            arithm::addScalar<short, float, short>,
-            arithm::addScalar<short, float, int>,
-            arithm::addScalar<short, float, float>,
-            arithm::addScalar<short, double, double>
-        },
-        {
-            0 /*arithm::addScalar<int, float, unsigned char>*/,
-            0 /*arithm::addScalar<int, float, signed char>*/,
-            0 /*arithm::addScalar<int, float, unsigned short>*/,
-            0 /*arithm::addScalar<int, float, short>*/,
-            arithm::addScalar<int, float, int>,
-            arithm::addScalar<int, float, float>,
-            arithm::addScalar<int, double, double>
-        },
-        {
-            0 /*arithm::addScalar<float, float, unsigned char>*/,
-            0 /*arithm::addScalar<float, float, signed char>*/,
-            0 /*arithm::addScalar<float, float, unsigned short>*/,
-            0 /*arithm::addScalar<float, float, short>*/,
-            0 /*arithm::addScalar<float, float, int>*/,
-            arithm::addScalar<float, float, float>,
-            arithm::addScalar<float, double, double>
-        },
-        {
-            0 /*arithm::addScalar<double, double, unsigned char>*/,
-            0 /*arithm::addScalar<double, double, signed char>*/,
-            0 /*arithm::addScalar<double, double, unsigned short>*/,
-            0 /*arithm::addScalar<double, double, short>*/,
-            0 /*arithm::addScalar<double, double, int>*/,
-            0 /*arithm::addScalar<double, double, float>*/,
-            arithm::addScalar<double, double, double>
-        }
-    };
-
-    typedef void (*npp_func_t)(const PtrStepSzb src, Scalar sc, PtrStepb dst, cudaStream_t stream);
-    static const npp_func_t npp_funcs[7][4] =
-    {
-        {NppArithmScalar<CV_8U , 1, nppiAddC_8u_C1RSfs >::call, 0                                                     , NppArithmScalar<CV_8U , 3, nppiAddC_8u_C3RSfs >::call, NppArithmScalar<CV_8U , 4, nppiAddC_8u_C4RSfs >::call},
-        {0                                                    , 0                                                     , 0                                                    , 0                                                    },
-        {NppArithmScalar<CV_16U, 1, nppiAddC_16u_C1RSfs>::call, 0                                                     , NppArithmScalar<CV_16U, 3, nppiAddC_16u_C3RSfs>::call, NppArithmScalar<CV_16U, 4, nppiAddC_16u_C4RSfs>::call},
-        {NppArithmScalar<CV_16S, 1, nppiAddC_16s_C1RSfs>::call, NppArithmScalar<CV_16S, 2, nppiAddC_16sc_C1RSfs>::call, NppArithmScalar<CV_16S, 3, nppiAddC_16s_C3RSfs>::call, NppArithmScalar<CV_16S, 4, nppiAddC_16s_C4RSfs>::call},
-        {NppArithmScalar<CV_32S, 1, nppiAddC_32s_C1RSfs>::call, NppArithmScalar<CV_32S, 2, nppiAddC_32sc_C1RSfs>::call, NppArithmScalar<CV_32S, 3, nppiAddC_32s_C3RSfs>::call, 0                                                    },
-        {NppArithmScalar<CV_32F, 1, nppiAddC_32f_C1R   >::call, NppArithmScalar<CV_32F, 2, nppiAddC_32fc_C1R   >::call, NppArithmScalar<CV_32F, 3, nppiAddC_32f_C3R   >::call, NppArithmScalar<CV_32F, 4, nppiAddC_32f_C4R   >::call},
-        {0                                                    , 0                                                     , 0                                                    , 0                                                    }
-    };
-
-    const int sdepth = src.depth();
-    const int ddepth = dst.depth();
-    const int cn = src.channels();
-
-    cudaStream_t stream = StreamAccessor::getStream(_stream);
-
-    const npp_func_t npp_func = npp_funcs[sdepth][cn - 1];
-    if (ddepth == sdepth && cn > 1 && npp_func != 0)
-    {
-        npp_func(src, val, dst, stream);
-        return;
-    }
-
-    CV_Assert( cn == 1 );
-
-    const func_t func = funcs[sdepth][ddepth];
-
-    if (!func)
-        CV_Error(cv::Error::StsUnsupportedFormat, "Unsupported combination of source and destination types");
-
-    func(src, val[0], dst, mask, stream);
-}
+void addScalar(const GpuMat& src, Scalar val, bool, GpuMat& dst, const GpuMat& mask, double, Stream& stream, int);
 
 void cv::cuda::add(InputArray src1, InputArray src2, OutputArray dst, InputArray mask, int dtype, Stream& stream)
 {
@@ -587,248 +174,9 @@ void cv::cuda::add(InputArray src1, InputArray src2, OutputArray dst, InputArray
 ////////////////////////////////////////////////////////////////////////
 // subtract
 
-namespace arithm
-{
-    void subMat_v4(PtrStepSz<unsigned int> src1, PtrStepSz<unsigned int> src2, PtrStepSz<unsigned int> dst, cudaStream_t stream);
-    void subMat_v2(PtrStepSz<unsigned int> src1, PtrStepSz<unsigned int> src2, PtrStepSz<unsigned int> dst, cudaStream_t stream);
+void subMat(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat& mask, double, Stream& _stream, int);
 
-    template <typename T, typename D>
-    void subMat(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-}
-
-static void subMat(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat& mask, double, Stream& _stream, int)
-{
-    typedef void (*func_t)(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    static const func_t funcs[7][7] =
-    {
-        {
-            arithm::subMat<unsigned char, unsigned char>,
-            arithm::subMat<unsigned char, signed char>,
-            arithm::subMat<unsigned char, unsigned short>,
-            arithm::subMat<unsigned char, short>,
-            arithm::subMat<unsigned char, int>,
-            arithm::subMat<unsigned char, float>,
-            arithm::subMat<unsigned char, double>
-        },
-        {
-            arithm::subMat<signed char, unsigned char>,
-            arithm::subMat<signed char, signed char>,
-            arithm::subMat<signed char, unsigned short>,
-            arithm::subMat<signed char, short>,
-            arithm::subMat<signed char, int>,
-            arithm::subMat<signed char, float>,
-            arithm::subMat<signed char, double>
-        },
-        {
-            0 /*arithm::subMat<unsigned short, unsigned char>*/,
-            0 /*arithm::subMat<unsigned short, signed char>*/,
-            arithm::subMat<unsigned short, unsigned short>,
-            arithm::subMat<unsigned short, short>,
-            arithm::subMat<unsigned short, int>,
-            arithm::subMat<unsigned short, float>,
-            arithm::subMat<unsigned short, double>
-        },
-        {
-            0 /*arithm::subMat<short, unsigned char>*/,
-            0 /*arithm::subMat<short, signed char>*/,
-            arithm::subMat<short, unsigned short>,
-            arithm::subMat<short, short>,
-            arithm::subMat<short, int>,
-            arithm::subMat<short, float>,
-            arithm::subMat<short, double>
-        },
-        {
-            0 /*arithm::subMat<int, unsigned char>*/,
-            0 /*arithm::subMat<int, signed char>*/,
-            0 /*arithm::subMat<int, unsigned short>*/,
-            0 /*arithm::subMat<int, short>*/,
-            arithm::subMat<int, int>,
-            arithm::subMat<int, float>,
-            arithm::subMat<int, double>
-        },
-        {
-            0 /*arithm::subMat<float, unsigned char>*/,
-            0 /*arithm::subMat<float, signed char>*/,
-            0 /*arithm::subMat<float, unsigned short>*/,
-            0 /*arithm::subMat<float, short>*/,
-            0 /*arithm::subMat<float, int>*/,
-            arithm::subMat<float, float>,
-            arithm::subMat<float, double>
-        },
-        {
-            0 /*arithm::subMat<double, unsigned char>*/,
-            0 /*arithm::subMat<double, signed char>*/,
-            0 /*arithm::subMat<double, unsigned short>*/,
-            0 /*arithm::subMat<double, short>*/,
-            0 /*arithm::subMat<double, int>*/,
-            0 /*arithm::subMat<double, float>*/,
-            arithm::subMat<double, double>
-        }
-    };
-
-    const int sdepth = src1.depth();
-    const int ddepth = dst.depth();
-    const int cn = src1.channels();
-
-    cudaStream_t stream = StreamAccessor::getStream(_stream);
-
-    PtrStepSzb src1_(src1.rows, src1.cols * cn, src1.data, src1.step);
-    PtrStepSzb src2_(src1.rows, src1.cols * cn, src2.data, src2.step);
-    PtrStepSzb dst_(src1.rows, src1.cols * cn, dst.data, dst.step);
-
-    if (mask.empty() && (sdepth == CV_8U || sdepth == CV_16U) && ddepth == sdepth)
-    {
-        const intptr_t src1ptr = reinterpret_cast<intptr_t>(src1_.data);
-        const intptr_t src2ptr = reinterpret_cast<intptr_t>(src2_.data);
-        const intptr_t dstptr = reinterpret_cast<intptr_t>(dst_.data);
-
-        const bool isAllAligned = (src1ptr & 31) == 0 && (src2ptr & 31) == 0 && (dstptr & 31) == 0;
-
-        if (isAllAligned)
-        {
-            if (sdepth == CV_8U && (src1_.cols & 3) == 0)
-            {
-                const int vcols = src1_.cols >> 2;
-
-                arithm::subMat_v4(PtrStepSz<unsigned int>(src1_.rows, vcols, (unsigned int*) src1_.data, src1_.step),
-                                  PtrStepSz<unsigned int>(src1_.rows, vcols, (unsigned int*) src2_.data, src2_.step),
-                                  PtrStepSz<unsigned int>(src1_.rows, vcols, (unsigned int*) dst_.data, dst_.step),
-                                  stream);
-
-                return;
-            }
-            else if (sdepth == CV_16U && (src1_.cols & 1) == 0)
-            {
-                const int vcols = src1_.cols >> 1;
-
-                arithm::subMat_v2(PtrStepSz<unsigned int>(src1_.rows, vcols, (unsigned int*) src1_.data, src1_.step),
-                                  PtrStepSz<unsigned int>(src1_.rows, vcols, (unsigned int*) src2_.data, src2_.step),
-                                  PtrStepSz<unsigned int>(src1_.rows, vcols, (unsigned int*) dst_.data, dst_.step),
-                                  stream);
-
-                return;
-            }
-        }
-    }
-
-    const func_t func = funcs[sdepth][ddepth];
-
-    if (!func)
-        CV_Error(cv::Error::StsUnsupportedFormat, "Unsupported combination of source and destination types");
-
-    func(src1_, src2_, dst_, mask, stream);
-}
-
-namespace arithm
-{
-    template <typename T, typename S, typename D>
-    void subScalar(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-}
-
-static void subScalar(const GpuMat& src, Scalar val, bool inv, GpuMat& dst, const GpuMat& mask, double, Stream& _stream, int)
-{
-    typedef void (*func_t)(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    static const func_t funcs[7][7] =
-    {
-        {
-            arithm::subScalar<unsigned char, float, unsigned char>,
-            arithm::subScalar<unsigned char, float, signed char>,
-            arithm::subScalar<unsigned char, float, unsigned short>,
-            arithm::subScalar<unsigned char, float, short>,
-            arithm::subScalar<unsigned char, float, int>,
-            arithm::subScalar<unsigned char, float, float>,
-            arithm::subScalar<unsigned char, double, double>
-        },
-        {
-            arithm::subScalar<signed char, float, unsigned char>,
-            arithm::subScalar<signed char, float, signed char>,
-            arithm::subScalar<signed char, float, unsigned short>,
-            arithm::subScalar<signed char, float, short>,
-            arithm::subScalar<signed char, float, int>,
-            arithm::subScalar<signed char, float, float>,
-            arithm::subScalar<signed char, double, double>
-        },
-        {
-            0 /*arithm::subScalar<unsigned short, float, unsigned char>*/,
-            0 /*arithm::subScalar<unsigned short, float, signed char>*/,
-            arithm::subScalar<unsigned short, float, unsigned short>,
-            arithm::subScalar<unsigned short, float, short>,
-            arithm::subScalar<unsigned short, float, int>,
-            arithm::subScalar<unsigned short, float, float>,
-            arithm::subScalar<unsigned short, double, double>
-        },
-        {
-            0 /*arithm::subScalar<short, float, unsigned char>*/,
-            0 /*arithm::subScalar<short, float, signed char>*/,
-            arithm::subScalar<short, float, unsigned short>,
-            arithm::subScalar<short, float, short>,
-            arithm::subScalar<short, float, int>,
-            arithm::subScalar<short, float, float>,
-            arithm::subScalar<short, double, double>
-        },
-        {
-            0 /*arithm::subScalar<int, float, unsigned char>*/,
-            0 /*arithm::subScalar<int, float, signed char>*/,
-            0 /*arithm::subScalar<int, float, unsigned short>*/,
-            0 /*arithm::subScalar<int, float, short>*/,
-            arithm::subScalar<int, float, int>,
-            arithm::subScalar<int, float, float>,
-            arithm::subScalar<int, double, double>
-        },
-        {
-            0 /*arithm::subScalar<float, float, unsigned char>*/,
-            0 /*arithm::subScalar<float, float, signed char>*/,
-            0 /*arithm::subScalar<float, float, unsigned short>*/,
-            0 /*arithm::subScalar<float, float, short>*/,
-            0 /*arithm::subScalar<float, float, int>*/,
-            arithm::subScalar<float, float, float>,
-            arithm::subScalar<float, double, double>
-        },
-        {
-            0 /*arithm::subScalar<double, double, unsigned char>*/,
-            0 /*arithm::subScalar<double, double, signed char>*/,
-            0 /*arithm::subScalar<double, double, unsigned short>*/,
-            0 /*arithm::subScalar<double, double, short>*/,
-            0 /*arithm::subScalar<double, double, int>*/,
-            0 /*arithm::subScalar<double, double, float>*/,
-            arithm::subScalar<double, double, double>
-        }
-    };
-
-    typedef void (*npp_func_t)(const PtrStepSzb src, Scalar sc, PtrStepb dst, cudaStream_t stream);
-    static const npp_func_t npp_funcs[7][4] =
-    {
-        {NppArithmScalar<CV_8U , 1, nppiSubC_8u_C1RSfs >::call, 0                                                     , NppArithmScalar<CV_8U , 3, nppiSubC_8u_C3RSfs >::call, NppArithmScalar<CV_8U , 4, nppiSubC_8u_C4RSfs >::call},
-        {0                                                    , 0                                                     , 0                                                    , 0                                                    },
-        {NppArithmScalar<CV_16U, 1, nppiSubC_16u_C1RSfs>::call, 0                                                     , NppArithmScalar<CV_16U, 3, nppiSubC_16u_C3RSfs>::call, NppArithmScalar<CV_16U, 4, nppiSubC_16u_C4RSfs>::call},
-        {NppArithmScalar<CV_16S, 1, nppiSubC_16s_C1RSfs>::call, NppArithmScalar<CV_16S, 2, nppiSubC_16sc_C1RSfs>::call, NppArithmScalar<CV_16S, 3, nppiSubC_16s_C3RSfs>::call, NppArithmScalar<CV_16S, 4, nppiSubC_16s_C4RSfs>::call},
-        {NppArithmScalar<CV_32S, 1, nppiSubC_32s_C1RSfs>::call, NppArithmScalar<CV_32S, 2, nppiSubC_32sc_C1RSfs>::call, NppArithmScalar<CV_32S, 3, nppiSubC_32s_C3RSfs>::call, 0                                                    },
-        {NppArithmScalar<CV_32F, 1, nppiSubC_32f_C1R   >::call, NppArithmScalar<CV_32F, 2, nppiSubC_32fc_C1R   >::call, NppArithmScalar<CV_32F, 3, nppiSubC_32f_C3R   >::call, NppArithmScalar<CV_32F, 4, nppiSubC_32f_C4R   >::call},
-        {0                                                    , 0                                                     , 0                                                    , 0                                                    }
-    };
-
-    const int sdepth = src.depth();
-    const int ddepth = dst.depth();
-    const int cn = src.channels();
-
-    cudaStream_t stream = StreamAccessor::getStream(_stream);
-
-    const npp_func_t npp_func = npp_funcs[sdepth][cn - 1];
-    if (ddepth == sdepth && cn > 1 && npp_func != 0 && !inv)
-    {
-        npp_func(src, val, dst, stream);
-        return;
-    }
-
-    CV_Assert( cn == 1 );
-
-    const func_t func = funcs[sdepth][ddepth];
-
-    if (!func)
-        CV_Error(cv::Error::StsUnsupportedFormat, "Unsupported combination of source and destination types");
-
-    func(src, val[0], inv, dst, mask, stream);
-}
+void subScalar(const GpuMat& src, Scalar val, bool inv, GpuMat& dst, const GpuMat& mask, double, Stream& stream, int);
 
 void cv::cuda::subtract(InputArray src1, InputArray src2, OutputArray dst, InputArray mask, int dtype, Stream& stream)
 {
@@ -838,219 +186,11 @@ void cv::cuda::subtract(InputArray src1, InputArray src2, OutputArray dst, Input
 ////////////////////////////////////////////////////////////////////////
 // multiply
 
-namespace arithm
-{
-    void mulMat_8uc4_32f(PtrStepSz<unsigned int> src1, PtrStepSzf src2, PtrStepSz<unsigned int> dst, cudaStream_t stream);
+void mulMat(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat&, double scale, Stream& stream, int);
+void mulMat_8uc4_32f(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, Stream& stream);
+void mulMat_16sc4_32f(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, Stream& stream);
 
-    void mulMat_16sc4_32f(PtrStepSz<short4> src1, PtrStepSzf src2, PtrStepSz<short4> dst, cudaStream_t stream);
-
-    template <typename T, typename S, typename D>
-    void mulMat(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-}
-
-static void mulMat(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat&, double scale, Stream& _stream, int)
-{
-    typedef void (*func_t)(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    static const func_t funcs[7][7] =
-    {
-        {
-            arithm::mulMat<unsigned char, float, unsigned char>,
-            arithm::mulMat<unsigned char, float, signed char>,
-            arithm::mulMat<unsigned char, float, unsigned short>,
-            arithm::mulMat<unsigned char, float, short>,
-            arithm::mulMat<unsigned char, float, int>,
-            arithm::mulMat<unsigned char, float, float>,
-            arithm::mulMat<unsigned char, double, double>
-        },
-        {
-            arithm::mulMat<signed char, float, unsigned char>,
-            arithm::mulMat<signed char, float, signed char>,
-            arithm::mulMat<signed char, float, unsigned short>,
-            arithm::mulMat<signed char, float, short>,
-            arithm::mulMat<signed char, float, int>,
-            arithm::mulMat<signed char, float, float>,
-            arithm::mulMat<signed char, double, double>
-        },
-        {
-            0 /*arithm::mulMat<unsigned short, float, unsigned char>*/,
-            0 /*arithm::mulMat<unsigned short, float, signed char>*/,
-            arithm::mulMat<unsigned short, float, unsigned short>,
-            arithm::mulMat<unsigned short, float, short>,
-            arithm::mulMat<unsigned short, float, int>,
-            arithm::mulMat<unsigned short, float, float>,
-            arithm::mulMat<unsigned short, double, double>
-        },
-        {
-            0 /*arithm::mulMat<short, float, unsigned char>*/,
-            0 /*arithm::mulMat<short, float, signed char>*/,
-            arithm::mulMat<short, float, unsigned short>,
-            arithm::mulMat<short, float, short>,
-            arithm::mulMat<short, float, int>,
-            arithm::mulMat<short, float, float>,
-            arithm::mulMat<short, double, double>
-        },
-        {
-            0 /*arithm::mulMat<int, float, unsigned char>*/,
-            0 /*arithm::mulMat<int, float, signed char>*/,
-            0 /*arithm::mulMat<int, float, unsigned short>*/,
-            0 /*arithm::mulMat<int, float, short>*/,
-            arithm::mulMat<int, float, int>,
-            arithm::mulMat<int, float, float>,
-            arithm::mulMat<int, double, double>
-        },
-        {
-            0 /*arithm::mulMat<float, float, unsigned char>*/,
-            0 /*arithm::mulMat<float, float, signed char>*/,
-            0 /*arithm::mulMat<float, float, unsigned short>*/,
-            0 /*arithm::mulMat<float, float, short>*/,
-            0 /*arithm::mulMat<float, float, int>*/,
-            arithm::mulMat<float, float, float>,
-            arithm::mulMat<float, double, double>
-        },
-        {
-            0 /*arithm::mulMat<double, double, unsigned char>*/,
-            0 /*arithm::mulMat<double, double, signed char>*/,
-            0 /*arithm::mulMat<double, double, unsigned short>*/,
-            0 /*arithm::mulMat<double, double, short>*/,
-            0 /*arithm::mulMat<double, double, int>*/,
-            0 /*arithm::mulMat<double, double, float>*/,
-            arithm::mulMat<double, double, double>
-        }
-    };
-
-    const int sdepth = src1.depth();
-    const int ddepth = dst.depth();
-    const int cn = src1.channels();
-
-    cudaStream_t stream = StreamAccessor::getStream(_stream);
-
-    PtrStepSzb src1_(src1.rows, src1.cols * cn, src1.data, src1.step);
-    PtrStepSzb src2_(src1.rows, src1.cols * cn, src2.data, src2.step);
-    PtrStepSzb dst_(src1.rows, src1.cols * cn, dst.data, dst.step);
-
-    const func_t func = funcs[sdepth][ddepth];
-
-    if (!func)
-        CV_Error(cv::Error::StsUnsupportedFormat, "Unsupported combination of source and destination types");
-
-    func(src1_, src2_, dst_, scale, stream);
-}
-
-namespace arithm
-{
-    template <typename T, typename S, typename D>
-    void mulScalar(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-}
-
-static void mulScalar(const GpuMat& src, Scalar val, bool, GpuMat& dst, const GpuMat&, double scale, Stream& _stream, int)
-{
-    typedef void (*func_t)(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-    static const func_t funcs[7][7] =
-    {
-        {
-            arithm::mulScalar<unsigned char, float, unsigned char>,
-            arithm::mulScalar<unsigned char, float, signed char>,
-            arithm::mulScalar<unsigned char, float, unsigned short>,
-            arithm::mulScalar<unsigned char, float, short>,
-            arithm::mulScalar<unsigned char, float, int>,
-            arithm::mulScalar<unsigned char, float, float>,
-            arithm::mulScalar<unsigned char, double, double>
-        },
-        {
-            arithm::mulScalar<signed char, float, unsigned char>,
-            arithm::mulScalar<signed char, float, signed char>,
-            arithm::mulScalar<signed char, float, unsigned short>,
-            arithm::mulScalar<signed char, float, short>,
-            arithm::mulScalar<signed char, float, int>,
-            arithm::mulScalar<signed char, float, float>,
-            arithm::mulScalar<signed char, double, double>
-        },
-        {
-            0 /*arithm::mulScalar<unsigned short, float, unsigned char>*/,
-            0 /*arithm::mulScalar<unsigned short, float, signed char>*/,
-            arithm::mulScalar<unsigned short, float, unsigned short>,
-            arithm::mulScalar<unsigned short, float, short>,
-            arithm::mulScalar<unsigned short, float, int>,
-            arithm::mulScalar<unsigned short, float, float>,
-            arithm::mulScalar<unsigned short, double, double>
-        },
-        {
-            0 /*arithm::mulScalar<short, float, unsigned char>*/,
-            0 /*arithm::mulScalar<short, float, signed char>*/,
-            arithm::mulScalar<short, float, unsigned short>,
-            arithm::mulScalar<short, float, short>,
-            arithm::mulScalar<short, float, int>,
-            arithm::mulScalar<short, float, float>,
-            arithm::mulScalar<short, double, double>
-        },
-        {
-            0 /*arithm::mulScalar<int, float, unsigned char>*/,
-            0 /*arithm::mulScalar<int, float, signed char>*/,
-            0 /*arithm::mulScalar<int, float, unsigned short>*/,
-            0 /*arithm::mulScalar<int, float, short>*/,
-            arithm::mulScalar<int, float, int>,
-            arithm::mulScalar<int, float, float>,
-            arithm::mulScalar<int, double, double>
-        },
-        {
-            0 /*arithm::mulScalar<float, float, unsigned char>*/,
-            0 /*arithm::mulScalar<float, float, signed char>*/,
-            0 /*arithm::mulScalar<float, float, unsigned short>*/,
-            0 /*arithm::mulScalar<float, float, short>*/,
-            0 /*arithm::mulScalar<float, float, int>*/,
-            arithm::mulScalar<float, float, float>,
-            arithm::mulScalar<float, double, double>
-        },
-        {
-            0 /*arithm::mulScalar<double, double, unsigned char>*/,
-            0 /*arithm::mulScalar<double, double, signed char>*/,
-            0 /*arithm::mulScalar<double, double, unsigned short>*/,
-            0 /*arithm::mulScalar<double, double, short>*/,
-            0 /*arithm::mulScalar<double, double, int>*/,
-            0 /*arithm::mulScalar<double, double, float>*/,
-            arithm::mulScalar<double, double, double>
-        }
-    };
-
-    typedef void (*npp_func_t)(const PtrStepSzb src, Scalar sc, PtrStepb dst, cudaStream_t stream);
-    static const npp_func_t npp_funcs[7][4] =
-    {
-        {NppArithmScalar<CV_8U , 1, nppiMulC_8u_C1RSfs >::call, 0, NppArithmScalar<CV_8U , 3, nppiMulC_8u_C3RSfs >::call, NppArithmScalar<CV_8U , 4, nppiMulC_8u_C4RSfs >::call},
-        {0                                                    , 0, 0                                                    , 0                                                    },
-        {NppArithmScalar<CV_16U, 1, nppiMulC_16u_C1RSfs>::call, 0, NppArithmScalar<CV_16U, 3, nppiMulC_16u_C3RSfs>::call, NppArithmScalar<CV_16U, 4, nppiMulC_16u_C4RSfs>::call},
-        {NppArithmScalar<CV_16S, 1, nppiMulC_16s_C1RSfs>::call, 0, NppArithmScalar<CV_16S, 3, nppiMulC_16s_C3RSfs>::call, NppArithmScalar<CV_16S, 4, nppiMulC_16s_C4RSfs>::call},
-        {NppArithmScalar<CV_32S, 1, nppiMulC_32s_C1RSfs>::call, 0, NppArithmScalar<CV_32S, 3, nppiMulC_32s_C3RSfs>::call, 0                                                    },
-        {NppArithmScalar<CV_32F, 1, nppiMulC_32f_C1R   >::call, 0, NppArithmScalar<CV_32F, 3, nppiMulC_32f_C3R   >::call, NppArithmScalar<CV_32F, 4, nppiMulC_32f_C4R   >::call},
-        {0                                                    , 0, 0                                                    , 0                                                    }
-    };
-
-    const int sdepth = src.depth();
-    const int ddepth = dst.depth();
-    const int cn = src.channels();
-
-    cudaStream_t stream = StreamAccessor::getStream(_stream);
-
-    val[0] *= scale;
-    val[1] *= scale;
-    val[2] *= scale;
-    val[3] *= scale;
-
-    const npp_func_t npp_func = npp_funcs[sdepth][cn - 1];
-    if (ddepth == sdepth && cn > 1 && npp_func != 0)
-    {
-        npp_func(src, val, dst, stream);
-        return;
-    }
-
-    CV_Assert( cn == 1 );
-
-    const func_t func = funcs[sdepth][ddepth];
-
-    if (!func)
-        CV_Error(cv::Error::StsUnsupportedFormat, "Unsupported combination of source and destination types");
-
-    func(src, val[0], dst, stream);
-}
+void mulScalar(const GpuMat& src, cv::Scalar val, bool, GpuMat& dst, const GpuMat& mask, double scale, Stream& stream, int);
 
 void cv::cuda::multiply(InputArray _src1, InputArray _src2, OutputArray _dst, double scale, int dtype, Stream& stream)
 {
@@ -1064,7 +204,7 @@ void cv::cuda::multiply(InputArray _src1, InputArray _src2, OutputArray _dst, do
         _dst.create(src1.size(), src1.type());
         GpuMat dst = _dst.getGpuMat();
 
-        arithm::mulMat_8uc4_32f(src1, src2, dst, StreamAccessor::getStream(stream));
+        mulMat_8uc4_32f(src1, src2, dst, stream);
     }
     else if (_src1.type() == CV_16SC4 && _src2.type() == CV_32FC1)
     {
@@ -1076,7 +216,7 @@ void cv::cuda::multiply(InputArray _src1, InputArray _src2, OutputArray _dst, do
         _dst.create(src1.size(), src1.type());
         GpuMat dst = _dst.getGpuMat();
 
-        arithm::mulMat_16sc4_32f(src1, src2, dst, StreamAccessor::getStream(stream));
+        mulMat_16sc4_32f(src1, src2, dst, stream);
     }
     else
     {
@@ -1087,229 +227,11 @@ void cv::cuda::multiply(InputArray _src1, InputArray _src2, OutputArray _dst, do
 ////////////////////////////////////////////////////////////////////////
 // divide
 
-namespace arithm
-{
-    void divMat_8uc4_32f(PtrStepSz<unsigned int> src1, PtrStepSzf src2, PtrStepSz<unsigned int> dst, cudaStream_t stream);
+void divMat(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat&, double scale, Stream& stream, int);
+void divMat_8uc4_32f(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, Stream& stream);
+void divMat_16sc4_32f(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, Stream& stream);
 
-    void divMat_16sc4_32f(PtrStepSz<short4> src1, PtrStepSzf src2, PtrStepSz<short4> dst, cudaStream_t stream);
-
-    template <typename T, typename S, typename D>
-    void divMat(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-}
-
-static void divMat(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat&, double scale, Stream& _stream, int)
-{
-    typedef void (*func_t)(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, double scale, cudaStream_t stream);
-    static const func_t funcs[7][7] =
-    {
-        {
-            arithm::divMat<unsigned char, float, unsigned char>,
-            arithm::divMat<unsigned char, float, signed char>,
-            arithm::divMat<unsigned char, float, unsigned short>,
-            arithm::divMat<unsigned char, float, short>,
-            arithm::divMat<unsigned char, float, int>,
-            arithm::divMat<unsigned char, float, float>,
-            arithm::divMat<unsigned char, double, double>
-        },
-        {
-            arithm::divMat<signed char, float, unsigned char>,
-            arithm::divMat<signed char, float, signed char>,
-            arithm::divMat<signed char, float, unsigned short>,
-            arithm::divMat<signed char, float, short>,
-            arithm::divMat<signed char, float, int>,
-            arithm::divMat<signed char, float, float>,
-            arithm::divMat<signed char, double, double>
-        },
-        {
-            0 /*arithm::divMat<unsigned short, float, unsigned char>*/,
-            0 /*arithm::divMat<unsigned short, float, signed char>*/,
-            arithm::divMat<unsigned short, float, unsigned short>,
-            arithm::divMat<unsigned short, float, short>,
-            arithm::divMat<unsigned short, float, int>,
-            arithm::divMat<unsigned short, float, float>,
-            arithm::divMat<unsigned short, double, double>
-        },
-        {
-            0 /*arithm::divMat<short, float, unsigned char>*/,
-            0 /*arithm::divMat<short, float, signed char>*/,
-            arithm::divMat<short, float, unsigned short>,
-            arithm::divMat<short, float, short>,
-            arithm::divMat<short, float, int>,
-            arithm::divMat<short, float, float>,
-            arithm::divMat<short, double, double>
-        },
-        {
-            0 /*arithm::divMat<int, float, unsigned char>*/,
-            0 /*arithm::divMat<int, float, signed char>*/,
-            0 /*arithm::divMat<int, float, unsigned short>*/,
-            0 /*arithm::divMat<int, float, short>*/,
-            arithm::divMat<int, float, int>,
-            arithm::divMat<int, float, float>,
-            arithm::divMat<int, double, double>
-        },
-        {
-            0 /*arithm::divMat<float, float, unsigned char>*/,
-            0 /*arithm::divMat<float, float, signed char>*/,
-            0 /*arithm::divMat<float, float, unsigned short>*/,
-            0 /*arithm::divMat<float, float, short>*/,
-            0 /*arithm::divMat<float, float, int>*/,
-            arithm::divMat<float, float, float>,
-            arithm::divMat<float, double, double>
-        },
-        {
-            0 /*arithm::divMat<double, double, unsigned char>*/,
-            0 /*arithm::divMat<double, double, signed char>*/,
-            0 /*arithm::divMat<double, double, unsigned short>*/,
-            0 /*arithm::divMat<double, double, short>*/,
-            0 /*arithm::divMat<double, double, int>*/,
-            0 /*arithm::divMat<double, double, float>*/,
-            arithm::divMat<double, double, double>
-        }
-    };
-
-    const int sdepth = src1.depth();
-    const int ddepth = dst.depth();
-    const int cn = src1.channels();
-
-    cudaStream_t stream = StreamAccessor::getStream(_stream);
-
-    PtrStepSzb src1_(src1.rows, src1.cols * cn, src1.data, src1.step);
-    PtrStepSzb src2_(src1.rows, src1.cols * cn, src2.data, src2.step);
-    PtrStepSzb dst_(src1.rows, src1.cols * cn, dst.data, dst.step);
-
-    const func_t func = funcs[sdepth][ddepth];
-
-    if (!func)
-        CV_Error(cv::Error::StsUnsupportedFormat, "Unsupported combination of source and destination types");
-
-    func(src1_, src2_, dst_, scale, stream);
-}
-
-namespace arithm
-{
-    template <typename T, typename S, typename D>
-    void divScalar(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-}
-
-static void divScalar(const GpuMat& src, Scalar val, bool inv, GpuMat& dst, const GpuMat&, double scale, Stream& _stream, int)
-{
-    typedef void (*func_t)(PtrStepSzb src1, double val, bool inv, PtrStepSzb dst, cudaStream_t stream);
-    static const func_t funcs[7][7] =
-    {
-        {
-            arithm::divScalar<unsigned char, float, unsigned char>,
-            arithm::divScalar<unsigned char, float, signed char>,
-            arithm::divScalar<unsigned char, float, unsigned short>,
-            arithm::divScalar<unsigned char, float, short>,
-            arithm::divScalar<unsigned char, float, int>,
-            arithm::divScalar<unsigned char, float, float>,
-            arithm::divScalar<unsigned char, double, double>
-        },
-        {
-            arithm::divScalar<signed char, float, unsigned char>,
-            arithm::divScalar<signed char, float, signed char>,
-            arithm::divScalar<signed char, float, unsigned short>,
-            arithm::divScalar<signed char, float, short>,
-            arithm::divScalar<signed char, float, int>,
-            arithm::divScalar<signed char, float, float>,
-            arithm::divScalar<signed char, double, double>
-        },
-        {
-            0 /*arithm::divScalar<unsigned short, float, unsigned char>*/,
-            0 /*arithm::divScalar<unsigned short, float, signed char>*/,
-            arithm::divScalar<unsigned short, float, unsigned short>,
-            arithm::divScalar<unsigned short, float, short>,
-            arithm::divScalar<unsigned short, float, int>,
-            arithm::divScalar<unsigned short, float, float>,
-            arithm::divScalar<unsigned short, double, double>
-        },
-        {
-            0 /*arithm::divScalar<short, float, unsigned char>*/,
-            0 /*arithm::divScalar<short, float, signed char>*/,
-            arithm::divScalar<short, float, unsigned short>,
-            arithm::divScalar<short, float, short>,
-            arithm::divScalar<short, float, int>,
-            arithm::divScalar<short, float, float>,
-            arithm::divScalar<short, double, double>
-        },
-        {
-            0 /*arithm::divScalar<int, float, unsigned char>*/,
-            0 /*arithm::divScalar<int, float, signed char>*/,
-            0 /*arithm::divScalar<int, float, unsigned short>*/,
-            0 /*arithm::divScalar<int, float, short>*/,
-            arithm::divScalar<int, float, int>,
-            arithm::divScalar<int, float, float>,
-            arithm::divScalar<int, double, double>
-        },
-        {
-            0 /*arithm::divScalar<float, float, unsigned char>*/,
-            0 /*arithm::divScalar<float, float, signed char>*/,
-            0 /*arithm::divScalar<float, float, unsigned short>*/,
-            0 /*arithm::divScalar<float, float, short>*/,
-            0 /*arithm::divScalar<float, float, int>*/,
-            arithm::divScalar<float, float, float>,
-            arithm::divScalar<float, double, double>
-        },
-        {
-            0 /*arithm::divScalar<double, double, unsigned char>*/,
-            0 /*arithm::divScalar<double, double, signed char>*/,
-            0 /*arithm::divScalar<double, double, unsigned short>*/,
-            0 /*arithm::divScalar<double, double, short>*/,
-            0 /*arithm::divScalar<double, double, int>*/,
-            0 /*arithm::divScalar<double, double, float>*/,
-            arithm::divScalar<double, double, double>
-        }
-    };
-
-    typedef void (*npp_func_t)(const PtrStepSzb src, Scalar sc, PtrStepb dst, cudaStream_t stream);
-    static const npp_func_t npp_funcs[7][4] =
-    {
-        {NppArithmScalar<CV_8U , 1, nppiDivC_8u_C1RSfs >::call, 0, NppArithmScalar<CV_8U , 3, nppiDivC_8u_C3RSfs >::call, NppArithmScalar<CV_8U , 4, nppiDivC_8u_C4RSfs >::call},
-        {0                                                    , 0, 0                                                    , 0                                                    },
-        {NppArithmScalar<CV_16U, 1, nppiDivC_16u_C1RSfs>::call, 0, NppArithmScalar<CV_16U, 3, nppiDivC_16u_C3RSfs>::call, NppArithmScalar<CV_16U, 4, nppiDivC_16u_C4RSfs>::call},
-        {NppArithmScalar<CV_16S, 1, nppiDivC_16s_C1RSfs>::call, 0, NppArithmScalar<CV_16S, 3, nppiDivC_16s_C3RSfs>::call, NppArithmScalar<CV_16S, 4, nppiDivC_16s_C4RSfs>::call},
-        {NppArithmScalar<CV_32S, 1, nppiDivC_32s_C1RSfs>::call, 0, NppArithmScalar<CV_32S, 3, nppiDivC_32s_C3RSfs>::call, 0                                                    },
-        {NppArithmScalar<CV_32F, 1, nppiDivC_32f_C1R   >::call, 0, NppArithmScalar<CV_32F, 3, nppiDivC_32f_C3R   >::call, NppArithmScalar<CV_32F, 4, nppiDivC_32f_C4R   >::call},
-        {0                                                    , 0, 0                                                    , 0                                                    }
-    };
-
-    const int sdepth = src.depth();
-    const int ddepth = dst.depth();
-    const int cn = src.channels();
-
-    cudaStream_t stream = StreamAccessor::getStream(_stream);
-
-    if (inv)
-    {
-        val[0] *= scale;
-        val[1] *= scale;
-        val[2] *= scale;
-        val[3] *= scale;
-    }
-    else
-    {
-        val[0] /= scale;
-        val[1] /= scale;
-        val[2] /= scale;
-        val[3] /= scale;
-    }
-
-    const npp_func_t npp_func = npp_funcs[sdepth][cn - 1];
-    if (ddepth == sdepth && cn > 1 && npp_func != 0 && !inv)
-    {
-        npp_func(src, val, dst, stream);
-        return;
-    }
-
-    CV_Assert( cn == 1 );
-
-    const func_t func = funcs[sdepth][ddepth];
-
-    if (!func)
-        CV_Error(cv::Error::StsUnsupportedFormat, "Unsupported combination of source and destination types");
-
-    func(src, val[0], inv, dst, stream);
-}
+void divScalar(const GpuMat& src, cv::Scalar val, bool inv, GpuMat& dst, const GpuMat& mask, double scale, Stream& stream, int);
 
 void cv::cuda::divide(InputArray _src1, InputArray _src2, OutputArray _dst, double scale, int dtype, Stream& stream)
 {
@@ -1323,7 +245,7 @@ void cv::cuda::divide(InputArray _src1, InputArray _src2, OutputArray _dst, doub
         _dst.create(src1.size(), src1.type());
         GpuMat dst = _dst.getGpuMat();
 
-        arithm::divMat_8uc4_32f(src1, src2, dst, StreamAccessor::getStream(stream));
+        divMat_8uc4_32f(src1, src2, dst, stream);
     }
     else if (_src1.type() == CV_16SC4 && _src2.type() == CV_32FC1)
     {
@@ -1335,7 +257,7 @@ void cv::cuda::divide(InputArray _src1, InputArray _src2, OutputArray _dst, doub
         _dst.create(src1.size(), src1.type());
         GpuMat dst = _dst.getGpuMat();
 
-        arithm::divMat_16sc4_32f(src1, src2, dst, StreamAccessor::getStream(stream));
+        divMat_16sc4_32f(src1, src2, dst, stream);
     }
     else
     {
@@ -1346,579 +268,27 @@ void cv::cuda::divide(InputArray _src1, InputArray _src2, OutputArray _dst, doub
 //////////////////////////////////////////////////////////////////////////////
 // absdiff
 
-namespace arithm
-{
-    void absDiffMat_v4(PtrStepSz<unsigned int> src1, PtrStepSz<unsigned int> src2, PtrStepSz<unsigned int> dst, cudaStream_t stream);
-    void absDiffMat_v2(PtrStepSz<unsigned int> src1, PtrStepSz<unsigned int> src2, PtrStepSz<unsigned int> dst, cudaStream_t stream);
+void absDiffMat(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat&, double, Stream& stream, int);
 
-    template <typename T>
-    void absDiffMat(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-}
-
-static void absDiffMat(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat&, double, Stream& _stream, int)
-{
-    typedef void (*func_t)(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    static const func_t funcs[] =
-    {
-        arithm::absDiffMat<unsigned char>,
-        arithm::absDiffMat<signed char>,
-        arithm::absDiffMat<unsigned short>,
-        arithm::absDiffMat<short>,
-        arithm::absDiffMat<int>,
-        arithm::absDiffMat<float>,
-        arithm::absDiffMat<double>
-    };
-
-    const int depth = src1.depth();
-    const int cn = src1.channels();
-
-    cudaStream_t stream = StreamAccessor::getStream(_stream);
-
-    PtrStepSzb src1_(src1.rows, src1.cols * cn, src1.data, src1.step);
-    PtrStepSzb src2_(src1.rows, src1.cols * cn, src2.data, src2.step);
-    PtrStepSzb dst_(src1.rows, src1.cols * cn, dst.data, dst.step);
-
-    if (depth == CV_8U || depth == CV_16U)
-    {
-        const intptr_t src1ptr = reinterpret_cast<intptr_t>(src1_.data);
-        const intptr_t src2ptr = reinterpret_cast<intptr_t>(src2_.data);
-        const intptr_t dstptr = reinterpret_cast<intptr_t>(dst_.data);
-
-        const bool isAllAligned = (src1ptr & 31) == 0 && (src2ptr & 31) == 0 && (dstptr & 31) == 0;
-
-        if (isAllAligned)
-        {
-            if (depth == CV_8U && (src1_.cols & 3) == 0)
-            {
-                const int vcols = src1_.cols >> 2;
-
-                arithm::absDiffMat_v4(PtrStepSz<unsigned int>(src1_.rows, vcols, (unsigned int*) src1_.data, src1_.step),
-                                      PtrStepSz<unsigned int>(src1_.rows, vcols, (unsigned int*) src2_.data, src2_.step),
-                                      PtrStepSz<unsigned int>(src1_.rows, vcols, (unsigned int*) dst_.data, dst_.step),
-                                      stream);
-
-                return;
-            }
-            else if (depth == CV_16U && (src1_.cols & 1) == 0)
-            {
-                const int vcols = src1_.cols >> 1;
-
-                arithm::absDiffMat_v2(PtrStepSz<unsigned int>(src1_.rows, vcols, (unsigned int*) src1_.data, src1_.step),
-                                      PtrStepSz<unsigned int>(src1_.rows, vcols, (unsigned int*) src2_.data, src2_.step),
-                                      PtrStepSz<unsigned int>(src1_.rows, vcols, (unsigned int*) dst_.data, dst_.step),
-                                      stream);
-
-                return;
-            }
-        }
-    }
-
-    const func_t func = funcs[depth];
-
-    if (!func)
-        CV_Error(cv::Error::StsUnsupportedFormat, "Unsupported combination of source and destination types");
-
-    func(src1_, src2_, dst_, stream);
-}
-
-namespace arithm
-{
-    template <typename T, typename S>
-    void absDiffScalar(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-}
-
-static void absDiffScalar(const GpuMat& src, Scalar val, bool, GpuMat& dst, const GpuMat&, double, Stream& stream, int)
-{
-    typedef void (*func_t)(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
-    static const func_t funcs[] =
-    {
-        arithm::absDiffScalar<unsigned char, float>,
-        arithm::absDiffScalar<signed char, float>,
-        arithm::absDiffScalar<unsigned short, float>,
-        arithm::absDiffScalar<short, float>,
-        arithm::absDiffScalar<int, float>,
-        arithm::absDiffScalar<float, float>,
-        arithm::absDiffScalar<double, double>
-    };
-
-    const int depth = src.depth();
-
-    funcs[depth](src, val[0], dst, StreamAccessor::getStream(stream));
-}
+void absDiffScalar(const GpuMat& src, cv::Scalar val, bool, GpuMat& dst, const GpuMat&, double, Stream& stream, int);
 
 void cv::cuda::absdiff(InputArray src1, InputArray src2, OutputArray dst, Stream& stream)
 {
     arithm_op(src1, src2, dst, noArray(), 1.0, -1, stream, absDiffMat, absDiffScalar);
 }
 
-//////////////////////////////////////////////////////////////////////////////
-// abs
-
-namespace arithm
-{
-    template <typename T>
-    void absMat(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
-}
-
-void cv::cuda::abs(InputArray _src, OutputArray _dst, Stream& stream)
-{
-    using namespace arithm;
-
-    typedef void (*func_t)(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
-    static const func_t funcs[] =
-    {
-        absMat<unsigned char>,
-        absMat<signed char>,
-        absMat<unsigned short>,
-        absMat<short>,
-        absMat<int>,
-        absMat<float>,
-        absMat<double>
-    };
-
-    GpuMat src = _src.getGpuMat();
-
-    const int depth = src.depth();
-
-    CV_Assert( depth <= CV_64F );
-    CV_Assert( src.channels() == 1 );
-
-    if (depth == CV_64F)
-    {
-        if (!deviceSupports(NATIVE_DOUBLE))
-            CV_Error(cv::Error::StsUnsupportedFormat, "The device doesn't support double");
-    }
-
-    _dst.create(src.size(), src.type());
-    GpuMat dst = _dst.getGpuMat();
-
-    funcs[depth](src, dst, StreamAccessor::getStream(stream));
-}
-
-//////////////////////////////////////////////////////////////////////////////
-// sqr
-
-namespace arithm
-{
-    template <typename T>
-    void sqrMat(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
-}
-
-void cv::cuda::sqr(InputArray _src, OutputArray _dst, Stream& stream)
-{
-    using namespace arithm;
-
-    typedef void (*func_t)(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
-    static const func_t funcs[] =
-    {
-        sqrMat<unsigned char>,
-        sqrMat<signed char>,
-        sqrMat<unsigned short>,
-        sqrMat<short>,
-        sqrMat<int>,
-        sqrMat<float>,
-        sqrMat<double>
-    };
-
-    GpuMat src = _src.getGpuMat();
-
-    const int depth = src.depth();
-
-    CV_Assert( depth <= CV_64F );
-    CV_Assert( src.channels() == 1 );
-
-    if (depth == CV_64F)
-    {
-        if (!deviceSupports(NATIVE_DOUBLE))
-            CV_Error(cv::Error::StsUnsupportedFormat, "The device doesn't support double");
-    }
-
-    _dst.create(src.size(), src.type());
-    GpuMat dst = _dst.getGpuMat();
-
-    funcs[depth](src, dst, StreamAccessor::getStream(stream));
-}
-
-//////////////////////////////////////////////////////////////////////////////
-// sqrt
-
-namespace arithm
-{
-    template <typename T>
-    void sqrtMat(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
-}
-
-void cv::cuda::sqrt(InputArray _src, OutputArray _dst, Stream& stream)
-{
-    using namespace arithm;
-
-    typedef void (*func_t)(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
-    static const func_t funcs[] =
-    {
-        sqrtMat<unsigned char>,
-        sqrtMat<signed char>,
-        sqrtMat<unsigned short>,
-        sqrtMat<short>,
-        sqrtMat<int>,
-        sqrtMat<float>,
-        sqrtMat<double>
-    };
-
-    GpuMat src = _src.getGpuMat();
-
-    const int depth = src.depth();
-
-    CV_Assert( depth <= CV_64F );
-    CV_Assert( src.channels() == 1 );
-
-    if (depth == CV_64F)
-    {
-        if (!deviceSupports(NATIVE_DOUBLE))
-            CV_Error(cv::Error::StsUnsupportedFormat, "The device doesn't support double");
-    }
-
-    _dst.create(src.size(), src.type());
-    GpuMat dst = _dst.getGpuMat();
-
-    funcs[depth](src, dst, StreamAccessor::getStream(stream));
-}
-
-////////////////////////////////////////////////////////////////////////
-// exp
-
-namespace arithm
-{
-    template <typename T>
-    void expMat(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
-}
-
-void cv::cuda::exp(InputArray _src, OutputArray _dst, Stream& stream)
-{
-    using namespace arithm;
-
-    typedef void (*func_t)(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
-    static const func_t funcs[] =
-    {
-        expMat<unsigned char>,
-        expMat<signed char>,
-        expMat<unsigned short>,
-        expMat<short>,
-        expMat<int>,
-        expMat<float>,
-        expMat<double>
-    };
-
-    GpuMat src = _src.getGpuMat();
-
-    const int depth = src.depth();
-
-    CV_Assert( depth <= CV_64F );
-    CV_Assert( src.channels() == 1 );
-
-    if (depth == CV_64F)
-    {
-        if (!deviceSupports(NATIVE_DOUBLE))
-            CV_Error(cv::Error::StsUnsupportedFormat, "The device doesn't support double");
-    }
-
-    _dst.create(src.size(), src.type());
-    GpuMat dst = _dst.getGpuMat();
-
-    funcs[depth](src, dst, StreamAccessor::getStream(stream));
-}
-
-////////////////////////////////////////////////////////////////////////
-// log
-
-namespace arithm
-{
-    template <typename T>
-    void logMat(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
-}
-
-void cv::cuda::log(InputArray _src, OutputArray _dst, Stream& stream)
-{
-    using namespace arithm;
-
-    typedef void (*func_t)(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream);
-    static const func_t funcs[] =
-    {
-        logMat<unsigned char>,
-        logMat<signed char>,
-        logMat<unsigned short>,
-        logMat<short>,
-        logMat<int>,
-        logMat<float>,
-        logMat<double>
-    };
-
-    GpuMat src = _src.getGpuMat();
-
-    const int depth = src.depth();
-
-    CV_Assert( depth <= CV_64F );
-    CV_Assert( src.channels() == 1 );
-
-    if (depth == CV_64F)
-    {
-        if (!deviceSupports(NATIVE_DOUBLE))
-            CV_Error(cv::Error::StsUnsupportedFormat, "The device doesn't support double");
-    }
-
-    _dst.create(src.size(), src.type());
-    GpuMat dst = _dst.getGpuMat();
-
-    funcs[depth](src, dst, StreamAccessor::getStream(stream));
-}
-
-////////////////////////////////////////////////////////////////////////
-// pow
-
-namespace arithm
-{
-    template<typename T> void pow(PtrStepSzb src, double power, PtrStepSzb dst, cudaStream_t stream);
-}
-
-void cv::cuda::pow(InputArray _src, double power, OutputArray _dst, Stream& stream)
-{
-    typedef void (*func_t)(PtrStepSzb src, double power, PtrStepSzb dst, cudaStream_t stream);
-    static const func_t funcs[] =
-    {
-        arithm::pow<unsigned char>,
-        arithm::pow<signed char>,
-        arithm::pow<unsigned short>,
-        arithm::pow<short>,
-        arithm::pow<int>,
-        arithm::pow<float>,
-        arithm::pow<double>
-    };
-
-    GpuMat src = _src.getGpuMat();
-
-    const int depth = src.depth();
-    const int cn = src.channels();
-
-    CV_Assert(depth <= CV_64F);
-
-    if (depth == CV_64F)
-    {
-        if (!deviceSupports(NATIVE_DOUBLE))
-            CV_Error(cv::Error::StsUnsupportedFormat, "The device doesn't support double");
-    }
-
-    _dst.create(src.size(), src.type());
-    GpuMat dst = _dst.getGpuMat();
-
-    PtrStepSzb src_(src.rows, src.cols * cn, src.data, src.step);
-    PtrStepSzb dst_(src.rows, src.cols * cn, dst.data, dst.step);
-
-    funcs[depth](src_, power, dst_, StreamAccessor::getStream(stream));
-}
-
 //////////////////////////////////////////////////////////////////////////////
 // compare
 
-namespace arithm
-{
-    void cmpMatEq_v4(PtrStepSz<uint> src1, PtrStepSz<uint> src2, PtrStepSz<uint> dst, cudaStream_t stream);
-    void cmpMatNe_v4(PtrStepSz<uint> src1, PtrStepSz<uint> src2, PtrStepSz<uint> dst, cudaStream_t stream);
-    void cmpMatLt_v4(PtrStepSz<uint> src1, PtrStepSz<uint> src2, PtrStepSz<uint> dst, cudaStream_t stream);
-    void cmpMatLe_v4(PtrStepSz<uint> src1, PtrStepSz<uint> src2, PtrStepSz<uint> dst, cudaStream_t stream);
+void cmpMat(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat&, double, Stream& stream, int cmpop);
 
-    template <typename T> void cmpMatEq(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template <typename T> void cmpMatNe(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template <typename T> void cmpMatLt(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template <typename T> void cmpMatLe(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-}
-
-static void cmpMat(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat&, double, Stream& _stream, int cmpop)
-{
-    using namespace arithm;
-
-    typedef void (*func_t)(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    static const func_t funcs[7][4] =
-    {
-        {cmpMatEq<unsigned char> , cmpMatNe<unsigned char> , cmpMatLt<unsigned char> , cmpMatLe<unsigned char> },
-        {cmpMatEq<signed char>   , cmpMatNe<signed char>   , cmpMatLt<signed char>   , cmpMatLe<signed char>   },
-        {cmpMatEq<unsigned short>, cmpMatNe<unsigned short>, cmpMatLt<unsigned short>, cmpMatLe<unsigned short>},
-        {cmpMatEq<short>         , cmpMatNe<short>         , cmpMatLt<short>         , cmpMatLe<short>         },
-        {cmpMatEq<int>           , cmpMatNe<int>           , cmpMatLt<int>           , cmpMatLe<int>           },
-        {cmpMatEq<float>         , cmpMatNe<float>         , cmpMatLt<float>         , cmpMatLe<float>         },
-        {cmpMatEq<double>        , cmpMatNe<double>        , cmpMatLt<double>        , cmpMatLe<double>        }
-    };
-
-    typedef void (*func_v4_t)(PtrStepSz<uint> src1, PtrStepSz<uint> src2, PtrStepSz<uint> dst, cudaStream_t stream);
-    static const func_v4_t funcs_v4[] =
-    {
-        cmpMatEq_v4, cmpMatNe_v4, cmpMatLt_v4, cmpMatLe_v4
-    };
-
-    const int depth = src1.depth();
-    const int cn = src1.channels();
-
-    cudaStream_t stream = StreamAccessor::getStream(_stream);
-
-    static const int codes[] =
-    {
-        0, 2, 3, 2, 3, 1
-    };
-    const GpuMat* psrc1[] =
-    {
-        &src1, &src2, &src2, &src1, &src1, &src1
-    };
-    const GpuMat* psrc2[] =
-    {
-        &src2, &src1, &src1, &src2, &src2, &src2
-    };
-
-    const int code = codes[cmpop];
-    PtrStepSzb src1_(src1.rows, src1.cols * cn, psrc1[cmpop]->data, psrc1[cmpop]->step);
-    PtrStepSzb src2_(src1.rows, src1.cols * cn, psrc2[cmpop]->data, psrc2[cmpop]->step);
-    PtrStepSzb dst_(src1.rows, src1.cols * cn, dst.data, dst.step);
-
-    if (depth == CV_8U && (src1_.cols & 3) == 0)
-    {
-        const intptr_t src1ptr = reinterpret_cast<intptr_t>(src1_.data);
-        const intptr_t src2ptr = reinterpret_cast<intptr_t>(src2_.data);
-        const intptr_t dstptr = reinterpret_cast<intptr_t>(dst_.data);
-
-        const bool isAllAligned = (src1ptr & 31) == 0 && (src2ptr & 31) == 0 && (dstptr & 31) == 0;
-
-        if (isAllAligned)
-        {
-            const int vcols = src1_.cols >> 2;
-
-            funcs_v4[code](PtrStepSz<unsigned int>(src1_.rows, vcols, (unsigned int*) src1_.data, src1_.step),
-                           PtrStepSz<unsigned int>(src1_.rows, vcols, (unsigned int*) src2_.data, src2_.step),
-                           PtrStepSz<unsigned int>(src1_.rows, vcols, (unsigned int*) dst_.data, dst_.step),
-                           stream);
-
-            return;
-        }
-    }
-
-    const func_t func = funcs[depth][code];
-
-    func(src1_, src2_, dst_, stream);
-}
-
-namespace arithm
-{
-    template <typename T> void cmpScalarEq(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-    template <typename T> void cmpScalarNe(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-    template <typename T> void cmpScalarLt(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-    template <typename T> void cmpScalarLe(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-    template <typename T> void cmpScalarGt(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-    template <typename T> void cmpScalarGe(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-}
-
-namespace
-{
-    template <typename T> void castScalar(Scalar& sc)
-    {
-        sc.val[0] = saturate_cast<T>(sc.val[0]);
-        sc.val[1] = saturate_cast<T>(sc.val[1]);
-        sc.val[2] = saturate_cast<T>(sc.val[2]);
-        sc.val[3] = saturate_cast<T>(sc.val[3]);
-    }
-}
-
-static void cmpScalar(const GpuMat& src, Scalar val, bool inv, GpuMat& dst, const GpuMat&, double, Stream& stream, int cmpop)
-{
-    using namespace arithm;
-
-    typedef void (*func_t)(PtrStepSzb src, int cn, double val[4], PtrStepSzb dst, cudaStream_t stream);
-    static const func_t funcs[7][6] =
-    {
-        {cmpScalarEq<unsigned char> , cmpScalarGt<unsigned char> , cmpScalarGe<unsigned char> , cmpScalarLt<unsigned char> , cmpScalarLe<unsigned char> , cmpScalarNe<unsigned char> },
-        {cmpScalarEq<signed char>   , cmpScalarGt<signed char>   , cmpScalarGe<signed char>   , cmpScalarLt<signed char>   , cmpScalarLe<signed char>   , cmpScalarNe<signed char>   },
-        {cmpScalarEq<unsigned short>, cmpScalarGt<unsigned short>, cmpScalarGe<unsigned short>, cmpScalarLt<unsigned short>, cmpScalarLe<unsigned short>, cmpScalarNe<unsigned short>},
-        {cmpScalarEq<short>         , cmpScalarGt<short>         , cmpScalarGe<short>         , cmpScalarLt<short>         , cmpScalarLe<short>         , cmpScalarNe<short>         },
-        {cmpScalarEq<int>           , cmpScalarGt<int>           , cmpScalarGe<int>           , cmpScalarLt<int>           , cmpScalarLe<int>           , cmpScalarNe<int>           },
-        {cmpScalarEq<float>         , cmpScalarGt<float>         , cmpScalarGe<float>         , cmpScalarLt<float>         , cmpScalarLe<float>         , cmpScalarNe<float>         },
-        {cmpScalarEq<double>        , cmpScalarGt<double>        , cmpScalarGe<double>        , cmpScalarLt<double>        , cmpScalarLe<double>        , cmpScalarNe<double>        }
-    };
-
-    typedef void (*cast_func_t)(Scalar& sc);
-    static const cast_func_t cast_func[] =
-    {
-        castScalar<unsigned char>, castScalar<signed char>, castScalar<unsigned short>, castScalar<short>, castScalar<int>, castScalar<float>, castScalar<double>
-    };
-
-    if (inv)
-    {
-        // src1 is a scalar; swap it with src2
-        cmpop = cmpop == CMP_LT ? CMP_GT : cmpop == CMP_LE ? CMP_GE :
-            cmpop == CMP_GE ? CMP_LE : cmpop == CMP_GT ? CMP_LT : cmpop;
-    }
-
-    const int depth = src.depth();
-    const int cn = src.channels();
-
-    cast_func[depth](val);
-
-    funcs[depth][cmpop](src, cn, val.val, dst, StreamAccessor::getStream(stream));
-}
+void cmpScalar(const GpuMat& src, Scalar val, bool inv, GpuMat& dst, const GpuMat&, double, Stream& stream, int cmpop);
 
 void cv::cuda::compare(InputArray src1, InputArray src2, OutputArray dst, int cmpop, Stream& stream)
 {
     arithm_op(src1, src2, dst, noArray(), 1.0, CV_8U, stream, cmpMat, cmpScalar, cmpop);
 }
 
-//////////////////////////////////////////////////////////////////////////////
-// bitwise_not
-
-namespace arithm
-{
-    template <typename T> void bitMatNot(PtrStepSzb src, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-}
-
-void cv::cuda::bitwise_not(InputArray _src, OutputArray _dst, InputArray _mask, Stream& _stream)
-{
-    using namespace arithm;
-
-    GpuMat src = _src.getGpuMat();
-    GpuMat mask = _mask.getGpuMat();
-
-    const int depth = src.depth();
-
-    CV_Assert( depth <= CV_64F );
-    CV_Assert( mask.empty() || (mask.type() == CV_8UC1 && mask.size() == src.size()) );
-
-    _dst.create(src.size(), src.type());
-    GpuMat dst = _dst.getGpuMat();
-
-    cudaStream_t stream = StreamAccessor::getStream(_stream);
-
-    const int bcols = (int) (src.cols * src.elemSize());
-
-    if ((bcols & 3) == 0)
-    {
-        const int vcols = bcols >> 2;
-
-        bitMatNot<unsigned int>(
-                    PtrStepSzb(src.rows, vcols, src.data, src.step),
-                    PtrStepSzb(src.rows, vcols, dst.data, dst.step),
-                    mask, stream);
-    }
-    else if ((bcols & 1) == 0)
-    {
-        const int vcols = bcols >> 1;
-
-        bitMatNot<unsigned short>(
-                    PtrStepSzb(src.rows, vcols, src.data, src.step),
-                    PtrStepSzb(src.rows, vcols, dst.data, dst.step),
-                    mask, stream);
-    }
-    else
-    {
-        bitMatNot<unsigned char>(
-                    PtrStepSzb(src.rows, bcols, src.data, src.step),
-                    PtrStepSzb(src.rows, bcols, dst.data, dst.step),
-                    mask, stream);
-    }
-}
-
 //////////////////////////////////////////////////////////////////////////////
 // Binary bitwise logical operations
 
@@ -1932,195 +302,9 @@ namespace
     };
 }
 
-namespace arithm
-{
-    template <typename T> void bitMatAnd(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template <typename T> void bitMatOr(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    template <typename T> void bitMatXor(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-}
+void bitMat(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat& mask, double, Stream& stream, int op);
 
-static void bitMat(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat& mask, double, Stream& _stream, int op)
-{
-    using namespace arithm;
-
-    typedef void (*func_t)(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
-    static const func_t funcs32[] =
-    {
-        bitMatAnd<uint>,
-        bitMatOr<uint>,
-        bitMatXor<uint>
-    };
-    static const func_t funcs16[] =
-    {
-        bitMatAnd<ushort>,
-        bitMatOr<ushort>,
-        bitMatXor<ushort>
-    };
-    static const func_t funcs8[] =
-    {
-        bitMatAnd<uchar>,
-        bitMatOr<uchar>,
-        bitMatXor<uchar>
-    };
-
-    cudaStream_t stream = StreamAccessor::getStream(_stream);
-
-    const int bcols = (int) (src1.cols * src1.elemSize());
-
-    if ((bcols & 3) == 0)
-    {
-        const int vcols = bcols >> 2;
-
-        funcs32[op](PtrStepSzb(src1.rows, vcols, src1.data, src1.step),
-                    PtrStepSzb(src1.rows, vcols, src2.data, src2.step),
-                    PtrStepSzb(src1.rows, vcols, dst.data, dst.step),
-                    mask, stream);
-    }
-    else if ((bcols & 1) == 0)
-    {
-        const int vcols = bcols >> 1;
-
-        funcs16[op](PtrStepSzb(src1.rows, vcols, src1.data, src1.step),
-                    PtrStepSzb(src1.rows, vcols, src2.data, src2.step),
-                    PtrStepSzb(src1.rows, vcols, dst.data, dst.step),
-                    mask, stream);
-    }
-    else
-    {
-
-        funcs8[op](PtrStepSzb(src1.rows, bcols, src1.data, src1.step),
-                   PtrStepSzb(src1.rows, bcols, src2.data, src2.step),
-                   PtrStepSzb(src1.rows, bcols, dst.data, dst.step),
-                   mask, stream);
-    }
-}
-
-namespace arithm
-{
-    template <typename T> void bitScalarAnd(PtrStepSzb src1, unsigned int src2, PtrStepSzb dst, cudaStream_t stream);
-    template <typename T> void bitScalarOr(PtrStepSzb src1, unsigned int src2, PtrStepSzb dst, cudaStream_t stream);
-    template <typename T> void bitScalarXor(PtrStepSzb src1, unsigned int src2, PtrStepSzb dst, cudaStream_t stream);
-}
-
-namespace
-{
-    typedef void (*bit_scalar_func_t)(PtrStepSzb src1, unsigned int src2, PtrStepSzb dst, cudaStream_t stream);
-
-    template <typename T, bit_scalar_func_t func> struct BitScalar
-    {
-        static void call(const GpuMat& src, Scalar sc, GpuMat& dst, cudaStream_t stream)
-        {
-            func(src, saturate_cast<T>(sc.val[0]), dst, stream);
-        }
-    };
-
-    template <bit_scalar_func_t func> struct BitScalar4
-    {
-        static void call(const GpuMat& src, Scalar sc, GpuMat& dst, cudaStream_t stream)
-        {
-            unsigned int packedVal = 0;
-
-            packedVal |= (saturate_cast<unsigned char>(sc.val[0]) & 0xffff);
-            packedVal |= (saturate_cast<unsigned char>(sc.val[1]) & 0xffff) << 8;
-            packedVal |= (saturate_cast<unsigned char>(sc.val[2]) & 0xffff) << 16;
-            packedVal |= (saturate_cast<unsigned char>(sc.val[3]) & 0xffff) << 24;
-
-            func(src, packedVal, dst, stream);
-        }
-    };
-
-    template <int DEPTH, int cn> struct NppBitwiseCFunc
-    {
-        typedef typename NppTypeTraits<DEPTH>::npp_t npp_t;
-
-        typedef NppStatus (*func_t)(const npp_t* pSrc1, int nSrc1Step, const npp_t* pConstants, npp_t* pDst, int nDstStep, NppiSize oSizeROI);
-    };
-    template <int DEPTH> struct NppBitwiseCFunc<DEPTH, 1>
-    {
-        typedef typename NppTypeTraits<DEPTH>::npp_t npp_t;
-
-        typedef NppStatus (*func_t)(const npp_t* pSrc1, int nSrc1Step, const npp_t pConstant, npp_t* pDst, int nDstStep, NppiSize oSizeROI);
-    };
-
-    template <int DEPTH, int cn, typename NppBitwiseCFunc<DEPTH, cn>::func_t func> struct NppBitwiseC
-    {
-        typedef typename NppBitwiseCFunc<DEPTH, cn>::npp_t npp_t;
-
-        static void call(const GpuMat& src, Scalar sc, GpuMat& dst, cudaStream_t stream)
-        {
-            NppStreamHandler h(stream);
-
-            NppiSize oSizeROI;
-            oSizeROI.width = src.cols;
-            oSizeROI.height = src.rows;
-
-            const npp_t pConstants[] = {saturate_cast<npp_t>(sc.val[0]), saturate_cast<npp_t>(sc.val[1]), saturate_cast<npp_t>(sc.val[2]), saturate_cast<npp_t>(sc.val[3])};
-
-            nppSafeCall( func(src.ptr<npp_t>(), static_cast<int>(src.step), pConstants, dst.ptr<npp_t>(), static_cast<int>(dst.step), oSizeROI) );
-
-            if (stream == 0)
-                cudaSafeCall( cudaDeviceSynchronize() );
-        }
-    };
-    template <int DEPTH, typename NppBitwiseCFunc<DEPTH, 1>::func_t func> struct NppBitwiseC<DEPTH, 1, func>
-    {
-        typedef typename NppBitwiseCFunc<DEPTH, 1>::npp_t npp_t;
-
-        static void call(const GpuMat& src, Scalar sc, GpuMat& dst, cudaStream_t stream)
-        {
-            NppStreamHandler h(stream);
-
-            NppiSize oSizeROI;
-            oSizeROI.width = src.cols;
-            oSizeROI.height = src.rows;
-
-            nppSafeCall( func(src.ptr<npp_t>(), static_cast<int>(src.step), saturate_cast<npp_t>(sc.val[0]), dst.ptr<npp_t>(), static_cast<int>(dst.step), oSizeROI) );
-
-            if (stream == 0)
-                cudaSafeCall( cudaDeviceSynchronize() );
-        }
-    };
-}
-
-static void bitScalar(const GpuMat& src, Scalar val, bool, GpuMat& dst, const GpuMat& mask, double, Stream& stream, int op)
-{
-    using namespace arithm;
-
-    typedef void (*func_t)(const GpuMat& src, Scalar sc, GpuMat& dst, cudaStream_t stream);
-    static const func_t funcs[3][5][4] =
-    {
-        {
-            {BitScalar<unsigned char, bitScalarAnd<unsigned char> >::call  , 0, NppBitwiseC<CV_8U , 3, nppiAndC_8u_C3R >::call, BitScalar4< bitScalarAnd<unsigned int> >::call},
-            {0,0,0,0},
-            {BitScalar<unsigned short, bitScalarAnd<unsigned short> >::call, 0, NppBitwiseC<CV_16U, 3, nppiAndC_16u_C3R>::call, NppBitwiseC<CV_16U, 4, nppiAndC_16u_C4R>::call},
-            {0,0,0,0},
-            {BitScalar<int, bitScalarAnd<int> >::call                      , 0, NppBitwiseC<CV_32S, 3, nppiAndC_32s_C3R>::call, NppBitwiseC<CV_32S, 4, nppiAndC_32s_C4R>::call}
-        },
-        {
-            {BitScalar<unsigned char, bitScalarOr<unsigned char> >::call  , 0, NppBitwiseC<CV_8U , 3, nppiOrC_8u_C3R >::call, BitScalar4< bitScalarOr<unsigned int> >::call},
-            {0,0,0,0},
-            {BitScalar<unsigned short, bitScalarOr<unsigned short> >::call, 0, NppBitwiseC<CV_16U, 3, nppiOrC_16u_C3R>::call, NppBitwiseC<CV_16U, 4, nppiOrC_16u_C4R>::call},
-            {0,0,0,0},
-            {BitScalar<int, bitScalarOr<int> >::call                      , 0, NppBitwiseC<CV_32S, 3, nppiOrC_32s_C3R>::call, NppBitwiseC<CV_32S, 4, nppiOrC_32s_C4R>::call}
-        },
-        {
-            {BitScalar<unsigned char, bitScalarXor<unsigned char> >::call  , 0, NppBitwiseC<CV_8U , 3, nppiXorC_8u_C3R >::call, BitScalar4< bitScalarXor<unsigned int> >::call},
-            {0,0,0,0},
-            {BitScalar<unsigned short, bitScalarXor<unsigned short> >::call, 0, NppBitwiseC<CV_16U, 3, nppiXorC_16u_C3R>::call, NppBitwiseC<CV_16U, 4, nppiXorC_16u_C4R>::call},
-            {0,0,0,0},
-            {BitScalar<int, bitScalarXor<int> >::call                      , 0, NppBitwiseC<CV_32S, 3, nppiXorC_32s_C3R>::call, NppBitwiseC<CV_32S, 4, nppiXorC_32s_C4R>::call}
-        }
-    };
-
-    const int depth = src.depth();
-    const int cn = src.channels();
-
-    CV_Assert( depth == CV_8U || depth == CV_16U || depth == CV_32S );
-    CV_Assert( cn == 1 || cn == 3 || cn == 4 );
-    CV_Assert( mask.empty() );
-
-    funcs[op][depth][cn - 1](src, val, dst, StreamAccessor::getStream(stream));
-}
+void bitScalar(const GpuMat& src, cv::Scalar value, bool, GpuMat& dst, const GpuMat& mask, double, Stream& stream, int op);
 
 void cv::cuda::bitwise_or(InputArray src1, InputArray src2, OutputArray dst, InputArray mask, Stream& stream)
 {
@@ -2144,20 +328,20 @@ namespace
 {
     template <int DEPTH, int cn> struct NppShiftFunc
     {
-        typedef typename NppTypeTraits<DEPTH>::npp_t npp_t;
+        typedef typename NPPTypeTraits<DEPTH>::npp_type npp_type;
 
-        typedef NppStatus (*func_t)(const npp_t* pSrc1, int nSrc1Step, const Npp32u* pConstants, npp_t* pDst,  int nDstStep,  NppiSize oSizeROI);
+        typedef NppStatus (*func_t)(const npp_type* pSrc1, int nSrc1Step, const Npp32u* pConstants, npp_type* pDst,  int nDstStep,  NppiSize oSizeROI);
     };
     template <int DEPTH> struct NppShiftFunc<DEPTH, 1>
     {
-        typedef typename NppTypeTraits<DEPTH>::npp_t npp_t;
+        typedef typename NPPTypeTraits<DEPTH>::npp_type npp_type;
 
-        typedef NppStatus (*func_t)(const npp_t* pSrc1, int nSrc1Step, const Npp32u pConstants, npp_t* pDst,  int nDstStep,  NppiSize oSizeROI);
+        typedef NppStatus (*func_t)(const npp_type* pSrc1, int nSrc1Step, const Npp32u pConstants, npp_type* pDst,  int nDstStep,  NppiSize oSizeROI);
     };
 
     template <int DEPTH, int cn, typename NppShiftFunc<DEPTH, cn>::func_t func> struct NppShift
     {
-        typedef typename NppTypeTraits<DEPTH>::npp_t npp_t;
+        typedef typename NPPTypeTraits<DEPTH>::npp_type npp_type;
 
         static void call(const GpuMat& src, Scalar_<Npp32u> sc, GpuMat& dst, cudaStream_t stream)
         {
@@ -2167,7 +351,7 @@ namespace
             oSizeROI.width = src.cols;
             oSizeROI.height = src.rows;
 
-            nppSafeCall( func(src.ptr<npp_t>(), static_cast<int>(src.step), sc.val, dst.ptr<npp_t>(), static_cast<int>(dst.step), oSizeROI) );
+            nppSafeCall( func(src.ptr<npp_type>(), static_cast<int>(src.step), sc.val, dst.ptr<npp_type>(), static_cast<int>(dst.step), oSizeROI) );
 
             if (stream == 0)
                 cudaSafeCall( cudaDeviceSynchronize() );
@@ -2175,7 +359,7 @@ namespace
     };
     template <int DEPTH, typename NppShiftFunc<DEPTH, 1>::func_t func> struct NppShift<DEPTH, 1, func>
     {
-        typedef typename NppTypeTraits<DEPTH>::npp_t npp_t;
+        typedef typename NPPTypeTraits<DEPTH>::npp_type npp_type;
 
         static void call(const GpuMat& src, Scalar_<Npp32u> sc, GpuMat& dst, cudaStream_t stream)
         {
@@ -2185,7 +369,7 @@ namespace
             oSizeROI.width = src.cols;
             oSizeROI.height = src.rows;
 
-            nppSafeCall( func(src.ptr<npp_t>(), static_cast<int>(src.step), sc.val[0], dst.ptr<npp_t>(), static_cast<int>(dst.step), oSizeROI) );
+            nppSafeCall( func(src.ptr<npp_type>(), static_cast<int>(src.step), sc.val[0], dst.ptr<npp_type>(), static_cast<int>(dst.step), oSizeROI) );
 
             if (stream == 0)
                 cudaSafeCall( cudaDeviceSynchronize() );
@@ -2251,158 +435,9 @@ namespace
     };
 }
 
-namespace arithm
-{
-    void minMat_v4(PtrStepSz<unsigned int> src1, PtrStepSz<unsigned int> src2, PtrStepSz<unsigned int> dst, cudaStream_t stream);
-    void minMat_v2(PtrStepSz<unsigned int> src1, PtrStepSz<unsigned int> src2, PtrStepSz<unsigned int> dst, cudaStream_t stream);
-    template <typename T> void minMat(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template <typename T> void minScalar(PtrStepSzb src1, double src2, PtrStepSzb dst, cudaStream_t stream);
+void minMaxMat(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat&, double, Stream& stream, int op);
 
-    void maxMat_v4(PtrStepSz<unsigned int> src1, PtrStepSz<unsigned int> src2, PtrStepSz<unsigned int> dst, cudaStream_t stream);
-    void maxMat_v2(PtrStepSz<unsigned int> src1, PtrStepSz<unsigned int> src2, PtrStepSz<unsigned int> dst, cudaStream_t stream);
-    template <typename T> void maxMat(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    template <typename T> void maxScalar(PtrStepSzb src1, double src2, PtrStepSzb dst, cudaStream_t stream);
-}
-
-void minMaxMat(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat&, double, Stream& _stream, int op)
-{
-    using namespace arithm;
-
-    typedef void (*func_t)(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream);
-    static const func_t funcs[2][7] =
-    {
-        {
-            minMat<unsigned char>,
-            minMat<signed char>,
-            minMat<unsigned short>,
-            minMat<short>,
-            minMat<int>,
-            minMat<float>,
-            minMat<double>
-        },
-        {
-            maxMat<unsigned char>,
-            maxMat<signed char>,
-            maxMat<unsigned short>,
-            maxMat<short>,
-            maxMat<int>,
-            maxMat<float>,
-            maxMat<double>
-        }
-    };
-
-    typedef void (*opt_func_t)(PtrStepSz<unsigned int> src1, PtrStepSz<unsigned int> src2, PtrStepSz<unsigned int> dst, cudaStream_t stream);
-    static const opt_func_t funcs_v4[2] =
-    {
-        minMat_v4, maxMat_v4
-    };
-    static const opt_func_t funcs_v2[2] =
-    {
-        minMat_v2, maxMat_v2
-    };
-
-    const int depth = src1.depth();
-    const int cn = src1.channels();
-
-    CV_Assert( depth <= CV_64F );
-
-    cudaStream_t stream = StreamAccessor::getStream(_stream);
-
-    PtrStepSzb src1_(src1.rows, src1.cols * cn, src1.data, src1.step);
-    PtrStepSzb src2_(src1.rows, src1.cols * cn, src2.data, src2.step);
-    PtrStepSzb dst_(src1.rows, src1.cols * cn, dst.data, dst.step);
-
-    if (depth == CV_8U || depth == CV_16U)
-    {
-        const intptr_t src1ptr = reinterpret_cast<intptr_t>(src1_.data);
-        const intptr_t src2ptr = reinterpret_cast<intptr_t>(src2_.data);
-        const intptr_t dstptr = reinterpret_cast<intptr_t>(dst_.data);
-
-        const bool isAllAligned = (src1ptr & 31) == 0 && (src2ptr & 31) == 0 && (dstptr & 31) == 0;
-
-        if (isAllAligned)
-        {
-            if (depth == CV_8U && (src1_.cols & 3) == 0)
-            {
-                const int vcols = src1_.cols >> 2;
-
-                funcs_v4[op](PtrStepSz<unsigned int>(src1_.rows, vcols, (unsigned int*) src1_.data, src1_.step),
-                             PtrStepSz<unsigned int>(src1_.rows, vcols, (unsigned int*) src2_.data, src2_.step),
-                             PtrStepSz<unsigned int>(src1_.rows, vcols, (unsigned int*) dst_.data, dst_.step),
-                             stream);
-
-                return;
-            }
-            else if (depth == CV_16U && (src1_.cols & 1) == 0)
-            {
-                const int vcols = src1_.cols >> 1;
-
-                funcs_v2[op](PtrStepSz<unsigned int>(src1_.rows, vcols, (unsigned int*) src1_.data, src1_.step),
-                             PtrStepSz<unsigned int>(src1_.rows, vcols, (unsigned int*) src2_.data, src2_.step),
-                             PtrStepSz<unsigned int>(src1_.rows, vcols, (unsigned int*) dst_.data, dst_.step),
-                             stream);
-
-                return;
-            }
-        }
-    }
-
-    const func_t func = funcs[op][depth];
-
-    if (!func)
-        CV_Error(cv::Error::StsUnsupportedFormat, "Unsupported combination of source and destination types");
-
-    func(src1_, src2_, dst_, stream);
-}
-
-namespace
-{
-    template <typename T> double castScalar(double val)
-    {
-        return saturate_cast<T>(val);
-    }
-}
-
-void minMaxScalar(const GpuMat& src, Scalar val, bool, GpuMat& dst, const GpuMat&, double, Stream& stream, int op)
-{
-    using namespace arithm;
-
-    typedef void (*func_t)(PtrStepSzb src1, double src2, PtrStepSzb dst, cudaStream_t stream);
-    static const func_t funcs[2][7] =
-    {
-        {
-            minScalar<unsigned char>,
-            minScalar<signed char>,
-            minScalar<unsigned short>,
-            minScalar<short>,
-            minScalar<int>,
-            minScalar<float>,
-            minScalar<double>
-        },
-        {
-            maxScalar<unsigned char>,
-            maxScalar<signed char>,
-            maxScalar<unsigned short>,
-            maxScalar<short>,
-            maxScalar<int>,
-            maxScalar<float>,
-            maxScalar<double>
-        }
-    };
-
-    typedef double (*cast_func_t)(double sc);
-    static const cast_func_t cast_func[] =
-    {
-        castScalar<unsigned char>, castScalar<signed char>, castScalar<unsigned short>, castScalar<short>, castScalar<int>, castScalar<float>, castScalar<double>
-    };
-
-    const int depth = src.depth();
-
-    CV_Assert( depth <= CV_64F );
-    CV_Assert( src.channels() == 1 );
-
-    funcs[op][depth](src, cast_func[depth](val[0]), dst, StreamAccessor::getStream(stream));
-}
+void minMaxScalar(const GpuMat& src, cv::Scalar value, bool, GpuMat& dst, const GpuMat&, double, Stream& stream, int op);
 
 void cv::cuda::min(InputArray src1, InputArray src2, OutputArray dst, Stream& stream)
 {
@@ -2414,586 +449,6 @@ void cv::cuda::max(InputArray src1, InputArray src2, OutputArray dst, Stream& st
     arithm_op(src1, src2, dst, noArray(), 1.0, -1, stream, minMaxMat, minMaxScalar, MAX_OP);
 }
 
-////////////////////////////////////////////////////////////////////////
-// addWeighted
-
-namespace arithm
-{
-    template <typename T1, typename T2, typename D>
-    void addWeighted(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-}
-
-void cv::cuda::addWeighted(InputArray _src1, double alpha, InputArray _src2, double beta, double gamma, OutputArray _dst, int ddepth, Stream& stream)
-{
-    typedef void (*func_t)(PtrStepSzb src1, double alpha, PtrStepSzb src2, double beta, double gamma, PtrStepSzb dst, cudaStream_t stream);
-    static const func_t funcs[7][7][7] =
-    {
-        {
-            {
-                arithm::addWeighted<unsigned char, unsigned char, unsigned char >,
-                arithm::addWeighted<unsigned char, unsigned char, signed char >,
-                arithm::addWeighted<unsigned char, unsigned char, unsigned short>,
-                arithm::addWeighted<unsigned char, unsigned char, short >,
-                arithm::addWeighted<unsigned char, unsigned char, int   >,
-                arithm::addWeighted<unsigned char, unsigned char, float >,
-                arithm::addWeighted<unsigned char, unsigned char, double>
-            },
-            {
-                arithm::addWeighted<unsigned char, signed char, unsigned char >,
-                arithm::addWeighted<unsigned char, signed char, signed char >,
-                arithm::addWeighted<unsigned char, signed char, unsigned short>,
-                arithm::addWeighted<unsigned char, signed char, short >,
-                arithm::addWeighted<unsigned char, signed char, int   >,
-                arithm::addWeighted<unsigned char, signed char, float >,
-                arithm::addWeighted<unsigned char, signed char, double>
-            },
-            {
-                arithm::addWeighted<unsigned char, unsigned short, unsigned char >,
-                arithm::addWeighted<unsigned char, unsigned short, signed char >,
-                arithm::addWeighted<unsigned char, unsigned short, unsigned short>,
-                arithm::addWeighted<unsigned char, unsigned short, short >,
-                arithm::addWeighted<unsigned char, unsigned short, int   >,
-                arithm::addWeighted<unsigned char, unsigned short, float >,
-                arithm::addWeighted<unsigned char, unsigned short, double>
-            },
-            {
-                arithm::addWeighted<unsigned char, short, unsigned char >,
-                arithm::addWeighted<unsigned char, short, signed char >,
-                arithm::addWeighted<unsigned char, short, unsigned short>,
-                arithm::addWeighted<unsigned char, short, short >,
-                arithm::addWeighted<unsigned char, short, int   >,
-                arithm::addWeighted<unsigned char, short, float >,
-                arithm::addWeighted<unsigned char, short, double>
-            },
-            {
-                arithm::addWeighted<unsigned char, int, unsigned char >,
-                arithm::addWeighted<unsigned char, int, signed char >,
-                arithm::addWeighted<unsigned char, int, unsigned short>,
-                arithm::addWeighted<unsigned char, int, short >,
-                arithm::addWeighted<unsigned char, int, int   >,
-                arithm::addWeighted<unsigned char, int, float >,
-                arithm::addWeighted<unsigned char, int, double>
-            },
-            {
-                arithm::addWeighted<unsigned char, float, unsigned char >,
-                arithm::addWeighted<unsigned char, float, signed char >,
-                arithm::addWeighted<unsigned char, float, unsigned short>,
-                arithm::addWeighted<unsigned char, float, short >,
-                arithm::addWeighted<unsigned char, float, int   >,
-                arithm::addWeighted<unsigned char, float, float >,
-                arithm::addWeighted<unsigned char, float, double>
-            },
-            {
-                arithm::addWeighted<unsigned char, double, unsigned char >,
-                arithm::addWeighted<unsigned char, double, signed char >,
-                arithm::addWeighted<unsigned char, double, unsigned short>,
-                arithm::addWeighted<unsigned char, double, short >,
-                arithm::addWeighted<unsigned char, double, int   >,
-                arithm::addWeighted<unsigned char, double, float >,
-                arithm::addWeighted<unsigned char, double, double>
-            }
-        },
-        {
-            {
-                0/*arithm::addWeighted<signed char, unsigned char, unsigned char >*/,
-                0/*arithm::addWeighted<signed char, unsigned char, signed char >*/,
-                0/*arithm::addWeighted<signed char, unsigned char, unsigned short>*/,
-                0/*arithm::addWeighted<signed char, unsigned char, short >*/,
-                0/*arithm::addWeighted<signed char, unsigned char, int   >*/,
-                0/*arithm::addWeighted<signed char, unsigned char, float >*/,
-                0/*arithm::addWeighted<signed char, unsigned char, double>*/
-            },
-            {
-                arithm::addWeighted<signed char, signed char, unsigned char >,
-                arithm::addWeighted<signed char, signed char, signed char >,
-                arithm::addWeighted<signed char, signed char, unsigned short>,
-                arithm::addWeighted<signed char, signed char, short >,
-                arithm::addWeighted<signed char, signed char, int   >,
-                arithm::addWeighted<signed char, signed char, float >,
-                arithm::addWeighted<signed char, signed char, double>
-            },
-            {
-                arithm::addWeighted<signed char, unsigned short, unsigned char >,
-                arithm::addWeighted<signed char, unsigned short, signed char >,
-                arithm::addWeighted<signed char, unsigned short, unsigned short>,
-                arithm::addWeighted<signed char, unsigned short, short >,
-                arithm::addWeighted<signed char, unsigned short, int   >,
-                arithm::addWeighted<signed char, unsigned short, float >,
-                arithm::addWeighted<signed char, unsigned short, double>
-            },
-            {
-                arithm::addWeighted<signed char, short, unsigned char >,
-                arithm::addWeighted<signed char, short, signed char >,
-                arithm::addWeighted<signed char, short, unsigned short>,
-                arithm::addWeighted<signed char, short, short >,
-                arithm::addWeighted<signed char, short, int   >,
-                arithm::addWeighted<signed char, short, float >,
-                arithm::addWeighted<signed char, short, double>
-            },
-            {
-                arithm::addWeighted<signed char, int, unsigned char >,
-                arithm::addWeighted<signed char, int, signed char >,
-                arithm::addWeighted<signed char, int, unsigned short>,
-                arithm::addWeighted<signed char, int, short >,
-                arithm::addWeighted<signed char, int, int   >,
-                arithm::addWeighted<signed char, int, float >,
-                arithm::addWeighted<signed char, int, double>
-            },
-            {
-                arithm::addWeighted<signed char, float, unsigned char >,
-                arithm::addWeighted<signed char, float, signed char >,
-                arithm::addWeighted<signed char, float, unsigned short>,
-                arithm::addWeighted<signed char, float, short >,
-                arithm::addWeighted<signed char, float, int   >,
-                arithm::addWeighted<signed char, float, float >,
-                arithm::addWeighted<signed char, float, double>
-            },
-            {
-                arithm::addWeighted<signed char, double, unsigned char >,
-                arithm::addWeighted<signed char, double, signed char >,
-                arithm::addWeighted<signed char, double, unsigned short>,
-                arithm::addWeighted<signed char, double, short >,
-                arithm::addWeighted<signed char, double, int   >,
-                arithm::addWeighted<signed char, double, float >,
-                arithm::addWeighted<signed char, double, double>
-            }
-        },
-        {
-            {
-                0/*arithm::addWeighted<unsigned short, unsigned char, unsigned char >*/,
-                0/*arithm::addWeighted<unsigned short, unsigned char, signed char >*/,
-                0/*arithm::addWeighted<unsigned short, unsigned char, unsigned short>*/,
-                0/*arithm::addWeighted<unsigned short, unsigned char, short >*/,
-                0/*arithm::addWeighted<unsigned short, unsigned char, int   >*/,
-                0/*arithm::addWeighted<unsigned short, unsigned char, float >*/,
-                0/*arithm::addWeighted<unsigned short, unsigned char, double>*/
-            },
-            {
-                0/*arithm::addWeighted<unsigned short, signed char, unsigned char >*/,
-                0/*arithm::addWeighted<unsigned short, signed char, signed char >*/,
-                0/*arithm::addWeighted<unsigned short, signed char, unsigned short>*/,
-                0/*arithm::addWeighted<unsigned short, signed char, short >*/,
-                0/*arithm::addWeighted<unsigned short, signed char, int   >*/,
-                0/*arithm::addWeighted<unsigned short, signed char, float >*/,
-                0/*arithm::addWeighted<unsigned short, signed char, double>*/
-            },
-            {
-                arithm::addWeighted<unsigned short, unsigned short, unsigned char >,
-                arithm::addWeighted<unsigned short, unsigned short, signed char >,
-                arithm::addWeighted<unsigned short, unsigned short, unsigned short>,
-                arithm::addWeighted<unsigned short, unsigned short, short >,
-                arithm::addWeighted<unsigned short, unsigned short, int   >,
-                arithm::addWeighted<unsigned short, unsigned short, float >,
-                arithm::addWeighted<unsigned short, unsigned short, double>
-            },
-            {
-                arithm::addWeighted<unsigned short, short, unsigned char >,
-                arithm::addWeighted<unsigned short, short, signed char >,
-                arithm::addWeighted<unsigned short, short, unsigned short>,
-                arithm::addWeighted<unsigned short, short, short >,
-                arithm::addWeighted<unsigned short, short, int   >,
-                arithm::addWeighted<unsigned short, short, float >,
-                arithm::addWeighted<unsigned short, short, double>
-            },
-            {
-                arithm::addWeighted<unsigned short, int, unsigned char >,
-                arithm::addWeighted<unsigned short, int, signed char >,
-                arithm::addWeighted<unsigned short, int, unsigned short>,
-                arithm::addWeighted<unsigned short, int, short >,
-                arithm::addWeighted<unsigned short, int, int   >,
-                arithm::addWeighted<unsigned short, int, float >,
-                arithm::addWeighted<unsigned short, int, double>
-            },
-            {
-                arithm::addWeighted<unsigned short, float, unsigned char >,
-                arithm::addWeighted<unsigned short, float, signed char >,
-                arithm::addWeighted<unsigned short, float, unsigned short>,
-                arithm::addWeighted<unsigned short, float, short >,
-                arithm::addWeighted<unsigned short, float, int   >,
-                arithm::addWeighted<unsigned short, float, float >,
-                arithm::addWeighted<unsigned short, float, double>
-            },
-            {
-                arithm::addWeighted<unsigned short, double, unsigned char >,
-                arithm::addWeighted<unsigned short, double, signed char >,
-                arithm::addWeighted<unsigned short, double, unsigned short>,
-                arithm::addWeighted<unsigned short, double, short >,
-                arithm::addWeighted<unsigned short, double, int   >,
-                arithm::addWeighted<unsigned short, double, float >,
-                arithm::addWeighted<unsigned short, double, double>
-            }
-        },
-        {
-            {
-                0/*arithm::addWeighted<short, unsigned char, unsigned char >*/,
-                0/*arithm::addWeighted<short, unsigned char, signed char >*/,
-                0/*arithm::addWeighted<short, unsigned char, unsigned short>*/,
-                0/*arithm::addWeighted<short, unsigned char, short >*/,
-                0/*arithm::addWeighted<short, unsigned char, int   >*/,
-                0/*arithm::addWeighted<short, unsigned char, float >*/,
-                0/*arithm::addWeighted<short, unsigned char, double>*/
-            },
-            {
-                0/*arithm::addWeighted<short, signed char, unsigned char >*/,
-                0/*arithm::addWeighted<short, signed char, signed char >*/,
-                0/*arithm::addWeighted<short, signed char, unsigned short>*/,
-                0/*arithm::addWeighted<short, signed char, short >*/,
-                0/*arithm::addWeighted<short, signed char, int   >*/,
-                0/*arithm::addWeighted<short, signed char, float >*/,
-                0/*arithm::addWeighted<short, signed char, double>*/
-            },
-            {
-                0/*arithm::addWeighted<short, unsigned short, unsigned char >*/,
-                0/*arithm::addWeighted<short, unsigned short, signed char >*/,
-                0/*arithm::addWeighted<short, unsigned short, unsigned short>*/,
-                0/*arithm::addWeighted<short, unsigned short, short >*/,
-                0/*arithm::addWeighted<short, unsigned short, int   >*/,
-                0/*arithm::addWeighted<short, unsigned short, float >*/,
-                0/*arithm::addWeighted<short, unsigned short, double>*/
-            },
-            {
-                arithm::addWeighted<short, short, unsigned char >,
-                arithm::addWeighted<short, short, signed char >,
-                arithm::addWeighted<short, short, unsigned short>,
-                arithm::addWeighted<short, short, short >,
-                arithm::addWeighted<short, short, int   >,
-                arithm::addWeighted<short, short, float >,
-                arithm::addWeighted<short, short, double>
-            },
-            {
-                arithm::addWeighted<short, int, unsigned char >,
-                arithm::addWeighted<short, int, signed char >,
-                arithm::addWeighted<short, int, unsigned short>,
-                arithm::addWeighted<short, int, short >,
-                arithm::addWeighted<short, int, int   >,
-                arithm::addWeighted<short, int, float >,
-                arithm::addWeighted<short, int, double>
-            },
-            {
-                arithm::addWeighted<short, float, unsigned char >,
-                arithm::addWeighted<short, float, signed char >,
-                arithm::addWeighted<short, float, unsigned short>,
-                arithm::addWeighted<short, float, short >,
-                arithm::addWeighted<short, float, int   >,
-                arithm::addWeighted<short, float, float >,
-                arithm::addWeighted<short, float, double>
-            },
-            {
-                arithm::addWeighted<short, double, unsigned char >,
-                arithm::addWeighted<short, double, signed char >,
-                arithm::addWeighted<short, double, unsigned short>,
-                arithm::addWeighted<short, double, short >,
-                arithm::addWeighted<short, double, int   >,
-                arithm::addWeighted<short, double, float >,
-                arithm::addWeighted<short, double, double>
-            }
-        },
-        {
-            {
-                0/*arithm::addWeighted<int, unsigned char, unsigned char >*/,
-                0/*arithm::addWeighted<int, unsigned char, signed char >*/,
-                0/*arithm::addWeighted<int, unsigned char, unsigned short>*/,
-                0/*arithm::addWeighted<int, unsigned char, short >*/,
-                0/*arithm::addWeighted<int, unsigned char, int   >*/,
-                0/*arithm::addWeighted<int, unsigned char, float >*/,
-                0/*arithm::addWeighted<int, unsigned char, double>*/
-            },
-            {
-                0/*arithm::addWeighted<int, signed char, unsigned char >*/,
-                0/*arithm::addWeighted<int, signed char, signed char >*/,
-                0/*arithm::addWeighted<int, signed char, unsigned short>*/,
-                0/*arithm::addWeighted<int, signed char, short >*/,
-                0/*arithm::addWeighted<int, signed char, int   >*/,
-                0/*arithm::addWeighted<int, signed char, float >*/,
-                0/*arithm::addWeighted<int, signed char, double>*/
-            },
-            {
-                0/*arithm::addWeighted<int, unsigned short, unsigned char >*/,
-                0/*arithm::addWeighted<int, unsigned short, signed char >*/,
-                0/*arithm::addWeighted<int, unsigned short, unsigned short>*/,
-                0/*arithm::addWeighted<int, unsigned short, short >*/,
-                0/*arithm::addWeighted<int, unsigned short, int   >*/,
-                0/*arithm::addWeighted<int, unsigned short, float >*/,
-                0/*arithm::addWeighted<int, unsigned short, double>*/
-            },
-            {
-                0/*arithm::addWeighted<int, short, unsigned char >*/,
-                0/*arithm::addWeighted<int, short, signed char >*/,
-                0/*arithm::addWeighted<int, short, unsigned short>*/,
-                0/*arithm::addWeighted<int, short, short >*/,
-                0/*arithm::addWeighted<int, short, int   >*/,
-                0/*arithm::addWeighted<int, short, float >*/,
-                0/*arithm::addWeighted<int, short, double>*/
-            },
-            {
-                arithm::addWeighted<int, int, unsigned char >,
-                arithm::addWeighted<int, int, signed char >,
-                arithm::addWeighted<int, int, unsigned short>,
-                arithm::addWeighted<int, int, short >,
-                arithm::addWeighted<int, int, int   >,
-                arithm::addWeighted<int, int, float >,
-                arithm::addWeighted<int, int, double>
-            },
-            {
-                arithm::addWeighted<int, float, unsigned char >,
-                arithm::addWeighted<int, float, signed char >,
-                arithm::addWeighted<int, float, unsigned short>,
-                arithm::addWeighted<int, float, short >,
-                arithm::addWeighted<int, float, int   >,
-                arithm::addWeighted<int, float, float >,
-                arithm::addWeighted<int, float, double>
-            },
-            {
-                arithm::addWeighted<int, double, unsigned char >,
-                arithm::addWeighted<int, double, signed char >,
-                arithm::addWeighted<int, double, unsigned short>,
-                arithm::addWeighted<int, double, short >,
-                arithm::addWeighted<int, double, int   >,
-                arithm::addWeighted<int, double, float >,
-                arithm::addWeighted<int, double, double>
-            }
-        },
-        {
-            {
-                0/*arithm::addWeighted<float, unsigned char, unsigned char >*/,
-                0/*arithm::addWeighted<float, unsigned char, signed char >*/,
-                0/*arithm::addWeighted<float, unsigned char, unsigned short>*/,
-                0/*arithm::addWeighted<float, unsigned char, short >*/,
-                0/*arithm::addWeighted<float, unsigned char, int   >*/,
-                0/*arithm::addWeighted<float, unsigned char, float >*/,
-                0/*arithm::addWeighted<float, unsigned char, double>*/
-            },
-            {
-                0/*arithm::addWeighted<float, signed char, unsigned char >*/,
-                0/*arithm::addWeighted<float, signed char, signed char >*/,
-                0/*arithm::addWeighted<float, signed char, unsigned short>*/,
-                0/*arithm::addWeighted<float, signed char, short >*/,
-                0/*arithm::addWeighted<float, signed char, int   >*/,
-                0/*arithm::addWeighted<float, signed char, float >*/,
-                0/*arithm::addWeighted<float, signed char, double>*/
-            },
-            {
-                0/*arithm::addWeighted<float, unsigned short, unsigned char >*/,
-                0/*arithm::addWeighted<float, unsigned short, signed char >*/,
-                0/*arithm::addWeighted<float, unsigned short, unsigned short>*/,
-                0/*arithm::addWeighted<float, unsigned short, short >*/,
-                0/*arithm::addWeighted<float, unsigned short, int   >*/,
-                0/*arithm::addWeighted<float, unsigned short, float >*/,
-                0/*arithm::addWeighted<float, unsigned short, double>*/
-            },
-            {
-                0/*arithm::addWeighted<float, short, unsigned char >*/,
-                0/*arithm::addWeighted<float, short, signed char >*/,
-                0/*arithm::addWeighted<float, short, unsigned short>*/,
-                0/*arithm::addWeighted<float, short, short >*/,
-                0/*arithm::addWeighted<float, short, int   >*/,
-                0/*arithm::addWeighted<float, short, float >*/,
-                0/*arithm::addWeighted<float, short, double>*/
-            },
-            {
-                0/*arithm::addWeighted<float, int, unsigned char >*/,
-                0/*arithm::addWeighted<float, int, signed char >*/,
-                0/*arithm::addWeighted<float, int, unsigned short>*/,
-                0/*arithm::addWeighted<float, int, short >*/,
-                0/*arithm::addWeighted<float, int, int   >*/,
-                0/*arithm::addWeighted<float, int, float >*/,
-                0/*arithm::addWeighted<float, int, double>*/
-            },
-            {
-                arithm::addWeighted<float, float, unsigned char >,
-                arithm::addWeighted<float, float, signed char >,
-                arithm::addWeighted<float, float, unsigned short>,
-                arithm::addWeighted<float, float, short >,
-                arithm::addWeighted<float, float, int   >,
-                arithm::addWeighted<float, float, float >,
-                arithm::addWeighted<float, float, double>
-            },
-            {
-                arithm::addWeighted<float, double, unsigned char >,
-                arithm::addWeighted<float, double, signed char >,
-                arithm::addWeighted<float, double, unsigned short>,
-                arithm::addWeighted<float, double, short >,
-                arithm::addWeighted<float, double, int   >,
-                arithm::addWeighted<float, double, float >,
-                arithm::addWeighted<float, double, double>
-            }
-        },
-        {
-            {
-                0/*arithm::addWeighted<double, unsigned char, unsigned char >*/,
-                0/*arithm::addWeighted<double, unsigned char, signed char >*/,
-                0/*arithm::addWeighted<double, unsigned char, unsigned short>*/,
-                0/*arithm::addWeighted<double, unsigned char, short >*/,
-                0/*arithm::addWeighted<double, unsigned char, int   >*/,
-                0/*arithm::addWeighted<double, unsigned char, float >*/,
-                0/*arithm::addWeighted<double, unsigned char, double>*/
-            },
-            {
-                0/*arithm::addWeighted<double, signed char, unsigned char >*/,
-                0/*arithm::addWeighted<double, signed char, signed char >*/,
-                0/*arithm::addWeighted<double, signed char, unsigned short>*/,
-                0/*arithm::addWeighted<double, signed char, short >*/,
-                0/*arithm::addWeighted<double, signed char, int   >*/,
-                0/*arithm::addWeighted<double, signed char, float >*/,
-                0/*arithm::addWeighted<double, signed char, double>*/
-            },
-            {
-                0/*arithm::addWeighted<double, unsigned short, unsigned char >*/,
-                0/*arithm::addWeighted<double, unsigned short, signed char >*/,
-                0/*arithm::addWeighted<double, unsigned short, unsigned short>*/,
-                0/*arithm::addWeighted<double, unsigned short, short >*/,
-                0/*arithm::addWeighted<double, unsigned short, int   >*/,
-                0/*arithm::addWeighted<double, unsigned short, float >*/,
-                0/*arithm::addWeighted<double, unsigned short, double>*/
-            },
-            {
-                0/*arithm::addWeighted<double, short, unsigned char >*/,
-                0/*arithm::addWeighted<double, short, signed char >*/,
-                0/*arithm::addWeighted<double, short, unsigned short>*/,
-                0/*arithm::addWeighted<double, short, short >*/,
-                0/*arithm::addWeighted<double, short, int   >*/,
-                0/*arithm::addWeighted<double, short, float >*/,
-                0/*arithm::addWeighted<double, short, double>*/
-            },
-            {
-                0/*arithm::addWeighted<double, int, unsigned char >*/,
-                0/*arithm::addWeighted<double, int, signed char >*/,
-                0/*arithm::addWeighted<double, int, unsigned short>*/,
-                0/*arithm::addWeighted<double, int, short >*/,
-                0/*arithm::addWeighted<double, int, int   >*/,
-                0/*arithm::addWeighted<double, int, float >*/,
-                0/*arithm::addWeighted<double, int, double>*/
-            },
-            {
-                0/*arithm::addWeighted<double, float, unsigned char >*/,
-                0/*arithm::addWeighted<double, float, signed char >*/,
-                0/*arithm::addWeighted<double, float, unsigned short>*/,
-                0/*arithm::addWeighted<double, float, short >*/,
-                0/*arithm::addWeighted<double, float, int   >*/,
-                0/*arithm::addWeighted<double, float, float >*/,
-                0/*arithm::addWeighted<double, float, double>*/
-            },
-            {
-                arithm::addWeighted<double, double, unsigned char >,
-                arithm::addWeighted<double, double, signed char >,
-                arithm::addWeighted<double, double, unsigned short>,
-                arithm::addWeighted<double, double, short >,
-                arithm::addWeighted<double, double, int   >,
-                arithm::addWeighted<double, double, float >,
-                arithm::addWeighted<double, double, double>
-            }
-        }
-    };
-
-    GpuMat src1 = _src1.getGpuMat();
-    GpuMat src2 = _src2.getGpuMat();
-
-    int sdepth1 = src1.depth();
-    int sdepth2 = src2.depth();
-    ddepth = ddepth >= 0 ? CV_MAT_DEPTH(ddepth) : std::max(sdepth1, sdepth2);
-    const int cn = src1.channels();
-
-    CV_Assert( src2.size() == src1.size() && src2.channels() == cn );
-    CV_Assert( sdepth1 <= CV_64F && sdepth2 <= CV_64F && ddepth <= CV_64F );
-
-    if (sdepth1 == CV_64F || sdepth2 == CV_64F || ddepth == CV_64F)
-    {
-        if (!deviceSupports(NATIVE_DOUBLE))
-            CV_Error(cv::Error::StsUnsupportedFormat, "The device doesn't support double");
-    }
-
-    _dst.create(src1.size(), CV_MAKE_TYPE(ddepth, cn));
-    GpuMat dst = _dst.getGpuMat();
-
-    PtrStepSzb src1_(src1.rows, src1.cols * cn, src1.data, src1.step);
-    PtrStepSzb src2_(src1.rows, src1.cols * cn, src2.data, src2.step);
-    PtrStepSzb dst_(src1.rows, src1.cols * cn, dst.data, dst.step);
-
-    if (sdepth1 > sdepth2)
-    {
-        std::swap(src1_.data, src2_.data);
-        std::swap(src1_.step, src2_.step);
-        std::swap(alpha, beta);
-        std::swap(sdepth1, sdepth2);
-    }
-
-    const func_t func = funcs[sdepth1][sdepth2][ddepth];
-
-    if (!func)
-        CV_Error(cv::Error::StsUnsupportedFormat, "Unsupported combination of source and destination types");
-
-    func(src1_, alpha, src2_, beta, gamma, dst_, StreamAccessor::getStream(stream));
-}
-
-////////////////////////////////////////////////////////////////////////
-// threshold
-
-namespace arithm
-{
-    template <typename T>
-    void threshold(PtrStepSzb src, PtrStepSzb dst, double thresh, double maxVal, int type, cudaStream_t stream);
-}
-
-double cv::cuda::threshold(InputArray _src, OutputArray _dst, double thresh, double maxVal, int type, Stream& _stream)
-{
-    GpuMat src = _src.getGpuMat();
-
-    const int depth = src.depth();
-
-    CV_Assert( src.channels() == 1 && depth <= CV_64F );
-    CV_Assert( type <= 4/*THRESH_TOZERO_INV*/ );
-
-    if (depth == CV_64F)
-    {
-        if (!deviceSupports(NATIVE_DOUBLE))
-            CV_Error(cv::Error::StsUnsupportedFormat, "The device doesn't support double");
-    }
-
-    _dst.create(src.size(), src.type());
-    GpuMat dst = _dst.getGpuMat();
-
-    cudaStream_t stream = StreamAccessor::getStream(_stream);
-
-    if (src.type() == CV_32FC1 && type == 2/*THRESH_TRUNC*/)
-    {
-        NppStreamHandler h(stream);
-
-        NppiSize sz;
-        sz.width  = src.cols;
-        sz.height = src.rows;
-
-        nppSafeCall( nppiThreshold_32f_C1R(src.ptr<Npp32f>(), static_cast<int>(src.step),
-            dst.ptr<Npp32f>(), static_cast<int>(dst.step), sz, static_cast<Npp32f>(thresh), NPP_CMP_GREATER) );
-
-        if (stream == 0)
-            cudaSafeCall( cudaDeviceSynchronize() );
-    }
-    else
-    {
-        typedef void (*func_t)(PtrStepSzb src, PtrStepSzb dst, double thresh, double maxVal, int type, cudaStream_t stream);
-        static const func_t funcs[] =
-        {
-            arithm::threshold<unsigned char>,
-            arithm::threshold<signed char>,
-            arithm::threshold<unsigned short>,
-            arithm::threshold<short>,
-            arithm::threshold<int>,
-            arithm::threshold<float>,
-            arithm::threshold<double>
-        };
-
-        if (depth != CV_32F && depth != CV_64F)
-        {
-            thresh = cvFloor(thresh);
-            maxVal = cvRound(maxVal);
-        }
-
-        funcs[depth](src, dst, thresh, maxVal, type, stream);
-    }
-
-    return thresh;
-}
-
 ////////////////////////////////////////////////////////////////////////
 // NPP magnitide
 
@@ -3038,110 +493,4 @@ void cv::cuda::magnitudeSqr(InputArray _src, OutputArray _dst, Stream& stream)
     npp_magnitude(src, dst, nppiMagnitudeSqr_32fc32f_C1R, StreamAccessor::getStream(stream));
 }
 
-////////////////////////////////////////////////////////////////////////
-// Polar <-> Cart
-
-namespace cv { namespace cuda { namespace device
-{
-    namespace mathfunc
-    {
-        void cartToPolar_gpu(PtrStepSzf x, PtrStepSzf y, PtrStepSzf mag, bool magSqr, PtrStepSzf angle, bool angleInDegrees, cudaStream_t stream);
-        void polarToCart_gpu(PtrStepSzf mag, PtrStepSzf angle, PtrStepSzf x, PtrStepSzf y, bool angleInDegrees, cudaStream_t stream);
-    }
-}}}
-
-namespace
-{
-    void cartToPolar_caller(const GpuMat& x, const GpuMat& y, GpuMat* mag, bool magSqr, GpuMat* angle, bool angleInDegrees, cudaStream_t stream)
-    {
-        using namespace ::cv::cuda::device::mathfunc;
-
-        CV_Assert(x.size() == y.size() && x.type() == y.type());
-        CV_Assert(x.depth() == CV_32F);
-
-        GpuMat x1cn = x.reshape(1);
-        GpuMat y1cn = y.reshape(1);
-        GpuMat mag1cn = mag ? mag->reshape(1) : GpuMat();
-        GpuMat angle1cn = angle ? angle->reshape(1) : GpuMat();
-
-        cartToPolar_gpu(x1cn, y1cn, mag1cn, magSqr, angle1cn, angleInDegrees, stream);
-    }
-
-    void polarToCart_caller(const GpuMat& mag, const GpuMat& angle, GpuMat& x, GpuMat& y, bool angleInDegrees, cudaStream_t stream)
-    {
-        using namespace ::cv::cuda::device::mathfunc;
-
-        CV_Assert((mag.empty() || mag.size() == angle.size()) && mag.type() == angle.type());
-        CV_Assert(mag.depth() == CV_32F);
-
-        GpuMat mag1cn = mag.reshape(1);
-        GpuMat angle1cn = angle.reshape(1);
-        GpuMat x1cn = x.reshape(1);
-        GpuMat y1cn = y.reshape(1);
-
-        polarToCart_gpu(mag1cn, angle1cn, x1cn, y1cn, angleInDegrees, stream);
-    }
-}
-
-void cv::cuda::magnitude(InputArray _x, InputArray _y, OutputArray _dst, Stream& stream)
-{
-    GpuMat x = _x.getGpuMat();
-    GpuMat y = _y.getGpuMat();
-
-    _dst.create(x.size(), CV_32FC1);
-    GpuMat dst = _dst.getGpuMat();
-
-    cartToPolar_caller(x, y, &dst, false, 0, false, StreamAccessor::getStream(stream));
-}
-
-void cv::cuda::magnitudeSqr(InputArray _x, InputArray _y, OutputArray _dst, Stream& stream)
-{
-    GpuMat x = _x.getGpuMat();
-    GpuMat y = _y.getGpuMat();
-
-    _dst.create(x.size(), CV_32FC1);
-    GpuMat dst = _dst.getGpuMat();
-
-    cartToPolar_caller(x, y, &dst, true, 0, false, StreamAccessor::getStream(stream));
-}
-
-void cv::cuda::phase(InputArray _x, InputArray _y, OutputArray _dst, bool angleInDegrees, Stream& stream)
-{
-    GpuMat x = _x.getGpuMat();
-    GpuMat y = _y.getGpuMat();
-
-    _dst.create(x.size(), CV_32FC1);
-    GpuMat dst = _dst.getGpuMat();
-
-    cartToPolar_caller(x, y, 0, false, &dst, angleInDegrees, StreamAccessor::getStream(stream));
-}
-
-void cv::cuda::cartToPolar(InputArray _x, InputArray _y, OutputArray _mag, OutputArray _angle, bool angleInDegrees, Stream& stream)
-{
-    GpuMat x = _x.getGpuMat();
-    GpuMat y = _y.getGpuMat();
-
-    _mag.create(x.size(), CV_32FC1);
-    GpuMat mag = _mag.getGpuMat();
-
-    _angle.create(x.size(), CV_32FC1);
-    GpuMat angle = _angle.getGpuMat();
-
-    cartToPolar_caller(x, y, &mag, false, &angle, angleInDegrees, StreamAccessor::getStream(stream));
-}
-
-void cv::cuda::polarToCart(InputArray _mag, InputArray _angle, OutputArray _x, OutputArray _y, bool angleInDegrees, Stream& stream)
-{
-    GpuMat mag = _mag.getGpuMat();
-    GpuMat angle = _angle.getGpuMat();
-
-    _x.create(mag.size(), CV_32FC1);
-    GpuMat x = _x.getGpuMat();
-
-    _y.create(mag.size(), CV_32FC1);
-    GpuMat y = _y.getGpuMat();
-
-    polarToCart_caller(mag, angle, x, y, angleInDegrees, StreamAccessor::getStream(stream));
-}
-
 #endif
diff --git a/modules/cudaarithm/src/precomp.hpp b/modules/cudaarithm/src/precomp.hpp
index 38fce941ee..6a756e9d02 100644
--- a/modules/cudaarithm/src/precomp.hpp
+++ b/modules/cudaarithm/src/precomp.hpp
@@ -52,13 +52,6 @@
 
 #include "opencv2/core/private.cuda.hpp"
 
-#include "opencv2/opencv_modules.hpp"
-
-#ifdef HAVE_OPENCV_CUDALEGACY
-#  include "opencv2/cudalegacy.hpp"
-#  include "opencv2/cudalegacy/private.hpp"
-#endif
-
 #ifdef HAVE_CUBLAS
 #  include <cublas.h>
 #endif
diff --git a/modules/cudaarithm/src/reductions.cpp b/modules/cudaarithm/src/reductions.cpp
index 9dd5817d99..c1e2af4ed3 100644
--- a/modules/cudaarithm/src/reductions.cpp
+++ b/modules/cudaarithm/src/reductions.cpp
@@ -133,513 +133,6 @@ double cv::cuda::norm(InputArray _src, int normType, InputArray _mask, GpuMat& b
     return std::max(std::abs(min_val), std::abs(max_val));
 }
 
-double cv::cuda::norm(InputArray _src1, InputArray _src2, GpuMat& buf, int normType)
-{
-#if CUDA_VERSION < 5050
-    (void) buf;
-
-    typedef NppStatus (*func_t)(const Npp8u* pSrc1, int nSrcStep1, const Npp8u* pSrc2, int nSrcStep2, NppiSize oSizeROI, Npp64f* pRetVal);
-
-    static const func_t funcs[] = {nppiNormDiff_Inf_8u_C1R, nppiNormDiff_L1_8u_C1R, nppiNormDiff_L2_8u_C1R};
-#else
-    typedef NppStatus (*func_t)(const Npp8u* pSrc1, int nSrcStep1, const Npp8u* pSrc2, int nSrcStep2,
-        NppiSize oSizeROI, Npp64f* pRetVal, Npp8u * pDeviceBuffer);
-
-    typedef NppStatus (*buf_size_func_t)(NppiSize oSizeROI, int* hpBufferSize);
-
-    static const func_t funcs[] = {nppiNormDiff_Inf_8u_C1R, nppiNormDiff_L1_8u_C1R, nppiNormDiff_L2_8u_C1R};
-
-    static const buf_size_func_t buf_size_funcs[] = {nppiNormDiffInfGetBufferHostSize_8u_C1R, nppiNormDiffL1GetBufferHostSize_8u_C1R, nppiNormDiffL2GetBufferHostSize_8u_C1R};
-#endif
-
-    GpuMat src1 = _src1.getGpuMat();
-    GpuMat src2 = _src2.getGpuMat();
-
-    CV_Assert( src1.type() == CV_8UC1 );
-    CV_Assert( src1.size() == src2.size() && src1.type() == src2.type() );
-    CV_Assert( normType == NORM_INF || normType == NORM_L1 || normType == NORM_L2 );
-
-    NppiSize sz;
-    sz.width  = src1.cols;
-    sz.height = src1.rows;
-
-    const int funcIdx = normType >> 1;
-
-    DeviceBuffer dbuf;
-
-#if CUDA_VERSION < 5050
-    nppSafeCall( funcs[funcIdx](src1.ptr<Npp8u>(), static_cast<int>(src1.step), src2.ptr<Npp8u>(), static_cast<int>(src2.step), sz, dbuf) );
-#else
-    int bufSize;
-    buf_size_funcs[funcIdx](sz, &bufSize);
-
-    ensureSizeIsEnough(1, bufSize, CV_8UC1, buf);
-
-    nppSafeCall( funcs[funcIdx](src1.ptr<Npp8u>(), static_cast<int>(src1.step), src2.ptr<Npp8u>(), static_cast<int>(src2.step), sz, dbuf, buf.data) );
-#endif
-
-    cudaSafeCall( cudaDeviceSynchronize() );
-
-    double retVal;
-    dbuf.download(&retVal);
-
-    return retVal;
-}
-
-////////////////////////////////////////////////////////////////////////
-// Sum
-
-namespace sum
-{
-    void getBufSize(int cols, int rows, int cn, int& bufcols, int& bufrows);
-
-    template <typename T, int cn>
-    void run(PtrStepSzb src, void* buf, double* sum, PtrStepSzb mask);
-
-    template <typename T, int cn>
-    void runAbs(PtrStepSzb src, void* buf, double* sum, PtrStepSzb mask);
-
-    template <typename T, int cn>
-    void runSqr(PtrStepSzb src, void* buf, double* sum, PtrStepSzb mask);
-}
-
-Scalar cv::cuda::sum(InputArray _src, InputArray _mask, GpuMat& buf)
-{
-    GpuMat src = _src.getGpuMat();
-    GpuMat mask = _mask.getGpuMat();
-
-    typedef void (*func_t)(PtrStepSzb src, void* buf, double* sum, PtrStepSzb mask);
-    static const func_t funcs[7][5] =
-    {
-        {0, ::sum::run<uchar , 1>, ::sum::run<uchar , 2>, ::sum::run<uchar , 3>, ::sum::run<uchar , 4>},
-        {0, ::sum::run<schar , 1>, ::sum::run<schar , 2>, ::sum::run<schar , 3>, ::sum::run<schar , 4>},
-        {0, ::sum::run<ushort, 1>, ::sum::run<ushort, 2>, ::sum::run<ushort, 3>, ::sum::run<ushort, 4>},
-        {0, ::sum::run<short , 1>, ::sum::run<short , 2>, ::sum::run<short , 3>, ::sum::run<short , 4>},
-        {0, ::sum::run<int   , 1>, ::sum::run<int   , 2>, ::sum::run<int   , 3>, ::sum::run<int   , 4>},
-        {0, ::sum::run<float , 1>, ::sum::run<float , 2>, ::sum::run<float , 3>, ::sum::run<float , 4>},
-        {0, ::sum::run<double, 1>, ::sum::run<double, 2>, ::sum::run<double, 3>, ::sum::run<double, 4>}
-    };
-
-    CV_Assert( mask.empty() || (mask.type() == CV_8UC1 && mask.size() == src.size()) );
-
-    if (src.depth() == CV_64F)
-    {
-        if (!deviceSupports(NATIVE_DOUBLE))
-            CV_Error(cv::Error::StsUnsupportedFormat, "The device doesn't support double");
-    }
-
-    Size buf_size;
-    ::sum::getBufSize(src.cols, src.rows, src.channels(), buf_size.width, buf_size.height);
-    ensureSizeIsEnough(buf_size, CV_8U, buf);
-    buf.setTo(Scalar::all(0));
-
-    const func_t func = funcs[src.depth()][src.channels()];
-
-    double result[4];
-    func(src, buf.data, result, mask);
-
-    return Scalar(result[0], result[1], result[2], result[3]);
-}
-
-Scalar cv::cuda::absSum(InputArray _src, InputArray _mask, GpuMat& buf)
-{
-    GpuMat src = _src.getGpuMat();
-    GpuMat mask = _mask.getGpuMat();
-
-    typedef void (*func_t)(PtrStepSzb src, void* buf, double* sum, PtrStepSzb mask);
-    static const func_t funcs[7][5] =
-    {
-        {0, ::sum::runAbs<uchar , 1>, ::sum::runAbs<uchar , 2>, ::sum::runAbs<uchar , 3>, ::sum::runAbs<uchar , 4>},
-        {0, ::sum::runAbs<schar , 1>, ::sum::runAbs<schar , 2>, ::sum::runAbs<schar , 3>, ::sum::runAbs<schar , 4>},
-        {0, ::sum::runAbs<ushort, 1>, ::sum::runAbs<ushort, 2>, ::sum::runAbs<ushort, 3>, ::sum::runAbs<ushort, 4>},
-        {0, ::sum::runAbs<short , 1>, ::sum::runAbs<short , 2>, ::sum::runAbs<short , 3>, ::sum::runAbs<short , 4>},
-        {0, ::sum::runAbs<int   , 1>, ::sum::runAbs<int   , 2>, ::sum::runAbs<int   , 3>, ::sum::runAbs<int   , 4>},
-        {0, ::sum::runAbs<float , 1>, ::sum::runAbs<float , 2>, ::sum::runAbs<float , 3>, ::sum::runAbs<float , 4>},
-        {0, ::sum::runAbs<double, 1>, ::sum::runAbs<double, 2>, ::sum::runAbs<double, 3>, ::sum::runAbs<double, 4>}
-    };
-
-    CV_Assert( mask.empty() || (mask.type() == CV_8UC1 && mask.size() == src.size()) );
-
-    if (src.depth() == CV_64F)
-    {
-        if (!deviceSupports(NATIVE_DOUBLE))
-            CV_Error(cv::Error::StsUnsupportedFormat, "The device doesn't support double");
-    }
-
-    Size buf_size;
-    ::sum::getBufSize(src.cols, src.rows, src.channels(), buf_size.width, buf_size.height);
-    ensureSizeIsEnough(buf_size, CV_8U, buf);
-    buf.setTo(Scalar::all(0));
-
-    const func_t func = funcs[src.depth()][src.channels()];
-
-    double result[4];
-    func(src, buf.data, result, mask);
-
-    return Scalar(result[0], result[1], result[2], result[3]);
-}
-
-Scalar cv::cuda::sqrSum(InputArray _src, InputArray _mask, GpuMat& buf)
-{
-    GpuMat src = _src.getGpuMat();
-    GpuMat mask = _mask.getGpuMat();
-
-    typedef void (*func_t)(PtrStepSzb src, void* buf, double* sum, PtrStepSzb mask);
-    static const func_t funcs[7][5] =
-    {
-        {0, ::sum::runSqr<uchar , 1>, ::sum::runSqr<uchar , 2>, ::sum::runSqr<uchar , 3>, ::sum::runSqr<uchar , 4>},
-        {0, ::sum::runSqr<schar , 1>, ::sum::runSqr<schar , 2>, ::sum::runSqr<schar , 3>, ::sum::runSqr<schar , 4>},
-        {0, ::sum::runSqr<ushort, 1>, ::sum::runSqr<ushort, 2>, ::sum::runSqr<ushort, 3>, ::sum::runSqr<ushort, 4>},
-        {0, ::sum::runSqr<short , 1>, ::sum::runSqr<short , 2>, ::sum::runSqr<short , 3>, ::sum::runSqr<short , 4>},
-        {0, ::sum::runSqr<int   , 1>, ::sum::runSqr<int   , 2>, ::sum::runSqr<int   , 3>, ::sum::runSqr<int   , 4>},
-        {0, ::sum::runSqr<float , 1>, ::sum::runSqr<float , 2>, ::sum::runSqr<float , 3>, ::sum::runSqr<float , 4>},
-        {0, ::sum::runSqr<double, 1>, ::sum::runSqr<double, 2>, ::sum::runSqr<double, 3>, ::sum::runSqr<double, 4>}
-    };
-
-    CV_Assert( mask.empty() || (mask.type() == CV_8UC1 && mask.size() == src.size()) );
-
-    if (src.depth() == CV_64F)
-    {
-        if (!deviceSupports(NATIVE_DOUBLE))
-            CV_Error(cv::Error::StsUnsupportedFormat, "The device doesn't support double");
-    }
-
-    Size buf_size;
-    ::sum::getBufSize(src.cols, src.rows, src.channels(), buf_size.width, buf_size.height);
-    ensureSizeIsEnough(buf_size, CV_8U, buf);
-    buf.setTo(Scalar::all(0));
-
-    const func_t func = funcs[src.depth()][src.channels()];
-
-    double result[4];
-    func(src, buf.data, result, mask);
-
-    return Scalar(result[0], result[1], result[2], result[3]);
-}
-
-////////////////////////////////////////////////////////////////////////
-// minMax
-
-namespace minMax
-{
-    void getBufSize(int cols, int rows, int& bufcols, int& bufrows);
-
-    template <typename T>
-    void run(const PtrStepSzb src, const PtrStepb mask, double* minval, double* maxval, PtrStepb buf);
-}
-
-void cv::cuda::minMax(InputArray _src, double* minVal, double* maxVal, InputArray _mask, GpuMat& buf)
-{
-    GpuMat src = _src.getGpuMat();
-    GpuMat mask = _mask.getGpuMat();
-
-    typedef void (*func_t)(const PtrStepSzb src, const PtrStepb mask, double* minval, double* maxval, PtrStepb buf);
-    static const func_t funcs[] =
-    {
-        ::minMax::run<uchar>,
-        ::minMax::run<schar>,
-        ::minMax::run<ushort>,
-        ::minMax::run<short>,
-        ::minMax::run<int>,
-        ::minMax::run<float>,
-        ::minMax::run<double>
-    };
-
-    CV_Assert( src.channels() == 1 );
-    CV_Assert( mask.empty() || (mask.size() == src.size() && mask.type() == CV_8U) );
-
-    if (src.depth() == CV_64F)
-    {
-        if (!deviceSupports(NATIVE_DOUBLE))
-            CV_Error(cv::Error::StsUnsupportedFormat, "The device doesn't support double");
-    }
-
-    Size buf_size;
-    ::minMax::getBufSize(src.cols, src.rows, buf_size.width, buf_size.height);
-    ensureSizeIsEnough(buf_size, CV_8U, buf);
-
-    const func_t func = funcs[src.depth()];
-
-    double temp1, temp2;
-    func(src, mask, minVal ? minVal : &temp1, maxVal ? maxVal : &temp2, buf);
-}
-
-////////////////////////////////////////////////////////////////////////
-// minMaxLoc
-
-namespace minMaxLoc
-{
-    void getBufSize(int cols, int rows, size_t elem_size, int& b1cols, int& b1rows, int& b2cols, int& b2rows);
-
-    template <typename T>
-    void run(const PtrStepSzb src, const PtrStepb mask, double* minval, double* maxval, int* minloc, int* maxloc, PtrStepb valbuf, PtrStep<unsigned int> locbuf);
-}
-
-void cv::cuda::minMaxLoc(InputArray _src, double* minVal, double* maxVal, Point* minLoc, Point* maxLoc,
-                        InputArray _mask, GpuMat& valBuf, GpuMat& locBuf)
-{
-    GpuMat src = _src.getGpuMat();
-    GpuMat mask = _mask.getGpuMat();
-
-    typedef void (*func_t)(const PtrStepSzb src, const PtrStepb mask, double* minval, double* maxval, int* minloc, int* maxloc, PtrStepb valbuf, PtrStep<unsigned int> locbuf);
-    static const func_t funcs[] =
-    {
-        ::minMaxLoc::run<uchar>,
-        ::minMaxLoc::run<schar>,
-        ::minMaxLoc::run<ushort>,
-        ::minMaxLoc::run<short>,
-        ::minMaxLoc::run<int>,
-        ::minMaxLoc::run<float>,
-        ::minMaxLoc::run<double>
-    };
-
-    CV_Assert( src.channels() == 1 );
-    CV_Assert( mask.empty() || (mask.size() == src.size() && mask.type() == CV_8U) );
-
-    if (src.depth() == CV_64F)
-    {
-        if (!deviceSupports(NATIVE_DOUBLE))
-            CV_Error(cv::Error::StsUnsupportedFormat, "The device doesn't support double");
-    }
-
-    Size valbuf_size, locbuf_size;
-    ::minMaxLoc::getBufSize(src.cols, src.rows, src.elemSize(), valbuf_size.width, valbuf_size.height, locbuf_size.width, locbuf_size.height);
-    ensureSizeIsEnough(valbuf_size, CV_8U, valBuf);
-    ensureSizeIsEnough(locbuf_size, CV_8U, locBuf);
-
-    const func_t func = funcs[src.depth()];
-
-    double temp1, temp2;
-    Point temp3, temp4;
-    func(src, mask, minVal ? minVal : &temp1, maxVal ? maxVal : &temp2, minLoc ? &minLoc->x : &temp3.x, maxLoc ? &maxLoc->x : &temp4.x, valBuf, locBuf);
-}
-
-//////////////////////////////////////////////////////////////////////////////
-// countNonZero
-
-namespace countNonZero
-{
-    void getBufSize(int cols, int rows, int& bufcols, int& bufrows);
-
-    template <typename T>
-    int run(const PtrStepSzb src, PtrStep<unsigned int> buf);
-}
-
-int cv::cuda::countNonZero(InputArray _src, GpuMat& buf)
-{
-    GpuMat src = _src.getGpuMat();
-
-    typedef int (*func_t)(const PtrStepSzb src, PtrStep<unsigned int> buf);
-    static const func_t funcs[] =
-    {
-        ::countNonZero::run<uchar>,
-        ::countNonZero::run<schar>,
-        ::countNonZero::run<ushort>,
-        ::countNonZero::run<short>,
-        ::countNonZero::run<int>,
-        ::countNonZero::run<float>,
-        ::countNonZero::run<double>
-    };
-
-    CV_Assert(src.channels() == 1);
-
-    if (src.depth() == CV_64F)
-    {
-        if (!deviceSupports(NATIVE_DOUBLE))
-            CV_Error(cv::Error::StsUnsupportedFormat, "The device doesn't support double");
-    }
-
-    Size buf_size;
-    ::countNonZero::getBufSize(src.cols, src.rows, buf_size.width, buf_size.height);
-    ensureSizeIsEnough(buf_size, CV_8U, buf);
-
-    const func_t func = funcs[src.depth()];
-
-    return func(src, buf);
-}
-
-//////////////////////////////////////////////////////////////////////////////
-// reduce
-
-namespace reduce
-{
-    template <typename T, typename S, typename D>
-    void rows(PtrStepSzb src, void* dst, int op, cudaStream_t stream);
-
-    template <typename T, typename S, typename D>
-    void cols(PtrStepSzb src, void* dst, int cn, int op, cudaStream_t stream);
-}
-
-void cv::cuda::reduce(InputArray _src, OutputArray _dst, int dim, int reduceOp, int dtype, Stream& stream)
-{
-    GpuMat src = _src.getGpuMat();
-
-    CV_Assert( src.channels() <= 4 );
-    CV_Assert( dim == 0 || dim == 1 );
-    CV_Assert( reduceOp == REDUCE_SUM || reduceOp == REDUCE_AVG || reduceOp == REDUCE_MAX || reduceOp == REDUCE_MIN );
-
-    if (dtype < 0)
-        dtype = src.depth();
-
-    _dst.create(1, dim == 0 ? src.cols : src.rows, CV_MAKE_TYPE(CV_MAT_DEPTH(dtype), src.channels()));
-    GpuMat dst = _dst.getGpuMat();
-
-    if (dim == 0)
-    {
-        typedef void (*func_t)(PtrStepSzb src, void* dst, int op, cudaStream_t stream);
-        static const func_t funcs[7][7] =
-        {
-            {
-                ::reduce::rows<unsigned char, int, unsigned char>,
-                0/*::reduce::rows<unsigned char, int, signed char>*/,
-                0/*::reduce::rows<unsigned char, int, unsigned short>*/,
-                0/*::reduce::rows<unsigned char, int, short>*/,
-                ::reduce::rows<unsigned char, int, int>,
-                ::reduce::rows<unsigned char, float, float>,
-                ::reduce::rows<unsigned char, double, double>
-            },
-            {
-                0/*::reduce::rows<signed char, int, unsigned char>*/,
-                0/*::reduce::rows<signed char, int, signed char>*/,
-                0/*::reduce::rows<signed char, int, unsigned short>*/,
-                0/*::reduce::rows<signed char, int, short>*/,
-                0/*::reduce::rows<signed char, int, int>*/,
-                0/*::reduce::rows<signed char, float, float>*/,
-                0/*::reduce::rows<signed char, double, double>*/
-            },
-            {
-                0/*::reduce::rows<unsigned short, int, unsigned char>*/,
-                0/*::reduce::rows<unsigned short, int, signed char>*/,
-                ::reduce::rows<unsigned short, int, unsigned short>,
-                0/*::reduce::rows<unsigned short, int, short>*/,
-                ::reduce::rows<unsigned short, int, int>,
-                ::reduce::rows<unsigned short, float, float>,
-                ::reduce::rows<unsigned short, double, double>
-            },
-            {
-                0/*::reduce::rows<short, int, unsigned char>*/,
-                0/*::reduce::rows<short, int, signed char>*/,
-                0/*::reduce::rows<short, int, unsigned short>*/,
-                ::reduce::rows<short, int, short>,
-                ::reduce::rows<short, int, int>,
-                ::reduce::rows<short, float, float>,
-                ::reduce::rows<short, double, double>
-            },
-            {
-                0/*::reduce::rows<int, int, unsigned char>*/,
-                0/*::reduce::rows<int, int, signed char>*/,
-                0/*::reduce::rows<int, int, unsigned short>*/,
-                0/*::reduce::rows<int, int, short>*/,
-                ::reduce::rows<int, int, int>,
-                ::reduce::rows<int, float, float>,
-                ::reduce::rows<int, double, double>
-            },
-            {
-                0/*::reduce::rows<float, float, unsigned char>*/,
-                0/*::reduce::rows<float, float, signed char>*/,
-                0/*::reduce::rows<float, float, unsigned short>*/,
-                0/*::reduce::rows<float, float, short>*/,
-                0/*::reduce::rows<float, float, int>*/,
-                ::reduce::rows<float, float, float>,
-                ::reduce::rows<float, double, double>
-            },
-            {
-                0/*::reduce::rows<double, double, unsigned char>*/,
-                0/*::reduce::rows<double, double, signed char>*/,
-                0/*::reduce::rows<double, double, unsigned short>*/,
-                0/*::reduce::rows<double, double, short>*/,
-                0/*::reduce::rows<double, double, int>*/,
-                0/*::reduce::rows<double, double, float>*/,
-                ::reduce::rows<double, double, double>
-            }
-        };
-
-        const func_t func = funcs[src.depth()][dst.depth()];
-
-        if (!func)
-            CV_Error(cv::Error::StsUnsupportedFormat, "Unsupported combination of input and output array formats");
-
-        func(src.reshape(1), dst.data, reduceOp, StreamAccessor::getStream(stream));
-    }
-    else
-    {
-        typedef void (*func_t)(PtrStepSzb src, void* dst, int cn, int op, cudaStream_t stream);
-        static const func_t funcs[7][7] =
-        {
-            {
-                ::reduce::cols<unsigned char, int, unsigned char>,
-                0/*::reduce::cols<unsigned char, int, signed char>*/,
-                0/*::reduce::cols<unsigned char, int, unsigned short>*/,
-                0/*::reduce::cols<unsigned char, int, short>*/,
-                ::reduce::cols<unsigned char, int, int>,
-                ::reduce::cols<unsigned char, float, float>,
-                ::reduce::cols<unsigned char, double, double>
-            },
-            {
-                0/*::reduce::cols<signed char, int, unsigned char>*/,
-                0/*::reduce::cols<signed char, int, signed char>*/,
-                0/*::reduce::cols<signed char, int, unsigned short>*/,
-                0/*::reduce::cols<signed char, int, short>*/,
-                0/*::reduce::cols<signed char, int, int>*/,
-                0/*::reduce::cols<signed char, float, float>*/,
-                0/*::reduce::cols<signed char, double, double>*/
-            },
-            {
-                0/*::reduce::cols<unsigned short, int, unsigned char>*/,
-                0/*::reduce::cols<unsigned short, int, signed char>*/,
-                ::reduce::cols<unsigned short, int, unsigned short>,
-                0/*::reduce::cols<unsigned short, int, short>*/,
-                ::reduce::cols<unsigned short, int, int>,
-                ::reduce::cols<unsigned short, float, float>,
-                ::reduce::cols<unsigned short, double, double>
-            },
-            {
-                0/*::reduce::cols<short, int, unsigned char>*/,
-                0/*::reduce::cols<short, int, signed char>*/,
-                0/*::reduce::cols<short, int, unsigned short>*/,
-                ::reduce::cols<short, int, short>,
-                ::reduce::cols<short, int, int>,
-                ::reduce::cols<short, float, float>,
-                ::reduce::cols<short, double, double>
-            },
-            {
-                0/*::reduce::cols<int, int, unsigned char>*/,
-                0/*::reduce::cols<int, int, signed char>*/,
-                0/*::reduce::cols<int, int, unsigned short>*/,
-                0/*::reduce::cols<int, int, short>*/,
-                ::reduce::cols<int, int, int>,
-                ::reduce::cols<int, float, float>,
-                ::reduce::cols<int, double, double>
-            },
-            {
-                0/*::reduce::cols<float, float, unsigned char>*/,
-                0/*::reduce::cols<float, float, signed char>*/,
-                0/*::reduce::cols<float, float, unsigned short>*/,
-                0/*::reduce::cols<float, float, short>*/,
-                0/*::reduce::cols<float, float, int>*/,
-                ::reduce::cols<float, float, float>,
-                ::reduce::cols<float, double, double>
-            },
-            {
-                0/*::reduce::cols<double, double, unsigned char>*/,
-                0/*::reduce::cols<double, double, signed char>*/,
-                0/*::reduce::cols<double, double, unsigned short>*/,
-                0/*::reduce::cols<double, double, short>*/,
-                0/*::reduce::cols<double, double, int>*/,
-                0/*::reduce::cols<double, double, float>*/,
-                ::reduce::cols<double, double, double>
-            }
-        };
-
-        const func_t func = funcs[src.depth()][dst.depth()];
-
-        if (!func)
-            CV_Error(cv::Error::StsUnsupportedFormat, "Unsupported combination of input and output array formats");
-
-        func(src, dst.data, src.channels(), reduceOp, StreamAccessor::getStream(stream));
-    }
-}
-
 ////////////////////////////////////////////////////////////////////////
 // meanStdDev
 
@@ -748,116 +241,4 @@ void cv::cuda::normalize(InputArray _src, OutputArray dst, double a, double b, i
     }
 }
 
-////////////////////////////////////////////////////////////////////////
-// integral
-
-namespace cv { namespace cuda { namespace device
-{
-    namespace imgproc
-    {
-        void shfl_integral_gpu(const PtrStepSzb& img, PtrStepSz<unsigned int> integral, cudaStream_t stream);
-    }
-}}}
-
-void cv::cuda::integral(InputArray _src, OutputArray _dst, GpuMat& buffer, Stream& _stream)
-{
-    GpuMat src = _src.getGpuMat();
-
-    CV_Assert( src.type() == CV_8UC1 );
-
-    cudaStream_t stream = StreamAccessor::getStream(_stream);
-
-    cv::Size whole;
-    cv::Point offset;
-    src.locateROI(whole, offset);
-
-    if (deviceSupports(WARP_SHUFFLE_FUNCTIONS) && src.cols <= 2048
-        && offset.x % 16 == 0 && ((src.cols + 63) / 64) * 64 <= (static_cast<int>(src.step) - offset.x))
-    {
-        ensureSizeIsEnough(((src.rows + 7) / 8) * 8, ((src.cols + 63) / 64) * 64, CV_32SC1, buffer);
-
-        cv::cuda::device::imgproc::shfl_integral_gpu(src, buffer, stream);
-
-        _dst.create(src.rows + 1, src.cols + 1, CV_32SC1);
-        GpuMat dst = _dst.getGpuMat();
-
-        dst.setTo(Scalar::all(0), _stream);
-
-        GpuMat inner = dst(Rect(1, 1, src.cols, src.rows));
-        GpuMat res = buffer(Rect(0, 0, src.cols, src.rows));
-
-        res.copyTo(inner, _stream);
-    }
-    else
-    {
-    #ifndef HAVE_OPENCV_CUDALEGACY
-        throw_no_cuda();
-    #else
-        _dst.create(src.rows + 1, src.cols + 1, CV_32SC1);
-        GpuMat dst = _dst.getGpuMat();
-
-        NcvSize32u roiSize;
-        roiSize.width = src.cols;
-        roiSize.height = src.rows;
-
-        cudaDeviceProp prop;
-        cudaSafeCall( cudaGetDeviceProperties(&prop, cv::cuda::getDevice()) );
-
-        Ncv32u bufSize;
-        ncvSafeCall( nppiStIntegralGetSize_8u32u(roiSize, &bufSize, prop) );
-        ensureSizeIsEnough(1, bufSize, CV_8UC1, buffer);
-
-        NppStStreamHandler h(stream);
-
-        ncvSafeCall( nppiStIntegral_8u32u_C1R(const_cast<Ncv8u*>(src.ptr<Ncv8u>()), static_cast<int>(src.step),
-            dst.ptr<Ncv32u>(), static_cast<int>(dst.step), roiSize, buffer.ptr<Ncv8u>(), bufSize, prop) );
-
-        if (stream == 0)
-            cudaSafeCall( cudaDeviceSynchronize() );
-    #endif
-    }
-}
-
-//////////////////////////////////////////////////////////////////////////////
-// sqrIntegral
-
-void cv::cuda::sqrIntegral(InputArray _src, OutputArray _dst, GpuMat& buf, Stream& _stream)
-{
-#ifndef HAVE_OPENCV_CUDALEGACY
-    (void) _src;
-    (void) _dst;
-    (void) _stream;
-    throw_no_cuda();
-#else
-    GpuMat src = _src.getGpuMat();
-
-    CV_Assert( src.type() == CV_8U );
-
-    NcvSize32u roiSize;
-    roiSize.width = src.cols;
-    roiSize.height = src.rows;
-
-    cudaDeviceProp prop;
-    cudaSafeCall( cudaGetDeviceProperties(&prop, cv::cuda::getDevice()) );
-
-    Ncv32u bufSize;
-    ncvSafeCall(nppiStSqrIntegralGetSize_8u64u(roiSize, &bufSize, prop));
-
-    ensureSizeIsEnough(1, bufSize, CV_8U, buf);
-
-    cudaStream_t stream = StreamAccessor::getStream(_stream);
-
-    NppStStreamHandler h(stream);
-
-    _dst.create(src.rows + 1, src.cols + 1, CV_64F);
-    GpuMat dst = _dst.getGpuMat();
-
-    ncvSafeCall(nppiStSqrIntegral_8u64u_C1R(const_cast<Ncv8u*>(src.ptr<Ncv8u>(0)), static_cast<int>(src.step),
-            dst.ptr<Ncv64u>(0), static_cast<int>(dst.step), roiSize, buf.ptr<Ncv8u>(0), bufSize, prop));
-
-    if (stream == 0)
-        cudaSafeCall( cudaDeviceSynchronize() );
-#endif
-}
-
 #endif
diff --git a/modules/cudaarithm/test/test_arithm.cpp b/modules/cudaarithm/test/test_arithm.cpp
index bd3f250f3d..0ee4e34699 100644
--- a/modules/cudaarithm/test/test_arithm.cpp
+++ b/modules/cudaarithm/test/test_arithm.cpp
@@ -125,43 +125,6 @@ INSTANTIATE_TEST_CASE_P(CUDA_Arithm, GEMM, testing::Combine(
     ALL_GEMM_FLAGS,
     WHOLE_SUBMAT));
 
-///////////////////////////////////////////////////////////////////////////////////////////////////////
-// Integral
-
-PARAM_TEST_CASE(Integral, cv::cuda::DeviceInfo, cv::Size, UseRoi)
-{
-    cv::cuda::DeviceInfo devInfo;
-    cv::Size size;
-    bool useRoi;
-
-    virtual void SetUp()
-    {
-        devInfo = GET_PARAM(0);
-        size = GET_PARAM(1);
-        useRoi = GET_PARAM(2);
-
-        cv::cuda::setDevice(devInfo.deviceID());
-    }
-};
-
-CUDA_TEST_P(Integral, Accuracy)
-{
-    cv::Mat src = randomMat(size, CV_8UC1);
-
-    cv::cuda::GpuMat dst = createMat(cv::Size(src.cols + 1, src.rows + 1), CV_32SC1, useRoi);
-    cv::cuda::integral(loadMat(src, useRoi), dst);
-
-    cv::Mat dst_gold;
-    cv::integral(src, dst_gold, CV_32S);
-
-    EXPECT_MAT_NEAR(dst_gold, dst, 0.0);
-}
-
-INSTANTIATE_TEST_CASE_P(CUDA_Arithm, Integral, testing::Combine(
-    ALL_DEVICES,
-    DIFFERENT_SIZES,
-    WHOLE_SUBMAT));
-
 ////////////////////////////////////////////////////////////////////////////
 // MulSpectrums
 
diff --git a/modules/cudaarithm/test/test_reductions.cpp b/modules/cudaarithm/test/test_reductions.cpp
index 69cb6aff33..68974bcefd 100644
--- a/modules/cudaarithm/test/test_reductions.cpp
+++ b/modules/cudaarithm/test/test_reductions.cpp
@@ -816,4 +816,78 @@ INSTANTIATE_TEST_CASE_P(CUDA_Arithm, MeanStdDev, testing::Combine(
     DIFFERENT_SIZES,
     WHOLE_SUBMAT));
 
+///////////////////////////////////////////////////////////////////////////////////////////////////////
+// Integral
+
+PARAM_TEST_CASE(Integral, cv::cuda::DeviceInfo, cv::Size, UseRoi)
+{
+    cv::cuda::DeviceInfo devInfo;
+    cv::Size size;
+    bool useRoi;
+
+    virtual void SetUp()
+    {
+        devInfo = GET_PARAM(0);
+        size = GET_PARAM(1);
+        useRoi = GET_PARAM(2);
+
+        cv::cuda::setDevice(devInfo.deviceID());
+    }
+};
+
+CUDA_TEST_P(Integral, Accuracy)
+{
+    cv::Mat src = randomMat(size, CV_8UC1);
+
+    cv::cuda::GpuMat dst = createMat(cv::Size(src.cols + 1, src.rows + 1), CV_32SC1, useRoi);
+    cv::cuda::integral(loadMat(src, useRoi), dst);
+
+    cv::Mat dst_gold;
+    cv::integral(src, dst_gold, CV_32S);
+
+    EXPECT_MAT_NEAR(dst_gold, dst, 0.0);
+}
+
+INSTANTIATE_TEST_CASE_P(CUDA_Arithm, Integral, testing::Combine(
+    ALL_DEVICES,
+    DIFFERENT_SIZES,
+    WHOLE_SUBMAT));
+
+///////////////////////////////////////////////////////////////////////////////////////////////////////
+// IntegralSqr
+
+PARAM_TEST_CASE(IntegralSqr, cv::cuda::DeviceInfo, cv::Size, UseRoi)
+{
+    cv::cuda::DeviceInfo devInfo;
+    cv::Size size;
+    bool useRoi;
+
+    virtual void SetUp()
+    {
+        devInfo = GET_PARAM(0);
+        size = GET_PARAM(1);
+        useRoi = GET_PARAM(2);
+
+        cv::cuda::setDevice(devInfo.deviceID());
+    }
+};
+
+CUDA_TEST_P(IntegralSqr, Accuracy)
+{
+    cv::Mat src = randomMat(size, CV_8UC1);
+
+    cv::cuda::GpuMat dst = createMat(cv::Size(src.cols + 1, src.rows + 1), CV_64FC1, useRoi);
+    cv::cuda::sqrIntegral(loadMat(src, useRoi), dst);
+
+    cv::Mat dst_gold, temp;
+    cv::integral(src, temp, dst_gold);
+
+    EXPECT_MAT_NEAR(dst_gold, dst, 0.0);
+}
+
+INSTANTIATE_TEST_CASE_P(CUDA_Arithm, IntegralSqr, testing::Combine(
+    ALL_DEVICES,
+    DIFFERENT_SIZES,
+    WHOLE_SUBMAT));
+
 #endif // HAVE_CUDA
diff --git a/modules/cudev/CMakeLists.txt b/modules/cudev/CMakeLists.txt
index 3c7b0592c7..3ea7790b9c 100644
--- a/modules/cudev/CMakeLists.txt
+++ b/modules/cudev/CMakeLists.txt
@@ -4,7 +4,7 @@ endif()
 
 set(the_description "CUDA device layer")
 
-ocv_warnings_disable(CMAKE_CXX_FLAGS /wd4189 /wd4505 -Wundef -Wmissing-declarations -Wunused-function -Wunused-variable)
+ocv_warnings_disable(CMAKE_CXX_FLAGS /wd4189 /wd4505 -Wundef -Wmissing-declarations -Wunused-function -Wunused-variable -Wenum-compare)
 
 ocv_add_module(cudev)
 
diff --git a/modules/cudev/include/opencv2/cudev.hpp b/modules/cudev/include/opencv2/cudev.hpp
index 63bfc95696..a5fb4f6967 100644
--- a/modules/cudev/include/opencv2/cudev.hpp
+++ b/modules/cudev/include/opencv2/cudev.hpp
@@ -73,7 +73,7 @@
 #include "cudev/block/vec_distance.hpp"
 
 #include "cudev/grid/copy.hpp"
-#include "cudev/grid/glob_reduce.hpp"
+#include "cudev/grid/reduce.hpp"
 #include "cudev/grid/histogram.hpp"
 #include "cudev/grid/integral.hpp"
 #include "cudev/grid/pyramids.hpp"
diff --git a/modules/cudev/include/opencv2/cudev/expr/reduction.hpp b/modules/cudev/include/opencv2/cudev/expr/reduction.hpp
index 0fdde02e53..1f0a3ff0e5 100644
--- a/modules/cudev/include/opencv2/cudev/expr/reduction.hpp
+++ b/modules/cudev/include/opencv2/cudev/expr/reduction.hpp
@@ -47,7 +47,7 @@
 #define __OPENCV_CUDEV_EXPR_REDUCTION_HPP__
 
 #include "../common.hpp"
-#include "../grid/glob_reduce.hpp"
+#include "../grid/reduce.hpp"
 #include "../grid/histogram.hpp"
 #include "../grid/integral.hpp"
 #include "../grid/reduce_to_vec.hpp"
diff --git a/modules/cudev/include/opencv2/cudev/functional/functional.hpp b/modules/cudev/include/opencv2/cudev/functional/functional.hpp
index 3ac532819a..7934f78b94 100644
--- a/modules/cudev/include/opencv2/cudev/functional/functional.hpp
+++ b/modules/cudev/include/opencv2/cudev/functional/functional.hpp
@@ -616,6 +616,30 @@ template <typename T> struct magnitude_func : binary_function<T, T, typename fun
     }
 };
 
+template <typename T> struct magnitude_sqr_func : binary_function<T, T, typename functional_detail::FloatType<T>::type>
+{
+    __device__ __forceinline__ typename functional_detail::FloatType<T>::type operator ()(typename TypeTraits<T>::parameter_type a, typename TypeTraits<T>::parameter_type b) const
+    {
+        return a * a + b * b;
+    }
+};
+
+template <typename T, bool angleInDegrees> struct direction_func : binary_function<T, T, T>
+{
+    __device__ T operator ()(T x, T y) const
+    {
+        atan2_func<T> f;
+        typename atan2_func<T>::result_type angle = f(y, x);
+
+        angle += (angle < 0) * (2.0f * CV_PI_F);
+
+        if (angleInDegrees)
+            angle *= (180.0f / CV_PI_F);
+
+        return saturate_cast<T>(angle);
+    }
+};
+
 template <typename T> struct pow_func : binary_function<T, float, float>
 {
     __device__ __forceinline__ float operator ()(T val, float power) const
diff --git a/modules/cudev/include/opencv2/cudev/grid/detail/integral.hpp b/modules/cudev/include/opencv2/cudev/grid/detail/integral.hpp
index b06d7ddbb6..5c90e99893 100644
--- a/modules/cudev/include/opencv2/cudev/grid/detail/integral.hpp
+++ b/modules/cudev/include/opencv2/cudev/grid/detail/integral.hpp
@@ -594,7 +594,7 @@ namespace integral_detail
             CV_CUDEV_SAFE_CALL( cudaDeviceSynchronize() );
     }
 
-    __host__ static void integral(const GlobPtr<uchar> src, GlobPtr<uint> dst, int rows, int cols, cudaStream_t stream)
+    __host__ static void integral(const GlobPtr<uchar>& src, const GlobPtr<uint>& dst, int rows, int cols, cudaStream_t stream)
     {
         if (deviceSupports(FEATURE_SET_COMPUTE_30)
             && (cols % 16 == 0)
@@ -614,7 +614,7 @@ namespace integral_detail
             CV_CUDEV_SAFE_CALL( cudaDeviceSynchronize() );
     }
 
-    __host__ static void integral(const GlobPtr<uchar> src, GlobPtr<int> dst, int rows, int cols, cudaStream_t stream)
+    __host__ __forceinline__ void integral(const GlobPtr<uchar>& src, const GlobPtr<int>& dst, int rows, int cols, cudaStream_t stream)
     {
         GlobPtr<uint> dstui = globPtr((uint*) dst.data, dst.step);
         integral(src, dstui, rows, cols, stream);
diff --git a/modules/cudev/include/opencv2/cudev/grid/detail/minmaxloc.hpp b/modules/cudev/include/opencv2/cudev/grid/detail/minmaxloc.hpp
new file mode 100644
index 0000000000..793dcc7f0a
--- /dev/null
+++ b/modules/cudev/include/opencv2/cudev/grid/detail/minmaxloc.hpp
@@ -0,0 +1,177 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_GRID_MINMAXLOC_DETAIL_HPP__
+#define __OPENCV_CUDEV_GRID_MINMAXLOC_DETAIL_HPP__
+
+#include "../../common.hpp"
+#include "../../util/vec_traits.hpp"
+#include "../../util/type_traits.hpp"
+#include "../../util/limits.hpp"
+#include "../../block/reduce.hpp"
+
+namespace cv { namespace cudev {
+
+namespace grid_minmaxloc_detail
+{
+    template <int BLOCK_SIZE, class SrcPtr, typename ResType, class MaskPtr>
+    __global__ void minMaxLoc_pass_1(const SrcPtr src, ResType* minVal, ResType* maxVal, int* minLoc, int* maxLoc, const MaskPtr mask, const int rows, const int cols, const int patch_y, const int patch_x)
+    {
+        __shared__ ResType sMinVal[BLOCK_SIZE];
+        __shared__ ResType sMaxVal[BLOCK_SIZE];
+        __shared__ uint sMinLoc[BLOCK_SIZE];
+        __shared__ uint sMaxLoc[BLOCK_SIZE];
+
+        const int x0 = blockIdx.x * blockDim.x * patch_x + threadIdx.x;
+        const int y0 = blockIdx.y * blockDim.y * patch_y + threadIdx.y;
+
+        ResType myMin = numeric_limits<ResType>::max();
+        ResType myMax = -numeric_limits<ResType>::max();
+        int myMinLoc = -1;
+        int myMaxLoc = -1;
+
+        for (int i = 0, y = y0; i < patch_y && y < rows; ++i, y += blockDim.y)
+        {
+            for (int j = 0, x = x0; j < patch_x && x < cols; ++j, x += blockDim.x)
+            {
+                if (mask(y, x))
+                {
+                    const ResType srcVal = src(y, x);
+
+                    if (srcVal < myMin)
+                    {
+                        myMin = srcVal;
+                        myMinLoc = y * cols + x;
+                    }
+
+                    if (srcVal > myMax)
+                    {
+                        myMax = srcVal;
+                        myMaxLoc = y * cols + x;
+                    }
+                }
+            }
+        }
+
+        const int tid = threadIdx.y * blockDim.x + threadIdx.x;
+
+        blockReduceKeyVal<BLOCK_SIZE>(smem_tuple(sMinVal, sMaxVal), tie(myMin, myMax),
+                                      smem_tuple(sMinLoc, sMaxLoc), tie(myMinLoc, myMaxLoc),
+                                      tid,
+                                      make_tuple(less<ResType>(), greater<ResType>()));
+
+        const int bid = blockIdx.y * gridDim.x + blockIdx.x;
+
+        if (tid == 0)
+        {
+            minVal[bid] = myMin;
+            maxVal[bid] = myMax;
+            minLoc[bid] = myMinLoc;
+            maxLoc[bid] = myMaxLoc;
+        }
+    }
+
+    template <int BLOCK_SIZE, typename T>
+    __global__ void minMaxLoc_pass_2(T* minMal, T* maxVal, int* minLoc, int* maxLoc, int count)
+    {
+        __shared__ T sMinVal[BLOCK_SIZE];
+        __shared__ T sMaxVal[BLOCK_SIZE];
+        __shared__ int sMinLoc[BLOCK_SIZE];
+        __shared__ int sMaxLoc[BLOCK_SIZE];
+
+        const int idx = ::min(threadIdx.x, count - 1);
+
+        T myMin = minMal[idx];
+        T myMax = maxVal[idx];
+        int myMinLoc = minLoc[idx];
+        int myMaxLoc = maxLoc[idx];
+
+        blockReduceKeyVal<BLOCK_SIZE>(smem_tuple(sMinVal, sMaxVal), tie(myMin, myMax),
+                                      smem_tuple(sMinLoc, sMaxLoc), tie(myMinLoc, myMaxLoc),
+                                      threadIdx.x,
+                                      make_tuple(less<T>(), greater<T>()));
+
+        if (threadIdx.x == 0)
+        {
+            minMal[0] = myMin;
+            maxVal[0] = myMax;
+            minLoc[0] = myMinLoc;
+            maxLoc[0] = myMaxLoc;
+        }
+    }
+
+    template <class Policy>
+    void getLaunchCfg(int rows, int cols, dim3& block, dim3& grid)
+    {
+        block = dim3(Policy::block_size_x, Policy::block_size_y);
+        grid = dim3(divUp(cols, block.x * Policy::patch_size_x), divUp(rows, block.y * Policy::patch_size_y));
+
+        grid.x = ::min(grid.x, block.x);
+        grid.y = ::min(grid.y, block.y);
+    }
+
+    template <class Policy, class SrcPtr, typename ResType, class MaskPtr>
+    __host__ void minMaxLoc(const SrcPtr& src, ResType* minVal, ResType* maxVal, int* minLoc, int* maxLoc, const MaskPtr& mask, int rows, int cols, cudaStream_t stream)
+    {
+        dim3 block, grid;
+        getLaunchCfg<Policy>(cols, rows, block, grid);
+
+        const int patch_x = divUp(divUp(cols, grid.x), block.x);
+        const int patch_y = divUp(divUp(rows, grid.y), block.y);
+
+        minMaxLoc_pass_1<Policy::block_size_x * Policy::block_size_y><<<grid, block, 0, stream>>>(src, minVal, maxVal, minLoc, maxLoc, mask, rows, cols, patch_y, patch_x);
+        CV_CUDEV_SAFE_CALL( cudaGetLastError() );
+
+        minMaxLoc_pass_2<Policy::block_size_x * Policy::block_size_y><<<1, Policy::block_size_x * Policy::block_size_y, 0, stream>>>(minVal, maxVal, minLoc, maxLoc, grid.x * grid.y);
+        CV_CUDEV_SAFE_CALL( cudaGetLastError() );
+
+        if (stream == 0)
+            CV_CUDEV_SAFE_CALL( cudaDeviceSynchronize() );
+    }
+}
+
+}}
+
+#endif
diff --git a/modules/cudev/include/opencv2/cudev/grid/detail/glob_reduce.hpp b/modules/cudev/include/opencv2/cudev/grid/detail/reduce.hpp
similarity index 90%
rename from modules/cudev/include/opencv2/cudev/grid/detail/glob_reduce.hpp
rename to modules/cudev/include/opencv2/cudev/grid/detail/reduce.hpp
index c9d715982c..7ccdd05f8a 100644
--- a/modules/cudev/include/opencv2/cudev/grid/detail/glob_reduce.hpp
+++ b/modules/cudev/include/opencv2/cudev/grid/detail/reduce.hpp
@@ -43,8 +43,8 @@
 
 #pragma once
 
-#ifndef __OPENCV_CUDEV_GRID_GLOB_REDUCE_DETAIL_HPP__
-#define __OPENCV_CUDEV_GRID_GLOB_REDUCE_DETAIL_HPP__
+#ifndef __OPENCV_CUDEV_GRID_REDUCE_DETAIL_HPP__
+#define __OPENCV_CUDEV_GRID_REDUCE_DETAIL_HPP__
 
 #include "../../common.hpp"
 #include "../../util/tuple.hpp"
@@ -59,7 +59,7 @@
 
 namespace cv { namespace cudev {
 
-namespace grid_glob_reduce_detail
+namespace grid_reduce_detail
 {
     // Unroll
 
@@ -389,7 +389,7 @@ namespace grid_glob_reduce_detail
     // glob_reduce
 
     template <class Reductor, int BLOCK_SIZE, int PATCH_X, int PATCH_Y, class SrcPtr, typename ResType, class MaskPtr>
-    __global__ void glob_reduce(const SrcPtr src, ResType* result, const MaskPtr mask, const int rows, const int cols)
+    __global__ void reduce(const SrcPtr src, ResType* result, const MaskPtr mask, const int rows, const int cols)
     {
         const int x0 = blockIdx.x * blockDim.x * PATCH_X + threadIdx.x;
         const int y0 = blockIdx.y * blockDim.y * PATCH_Y + threadIdx.y;
@@ -413,14 +413,12 @@ namespace grid_glob_reduce_detail
     }
 
     template <class Reductor, class Policy, class SrcPtr, typename ResType, class MaskPtr>
-    __host__ void glob_reduce(const SrcPtr& src, ResType* result, const MaskPtr& mask, int rows, int cols, cudaStream_t stream)
+    __host__ void reduce(const SrcPtr& src, ResType* result, const MaskPtr& mask, int rows, int cols, cudaStream_t stream)
     {
         const dim3 block(Policy::block_size_x, Policy::block_size_y);
         const dim3 grid(divUp(cols, block.x * Policy::patch_size_x), divUp(rows, block.y * Policy::patch_size_y));
 
-        const int BLOCK_SIZE = Policy::block_size_x * Policy::block_size_y;
-
-        glob_reduce<Reductor, BLOCK_SIZE, Policy::patch_size_x, Policy::patch_size_y><<<grid, block, 0, stream>>>(src, result, mask, rows, cols);
+        reduce<Reductor, Policy::block_size_x * Policy::block_size_y, Policy::patch_size_x, Policy::patch_size_y><<<grid, block, 0, stream>>>(src, result, mask, rows, cols);
         CV_CUDEV_SAFE_CALL( cudaGetLastError() );
 
         if (stream == 0)
@@ -433,40 +431,33 @@ namespace grid_glob_reduce_detail
     __host__ void sum(const SrcPtr& src, ResType* result, const MaskPtr& mask, int rows, int cols, cudaStream_t stream)
     {
         typedef typename PtrTraits<SrcPtr>::value_type src_type;
-        const int cn = VecTraits<src_type>::cn;
-        typedef typename MakeVec<ResType, cn>::type work_type;
+        typedef typename VecTraits<ResType>::elem_type res_elem_type;
 
-        glob_reduce<SumReductor<src_type, work_type>, Policy>(src, result, mask, rows, cols, stream);
+        reduce<SumReductor<src_type, ResType>, Policy>(src, (res_elem_type*) result, mask, rows, cols, stream);
     }
 
     template <class Policy, class SrcPtr, typename ResType, class MaskPtr>
     __host__ void minVal(const SrcPtr& src, ResType* result, const MaskPtr& mask, int rows, int cols, cudaStream_t stream)
     {
         typedef typename PtrTraits<SrcPtr>::value_type src_type;
-        const int cn = VecTraits<src_type>::cn;
-        typedef typename MakeVec<ResType, cn>::type work_type;
 
-        glob_reduce<MinMaxReductor<minop<work_type>, src_type, work_type>, Policy>(src, result, mask, rows, cols, stream);
+        reduce<MinMaxReductor<minop<ResType>, src_type, ResType>, Policy>(src, result, mask, rows, cols, stream);
     }
 
     template <class Policy, class SrcPtr, typename ResType, class MaskPtr>
     __host__ void maxVal(const SrcPtr& src, ResType* result, const MaskPtr& mask, int rows, int cols, cudaStream_t stream)
     {
         typedef typename PtrTraits<SrcPtr>::value_type src_type;
-        const int cn = VecTraits<src_type>::cn;
-        typedef typename MakeVec<ResType, cn>::type work_type;
 
-        glob_reduce<MinMaxReductor<maxop<work_type>, src_type, work_type>, Policy>(src, result, mask, rows, cols, stream);
+        reduce<MinMaxReductor<maxop<ResType>, src_type, ResType>, Policy>(src, result, mask, rows, cols, stream);
     }
 
     template <class Policy, class SrcPtr, typename ResType, class MaskPtr>
     __host__ void minMaxVal(const SrcPtr& src, ResType* result, const MaskPtr& mask, int rows, int cols, cudaStream_t stream)
     {
         typedef typename PtrTraits<SrcPtr>::value_type src_type;
-        const int cn = VecTraits<src_type>::cn;
-        typedef typename MakeVec<ResType, cn>::type work_type;
 
-        glob_reduce<MinMaxReductor<both, src_type, work_type>, Policy>(src, result, mask, rows, cols, stream);
+        reduce<MinMaxReductor<both, src_type, ResType>, Policy>(src, result, mask, rows, cols, stream);
     }
 }
 
diff --git a/modules/cudev/include/opencv2/cudev/grid/detail/reduce_to_column.hpp b/modules/cudev/include/opencv2/cudev/grid/detail/reduce_to_column.hpp
index b257e75039..c4852949e6 100644
--- a/modules/cudev/include/opencv2/cudev/grid/detail/reduce_to_column.hpp
+++ b/modules/cudev/include/opencv2/cudev/grid/detail/reduce_to_column.hpp
@@ -54,12 +54,52 @@ namespace cv { namespace cudev {
 
 namespace grid_reduce_to_vec_detail
 {
+    template <int BLOCK_SIZE, typename work_type, typename work_elem_type, class Reductor, int cn> struct Reduce;
+
+    template <int BLOCK_SIZE, typename work_type, typename work_elem_type, class Reductor> struct Reduce<BLOCK_SIZE, work_type, work_elem_type, Reductor, 1>
+    {
+        __device__ __forceinline__ static void call(work_elem_type smem[1][BLOCK_SIZE], work_type& myVal)
+        {
+            typename Reductor::template rebind<work_elem_type>::other op;
+            blockReduce<BLOCK_SIZE>(smem[0], myVal, threadIdx.x, op);
+        }
+    };
+
+    template <int BLOCK_SIZE, typename work_type, typename work_elem_type, class Reductor> struct Reduce<BLOCK_SIZE, work_type, work_elem_type, Reductor, 2>
+    {
+        __device__ __forceinline__ static void call(work_elem_type smem[2][BLOCK_SIZE], work_type& myVal)
+        {
+            typename Reductor::template rebind<work_elem_type>::other op;
+            blockReduce<BLOCK_SIZE>(smem_tuple(smem[0], smem[1]), tie(myVal.x, myVal.y), threadIdx.x, make_tuple(op, op));
+        }
+    };
+
+    template <int BLOCK_SIZE, typename work_type, typename work_elem_type, class Reductor> struct Reduce<BLOCK_SIZE, work_type, work_elem_type, Reductor, 3>
+    {
+        __device__ __forceinline__ static void call(work_elem_type smem[3][BLOCK_SIZE], work_type& myVal)
+        {
+            typename Reductor::template rebind<work_elem_type>::other op;
+            blockReduce<BLOCK_SIZE>(smem_tuple(smem[0], smem[1], smem[2]), tie(myVal.x, myVal.y, myVal.z), threadIdx.x, make_tuple(op, op, op));
+        }
+    };
+
+    template <int BLOCK_SIZE, typename work_type, typename work_elem_type, class Reductor> struct Reduce<BLOCK_SIZE, work_type, work_elem_type, Reductor, 4>
+    {
+        __device__ __forceinline__ static void call(work_elem_type smem[4][BLOCK_SIZE], work_type& myVal)
+        {
+            typename Reductor::template rebind<work_elem_type>::other op;
+            blockReduce<BLOCK_SIZE>(smem_tuple(smem[0], smem[1], smem[2], smem[3]), tie(myVal.x, myVal.y, myVal.z, myVal.w), threadIdx.x, make_tuple(op, op, op, op));
+        }
+    };
+
     template <class Reductor, int BLOCK_SIZE, class SrcPtr, typename ResType, class MaskPtr>
     __global__ void reduceToColumn(const SrcPtr src, ResType* dst, const MaskPtr mask, const int cols)
     {
         typedef typename Reductor::work_type work_type;
+        typedef typename VecTraits<work_type>::elem_type work_elem_type;
+        const int cn = VecTraits<work_type>::cn;
 
-        __shared__ work_type smem[BLOCK_SIZE];
+        __shared__ work_elem_type smem[cn][BLOCK_SIZE];
 
         const int y = blockIdx.x;
 
@@ -75,7 +115,7 @@ namespace grid_reduce_to_vec_detail
             }
         }
 
-        blockReduce<BLOCK_SIZE>(smem, myVal, threadIdx.x, op);
+        Reduce<BLOCK_SIZE, work_type, work_elem_type, Reductor, cn>::call(smem, myVal);
 
         if (threadIdx.x == 0)
             dst[y] = saturate_cast<ResType>(Reductor::result(myVal, cols));
diff --git a/modules/cudev/include/opencv2/cudev/grid/detail/transform.hpp b/modules/cudev/include/opencv2/cudev/grid/detail/transform.hpp
index 0e0df0794a..d13585398b 100644
--- a/modules/cudev/include/opencv2/cudev/grid/detail/transform.hpp
+++ b/modules/cudev/include/opencv2/cudev/grid/detail/transform.hpp
@@ -217,7 +217,7 @@ namespace grid_transform_detail
     }
 
     template <int SHIFT, typename SrcType1, typename SrcType2, typename DstType, class BinOp, class MaskPtr>
-    __global__ void transformSmart(const GlobPtr<SrcType1> src1_, const GlobPtr<SrcType2> src2_, PtrStep<DstType> dst_, const BinOp op, const MaskPtr mask, const int rows, const int cols)
+    __global__ void transformSmart(const GlobPtr<SrcType1> src1_, const GlobPtr<SrcType2> src2_, GlobPtr<DstType> dst_, const BinOp op, const MaskPtr mask, const int rows, const int cols)
     {
         typedef typename MakeVec<SrcType1, SHIFT>::type read_type1;
         typedef typename MakeVec<SrcType2, SHIFT>::type read_type2;
@@ -345,25 +345,25 @@ namespace grid_transform_detail
     };
 
     template <class Policy, class SrcPtr, typename DstType, class UnOp, class MaskPtr>
-    __host__ void transform(const SrcPtr& src, const GlobPtr<DstType>& dst, const UnOp& op, const MaskPtr& mask, int rows, int cols, cudaStream_t stream)
+    __host__ void transform_unary(const SrcPtr& src, const GlobPtr<DstType>& dst, const UnOp& op, const MaskPtr& mask, int rows, int cols, cudaStream_t stream)
     {
         TransformDispatcher<false, Policy>::call(src, dst, op, mask, rows, cols, stream);
     }
 
     template <class Policy, class SrcPtr1, class SrcPtr2, typename DstType, class BinOp, class MaskPtr>
-    __host__ void transform(const SrcPtr1& src1, const SrcPtr2& src2, const GlobPtr<DstType>& dst, const BinOp& op, const MaskPtr& mask, int rows, int cols, cudaStream_t stream)
+    __host__ void transform_binary(const SrcPtr1& src1, const SrcPtr2& src2, const GlobPtr<DstType>& dst, const BinOp& op, const MaskPtr& mask, int rows, int cols, cudaStream_t stream)
     {
         TransformDispatcher<false, Policy>::call(src1, src2, dst, op, mask, rows, cols, stream);
     }
 
     template <class Policy, typename SrcType, typename DstType, class UnOp, class MaskPtr>
-    __host__ void transform(const GlobPtr<SrcType>& src, const GlobPtr<DstType>& dst, const UnOp& op, const MaskPtr& mask, int rows, int cols, cudaStream_t stream)
+    __host__ void transform_unary(const GlobPtr<SrcType>& src, const GlobPtr<DstType>& dst, const UnOp& op, const MaskPtr& mask, int rows, int cols, cudaStream_t stream)
     {
         TransformDispatcher<VecTraits<SrcType>::cn == 1 && VecTraits<DstType>::cn == 1 && Policy::shift != 1, Policy>::call(src, dst, op, mask, rows, cols, stream);
     }
 
     template <class Policy, typename SrcType1, typename SrcType2, typename DstType, class BinOp, class MaskPtr>
-    __host__ void transform(const GlobPtr<SrcType1>& src1, const GlobPtr<SrcType2>& src2, const GlobPtr<DstType>& dst, const BinOp& op, const MaskPtr& mask, int rows, int cols, cudaStream_t stream)
+    __host__ void transform_binary(const GlobPtr<SrcType1>& src1, const GlobPtr<SrcType2>& src2, const GlobPtr<DstType>& dst, const BinOp& op, const MaskPtr& mask, int rows, int cols, cudaStream_t stream)
     {
         TransformDispatcher<VecTraits<SrcType1>::cn == 1 && VecTraits<SrcType2>::cn == 1 && VecTraits<DstType>::cn == 1 && Policy::shift != 1, Policy>::call(src1, src2, dst, op, mask, rows, cols, stream);
     }
diff --git a/modules/cudev/include/opencv2/cudev/grid/detail/transpose.hpp b/modules/cudev/include/opencv2/cudev/grid/detail/transpose.hpp
index 3854a17402..83ee96b790 100644
--- a/modules/cudev/include/opencv2/cudev/grid/detail/transpose.hpp
+++ b/modules/cudev/include/opencv2/cudev/grid/detail/transpose.hpp
@@ -55,15 +55,12 @@ namespace cv { namespace cudev {
 
 namespace transpose_detail
 {
-    const int TRANSPOSE_TILE_DIM   = 16;
-    const int TRANSPOSE_BLOCK_ROWS = 16;
-
-    template <class SrcPtr, typename DstType>
+    template <int TILE_DIM, int BLOCK_DIM_Y, class SrcPtr, typename DstType>
     __global__ void transpose(const SrcPtr src, GlobPtr<DstType> dst, const int rows, const int cols)
     {
         typedef typename PtrTraits<SrcPtr>::value_type src_type;
 
-        __shared__ src_type tile[TRANSPOSE_TILE_DIM][TRANSPOSE_TILE_DIM + 1];
+        __shared__ src_type tile[TILE_DIM][TILE_DIM + 1];
 
         int blockIdx_x, blockIdx_y;
 
@@ -80,12 +77,12 @@ namespace transpose_detail
             blockIdx_x = ((bid / gridDim.y) + blockIdx_y) % gridDim.x;
         }
 
-        int xIndex = blockIdx_x * TRANSPOSE_TILE_DIM + threadIdx.x;
-        int yIndex = blockIdx_y * TRANSPOSE_TILE_DIM + threadIdx.y;
+        int xIndex = blockIdx_x * TILE_DIM + threadIdx.x;
+        int yIndex = blockIdx_y * TILE_DIM + threadIdx.y;
 
         if (xIndex < cols)
         {
-            for (int i = 0; i < TRANSPOSE_TILE_DIM; i += TRANSPOSE_BLOCK_ROWS)
+            for (int i = 0; i < TILE_DIM; i += BLOCK_DIM_Y)
             {
                 if (yIndex + i < rows)
                 {
@@ -96,12 +93,12 @@ namespace transpose_detail
 
         __syncthreads();
 
-        xIndex = blockIdx_y * TRANSPOSE_TILE_DIM + threadIdx.x;
-        yIndex = blockIdx_x * TRANSPOSE_TILE_DIM + threadIdx.y;
+        xIndex = blockIdx_y * TILE_DIM + threadIdx.x;
+        yIndex = blockIdx_x * TILE_DIM + threadIdx.y;
 
         if (xIndex < rows)
         {
-            for (int i = 0; i < TRANSPOSE_TILE_DIM; i += TRANSPOSE_BLOCK_ROWS)
+            for (int i = 0; i < TILE_DIM; i += BLOCK_DIM_Y)
             {
                 if (yIndex + i < cols)
                 {
@@ -111,13 +108,13 @@ namespace transpose_detail
         }
     }
 
-    template <class SrcPtr, typename DstType>
+    template <class Policy, class SrcPtr, typename DstType>
     __host__ void transpose(const SrcPtr& src, const GlobPtr<DstType>& dst, int rows, int cols, cudaStream_t stream)
     {
-        const dim3 block(TRANSPOSE_TILE_DIM, TRANSPOSE_TILE_DIM);
+        const dim3 block(Policy::tile_dim, Policy::block_dim_y);
         const dim3 grid(divUp(cols, block.x), divUp(rows, block.y));
 
-        transpose<<<grid, block, 0, stream>>>(src, dst, rows, cols);
+        transpose<Policy::tile_dim, Policy::block_dim_y><<<grid, block, 0, stream>>>(src, dst, rows, cols);
         CV_CUDEV_SAFE_CALL( cudaGetLastError() );
 
         if (stream == 0)
diff --git a/modules/cudev/include/opencv2/cudev/grid/glob_reduce.hpp b/modules/cudev/include/opencv2/cudev/grid/reduce.hpp
similarity index 62%
rename from modules/cudev/include/opencv2/cudev/grid/glob_reduce.hpp
rename to modules/cudev/include/opencv2/cudev/grid/reduce.hpp
index 8d3e0826d1..3861ae2281 100644
--- a/modules/cudev/include/opencv2/cudev/grid/glob_reduce.hpp
+++ b/modules/cudev/include/opencv2/cudev/grid/reduce.hpp
@@ -43,8 +43,8 @@
 
 #pragma once
 
-#ifndef __OPENCV_CUDEV_GRID_GLOB_REDUCE_HPP__
-#define __OPENCV_CUDEV_GRID_GLOB_REDUCE_HPP__
+#ifndef __OPENCV_CUDEV_GRID_REDUCE_HPP__
+#define __OPENCV_CUDEV_GRID_REDUCE_HPP__
 
 #include <limits>
 #include "../common.hpp"
@@ -52,13 +52,18 @@
 #include "../ptr2d/gpumat.hpp"
 #include "../ptr2d/mask.hpp"
 #include "../ptr2d/transform.hpp"
-#include "detail/glob_reduce.hpp"
+#include "detail/reduce.hpp"
+#include "detail/minmaxloc.hpp"
 
 namespace cv { namespace cudev {
 
 template <class Policy, class SrcPtr, typename ResType, class MaskPtr>
 __host__ void gridCalcSum_(const SrcPtr& src, GpuMat_<ResType>& dst, const MaskPtr& mask, Stream& stream = Stream::Null())
 {
+    typedef typename PtrTraits<SrcPtr>::value_type src_type;
+
+    CV_StaticAssert( unsigned(VecTraits<src_type>::cn) == unsigned(VecTraits<ResType>::cn), "" );
+
     dst.create(1, 1);
     dst.setTo(0, stream);
 
@@ -67,27 +72,31 @@ __host__ void gridCalcSum_(const SrcPtr& src, GpuMat_<ResType>& dst, const MaskP
 
     CV_Assert( getRows(mask) == rows && getCols(mask) == cols );
 
-    grid_glob_reduce_detail::sum<Policy>(shrinkPtr(src),
-                                         dst[0],
-                                         shrinkPtr(mask),
-                                         rows, cols,
-                                         StreamAccessor::getStream(stream));
+    grid_reduce_detail::sum<Policy>(shrinkPtr(src),
+                                    dst[0],
+                                    shrinkPtr(mask),
+                                    rows, cols,
+                                    StreamAccessor::getStream(stream));
 }
 
 template <class Policy, class SrcPtr, typename ResType>
 __host__ void gridCalcSum_(const SrcPtr& src, GpuMat_<ResType>& dst, Stream& stream = Stream::Null())
 {
+    typedef typename PtrTraits<SrcPtr>::value_type src_type;
+
+    CV_StaticAssert( unsigned(VecTraits<src_type>::cn) == unsigned(VecTraits<ResType>::cn), "" );
+
     dst.create(1, 1);
     dst.setTo(0, stream);
 
     const int rows = getRows(src);
     const int cols = getCols(src);
 
-    grid_glob_reduce_detail::sum<Policy>(shrinkPtr(src),
-                                         dst[0],
-                                         WithOutMask(),
-                                         rows, cols,
-                                         StreamAccessor::getStream(stream));
+    grid_reduce_detail::sum<Policy>(shrinkPtr(src),
+                                    dst[0],
+                                    WithOutMask(),
+                                    rows, cols,
+                                    StreamAccessor::getStream(stream));
 }
 
 template <class Policy, class SrcPtr, typename ResType, class MaskPtr>
@@ -101,11 +110,11 @@ __host__ void gridFindMinVal_(const SrcPtr& src, GpuMat_<ResType>& dst, const Ma
 
     CV_Assert( getRows(mask) == rows && getCols(mask) == cols );
 
-    grid_glob_reduce_detail::minVal<Policy>(shrinkPtr(src),
-                                            dst[0],
-                                            shrinkPtr(mask),
-                                            rows, cols,
-                                            StreamAccessor::getStream(stream));
+    grid_reduce_detail::minVal<Policy>(shrinkPtr(src),
+                                       dst[0],
+                                       shrinkPtr(mask),
+                                       rows, cols,
+                                       StreamAccessor::getStream(stream));
 }
 
 template <class Policy, class SrcPtr, typename ResType>
@@ -117,11 +126,11 @@ __host__ void gridFindMinVal_(const SrcPtr& src, GpuMat_<ResType>& dst, Stream&
     const int rows = getRows(src);
     const int cols = getCols(src);
 
-    grid_glob_reduce_detail::minVal<Policy>(shrinkPtr(src),
-                                            dst[0],
-                                            WithOutMask(),
-                                            rows, cols,
-                                            StreamAccessor::getStream(stream));
+    grid_reduce_detail::minVal<Policy>(shrinkPtr(src),
+                                       dst[0],
+                                       WithOutMask(),
+                                       rows, cols,
+                                       StreamAccessor::getStream(stream));
 }
 
 template <class Policy, class SrcPtr, typename ResType, class MaskPtr>
@@ -135,11 +144,11 @@ __host__ void gridFindMaxVal_(const SrcPtr& src, GpuMat_<ResType>& dst, const Ma
 
     CV_Assert( getRows(mask) == rows && getCols(mask) == cols );
 
-    grid_glob_reduce_detail::maxVal<Policy>(shrinkPtr(src),
-                                            dst[0],
-                                            shrinkPtr(mask),
-                                            rows, cols,
-                                            StreamAccessor::getStream(stream));
+    grid_reduce_detail::maxVal<Policy>(shrinkPtr(src),
+                                       dst[0],
+                                       shrinkPtr(mask),
+                                       rows, cols,
+                                       StreamAccessor::getStream(stream));
 }
 
 template <class Policy, class SrcPtr, typename ResType>
@@ -151,11 +160,11 @@ __host__ void gridFindMaxVal_(const SrcPtr& src, GpuMat_<ResType>& dst, Stream&
     const int rows = getRows(src);
     const int cols = getCols(src);
 
-    grid_glob_reduce_detail::maxVal<Policy>(shrinkPtr(src),
-                                            dst[0],
-                                            WithOutMask(),
-                                            rows, cols,
-                                            StreamAccessor::getStream(stream));
+    grid_reduce_detail::maxVal<Policy>(shrinkPtr(src),
+                                       dst[0],
+                                       WithOutMask(),
+                                       rows, cols,
+                                       StreamAccessor::getStream(stream));
 }
 
 template <class Policy, class SrcPtr, typename ResType, class MaskPtr>
@@ -170,11 +179,11 @@ __host__ void gridFindMinMaxVal_(const SrcPtr& src, GpuMat_<ResType>& dst, const
 
     CV_Assert( getRows(mask) == rows && getCols(mask) == cols );
 
-    grid_glob_reduce_detail::minMaxVal<Policy>(shrinkPtr(src),
-                                               dst[0],
-                                               shrinkPtr(mask),
-                                               rows, cols,
-                                               StreamAccessor::getStream(stream));
+    grid_reduce_detail::minMaxVal<Policy>(shrinkPtr(src),
+                                          dst[0],
+                                          shrinkPtr(mask),
+                                          rows, cols,
+                                          StreamAccessor::getStream(stream));
 }
 
 template <class Policy, class SrcPtr, typename ResType>
@@ -187,11 +196,51 @@ __host__ void gridFindMinMaxVal_(const SrcPtr& src, GpuMat_<ResType>& dst, Strea
     const int rows = getRows(src);
     const int cols = getCols(src);
 
-    grid_glob_reduce_detail::minMaxVal<Policy>(shrinkPtr(src),
-                                               dst[0],
-                                               WithOutMask(),
-                                               rows, cols,
-                                               StreamAccessor::getStream(stream));
+    grid_reduce_detail::minMaxVal<Policy>(shrinkPtr(src),
+                                          dst[0],
+                                          WithOutMask(),
+                                          rows, cols,
+                                          StreamAccessor::getStream(stream));
+}
+
+template <class Policy, class SrcPtr, typename ResType, class MaskPtr>
+__host__ void gridMinMaxLoc_(const SrcPtr& src, GpuMat_<ResType>& valBuf, GpuMat_<int>& locBuf, const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    CV_Assert( getRows(mask) == rows && getCols(mask) == cols );
+
+    dim3 grid, block;
+    grid_minmaxloc_detail::getLaunchCfg<Policy>(rows, cols, block, grid);
+
+    valBuf.create(2, grid.x * grid.y);
+    locBuf.create(2, grid.x * grid.y);
+
+    grid_minmaxloc_detail::minMaxLoc<Policy>(shrinkPtr(src),
+                                             valBuf[0], valBuf[1], locBuf[0], locBuf[1],
+                                             shrinkPtr(mask),
+                                             rows, cols,
+                                             StreamAccessor::getStream(stream));
+}
+
+template <class Policy, class SrcPtr, typename ResType>
+__host__ void gridMinMaxLoc_(const SrcPtr& src, GpuMat_<ResType>& valBuf, GpuMat_<int>& locBuf, Stream& stream = Stream::Null())
+{
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    dim3 grid, block;
+    grid_minmaxloc_detail::getLaunchCfg<Policy>(rows, cols, block, grid);
+
+    valBuf.create(2, grid.x * grid.y);
+    locBuf.create(2, grid.x * grid.y);
+
+    grid_minmaxloc_detail::minMaxLoc<Policy>(shrinkPtr(src),
+                                             valBuf[0], valBuf[1], locBuf[0], locBuf[1],
+                                             WithOutMask(),
+                                             rows, cols,
+                                             StreamAccessor::getStream(stream));
 }
 
 template <class Policy, class SrcPtr, typename ResType, class MaskPtr>
@@ -209,11 +258,11 @@ __host__ void gridCountNonZero_(const SrcPtr& src, GpuMat_<ResType>& dst, const
     not_equal_to<src_type> ne_op;
     const src_type zero = VecTraits<src_type>::all(0);
 
-    grid_glob_reduce_detail::sum<Policy>(shrinkPtr(transformPtr(src, bind2nd(ne_op, zero))),
-                                         dst[0],
-                                         shrinkPtr(mask),
-                                         rows, cols,
-                                         StreamAccessor::getStream(stream));
+    grid_reduce_detail::sum<Policy>(shrinkPtr(transformPtr(src, bind2nd(ne_op, zero))),
+                                    dst[0],
+                                    shrinkPtr(mask),
+                                    rows, cols,
+                                    StreamAccessor::getStream(stream));
 }
 
 template <class Policy, class SrcPtr, typename ResType>
@@ -229,11 +278,11 @@ __host__ void gridCountNonZero_(const SrcPtr& src, GpuMat_<ResType>& dst, Stream
     not_equal_to<src_type> ne_op;
     const src_type zero = VecTraits<src_type>::all(0);
 
-    grid_glob_reduce_detail::sum<Policy>(shrinkPtr(transformPtr(src, bind2nd(ne_op, zero))),
-                                         dst[0],
-                                         WithOutMask(),
-                                         rows, cols,
-                                         StreamAccessor::getStream(stream));
+    grid_reduce_detail::sum<Policy>(shrinkPtr(transformPtr(src, bind2nd(ne_op, zero))),
+                                    dst[0],
+                                    WithOutMask(),
+                                    rows, cols,
+                                    StreamAccessor::getStream(stream));
 }
 
 // default policy
@@ -297,6 +346,18 @@ __host__ void gridFindMinMaxVal(const SrcPtr& src, GpuMat_<ResType>& dst, Stream
     gridFindMinMaxVal_<DefaultGlobReducePolicy>(src, dst, stream);
 }
 
+template <class SrcPtr, typename ResType, class MaskPtr>
+__host__ void gridMinMaxLoc(const SrcPtr& src, GpuMat_<ResType>& valBuf, GpuMat_<int>& locBuf, const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    gridMinMaxLoc_<DefaultGlobReducePolicy>(src, valBuf, locBuf, mask, stream);
+}
+
+template <class SrcPtr, typename ResType>
+__host__ void gridMinMaxLoc(const SrcPtr& src, GpuMat_<ResType>& valBuf, GpuMat_<int>& locBuf, Stream& stream = Stream::Null())
+{
+    gridMinMaxLoc_<DefaultGlobReducePolicy>(src, valBuf, locBuf, stream);
+}
+
 template <class SrcPtr, typename ResType, class MaskPtr>
 __host__ void gridCountNonZero(const SrcPtr& src, GpuMat_<ResType>& dst, const MaskPtr& mask, Stream& stream = Stream::Null())
 {
diff --git a/modules/cudev/include/opencv2/cudev/grid/reduce_to_vec.hpp b/modules/cudev/include/opencv2/cudev/grid/reduce_to_vec.hpp
index f9e3512428..361d40d1c8 100644
--- a/modules/cudev/include/opencv2/cudev/grid/reduce_to_vec.hpp
+++ b/modules/cudev/include/opencv2/cudev/grid/reduce_to_vec.hpp
@@ -49,6 +49,7 @@
 #include "../common.hpp"
 #include "../util/vec_traits.hpp"
 #include "../util/limits.hpp"
+#include "../util/saturate_cast.hpp"
 #include "../ptr2d/traits.hpp"
 #include "../ptr2d/gpumat.hpp"
 #include "../ptr2d/mask.hpp"
@@ -62,6 +63,11 @@ template <typename T> struct Sum : plus<T>
 {
     typedef T work_type;
 
+    template <typename U> struct rebind
+    {
+        typedef Sum<U> other;
+    };
+
     __device__ __forceinline__ static T initialValue()
     {
         return VecTraits<T>::all(0);
@@ -77,14 +83,19 @@ template <typename T> struct Avg : plus<T>
 {
     typedef T work_type;
 
+    template <typename U> struct rebind
+    {
+        typedef Avg<U> other;
+    };
+
     __device__ __forceinline__ static T initialValue()
     {
         return VecTraits<T>::all(0);
     }
 
-    __device__ __forceinline__ static T result(T r, int sz)
+    __device__ __forceinline__ static T result(T r, float sz)
     {
-        return r / sz;
+        return saturate_cast<T>(r / sz);
     }
 };
 
@@ -92,6 +103,11 @@ template <typename T> struct Min : minimum<T>
 {
     typedef T work_type;
 
+    template <typename U> struct rebind
+    {
+        typedef Min<U> other;
+    };
+
     __device__ __forceinline__ static T initialValue()
     {
         return VecTraits<T>::all(numeric_limits<typename VecTraits<T>::elem_type>::max());
@@ -107,6 +123,11 @@ template <typename T> struct Max : maximum<T>
 {
     typedef T work_type;
 
+    template <typename U> struct rebind
+    {
+        typedef Max<U> other;
+    };
+
     __device__ __forceinline__ static T initialValue()
     {
         return VecTraits<T>::all(-numeric_limits<typename VecTraits<T>::elem_type>::max());
@@ -158,7 +179,7 @@ __host__ void gridReduceToColumn_(const SrcPtr& src, GpuMat_<ResType>& dst, cons
 
     CV_Assert( getRows(mask) == rows && getCols(mask) == cols );
 
-    createContinuous(rows, 1, DataType<ResType>::type, dst);
+    dst.create(1, rows);
 
     grid_reduce_to_vec_detail::reduceToColumn<Reductor, Policy>(shrinkPtr(src),
                                                                 dst[0],
@@ -173,7 +194,7 @@ __host__ void gridReduceToColumn_(const SrcPtr& src, GpuMat_<ResType>& dst, Stre
     const int rows = getRows(src);
     const int cols = getCols(src);
 
-    createContinuous(rows, 1, DataType<ResType>::type, dst);
+    dst.create(1, rows);
 
     grid_reduce_to_vec_detail::reduceToColumn<Reductor, Policy>(shrinkPtr(src),
                                                                 dst[0],
diff --git a/modules/cudev/include/opencv2/cudev/grid/split_merge.hpp b/modules/cudev/include/opencv2/cudev/grid/split_merge.hpp
index 78584aad84..ed7e8ee605 100644
--- a/modules/cudev/include/opencv2/cudev/grid/split_merge.hpp
+++ b/modules/cudev/include/opencv2/cudev/grid/split_merge.hpp
@@ -51,6 +51,7 @@
 #include "../util/vec_traits.hpp"
 #include "../ptr2d/traits.hpp"
 #include "../ptr2d/gpumat.hpp"
+#include "../ptr2d/glob.hpp"
 #include "../ptr2d/mask.hpp"
 #include "detail/split_merge.hpp"
 
@@ -75,6 +76,24 @@ __host__ void gridMerge_(const SrcPtrTuple& src, GpuMat_<DstType>& dst, const Ma
                                                                               StreamAccessor::getStream(stream));
 }
 
+template <class Policy, class SrcPtrTuple, typename DstType, class MaskPtr>
+__host__ void gridMerge_(const SrcPtrTuple& src, const GlobPtrSz<DstType>& dst, const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    CV_StaticAssert( VecTraits<DstType>::cn == tuple_size<SrcPtrTuple>::value, "" );
+
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    CV_Assert( getRows(dst) == rows && getCols(dst) == cols );
+    CV_Assert( getRows(mask) == rows && getCols(mask) == cols );
+
+    grid_split_merge_detail::MergeImpl<VecTraits<DstType>::cn, Policy>::merge(shrinkPtr(src),
+                                                                              shrinkPtr(dst),
+                                                                              shrinkPtr(mask),
+                                                                              rows, cols,
+                                                                              StreamAccessor::getStream(stream));
+}
+
 template <class Policy, class SrcPtrTuple, typename DstType>
 __host__ void gridMerge_(const SrcPtrTuple& src, GpuMat_<DstType>& dst, Stream& stream = Stream::Null())
 {
@@ -92,6 +111,23 @@ __host__ void gridMerge_(const SrcPtrTuple& src, GpuMat_<DstType>& dst, Stream&
                                                                               StreamAccessor::getStream(stream));
 }
 
+template <class Policy, class SrcPtrTuple, typename DstType>
+__host__ void gridMerge_(const SrcPtrTuple& src, const GlobPtrSz<DstType>& dst, Stream& stream = Stream::Null())
+{
+    CV_StaticAssert( VecTraits<DstType>::cn == tuple_size<SrcPtrTuple>::value, "" );
+
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    CV_Assert( getRows(dst) == rows && getCols(dst) == cols );
+
+    grid_split_merge_detail::MergeImpl<VecTraits<DstType>::cn, Policy>::merge(shrinkPtr(src),
+                                                                              shrinkPtr(dst),
+                                                                              WithOutMask(),
+                                                                              rows, cols,
+                                                                              StreamAccessor::getStream(stream));
+}
+
 template <class Policy, class SrcPtr, typename DstType, class MaskPtr>
 __host__ void gridSplit_(const SrcPtr& src, const tuple< GpuMat_<DstType>&, GpuMat_<DstType>& >& dst, const MaskPtr& mask, Stream& stream = Stream::Null())
 {
@@ -132,6 +168,25 @@ __host__ void gridSplit_(const SrcPtr& src, GpuMat_<DstType> (&dst)[2], const Ma
                                            StreamAccessor::getStream(stream));
 }
 
+template <class Policy, class SrcPtr, typename DstType, class MaskPtr>
+__host__ void gridSplit_(const SrcPtr& src, GlobPtrSz<DstType> (&dst)[2], const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    CV_StaticAssert( VecTraits<typename PtrTraits<SrcPtr>::value_type>::cn == 2, "" );
+
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    CV_Assert( getRows(dst[0]) == rows && getCols(dst[0]) == cols );
+    CV_Assert( getRows(dst[1]) == rows && getCols(dst[1]) == cols );
+    CV_Assert( getRows(mask) == rows && getCols(mask) == cols );
+
+    grid_split_merge_detail::split<Policy>(shrinkPtr(src),
+                                           shrinkPtr(dst[0]), shrinkPtr(dst[1]),
+                                           shrinkPtr(mask),
+                                           rows, cols,
+                                           StreamAccessor::getStream(stream));
+}
+
 template <class Policy, class SrcPtr, typename DstType>
 __host__ void gridSplit_(const SrcPtr& src, const tuple< GpuMat_<DstType>&, GpuMat_<DstType>& >& dst, Stream& stream = Stream::Null())
 {
@@ -168,6 +223,24 @@ __host__ void gridSplit_(const SrcPtr& src, GpuMat_<DstType> (&dst)[2], Stream&
                                            StreamAccessor::getStream(stream));
 }
 
+template <class Policy, class SrcPtr, typename DstType>
+__host__ void gridSplit_(const SrcPtr& src, GlobPtrSz<DstType> (&dst)[2], Stream& stream = Stream::Null())
+{
+    CV_StaticAssert( VecTraits<typename PtrTraits<SrcPtr>::value_type>::cn == 2, "" );
+
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    CV_Assert( getRows(dst[0]) == rows && getCols(dst[0]) == cols );
+    CV_Assert( getRows(dst[1]) == rows && getCols(dst[1]) == cols );
+
+    grid_split_merge_detail::split<Policy>(shrinkPtr(src),
+                                           shrinkPtr(dst[0]), shrinkPtr(dst[1]),
+                                           WithOutMask(),
+                                           rows, cols,
+                                           StreamAccessor::getStream(stream));
+}
+
 template <class Policy, class SrcPtr, typename DstType, class MaskPtr>
 __host__ void gridSplit_(const SrcPtr& src, const tuple< GpuMat_<DstType>&, GpuMat_<DstType>&, GpuMat_<DstType>& >& dst, const MaskPtr& mask, Stream& stream = Stream::Null())
 {
@@ -210,6 +283,26 @@ __host__ void gridSplit_(const SrcPtr& src, GpuMat_<DstType> (&dst)[3], const Ma
                                            StreamAccessor::getStream(stream));
 }
 
+template <class Policy, class SrcPtr, typename DstType, class MaskPtr>
+__host__ void gridSplit_(const SrcPtr& src, GlobPtrSz<DstType> (&dst)[3], const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    CV_StaticAssert( VecTraits<typename PtrTraits<SrcPtr>::value_type>::cn == 3, "" );
+
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    CV_Assert( getRows(dst[0]) == rows && getCols(dst[0]) == cols );
+    CV_Assert( getRows(dst[1]) == rows && getCols(dst[1]) == cols );
+    CV_Assert( getRows(dst[2]) == rows && getCols(dst[2]) == cols );
+    CV_Assert( getRows(mask) == rows && getCols(mask) == cols );
+
+    grid_split_merge_detail::split<Policy>(shrinkPtr(src),
+                                           shrinkPtr(dst[0]), shrinkPtr(dst[1]), shrinkPtr(dst[2]),
+                                           shrinkPtr(mask),
+                                           rows, cols,
+                                           StreamAccessor::getStream(stream));
+}
+
 template <class Policy, class SrcPtr, typename DstType>
 __host__ void gridSplit_(const SrcPtr& src, const tuple< GpuMat_<DstType>&, GpuMat_<DstType>&, GpuMat_<DstType>& >& dst, Stream& stream = Stream::Null())
 {
@@ -248,6 +341,25 @@ __host__ void gridSplit_(const SrcPtr& src, GpuMat_<DstType> (&dst)[3], Stream&
                                            StreamAccessor::getStream(stream));
 }
 
+template <class Policy, class SrcPtr, typename DstType>
+__host__ void gridSplit_(const SrcPtr& src, GlobPtrSz<DstType> (&dst)[3], Stream& stream = Stream::Null())
+{
+    CV_StaticAssert( VecTraits<typename PtrTraits<SrcPtr>::value_type>::cn == 3, "" );
+
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    CV_Assert( getRows(dst[0]) == rows && getCols(dst[0]) == cols );
+    CV_Assert( getRows(dst[1]) == rows && getCols(dst[1]) == cols );
+    CV_Assert( getRows(dst[2]) == rows && getCols(dst[2]) == cols );
+
+    grid_split_merge_detail::split<Policy>(shrinkPtr(src),
+                                           shrinkPtr(dst[0]), shrinkPtr(dst[1]), shrinkPtr(dst[2]),
+                                           WithOutMask(),
+                                           rows, cols,
+                                           StreamAccessor::getStream(stream));
+}
+
 template <class Policy, class SrcPtr, typename DstType, class MaskPtr>
 __host__ void gridSplit_(const SrcPtr& src, const tuple< GpuMat_<DstType>&, GpuMat_<DstType>&, GpuMat_<DstType>&, GpuMat_<DstType>& >& dst, const MaskPtr& mask, Stream& stream = Stream::Null())
 {
@@ -283,10 +395,31 @@ __host__ void gridSplit_(const SrcPtr& src, GpuMat_<DstType> (&dst)[4], const Ma
     dst[0].create(rows, cols);
     dst[1].create(rows, cols);
     dst[2].create(rows, cols);
-    dst[4].create(rows, cols);
+    dst[3].create(rows, cols);
 
     grid_split_merge_detail::split<Policy>(shrinkPtr(src),
-                                           shrinkPtr(dst[0]), shrinkPtr(dst[1]), shrinkPtr(dst[2]), shrinkPtr(dst[4]),
+                                           shrinkPtr(dst[0]), shrinkPtr(dst[1]), shrinkPtr(dst[2]), shrinkPtr(dst[3]),
+                                           shrinkPtr(mask),
+                                           rows, cols,
+                                           StreamAccessor::getStream(stream));
+}
+
+template <class Policy, class SrcPtr, typename DstType, class MaskPtr>
+__host__ void gridSplit_(const SrcPtr& src, GlobPtrSz<DstType> (&dst)[4], const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    CV_StaticAssert( VecTraits<typename PtrTraits<SrcPtr>::value_type>::cn == 4, "" );
+
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    CV_Assert( getRows(dst[0]) == rows && getCols(dst[0]) == cols );
+    CV_Assert( getRows(dst[1]) == rows && getCols(dst[1]) == cols );
+    CV_Assert( getRows(dst[2]) == rows && getCols(dst[2]) == cols );
+    CV_Assert( getRows(dst[3]) == rows && getCols(dst[3]) == cols );
+    CV_Assert( getRows(mask) == rows && getCols(mask) == cols );
+
+    grid_split_merge_detail::split<Policy>(shrinkPtr(src),
+                                           shrinkPtr(dst[0]), shrinkPtr(dst[1]), shrinkPtr(dst[2]), shrinkPtr(dst[3]),
                                            shrinkPtr(mask),
                                            rows, cols,
                                            StreamAccessor::getStream(stream));
@@ -323,10 +456,30 @@ __host__ void gridSplit_(const SrcPtr& src, GpuMat_<DstType> (&dst)[4], Stream&
     dst[0].create(rows, cols);
     dst[1].create(rows, cols);
     dst[2].create(rows, cols);
-    dst[4].create(rows, cols);
+    dst[3].create(rows, cols);
 
     grid_split_merge_detail::split<Policy>(shrinkPtr(src),
-                                           shrinkPtr(dst[0]), shrinkPtr(dst[1]), shrinkPtr(dst[2]), shrinkPtr(dst[4]),
+                                           shrinkPtr(dst[0]), shrinkPtr(dst[1]), shrinkPtr(dst[2]), shrinkPtr(dst[3]),
+                                           WithOutMask(),
+                                           rows, cols,
+                                           StreamAccessor::getStream(stream));
+}
+
+template <class Policy, class SrcPtr, typename DstType>
+__host__ void gridSplit_(const SrcPtr& src, GlobPtrSz<DstType> (&dst)[4], Stream& stream = Stream::Null())
+{
+    CV_StaticAssert( VecTraits<typename PtrTraits<SrcPtr>::value_type>::cn == 4, "" );
+
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    CV_Assert( getRows(dst[0]) == rows && getCols(dst[0]) == cols );
+    CV_Assert( getRows(dst[1]) == rows && getCols(dst[1]) == cols );
+    CV_Assert( getRows(dst[2]) == rows && getCols(dst[2]) == cols );
+    CV_Assert( getRows(dst[3]) == rows && getCols(dst[3]) == cols );
+
+    grid_split_merge_detail::split<Policy>(shrinkPtr(src),
+                                           shrinkPtr(dst[0]), shrinkPtr(dst[1]), shrinkPtr(dst[2]), shrinkPtr(dst[3]),
                                            WithOutMask(),
                                            rows, cols,
                                            StreamAccessor::getStream(stream));
@@ -348,12 +501,24 @@ __host__ void gridMerge(const SrcPtrTuple& src, GpuMat_<DstType>& dst, const Mas
     gridMerge_<DefaultSplitMergePolicy>(src, dst, mask, stream);
 }
 
+template <class SrcPtrTuple, typename DstType, class MaskPtr>
+__host__ void gridMerge(const SrcPtrTuple& src, const GlobPtrSz<DstType>& dst, const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    gridMerge_<DefaultSplitMergePolicy>(src, dst, mask, stream);
+}
+
 template <class SrcPtrTuple, typename DstType>
 __host__ void gridMerge(const SrcPtrTuple& src, GpuMat_<DstType>& dst, Stream& stream = Stream::Null())
 {
     gridMerge_<DefaultSplitMergePolicy>(src, dst, stream);
 }
 
+template <class SrcPtrTuple, typename DstType>
+__host__ void gridMerge(const SrcPtrTuple& src, const GlobPtrSz<DstType>& dst, Stream& stream = Stream::Null())
+{
+    gridMerge_<DefaultSplitMergePolicy>(src, dst, stream);
+}
+
 template <class SrcPtr, typename DstType, class MaskPtr>
 __host__ void gridSplit(const SrcPtr& src, const tuple< GpuMat_<DstType>&, GpuMat_<DstType>& >& dst, const MaskPtr& mask, Stream& stream = Stream::Null())
 {
@@ -396,12 +561,24 @@ __host__ void gridSplit(const SrcPtr& src, GpuMat_<DstType> (&dst)[COUNT], const
     gridSplit_<DefaultSplitMergePolicy>(src, dst, mask, stream);
 }
 
+template <class SrcPtr, typename DstType, int COUNT, class MaskPtr>
+__host__ void gridSplit(const SrcPtr& src, GlobPtrSz<DstType> (&dst)[COUNT], const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    gridSplit_<DefaultSplitMergePolicy>(src, dst, mask, stream);
+}
+
 template <class SrcPtr, typename DstType, int COUNT>
 __host__ void gridSplit(const SrcPtr& src, GpuMat_<DstType> (&dst)[COUNT], Stream& stream = Stream::Null())
 {
     gridSplit_<DefaultSplitMergePolicy>(src, dst, stream);
 }
 
+template <class SrcPtr, typename DstType, int COUNT>
+__host__ void gridSplit(const SrcPtr& src, GlobPtrSz<DstType> (&dst)[COUNT], Stream& stream = Stream::Null())
+{
+    gridSplit_<DefaultSplitMergePolicy>(src, dst, stream);
+}
+
 }}
 
 #endif
diff --git a/modules/cudev/include/opencv2/cudev/grid/transform.hpp b/modules/cudev/include/opencv2/cudev/grid/transform.hpp
index 0da5e2d8a4..62555ab5ae 100644
--- a/modules/cudev/include/opencv2/cudev/grid/transform.hpp
+++ b/modules/cudev/include/opencv2/cudev/grid/transform.hpp
@@ -58,7 +58,7 @@
 namespace cv { namespace cudev {
 
 template <class Policy, class SrcPtr, typename DstType, class UnOp, class MaskPtr>
-__host__ void gridTransform_(const SrcPtr& src, GpuMat_<DstType>& dst, const UnOp& op, const MaskPtr& mask, Stream& stream = Stream::Null())
+__host__ void gridTransformUnary_(const SrcPtr& src, GpuMat_<DstType>& dst, const UnOp& op, const MaskPtr& mask, Stream& stream = Stream::Null())
 {
     const int rows = getRows(src);
     const int cols = getCols(src);
@@ -67,11 +67,11 @@ __host__ void gridTransform_(const SrcPtr& src, GpuMat_<DstType>& dst, const UnO
 
     dst.create(rows, cols);
 
-    grid_transform_detail::transform<Policy>(shrinkPtr(src), shrinkPtr(dst), op, shrinkPtr(mask), rows, cols, StreamAccessor::getStream(stream));
+    grid_transform_detail::transform_unary<Policy>(shrinkPtr(src), shrinkPtr(dst), op, shrinkPtr(mask), rows, cols, StreamAccessor::getStream(stream));
 }
 
 template <class Policy, class SrcPtr, typename DstType, class UnOp, class MaskPtr>
-__host__ void gridTransform_(const SrcPtr& src, const GlobPtrSz<DstType>& dst, const UnOp& op, const MaskPtr& mask, Stream& stream = Stream::Null())
+__host__ void gridTransformUnary_(const SrcPtr& src, const GlobPtrSz<DstType>& dst, const UnOp& op, const MaskPtr& mask, Stream& stream = Stream::Null())
 {
     const int rows = getRows(src);
     const int cols = getCols(src);
@@ -79,33 +79,33 @@ __host__ void gridTransform_(const SrcPtr& src, const GlobPtrSz<DstType>& dst, c
     CV_Assert( getRows(dst) == rows && getCols(dst) == cols );
     CV_Assert( getRows(mask) == rows && getCols(mask) == cols );
 
-    grid_transform_detail::transform<Policy>(shrinkPtr(src), shrinkPtr(dst), op, shrinkPtr(mask), rows, cols, StreamAccessor::getStream(stream));
+    grid_transform_detail::transform_unary<Policy>(shrinkPtr(src), shrinkPtr(dst), op, shrinkPtr(mask), rows, cols, StreamAccessor::getStream(stream));
 }
 
 template <class Policy, class SrcPtr, typename DstType, class UnOp>
-__host__ void gridTransform_(const SrcPtr& src, GpuMat_<DstType>& dst, const UnOp& op, Stream& stream = Stream::Null())
+__host__ void gridTransformUnary_(const SrcPtr& src, GpuMat_<DstType>& dst, const UnOp& op, Stream& stream = Stream::Null())
 {
     const int rows = getRows(src);
     const int cols = getCols(src);
 
     dst.create(rows, cols);
 
-    grid_transform_detail::transform<Policy>(shrinkPtr(src), shrinkPtr(dst), op, WithOutMask(), rows, cols, StreamAccessor::getStream(stream));
+    grid_transform_detail::transform_unary<Policy>(shrinkPtr(src), shrinkPtr(dst), op, WithOutMask(), rows, cols, StreamAccessor::getStream(stream));
 }
 
 template <class Policy, class SrcPtr, typename DstType, class UnOp>
-__host__ void gridTransform_(const SrcPtr& src, const GlobPtrSz<DstType>& dst, const UnOp& op, Stream& stream = Stream::Null())
+__host__ void gridTransformUnary_(const SrcPtr& src, const GlobPtrSz<DstType>& dst, const UnOp& op, Stream& stream = Stream::Null())
 {
     const int rows = getRows(src);
     const int cols = getCols(src);
 
     CV_Assert( getRows(dst) == rows && getCols(dst) == cols );
 
-    grid_transform_detail::transform<Policy>(shrinkPtr(src), shrinkPtr(dst), op, WithOutMask(), rows, cols, StreamAccessor::getStream(stream));
+    grid_transform_detail::transform_unary<Policy>(shrinkPtr(src), shrinkPtr(dst), op, WithOutMask(), rows, cols, StreamAccessor::getStream(stream));
 }
 
 template <class Policy, class SrcPtr1, class SrcPtr2, typename DstType, class BinOp, class MaskPtr>
-__host__ void gridTransform_(const SrcPtr1& src1, const SrcPtr2& src2, GpuMat_<DstType>& dst, const BinOp& op, const MaskPtr& mask, Stream& stream = Stream::Null())
+__host__ void gridTransformBinary_(const SrcPtr1& src1, const SrcPtr2& src2, GpuMat_<DstType>& dst, const BinOp& op, const MaskPtr& mask, Stream& stream = Stream::Null())
 {
     const int rows = getRows(src1);
     const int cols = getCols(src1);
@@ -115,11 +115,11 @@ __host__ void gridTransform_(const SrcPtr1& src1, const SrcPtr2& src2, GpuMat_<D
 
     dst.create(rows, cols);
 
-    grid_transform_detail::transform<Policy>(shrinkPtr(src1), shrinkPtr(src2), shrinkPtr(dst), op, shrinkPtr(mask), rows, cols, StreamAccessor::getStream(stream));
+    grid_transform_detail::transform_binary<Policy>(shrinkPtr(src1), shrinkPtr(src2), shrinkPtr(dst), op, shrinkPtr(mask), rows, cols, StreamAccessor::getStream(stream));
 }
 
 template <class Policy, class SrcPtr1, class SrcPtr2, typename DstType, class BinOp, class MaskPtr>
-__host__ void gridTransform_(const SrcPtr1& src1, const SrcPtr2& src2, const GlobPtrSz<DstType>& dst, const BinOp& op, const MaskPtr& mask, Stream& stream = Stream::Null())
+__host__ void gridTransformBinary_(const SrcPtr1& src1, const SrcPtr2& src2, const GlobPtrSz<DstType>& dst, const BinOp& op, const MaskPtr& mask, Stream& stream = Stream::Null())
 {
     const int rows = getRows(src1);
     const int cols = getCols(src1);
@@ -128,11 +128,11 @@ __host__ void gridTransform_(const SrcPtr1& src1, const SrcPtr2& src2, const Glo
     CV_Assert( getRows(src2) == rows && getCols(src2) == cols );
     CV_Assert( getRows(mask) == rows && getCols(mask) == cols );
 
-    grid_transform_detail::transform<Policy>(shrinkPtr(src1), shrinkPtr(src2), shrinkPtr(dst), op, shrinkPtr(mask), rows, cols, StreamAccessor::getStream(stream));
+    grid_transform_detail::transform_binary<Policy>(shrinkPtr(src1), shrinkPtr(src2), shrinkPtr(dst), op, shrinkPtr(mask), rows, cols, StreamAccessor::getStream(stream));
 }
 
 template <class Policy, class SrcPtr1, class SrcPtr2, typename DstType, class BinOp>
-__host__ void gridTransform_(const SrcPtr1& src1, const SrcPtr2& src2, GpuMat_<DstType>& dst, const BinOp& op, Stream& stream = Stream::Null())
+__host__ void gridTransformBinary_(const SrcPtr1& src1, const SrcPtr2& src2, GpuMat_<DstType>& dst, const BinOp& op, Stream& stream = Stream::Null())
 {
     const int rows = getRows(src1);
     const int cols = getCols(src1);
@@ -141,11 +141,11 @@ __host__ void gridTransform_(const SrcPtr1& src1, const SrcPtr2& src2, GpuMat_<D
 
     dst.create(rows, cols);
 
-    grid_transform_detail::transform<Policy>(shrinkPtr(src1), shrinkPtr(src2), shrinkPtr(dst), op, WithOutMask(), rows, cols, StreamAccessor::getStream(stream));
+    grid_transform_detail::transform_binary<Policy>(shrinkPtr(src1), shrinkPtr(src2), shrinkPtr(dst), op, WithOutMask(), rows, cols, StreamAccessor::getStream(stream));
 }
 
 template <class Policy, class SrcPtr1, class SrcPtr2, typename DstType, class BinOp>
-__host__ void gridTransform_(const SrcPtr1& src1, const SrcPtr2& src2, GlobPtrSz<DstType>& dst, const BinOp& op, Stream& stream = Stream::Null())
+__host__ void gridTransformBinary_(const SrcPtr1& src1, const SrcPtr2& src2, const GlobPtrSz<DstType>& dst, const BinOp& op, Stream& stream = Stream::Null())
 {
     const int rows = getRows(src1);
     const int cols = getCols(src1);
@@ -153,11 +153,11 @@ __host__ void gridTransform_(const SrcPtr1& src1, const SrcPtr2& src2, GlobPtrSz
     CV_Assert( getRows(dst) == rows && getCols(dst) == cols );
     CV_Assert( getRows(src2) == rows && getCols(src2) == cols );
 
-    grid_transform_detail::transform<Policy>(shrinkPtr(src1), shrinkPtr(src2), shrinkPtr(dst), op, WithOutMask(), rows, cols, StreamAccessor::getStream(stream));
+    grid_transform_detail::transform_binary<Policy>(shrinkPtr(src1), shrinkPtr(src2), shrinkPtr(dst), op, WithOutMask(), rows, cols, StreamAccessor::getStream(stream));
 }
 
 template <class Policy, class SrcPtr, typename D0, typename D1, class OpTuple, class MaskPtr>
-__host__ void gridTransform_(const SrcPtr& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>& >& dst, const OpTuple& op, const MaskPtr& mask, Stream& stream = Stream::Null())
+__host__ void gridTransformTuple_(const SrcPtr& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>& >& dst, const OpTuple& op, const MaskPtr& mask, Stream& stream = Stream::Null())
 {
     CV_StaticAssert( tuple_size<OpTuple>::value == 2, "" );
 
@@ -178,7 +178,7 @@ __host__ void gridTransform_(const SrcPtr& src, const tuple< GpuMat_<D0>&, GpuMa
 }
 
 template <class Policy, class SrcPtr, typename D0, typename D1, class OpTuple, class MaskPtr>
-__host__ void gridTransform_(const SrcPtr& src, const tuple< GlobPtrSz<D0>, GlobPtrSz<D1> >& dst, const OpTuple& op, const MaskPtr& mask, Stream& stream = Stream::Null())
+__host__ void gridTransformTuple_(const SrcPtr& src, const tuple< GlobPtrSz<D0>, GlobPtrSz<D1> >& dst, const OpTuple& op, const MaskPtr& mask, Stream& stream = Stream::Null())
 {
     CV_StaticAssert( tuple_size<OpTuple>::value == 2, "" );
 
@@ -198,7 +198,7 @@ __host__ void gridTransform_(const SrcPtr& src, const tuple< GlobPtrSz<D0>, Glob
 }
 
 template <class Policy, class SrcPtr, typename D0, typename D1, class OpTuple>
-__host__ void gridTransform_(const SrcPtr& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>& >& dst, const OpTuple& op, Stream& stream = Stream::Null())
+__host__ void gridTransformTuple_(const SrcPtr& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>& >& dst, const OpTuple& op, Stream& stream = Stream::Null())
 {
     CV_StaticAssert( tuple_size<OpTuple>::value == 2, "" );
 
@@ -217,7 +217,7 @@ __host__ void gridTransform_(const SrcPtr& src, const tuple< GpuMat_<D0>&, GpuMa
 }
 
 template <class Policy, class SrcPtr, typename D0, typename D1, class OpTuple>
-__host__ void gridTransform_(const SrcPtr& src, const tuple< GlobPtrSz<D0>, GlobPtrSz<D1> >& dst, const OpTuple& op, Stream& stream = Stream::Null())
+__host__ void gridTransformTuple_(const SrcPtr& src, const tuple< GlobPtrSz<D0>, GlobPtrSz<D1> >& dst, const OpTuple& op, Stream& stream = Stream::Null())
 {
     CV_StaticAssert( tuple_size<OpTuple>::value == 2, "" );
 
@@ -236,7 +236,7 @@ __host__ void gridTransform_(const SrcPtr& src, const tuple< GlobPtrSz<D0>, Glob
 }
 
 template <class Policy, class SrcPtr, typename D0, typename D1, typename D2, class OpTuple, class MaskPtr>
-__host__ void gridTransform_(const SrcPtr& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>&, GpuMat_<D2>& >& dst, const OpTuple& op, const MaskPtr& mask, Stream& stream = Stream::Null())
+__host__ void gridTransformTuple_(const SrcPtr& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>&, GpuMat_<D2>& >& dst, const OpTuple& op, const MaskPtr& mask, Stream& stream = Stream::Null())
 {
     CV_StaticAssert( tuple_size<OpTuple>::value == 3, "" );
 
@@ -258,7 +258,7 @@ __host__ void gridTransform_(const SrcPtr& src, const tuple< GpuMat_<D0>&, GpuMa
 }
 
 template <class Policy, class SrcPtr, typename D0, typename D1, typename D2, class OpTuple, class MaskPtr>
-__host__ void gridTransform_(const SrcPtr& src, const tuple< GlobPtrSz<D0>, GlobPtrSz<D1>, GlobPtrSz<D2> >& dst, const OpTuple& op, const MaskPtr& mask, Stream& stream = Stream::Null())
+__host__ void gridTransformTuple_(const SrcPtr& src, const tuple< GlobPtrSz<D0>, GlobPtrSz<D1>, GlobPtrSz<D2> >& dst, const OpTuple& op, const MaskPtr& mask, Stream& stream = Stream::Null())
 {
     CV_StaticAssert( tuple_size<OpTuple>::value == 3, "" );
 
@@ -279,7 +279,7 @@ __host__ void gridTransform_(const SrcPtr& src, const tuple< GlobPtrSz<D0>, Glob
 }
 
 template <class Policy, class SrcPtr, typename D0, typename D1, typename D2, class OpTuple>
-__host__ void gridTransform_(const SrcPtr& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>&, GpuMat_<D2>& >& dst, const OpTuple& op, Stream& stream = Stream::Null())
+__host__ void gridTransformTuple_(const SrcPtr& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>&, GpuMat_<D2>& >& dst, const OpTuple& op, Stream& stream = Stream::Null())
 {
     CV_StaticAssert( tuple_size<OpTuple>::value == 3, "" );
 
@@ -299,7 +299,7 @@ __host__ void gridTransform_(const SrcPtr& src, const tuple< GpuMat_<D0>&, GpuMa
 }
 
 template <class Policy, class SrcPtr, typename D0, typename D1, typename D2, class OpTuple>
-__host__ void gridTransform_(const SrcPtr& src, const tuple< GlobPtrSz<D0>, GlobPtrSz<D1>, GlobPtrSz<D2> >& dst, const OpTuple& op, Stream& stream = Stream::Null())
+__host__ void gridTransformTuple_(const SrcPtr& src, const tuple< GlobPtrSz<D0>, GlobPtrSz<D1>, GlobPtrSz<D2> >& dst, const OpTuple& op, Stream& stream = Stream::Null())
 {
     CV_StaticAssert( tuple_size<OpTuple>::value == 3, "" );
 
@@ -319,7 +319,7 @@ __host__ void gridTransform_(const SrcPtr& src, const tuple< GlobPtrSz<D0>, Glob
 }
 
 template <class Policy, class SrcPtr, typename D0, typename D1, typename D2, typename D3, class OpTuple, class MaskPtr>
-__host__ void gridTransform_(const SrcPtr& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>&, GpuMat_<D2>&, GpuMat_<D3>& >& dst, const OpTuple& op, const MaskPtr& mask, Stream& stream = Stream::Null())
+__host__ void gridTransformTuple_(const SrcPtr& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>&, GpuMat_<D2>&, GpuMat_<D3>& >& dst, const OpTuple& op, const MaskPtr& mask, Stream& stream = Stream::Null())
 {
     CV_StaticAssert( tuple_size<OpTuple>::value == 4, "" );
 
@@ -342,7 +342,7 @@ __host__ void gridTransform_(const SrcPtr& src, const tuple< GpuMat_<D0>&, GpuMa
 }
 
 template <class Policy, class SrcPtr, typename D0, typename D1, typename D2, typename D3, class OpTuple, class MaskPtr>
-__host__ void gridTransform_(const SrcPtr& src, const tuple< GlobPtrSz<D0>, GlobPtrSz<D1>, GlobPtrSz<D2>, GlobPtrSz<D3> >& dst, const OpTuple& op, const MaskPtr& mask, Stream& stream = Stream::Null())
+__host__ void gridTransformTuple_(const SrcPtr& src, const tuple< GlobPtrSz<D0>, GlobPtrSz<D1>, GlobPtrSz<D2>, GlobPtrSz<D3> >& dst, const OpTuple& op, const MaskPtr& mask, Stream& stream = Stream::Null())
 {
     CV_StaticAssert( tuple_size<OpTuple>::value == 4, "" );
 
@@ -364,7 +364,7 @@ __host__ void gridTransform_(const SrcPtr& src, const tuple< GlobPtrSz<D0>, Glob
 }
 
 template <class Policy, class SrcPtr, typename D0, typename D1, typename D2, typename D3, class OpTuple>
-__host__ void gridTransform_(const SrcPtr& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>&, GpuMat_<D2>&, GpuMat_<D3>& >& dst, const OpTuple& op, Stream& stream = Stream::Null())
+__host__ void gridTransformTuple_(const SrcPtr& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>&, GpuMat_<D2>&, GpuMat_<D3>& >& dst, const OpTuple& op, Stream& stream = Stream::Null())
 {
     CV_StaticAssert( tuple_size<OpTuple>::value == 4, "" );
 
@@ -385,7 +385,7 @@ __host__ void gridTransform_(const SrcPtr& src, const tuple< GpuMat_<D0>&, GpuMa
 }
 
 template <class Policy, class SrcPtr, typename D0, typename D1, typename D2, typename D3, class OpTuple>
-__host__ void gridTransform_(const SrcPtr& src, const tuple< GlobPtrSz<D0>, GlobPtrSz<D1>, GlobPtrSz<D2>, GlobPtrSz<D3> >& dst, const OpTuple& op, Stream& stream = Stream::Null())
+__host__ void gridTransformTuple_(const SrcPtr& src, const tuple< GlobPtrSz<D0>, GlobPtrSz<D1>, GlobPtrSz<D2>, GlobPtrSz<D3> >& dst, const OpTuple& op, Stream& stream = Stream::Null())
 {
     CV_StaticAssert( tuple_size<OpTuple>::value == 4, "" );
 
@@ -417,123 +417,123 @@ struct DefaultTransformPolicy
 };
 
 template <class SrcPtr, typename DstType, class Op, class MaskPtr>
-__host__ void gridTransform(const SrcPtr& src, GpuMat_<DstType>& dst, const Op& op, const MaskPtr& mask, Stream& stream = Stream::Null())
+__host__ void gridTransformUnary(const SrcPtr& src, GpuMat_<DstType>& dst, const Op& op, const MaskPtr& mask, Stream& stream = Stream::Null())
 {
-    gridTransform_<DefaultTransformPolicy>(src, dst, op, mask, stream);
+    gridTransformUnary_<DefaultTransformPolicy>(src, dst, op, mask, stream);
 }
 
 template <class SrcPtr, typename DstType, class Op, class MaskPtr>
-__host__ void gridTransform(const SrcPtr& src, const GlobPtrSz<DstType>& dst, const Op& op, const MaskPtr& mask, Stream& stream = Stream::Null())
+__host__ void gridTransformUnary(const SrcPtr& src, const GlobPtrSz<DstType>& dst, const Op& op, const MaskPtr& mask, Stream& stream = Stream::Null())
 {
-    gridTransform_<DefaultTransformPolicy>(src, dst, op, mask, stream);
+    gridTransformUnary_<DefaultTransformPolicy>(src, dst, op, mask, stream);
 }
 
 template <class SrcPtr, typename DstType, class Op>
-__host__ void gridTransform(const SrcPtr& src, GpuMat_<DstType>& dst, const Op& op, Stream& stream = Stream::Null())
+__host__ void gridTransformUnary(const SrcPtr& src, GpuMat_<DstType>& dst, const Op& op, Stream& stream = Stream::Null())
 {
-    gridTransform_<DefaultTransformPolicy>(src, dst, op, stream);
+    gridTransformUnary_<DefaultTransformPolicy>(src, dst, op, stream);
 }
 
 template <class SrcPtr, typename DstType, class Op>
-__host__ void gridTransform(const SrcPtr& src, const GlobPtrSz<DstType>& dst, const Op& op, Stream& stream = Stream::Null())
+__host__ void gridTransformUnary(const SrcPtr& src, const GlobPtrSz<DstType>& dst, const Op& op, Stream& stream = Stream::Null())
 {
-    gridTransform_<DefaultTransformPolicy>(src, dst, op, stream);
+    gridTransformUnary_<DefaultTransformPolicy>(src, dst, op, stream);
 }
 
 template <class SrcPtr1, class SrcPtr2, typename DstType, class Op, class MaskPtr>
-__host__ void gridTransform(const SrcPtr1& src1, const SrcPtr1& src2, GpuMat_<DstType>& dst, const Op& op, const MaskPtr& mask, Stream& stream = Stream::Null())
+__host__ void gridTransformBinary(const SrcPtr1& src1, const SrcPtr2& src2, GpuMat_<DstType>& dst, const Op& op, const MaskPtr& mask, Stream& stream = Stream::Null())
 {
-    gridTransform_<DefaultTransformPolicy>(src1, src2, dst, op, mask, stream);
+    gridTransformBinary_<DefaultTransformPolicy>(src1, src2, dst, op, mask, stream);
 }
 
 template <class SrcPtr1, class SrcPtr2, typename DstType, class Op, class MaskPtr>
-__host__ void gridTransform(const SrcPtr1& src1, const SrcPtr1& src2, const GlobPtrSz<DstType>& dst, const Op& op, const MaskPtr& mask, Stream& stream = Stream::Null())
+__host__ void gridTransformBinary(const SrcPtr1& src1, const SrcPtr2& src2, const GlobPtrSz<DstType>& dst, const Op& op, const MaskPtr& mask, Stream& stream = Stream::Null())
 {
-    gridTransform_<DefaultTransformPolicy>(src1, src2, dst, op, mask, stream);
+    gridTransformBinary_<DefaultTransformPolicy>(src1, src2, dst, op, mask, stream);
 }
 
 template <class SrcPtr1, class SrcPtr2, typename DstType, class Op>
-__host__ void gridTransform(const SrcPtr1& src1, const SrcPtr1& src2, GpuMat_<DstType>& dst, const Op& op, Stream& stream = Stream::Null())
+__host__ void gridTransformBinary(const SrcPtr1& src1, const SrcPtr2& src2, GpuMat_<DstType>& dst, const Op& op, Stream& stream = Stream::Null())
 {
-    gridTransform_<DefaultTransformPolicy>(src1, src2, dst, op, stream);
+    gridTransformBinary_<DefaultTransformPolicy>(src1, src2, dst, op, stream);
 }
 
 template <class SrcPtr1, class SrcPtr2, typename DstType, class Op>
-__host__ void gridTransform(const SrcPtr1& src1, const SrcPtr1& src2, const GlobPtrSz<DstType>& dst, const Op& op, Stream& stream = Stream::Null())
+__host__ void gridTransformBinary(const SrcPtr1& src1, const SrcPtr2& src2, const GlobPtrSz<DstType>& dst, const Op& op, Stream& stream = Stream::Null())
 {
-    gridTransform_<DefaultTransformPolicy>(src1, src2, dst, op, stream);
+    gridTransformBinary_<DefaultTransformPolicy>(src1, src2, dst, op, stream);
 }
 
 template <class SrcPtr, typename D0, typename D1, class OpTuple, class MaskPtr>
-__host__ void gridTransform(const SrcPtr& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>& >& dst, const OpTuple& op, const MaskPtr& mask, Stream& stream = Stream::Null())
+__host__ void gridTransformTuple(const SrcPtr& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>& >& dst, const OpTuple& op, const MaskPtr& mask, Stream& stream = Stream::Null())
 {
-    gridTransform_<DefaultTransformPolicy>(src, dst, op, mask, stream);
+    gridTransformTuple_<DefaultTransformPolicy>(src, dst, op, mask, stream);
 }
 
 template <class SrcPtr, typename D0, typename D1, class OpTuple, class MaskPtr>
-__host__ void gridTransform(const SrcPtr& src, const tuple< GlobPtrSz<D0>, GlobPtrSz<D1> >& dst, const OpTuple& op, const MaskPtr& mask, Stream& stream = Stream::Null())
+__host__ void gridTransformTuple(const SrcPtr& src, const tuple< GlobPtrSz<D0>, GlobPtrSz<D1> >& dst, const OpTuple& op, const MaskPtr& mask, Stream& stream = Stream::Null())
 {
-    gridTransform_<DefaultTransformPolicy>(src, dst, op, mask, stream);
+    gridTransformTuple_<DefaultTransformPolicy>(src, dst, op, mask, stream);
 }
 
 template <class SrcPtr, typename D0, typename D1, class OpTuple>
-__host__ void gridTransform(const SrcPtr& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>& >& dst, const OpTuple& op, Stream& stream = Stream::Null())
+__host__ void gridTransformTuple(const SrcPtr& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>& >& dst, const OpTuple& op, Stream& stream = Stream::Null())
 {
-    gridTransform_<DefaultTransformPolicy>(src, dst, op, stream);
+    gridTransformTuple_<DefaultTransformPolicy>(src, dst, op, stream);
 }
 
 template <class SrcPtr, typename D0, typename D1, class OpTuple>
-__host__ void gridTransform(const SrcPtr& src, const tuple< GlobPtrSz<D0>, GlobPtrSz<D1> >& dst, const OpTuple& op, Stream& stream = Stream::Null())
+__host__ void gridTransformTuple(const SrcPtr& src, const tuple< GlobPtrSz<D0>, GlobPtrSz<D1> >& dst, const OpTuple& op, Stream& stream = Stream::Null())
 {
-    gridTransform_<DefaultTransformPolicy>(src, dst, op, stream);
+    gridTransformTuple_<DefaultTransformPolicy>(src, dst, op, stream);
 }
 
 template <class SrcPtr, typename D0, typename D1, typename D2, class OpTuple, class MaskPtr>
-__host__ void gridTransform(const SrcPtr& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>&, GpuMat_<D2>& >& dst, const OpTuple& op, const MaskPtr& mask, Stream& stream = Stream::Null())
+__host__ void gridTransformTuple(const SrcPtr& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>&, GpuMat_<D2>& >& dst, const OpTuple& op, const MaskPtr& mask, Stream& stream = Stream::Null())
 {
-    gridTransform_<DefaultTransformPolicy>(src, dst, op, mask, stream);
+    gridTransformTuple_<DefaultTransformPolicy>(src, dst, op, mask, stream);
 }
 
 template <class SrcPtr, typename D0, typename D1, typename D2, class OpTuple, class MaskPtr>
-__host__ void gridTransform(const SrcPtr& src, const tuple< GlobPtrSz<D0>, GlobPtrSz<D1>, GlobPtrSz<D2> >& dst, const OpTuple& op, const MaskPtr& mask, Stream& stream = Stream::Null())
+__host__ void gridTransformTuple(const SrcPtr& src, const tuple< GlobPtrSz<D0>, GlobPtrSz<D1>, GlobPtrSz<D2> >& dst, const OpTuple& op, const MaskPtr& mask, Stream& stream = Stream::Null())
 {
-    gridTransform_<DefaultTransformPolicy>(src, dst, op, mask, stream);
+    gridTransformTuple_<DefaultTransformPolicy>(src, dst, op, mask, stream);
 }
 
 template <class SrcPtr, typename D0, typename D1, typename D2, class OpTuple>
-__host__ void gridTransform(const SrcPtr& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>&, GpuMat_<D2>& >& dst, const OpTuple& op, Stream& stream = Stream::Null())
+__host__ void gridTransformTuple(const SrcPtr& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>&, GpuMat_<D2>& >& dst, const OpTuple& op, Stream& stream = Stream::Null())
 {
-    gridTransform_<DefaultTransformPolicy>(src, dst, op, stream);
+    gridTransformTuple_<DefaultTransformPolicy>(src, dst, op, stream);
 }
 
 template <class SrcPtr, typename D0, typename D1, typename D2, class OpTuple>
-__host__ void gridTransform(const SrcPtr& src, const tuple< GlobPtrSz<D0>, GlobPtrSz<D1>, GlobPtrSz<D2> >& dst, const OpTuple& op, Stream& stream = Stream::Null())
+__host__ void gridTransformTuple(const SrcPtr& src, const tuple< GlobPtrSz<D0>, GlobPtrSz<D1>, GlobPtrSz<D2> >& dst, const OpTuple& op, Stream& stream = Stream::Null())
 {
-    gridTransform_<DefaultTransformPolicy>(src, dst, op, stream);
+    gridTransformTuple_<DefaultTransformPolicy>(src, dst, op, stream);
 }
 
 template <class SrcPtr, typename D0, typename D1, typename D2, typename D3, class OpTuple, class MaskPtr>
-__host__ void gridTransform(const SrcPtr& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>&, GpuMat_<D2>&, GpuMat_<D3>& >& dst, const OpTuple& op, const MaskPtr& mask, Stream& stream = Stream::Null())
+__host__ void gridTransformTuple(const SrcPtr& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>&, GpuMat_<D2>&, GpuMat_<D3>& >& dst, const OpTuple& op, const MaskPtr& mask, Stream& stream = Stream::Null())
 {
-    gridTransform_<DefaultTransformPolicy>(src, dst, op, mask, stream);
+    gridTransformTuple_<DefaultTransformPolicy>(src, dst, op, mask, stream);
 }
 
 template <class SrcPtr, typename D0, typename D1, typename D2, typename D3, class OpTuple, class MaskPtr>
-__host__ void gridTransform(const SrcPtr& src, const tuple< GlobPtrSz<D0>, GlobPtrSz<D1>, GlobPtrSz<D2>, GlobPtrSz<D3> >& dst, const OpTuple& op, const MaskPtr& mask, Stream& stream = Stream::Null())
+__host__ void gridTransformTuple(const SrcPtr& src, const tuple< GlobPtrSz<D0>, GlobPtrSz<D1>, GlobPtrSz<D2>, GlobPtrSz<D3> >& dst, const OpTuple& op, const MaskPtr& mask, Stream& stream = Stream::Null())
 {
-    gridTransform_<DefaultTransformPolicy>(src, dst, op, mask, stream);
+    gridTransformTuple_<DefaultTransformPolicy>(src, dst, op, mask, stream);
 }
 
 template <class SrcPtr, typename D0, typename D1, typename D2, typename D3, class OpTuple>
-__host__ void gridTransform(const SrcPtr& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>&, GpuMat_<D2>&, GpuMat_<D3>& >& dst, const OpTuple& op, Stream& stream = Stream::Null())
+__host__ void gridTransformTuple(const SrcPtr& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>&, GpuMat_<D2>&, GpuMat_<D3>& >& dst, const OpTuple& op, Stream& stream = Stream::Null())
 {
-    gridTransform_<DefaultTransformPolicy>(src, dst, op, stream);
+    gridTransformTuple_<DefaultTransformPolicy>(src, dst, op, stream);
 }
 
 template <class SrcPtr, typename D0, typename D1, typename D2, typename D3, class OpTuple>
-__host__ void gridTransform(const SrcPtr& src, const tuple< GlobPtrSz<D0>, GlobPtrSz<D1>, GlobPtrSz<D2>, GlobPtrSz<D3> >& dst, const OpTuple& op, Stream& stream = Stream::Null())
+__host__ void gridTransformTuple(const SrcPtr& src, const tuple< GlobPtrSz<D0>, GlobPtrSz<D1>, GlobPtrSz<D2>, GlobPtrSz<D3> >& dst, const OpTuple& op, Stream& stream = Stream::Null())
 {
-    gridTransform_<DefaultTransformPolicy>(src, dst, op, stream);
+    gridTransformTuple_<DefaultTransformPolicy>(src, dst, op, stream);
 }
 
 }}
diff --git a/modules/cudev/include/opencv2/cudev/grid/transpose.hpp b/modules/cudev/include/opencv2/cudev/grid/transpose.hpp
index 5e8f24868f..cf1bf8303e 100644
--- a/modules/cudev/include/opencv2/cudev/grid/transpose.hpp
+++ b/modules/cudev/include/opencv2/cudev/grid/transpose.hpp
@@ -49,19 +49,53 @@
 #include "../common.hpp"
 #include "../ptr2d/traits.hpp"
 #include "../ptr2d/gpumat.hpp"
+#include "../ptr2d/glob.hpp"
 #include "detail/transpose.hpp"
 
 namespace cv { namespace cudev {
 
-template <class SrcPtr, typename DstType>
-__host__ void gridTranspose(const SrcPtr& src, GpuMat_<DstType>& dst, Stream& stream = Stream::Null())
+template <class Policy, class SrcPtr, typename DstType>
+__host__ void gridTranspose_(const SrcPtr& src, GpuMat_<DstType>& dst, Stream& stream = Stream::Null())
 {
     const int rows = getRows(src);
     const int cols = getCols(src);
 
     dst.create(cols, rows);
 
-    transpose_detail::transpose(shrinkPtr(src), shrinkPtr(dst), rows, cols, StreamAccessor::getStream(stream));
+    transpose_detail::transpose<Policy>(shrinkPtr(src), shrinkPtr(dst), rows, cols, StreamAccessor::getStream(stream));
+}
+
+template <class Policy, class SrcPtr, typename DstType>
+__host__ void gridTranspose_(const SrcPtr& src, const GlobPtrSz<DstType>& dst, Stream& stream = Stream::Null())
+{
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    CV_Assert( getRows(dst) == cols && getCols(dst) == rows );
+
+    transpose_detail::transpose<Policy>(shrinkPtr(src), shrinkPtr(dst), rows, cols, StreamAccessor::getStream(stream));
+}
+
+// Default Policy
+
+struct DefaultTransposePolicy
+{
+    enum {
+        tile_dim    = 16,
+        block_dim_y = 16
+    };
+};
+
+template <class SrcPtr, typename DstType>
+__host__ void gridTranspose(const SrcPtr& src, GpuMat_<DstType>& dst, Stream& stream = Stream::Null())
+{
+    gridTranspose_<DefaultTransposePolicy>(src, dst, stream);
+}
+
+template <class SrcPtr, typename DstType>
+__host__ void gridTranspose(const SrcPtr& src, const GlobPtrSz<DstType>& dst, Stream& stream = Stream::Null())
+{
+    gridTranspose_<DefaultTransposePolicy>(src, dst, stream);
 }
 
 }}
diff --git a/modules/cudev/include/opencv2/cudev/ptr2d/lut.hpp b/modules/cudev/include/opencv2/cudev/ptr2d/lut.hpp
index e47719c6ec..accf545617 100644
--- a/modules/cudev/include/opencv2/cudev/ptr2d/lut.hpp
+++ b/modules/cudev/include/opencv2/cudev/ptr2d/lut.hpp
@@ -47,6 +47,7 @@
 #define __OPENCV_CUDEV_PTR2D_LUT_HPP__
 
 #include "../common.hpp"
+#include "../util/vec_traits.hpp"
 #include "../grid/copy.hpp"
 #include "traits.hpp"
 #include "gpumat.hpp"
@@ -63,7 +64,8 @@ template <class SrcPtr, class TablePtr> struct LutPtr
 
     __device__ __forceinline__ typename PtrTraits<TablePtr>::value_type operator ()(typename PtrTraits<SrcPtr>::index_type y, typename PtrTraits<SrcPtr>::index_type x) const
     {
-        return tbl(0, src(y, x));
+        typedef typename PtrTraits<TablePtr>::index_type tbl_index_type;
+        return tbl(VecTraits<tbl_index_type>::all(0), src(y, x));
     }
 };
 
@@ -81,8 +83,6 @@ template <class SrcPtr, class TablePtr> struct LutPtrSz : LutPtr<SrcPtr, TablePt
 template <class SrcPtr, class TablePtr>
 __host__ LutPtrSz<typename PtrTraits<SrcPtr>::ptr_type, typename PtrTraits<TablePtr>::ptr_type> lutPtr(const SrcPtr& src, const TablePtr& tbl)
 {
-    CV_Assert( getRows(tbl) == 1 );
-
     LutPtrSz<typename PtrTraits<SrcPtr>::ptr_type, typename PtrTraits<TablePtr>::ptr_type> ptr;
     ptr.src = shrinkPtr(src);
     ptr.tbl = shrinkPtr(tbl);
diff --git a/modules/cudev/include/opencv2/cudev/ptr2d/mask.hpp b/modules/cudev/include/opencv2/cudev/ptr2d/mask.hpp
index 1ff3371931..d425fa3182 100644
--- a/modules/cudev/include/opencv2/cudev/ptr2d/mask.hpp
+++ b/modules/cudev/include/opencv2/cudev/ptr2d/mask.hpp
@@ -62,6 +62,42 @@ struct WithOutMask
     }
 };
 
+template <class MaskPtr> struct SingleMaskChannels
+{
+    typedef typename PtrTraits<MaskPtr>::value_type value_type;
+    typedef typename PtrTraits<MaskPtr>::index_type index_type;
+
+    MaskPtr mask;
+    int channels;
+
+    __device__ __forceinline__ value_type operator()(index_type y, index_type x) const
+    {
+        return mask(y, x / channels);
+    }
+
+};
+
+template <class MaskPtr> struct SingleMaskChannelsSz : SingleMaskChannels<MaskPtr>
+{
+    int rows, cols;
+};
+
+template <class MaskPtr>
+__host__ SingleMaskChannelsSz<typename PtrTraits<MaskPtr>::ptr_type>
+singleMaskChannels(const MaskPtr& mask, int channels)
+{
+    SingleMaskChannelsSz<typename PtrTraits<MaskPtr>::ptr_type> ptr;
+    ptr.mask = shrinkPtr(mask);
+    ptr.channels = channels;
+    ptr.rows = getRows(mask);
+    ptr.cols = getCols(mask) * channels;
+    return ptr;
+}
+
+template <class MaskPtr> struct PtrTraits< SingleMaskChannelsSz<MaskPtr> > : PtrTraitsBase<SingleMaskChannelsSz<MaskPtr>, SingleMaskChannels<MaskPtr> >
+{
+};
+
 }}
 
 #endif
diff --git a/modules/cudev/include/opencv2/cudev/util/vec_math.hpp b/modules/cudev/include/opencv2/cudev/util/vec_math.hpp
index 3ce265e277..361ef7b294 100644
--- a/modules/cudev/include/opencv2/cudev/util/vec_math.hpp
+++ b/modules/cudev/include/opencv2/cudev/util/vec_math.hpp
@@ -194,10 +194,23 @@ CV_CUDEV_IMPLEMENT_VEC_UNARY_OP(~, uint, uint)
         return VecTraits<output_type ## 4>::make(func (a.x), func (a.y), func (a.z), func (a.w)); \
     }
 
+namespace vec_math_detail
+{
+    __device__ __forceinline__ schar abs_(schar val)
+    {
+        return (schar) ::abs((int) val);
+    }
+
+    __device__ __forceinline__ short abs_(short val)
+    {
+        return (short) ::abs((int) val);
+    }
+}
+
 CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(abs, /*::abs*/, uchar, uchar)
-CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(abs, ::abs, char, char)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(abs, vec_math_detail::abs_, char, char)
 CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(abs, /*::abs*/, ushort, ushort)
-CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(abs, ::abs, short, short)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(abs, vec_math_detail::abs_, short, short)
 CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(abs, ::abs, int, int)
 CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(abs, /*::abs*/, uint, uint)
 CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(abs, ::fabsf, float, float)
diff --git a/modules/cudev/include/opencv2/cudev/util/vec_traits.hpp b/modules/cudev/include/opencv2/cudev/util/vec_traits.hpp
index bdff958d87..585423dd52 100644
--- a/modules/cudev/include/opencv2/cudev/util/vec_traits.hpp
+++ b/modules/cudev/include/opencv2/cudev/util/vec_traits.hpp
@@ -70,7 +70,7 @@ CV_CUDEV_MAKE_VEC_INST(double)
 
 #undef CV_CUDEV_MAKE_VEC_INST
 
-template<> struct MakeVec<schar, 1> { typedef char  type; };
+template<> struct MakeVec<schar, 1> { typedef schar type; };
 template<> struct MakeVec<schar, 2> { typedef char2 type; };
 template<> struct MakeVec<schar, 3> { typedef char3 type; };
 template<> struct MakeVec<schar, 4> { typedef char4 type; };
diff --git a/modules/cudev/test/test_reduction.cu b/modules/cudev/test/test_reduction.cu
index 03c78def15..c376059870 100644
--- a/modules/cudev/test/test_reduction.cu
+++ b/modules/cudev/test/test_reduction.cu
@@ -228,6 +228,9 @@ TEST(ReduceToColumn, Sum)
 
     Mat dst_gold;
     cv::reduce(src, dst_gold, 1, REDUCE_SUM, CV_32S);
+    dst_gold.cols = dst_gold.rows;
+    dst_gold.rows = 1;
+    dst_gold.step = dst_gold.cols * dst_gold.elemSize();
 
     EXPECT_MAT_NEAR(dst_gold, dst, 0.0);
 }
@@ -244,6 +247,9 @@ TEST(ReduceToColumn, Avg)
 
     Mat dst_gold;
     cv::reduce(src, dst_gold, 1, REDUCE_AVG, CV_32F);
+    dst_gold.cols = dst_gold.rows;
+    dst_gold.rows = 1;
+    dst_gold.step = dst_gold.cols * dst_gold.elemSize();
 
     EXPECT_MAT_NEAR(dst_gold, dst, 1e-4);
 }
@@ -260,6 +266,9 @@ TEST(ReduceToColumn, Min)
 
     Mat dst_gold;
     cv::reduce(src, dst_gold, 1, REDUCE_MIN);
+    dst_gold.cols = dst_gold.rows;
+    dst_gold.rows = 1;
+    dst_gold.step = dst_gold.cols * dst_gold.elemSize();
 
     EXPECT_MAT_NEAR(dst_gold, dst, 0.0);
 }
@@ -276,6 +285,9 @@ TEST(ReduceToColumn, Max)
 
     Mat dst_gold;
     cv::reduce(src, dst_gold, 1, REDUCE_MAX);
+    dst_gold.cols = dst_gold.rows;
+    dst_gold.rows = 1;
+    dst_gold.step = dst_gold.cols * dst_gold.elemSize();
 
     EXPECT_MAT_NEAR(dst_gold, dst, 0.0);
 }