gpuarithm module for arithmetics operations on matrices

modules/gpuarithm/CMakeLists.txt

@@ -0,0 +1,13 @@
+if(ANDROID OR IOS)
+  ocv_module_disable(gpuarithm)
+endif()
+
+set(the_description "GPU-accelerated Operations on Matrices")
+
+ocv_warnings_disable(CMAKE_CXX_FLAGS -Wundef -Wmissing-declarations)
+
+ocv_define_module(gpuarithm opencv_core)
+
+if(HAVE_CUBLAS)
+  CUDA_ADD_CUBLAS_TO_TARGET(${the_module})
+endif()

modules/gpuarithm/doc/gpuarithm.rst

@@ -0,0 +1,10 @@
+*******************************************
+gpu. GPU-accelerated Operations on Matrices
+*******************************************
+
+.. toctree::
+    :maxdepth: 1
+
+    operations_on_matrices
+    per_element_operations
+    matrix_reductions

modules/gpuarithm/doc/matrix_reductions.rst

@@ -0,0 +1,207 @@
+Matrix Reductions
+=================
+
+.. highlight:: cpp
+
+
+
+gpu::meanStdDev
+-------------------
+Computes a mean value and a standard deviation of matrix elements.
+
+.. ocv:function:: void gpu::meanStdDev(const GpuMat& mtx, Scalar& mean, Scalar& stddev)
+.. ocv:function:: void gpu::meanStdDev(const GpuMat& mtx, Scalar& mean, Scalar& stddev, GpuMat& buf)
+
+    :param mtx: Source matrix.  ``CV_8UC1``  matrices are supported for now.
+
+    :param mean: Mean value.
+
+    :param stddev: Standard deviation value.
+
+    :param buf: Optional buffer to avoid extra memory allocations. It is resized automatically.
+
+.. seealso:: :ocv:func:`meanStdDev`
+
+
+
+gpu::norm
+-------------
+Returns the norm of a matrix (or difference of two matrices).
+
+.. ocv:function:: double gpu::norm(const GpuMat& src1, int normType=NORM_L2)
+
+.. ocv:function:: double gpu::norm(const GpuMat& src1, int normType, GpuMat& buf)
+
+.. ocv:function:: double gpu::norm(const GpuMat& src1, int normType, const GpuMat& mask, GpuMat& buf)
+
+.. ocv:function:: double gpu::norm(const GpuMat& src1, const GpuMat& src2, int normType=NORM_L2)
+
+    :param src1: Source matrix. Any matrices except 64F are supported.
+
+    :param src2: Second source matrix (if any) with the same size and type as ``src1``.
+
+    :param normType: Norm type.  ``NORM_L1`` ,  ``NORM_L2`` , and  ``NORM_INF``  are supported for now.
+
+    :param mask: optional operation mask; it must have the same size as ``src1`` and ``CV_8UC1`` type.
+
+    :param buf: Optional buffer to avoid extra memory allocations. It is resized automatically.
+
+.. seealso:: :ocv:func:`norm`
+
+
+
+gpu::sum
+------------
+Returns the sum of matrix elements.
+
+.. ocv:function:: Scalar gpu::sum(const GpuMat& src)
+
+.. ocv:function:: Scalar gpu::sum(const GpuMat& src, GpuMat& buf)
+
+.. ocv:function:: Scalar gpu::sum(const GpuMat& src, const GpuMat& mask, GpuMat& buf)
+
+    :param src: Source image of any depth except for ``CV_64F`` .
+
+    :param mask: optional operation mask; it must have the same size as ``src1`` and ``CV_8UC1`` type.
+
+    :param buf: Optional buffer to avoid extra memory allocations. It is resized automatically.
+
+.. seealso:: :ocv:func:`sum`
+
+
+
+gpu::absSum
+---------------
+Returns the sum of absolute values for matrix elements.
+
+.. ocv:function:: Scalar gpu::absSum(const GpuMat& src)
+
+.. ocv:function:: Scalar gpu::absSum(const GpuMat& src, GpuMat& buf)
+
+.. ocv:function:: Scalar gpu::absSum(const GpuMat& src, const GpuMat& mask, GpuMat& buf)
+
+    :param src: Source image of any depth except for ``CV_64F`` .
+
+    :param mask: optional operation mask; it must have the same size as ``src1`` and ``CV_8UC1`` type.
+
+    :param buf: Optional buffer to avoid extra memory allocations. It is resized automatically.
+
+
+
+gpu::sqrSum
+---------------
+Returns the squared sum of matrix elements.
+
+.. ocv:function:: Scalar gpu::sqrSum(const GpuMat& src)
+
+.. ocv:function:: Scalar gpu::sqrSum(const GpuMat& src, GpuMat& buf)
+
+.. ocv:function:: Scalar gpu::sqrSum(const GpuMat& src, const GpuMat& mask, GpuMat& buf)
+
+    :param src: Source image of any depth except for ``CV_64F`` .
+
+    :param mask: optional operation mask; it must have the same size as ``src1`` and ``CV_8UC1`` type.
+
+    :param buf: Optional buffer to avoid extra memory allocations. It is resized automatically.
+
+
+
+gpu::minMax
+---------------
+Finds global minimum and maximum matrix elements and returns their values.
+
+.. ocv:function:: void gpu::minMax(const GpuMat& src, double* minVal, double* maxVal=0, const GpuMat& mask=GpuMat())
+
+.. ocv:function:: void gpu::minMax(const GpuMat& src, double* minVal, double* maxVal, const GpuMat& mask, GpuMat& buf)
+
+    :param src: Single-channel source image.
+
+    :param minVal: Pointer to the returned minimum value.  Use ``NULL``  if not required.
+
+    :param maxVal: Pointer to the returned maximum value.  Use ``NULL``  if not required.
+
+    :param mask: Optional mask to select a sub-matrix.
+
+    :param buf: Optional buffer to avoid extra memory allocations. It is resized automatically.
+
+The function does not work with ``CV_64F`` images on GPUs with the compute capability < 1.3.
+
+.. seealso:: :ocv:func:`minMaxLoc`
+
+
+
+gpu::minMaxLoc
+------------------
+Finds global minimum and maximum matrix elements and returns their values with locations.
+
+.. ocv:function:: void gpu::minMaxLoc(const GpuMat& src, double* minVal, double* maxVal=0, Point* minLoc=0, Point* maxLoc=0, const GpuMat& mask=GpuMat())
+
+.. ocv:function:: void gpu::minMaxLoc(const GpuMat& src, double* minVal, double* maxVal, Point* minLoc, Point* maxLoc, const GpuMat& mask, GpuMat& valbuf, GpuMat& locbuf)
+
+    :param src: Single-channel source image.
+
+    :param minVal: Pointer to the returned minimum value. Use ``NULL``  if not required.
+
+    :param maxVal: Pointer to the returned maximum value. Use ``NULL``  if not required.
+
+    :param minLoc: Pointer to the returned minimum location. Use ``NULL``  if not required.
+
+    :param maxLoc: Pointer to the returned maximum location. Use ``NULL``  if not required.
+
+    :param mask: Optional mask to select a sub-matrix.
+
+    :param valbuf: Optional values buffer to avoid extra memory allocations. It is resized automatically.
+
+    :param locbuf: Optional locations buffer to avoid extra memory allocations. It is resized automatically.
+
+    The function does not work with ``CV_64F`` images on GPU with the compute capability < 1.3.
+
+.. seealso:: :ocv:func:`minMaxLoc`
+
+
+
+gpu::countNonZero
+---------------------
+Counts non-zero matrix elements.
+
+.. ocv:function:: int gpu::countNonZero(const GpuMat& src)
+
+.. ocv:function:: int gpu::countNonZero(const GpuMat& src, GpuMat& buf)
+
+    :param src: Single-channel source image.
+
+    :param buf: Optional buffer to avoid extra memory allocations. It is resized automatically.
+
+The function does not work with ``CV_64F`` images on GPUs with the compute capability < 1.3.
+
+.. seealso:: :ocv:func:`countNonZero`
+
+
+
+gpu::reduce
+-----------
+Reduces a matrix to a vector.
+
+.. ocv:function:: void gpu::reduce(const GpuMat& mtx, GpuMat& vec, int dim, int reduceOp, int dtype = -1, Stream& stream = Stream::Null())
+
+    :param mtx: Source 2D matrix.
+
+    :param vec: Destination vector. Its size and type is defined by  ``dim``  and  ``dtype``  parameters.
+
+    :param dim: Dimension index along which the matrix is reduced. 0 means that the matrix is reduced to a single row. 1 means that the matrix is reduced to a single column.
+
+    :param reduceOp: Reduction operation that could be one of the following:
+
+            * **CV_REDUCE_SUM** The output is the sum of all rows/columns of the matrix.
+
+            * **CV_REDUCE_AVG** The output is the mean vector of all rows/columns of the matrix.
+
+            * **CV_REDUCE_MAX** The output is the maximum (column/row-wise) of all rows/columns of the matrix.
+
+            * **CV_REDUCE_MIN** The output is the minimum (column/row-wise) of all rows/columns of the matrix.
+
+    :param dtype: When it is negative, the destination vector will have the same type as the source matrix. Otherwise, its type will be  ``CV_MAKE_TYPE(CV_MAT_DEPTH(dtype), mtx.channels())`` .
+
+The function ``reduce`` reduces the matrix to a vector by treating the matrix rows/columns as a set of 1D vectors and performing the specified operation on the vectors until a single row/column is obtained. For example, the function can be used to compute horizontal and vertical projections of a raster image. In case of ``CV_REDUCE_SUM`` and ``CV_REDUCE_AVG`` , the output may have a larger element bit-depth to preserve accuracy. And multi-channel arrays are also supported in these two reduction modes.
+
+.. seealso:: :ocv:func:`reduce`

modules/gpuarithm/doc/operations_on_matrices.rst

@@ -0,0 +1,274 @@
+Operations on Matrices
+======================
+
+.. highlight:: cpp
+
+
+
+gpu::gemm
+------------------
+Performs generalized matrix multiplication.
+
+.. ocv:function:: void gpu::gemm(const GpuMat& src1, const GpuMat& src2, double alpha, const GpuMat& src3, double beta, GpuMat& dst, int flags = 0, Stream& stream = Stream::Null())
+
+    :param src1: First multiplied input matrix that should have  ``CV_32FC1`` , ``CV_64FC1`` , ``CV_32FC2`` , or  ``CV_64FC2``  type.
+
+    :param src2: Second multiplied input matrix of the same type as  ``src1`` .
+
+    :param alpha: Weight of the matrix product.
+
+    :param src3: Third optional delta matrix added to the matrix product. It should have the same type as  ``src1``  and  ``src2`` .
+
+    :param beta: Weight of  ``src3`` .
+
+    :param dst: Destination matrix. It has the proper size and the same type as input matrices.
+
+    :param flags: Operation flags:
+
+            * **GEMM_1_T** transpose  ``src1``
+            * **GEMM_2_T** transpose  ``src2``
+            * **GEMM_3_T** transpose  ``src3``
+
+    :param stream: Stream for the asynchronous version.
+
+The function performs generalized matrix multiplication similar to the ``gemm`` functions in BLAS level 3. For example, ``gemm(src1, src2, alpha, src3, beta, dst, GEMM_1_T + GEMM_3_T)`` corresponds to
+
+.. math::
+
+    \texttt{dst} =  \texttt{alpha} \cdot \texttt{src1} ^T  \cdot \texttt{src2} +  \texttt{beta} \cdot \texttt{src3} ^T
+
+.. note:: Transposition operation doesn't support  ``CV_64FC2``  input type.
+
+.. seealso:: :ocv:func:`gemm`
+
+
+
+gpu::transpose
+------------------
+Transposes a matrix.
+
+.. ocv:function:: void gpu::transpose( const GpuMat& src1, GpuMat& dst, Stream& stream=Stream::Null() )
+
+    :param src1: Source matrix. 1-, 4-, 8-byte element sizes are supported for now (CV_8UC1, CV_8UC4, CV_16UC2, CV_32FC1, etc).
+
+    :param dst: Destination matrix.
+
+    :param stream: Stream for the asynchronous version.
+
+.. seealso:: :ocv:func:`transpose`
+
+
+
+gpu::flip
+-------------
+Flips a 2D matrix around vertical, horizontal, or both axes.
+
+.. ocv:function:: void gpu::flip( const GpuMat& a, GpuMat& b, int flipCode, Stream& stream=Stream::Null() )
+
+    :param a: Source matrix. Supports 1, 3 and 4 channels images with ``CV_8U``, ``CV_16U``, ``CV_32S`` or ``CV_32F`` depth.
+
+    :param b: Destination matrix.
+
+    :param flipCode: Flip mode for the source:
+
+        * ``0`` Flips around x-axis.
+
+        * ``>0`` Flips around y-axis.
+
+        * ``<0`` Flips around both axes.
+
+    :param stream: Stream for the asynchronous version.
+
+.. seealso:: :ocv:func:`flip`
+
+
+
+gpu::LUT
+------------
+Transforms the source matrix into the destination matrix using the given look-up table: ``dst(I) = lut(src(I))``
+
+.. ocv:function:: void gpu::LUT(const GpuMat& src, const Mat& lut, GpuMat& dst, Stream& stream = Stream::Null())
+
+    :param src: Source matrix.  ``CV_8UC1``  and  ``CV_8UC3``  matrices are supported for now.
+
+    :param lut: Look-up table of 256 elements. It is a continuous ``CV_8U`` matrix.
+
+    :param dst: Destination matrix with the same depth as  ``lut``  and the same number of channels as  ``src`` .
+
+    :param stream: Stream for the asynchronous version.
+
+.. seealso:: :ocv:func:`LUT`
+
+
+
+gpu::merge
+--------------
+Makes a multi-channel matrix out of several single-channel matrices.
+
+.. ocv:function:: void gpu::merge(const GpuMat* src, size_t n, GpuMat& dst, Stream& stream = Stream::Null())
+
+.. ocv:function:: void gpu::merge(const vector<GpuMat>& src, GpuMat& dst, Stream& stream = Stream::Null())
+
+    :param src: Array/vector of source matrices.
+
+    :param n: Number of source matrices.
+
+    :param dst: Destination matrix.
+
+    :param stream: Stream for the asynchronous version.
+
+.. seealso:: :ocv:func:`merge`
+
+
+
+gpu::split
+--------------
+Copies each plane of a multi-channel matrix into an array.
+
+.. ocv:function:: void gpu::split(const GpuMat& src, GpuMat* dst, Stream& stream = Stream::Null())
+
+.. ocv:function:: void gpu::split(const GpuMat& src, vector<GpuMat>& dst, Stream& stream = Stream::Null())
+
+    :param src: Source matrix.
+
+    :param dst: Destination array/vector of single-channel matrices.
+
+    :param stream: Stream for the asynchronous version.
+
+.. seealso:: :ocv:func:`split`
+
+
+
+gpu::magnitude
+------------------
+Computes magnitudes of complex matrix elements.
+
+.. ocv:function:: void gpu::magnitude( const GpuMat& xy, GpuMat& magnitude, Stream& stream=Stream::Null() )
+
+.. ocv:function:: void gpu::magnitude(const GpuMat& x, const GpuMat& y, GpuMat& magnitude, Stream& stream = Stream::Null())
+
+    :param xy: Source complex matrix in the interleaved format ( ``CV_32FC2`` ).
+
+    :param x: Source matrix containing real components ( ``CV_32FC1`` ).
+
+    :param y: Source matrix containing imaginary components ( ``CV_32FC1`` ).
+
+    :param magnitude: Destination matrix of float magnitudes ( ``CV_32FC1`` ).
+
+    :param stream: Stream for the asynchronous version.
+
+.. seealso:: :ocv:func:`magnitude`
+
+
+
+gpu::magnitudeSqr
+---------------------
+Computes squared magnitudes of complex matrix elements.
+
+.. ocv:function:: void gpu::magnitudeSqr( const GpuMat& xy, GpuMat& magnitude, Stream& stream=Stream::Null() )
+
+.. ocv:function:: void gpu::magnitudeSqr(const GpuMat& x, const GpuMat& y, GpuMat& magnitude, Stream& stream = Stream::Null())
+
+    :param xy: Source complex matrix in the interleaved format ( ``CV_32FC2`` ).
+
+    :param x: Source matrix containing real components ( ``CV_32FC1`` ).
+
+    :param y: Source matrix containing imaginary components ( ``CV_32FC1`` ).
+
+    :param magnitude: Destination matrix of float magnitude squares ( ``CV_32FC1`` ).
+
+    :param stream: Stream for the asynchronous version.
+
+
+
+gpu::phase
+--------------
+Computes polar angles of complex matrix elements.
+
+.. ocv:function:: void gpu::phase(const GpuMat& x, const GpuMat& y, GpuMat& angle, bool angleInDegrees=false, Stream& stream = Stream::Null())
+
+    :param x: Source matrix containing real components ( ``CV_32FC1`` ).
+
+    :param y: Source matrix containing imaginary components ( ``CV_32FC1`` ).
+
+    :param angle: Destination matrix of angles ( ``CV_32FC1`` ).
+
+    :param angleInDegrees: Flag for angles that must be evaluated in degrees.
+
+    :param stream: Stream for the asynchronous version.
+
+.. seealso:: :ocv:func:`phase`
+
+
+
+gpu::cartToPolar
+--------------------
+Converts Cartesian coordinates into polar.
+
+.. ocv:function:: void gpu::cartToPolar(const GpuMat& x, const GpuMat& y, GpuMat& magnitude, GpuMat& angle, bool angleInDegrees=false, Stream& stream = Stream::Null())
+
+    :param x: Source matrix containing real components ( ``CV_32FC1`` ).
+
+    :param y: Source matrix containing imaginary components ( ``CV_32FC1`` ).
+
+    :param magnitude: Destination matrix of float magnitudes ( ``CV_32FC1`` ).
+
+    :param angle: Destination matrix of angles ( ``CV_32FC1`` ).
+
+    :param angleInDegrees: Flag for angles that must be evaluated in degrees.
+
+    :param stream: Stream for the asynchronous version.
+
+.. seealso:: :ocv:func:`cartToPolar`
+
+
+
+gpu::polarToCart
+--------------------
+Converts polar coordinates into Cartesian.
+
+.. ocv:function:: void gpu::polarToCart(const GpuMat& magnitude, const GpuMat& angle, GpuMat& x, GpuMat& y, bool angleInDegrees=false, Stream& stream = Stream::Null())
+
+    :param magnitude: Source matrix containing magnitudes ( ``CV_32FC1`` ).
+
+    :param angle: Source matrix containing angles ( ``CV_32FC1`` ).
+
+    :param x: Destination matrix of real components ( ``CV_32FC1`` ).
+
+    :param y: Destination matrix of imaginary components ( ``CV_32FC1`` ).
+
+    :param angleInDegrees: Flag that indicates angles in degrees.
+
+    :param stream: Stream for the asynchronous version.
+
+.. seealso:: :ocv:func:`polarToCart`
+
+
+
+gpu::normalize
+--------------
+Normalizes the norm or value range of an array.
+
+.. ocv:function:: void gpu::normalize(const GpuMat& src, GpuMat& dst, double alpha = 1, double beta = 0, int norm_type = NORM_L2, int dtype = -1, const GpuMat& mask = GpuMat())
+
+.. ocv:function:: void gpu::normalize(const GpuMat& src, GpuMat& dst, double a, double b, int norm_type, int dtype, const GpuMat& mask, GpuMat& norm_buf, GpuMat& cvt_buf)
+
+    :param src: input array.
+
+    :param dst: output array of the same size as  ``src`` .
+
+    :param alpha: norm value to normalize to or the lower range boundary in case of the range normalization.
+
+    :param beta: upper range boundary in case of the range normalization; it is not used for the norm normalization.
+
+    :param normType: normalization type (see the details below).
+
+    :param dtype: when negative, the output array has the same type as ``src``; otherwise, it has the same number of channels as  ``src`` and the depth ``=CV_MAT_DEPTH(dtype)``.
+
+    :param mask: optional operation mask.
+
+    :param norm_buf: Optional buffer to avoid extra memory allocations. It is resized automatically.
+
+    :param cvt_buf: Optional buffer to avoid extra memory allocations. It is resized automatically.
+
+.. seealso:: :ocv:func:`normalize`

modules/gpuarithm/doc/per_element_operations.rst

@@ -0,0 +1,467 @@
+Per-element Operations
+=======================
+
+.. highlight:: cpp
+
+
+
+gpu::add
+------------
+Computes a matrix-matrix or matrix-scalar sum.
+
+.. ocv:function:: void gpu::add( const GpuMat& a, const GpuMat& b, GpuMat& c, const GpuMat& mask=GpuMat(), int dtype=-1, Stream& stream=Stream::Null() )
+
+.. ocv:function:: void gpu::add( const GpuMat& a, const Scalar& sc, GpuMat& c, const GpuMat& mask=GpuMat(), int dtype=-1, Stream& stream=Stream::Null() )
+
+    :param a: First source matrix.
+
+    :param b: Second source matrix to be added to ``a`` . Matrix should have the same size and type as ``a`` .
+
+    :param sc: A scalar to be added to ``a`` .
+
+    :param c: Destination matrix that has the same size and number of channels as the input array(s). The depth is defined by ``dtype`` or ``a`` depth.
+
+    :param mask: Optional operation mask, 8-bit single channel array, that specifies elements of the destination array to be changed.
+
+    :param dtype: Optional depth of the output array.
+
+    :param stream: Stream for the asynchronous version.
+
+.. seealso:: :ocv:func:`add`
+
+
+
+gpu::subtract
+-----------------
+Computes a matrix-matrix or matrix-scalar difference.
+
+.. ocv:function:: void gpu::subtract( const GpuMat& a, const GpuMat& b, GpuMat& c, const GpuMat& mask=GpuMat(), int dtype=-1, Stream& stream=Stream::Null() )
+
+.. ocv:function:: void gpu::subtract( const GpuMat& a, const Scalar& sc, GpuMat& c, const GpuMat& mask=GpuMat(), int dtype=-1, Stream& stream=Stream::Null() )
+
+    :param a: First source matrix.
+
+    :param b: Second source matrix to be added to ``a`` . Matrix should have the same size and type as ``a`` .
+
+    :param sc: A scalar to be added to ``a`` .
+
+    :param c: Destination matrix that has the same size and number of channels as the input array(s). The depth is defined by ``dtype`` or ``a`` depth.
+
+    :param mask: Optional operation mask, 8-bit single channel array, that specifies elements of the destination array to be changed.
+
+    :param dtype: Optional depth of the output array.
+
+    :param stream: Stream for the asynchronous version.
+
+.. seealso:: :ocv:func:`subtract`
+
+
+
+gpu::multiply
+-----------------
+Computes a matrix-matrix or matrix-scalar per-element product.
+
+.. ocv:function:: void gpu::multiply( const GpuMat& a, const GpuMat& b, GpuMat& c, double scale=1, int dtype=-1, Stream& stream=Stream::Null() )
+
+.. ocv:function:: void gpu::multiply( const GpuMat& a, const Scalar& sc, GpuMat& c, double scale=1, int dtype=-1, Stream& stream=Stream::Null() )
+
+    :param a: First source matrix.
+
+    :param b: Second source matrix to be multiplied by ``a`` elements.
+
+    :param sc: A scalar to be multiplied by ``a`` elements.
+
+    :param c: Destination matrix that has the same size and number of channels as the input array(s). The depth is defined by ``dtype`` or ``a`` depth.
+
+    :param scale: Optional scale factor.
+
+    :param dtype: Optional depth of the output array.
+
+    :param stream: Stream for the asynchronous version.
+
+.. seealso:: :ocv:func:`multiply`
+
+
+
+gpu::divide
+-----------
+Computes a matrix-matrix or matrix-scalar division.
+
+.. ocv:function:: void gpu::divide( const GpuMat& a, const GpuMat& b, GpuMat& c, double scale=1, int dtype=-1, Stream& stream=Stream::Null() )
+
+.. ocv:function:: void gpu::divide(const GpuMat& a, const Scalar& sc, GpuMat& c, double scale = 1, int dtype = -1, Stream& stream = Stream::Null())
+
+.. ocv:function:: void gpu::divide( double scale, const GpuMat& b, GpuMat& c, int dtype=-1, Stream& stream=Stream::Null() )
+
+    :param a: First source matrix or a scalar.
+
+    :param b: Second source matrix. The ``a`` elements are divided by it.
+
+    :param sc: A scalar to be divided by the elements of ``a`` matrix.
+
+    :param c: Destination matrix that has the same size and number of channels as the input array(s). The depth is defined by ``dtype`` or ``a`` depth.
+
+    :param scale: Optional scale factor.
+
+    :param dtype: Optional depth of the output array.
+
+    :param stream: Stream for the asynchronous version.
+
+This function, in contrast to :ocv:func:`divide`, uses a round-down rounding mode.
+
+.. seealso:: :ocv:func:`divide`
+
+
+gpu::addWeighted
+----------------
+Computes the weighted sum of two arrays.
+
+.. ocv:function:: void gpu::addWeighted(const GpuMat& src1, double alpha, const GpuMat& src2, double beta, double gamma, GpuMat& dst, int dtype = -1, Stream& stream = Stream::Null())
+
+    :param src1: First source array.
+
+    :param alpha: Weight for the first array elements.
+
+    :param src2: Second source array of the same size and channel number as  ``src1`` .
+
+    :param beta: Weight for the second array elements.
+
+    :param dst: Destination array that has the same size and number of channels as the input arrays.
+
+    :param gamma: Scalar added to each sum.
+
+    :param dtype: Optional depth of the destination array. When both input arrays have the same depth, ``dtype`` can be set to ``-1``, which will be equivalent to ``src1.depth()``.
+
+    :param stream: Stream for the asynchronous version.
+
+The function ``addWeighted`` calculates the weighted sum of two arrays as follows:
+
+.. math::
+
+    \texttt{dst} (I)= \texttt{saturate} ( \texttt{src1} (I)* \texttt{alpha} +  \texttt{src2} (I)* \texttt{beta} +  \texttt{gamma} )
+
+where ``I`` is a multi-dimensional index of array elements. In case of multi-channel arrays, each channel is processed independently.
+
+.. seealso:: :ocv:func:`addWeighted`
+
+
+
+gpu::abs
+------------
+Computes an absolute value of each matrix element.
+
+.. ocv:function:: void gpu::abs(const GpuMat& src, GpuMat& dst, Stream& stream = Stream::Null())
+
+    :param src: Source matrix. Supports ``CV_16S`` and ``CV_32F`` depth.
+
+    :param dst: Destination matrix with the same size and type as ``src`` .
+
+    :param stream: Stream for the asynchronous version.
+
+.. seealso:: :ocv:func:`abs`
+
+
+
+gpu::sqr
+------------
+Computes a square value of each matrix element.
+
+.. ocv:function:: void gpu::sqr(const GpuMat& src, GpuMat& dst, Stream& stream = Stream::Null())
+
+    :param src: Source matrix. Supports ``CV_8U`` , ``CV_16U`` , ``CV_16S`` and ``CV_32F`` depth.
+
+    :param dst: Destination matrix with the same size and type as ``src`` .
+
+    :param stream: Stream for the asynchronous version.
+
+
+
+gpu::sqrt
+------------
+Computes a square root of each matrix element.
+
+.. ocv:function:: void gpu::sqrt(const GpuMat& src, GpuMat& dst, Stream& stream = Stream::Null())
+
+    :param src: Source matrix. Supports ``CV_8U`` , ``CV_16U`` , ``CV_16S`` and ``CV_32F`` depth.
+
+    :param dst: Destination matrix with the same size and type as ``src`` .
+
+    :param stream: Stream for the asynchronous version.
+
+.. seealso:: :ocv:func:`sqrt`
+
+
+
+gpu::exp
+------------
+Computes an exponent of each matrix element.
+
+.. ocv:function:: void gpu::exp( const GpuMat& a, GpuMat& b, Stream& stream=Stream::Null() )
+
+    :param a: Source matrix. Supports ``CV_8U`` , ``CV_16U`` , ``CV_16S`` and ``CV_32F`` depth.
+
+    :param b: Destination matrix with the same size and type as ``a`` .
+
+    :param stream: Stream for the asynchronous version.
+
+.. seealso:: :ocv:func:`exp`
+
+
+
+gpu::log
+------------
+Computes a natural logarithm of absolute value of each matrix element.
+
+.. ocv:function:: void gpu::log( const GpuMat& a, GpuMat& b, Stream& stream=Stream::Null() )
+
+    :param a: Source matrix. Supports ``CV_8U`` , ``CV_16U`` , ``CV_16S`` and ``CV_32F`` depth.
+
+    :param b: Destination matrix with the same size and type as ``a`` .
+
+    :param stream: Stream for the asynchronous version.
+
+.. seealso:: :ocv:func:`log`
+
+
+
+gpu::pow
+------------
+Raises every matrix element to a power.
+
+.. ocv:function:: void gpu::pow(const GpuMat& src, double power, GpuMat& dst, Stream& stream = Stream::Null())
+
+    :param src: Source matrix. Supports all type, except ``CV_64F`` depth.
+
+    :param power: Exponent of power.
+
+    :param dst: Destination matrix with the same size and type as ``src`` .
+
+    :param stream: Stream for the asynchronous version.
+
+The function ``pow`` raises every element of the input matrix to ``p`` :
+
+.. math::
+
+    \texttt{dst} (I) =  \fork{\texttt{src}(I)^p}{if \texttt{p} is integer}{|\texttt{src}(I)|^p}{otherwise}
+
+.. seealso:: :ocv:func:`pow`
+
+
+
+gpu::absdiff
+----------------
+Computes per-element absolute difference of two matrices (or of a matrix and scalar).
+
+.. ocv:function:: void gpu::absdiff( const GpuMat& a, const GpuMat& b, GpuMat& c, Stream& stream=Stream::Null() )
+
+.. ocv:function:: void gpu::absdiff( const GpuMat& a, const Scalar& s, GpuMat& c, Stream& stream=Stream::Null() )
+
+    :param a: First source matrix.
+
+    :param b: Second source matrix to be added to ``a`` .
+
+    :param s: A scalar to be added to ``a`` .
+
+    :param c: Destination matrix with the same size and type as ``a`` .
+
+    :param stream: Stream for the asynchronous version.
+
+.. seealso:: :ocv:func:`absdiff`
+
+
+
+gpu::compare
+----------------
+Compares elements of two matrices.
+
+.. ocv:function:: void gpu::compare( const GpuMat& a, const GpuMat& b, GpuMat& c, int cmpop, Stream& stream=Stream::Null() )
+
+.. ocv:function:: void gpu::compare(const GpuMat& a, Scalar sc, GpuMat& c, int cmpop, Stream& stream = Stream::Null())
+
+    :param a: First source matrix.
+
+    :param b: Second source matrix with the same size and type as ``a`` .
+
+    :param sc: A scalar to be compared with ``a`` .
+
+    :param c: Destination matrix with the same size as ``a`` and the ``CV_8UC1`` type.
+
+    :param cmpop: Flag specifying the relation between the elements to be checked:
+
+            * **CMP_EQ:** ``a(.) == b(.)``
+            * **CMP_GT:** ``a(.) < b(.)``
+            * **CMP_GE:** ``a(.) <= b(.)``
+            * **CMP_LT:** ``a(.) < b(.)``
+            * **CMP_LE:** ``a(.) <= b(.)``
+            * **CMP_NE:** ``a(.) != b(.)``
+
+    :param stream: Stream for the asynchronous version.
+
+.. seealso:: :ocv:func:`compare`
+
+
+
+gpu::bitwise_not
+--------------------
+Performs a per-element bitwise inversion.
+
+.. ocv:function:: void gpu::bitwise_not(const GpuMat& src, GpuMat& dst, const GpuMat& mask=GpuMat(), Stream& stream = Stream::Null())
+
+    :param src: Source matrix.
+
+    :param dst: Destination matrix with the same size and type as ``src`` .
+
+    :param mask: Optional operation mask. 8-bit single channel image.
+
+    :param stream: Stream for the asynchronous version.
+
+
+
+gpu::bitwise_or
+-------------------
+Performs a per-element bitwise disjunction of two matrices or of matrix and scalar.
+
+.. ocv:function:: void gpu::bitwise_or(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat& mask=GpuMat(), Stream& stream = Stream::Null())
+.. ocv:function:: void gpu::bitwise_or(const GpuMat& src1, const Scalar& sc, GpuMat& dst, Stream& stream = Stream::Null())
+
+    :param src1: First source matrix.
+
+    :param src2: Second source matrix with the same size and type as ``src1`` .
+
+    :param dst: Destination matrix with the same size and type as ``src1`` .
+
+    :param mask: Optional operation mask. 8-bit single channel image.
+
+    :param stream: Stream for the asynchronous version.
+
+
+
+gpu::bitwise_and
+--------------------
+Performs a per-element bitwise conjunction of two matrices or of matrix and scalar.
+
+.. ocv:function:: void gpu::bitwise_and(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat& mask=GpuMat(), Stream& stream = Stream::Null())
+.. ocv:function:: void gpu::bitwise_and(const GpuMat& src1, const Scalar& sc, GpuMat& dst, Stream& stream = Stream::Null())
+
+    :param src1: First source matrix.
+
+    :param src2: Second source matrix with the same size and type as ``src1`` .
+
+    :param dst: Destination matrix with the same size and type as ``src1`` .
+
+    :param mask: Optional operation mask. 8-bit single channel image.
+
+    :param stream: Stream for the asynchronous version.
+
+
+
+gpu::bitwise_xor
+--------------------
+Performs a per-element bitwise ``exclusive or`` operation of two matrices of matrix and scalar.
+
+.. ocv:function:: void gpu::bitwise_xor(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat& mask=GpuMat(), Stream& stream = Stream::Null())
+.. ocv:function:: void gpu::bitwise_xor(const GpuMat& src1, const Scalar& sc, GpuMat& dst, Stream& stream = Stream::Null())
+
+    :param src1: First source matrix.
+
+    :param src2: Second source matrix with the same size and type as ``src1`` .
+
+    :param dst: Destination matrix with the same size and type as ``src1`` .
+
+    :param mask: Optional operation mask. 8-bit single channel image.
+
+    :param stream: Stream for the asynchronous version.
+
+
+
+gpu::rshift
+--------------------
+Performs pixel by pixel right shift of an image by a constant value.
+
+.. ocv:function:: void gpu::rshift( const GpuMat& src, Scalar_<int> sc, GpuMat& dst, Stream& stream=Stream::Null() )
+
+    :param src: Source matrix. Supports 1, 3 and 4 channels images with integers elements.
+
+    :param sc: Constant values, one per channel.
+
+    :param dst: Destination matrix with the same size and type as ``src`` .
+
+    :param stream: Stream for the asynchronous version.
+
+
+
+gpu::lshift
+--------------------
+Performs pixel by pixel right left of an image by a constant value.
+
+.. ocv:function:: void gpu::lshift( const GpuMat& src, Scalar_<int> sc, GpuMat& dst, Stream& stream=Stream::Null() )
+
+    :param src: Source matrix. Supports 1, 3 and 4 channels images with ``CV_8U`` , ``CV_16U`` or ``CV_32S`` depth.
+
+    :param sc: Constant values, one per channel.
+
+    :param dst: Destination matrix with the same size and type as ``src`` .
+
+    :param stream: Stream for the asynchronous version.
+
+
+
+gpu::min
+------------
+Computes the per-element minimum of two matrices (or a matrix and a scalar).
+
+.. ocv:function:: void gpu::min(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, Stream& stream = Stream::Null())
+
+.. ocv:function:: void gpu::min(const GpuMat& src1, double src2, GpuMat& dst, Stream& stream = Stream::Null())
+
+    :param src1: First source matrix.
+
+    :param src2: Second source matrix or a scalar to compare ``src1`` elements with.
+
+    :param dst: Destination matrix with the same size and type as ``src1`` .
+
+    :param stream: Stream for the asynchronous version.
+
+.. seealso:: :ocv:func:`min`
+
+
+
+gpu::max
+------------
+Computes the per-element maximum of two matrices (or a matrix and a scalar).
+
+.. ocv:function:: void gpu::max(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, Stream& stream = Stream::Null())
+
+.. ocv:function:: void gpu::max(const GpuMat& src1, double src2, GpuMat& dst, Stream& stream = Stream::Null())
+
+    :param src1: First source matrix.
+
+    :param src2: Second source matrix or a scalar to compare ``src1`` elements with.
+
+    :param dst: Destination matrix with the same size and type as ``src1`` .
+
+    :param stream: Stream for the asynchronous version.
+
+.. seealso:: :ocv:func:`max`
+
+
+
+gpu::threshold
+------------------
+Applies a fixed-level threshold to each array element.
+
+.. ocv:function:: double gpu::threshold(const GpuMat& src, GpuMat& dst, double thresh, double maxval, int type, Stream& stream = Stream::Null())
+
+    :param src: Source array (single-channel).
+
+    :param dst: Destination array with the same size and type as  ``src`` .
+
+    :param thresh: Threshold value.
+
+    :param maxval: Maximum value to use with  ``THRESH_BINARY`` and  ``THRESH_BINARY_INV`` threshold types.
+
+    :param type: Threshold type. For details, see  :ocv:func:`threshold` . The ``THRESH_OTSU`` threshold type is not supported.
+
+    :param stream: Stream for the asynchronous version.
+
+.. seealso:: :ocv:func:`threshold`

modules/gpuarithm/include/opencv2/gpuarithm.hpp

@@ -0,0 +1,279 @@
+/*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 __OPENCV_GPUARITHM_HPP__
+#define __OPENCV_GPUARITHM_HPP__
+
+#include "opencv2/core/gpumat.hpp"
+
+namespace cv { namespace gpu {
+
+//! adds one matrix to another (c = a + b)
+CV_EXPORTS void add(const GpuMat& a, const GpuMat& b, GpuMat& c, const GpuMat& mask = GpuMat(), int dtype = -1, Stream& stream = Stream::Null());
+//! adds scalar to a matrix (c = a + s)
+CV_EXPORTS void add(const GpuMat& a, const Scalar& sc, GpuMat& c, const GpuMat& mask = GpuMat(), int dtype = -1, Stream& stream = Stream::Null());
+
+//! subtracts one matrix from another (c = a - b)
+CV_EXPORTS void subtract(const GpuMat& a, const GpuMat& b, GpuMat& c, const GpuMat& mask = GpuMat(), int dtype = -1, Stream& stream = Stream::Null());
+//! subtracts scalar from a matrix (c = a - s)
+CV_EXPORTS void subtract(const GpuMat& a, const Scalar& sc, GpuMat& c, const GpuMat& mask = GpuMat(), int dtype = -1, Stream& stream = Stream::Null());
+
+//! computes element-wise weighted product of the two arrays (c = scale * a * b)
+CV_EXPORTS void multiply(const GpuMat& a, const GpuMat& b, GpuMat& c, double scale = 1, int dtype = -1, Stream& stream = Stream::Null());
+//! weighted multiplies matrix to a scalar (c = scale * a * s)
+CV_EXPORTS void multiply(const GpuMat& a, const Scalar& sc, GpuMat& c, double scale = 1, int dtype = -1, Stream& stream = Stream::Null());
+
+//! computes element-wise weighted quotient of the two arrays (c = a / b)
+CV_EXPORTS void divide(const GpuMat& a, const GpuMat& b, GpuMat& c, double scale = 1, int dtype = -1, Stream& stream = Stream::Null());
+//! computes element-wise weighted quotient of matrix and scalar (c = a / s)
+CV_EXPORTS void divide(const GpuMat& a, const Scalar& sc, GpuMat& c, double scale = 1, int dtype = -1, Stream& stream = Stream::Null());
+//! computes element-wise weighted reciprocal of an array (dst = scale/src2)
+CV_EXPORTS void divide(double scale, const GpuMat& b, GpuMat& c, int dtype = -1, Stream& stream = Stream::Null());
+
+//! computes the weighted sum of two arrays (dst = alpha*src1 + beta*src2 + gamma)
+CV_EXPORTS void addWeighted(const GpuMat& src1, double alpha, const GpuMat& src2, double beta, double gamma, GpuMat& dst,
+                            int dtype = -1, Stream& stream = Stream::Null());
+
+//! adds scaled array to another one (dst = alpha*src1 + src2)
+static inline void scaleAdd(const GpuMat& src1, double alpha, const GpuMat& src2, GpuMat& dst, Stream& stream = Stream::Null())
+{
+    addWeighted(src1, alpha, src2, 1.0, 0.0, dst, -1, stream);
+}
+
+//! computes element-wise absolute difference of two arrays (c = abs(a - b))
+CV_EXPORTS void absdiff(const GpuMat& a, const GpuMat& b, GpuMat& c, Stream& stream = Stream::Null());
+//! computes element-wise absolute difference of array and scalar (c = abs(a - s))
+CV_EXPORTS void absdiff(const GpuMat& a, const Scalar& s, GpuMat& c, Stream& stream = Stream::Null());
+
+//! computes absolute value of each matrix element
+//! supports CV_16S and CV_32F depth
+CV_EXPORTS void abs(const GpuMat& src, GpuMat& dst, Stream& stream = Stream::Null());
+
+//! computes square of each pixel in an image
+//! supports CV_8U, CV_16U, CV_16S and CV_32F depth
+CV_EXPORTS void sqr(const GpuMat& src, GpuMat& dst, Stream& stream = Stream::Null());
+
+//! computes square root of each pixel in an image
+//! supports CV_8U, CV_16U, CV_16S and CV_32F depth
+CV_EXPORTS void sqrt(const GpuMat& src, GpuMat& dst, Stream& stream = Stream::Null());
+
+//! computes exponent of each matrix element (b = e**a)
+//! supports CV_8U, CV_16U, CV_16S and CV_32F depth
+CV_EXPORTS void exp(const GpuMat& a, GpuMat& b, Stream& stream = Stream::Null());
+
+//! computes natural logarithm of absolute value of each matrix element: b = log(abs(a))
+//! supports CV_8U, CV_16U, CV_16S and CV_32F depth
+CV_EXPORTS void log(const GpuMat& a, GpuMat& b, Stream& stream = Stream::Null());
+
+//! computes power of each matrix element:
+//    (dst(i,j) = pow(     src(i,j) , power), if src.type() is integer
+//    (dst(i,j) = pow(fabs(src(i,j)), power), otherwise
+//! supports all, except depth == CV_64F
+CV_EXPORTS void pow(const GpuMat& src, double power, GpuMat& dst, Stream& stream = Stream::Null());
+
+//! compares elements of two arrays (c = a <cmpop> b)
+CV_EXPORTS void compare(const GpuMat& a, const GpuMat& b, GpuMat& c, int cmpop, Stream& stream = Stream::Null());
+CV_EXPORTS void compare(const GpuMat& a, Scalar sc, GpuMat& c, int cmpop, Stream& stream = Stream::Null());
+
+//! performs per-elements bit-wise inversion
+CV_EXPORTS void bitwise_not(const GpuMat& src, GpuMat& dst, const GpuMat& mask=GpuMat(), Stream& stream = Stream::Null());
+
+//! calculates per-element bit-wise disjunction of two arrays
+CV_EXPORTS void bitwise_or(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat& mask=GpuMat(), Stream& stream = Stream::Null());
+//! calculates per-element bit-wise disjunction of array and scalar
+//! supports 1, 3 and 4 channels images with CV_8U, CV_16U or CV_32S depth
+CV_EXPORTS void bitwise_or(const GpuMat& src1, const Scalar& sc, GpuMat& dst, Stream& stream = Stream::Null());
+
+//! calculates per-element bit-wise conjunction of two arrays
+CV_EXPORTS void bitwise_and(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat& mask=GpuMat(), Stream& stream = Stream::Null());
+//! calculates per-element bit-wise conjunction of array and scalar
+//! supports 1, 3 and 4 channels images with CV_8U, CV_16U or CV_32S depth
+CV_EXPORTS void bitwise_and(const GpuMat& src1, const Scalar& sc, GpuMat& dst, Stream& stream = Stream::Null());
+
+//! calculates per-element bit-wise "exclusive or" operation
+CV_EXPORTS void bitwise_xor(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat& mask=GpuMat(), Stream& stream = Stream::Null());
+//! calculates per-element bit-wise "exclusive or" of array and scalar
+//! supports 1, 3 and 4 channels images with CV_8U, CV_16U or CV_32S depth
+CV_EXPORTS void bitwise_xor(const GpuMat& src1, const Scalar& sc, GpuMat& dst, Stream& stream = Stream::Null());
+
+//! pixel by pixel right shift of an image by a constant value
+//! supports 1, 3 and 4 channels images with integers elements
+CV_EXPORTS void rshift(const GpuMat& src, Scalar_<int> sc, GpuMat& dst, Stream& stream = Stream::Null());
+
+//! pixel by pixel left shift of an image by a constant value
+//! supports 1, 3 and 4 channels images with CV_8U, CV_16U or CV_32S depth
+CV_EXPORTS void lshift(const GpuMat& src, Scalar_<int> sc, GpuMat& dst, Stream& stream = Stream::Null());
+
+//! computes per-element minimum of two arrays (dst = min(src1, src2))
+CV_EXPORTS void min(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, Stream& stream = Stream::Null());
+
+//! computes per-element minimum of array and scalar (dst = min(src1, src2))
+CV_EXPORTS void min(const GpuMat& src1, double src2, GpuMat& dst, Stream& stream = Stream::Null());
+
+//! computes per-element maximum of two arrays (dst = max(src1, src2))
+CV_EXPORTS void max(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, Stream& stream = Stream::Null());
+
+//! computes per-element maximum of array and scalar (dst = max(src1, src2))
+CV_EXPORTS void max(const GpuMat& src1, double src2, GpuMat& dst, Stream& stream = Stream::Null());
+
+//! implements generalized matrix product algorithm GEMM from BLAS
+CV_EXPORTS void gemm(const GpuMat& src1, const GpuMat& src2, double alpha,
+    const GpuMat& src3, double beta, GpuMat& dst, int flags = 0, Stream& stream = Stream::Null());
+
+//! transposes the matrix
+//! supports matrix with element size = 1, 4 and 8 bytes (CV_8UC1, CV_8UC4, CV_16UC2, CV_32FC1, etc)
+CV_EXPORTS void transpose(const GpuMat& src1, GpuMat& dst, Stream& stream = Stream::Null());
+
+//! reverses the order of the rows, columns or both in a matrix
+//! supports 1, 3 and 4 channels images with CV_8U, CV_16U, CV_32S or CV_32F depth
+CV_EXPORTS void flip(const GpuMat& a, GpuMat& b, int flipCode, Stream& stream = Stream::Null());
+
+//! transforms 8-bit unsigned integers using lookup table: dst(i)=lut(src(i))
+//! destination array will have the depth type as lut and the same channels number as source
+//! supports CV_8UC1, CV_8UC3 types
+CV_EXPORTS void LUT(const GpuMat& src, const Mat& lut, GpuMat& dst, Stream& stream = Stream::Null());
+
+//! makes multi-channel array out of several single-channel arrays
+CV_EXPORTS void merge(const GpuMat* src, size_t n, GpuMat& dst, Stream& stream = Stream::Null());
+
+//! makes multi-channel array out of several single-channel arrays
+CV_EXPORTS void merge(const std::vector<GpuMat>& src, GpuMat& dst, Stream& stream = Stream::Null());
+
+//! copies each plane of a multi-channel array to a dedicated array
+CV_EXPORTS void split(const GpuMat& src, GpuMat* dst, Stream& stream = Stream::Null());
+
+//! copies each plane of a multi-channel array to a dedicated array
+CV_EXPORTS void split(const GpuMat& src, std::vector<GpuMat>& dst, Stream& stream = Stream::Null());
+
+//! computes magnitude of complex (x(i).re, x(i).im) vector
+//! supports only CV_32FC2 type
+CV_EXPORTS void magnitude(const GpuMat& xy, GpuMat& magnitude, Stream& stream = Stream::Null());
+
+//! computes squared magnitude of complex (x(i).re, x(i).im) vector
+//! supports only CV_32FC2 type
+CV_EXPORTS void magnitudeSqr(const GpuMat& xy, GpuMat& magnitude, Stream& stream = Stream::Null());
+
+//! computes magnitude of each (x(i), y(i)) vector
+//! supports only floating-point source
+CV_EXPORTS void magnitude(const GpuMat& x, const GpuMat& y, GpuMat& magnitude, Stream& stream = Stream::Null());
+
+//! computes squared magnitude of each (x(i), y(i)) vector
+//! supports only floating-point source
+CV_EXPORTS void magnitudeSqr(const GpuMat& x, const GpuMat& y, GpuMat& magnitude, Stream& stream = Stream::Null());
+
+//! computes angle (angle(i)) of each (x(i), y(i)) vector
+//! supports only floating-point source
+CV_EXPORTS void phase(const GpuMat& x, const GpuMat& y, GpuMat& angle, bool angleInDegrees = false, Stream& stream = Stream::Null());
+
+//! converts Cartesian coordinates to polar
+//! supports only floating-point source
+CV_EXPORTS void cartToPolar(const GpuMat& x, const GpuMat& y, GpuMat& magnitude, GpuMat& angle, bool angleInDegrees = false, Stream& stream = Stream::Null());
+
+//! converts polar coordinates to Cartesian
+//! supports only floating-point source
+CV_EXPORTS void polarToCart(const GpuMat& magnitude, const GpuMat& angle, GpuMat& x, GpuMat& y, bool angleInDegrees = false, Stream& stream = Stream::Null());
+
+//! scales and shifts array elements so that either the specified norm (alpha) or the minimum (alpha) and maximum (beta) array values get the specified values
+CV_EXPORTS void normalize(const GpuMat& src, GpuMat& dst, double alpha = 1, double beta = 0,
+                          int norm_type = NORM_L2, int dtype = -1, const GpuMat& mask = GpuMat());
+CV_EXPORTS void normalize(const GpuMat& src, GpuMat& dst, double a, double b,
+                          int norm_type, int dtype, const GpuMat& mask, GpuMat& norm_buf, GpuMat& cvt_buf);
+
+//! computes mean value and standard deviation of all or selected array elements
+//! supports only CV_8UC1 type
+CV_EXPORTS void meanStdDev(const GpuMat& mtx, Scalar& mean, Scalar& stddev);
+//! buffered version
+CV_EXPORTS void meanStdDev(const GpuMat& mtx, Scalar& mean, Scalar& stddev, GpuMat& buf);
+
+//! computes norm of array
+//! supports NORM_INF, NORM_L1, NORM_L2
+//! supports all matrices except 64F
+CV_EXPORTS double norm(const GpuMat& src1, int normType=NORM_L2);
+CV_EXPORTS double norm(const GpuMat& src1, int normType, GpuMat& buf);
+CV_EXPORTS double norm(const GpuMat& src1, int normType, const GpuMat& mask, GpuMat& buf);
+
+//! computes norm of the difference between two arrays
+//! supports NORM_INF, NORM_L1, NORM_L2
+//! supports only CV_8UC1 type
+CV_EXPORTS double norm(const GpuMat& src1, const GpuMat& src2, int normType=NORM_L2);
+
+//! computes sum of array elements
+//! supports only single channel images
+CV_EXPORTS Scalar sum(const GpuMat& src);
+CV_EXPORTS Scalar sum(const GpuMat& src, GpuMat& buf);
+CV_EXPORTS Scalar sum(const GpuMat& src, const GpuMat& mask, GpuMat& buf);
+
+//! computes sum of array elements absolute values
+//! supports only single channel images
+CV_EXPORTS Scalar absSum(const GpuMat& src);
+CV_EXPORTS Scalar absSum(const GpuMat& src, GpuMat& buf);
+CV_EXPORTS Scalar absSum(const GpuMat& src, const GpuMat& mask, GpuMat& buf);
+
+//! computes squared sum of array elements
+//! supports only single channel images
+CV_EXPORTS Scalar sqrSum(const GpuMat& src);
+CV_EXPORTS Scalar sqrSum(const GpuMat& src, GpuMat& buf);
+CV_EXPORTS Scalar sqrSum(const GpuMat& src, const GpuMat& mask, GpuMat& buf);
+
+//! finds global minimum and maximum array elements and returns their values
+CV_EXPORTS void minMax(const GpuMat& src, double* minVal, double* maxVal=0, const GpuMat& mask=GpuMat());
+CV_EXPORTS void minMax(const GpuMat& src, double* minVal, double* maxVal, const GpuMat& mask, GpuMat& buf);
+
+//! finds global minimum and maximum array elements and returns their values with locations
+CV_EXPORTS void minMaxLoc(const GpuMat& src, double* minVal, double* maxVal=0, Point* minLoc=0, Point* maxLoc=0,
+                          const GpuMat& mask=GpuMat());
+CV_EXPORTS void minMaxLoc(const GpuMat& src, double* minVal, double* maxVal, Point* minLoc, Point* maxLoc,
+                          const GpuMat& mask, GpuMat& valbuf, GpuMat& locbuf);
+
+//! counts non-zero array elements
+CV_EXPORTS int countNonZero(const GpuMat& src);
+CV_EXPORTS int countNonZero(const GpuMat& src, GpuMat& buf);
+
+//! reduces a matrix to a vector
+CV_EXPORTS void reduce(const GpuMat& mtx, GpuMat& vec, int dim, int reduceOp, int dtype = -1, Stream& stream = Stream::Null());
+
+//! applies fixed threshold to the image
+CV_EXPORTS double threshold(const GpuMat& src, GpuMat& dst, double thresh, double maxval, int type, Stream& stream = Stream::Null());
+
+}} // namespace cv { namespace gpu {
+
+#endif /* __OPENCV_GPUARITHM_HPP__ */

modules/gpuarithm/perf/perf_main.cpp

@@ -0,0 +1,47 @@
+/*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 "perf_precomp.hpp"
+
+using namespace perf;
+
+CV_PERF_TEST_MAIN(gpuarithm, printCudaInfo())

modules/gpuarithm/perf/perf_precomp.cpp

@@ -0,0 +1,43 @@
+/*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 "perf_precomp.hpp"

modules/gpuarithm/perf/perf_precomp.hpp

@@ -0,0 +1,64 @@
+/*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*/
+
+#ifdef __GNUC__
+#  pragma GCC diagnostic ignored "-Wmissing-declarations"
+#  if defined __clang__ || defined __APPLE__
+#    pragma GCC diagnostic ignored "-Wmissing-prototypes"
+#    pragma GCC diagnostic ignored "-Wextra"
+#  endif
+#endif
+
+#ifndef __OPENCV_PERF_PRECOMP_HPP__
+#define __OPENCV_PERF_PRECOMP_HPP__
+
+#include "opencv2/ts.hpp"
+#include "opencv2/ts/gpu_perf.hpp"
+
+#include "opencv2/core.hpp"
+#include "opencv2/gpuarithm.hpp"
+
+#ifdef GTEST_CREATE_SHARED_LIBRARY
+#error no modules except ts should have GTEST_CREATE_SHARED_LIBRARY defined
+#endif
+
+#endif

modules/gpuarithm/src/arithm.cpp

@@ -0,0 +1,611 @@
+/*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 "precomp.hpp"
+
+using namespace cv;
+using namespace cv::gpu;
+
+#if !defined (HAVE_CUDA) || defined (CUDA_DISABLER)
+
+void cv::gpu::gemm(const GpuMat&, const GpuMat&, double, const GpuMat&, double, GpuMat&, int, Stream&) { throw_no_cuda(); }
+void cv::gpu::transpose(const GpuMat&, GpuMat&, Stream&) { throw_no_cuda(); }
+void cv::gpu::flip(const GpuMat&, GpuMat&, int, Stream&) { throw_no_cuda(); }
+void cv::gpu::LUT(const GpuMat&, const Mat&, GpuMat&, Stream&) { throw_no_cuda(); }
+void cv::gpu::magnitude(const GpuMat&, GpuMat&, Stream&) { throw_no_cuda(); }
+void cv::gpu::magnitudeSqr(const GpuMat&, GpuMat&, Stream&) { throw_no_cuda(); }
+void cv::gpu::magnitude(const GpuMat&, const GpuMat&, GpuMat&, Stream&) { throw_no_cuda(); }
+void cv::gpu::magnitudeSqr(const GpuMat&, const GpuMat&, GpuMat&, Stream&) { throw_no_cuda(); }
+void cv::gpu::phase(const GpuMat&, const GpuMat&, GpuMat&, bool, Stream&) { throw_no_cuda(); }
+void cv::gpu::cartToPolar(const GpuMat&, const GpuMat&, GpuMat&, GpuMat&, bool, Stream&) { throw_no_cuda(); }
+void cv::gpu::polarToCart(const GpuMat&, const GpuMat&, GpuMat&, GpuMat&, bool, Stream&) { throw_no_cuda(); }
+void cv::gpu::normalize(const GpuMat&, GpuMat&, double, double, int, int, const GpuMat&) { throw_no_cuda(); }
+void cv::gpu::normalize(const GpuMat&, GpuMat&, double, double, int, int, const GpuMat&, GpuMat&, GpuMat&) { throw_no_cuda(); }
+
+#else /* !defined (HAVE_CUDA) */
+
+////////////////////////////////////////////////////////////////////////
+// gemm
+
+#ifdef HAVE_CUBLAS
+
+namespace
+{
+    #define error_entry(entry)  { entry, #entry }
+
+    struct ErrorEntry
+    {
+        int code;
+        const char* str;
+    };
+
+    struct ErrorEntryComparer
+    {
+        int code;
+        ErrorEntryComparer(int code_) : code(code_) {}
+        bool operator()(const ErrorEntry& e) const { return e.code == code; }
+    };
+
+    const ErrorEntry cublas_errors[] =
+    {
+        error_entry( CUBLAS_STATUS_SUCCESS ),
+        error_entry( CUBLAS_STATUS_NOT_INITIALIZED ),
+        error_entry( CUBLAS_STATUS_ALLOC_FAILED ),
+        error_entry( CUBLAS_STATUS_INVALID_VALUE ),
+        error_entry( CUBLAS_STATUS_ARCH_MISMATCH ),
+        error_entry( CUBLAS_STATUS_MAPPING_ERROR ),
+        error_entry( CUBLAS_STATUS_EXECUTION_FAILED ),
+        error_entry( CUBLAS_STATUS_INTERNAL_ERROR )
+    };
+
+    const size_t cublas_error_num = sizeof(cublas_errors) / sizeof(cublas_errors[0]);
+
+    static inline void ___cublasSafeCall(cublasStatus_t err, const char* file, const int line, const char* func)
+    {
+        if (CUBLAS_STATUS_SUCCESS != err)
+        {
+            size_t idx = std::find_if(cublas_errors, cublas_errors + cublas_error_num, ErrorEntryComparer(err)) - cublas_errors;
+
+            const char* msg = (idx != cublas_error_num) ? cublas_errors[idx].str : "Unknown error code";
+            String str = cv::format("%s [Code = %d]", msg, err);
+
+            cv::error(cv::Error::GpuApiCallError, str, func, file, line);
+        }
+    }
+}
+
+#if defined(__GNUC__)
+    #define cublasSafeCall(expr)  ___cublasSafeCall(expr, __FILE__, __LINE__, __func__)
+#else /* defined(__CUDACC__) || defined(__MSVC__) */
+    #define cublasSafeCall(expr)  ___cublasSafeCall(expr, __FILE__, __LINE__, "")
+#endif
+
+#endif
+
+void cv::gpu::gemm(const GpuMat& src1, const GpuMat& src2, double alpha, const GpuMat& src3, double beta, GpuMat& dst, int flags, Stream& stream)
+{
+#ifndef HAVE_CUBLAS
+    (void)src1;
+    (void)src2;
+    (void)alpha;
+    (void)src3;
+    (void)beta;
+    (void)dst;
+    (void)flags;
+    (void)stream;
+    CV_Error(cv::Error::StsNotImplemented, "The library was build without CUBLAS");
+#else
+    // CUBLAS works with column-major matrices
+
+    CV_Assert(src1.type() == CV_32FC1 || src1.type() == CV_32FC2 || src1.type() == CV_64FC1 || src1.type() == CV_64FC2);
+    CV_Assert(src2.type() == src1.type() && (src3.empty() || src3.type() == src1.type()));
+
+    if (src1.depth() == CV_64F)
+    {
+        if (!deviceSupports(NATIVE_DOUBLE))
+            CV_Error(cv::Error::StsUnsupportedFormat, "The device doesn't support double");
+    }
+
+    bool tr1 = (flags & GEMM_1_T) != 0;
+    bool tr2 = (flags & GEMM_2_T) != 0;
+    bool tr3 = (flags & GEMM_3_T) != 0;
+
+    if (src1.type() == CV_64FC2)
+    {
+        if (tr1 || tr2 || tr3)
+            CV_Error(cv::Error::StsNotImplemented, "transpose operation doesn't implemented for CV_64FC2 type");
+    }
+
+    Size src1Size = tr1 ? Size(src1.rows, src1.cols) : src1.size();
+    Size src2Size = tr2 ? Size(src2.rows, src2.cols) : src2.size();
+    Size src3Size = tr3 ? Size(src3.rows, src3.cols) : src3.size();
+    Size dstSize(src2Size.width, src1Size.height);
+
+    CV_Assert(src1Size.width == src2Size.height);
+    CV_Assert(src3.empty() || src3Size == dstSize);
+
+    dst.create(dstSize, src1.type());
+
+    if (beta != 0)
+    {
+        if (src3.empty())
+        {
+            if (stream)
+                stream.enqueueMemSet(dst, Scalar::all(0));
+            else
+                dst.setTo(Scalar::all(0));
+        }
+        else
+        {
+            if (tr3)
+            {
+                transpose(src3, dst, stream);
+            }
+            else
+            {
+                if (stream)
+                    stream.enqueueCopy(src3, dst);
+                else
+                    src3.copyTo(dst);
+            }
+        }
+    }
+
+    cublasHandle_t handle;
+    cublasSafeCall( cublasCreate_v2(&handle) );
+
+    cublasSafeCall( cublasSetStream_v2(handle, StreamAccessor::getStream(stream)) );
+
+    cublasSafeCall( cublasSetPointerMode_v2(handle, CUBLAS_POINTER_MODE_HOST) );
+
+    const float alphaf = static_cast<float>(alpha);
+    const float betaf = static_cast<float>(beta);
+
+    const cuComplex alphacf = make_cuComplex(alphaf, 0);
+    const cuComplex betacf = make_cuComplex(betaf, 0);
+
+    const cuDoubleComplex alphac = make_cuDoubleComplex(alpha, 0);
+    const cuDoubleComplex betac = make_cuDoubleComplex(beta, 0);
+
+    cublasOperation_t transa = tr2 ? CUBLAS_OP_T : CUBLAS_OP_N;
+    cublasOperation_t transb = tr1 ? CUBLAS_OP_T : CUBLAS_OP_N;
+
+    switch (src1.type())
+    {
+    case CV_32FC1:
+        cublasSafeCall( cublasSgemm_v2(handle, transa, transb, tr2 ? src2.rows : src2.cols, tr1 ? src1.cols : src1.rows, tr2 ? src2.cols : src2.rows,
+            &alphaf,
+            src2.ptr<float>(), static_cast<int>(src2.step / sizeof(float)),
+            src1.ptr<float>(), static_cast<int>(src1.step / sizeof(float)),
+            &betaf,
+            dst.ptr<float>(), static_cast<int>(dst.step / sizeof(float))) );
+        break;
+
+    case CV_64FC1:
+        cublasSafeCall( cublasDgemm_v2(handle, transa, transb, tr2 ? src2.rows : src2.cols, tr1 ? src1.cols : src1.rows, tr2 ? src2.cols : src2.rows,
+            &alpha,
+            src2.ptr<double>(), static_cast<int>(src2.step / sizeof(double)),
+            src1.ptr<double>(), static_cast<int>(src1.step / sizeof(double)),
+            &beta,
+            dst.ptr<double>(), static_cast<int>(dst.step / sizeof(double))) );
+        break;
+
+    case CV_32FC2:
+        cublasSafeCall( cublasCgemm_v2(handle, transa, transb, tr2 ? src2.rows : src2.cols, tr1 ? src1.cols : src1.rows, tr2 ? src2.cols : src2.rows,
+            &alphacf,
+            src2.ptr<cuComplex>(), static_cast<int>(src2.step / sizeof(cuComplex)),
+            src1.ptr<cuComplex>(), static_cast<int>(src1.step / sizeof(cuComplex)),
+            &betacf,
+            dst.ptr<cuComplex>(), static_cast<int>(dst.step / sizeof(cuComplex))) );
+        break;
+
+    case CV_64FC2:
+        cublasSafeCall( cublasZgemm_v2(handle, transa, transb, tr2 ? src2.rows : src2.cols, tr1 ? src1.cols : src1.rows, tr2 ? src2.cols : src2.rows,
+            &alphac,
+            src2.ptr<cuDoubleComplex>(), static_cast<int>(src2.step / sizeof(cuDoubleComplex)),
+            src1.ptr<cuDoubleComplex>(), static_cast<int>(src1.step / sizeof(cuDoubleComplex)),
+            &betac,
+            dst.ptr<cuDoubleComplex>(), static_cast<int>(dst.step / sizeof(cuDoubleComplex))) );
+        break;
+    }
+
+    cublasSafeCall( cublasDestroy_v2(handle) );
+#endif
+}
+
+////////////////////////////////////////////////////////////////////////
+// transpose
+
+namespace arithm
+{
+    template <typename T> void transpose(PtrStepSz<T> src, PtrStepSz<T> dst, cudaStream_t stream);
+}
+
+void cv::gpu::transpose(const GpuMat& src, GpuMat& dst, Stream& s)
+{
+    CV_Assert( src.elemSize() == 1 || src.elemSize() == 4 || src.elemSize() == 8 );
+
+    dst.create( src.cols, src.rows, src.type() );
+
+    cudaStream_t stream = StreamAccessor::getStream(s);
+
+    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
+
+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; };
+    template<> struct NppTypeTraits<CV_32S> { typedef Npp32s npp_t; };
+    template<> struct NppTypeTraits<CV_32F> { typedef Npp32f npp_t; };
+    template<> struct NppTypeTraits<CV_64F> { typedef Npp64f npp_t; };
+
+    template <int DEPTH> struct NppMirrorFunc
+    {
+        typedef typename NppTypeTraits<DEPTH>::npp_t npp_t;
+
+        typedef NppStatus (*func_t)(const npp_t* pSrc, int nSrcStep, npp_t* pDst, int nDstStep, NppiSize oROI, NppiAxis flip);
+    };
+
+    template <int DEPTH, typename NppMirrorFunc<DEPTH>::func_t func> struct NppMirror
+    {
+        typedef typename NppMirrorFunc<DEPTH>::npp_t npp_t;
+
+        static void call(const GpuMat& src, GpuMat& dst, int flipCode, cudaStream_t stream)
+        {
+            NppStreamHandler h(stream);
+
+            NppiSize sz;
+            sz.width  = src.cols;
+            sz.height = src.rows;
+
+            nppSafeCall( func(src.ptr<npp_t>(), static_cast<int>(src.step),
+                dst.ptr<npp_t>(), static_cast<int>(dst.step), sz,
+                (flipCode == 0 ? NPP_HORIZONTAL_AXIS : (flipCode > 0 ? NPP_VERTICAL_AXIS : NPP_BOTH_AXIS))) );
+
+            if (stream == 0)
+                cudaSafeCall( cudaDeviceSynchronize() );
+        }
+    };
+}
+
+void cv::gpu::flip(const GpuMat& src, GpuMat& dst, int flipCode, Stream& stream)
+{
+    typedef void (*func_t)(const GpuMat& src, GpuMat& dst, int flipCode, cudaStream_t stream);
+    static const func_t funcs[6][4] =
+    {
+        {NppMirror<CV_8U, nppiMirror_8u_C1R>::call, 0, NppMirror<CV_8U, nppiMirror_8u_C3R>::call, NppMirror<CV_8U, nppiMirror_8u_C4R>::call},
+        {0,0,0,0},
+        {NppMirror<CV_16U, nppiMirror_16u_C1R>::call, 0, NppMirror<CV_16U, nppiMirror_16u_C3R>::call, NppMirror<CV_16U, nppiMirror_16u_C4R>::call},
+        {0,0,0,0},
+        {NppMirror<CV_32S, nppiMirror_32s_C1R>::call, 0, NppMirror<CV_32S, nppiMirror_32s_C3R>::call, NppMirror<CV_32S, nppiMirror_32s_C4R>::call},
+        {NppMirror<CV_32F, nppiMirror_32f_C1R>::call, 0, NppMirror<CV_32F, nppiMirror_32f_C3R>::call, NppMirror<CV_32F, nppiMirror_32f_C4R>::call}
+    };
+
+    CV_Assert(src.depth() == CV_8U || src.depth() == CV_16U || src.depth() == CV_32S || src.depth() == CV_32F);
+    CV_Assert(src.channels() == 1 || src.channels() == 3 || src.channels() == 4);
+
+    dst.create(src.size(), src.type());
+
+    funcs[src.depth()][src.channels() - 1](src, dst, flipCode, StreamAccessor::getStream(stream));
+}
+
+////////////////////////////////////////////////////////////////////////
+// LUT
+
+void cv::gpu::LUT(const GpuMat& src, const Mat& lut, GpuMat& dst, Stream& s)
+{
+    const int cn = src.channels();
+
+    CV_Assert( src.type() == CV_8UC1 || src.type() == CV_8UC3 );
+    CV_Assert( lut.depth() == CV_8U );
+    CV_Assert( lut.channels() == 1 || lut.channels() == cn );
+    CV_Assert( lut.rows * lut.cols == 256 && lut.isContinuous() );
+
+    dst.create(src.size(), CV_MAKE_TYPE(lut.depth(), cn));
+
+    NppiSize sz;
+    sz.height = src.rows;
+    sz.width = src.cols;
+
+    Mat nppLut;
+    lut.convertTo(nppLut, CV_32S);
+
+    int nValues3[] = {256, 256, 256};
+
+    Npp32s pLevels[256];
+    for (int i = 0; i < 256; ++i)
+        pLevels[i] = i;
+
+    const Npp32s* pLevels3[3];
+
+#if (CUDA_VERSION <= 4020)
+    pLevels3[0] = pLevels3[1] = pLevels3[2] = pLevels;
+#else
+    GpuMat d_pLevels;
+    d_pLevels.upload(Mat(1, 256, CV_32S, pLevels));
+    pLevels3[0] = pLevels3[1] = pLevels3[2] = d_pLevels.ptr<Npp32s>();
+#endif
+
+    cudaStream_t stream = StreamAccessor::getStream(s);
+    NppStreamHandler h(stream);
+
+    if (src.type() == CV_8UC1)
+    {
+#if (CUDA_VERSION <= 4020)
+        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
+        GpuMat d_nppLut(Mat(1, 256, CV_32S, nppLut.data));
+        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) );
+#endif
+    }
+    else
+    {
+        const Npp32s* pValues3[3];
+
+        Mat nppLut3[3];
+        if (nppLut.channels() == 1)
+        {
+#if (CUDA_VERSION <= 4020)
+            pValues3[0] = pValues3[1] = pValues3[2] = nppLut.ptr<Npp32s>();
+#else
+            GpuMat d_nppLut(Mat(1, 256, CV_32S, nppLut.data));
+            pValues3[0] = pValues3[1] = pValues3[2] = d_nppLut.ptr<Npp32s>();
+#endif
+        }
+        else
+        {
+            cv::split(nppLut, nppLut3);
+
+#if (CUDA_VERSION <= 4020)
+            pValues3[0] = nppLut3[0].ptr<Npp32s>();
+            pValues3[1] = nppLut3[1].ptr<Npp32s>();
+            pValues3[2] = nppLut3[2].ptr<Npp32s>();
+#else
+            GpuMat d_nppLut0(Mat(1, 256, CV_32S, nppLut3[0].data));
+            GpuMat d_nppLut1(Mat(1, 256, CV_32S, nppLut3[1].data));
+            GpuMat d_nppLut2(Mat(1, 256, CV_32S, nppLut3[2].data));
+
+            pValues3[0] = d_nppLut0.ptr<Npp32s>();
+            pValues3[1] = d_nppLut1.ptr<Npp32s>();
+            pValues3[2] = d_nppLut2.ptr<Npp32s>();
+#endif
+        }
+
+        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() );
+}
+
+////////////////////////////////////////////////////////////////////////
+// NPP magnitide
+
+namespace
+{
+    typedef NppStatus (*nppMagnitude_t)(const Npp32fc* pSrc, int nSrcStep, Npp32f* pDst, int nDstStep, NppiSize oSizeROI);
+
+    inline void npp_magnitude(const GpuMat& src, GpuMat& dst, nppMagnitude_t func, cudaStream_t stream)
+    {
+        CV_Assert(src.type() == CV_32FC2);
+
+        dst.create(src.size(), CV_32FC1);
+
+        NppiSize sz;
+        sz.width = src.cols;
+        sz.height = src.rows;
+
+        NppStreamHandler h(stream);
+
+        nppSafeCall( func(src.ptr<Npp32fc>(), static_cast<int>(src.step), dst.ptr<Npp32f>(), static_cast<int>(dst.step), sz) );
+
+        if (stream == 0)
+            cudaSafeCall( cudaDeviceSynchronize() );
+    }
+}
+
+void cv::gpu::magnitude(const GpuMat& src, GpuMat& dst, Stream& stream)
+{
+    npp_magnitude(src, dst, nppiMagnitude_32fc32f_C1R, StreamAccessor::getStream(stream));
+}
+
+void cv::gpu::magnitudeSqr(const GpuMat& src, GpuMat& dst, Stream& stream)
+{
+    npp_magnitude(src, dst, nppiMagnitudeSqr_32fc32f_C1R, StreamAccessor::getStream(stream));
+}
+
+////////////////////////////////////////////////////////////////////////
+// Polar <-> Cart
+
+namespace cv { namespace gpu { namespace cudev
+{
+    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
+{
+    inline void cartToPolar_caller(const GpuMat& x, const GpuMat& y, GpuMat* mag, bool magSqr, GpuMat* angle, bool angleInDegrees, cudaStream_t stream)
+    {
+        using namespace ::cv::gpu::cudev::mathfunc;
+
+        CV_Assert(x.size() == y.size() && x.type() == y.type());
+        CV_Assert(x.depth() == CV_32F);
+
+        if (mag)
+            mag->create(x.size(), x.type());
+        if (angle)
+            angle->create(x.size(), x.type());
+
+        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);
+    }
+
+    inline void polarToCart_caller(const GpuMat& mag, const GpuMat& angle, GpuMat& x, GpuMat& y, bool angleInDegrees, cudaStream_t stream)
+    {
+        using namespace ::cv::gpu::cudev::mathfunc;
+
+        CV_Assert((mag.empty() || mag.size() == angle.size()) && mag.type() == angle.type());
+        CV_Assert(mag.depth() == CV_32F);
+
+        x.create(mag.size(), mag.type());
+        y.create(mag.size(), mag.type());
+
+        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::gpu::magnitude(const GpuMat& x, const GpuMat& y, GpuMat& dst, Stream& stream)
+{
+    cartToPolar_caller(x, y, &dst, false, 0, false, StreamAccessor::getStream(stream));
+}
+
+void cv::gpu::magnitudeSqr(const GpuMat& x, const GpuMat& y, GpuMat& dst, Stream& stream)
+{
+    cartToPolar_caller(x, y, &dst, true, 0, false, StreamAccessor::getStream(stream));
+}
+
+void cv::gpu::phase(const GpuMat& x, const GpuMat& y, GpuMat& angle, bool angleInDegrees, Stream& stream)
+{
+    cartToPolar_caller(x, y, 0, false, &angle, angleInDegrees, StreamAccessor::getStream(stream));
+}
+
+void cv::gpu::cartToPolar(const GpuMat& x, const GpuMat& y, GpuMat& mag, GpuMat& angle, bool angleInDegrees, Stream& stream)
+{
+    cartToPolar_caller(x, y, &mag, false, &angle, angleInDegrees, StreamAccessor::getStream(stream));
+}
+
+void cv::gpu::polarToCart(const GpuMat& magnitude, const GpuMat& angle, GpuMat& x, GpuMat& y, bool angleInDegrees, Stream& stream)
+{
+    polarToCart_caller(magnitude, angle, x, y, angleInDegrees, StreamAccessor::getStream(stream));
+}
+
+////////////////////////////////////////////////////////////////////////
+// normalize
+
+void cv::gpu::normalize(const GpuMat& src, GpuMat& dst, double a, double b, int norm_type, int dtype, const GpuMat& mask)
+{
+    GpuMat norm_buf;
+    GpuMat cvt_buf;
+    normalize(src, dst, a, b, norm_type, dtype, mask, norm_buf, cvt_buf);
+}
+
+void cv::gpu::normalize(const GpuMat& src, GpuMat& dst, double a, double b, int norm_type, int dtype, const GpuMat& mask, GpuMat& norm_buf, GpuMat& cvt_buf)
+{
+    double scale = 1, shift = 0;
+    if (norm_type == NORM_MINMAX)
+    {
+        double smin = 0, smax = 0;
+        double dmin = std::min(a, b), dmax = std::max(a, b);
+        minMax(src, &smin, &smax, mask, norm_buf);
+        scale = (dmax - dmin) * (smax - smin > std::numeric_limits<double>::epsilon() ? 1.0 / (smax - smin) : 0.0);
+        shift = dmin - smin * scale;
+    }
+    else if (norm_type == NORM_L2 || norm_type == NORM_L1 || norm_type == NORM_INF)
+    {
+        scale = norm(src, norm_type, mask, norm_buf);
+        scale = scale > std::numeric_limits<double>::epsilon() ? a / scale : 0.0;
+        shift = 0;
+    }
+    else
+    {
+        CV_Error(cv::Error::StsBadArg, "Unknown/unsupported norm type");
+    }
+
+    if (mask.empty())
+    {
+        src.convertTo(dst, dtype, scale, shift);
+    }
+    else
+    {
+        src.convertTo(cvt_buf, dtype, scale, shift);
+        cvt_buf.copyTo(dst, mask);
+    }
+}
+
+#endif /* !defined (HAVE_CUDA) */

modules/gpuarithm/src/cuda/absdiff_mat.cu

@@ -0,0 +1,147 @@
+/*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*/
+
+#if !defined CUDA_DISABLER
+
+#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 "arithm_func_traits.hpp"
+
+using namespace cv::gpu;
+using namespace cv::gpu::cudev;
+
+namespace arithm
+{
+    struct VAbsDiff4 : binary_function<uint, uint, uint>
+    {
+        __device__ __forceinline__ uint operator ()(uint a, uint b) const
+        {
+            return vabsdiff4(a, b);
+        }
+
+        __device__ __forceinline__ VAbsDiff4() {}
+        __device__ __forceinline__ VAbsDiff4(const VAbsDiff4& other) {}
+    };
+
+    struct VAbsDiff2 : binary_function<uint, uint, uint>
+    {
+        __device__ __forceinline__ uint operator ()(uint a, uint b) const
+        {
+            return vabsdiff2(a, b);
+        }
+
+        __device__ __forceinline__ VAbsDiff2() {}
+        __device__ __forceinline__ VAbsDiff2(const VAbsDiff2& other) {}
+    };
+
+    __device__ __forceinline__ int _abs(int a)
+    {
+        return ::abs(a);
+    }
+    __device__ __forceinline__ float _abs(float a)
+    {
+        return ::fabsf(a);
+    }
+    __device__ __forceinline__ double _abs(double a)
+    {
+        return ::fabs(a);
+    }
+
+    template <typename T> struct AbsDiffMat : binary_function<T, T, T>
+    {
+        __device__ __forceinline__ T operator ()(T a, T b) const
+        {
+            return saturate_cast<T>(_abs(a - b));
+        }
+
+        __device__ __forceinline__ AbsDiffMat() {}
+        __device__ __forceinline__ AbsDiffMat(const AbsDiffMat& other) {}
+    };
+}
+
+namespace cv { namespace gpu { namespace cudev
+{
+    template <> struct TransformFunctorTraits< arithm::VAbsDiff4 > : arithm::ArithmFuncTraits<sizeof(uint), sizeof(uint)>
+    {
+    };
+
+    template <> struct TransformFunctorTraits< arithm::VAbsDiff2 > : arithm::ArithmFuncTraits<sizeof(uint), sizeof(uint)>
+    {
+    };
+
+    template <typename T> struct TransformFunctorTraits< arithm::AbsDiffMat<T> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(T)>
+    {
+    };
+}}}
+
+namespace arithm
+{
+    void absDiffMat_v4(PtrStepSz<uint> src1, PtrStepSz<uint> src2, PtrStepSz<uint> dst, cudaStream_t stream)
+    {
+        cudev::transform(src1, src2, dst, VAbsDiff4(), WithOutMask(), stream);
+    }
+
+    void absDiffMat_v2(PtrStepSz<uint> src1, PtrStepSz<uint> src2, PtrStepSz<uint> dst, cudaStream_t stream)
+    {
+        cudev::transform(src1, src2, dst, VAbsDiff2(), WithOutMask(), stream);
+    }
+
+    template <typename T>
+    void absDiffMat(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream)
+    {
+        cudev::transform((PtrStepSz<T>) src1, (PtrStepSz<T>) src2, (PtrStepSz<T>) dst, AbsDiffMat<T>(), WithOutMask(), 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);
+}
+
+#endif // CUDA_DISABLER

modules/gpuarithm/src/cuda/absdiff_scalar.cu

@@ -0,0 +1,98 @@
+/*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*/
+
+#if !defined CUDA_DISABLER
+
+#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 "arithm_func_traits.hpp"
+
+using namespace cv::gpu;
+using namespace cv::gpu::cudev;
+
+namespace arithm
+{
+    template <typename T, typename S> struct AbsDiffScalar : unary_function<T, T>
+    {
+        S val;
+
+        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 gpu { namespace cudev
+{
+    template <typename T, typename S> struct TransformFunctorTraits< arithm::AbsDiffScalar<T, S> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(T)>
+    {
+    };
+}}}
+
+namespace arithm
+{
+    template <typename T, typename S>
+    void absDiffScalar(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream)
+    {
+        AbsDiffScalar<T, S> op(static_cast<S>(val));
+
+        cudev::transform((PtrStepSz<T>) src1, (PtrStepSz<T>) dst, op, WithOutMask(), 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

modules/gpuarithm/src/cuda/add_mat.cu

@@ -0,0 +1,185 @@
+/*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*/
+
+#if !defined CUDA_DISABLER
+
+#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 "arithm_func_traits.hpp"
+
+using namespace cv::gpu;
+using namespace cv::gpu::cudev;
+
+namespace arithm
+{
+    struct VAdd4 : binary_function<uint, uint, uint>
+    {
+        __device__ __forceinline__ uint operator ()(uint a, uint b) const
+        {
+            return vadd4(a, b);
+        }
+
+        __device__ __forceinline__ VAdd4() {}
+        __device__ __forceinline__ VAdd4(const VAdd4& other) {}
+    };
+
+    struct VAdd2 : binary_function<uint, uint, uint>
+    {
+        __device__ __forceinline__ uint operator ()(uint a, uint b) const
+        {
+            return vadd2(a, b);
+        }
+
+        __device__ __forceinline__ VAdd2() {}
+        __device__ __forceinline__ VAdd2(const VAdd2& other) {}
+    };
+
+    template <typename T, typename D> struct AddMat : binary_function<T, T, D>
+    {
+        __device__ __forceinline__ D operator ()(T a, T b) const
+        {
+            return saturate_cast<D>(a + b);
+        }
+
+        __device__ __forceinline__ AddMat() {}
+        __device__ __forceinline__ AddMat(const AddMat& other) {}
+    };
+}
+
+namespace cv { namespace gpu { namespace cudev
+{
+    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)
+    {
+        cudev::transform(src1, src2, dst, VAdd4(), WithOutMask(), stream);
+    }
+
+    void addMat_v2(PtrStepSz<uint> src1, PtrStepSz<uint> src2, PtrStepSz<uint> dst, cudaStream_t stream)
+    {
+        cudev::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)
+    {
+        if (mask.data)
+            cudev::transform((PtrStepSz<T>) src1, (PtrStepSz<T>) src2, (PtrStepSz<D>) dst, AddMat<T, D>(), mask, stream);
+        else
+            cudev::transform((PtrStepSz<T>) src1, (PtrStepSz<T>) src2, (PtrStepSz<D>) dst, AddMat<T, D>(), WithOutMask(), 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);
+
+    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);
+
+    //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);
+
+    //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);
+
+    //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);
+
+    //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);
+
+    //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);
+}
+
+#endif // CUDA_DISABLER

modules/gpuarithm/src/cuda/add_scalar.cu

@@ -0,0 +1,148 @@
+/*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*/
+
+#if !defined CUDA_DISABLER
+
+#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 "arithm_func_traits.hpp"
+
+using namespace cv::gpu;
+using namespace cv::gpu::cudev;
+
+namespace arithm
+{
+    template <typename T, typename S, typename D> struct AddScalar : unary_function<T, D>
+    {
+        S val;
+
+        explicit AddScalar(S val_) : val(val_) {}
+
+        __device__ __forceinline__ D operator ()(T a) const
+        {
+            return saturate_cast<D>(a + val);
+        }
+    };
+}
+
+namespace cv { namespace gpu { namespace cudev
+{
+    template <typename T, typename S, typename D> struct TransformFunctorTraits< arithm::AddScalar<T, S, D> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(D)>
+    {
+    };
+}}}
+
+namespace arithm
+{
+    template <typename T, typename S, typename D>
+    void addScalar(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream)
+    {
+        AddScalar<T, S, D> op(static_cast<S>(val));
+
+        if (mask.data)
+            cudev::transform((PtrStepSz<T>) src1, (PtrStepSz<D>) dst, op, mask, stream);
+        else
+            cudev::transform((PtrStepSz<T>) src1, (PtrStepSz<D>) dst, op, WithOutMask(), 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

modules/gpuarithm/src/cuda/add_weighted.cu

@@ -0,0 +1,364 @@
+/*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*/
+
+#if !defined CUDA_DISABLER
+
+#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 "arithm_func_traits.hpp"
+
+using namespace cv::gpu;
+using namespace cv::gpu::cudev;
+
+namespace arithm
+{
+    template <typename T> struct UseDouble_
+    {
+        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;
+
+        AddWeighted_(double alpha_, double beta_, double gamma_) : alpha(static_cast<float>(alpha_)), beta(static_cast<float>(beta_)), gamma(static_cast<float>(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;
+
+        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>
+    {
+        AddWeighted(double alpha_, double beta_, double gamma_) : AddWeighted_<T1, T2, D, UseDouble<T1, T2, D>::value>(alpha_, beta_, gamma_) {}
+    };
+}
+
+namespace cv { namespace gpu { namespace cudev
+{
+    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>
+    {
+    };
+
+    template <typename T1, typename T2, typename D> struct TransformFunctorTraits< arithm::AddWeighted<T1, T2, D> > : AddWeightedTraits<T1, T2, D, sizeof(T1), sizeof(T2), sizeof(D)>
+    {
+    };
+}}}
+
+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);
+
+        cudev::transform((PtrStepSz<T1>) src1, (PtrStepSz<T2>) src2, (PtrStepSz<D>) dst, op, WithOutMask(), stream);
+    }
+
+    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);
+
+    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);
+
+    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);
+}
+
+#endif /* CUDA_DISABLER */

modules/gpuarithm/src/cuda/arithm_func_traits.hpp

@@ -0,0 +1,145 @@
+/*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__

modules/gpuarithm/src/cuda/bitwise_mat.cu

@@ -0,0 +1,126 @@
+/*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*/
+
+#if !defined CUDA_DISABLER
+
+#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 "arithm_func_traits.hpp"
+
+using namespace cv::gpu;
+using namespace cv::gpu::cudev;
+
+namespace cv { namespace gpu { namespace cudev
+{
+    template <typename T> struct TransformFunctorTraits< bit_not<T> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(T)>
+    {
+    };
+
+    template <typename T> struct TransformFunctorTraits< bit_and<T> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(T)>
+    {
+    };
+
+    template <typename T> struct TransformFunctorTraits< bit_or<T> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(T)>
+    {
+    };
+
+    template <typename T> struct TransformFunctorTraits< bit_xor<T> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(T)>
+    {
+    };
+}}}
+
+namespace arithm
+{
+    template <typename T> void bitMatNot(PtrStepSzb src, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream)
+    {
+        if (mask.data)
+            cudev::transform((PtrStepSz<T>) src, (PtrStepSz<T>) dst, bit_not<T>(), mask, stream);
+        else
+            cudev::transform((PtrStepSz<T>) src, (PtrStepSz<T>) dst, bit_not<T>(), WithOutMask(), stream);
+    }
+
+    template <typename T> void bitMatAnd(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream)
+    {
+        if (mask.data)
+            cudev::transform((PtrStepSz<T>) src1, (PtrStepSz<T>) src2, (PtrStepSz<T>) dst, bit_and<T>(), mask, stream);
+        else
+            cudev::transform((PtrStepSz<T>) src1, (PtrStepSz<T>) src2, (PtrStepSz<T>) dst, bit_and<T>(), WithOutMask(), stream);
+    }
+
+    template <typename T> void bitMatOr(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream)
+    {
+        if (mask.data)
+            cudev::transform((PtrStepSz<T>) src1, (PtrStepSz<T>) src2, (PtrStepSz<T>) dst, bit_or<T>(), mask, stream);
+        else
+            cudev::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)
+            cudev::transform((PtrStepSz<T>) src1, (PtrStepSz<T>) src2, (PtrStepSz<T>) dst, bit_xor<T>(), mask, stream);
+        else
+            cudev::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

modules/gpuarithm/src/cuda/bitwise_scalar.cu

@@ -0,0 +1,104 @@
+/*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*/
+
+#if !defined CUDA_DISABLER
+
+#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 "arithm_func_traits.hpp"
+
+using namespace cv::gpu;
+using namespace cv::gpu::cudev;
+
+namespace cv { namespace gpu { namespace cudev
+{
+    template <typename T> struct TransformFunctorTraits< binder2nd< bit_and<T> > > : arithm::ArithmFuncTraits<sizeof(T), sizeof(T)>
+    {
+    };
+
+    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)
+    {
+        cudev::transform((PtrStepSz<T>) src1, (PtrStepSz<T>) dst, cv::gpu::cudev::bind2nd(bit_and<T>(), src2), WithOutMask(), stream);
+    }
+
+    template <typename T> void bitScalarOr(PtrStepSzb src1, uint src2, PtrStepSzb dst, cudaStream_t stream)
+    {
+        cudev::transform((PtrStepSz<T>) src1, (PtrStepSz<T>) dst, cv::gpu::cudev::bind2nd(bit_or<T>(), src2), WithOutMask(), stream);
+    }
+
+    template <typename T> void bitScalarXor(PtrStepSzb src1, uint src2, PtrStepSzb dst, cudaStream_t stream)
+    {
+        cudev::transform((PtrStepSz<T>) src1, (PtrStepSz<T>) dst, cv::gpu::cudev::bind2nd(bit_xor<T>(), src2), WithOutMask(), stream);
+    }
+
+    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);
+
+    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);
+
+    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);
+}
+
+#endif // CUDA_DISABLER

modules/gpuarithm/src/cuda/cmp_mat.cu

@@ -0,0 +1,206 @@
+/*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*/
+
+#if !defined CUDA_DISABLER
+
+#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 "arithm_func_traits.hpp"
+
+using namespace cv::gpu;
+using namespace cv::gpu::cudev;
+
+namespace arithm
+{
+    struct VCmpEq4 : binary_function<uint, uint, uint>
+    {
+        __device__ __forceinline__ uint operator ()(uint a, uint b) const
+        {
+            return vcmpeq4(a, b);
+        }
+
+        __device__ __forceinline__ VCmpEq4() {}
+        __device__ __forceinline__ VCmpEq4(const VCmpEq4& other) {}
+    };
+    struct VCmpNe4 : binary_function<uint, uint, uint>
+    {
+        __device__ __forceinline__ uint operator ()(uint a, uint b) const
+        {
+            return vcmpne4(a, b);
+        }
+
+        __device__ __forceinline__ VCmpNe4() {}
+        __device__ __forceinline__ VCmpNe4(const VCmpNe4& other) {}
+    };
+    struct VCmpLt4 : binary_function<uint, uint, uint>
+    {
+        __device__ __forceinline__ uint operator ()(uint a, uint b) const
+        {
+            return vcmplt4(a, b);
+        }
+
+        __device__ __forceinline__ VCmpLt4() {}
+        __device__ __forceinline__ VCmpLt4(const VCmpLt4& other) {}
+    };
+    struct VCmpLe4 : binary_function<uint, uint, uint>
+    {
+        __device__ __forceinline__ uint operator ()(uint a, uint b) const
+        {
+            return vcmple4(a, b);
+        }
+
+        __device__ __forceinline__ VCmpLe4() {}
+        __device__ __forceinline__ VCmpLe4(const VCmpLe4& other) {}
+    };
+
+    template <class Op, typename T>
+    struct Cmp : binary_function<T, T, uchar>
+    {
+        __device__ __forceinline__ uchar operator()(T a, T b) const
+        {
+            Op op;
+            return -op(a, b);
+        }
+    };
+}
+
+namespace cv { namespace gpu { namespace cudev
+{
+    template <> struct TransformFunctorTraits< arithm::VCmpEq4 > : arithm::ArithmFuncTraits<sizeof(uint), sizeof(uint)>
+    {
+    };
+    template <> struct TransformFunctorTraits< arithm::VCmpNe4 > : arithm::ArithmFuncTraits<sizeof(uint), sizeof(uint)>
+    {
+    };
+    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)
+    {
+        cudev::transform(src1, src2, dst, VCmpEq4(), WithOutMask(), stream);
+    }
+    void cmpMatNe_v4(PtrStepSz<uint> src1, PtrStepSz<uint> src2, PtrStepSz<uint> dst, cudaStream_t stream)
+    {
+        cudev::transform(src1, src2, dst, VCmpNe4(), WithOutMask(), stream);
+    }
+    void cmpMatLt_v4(PtrStepSz<uint> src1, PtrStepSz<uint> src2, PtrStepSz<uint> dst, cudaStream_t stream)
+    {
+        cudev::transform(src1, src2, dst, VCmpLt4(), WithOutMask(), stream);
+    }
+    void cmpMatLe_v4(PtrStepSz<uint> src1, PtrStepSz<uint> src2, PtrStepSz<uint> dst, cudaStream_t stream)
+    {
+        cudev::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)
+    {
+        Cmp<Op<T>, T> op;
+        cudev::transform((PtrStepSz<T>) src1, (PtrStepSz<T>) src2, dst, op, WithOutMask(), stream);
+    }
+
+    template <typename T> void cmpMatEq(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream)
+    {
+        cmpMat<equal_to, T>(src1, src2, dst, stream);
+    }
+    template <typename T> void cmpMatNe(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream)
+    {
+        cmpMat<not_equal_to, T>(src1, src2, dst, stream);
+    }
+    template <typename T> void cmpMatLt(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream)
+    {
+        cmpMat<less, T>(src1, src2, dst, stream);
+    }
+    template <typename T> void cmpMatLe(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream)
+    {
+        cmpMat<less_equal, T>(src1, src2, dst, stream);
+    }
+
+    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);
+
+    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);
+
+    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);
+
+    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);
+}
+
+#endif // CUDA_DISABLER

modules/gpuarithm/src/cuda/cmp_scalar.cu

@@ -0,0 +1,284 @@
+/*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*/
+
+#if !defined CUDA_DISABLER
+
+#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"
+
+#include "arithm_func_traits.hpp"
+
+using namespace cv::gpu;
+using namespace cv::gpu::cudev;
+
+namespace arithm
+{
+    template <class Op, typename T>
+    struct Cmp : binary_function<T, T, uchar>
+    {
+        __device__ __forceinline__ uchar operator()(T a, T b) const
+        {
+            Op op;
+            return -op(a, b);
+        }
+    };
+
+#define TYPE_VEC(type, cn) typename TypeVec<type, cn>::vec_type
+
+    template <class Op, typename T, int cn> struct CmpScalar;
+    template <class Op, typename T>
+    struct CmpScalar<Op, T, 1> : unary_function<T, uchar>
+    {
+        const T val;
+
+        __host__ explicit CmpScalar(T val_) : val(val_) {}
+
+        __device__ __forceinline__ uchar operator()(T src) const
+        {
+            Cmp<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)>
+    {
+        const TYPE_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
+        {
+            Cmp<Op, T> op;
+            return VecTraits<TYPE_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)>
+    {
+        const TYPE_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
+        {
+            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));
+        }
+    };
+    template <class Op, typename T>
+    struct CmpScalar<Op, T, 4> : unary_function<TYPE_VEC(T, 4), TYPE_VEC(uchar, 4)>
+    {
+        const TYPE_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
+        {
+            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));
+        }
+    };
+
+#undef TYPE_VEC
+}
+
+namespace cv { namespace gpu { namespace cudev
+{
+    template <class Op, typename T> struct TransformFunctorTraits< arithm::CmpScalar<Op, T, 1> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(uchar)>
+    {
+    };
+}}}
+
+namespace arithm
+{
+    template <template <typename> class Op, typename T, int cn>
+    void cmpScalar(PtrStepSzb src, double val[4], PtrStepSzb dst, cudaStream_t stream)
+    {
+        typedef typename TypeVec<T, cn>::vec_type src_t;
+        typedef typename TypeVec<uchar, cn>::vec_type dst_t;
+
+        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);
+
+        CmpScalar<Op<T>, T, cn> op(val1);
+        cudev::transform((PtrStepSz<src_t>) src, (PtrStepSz<dst_t>) dst, op, WithOutMask(), 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

modules/gpuarithm/src/cuda/countnonzero.cu

@@ -0,0 +1,175 @@
+/*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*/
+
+#if !defined CUDA_DISABLER
+
+#include "opencv2/core/cuda/common.hpp"
+#include "opencv2/core/cuda/vec_traits.hpp"
+#include "opencv2/core/cuda/vec_math.hpp"
+#include "opencv2/core/cuda/reduce.hpp"
+#include "opencv2/core/cuda/emulation.hpp"
+
+using namespace cv::gpu;
+using namespace cv::gpu::cudev;
+
+namespace countNonZero
+{
+    __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);
+            }
+        }
+
+        cudev::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;
+
+            cudev::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)
+    {
+        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);
+
+        unsigned int* count_buf = buf.ptr(0);
+
+        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;
+    }
+
+    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

modules/gpuarithm/src/cuda/div_inv.cu

@@ -0,0 +1,144 @@
+/*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*/
+
+#if !defined CUDA_DISABLER
+
+#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 "arithm_func_traits.hpp"
+
+using namespace cv::gpu;
+using namespace cv::gpu::cudev;
+
+namespace arithm
+{
+    template <typename T, typename S, typename D> struct DivInv : unary_function<T, D>
+    {
+        S val;
+
+        explicit DivInv(S val_) : val(val_) {}
+
+        __device__ __forceinline__ D operator ()(T a) const
+        {
+            return a != 0 ? saturate_cast<D>(val / a) : 0;
+        }
+    };
+}
+
+namespace cv { namespace gpu { namespace cudev
+{
+    template <typename T, typename S, typename D> struct TransformFunctorTraits< arithm::DivInv<T, S, D> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(D)>
+    {
+    };
+}}}
+
+namespace arithm
+{
+    template <typename T, typename S, typename D>
+    void divInv(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream)
+    {
+        DivInv<T, S, D> op(static_cast<S>(val));
+        cudev::transform((PtrStepSz<T>) src1, (PtrStepSz<D>) dst, op, WithOutMask(), stream);
+    }
+
+    template void divInv<uchar, float, uchar>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divInv<uchar, float, schar>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divInv<uchar, float, ushort>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divInv<uchar, float, short>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divInv<uchar, float, int>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divInv<uchar, float, float>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divInv<uchar, double, double>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+
+    template void divInv<schar, float, uchar>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divInv<schar, float, schar>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divInv<schar, float, ushort>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divInv<schar, float, short>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divInv<schar, float, int>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divInv<schar, float, float>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divInv<schar, double, double>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+
+    //template void divInv<ushort, float, uchar>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    //template void divInv<ushort, float, schar>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divInv<ushort, float, ushort>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divInv<ushort, float, short>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divInv<ushort, float, int>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divInv<ushort, float, float>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divInv<ushort, double, double>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+
+    //template void divInv<short, float, uchar>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    //template void divInv<short, float, schar>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divInv<short, float, ushort>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divInv<short, float, short>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divInv<short, float, int>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divInv<short, float, float>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divInv<short, double, double>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+
+    //template void divInv<int, float, uchar>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    //template void divInv<int, float, schar>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    //template void divInv<int, float, ushort>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    //template void divInv<int, float, short>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divInv<int, float, int>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divInv<int, float, float>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divInv<int, double, double>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+
+    //template void divInv<float, float, uchar>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    //template void divInv<float, float, schar>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    //template void divInv<float, float, ushort>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    //template void divInv<float, float, short>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    //template void divInv<float, float, int>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divInv<float, float, float>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divInv<float, double, double>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+
+    //template void divInv<double, double, uchar>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    //template void divInv<double, double, schar>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    //template void divInv<double, double, ushort>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    //template void divInv<double, double, short>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    //template void divInv<double, double, int>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    //template void divInv<double, double, float>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divInv<double, double, double>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+}
+
+#endif // CUDA_DISABLER

modules/gpuarithm/src/cuda/div_mat.cu

@@ -0,0 +1,230 @@
+/*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*/
+
+#if !defined CUDA_DISABLER
+
+#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 "arithm_func_traits.hpp"
+
+using namespace cv::gpu;
+using namespace cv::gpu::cudev;
+
+namespace arithm
+{
+    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>
+    {
+        __device__ __forceinline__ D operator ()(T a, T b) const
+        {
+            return b != 0 ? saturate_cast<D>(a / b) : 0;
+        }
+
+        __device__ __forceinline__ Div() {}
+        __device__ __forceinline__ Div(const Div& other) {}
+    };
+    template <typename T> struct Div<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;
+        }
+
+        __device__ __forceinline__ Div() {}
+        __device__ __forceinline__ Div(const Div& other) {}
+    };
+    template <typename T> struct Div<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;
+        }
+
+        __device__ __forceinline__ Div() {}
+        __device__ __forceinline__ Div(const Div& other) {}
+    };
+
+    template <typename T, typename S, typename D> struct DivScale : binary_function<T, T, D>
+    {
+        S scale;
+
+        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 gpu { namespace cudev
+{
+    template <> struct TransformFunctorTraits<arithm::Div_8uc4_32f> : arithm::ArithmFuncTraits<sizeof(uint), sizeof(uint)>
+    {
+    };
+
+    template <typename T, typename D> struct TransformFunctorTraits< arithm::Div<T, D> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(D)>
+    {
+    };
+
+    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)
+    {
+        cudev::transform(src1, src2, dst, Div_8uc4_32f(), WithOutMask(), stream);
+    }
+
+    void divMat_16sc4_32f(PtrStepSz<short4> src1, PtrStepSzf src2, PtrStepSz<short4> dst, cudaStream_t stream)
+    {
+        cudev::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)
+    {
+        if (scale == 1)
+        {
+            Div<T, D> op;
+            cudev::transform((PtrStepSz<T>) src1, (PtrStepSz<T>) src2, (PtrStepSz<D>) dst, op, WithOutMask(), stream);
+        }
+        else
+        {
+            DivScale<T, S, D> op(static_cast<S>(scale));
+            cudev::transform((PtrStepSz<T>) src1, (PtrStepSz<T>) src2, (PtrStepSz<D>) dst, op, WithOutMask(), 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

modules/gpuarithm/src/cuda/div_scalar.cu

@@ -0,0 +1,144 @@
+/*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*/
+
+#if !defined CUDA_DISABLER
+
+#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 "arithm_func_traits.hpp"
+
+using namespace cv::gpu;
+using namespace cv::gpu::cudev;
+
+namespace arithm
+{
+    template <typename T, typename S, typename D> struct DivScalar : unary_function<T, D>
+    {
+        S val;
+
+        explicit DivScalar(S val_) : val(val_) {}
+
+        __device__ __forceinline__ D operator ()(T a) const
+        {
+            return saturate_cast<D>(a / val);
+        }
+    };
+}
+
+namespace cv { namespace gpu { namespace cudev
+{
+    template <typename T, typename S, typename D> struct TransformFunctorTraits< arithm::DivScalar<T, S, D> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(D)>
+    {
+    };
+}}}
+
+namespace arithm
+{
+    template <typename T, typename S, typename D>
+    void divScalar(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream)
+    {
+        DivScalar<T, S, D> op(static_cast<S>(val));
+        cudev::transform((PtrStepSz<T>) src1, (PtrStepSz<D>) dst, op, WithOutMask(), stream);
+    }
+
+    template void divScalar<uchar, float, uchar>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divScalar<uchar, float, schar>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divScalar<uchar, float, ushort>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divScalar<uchar, float, short>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divScalar<uchar, float, int>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divScalar<uchar, float, float>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divScalar<uchar, double, double>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+
+    template void divScalar<schar, float, uchar>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divScalar<schar, float, schar>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divScalar<schar, float, ushort>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divScalar<schar, float, short>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divScalar<schar, float, int>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divScalar<schar, float, float>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divScalar<schar, double, double>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+
+    //template void divScalar<ushort, float, uchar>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    //template void divScalar<ushort, float, schar>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divScalar<ushort, float, ushort>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divScalar<ushort, float, short>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divScalar<ushort, float, int>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divScalar<ushort, float, float>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divScalar<ushort, double, double>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+
+    //template void divScalar<short, float, uchar>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    //template void divScalar<short, float, schar>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divScalar<short, float, ushort>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divScalar<short, float, short>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divScalar<short, float, int>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divScalar<short, float, float>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divScalar<short, double, double>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+
+    //template void divScalar<int, float, uchar>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    //template void divScalar<int, float, schar>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    //template void divScalar<int, float, ushort>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    //template void divScalar<int, float, short>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divScalar<int, float, int>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divScalar<int, float, float>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divScalar<int, double, double>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+
+    //template void divScalar<float, float, uchar>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    //template void divScalar<float, float, schar>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    //template void divScalar<float, float, ushort>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    //template void divScalar<float, float, short>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    //template void divScalar<float, float, int>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divScalar<float, float, float>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divScalar<float, double, double>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+
+    //template void divScalar<double, double, uchar>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    //template void divScalar<double, double, schar>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    //template void divScalar<double, double, ushort>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    //template void divScalar<double, double, short>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    //template void divScalar<double, double, int>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    //template void divScalar<double, double, float>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+    template void divScalar<double, double, double>(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream);
+}
+
+#endif // CUDA_DISABLER

modules/gpuarithm/src/cuda/math.cu

@@ -0,0 +1,302 @@
+/*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*/
+
+#if !defined CUDA_DISABLER
+
+#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"
+
+#include "arithm_func_traits.hpp"
+
+using namespace cv::gpu;
+using namespace cv::gpu::cudev;
+
+//////////////////////////////////////////////////////////////////////////
+// absMat
+
+namespace cv { namespace gpu { namespace cudev
+{
+    template <typename T> struct TransformFunctorTraits< abs_func<T> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(T)>
+    {
+    };
+}}}
+
+namespace arithm
+{
+    template <typename T>
+    void absMat(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream)
+    {
+        cudev::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);
+}
+
+//////////////////////////////////////////////////////////////////////////
+// sqrMat
+
+namespace arithm
+{
+    template <typename T> struct Sqr : unary_function<T, T>
+    {
+        __device__ __forceinline__ T operator ()(T x) const
+        {
+            return saturate_cast<T>(x * x);
+        }
+
+        __device__ __forceinline__ Sqr() {}
+        __device__ __forceinline__ Sqr(const Sqr& other) {}
+    };
+}
+
+namespace cv { namespace gpu { namespace cudev
+{
+    template <typename T> struct TransformFunctorTraits< arithm::Sqr<T> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(T)>
+    {
+    };
+}}}
+
+namespace arithm
+{
+    template <typename T>
+    void sqrMat(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream)
+    {
+        cudev::transform((PtrStepSz<T>) src, (PtrStepSz<T>) dst, Sqr<T>(), WithOutMask(), 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 gpu { namespace cudev
+{
+    template <typename T> struct TransformFunctorTraits< sqrt_func<T> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(T)>
+    {
+    };
+}}}
+
+namespace arithm
+{
+    template <typename T>
+    void sqrtMat(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream)
+    {
+        cudev::transform((PtrStepSz<T>) src, (PtrStepSz<T>) dst, sqrt_func<T>(), WithOutMask(), stream);
+    }
+
+    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);
+}
+
+//////////////////////////////////////////////////////////////////////////
+// logMat
+
+namespace cv { namespace gpu { namespace cudev
+{
+    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)
+    {
+        cudev::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>
+    {
+        __device__ __forceinline__ T operator ()(T x) const
+        {
+            exp_func<T> f;
+            return saturate_cast<T>(f(x));
+        }
+
+        __device__ __forceinline__ Exp() {}
+        __device__ __forceinline__ Exp(const Exp& other) {}
+    };
+}
+
+namespace cv { namespace gpu { namespace cudev
+{
+    template <typename T> struct TransformFunctorTraits< arithm::Exp<T> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(T)>
+    {
+    };
+}}}
+
+namespace arithm
+{
+    template <typename T>
+    void expMat(PtrStepSzb src, PtrStepSzb dst, cudaStream_t stream)
+    {
+        cudev::transform((PtrStepSz<T>) src, (PtrStepSz<T>) dst, Exp<T>(), WithOutMask(), 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);
+}
+
+//////////////////////////////////////////////////////////////////////////
+// pow
+
+namespace arithm
+{
+    template<typename T, bool Signed = numeric_limits<T>::is_signed> struct PowOp : unary_function<T, T>
+    {
+        float power;
+
+        PowOp(double power_) : power(static_cast<float>(power_)) {}
+
+        __device__ __forceinline__ T operator()(T e) const
+        {
+            return saturate_cast<T>(__powf((float)e, power));
+        }
+    };
+    template<typename T> struct PowOp<T, true> : unary_function<T, T>
+    {
+        float power;
+
+        PowOp(double power_) : power(static_cast<float>(power_)) {}
+
+        __device__ __forceinline__ T operator()(T e) const
+        {
+            T res = saturate_cast<T>(__powf((float)e, power));
+
+            if ((e < 0) && (1 & static_cast<int>(power)))
+                res *= -1;
+
+            return res;
+        }
+    };
+    template<> struct PowOp<float> : unary_function<float, float>
+    {
+        const float power;
+
+        PowOp(double power_) : power(static_cast<float>(power_)) {}
+
+        __device__ __forceinline__ float operator()(float e) const
+        {
+            return __powf(::fabs(e), power);
+        }
+    };
+    template<> struct PowOp<double> : unary_function<double, double>
+    {
+        double power;
+
+        PowOp(double power_) : power(power_) {}
+
+        __device__ __forceinline__ double operator()(double e) const
+        {
+            return ::pow(::fabs(e), power);
+        }
+    };
+}
+
+namespace cv { namespace gpu { namespace cudev
+{
+    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)
+    {
+        cudev::transform((PtrStepSz<T>) src, (PtrStepSz<T>) dst, PowOp<T>(power), WithOutMask(), stream);
+    }
+
+    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);
+}
+
+#endif // CUDA_DISABLER

modules/gpuarithm/src/cuda/minmax.cu

@@ -0,0 +1,246 @@
+/*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*/
+
+#if !defined CUDA_DISABLER
+
+#include "opencv2/core/cuda/common.hpp"
+#include "opencv2/core/cuda/vec_traits.hpp"
+#include "opencv2/core/cuda/vec_math.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"
+
+using namespace cv::gpu;
+using namespace cv::gpu::cudev;
+
+namespace minMax
+{
+    __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;
+                cudev::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);
+                }
+            }
+        }
+
+        cudev::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)
+    {
+        typedef typename MinMaxTypeTraits<T>::best_type R;
+
+        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);
+
+        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);
+        else
+            kernel<threads_x * threads_y><<<grid, block>>>((PtrStepSz<T>) src, WithOutMask(), minval_buf, maxval_buf, twidth, theight);
+
+        cudaSafeCall( cudaGetLastError() );
+
+        cudaSafeCall( cudaDeviceSynchronize() );
+
+        R minval_, maxval_;
+        cudaSafeCall( cudaMemcpy(&minval_, minval_buf, sizeof(R), cudaMemcpyDeviceToHost) );
+        cudaSafeCall( cudaMemcpy(&maxval_, maxval_buf, sizeof(R), cudaMemcpyDeviceToHost) );
+        *minval = minval_;
+        *maxval = maxval_;
+    }
+
+    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

modules/gpuarithm/src/cuda/minmax_mat.cu

@@ -0,0 +1,228 @@
+/*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*/
+
+#if !defined CUDA_DISABLER
+
+#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 "arithm_func_traits.hpp"
+
+using namespace cv::gpu;
+using namespace cv::gpu::cudev;
+
+//////////////////////////////////////////////////////////////////////////
+// min
+
+namespace arithm
+{
+    struct VMin4 : binary_function<uint, uint, uint>
+    {
+        __device__ __forceinline__ uint operator ()(uint a, uint b) const
+        {
+            return vmin4(a, b);
+        }
+
+        __device__ __forceinline__ VMin4() {}
+        __device__ __forceinline__ VMin4(const VMin4& other) {}
+    };
+
+    struct VMin2 : binary_function<uint, uint, uint>
+    {
+        __device__ __forceinline__ uint operator ()(uint a, uint b) const
+        {
+            return vmin2(a, b);
+        }
+
+        __device__ __forceinline__ VMin2() {}
+        __device__ __forceinline__ VMin2(const VMin2& other) {}
+    };
+}
+
+namespace cv { namespace gpu { namespace cudev
+{
+    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)
+    {
+        cudev::transform(src1, src2, dst, VMin4(), WithOutMask(), stream);
+    }
+
+    void minMat_v2(PtrStepSz<uint> src1, PtrStepSz<uint> src2, PtrStepSz<uint> dst, cudaStream_t stream)
+    {
+        cudev::transform(src1, src2, dst, VMin2(), WithOutMask(), stream);
+    }
+
+    template <typename T> void minMat(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream)
+    {
+        cudev::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)
+    {
+        cudev::transform((PtrStepSz<T>) src1, (PtrStepSz<T>) dst, cv::gpu::cudev::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);
+        }
+
+        __device__ __forceinline__ VMax4() {}
+        __device__ __forceinline__ VMax4(const VMax4& other) {}
+    };
+
+    struct VMax2 : binary_function<uint, uint, uint>
+    {
+        __device__ __forceinline__ uint operator ()(uint a, uint b) const
+        {
+            return vmax2(a, b);
+        }
+
+        __device__ __forceinline__ VMax2() {}
+        __device__ __forceinline__ VMax2(const VMax2& other) {}
+    };
+}
+
+namespace cv { namespace gpu { namespace cudev
+{
+    template <> struct TransformFunctorTraits< arithm::VMax4 > : arithm::ArithmFuncTraits<sizeof(uint), sizeof(uint)>
+    {
+    };
+
+    template <> struct TransformFunctorTraits< arithm::VMax2 > : arithm::ArithmFuncTraits<sizeof(uint), sizeof(uint)>
+    {
+    };
+
+    template <typename T> struct TransformFunctorTraits< maximum<T> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(T)>
+    {
+    };
+
+    template <typename T> struct TransformFunctorTraits< binder2nd< maximum<T> > > : arithm::ArithmFuncTraits<sizeof(T), sizeof(T)>
+    {
+    };
+}}}
+
+namespace arithm
+{
+    void maxMat_v4(PtrStepSz<uint> src1, PtrStepSz<uint> src2, PtrStepSz<uint> dst, cudaStream_t stream)
+    {
+        cudev::transform(src1, src2, dst, VMax4(), WithOutMask(), stream);
+    }
+
+    void maxMat_v2(PtrStepSz<uint> src1, PtrStepSz<uint> src2, PtrStepSz<uint> dst, cudaStream_t stream)
+    {
+        cudev::transform(src1, src2, dst, VMax2(), WithOutMask(), stream);
+    }
+
+    template <typename T> void maxMat(PtrStepSzb src1, PtrStepSzb src2, PtrStepSzb dst, cudaStream_t stream)
+    {
+        cudev::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)
+    {
+        cudev::transform((PtrStepSz<T>) src1, (PtrStepSz<T>) dst, cv::gpu::cudev::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);
+}
+
+#endif // CUDA_DISABLER

modules/gpuarithm/src/cuda/minmaxloc.cu

@@ -0,0 +1,235 @@
+/*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*/
+
+#if !defined CUDA_DISABLER
+
+#include "opencv2/core/cuda/common.hpp"
+#include "opencv2/core/cuda/vec_traits.hpp"
+#include "opencv2/core/cuda/vec_math.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"
+
+using namespace cv::gpu;
+using namespace cv::gpu::cudev;
+
+namespace minMaxLoc
+{
+    // 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)
+    {
+        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);
+
+        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);
+        else
+            kernel_pass_1<threads_x * threads_y><<<grid, block>>>((PtrStepSz<T>) src, WithOutMask(), minval_buf, maxval_buf, minloc_buf, maxloc_buf, twidth, theight);
+
+        cudaSafeCall( cudaGetLastError() );
+
+        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() );
+
+        cudaSafeCall( cudaDeviceSynchronize() );
+
+        T minval_, maxval_;
+        cudaSafeCall( cudaMemcpy(&minval_, minval_buf, sizeof(T), cudaMemcpyDeviceToHost) );
+        cudaSafeCall( cudaMemcpy(&maxval_, maxval_buf, sizeof(T), cudaMemcpyDeviceToHost) );
+        *minval = minval_;
+        *maxval = maxval_;
+
+        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;
+    }
+
+    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

modules/gpuarithm/src/cuda/mul_mat.cu

@@ -0,0 +1,211 @@
+/*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*/
+
+#if !defined CUDA_DISABLER
+
+#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 "arithm_func_traits.hpp"
+
+using namespace cv::gpu;
+using namespace cv::gpu::cudev;
+
+namespace arithm
+{
+    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;
+        }
+
+        __device__ __forceinline__ Mul_8uc4_32f() {}
+        __device__ __forceinline__ Mul_8uc4_32f(const Mul_8uc4_32f& other) {}
+    };
+
+    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));
+        }
+
+        __device__ __forceinline__ Mul_16sc4_32f() {}
+        __device__ __forceinline__ Mul_16sc4_32f(const Mul_16sc4_32f& other) {}
+    };
+
+    template <typename T, typename D> struct Mul : binary_function<T, T, D>
+    {
+        __device__ __forceinline__ D operator ()(T a, T b) const
+        {
+            return saturate_cast<D>(a * b);
+        }
+
+        __device__ __forceinline__ Mul() {}
+        __device__ __forceinline__ Mul(const Mul& other) {}
+    };
+
+    template <typename T, typename S, typename D> struct MulScale : binary_function<T, T, D>
+    {
+        S scale;
+
+        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 gpu { namespace cudev
+{
+    template <> struct TransformFunctorTraits<arithm::Mul_8uc4_32f> : arithm::ArithmFuncTraits<sizeof(uint), sizeof(uint)>
+    {
+    };
+
+    template <typename T, typename D> struct TransformFunctorTraits< arithm::Mul<T, D> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(D)>
+    {
+    };
+
+    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)
+    {
+        cudev::transform(src1, src2, dst, Mul_8uc4_32f(), WithOutMask(), stream);
+    }
+
+    void mulMat_16sc4_32f(PtrStepSz<short4> src1, PtrStepSzf src2, PtrStepSz<short4> dst, cudaStream_t stream)
+    {
+        cudev::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)
+    {
+        if (scale == 1)
+        {
+            Mul<T, D> op;
+            cudev::transform((PtrStepSz<T>) src1, (PtrStepSz<T>) src2, (PtrStepSz<D>) dst, op, WithOutMask(), stream);
+        }
+        else
+        {
+            MulScale<T, S, D> op(static_cast<S>(scale));
+            cudev::transform((PtrStepSz<T>) src1, (PtrStepSz<T>) src2, (PtrStepSz<D>) dst, op, WithOutMask(), 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

modules/gpuarithm/src/cuda/mul_scalar.cu

@@ -0,0 +1,144 @@
+/*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*/
+
+#if !defined CUDA_DISABLER
+
+#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 "arithm_func_traits.hpp"
+
+using namespace cv::gpu;
+using namespace cv::gpu::cudev;
+
+namespace arithm
+{
+    template <typename T, typename S, typename D> struct MulScalar : unary_function<T, D>
+    {
+        S val;
+
+        explicit MulScalar(S val_) : val(val_) {}
+
+        __device__ __forceinline__ D operator ()(T a) const
+        {
+            return saturate_cast<D>(a * val);
+        }
+    };
+}
+
+namespace cv { namespace gpu { namespace cudev
+{
+    template <typename T, typename S, typename D> struct TransformFunctorTraits< arithm::MulScalar<T, S, D> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(D)>
+    {
+    };
+}}}
+
+namespace arithm
+{
+    template <typename T, typename S, typename D>
+    void mulScalar(PtrStepSzb src1, double val, PtrStepSzb dst, cudaStream_t stream)
+    {
+        MulScalar<T, S, D> op(static_cast<S>(val));
+        cudev::transform((PtrStepSz<T>) src1, (PtrStepSz<D>) dst, op, WithOutMask(), 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

modules/gpuarithm/src/cuda/polar_cart.cu

@@ -0,0 +1,217 @@
+/*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*/
+
+#if !defined CUDA_DISABLER
+
+#include "opencv2/core/cuda/common.hpp"
+
+namespace cv { namespace gpu { namespace cudev
+{
+    namespace mathfunc
+    {
+        //////////////////////////////////////////////////////////////////////////////////////
+        // Cart <-> Polar
+
+        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;
+
+            if (x < width && y < height)
+            {
+                float x_data = xptr[y * x_step + x];
+                float y_data = yptr[y * y_step + x];
+
+                Mag::calc(x, y, x_data, y_data, mag, mag_step, scale);
+                Angle::calc(x, y, x_data, y_data, angle, angle_step, scale);
+            }
+        }
+
+        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;
+
+            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;
+
+                ::sincosf(scale * angle_data, &sin_a, &cos_a);
+
+                xptr[y * x_step + x] = mag_data * cos_a;
+                yptr[y * y_step + x] = mag_data * sin_a;
+            }
+        }
+
+        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);
+
+            grid.x = divUp(x.cols, threads.x);
+            grid.y = divUp(x.rows, threads.y);
+
+            const float scale = angleInDegrees ? (180.0f / CV_PI_F) : 1.f;
+
+            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() );
+
+            if (stream == 0)
+                cudaSafeCall( cudaDeviceSynchronize() );
+        }
+
+        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>,
+                    }
+                }
+            };
+
+            callers[mag.data == 0][magSqr][angle.data == 0](x, y, mag, angle, angleInDegrees, stream);
+        }
+
+        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);
+
+            grid.x = divUp(mag.cols, threads.x);
+            grid.y = divUp(mag.rows, threads.y);
+
+            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 gpu { namespace cudev
+
+#endif /* CUDA_DISABLER */

modules/gpuarithm/src/cuda/reduce.cu

@@ -0,0 +1,330 @@
+/*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*/
+
+#if !defined CUDA_DISABLER
+
+#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/reduce.hpp"
+#include "opencv2/core/cuda/limits.hpp"
+
+#include "unroll_detail.hpp"
+
+using namespace cv::gpu;
+using namespace cv::gpu::cudev;
+
+namespace reduce
+{
+    struct Sum
+    {
+        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__ T result(T r, double) const
+        {
+            return r;
+        }
+
+        __device__ __forceinline__ Sum() {}
+        __device__ __forceinline__ Sum(const Sum&) {}
+    };
+
+    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<double, VecTraits<T>::cn>::vec_type result(T r, double sz) const
+        {
+            return r / sz;
+        }
+
+        __device__ __forceinline__ Avg() {}
+        __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, double) const
+        {
+            return r;
+        }
+
+        __device__ __forceinline__ Min() {}
+        __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, double) const
+        {
+            return r;
+        }
+
+        __device__ __forceinline__ Max() {}
+        __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];
+        cudev::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() );
+    }
+
+    template <typename T, typename S, typename D>
+    void rows(PtrStepSzb src, void* dst, int op, cudaStream_t stream)
+    {
+        typedef void (*func_t)(PtrStepSz<T> src, D* dst, cudaStream_t stream);
+        static const func_t funcs[] =
+        {
+            rowsCaller<T, S, D, Sum>,
+            rowsCaller<T, S, D, Avg>,
+            rowsCaller<T, S, D, Max>,
+            rowsCaller<T, S, D, Min>
+        };
+
+        funcs[op]((PtrStepSz<T>) src, (D*) dst, 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]));
+
+        cudev::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 */

modules/gpuarithm/src/cuda/split_merge.cu

@@ -0,0 +1,511 @@
+/*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*/
+
+#if !defined CUDA_DISABLER
+
+#include "opencv2/core/cuda/common.hpp"
+
+namespace cv { namespace gpu { namespace cudev
+{
+    namespace split_merge
+    {
+        template <typename T, size_t elem_size = sizeof(T)>
+        struct TypeTraits
+        {
+            typedef T type;
+            typedef T type2;
+            typedef T type3;
+            typedef T type4;
+        };
+
+        template <typename T>
+        struct TypeTraits<T, 1>
+        {
+            typedef char type;
+            typedef char2 type2;
+            typedef char3 type3;
+            typedef char4 type4;
+        };
+
+        template <typename T>
+        struct TypeTraits<T, 2>
+        {
+            typedef short type;
+            typedef short2 type2;
+            typedef short3 type3;
+            typedef short4 type4;
+        };
+
+        template <typename T>
+        struct TypeTraits<T, 4>
+        {
+            typedef int type;
+            typedef int2 type2;
+            typedef int3 type3;
+            typedef int4 type4;
+        };
+
+        template <typename T>
+        struct TypeTraits<T, 8>
+        {
+            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)
+        {
+            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)
+            {
+                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_caller(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>,
+            };
+
+            size_t merge_func_id = (total_channels - 2) * 5 + (elem_size >> 1);
+            MergeFunction merge_func = merge_func_tbl[merge_func_id];
+
+            if (merge_func == 0)
+                CV_Error(cv::Error::StsUnsupportedFormat, "Unsupported channel count or data type");
+
+            merge_func(src, dst, stream);
+        }
+
+
+
+        //------------------------------------------------------------
+        // Split
+
+
+        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)
+        {
+            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)
+            {
+                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_caller(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>,
+            };
+
+            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);
+        }
+    } // namespace split_merge
+}}} // namespace cv { namespace gpu { namespace cudev
+
+
+#endif /* CUDA_DISABLER */

modules/gpuarithm/src/cuda/sub_mat.cu

@@ -0,0 +1,185 @@
+/*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*/
+
+#if !defined CUDA_DISABLER
+
+#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 "arithm_func_traits.hpp"
+
+using namespace cv::gpu;
+using namespace cv::gpu::cudev;
+
+namespace arithm
+{
+    struct VSub4 : binary_function<uint, uint, uint>
+    {
+        __device__ __forceinline__ uint operator ()(uint a, uint b) const
+        {
+            return vsub4(a, b);
+        }
+
+        __device__ __forceinline__ VSub4() {}
+        __device__ __forceinline__ VSub4(const VSub4& other) {}
+    };
+
+    struct VSub2 : binary_function<uint, uint, uint>
+    {
+        __device__ __forceinline__ uint operator ()(uint a, uint b) const
+        {
+            return vsub2(a, b);
+        }
+
+        __device__ __forceinline__ VSub2() {}
+        __device__ __forceinline__ VSub2(const VSub2& other) {}
+    };
+
+    template <typename T, typename D> struct SubMat : binary_function<T, T, D>
+    {
+        __device__ __forceinline__ D operator ()(T a, T b) const
+        {
+            return saturate_cast<D>(a - b);
+        }
+
+        __device__ __forceinline__ SubMat() {}
+        __device__ __forceinline__ SubMat(const SubMat& other) {}
+    };
+}
+
+namespace cv { namespace gpu { namespace cudev
+{
+    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)
+    {
+        cudev::transform(src1, src2, dst, VSub4(), WithOutMask(), stream);
+    }
+
+    void subMat_v2(PtrStepSz<uint> src1, PtrStepSz<uint> src2, PtrStepSz<uint> dst, cudaStream_t stream)
+    {
+        cudev::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)
+    {
+        if (mask.data)
+            cudev::transform((PtrStepSz<T>) src1, (PtrStepSz<T>) src2, (PtrStepSz<D>) dst, SubMat<T, D>(), mask, stream);
+        else
+            cudev::transform((PtrStepSz<T>) src1, (PtrStepSz<T>) src2, (PtrStepSz<D>) dst, SubMat<T, D>(), WithOutMask(), 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);
+
+    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);
+
+    //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);
+
+    //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);
+
+    //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);
+
+    //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);
+
+    //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);
+}
+
+#endif // CUDA_DISABLER

modules/gpuarithm/src/cuda/sub_scalar.cu

@@ -0,0 +1,148 @@
+/*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*/
+
+#if !defined CUDA_DISABLER
+
+#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 "arithm_func_traits.hpp"
+
+using namespace cv::gpu;
+using namespace cv::gpu::cudev;
+
+namespace arithm
+{
+    template <typename T, typename S, typename D> struct SubScalar : unary_function<T, D>
+    {
+        S val;
+
+        explicit SubScalar(S val_) : val(val_) {}
+
+        __device__ __forceinline__ D operator ()(T a) const
+        {
+            return saturate_cast<D>(a - val);
+        }
+    };
+}
+
+namespace cv { namespace gpu { namespace cudev
+{
+    template <typename T, typename S, typename D> struct TransformFunctorTraits< arithm::SubScalar<T, S, D> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(D)>
+    {
+    };
+}}}
+
+namespace arithm
+{
+    template <typename T, typename S, typename D>
+    void subScalar(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream)
+    {
+        SubScalar<T, S, D> op(static_cast<S>(val));
+
+        if (mask.data)
+            cudev::transform((PtrStepSz<T>) src1, (PtrStepSz<D>) dst, op, mask, stream);
+        else
+            cudev::transform((PtrStepSz<T>) src1, (PtrStepSz<D>) dst, op, WithOutMask(), stream);
+    }
+
+    template void subScalar<uchar, float, uchar>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    template void subScalar<uchar, float, schar>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    template void subScalar<uchar, float, ushort>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    template void subScalar<uchar, float, short>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    template void subScalar<uchar, float, int>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    template void subScalar<uchar, float, float>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    template void subScalar<uchar, double, double>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+
+    template void subScalar<schar, float, uchar>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    template void subScalar<schar, float, schar>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    template void subScalar<schar, float, ushort>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    template void subScalar<schar, float, short>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    template void subScalar<schar, float, int>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    template void subScalar<schar, float, float>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    template void subScalar<schar, double, double>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+
+    //template void subScalar<ushort, float, uchar>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    //template void subScalar<ushort, float, schar>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    template void subScalar<ushort, float, ushort>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    template void subScalar<ushort, float, short>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    template void subScalar<ushort, float, int>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    template void subScalar<ushort, float, float>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    template void subScalar<ushort, double, double>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+
+    //template void subScalar<short, float, uchar>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    //template void subScalar<short, float, schar>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    template void subScalar<short, float, ushort>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    template void subScalar<short, float, short>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    template void subScalar<short, float, int>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    template void subScalar<short, float, float>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    template void subScalar<short, double, double>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+
+    //template void subScalar<int, float, uchar>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    //template void subScalar<int, float, schar>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    //template void subScalar<int, float, ushort>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    //template void subScalar<int, float, short>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    template void subScalar<int, float, int>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    template void subScalar<int, float, float>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    template void subScalar<int, double, double>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+
+    //template void subScalar<float, float, uchar>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    //template void subScalar<float, float, schar>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    //template void subScalar<float, float, ushort>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    //template void subScalar<float, float, short>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    //template void subScalar<float, float, int>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    template void subScalar<float, float, float>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    template void subScalar<float, double, double>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+
+    //template void subScalar<double, double, uchar>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    //template void subScalar<double, double, schar>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    //template void subScalar<double, double, ushort>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    //template void subScalar<double, double, short>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    //template void subScalar<double, double, int>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    //template void subScalar<double, double, float>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+    template void subScalar<double, double, double>(PtrStepSzb src1, double val, PtrStepSzb dst, PtrStepb mask, cudaStream_t stream);
+}
+
+#endif // CUDA_DISABLER

modules/gpuarithm/src/cuda/sum.cu

@@ -0,0 +1,380 @@
+/*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*/
+
+#if !defined CUDA_DISABLER
+
+#include "opencv2/core/cuda/common.hpp"
+#include "opencv2/core/cuda/vec_traits.hpp"
+#include "opencv2/core/cuda/vec_math.hpp"
+#include "opencv2/core/cuda/reduce.hpp"
+#include "opencv2/core/cuda/emulation.hpp"
+#include "opencv2/core/cuda/utility.hpp"
+
+#include "unroll_detail.hpp"
+
+using namespace cv::gpu;
+using namespace cv::gpu::cudev;
+
+namespace sum
+{
+    __device__ unsigned int blocks_finished = 0;
+
+    template <typename R, int cn> struct AtomicAdd;
+    template <typename R> struct AtomicAdd<R, 1>
+    {
+        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;
+
+        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;
+
+        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);
+
+                cudev::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));
+                }
+            }
+        }
+
+        cudev::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);
+        else
+            kernel<threads_x * threads_y><<<grid, block>>>(src, buf, WithOutMask(), op, twidth, theight);
+        cudaSafeCall( cudaGetLastError() );
+
+        cudaSafeCall( cudaDeviceSynchronize() );
+
+        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];
+    }
+
+    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)
+    {
+        typedef typename SumType<T>::R R;
+        caller<T, R, cn, identity>(src, buf, out, mask);
+    }
+
+    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)
+    {
+        typedef typename SumType<T>::R R;
+        caller<T, R, cn, abs_func>(src, buf, out, mask);
+    }
+
+    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

modules/gpuarithm/src/cuda/threshold.cu

@@ -0,0 +1,114 @@
+/*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*/
+
+#if !defined CUDA_DISABLER
+
+#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 "arithm_func_traits.hpp"
+
+using namespace cv::gpu;
+using namespace cv::gpu::cudev;
+
+namespace cv { namespace gpu { namespace cudev
+{
+    template <typename T> struct TransformFunctorTraits< thresh_binary_func<T> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(T)>
+    {
+    };
+
+    template <typename T> struct TransformFunctorTraits< thresh_binary_inv_func<T> > : arithm::ArithmFuncTraits<sizeof(T), sizeof(T)>
+    {
+    };
+
+    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);
+        cudev::transform(src, dst, op, WithOutMask(), stream);
+    }
+
+    template <typename T>
+    void threshold(PtrStepSzb src, PtrStepSzb dst, double thresh, double maxVal, int type, cudaStream_t stream)
+    {
+        typedef void (*caller_t)(PtrStepSz<T> src, PtrStepSz<T> dst, T thresh, T maxVal, cudaStream_t stream);
+
+        static const caller_t callers[] =
+        {
+            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>
+        };
+
+        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

modules/gpuarithm/src/cuda/transpose.cu

@@ -0,0 +1,122 @@
+/*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*/
+
+#if !defined CUDA_DISABLER
+
+#include "opencv2/core/cuda/common.hpp"
+
+using namespace cv::gpu;
+using namespace cv::gpu::cudev;
+
+namespace arithm
+{
+    const int TRANSPOSE_TILE_DIM   = 16;
+    const int TRANSPOSE_BLOCK_ROWS = 16;
+
+    template <typename T>
+    __global__ void transposeKernel(const PtrStepSz<T> src, PtrStep<T> dst)
+    {
+        __shared__ T tile[TRANSPOSE_TILE_DIM][TRANSPOSE_TILE_DIM + 1];
+
+        int blockIdx_x, blockIdx_y;
+
+        // 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;
+        }
+
+        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];
+                }
+            }
+        }
+    }
+
+    template <typename T> void transpose(PtrStepSz<T> src, PtrStepSz<T> dst, cudaStream_t stream)
+    {
+        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() );
+    }
+
+    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

modules/gpuarithm/src/cuda/unroll_detail.hpp

@@ -0,0 +1,135 @@
+/*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::gpu::cudev::smem_tuple(smem, smem + BLOCK_SIZE);
+        }
+
+        template <typename R>
+        static __device__ __forceinline__ thrust::tuple<typename cv::gpu::cudev::VecTraits<R>::elem_type&, typename cv::gpu::cudev::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::gpu::cudev::smem_tuple(smem, smem + BLOCK_SIZE, smem + 2 * BLOCK_SIZE);
+        }
+
+        template <typename R>
+        static __device__ __forceinline__ thrust::tuple<typename cv::gpu::cudev::VecTraits<R>::elem_type&, typename cv::gpu::cudev::VecTraits<R>::elem_type&, typename cv::gpu::cudev::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::gpu::cudev::smem_tuple(smem, smem + BLOCK_SIZE, smem + 2 * BLOCK_SIZE, smem + 3 * BLOCK_SIZE);
+        }
+
+        template <typename R>
+        static __device__ __forceinline__ thrust::tuple<typename cv::gpu::cudev::VecTraits<R>::elem_type&, typename cv::gpu::cudev::VecTraits<R>::elem_type&, typename cv::gpu::cudev::VecTraits<R>::elem_type&, typename cv::gpu::cudev::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__

modules/gpuarithm/src/matrix_reductions.cpp

@@ -0,0 +1,700 @@
+/*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 "precomp.hpp"
+
+using namespace cv;
+using namespace cv::gpu;
+
+#if !defined (HAVE_CUDA) || defined (CUDA_DISABLER)
+
+void cv::gpu::meanStdDev(const GpuMat&, Scalar&, Scalar&) { throw_no_cuda(); }
+void cv::gpu::meanStdDev(const GpuMat&, Scalar&, Scalar&, GpuMat&) { throw_no_cuda(); }
+double cv::gpu::norm(const GpuMat&, int) { throw_no_cuda(); return 0.0; }
+double cv::gpu::norm(const GpuMat&, int, GpuMat&) { throw_no_cuda(); return 0.0; }
+double cv::gpu::norm(const GpuMat&, int, const GpuMat&, GpuMat&) { throw_no_cuda(); return 0.0; }
+double cv::gpu::norm(const GpuMat&, const GpuMat&, int) { throw_no_cuda(); return 0.0; }
+Scalar cv::gpu::sum(const GpuMat&) { throw_no_cuda(); return Scalar(); }
+Scalar cv::gpu::sum(const GpuMat&, GpuMat&) { throw_no_cuda(); return Scalar(); }
+Scalar cv::gpu::sum(const GpuMat&, const GpuMat&, GpuMat&) { throw_no_cuda(); return Scalar(); }
+Scalar cv::gpu::absSum(const GpuMat&) { throw_no_cuda(); return Scalar(); }
+Scalar cv::gpu::absSum(const GpuMat&, GpuMat&) { throw_no_cuda(); return Scalar(); }
+Scalar cv::gpu::absSum(const GpuMat&, const GpuMat&, GpuMat&) { throw_no_cuda(); return Scalar(); }
+Scalar cv::gpu::sqrSum(const GpuMat&) { throw_no_cuda(); return Scalar(); }
+Scalar cv::gpu::sqrSum(const GpuMat&, GpuMat&) { throw_no_cuda(); return Scalar(); }
+Scalar cv::gpu::sqrSum(const GpuMat&, const GpuMat&, GpuMat&) { throw_no_cuda(); return Scalar(); }
+void cv::gpu::minMax(const GpuMat&, double*, double*, const GpuMat&) { throw_no_cuda(); }
+void cv::gpu::minMax(const GpuMat&, double*, double*, const GpuMat&, GpuMat&) { throw_no_cuda(); }
+void cv::gpu::minMaxLoc(const GpuMat&, double*, double*, Point*, Point*, const GpuMat&) { throw_no_cuda(); }
+void cv::gpu::minMaxLoc(const GpuMat&, double*, double*, Point*, Point*, const GpuMat&, GpuMat&, GpuMat&) { throw_no_cuda(); }
+int cv::gpu::countNonZero(const GpuMat&) { throw_no_cuda(); return 0; }
+int cv::gpu::countNonZero(const GpuMat&, GpuMat&) { throw_no_cuda(); return 0; }
+void cv::gpu::reduce(const GpuMat&, GpuMat&, int, int, int, Stream&) { throw_no_cuda(); }
+
+#else
+#include "opencv2/core/utility.hpp"
+
+namespace
+{
+    class DeviceBuffer
+    {
+    public:
+        explicit DeviceBuffer(int count_ = 1) : count(count_)
+        {
+            cudaSafeCall( cudaMalloc(&pdev, count * sizeof(double)) );
+        }
+        ~DeviceBuffer()
+        {
+            cudaSafeCall( cudaFree(pdev) );
+        }
+
+        operator double*() {return pdev;}
+
+        void download(double* hptr)
+        {
+            double hbuf;
+            cudaSafeCall( cudaMemcpy(&hbuf, pdev, sizeof(double), cudaMemcpyDeviceToHost) );
+            *hptr = hbuf;
+        }
+        void download(double** hptrs)
+        {
+            AutoBuffer<double, 2 * sizeof(double)> hbuf(count);
+            cudaSafeCall( cudaMemcpy((void*)hbuf, pdev, count * sizeof(double), cudaMemcpyDeviceToHost) );
+            for (int i = 0; i < count; ++i)
+                *hptrs[i] = hbuf[i];
+        }
+
+    private:
+        double* pdev;
+        int count;
+    };
+}
+
+
+////////////////////////////////////////////////////////////////////////
+// meanStdDev
+
+void cv::gpu::meanStdDev(const GpuMat& src, Scalar& mean, Scalar& stddev)
+{
+    GpuMat buf;
+    meanStdDev(src, mean, stddev, buf);
+}
+
+void cv::gpu::meanStdDev(const GpuMat& src, Scalar& mean, Scalar& stddev, GpuMat& buf)
+{
+    CV_Assert(src.type() == CV_8UC1);
+
+    if (!deviceSupports(FEATURE_SET_COMPUTE_13))
+        CV_Error(cv::Error::StsNotImplemented, "Not sufficient compute capebility");
+
+    NppiSize sz;
+    sz.width  = src.cols;
+    sz.height = src.rows;
+
+    DeviceBuffer dbuf(2);
+
+    int bufSize;
+#if (CUDA_VERSION <= 4020)
+    nppSafeCall( nppiMeanStdDev8uC1RGetBufferHostSize(sz, &bufSize) );
+#else
+    nppSafeCall( nppiMeanStdDevGetBufferHostSize_8u_C1R(sz, &bufSize) );
+#endif
+
+    ensureSizeIsEnough(1, bufSize, CV_8UC1, buf);
+
+    nppSafeCall( nppiMean_StdDev_8u_C1R(src.ptr<Npp8u>(), static_cast<int>(src.step), sz, buf.ptr<Npp8u>(), dbuf, (double*)dbuf + 1) );
+
+    cudaSafeCall( cudaDeviceSynchronize() );
+
+    double* ptrs[2] = {mean.val, stddev.val};
+    dbuf.download(ptrs);
+}
+
+////////////////////////////////////////////////////////////////////////
+// norm
+
+double cv::gpu::norm(const GpuMat& src, int normType)
+{
+    GpuMat buf;
+    return norm(src, normType, GpuMat(), buf);
+}
+
+double cv::gpu::norm(const GpuMat& src, int normType, GpuMat& buf)
+{
+    return norm(src, normType, GpuMat(), buf);
+}
+
+double cv::gpu::norm(const GpuMat& src, int normType, const GpuMat& mask, GpuMat& buf)
+{
+    CV_Assert(normType == NORM_INF || normType == NORM_L1 || normType == NORM_L2);
+    CV_Assert(mask.empty() || (mask.type() == CV_8UC1 && mask.size() == src.size() && src.channels() == 1));
+
+    GpuMat src_single_channel = src.reshape(1);
+
+    if (normType == NORM_L1)
+        return absSum(src_single_channel, mask, buf)[0];
+
+    if (normType == NORM_L2)
+        return std::sqrt(sqrSum(src_single_channel, mask, buf)[0]);
+
+    // NORM_INF
+    double min_val, max_val;
+    minMax(src_single_channel, &min_val, &max_val, mask, buf);
+    return std::max(std::abs(min_val), std::abs(max_val));
+}
+
+double cv::gpu::norm(const GpuMat& src1, const GpuMat& src2, int normType)
+{
+    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);
+
+    typedef NppStatus (*npp_norm_diff_func_t)(const Npp8u* pSrc1, int nSrcStep1, const Npp8u* pSrc2, int nSrcStep2,
+        NppiSize oSizeROI, Npp64f* pRetVal);
+
+    static const npp_norm_diff_func_t npp_norm_diff_func[] = {nppiNormDiff_Inf_8u_C1R, nppiNormDiff_L1_8u_C1R, nppiNormDiff_L2_8u_C1R};
+
+    NppiSize sz;
+    sz.width  = src1.cols;
+    sz.height = src1.rows;
+
+    int funcIdx = normType >> 1;
+
+    double retVal;
+
+    DeviceBuffer dbuf;
+
+    nppSafeCall( npp_norm_diff_func[funcIdx](src1.ptr<Npp8u>(), static_cast<int>(src1.step), src2.ptr<Npp8u>(), static_cast<int>(src2.step), sz, dbuf) );
+
+    cudaSafeCall( cudaDeviceSynchronize() );
+
+    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::gpu::sum(const GpuMat& src)
+{
+    GpuMat buf;
+    return sum(src, GpuMat(), buf);
+}
+
+Scalar cv::gpu::sum(const GpuMat& src, GpuMat& buf)
+{
+    return sum(src, GpuMat(), buf);
+}
+
+Scalar cv::gpu::sum(const GpuMat& src, const GpuMat& mask, GpuMat& buf)
+{
+    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::gpu::absSum(const GpuMat& src)
+{
+    GpuMat buf;
+    return absSum(src, GpuMat(), buf);
+}
+
+Scalar cv::gpu::absSum(const GpuMat& src, GpuMat& buf)
+{
+    return absSum(src, GpuMat(), buf);
+}
+
+Scalar cv::gpu::absSum(const GpuMat& src, const GpuMat& mask, GpuMat& buf)
+{
+    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::gpu::sqrSum(const GpuMat& src)
+{
+    GpuMat buf;
+    return sqrSum(src, GpuMat(), buf);
+}
+
+Scalar cv::gpu::sqrSum(const GpuMat& src, GpuMat& buf)
+{
+    return sqrSum(src, GpuMat(), buf);
+}
+
+Scalar cv::gpu::sqrSum(const GpuMat& src, const GpuMat& mask, GpuMat& buf)
+{
+    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::gpu::minMax(const GpuMat& src, double* minVal, double* maxVal, const GpuMat& mask)
+{
+    GpuMat buf;
+    minMax(src, minVal, maxVal, mask, buf);
+}
+
+void cv::gpu::minMax(const GpuMat& src, double* minVal, double* maxVal, const GpuMat& mask, GpuMat& buf)
+{
+    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::gpu::minMaxLoc(const GpuMat& src, double* minVal, double* maxVal, Point* minLoc, Point* maxLoc, const GpuMat& mask)
+{
+    GpuMat valBuf, locBuf;
+    minMaxLoc(src, minVal, maxVal, minLoc, maxLoc, mask, valBuf, locBuf);
+}
+
+void cv::gpu::minMaxLoc(const GpuMat& src, double* minVal, double* maxVal, Point* minLoc, Point* maxLoc,
+                        const GpuMat& mask, GpuMat& valBuf, GpuMat& locBuf)
+{
+    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::gpu::countNonZero(const GpuMat& src)
+{
+    GpuMat buf;
+    return countNonZero(src, buf);
+}
+
+int cv::gpu::countNonZero(const GpuMat& src, GpuMat& buf)
+{
+    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::gpu::reduce(const GpuMat& src, GpuMat& dst, int dim, int reduceOp, int dtype, Stream& stream)
+{
+    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()));
+
+    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));
+    }
+}
+
+#endif

modules/gpuarithm/src/precomp.cpp

@@ -0,0 +1,43 @@
+/*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 "precomp.hpp"

modules/gpuarithm/src/precomp.hpp

@@ -0,0 +1,58 @@
+/*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 __OPENCV_PRECOMP_H__
+#define __OPENCV_PRECOMP_H__
+
+#include <limits>
+
+#include "opencv2/gpuarithm.hpp"
+#include "opencv2/core/utility.hpp"
+#include "opencv2/core/core_c.h"
+
+#include "opencv2/core/gpu_private.hpp"
+
+#ifdef HAVE_CUBLAS
+    #include <cublas.h>
+#endif
+
+#endif /* __OPENCV_PRECOMP_H__ */

modules/gpuarithm/src/split_merge.cpp

@@ -0,0 +1,171 @@
+/*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 "precomp.hpp"
+
+using namespace cv;
+using namespace cv::gpu;
+
+#if !defined (HAVE_CUDA) || defined (CUDA_DISABLER)
+
+void cv::gpu::merge(const GpuMat* /*src*/, size_t /*count*/, GpuMat& /*dst*/, Stream& /*stream*/) { throw_no_cuda(); }
+void cv::gpu::merge(const std::vector<GpuMat>& /*src*/, GpuMat& /*dst*/, Stream& /*stream*/) { throw_no_cuda(); }
+void cv::gpu::split(const GpuMat& /*src*/, GpuMat* /*dst*/, Stream& /*stream*/) { throw_no_cuda(); }
+void cv::gpu::split(const GpuMat& /*src*/, std::vector<GpuMat>& /*dst*/, Stream& /*stream*/) { throw_no_cuda(); }
+
+#else /* !defined (HAVE_CUDA) */
+
+namespace cv { namespace gpu { namespace cudev
+{
+    namespace split_merge
+    {
+        void merge_caller(const PtrStepSzb* src, PtrStepSzb& dst, int total_channels, size_t elem_size, const cudaStream_t& stream);
+        void split_caller(const PtrStepSzb& src, PtrStepSzb* dst, int num_channels, size_t elem_size1, const cudaStream_t& stream);
+    }
+}}}
+
+namespace
+{
+    void merge(const GpuMat* src, size_t n, GpuMat& dst, const cudaStream_t& stream)
+    {
+        using namespace ::cv::gpu::cudev::split_merge;
+
+        CV_Assert(src);
+        CV_Assert(n > 0);
+
+        int depth = src[0].depth();
+        Size size = src[0].size();
+
+        if (depth == CV_64F)
+        {
+            if (!deviceSupports(NATIVE_DOUBLE))
+                CV_Error(cv::Error::StsUnsupportedFormat, "The device doesn't support double");
+        }
+
+        bool single_channel_only = true;
+        int total_channels = 0;
+
+        for (size_t i = 0; i < n; ++i)
+        {
+            CV_Assert(src[i].size() == size);
+            CV_Assert(src[i].depth() == depth);
+            single_channel_only = single_channel_only && src[i].channels() == 1;
+            total_channels += src[i].channels();
+        }
+
+        CV_Assert(single_channel_only);
+        CV_Assert(total_channels <= 4);
+
+        if (total_channels == 1)
+            src[0].copyTo(dst);
+        else
+        {
+            dst.create(size, CV_MAKETYPE(depth, total_channels));
+
+            PtrStepSzb src_as_devmem[4];
+            for(size_t i = 0; i < n; ++i)
+                src_as_devmem[i] = src[i];
+
+            PtrStepSzb dst_as_devmem(dst);
+            merge_caller(src_as_devmem, dst_as_devmem, total_channels, CV_ELEM_SIZE(depth), stream);
+        }
+    }
+
+    void split(const GpuMat& src, GpuMat* dst, const cudaStream_t& stream)
+    {
+        using namespace ::cv::gpu::cudev::split_merge;
+
+        CV_Assert(dst);
+
+        int depth = src.depth();
+        int num_channels = src.channels();
+
+        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]);
+            return;
+        }
+
+        for (int i = 0; i < num_channels; ++i)
+            dst[i].create(src.size(), depth);
+
+        CV_Assert(num_channels <= 4);
+
+        PtrStepSzb dst_as_devmem[4];
+        for (int i = 0; i < num_channels; ++i)
+            dst_as_devmem[i] = dst[i];
+
+        PtrStepSzb src_as_devmem(src);
+        split_caller(src_as_devmem, dst_as_devmem, num_channels, src.elemSize1(), stream);
+    }
+}
+
+void cv::gpu::merge(const GpuMat* src, size_t n, GpuMat& dst, Stream& stream)
+{
+    ::merge(src, n, dst, StreamAccessor::getStream(stream));
+}
+
+
+void cv::gpu::merge(const std::vector<GpuMat>& src, GpuMat& dst, Stream& stream)
+{
+    ::merge(&src[0], src.size(), dst, StreamAccessor::getStream(stream));
+}
+
+void cv::gpu::split(const GpuMat& src, GpuMat* dst, Stream& stream)
+{
+    ::split(src, dst, StreamAccessor::getStream(stream));
+}
+
+void cv::gpu::split(const GpuMat& src, std::vector<GpuMat>& dst, Stream& stream)
+{
+    dst.resize(src.channels());
+    if(src.channels() > 0)
+        ::split(src, &dst[0], StreamAccessor::getStream(stream));
+}
+
+#endif /* !defined (HAVE_CUDA) */

modules/gpuarithm/test/test_main.cpp

@@ -0,0 +1,120 @@
+/*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 "test_precomp.hpp"
+
+#ifdef HAVE_CUDA
+
+using namespace std;
+using namespace cv;
+using namespace cv::gpu;
+using namespace cvtest;
+using namespace testing;
+
+int main(int argc, char** argv)
+{
+    try
+    {
+        const std::string keys =
+                "{ h help ?            |      | Print help}"
+                "{ i info              |      | Print information about system and exit }"
+                "{ device              | -1   | Device on which tests will be executed (-1 means all devices) }"
+                ;
+
+        CommandLineParser cmd(argc, (const char**)argv, keys);
+
+        if (cmd.has("help"))
+        {
+            cmd.printMessage();
+            return 0;
+        }
+
+        printCudaInfo();
+
+        if (cmd.has("info"))
+        {
+            return 0;
+        }
+
+        int device = cmd.get<int>("device");
+        if (device < 0)
+        {
+            DeviceManager::instance().loadAll();
+
+            cout << "Run tests on all supported devices \n" << endl;
+        }
+        else
+        {
+            DeviceManager::instance().load(device);
+
+            DeviceInfo info(device);
+            cout << "Run tests on device " << device << " [" << info.name() << "] \n" << endl;
+        }
+
+        TS::ptr()->init("gpu");
+        InitGoogleTest(&argc, argv);
+
+        return RUN_ALL_TESTS();
+    }
+    catch (const exception& e)
+    {
+        cerr << e.what() << endl;
+        return -1;
+    }
+    catch (...)
+    {
+        cerr << "Unknown error" << endl;
+        return -1;
+    }
+
+    return 0;
+}
+
+#else // HAVE_CUDA
+
+int main()
+{
+    printf("OpenCV was built without CUDA support\n");
+    return 0;
+}
+
+#endif // HAVE_CUDA

modules/gpuarithm/test/test_precomp.cpp

@@ -0,0 +1,43 @@
+/*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 "test_precomp.hpp"

modules/gpuarithm/test/test_precomp.hpp

@@ -0,0 +1,60 @@
+/*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*/
+
+#ifdef __GNUC__
+#  pragma GCC diagnostic ignored "-Wmissing-declarations"
+#  if defined __clang__ || defined __APPLE__
+#    pragma GCC diagnostic ignored "-Wmissing-prototypes"
+#    pragma GCC diagnostic ignored "-Wextra"
+#  endif
+#endif
+
+#ifndef __OPENCV_TEST_PRECOMP_HPP__
+#define __OPENCV_TEST_PRECOMP_HPP__
+
+#include "opencv2/ts.hpp"
+#include "opencv2/ts/gpu_test.hpp"
+
+#include "opencv2/core.hpp"
+#include "opencv2/gpuarithm.hpp"
+
+#endif