Merge pull request #1083 from jet47:cuda-device-layer

modules/cudev/CMakeLists.txt

@@ -0,0 +1,36 @@
+if(NOT HAVE_CUDA)
+  ocv_module_disable(cudev)
+endif()
+
+set(the_description "CUDA device layer")
+
+ocv_warnings_disable(CMAKE_CXX_FLAGS /wd4189 /wd4505 -Wundef -Wmissing-declarations -Wunused-function -Wunused-variable)
+
+ocv_add_module(cudev)
+
+file(GLOB_RECURSE lib_hdrs "include/opencv2/*.hpp")
+
+add_custom_target(${the_module} SOURCES ${lib_hdrs})
+
+if(ENABLE_SOLUTION_FOLDERS)
+  set_target_properties(${the_module} PROPERTIES FOLDER "modules")
+endif()
+
+foreach(var CMAKE_CXX_FLAGS CMAKE_CXX_FLAGS_RELEASE CMAKE_CXX_FLAGS_DEBUG)
+  # we remove /EHa as it generates warnings under windows
+  string(REPLACE "/EHa" "" ${var} "${${var}}")
+
+  # we remove -ggdb3 flag as it leads to preprocessor errors when compiling CUDA files (CUDA 4.1)
+  string(REPLACE "-ggdb3" "" ${var} "${${var}}")
+
+  # we remove -Wsign-promo as it generates warnings under linux
+  string(REPLACE "-Wsign-promo" "" ${var} "${${var}}")
+
+  # we remove -fvisibility-inlines-hidden because it's used for C++ compiler
+  # but NVCC uses C compiler by default
+  string(REPLACE "-fvisibility-inlines-hidden" "" ${var} "${${var}}")
+endforeach()
+
+if(BUILD_TESTS)
+  add_subdirectory(test)
+endif()

modules/cudev/include/opencv2/cudev.hpp

@@ -0,0 +1,112 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_HPP__
+#define __OPENCV_CUDEV_HPP__
+
+#include "cudev/common.hpp"
+
+#include "cudev/util/atomic.hpp"
+#include "cudev/util/limits.hpp"
+#include "cudev/util/saturate_cast.hpp"
+#include "cudev/util/simd_functions.hpp"
+#include "cudev/util/tuple.hpp"
+#include "cudev/util/type_traits.hpp"
+#include "cudev/util/vec_math.hpp"
+#include "cudev/util/vec_traits.hpp"
+
+#include "cudev/functional/color_cvt.hpp"
+#include "cudev/functional/functional.hpp"
+#include "cudev/functional/tuple_adapter.hpp"
+
+#include "cudev/warp/reduce.hpp"
+#include "cudev/warp/scan.hpp"
+#include "cudev/warp/shuffle.hpp"
+#include "cudev/warp/warp.hpp"
+
+#include "cudev/block/block.hpp"
+#include "cudev/block/dynamic_smem.hpp"
+#include "cudev/block/reduce.hpp"
+#include "cudev/block/scan.hpp"
+#include "cudev/block/vec_distance.hpp"
+
+#include "cudev/grid/copy.hpp"
+#include "cudev/grid/glob_reduce.hpp"
+#include "cudev/grid/histogram.hpp"
+#include "cudev/grid/integral.hpp"
+#include "cudev/grid/pyramids.hpp"
+#include "cudev/grid/reduce_to_vec.hpp"
+#include "cudev/grid/split_merge.hpp"
+#include "cudev/grid/transform.hpp"
+#include "cudev/grid/transpose.hpp"
+
+#include "cudev/ptr2d/constant.hpp"
+#include "cudev/ptr2d/deriv.hpp"
+#include "cudev/ptr2d/extrapolation.hpp"
+#include "cudev/ptr2d/glob.hpp"
+#include "cudev/ptr2d/gpumat.hpp"
+#include "cudev/ptr2d/interpolation.hpp"
+#include "cudev/ptr2d/lut.hpp"
+#include "cudev/ptr2d/mask.hpp"
+#include "cudev/ptr2d/remap.hpp"
+#include "cudev/ptr2d/resize.hpp"
+#include "cudev/ptr2d/texture.hpp"
+#include "cudev/ptr2d/traits.hpp"
+#include "cudev/ptr2d/transform.hpp"
+#include "cudev/ptr2d/warping.hpp"
+#include "cudev/ptr2d/zip.hpp"
+
+#include "cudev/expr/binary_func.hpp"
+#include "cudev/expr/binary_op.hpp"
+#include "cudev/expr/color.hpp"
+#include "cudev/expr/deriv.hpp"
+#include "cudev/expr/expr.hpp"
+#include "cudev/expr/per_element_func.hpp"
+#include "cudev/expr/reduction.hpp"
+#include "cudev/expr/unary_func.hpp"
+#include "cudev/expr/unary_op.hpp"
+#include "cudev/expr/warping.hpp"
+
+#endif

modules/cudev/include/opencv2/cudev/block/block.hpp

@@ -0,0 +1,127 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_BLOCK_BLOCK_HPP__
+#define __OPENCV_CUDEV_BLOCK_BLOCK_HPP__
+
+#include "../common.hpp"
+
+namespace cv { namespace cudev {
+
+struct Block
+{
+    __device__ __forceinline__ static uint blockId()
+    {
+        return (blockIdx.z * gridDim.y + blockIdx.y) * gridDim.x + blockIdx.x;
+    }
+
+    __device__ __forceinline__ static uint blockSize()
+    {
+        return blockDim.x * blockDim.y * blockDim.z;
+    }
+
+    __device__ __forceinline__ static uint threadLineId()
+    {
+        return (threadIdx.z * blockDim.y + threadIdx.y) * blockDim.x + threadIdx.x;
+    }
+};
+
+template <class It, typename T>
+__device__ __forceinline__ static void blockFill(It beg, It end, const T& value)
+{
+    uint STRIDE = Block::blockSize();
+    It t = beg + Block::threadLineId();
+
+    for(; t < end; t += STRIDE)
+        *t = value;
+}
+
+template <class OutIt, typename T>
+__device__ __forceinline__ static void blockYota(OutIt beg, OutIt end, T value)
+{
+    uint STRIDE = Block::blockSize();
+    uint tid = Block::threadLineId();
+    value += tid;
+
+    for(OutIt t = beg + tid; t < end; t += STRIDE, value += STRIDE)
+        *t = value;
+}
+
+template <class InIt, class OutIt>
+__device__ __forceinline__ static void blockCopy(InIt beg, InIt end, OutIt out)
+{
+    uint STRIDE = Block::blockSize();
+    InIt  t = beg + Block::threadLineId();
+    OutIt o = out + (t - beg);
+
+    for(; t < end; t += STRIDE, o += STRIDE)
+        *o = *t;
+}
+
+template <class InIt, class OutIt, class UnOp>
+__device__ __forceinline__ static void blockTransfrom(InIt beg, InIt end, OutIt out, const UnOp& op)
+{
+    uint STRIDE = Block::blockSize();
+    InIt  t = beg + Block::threadLineId();
+    OutIt o = out + (t - beg);
+
+    for(; t < end; t += STRIDE, o += STRIDE)
+        *o = op(*t);
+}
+
+template <class InIt1, class InIt2, class OutIt, class BinOp>
+__device__ __forceinline__ static void blockTransfrom(InIt1 beg1, InIt1 end1, InIt2 beg2, OutIt out, const BinOp& op)
+{
+    uint STRIDE = Block::blockSize();
+    InIt1 t1 = beg1 + Block::threadLineId();
+    InIt2 t2 = beg2 + Block::threadLineId();
+    OutIt o  = out + (t1 - beg1);
+
+    for(; t1 < end1; t1 += STRIDE, t2 += STRIDE, o += STRIDE)
+        *o = op(*t1, *t2);
+}
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/block/detail/reduce.hpp

@@ -0,0 +1,392 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_BLOCK_REDUCE_DETAIL_HPP__
+#define __OPENCV_CUDEV_BLOCK_REDUCE_DETAIL_HPP__
+
+#include "../../common.hpp"
+#include "../../util/tuple.hpp"
+#include "../../util/type_traits.hpp"
+#include "../../warp/warp.hpp"
+#include "../../warp/shuffle.hpp"
+
+namespace cv { namespace cudev {
+
+namespace block_reduce_detail
+{
+    // GetType
+
+    template <typename T> struct GetType;
+
+    template <typename T> struct GetType<T*>
+    {
+        typedef T type;
+    };
+
+    template <typename T> struct GetType<volatile T*>
+    {
+        typedef T type;
+    };
+
+    template <typename T> struct GetType<T&>
+    {
+        typedef T type;
+    };
+
+    // For
+
+    template <int I, int N> struct For
+    {
+        template <class PointerTuple, class ValTuple>
+        __device__ static void loadToSmem(const PointerTuple& smem, const ValTuple& val, uint tid)
+        {
+            get<I>(smem)[tid] = get<I>(val);
+
+            For<I + 1, N>::loadToSmem(smem, val, tid);
+        }
+
+        template <class PointerTuple, class ValTuple>
+        __device__ static void loadFromSmem(const PointerTuple& smem, const ValTuple& val, uint tid)
+        {
+            get<I>(val) = get<I>(smem)[tid];
+
+            For<I + 1, N>::loadFromSmem(smem, val, tid);
+        }
+
+        template <class PointerTuple, class ValTuple, class OpTuple>
+        __device__ static void merge(const PointerTuple& smem, const ValTuple& val, uint tid, uint delta, const OpTuple& op)
+        {
+            typename GetType<typename tuple_element<I, PointerTuple>::type>::type reg = get<I>(smem)[tid + delta];
+            get<I>(smem)[tid] = get<I>(val) = get<I>(op)(get<I>(val), reg);
+
+            For<I + 1, N>::merge(smem, val, tid, delta, op);
+        }
+
+#if CV_CUDEV_ARCH >= 300
+        template <class ValTuple, class OpTuple>
+        __device__ static void mergeShfl(const ValTuple& val, uint delta, uint width, const OpTuple& op)
+        {
+            typename GetType<typename tuple_element<I, ValTuple>::type>::type reg = shfl_down(get<I>(val), delta, width);
+            get<I>(val) = get<I>(op)(get<I>(val), reg);
+
+            For<I + 1, N>::mergeShfl(val, delta, width, op);
+        }
+#endif
+    };
+
+    template <int N> struct For<N, N>
+    {
+        template <class PointerTuple, class ValTuple>
+        __device__ __forceinline__ static void loadToSmem(const PointerTuple&, const ValTuple&, uint)
+        {
+        }
+        template <class PointerTuple, class ValTuple>
+        __device__ __forceinline__ static void loadFromSmem(const PointerTuple&, const ValTuple&, uint)
+        {
+        }
+
+        template <class PointerTuple, class ValTuple, class OpTuple>
+        __device__ __forceinline__ static void merge(const PointerTuple&, const ValTuple&, uint, uint, const OpTuple&)
+        {
+        }
+
+#if CV_CUDEV_ARCH >= 300
+        template <class ValTuple, class OpTuple>
+        __device__ __forceinline__ static void mergeShfl(const ValTuple&, uint, uint, const OpTuple&)
+        {
+        }
+#endif
+    };
+
+    // loadToSmem / loadFromSmem
+
+    template <typename T>
+    __device__ __forceinline__ void loadToSmem(volatile T* smem, T& val, uint tid)
+    {
+        smem[tid] = val;
+    }
+
+    template <typename T>
+    __device__ __forceinline__ void loadFromSmem(volatile T* smem, T& val, uint tid)
+    {
+        val = smem[tid];
+    }
+
+    template <typename P0, typename P1, typename P2, typename P3, typename P4, typename P5, typename P6, typename P7, typename P8, typename P9,
+              typename R0, typename R1, typename R2, typename R3, typename R4, typename R5, typename R6, typename R7, typename R8, typename R9>
+    __device__ __forceinline__ void loadToSmem(const tuple<P0, P1, P2, P3, P4, P5, P6, P7, P8, P9>& smem,
+                                               const tuple<R0, R1, R2, R3, R4, R5, R6, R7, R8, R9>& val,
+                                               uint tid)
+    {
+        For<0, tuple_size<tuple<P0, P1, P2, P3, P4, P5, P6, P7, P8, P9> >::value>::loadToSmem(smem, val, tid);
+    }
+
+    template <typename P0, typename P1, typename P2, typename P3, typename P4, typename P5, typename P6, typename P7, typename P8, typename P9,
+              typename R0, typename R1, typename R2, typename R3, typename R4, typename R5, typename R6, typename R7, typename R8, typename R9>
+    __device__ __forceinline__ void loadFromSmem(const tuple<P0, P1, P2, P3, P4, P5, P6, P7, P8, P9>& smem,
+                                                     const tuple<R0, R1, R2, R3, R4, R5, R6, R7, R8, R9>& val,
+                                                     uint tid)
+    {
+        For<0, tuple_size<tuple<P0, P1, P2, P3, P4, P5, P6, P7, P8, P9> >::value>::loadFromSmem(smem, val, tid);
+    }
+
+    // merge
+
+    template <typename T, class Op>
+    __device__ __forceinline__ void merge(volatile T* smem, T& val, uint tid, uint delta, const Op& op)
+    {
+        T reg = smem[tid + delta];
+        smem[tid] = val = op(val, reg);
+    }
+
+    template <typename P0, typename P1, typename P2, typename P3, typename P4, typename P5, typename P6, typename P7, typename P8, typename P9,
+              typename R0, typename R1, typename R2, typename R3, typename R4, typename R5, typename R6, typename R7, typename R8, typename R9,
+              class Op0, class Op1, class Op2, class Op3, class Op4, class Op5, class Op6, class Op7, class Op8, class Op9>
+    __device__ __forceinline__ void merge(const tuple<P0, P1, P2, P3, P4, P5, P6, P7, P8, P9>& smem,
+                                          const tuple<R0, R1, R2, R3, R4, R5, R6, R7, R8, R9>& val,
+                                          uint tid,
+                                          uint delta,
+                                          const tuple<Op0, Op1, Op2, Op3, Op4, Op5, Op6, Op7, Op8, Op9>& op)
+    {
+        For<0, tuple_size<tuple<P0, P1, P2, P3, P4, P5, P6, P7, P8, P9> >::value>::merge(smem, val, tid, delta, op);
+    }
+
+    // mergeShfl
+
+#if CV_CUDEV_ARCH >= 300
+    template <typename T, class Op>
+    __device__ __forceinline__ void mergeShfl(T& val, uint delta, uint width, const Op& op)
+    {
+        T reg = shfl_down(val, delta, width);
+        val = op(val, reg);
+    }
+
+    template <typename R0, typename R1, typename R2, typename R3, typename R4, typename R5, typename R6, typename R7, typename R8, typename R9,
+              class Op0, class Op1, class Op2, class Op3, class Op4, class Op5, class Op6, class Op7, class Op8, class Op9>
+    __device__ __forceinline__ void mergeShfl(const tuple<R0, R1, R2, R3, R4, R5, R6, R7, R8, R9>& val,
+                                              uint delta,
+                                              uint width,
+                                              const tuple<Op0, Op1, Op2, Op3, Op4, Op5, Op6, Op7, Op8, Op9>& op)
+    {
+        For<0, tuple_size<tuple<R0, R1, R2, R3, R4, R5, R6, R7, R8, R9> >::value>::mergeShfl(val, delta, width, op);
+    }
+#endif
+
+    // Generic
+
+    template <int N> struct Generic
+    {
+        template <typename Pointer, typename Reference, class Op>
+        __device__ static void reduce(Pointer smem, Reference val, uint tid, Op op)
+        {
+            loadToSmem(smem, val, tid);
+            if (N >= 32)
+                __syncthreads();
+
+            if (N >= 2048)
+            {
+                if (tid < 1024)
+                    merge(smem, val, tid, 1024, op);
+
+                __syncthreads();
+            }
+            if (N >= 1024)
+            {
+                if (tid < 512)
+                    merge(smem, val, tid, 512, op);
+
+                __syncthreads();
+            }
+            if (N >= 512)
+            {
+                if (tid < 256)
+                    merge(smem, val, tid, 256, op);
+
+                __syncthreads();
+            }
+            if (N >= 256)
+            {
+                if (tid < 128)
+                    merge(smem, val, tid, 128, op);
+
+                __syncthreads();
+            }
+            if (N >= 128)
+            {
+                if (tid < 64)
+                    merge(smem, val, tid, 64, op);
+
+                __syncthreads();
+            }
+            if (N >= 64)
+            {
+                if (tid < 32)
+                    merge(smem, val, tid, 32, op);
+            }
+
+            if (tid < 16)
+            {
+                merge(smem, val, tid, 16, op);
+                merge(smem, val, tid, 8, op);
+                merge(smem, val, tid, 4, op);
+                merge(smem, val, tid, 2, op);
+                merge(smem, val, tid, 1, op);
+            }
+        }
+    };
+
+    // Unroll
+
+    template <int I, typename Pointer, typename Reference, class Op> struct Unroll
+    {
+        __device__ static void loop(Pointer smem, Reference val, uint tid, Op op)
+        {
+            merge(smem, val, tid, I, op);
+            Unroll<I / 2, Pointer, Reference, Op>::loop(smem, val, tid, op);
+        }
+
+#if CV_CUDEV_ARCH >= 300
+        __device__ static void loopShfl(Reference val, Op op, uint N)
+        {
+            mergeShfl(val, I, N, op);
+            Unroll<I / 2, Pointer, Reference, Op>::loopShfl(val, op, N);
+        }
+#endif
+    };
+
+    template <typename Pointer, typename Reference, class Op> struct Unroll<0, Pointer, Reference, Op>
+    {
+        __device__ __forceinline__ static void loop(Pointer, Reference, uint, Op)
+        {
+        }
+
+#if CV_CUDEV_ARCH >= 300
+        __device__ __forceinline__ static void loopShfl(Reference, Op, uint)
+        {
+        }
+#endif
+    };
+
+    // WarpOptimized
+
+    template <int N> struct WarpOptimized
+    {
+        template <typename Pointer, typename Reference, class Op>
+        __device__ static void reduce(Pointer smem, Reference val, uint tid, Op op)
+        {
+        #if CV_CUDEV_ARCH >= 300
+            (void) smem;
+            (void) tid;
+
+            Unroll<N / 2, Pointer, Reference, Op>::loopShfl(val, op, N);
+        #else
+            loadToSmem(smem, val, tid);
+
+            if (tid < N / 2)
+                Unroll<N / 2, Pointer, Reference, Op>::loop(smem, val, tid, op);
+        #endif
+        }
+    };
+
+    // GenericOptimized32
+
+    template <int N> struct GenericOptimized32
+    {
+        enum { M = N / 32 };
+
+        template <typename Pointer, typename Reference, class Op>
+        __device__ static void reduce(Pointer smem, Reference val, uint tid, Op op)
+        {
+            const uint laneId = Warp::laneId();
+
+        #if CV_CUDEV_ARCH >= 300
+            Unroll<16, Pointer, Reference, Op>::loopShfl(val, op, warpSize);
+
+            if (laneId == 0)
+                loadToSmem(smem, val, tid / 32);
+        #else
+            loadToSmem(smem, val, tid);
+
+            if (laneId < 16)
+                Unroll<16, Pointer, Reference, Op>::loop(smem, val, tid, op);
+
+            __syncthreads();
+
+            if (laneId == 0)
+                loadToSmem(smem, val, tid / 32);
+        #endif
+
+            __syncthreads();
+
+            loadFromSmem(smem, val, tid);
+
+            if (tid < 32)
+            {
+        #if CV_CUDEV_ARCH >= 300
+                Unroll<M / 2, Pointer, Reference, Op>::loopShfl(val, op, M);
+        #else
+                Unroll<M / 2, Pointer, Reference, Op>::loop(smem, val, tid, op);
+        #endif
+            }
+        }
+    };
+
+    template <int N> struct Dispatcher
+    {
+        typedef typename SelectIf<
+            (N <= 32) && IsPowerOf2<N>::value,
+            WarpOptimized<N>,
+            typename SelectIf<
+                (N <= 1024) && IsPowerOf2<N>::value,
+                GenericOptimized32<N>,
+                Generic<N>
+            >::type
+        >::type reductor;
+    };
+}
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/block/detail/reduce_key_val.hpp

@@ -0,0 +1,394 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_BLOCK_REDUCE_KEY_VAL_DETAIL_HPP__
+#define __OPENCV_CUDEV_BLOCK_REDUCE_KEY_VAL_DETAIL_HPP__
+
+#include "../../common.hpp"
+#include "../../util/tuple.hpp"
+#include "../../util/type_traits.hpp"
+#include "../../warp/warp.hpp"
+
+namespace cv { namespace cudev {
+
+namespace block_reduce_key_val_detail
+{
+    // GetType
+
+    template <typename T> struct GetType;
+
+    template <typename T> struct GetType<T*>
+    {
+        typedef T type;
+    };
+
+    template <typename T> struct GetType<volatile T*>
+    {
+        typedef T type;
+    };
+
+    template <typename T> struct GetType<T&>
+    {
+        typedef T type;
+    };
+
+    // For
+
+    template <int I, int N> struct For
+    {
+        template <class PointerTuple, class ReferenceTuple>
+        __device__ static void loadToSmem(const PointerTuple& smem, const ReferenceTuple& data, uint tid)
+        {
+            get<I>(smem)[tid] = get<I>(data);
+
+            For<I + 1, N>::loadToSmem(smem, data, tid);
+        }
+
+        template <class PointerTuple, class ReferenceTuple>
+        __device__ static void loadFromSmem(const PointerTuple& smem, const ReferenceTuple& data, uint tid)
+        {
+            get<I>(data) = get<I>(smem)[tid];
+
+            For<I + 1, N>::loadFromSmem(smem, data, tid);
+        }
+
+        template <class PointerTuple, class ReferenceTuple>
+        __device__ static void copy(const PointerTuple& svals, const ReferenceTuple& val, uint tid, uint delta)
+        {
+            get<I>(svals)[tid] = get<I>(val) = get<I>(svals)[tid + delta];
+
+            For<I + 1, N>::copy(svals, val, tid, delta);
+        }
+
+        template <class KeyPointerTuple, class KeyReferenceTuple, class ValPointerTuple, class ValReferenceTuple, class CmpTuple>
+        __device__ static void merge(const KeyPointerTuple& skeys, const KeyReferenceTuple& key,
+                                     const ValPointerTuple& svals, const ValReferenceTuple& val,
+                                     const CmpTuple& cmp,
+                                     uint tid, uint delta)
+        {
+            typename GetType<typename tuple_element<I, KeyPointerTuple>::type>::type reg = get<I>(skeys)[tid + delta];
+
+            if (get<I>(cmp)(reg, get<I>(key)))
+            {
+                get<I>(skeys)[tid] = get<I>(key) = reg;
+                get<I>(svals)[tid] = get<I>(val) = get<I>(svals)[tid + delta];
+            }
+
+            For<I + 1, N>::merge(skeys, key, svals, val, cmp, tid, delta);
+        }
+    };
+
+    template <int N> struct For<N, N>
+    {
+        template <class PointerTuple, class ReferenceTuple>
+        __device__ static void loadToSmem(const PointerTuple&, const ReferenceTuple&, uint)
+        {
+        }
+
+        template <class PointerTuple, class ReferenceTuple>
+        __device__ static void loadFromSmem(const PointerTuple&, const ReferenceTuple&, uint)
+        {
+        }
+
+        template <class PointerTuple, class ReferenceTuple>
+        __device__ static void copy(const PointerTuple&, const ReferenceTuple&, uint, uint)
+        {
+        }
+
+        template <class KeyPointerTuple, class KeyReferenceTuple, class ValPointerTuple, class ValReferenceTuple, class CmpTuple>
+        __device__ static void merge(const KeyPointerTuple&, const KeyReferenceTuple&,
+                                     const ValPointerTuple&, const ValReferenceTuple&,
+                                     const CmpTuple&,
+                                     uint, uint)
+        {
+        }
+    };
+
+    // loadToSmem / loadFromSmem
+
+    template <typename T>
+    __device__ __forceinline__ void loadToSmem(volatile T* smem, T& data, uint tid)
+    {
+        smem[tid] = data;
+    }
+
+    template <typename T>
+    __device__ __forceinline__ void loadFromSmem(volatile T* smem, T& data, uint tid)
+    {
+        data = smem[tid];
+    }
+
+    template <typename VP0, typename VP1, typename VP2, typename VP3, typename VP4, typename VP5, typename VP6, typename VP7, typename VP8, typename VP9,
+              typename VR0, typename VR1, typename VR2, typename VR3, typename VR4, typename VR5, typename VR6, typename VR7, typename VR8, typename VR9>
+    __device__ __forceinline__ void loadToSmem(const tuple<VP0, VP1, VP2, VP3, VP4, VP5, VP6, VP7, VP8, VP9>& smem,
+                                               const tuple<VR0, VR1, VR2, VR3, VR4, VR5, VR6, VR7, VR8, VR9>& data,
+                                               uint tid)
+    {
+        For<0, tuple_size<tuple<VP0, VP1, VP2, VP3, VP4, VP5, VP6, VP7, VP8, VP9> >::value>::loadToSmem(smem, data, tid);
+    }
+
+    template <typename VP0, typename VP1, typename VP2, typename VP3, typename VP4, typename VP5, typename VP6, typename VP7, typename VP8, typename VP9,
+              typename VR0, typename VR1, typename VR2, typename VR3, typename VR4, typename VR5, typename VR6, typename VR7, typename VR8, typename VR9>
+    __device__ __forceinline__ void loadFromSmem(const tuple<VP0, VP1, VP2, VP3, VP4, VP5, VP6, VP7, VP8, VP9>& smem,
+                                                 const tuple<VR0, VR1, VR2, VR3, VR4, VR5, VR6, VR7, VR8, VR9>& data,
+                                                 uint tid)
+    {
+        For<0, tuple_size<tuple<VP0, VP1, VP2, VP3, VP4, VP5, VP6, VP7, VP8, VP9> >::value>::loadFromSmem(smem, data, tid);
+    }
+
+    // copyVals
+
+    template <typename V>
+    __device__ __forceinline__ void copyVals(volatile V* svals, V& val, uint tid, uint delta)
+    {
+        svals[tid] = val = svals[tid + delta];
+    }
+
+    template <typename VP0, typename VP1, typename VP2, typename VP3, typename VP4, typename VP5, typename VP6, typename VP7, typename VP8, typename VP9,
+              typename VR0, typename VR1, typename VR2, typename VR3, typename VR4, typename VR5, typename VR6, typename VR7, typename VR8, typename VR9>
+    __device__ __forceinline__ void copyVals(const tuple<VP0, VP1, VP2, VP3, VP4, VP5, VP6, VP7, VP8, VP9>& svals,
+                                             const tuple<VR0, VR1, VR2, VR3, VR4, VR5, VR6, VR7, VR8, VR9>& val,
+                                             uint tid, uint delta)
+    {
+        For<0, tuple_size<tuple<VP0, VP1, VP2, VP3, VP4, VP5, VP6, VP7, VP8, VP9> >::value>::copy(svals, val, tid, delta);
+    }
+
+    // merge
+
+    template <typename K, typename V, class Cmp>
+    __device__ void merge(volatile K* skeys, K& key, volatile V* svals, V& val, const Cmp& cmp, uint tid, uint delta)
+    {
+        K reg = skeys[tid + delta];
+
+        if (cmp(reg, key))
+        {
+            skeys[tid] = key = reg;
+            copyVals(svals, val, tid, delta);
+        }
+    }
+
+    template <typename K,
+              typename VP0, typename VP1, typename VP2, typename VP3, typename VP4, typename VP5, typename VP6, typename VP7, typename VP8, typename VP9,
+              typename VR0, typename VR1, typename VR2, typename VR3, typename VR4, typename VR5, typename VR6, typename VR7, typename VR8, typename VR9,
+              class Cmp>
+    __device__ void merge(volatile K* skeys, K& key,
+                          const tuple<VP0, VP1, VP2, VP3, VP4, VP5, VP6, VP7, VP8, VP9>& svals,
+                          const tuple<VR0, VR1, VR2, VR3, VR4, VR5, VR6, VR7, VR8, VR9>& val,
+                          const Cmp& cmp, uint tid, uint delta)
+    {
+        K reg = skeys[tid + delta];
+
+        if (cmp(reg, key))
+        {
+            skeys[tid] = key = reg;
+            copyVals(svals, val, tid, delta);
+        }
+    }
+
+    template <typename KP0, typename KP1, typename KP2, typename KP3, typename KP4, typename KP5, typename KP6, typename KP7, typename KP8, typename KP9,
+              typename KR0, typename KR1, typename KR2, typename KR3, typename KR4, typename KR5, typename KR6, typename KR7, typename KR8, typename KR9,
+              typename VP0, typename VP1, typename VP2, typename VP3, typename VP4, typename VP5, typename VP6, typename VP7, typename VP8, typename VP9,
+              typename VR0, typename VR1, typename VR2, typename VR3, typename VR4, typename VR5, typename VR6, typename VR7, typename VR8, typename VR9,
+              class Cmp0, class Cmp1, class Cmp2, class Cmp3, class Cmp4, class Cmp5, class Cmp6, class Cmp7, class Cmp8, class Cmp9>
+    __device__ __forceinline__ void merge(const tuple<KP0, KP1, KP2, KP3, KP4, KP5, KP6, KP7, KP8, KP9>& skeys,
+                                          const tuple<KR0, KR1, KR2, KR3, KR4, KR5, KR6, KR7, KR8, KR9>& key,
+                                          const tuple<VP0, VP1, VP2, VP3, VP4, VP5, VP6, VP7, VP8, VP9>& svals,
+                                          const tuple<VR0, VR1, VR2, VR3, VR4, VR5, VR6, VR7, VR8, VR9>& val,
+                                          const tuple<Cmp0, Cmp1, Cmp2, Cmp3, Cmp4, Cmp5, Cmp6, Cmp7, Cmp8, Cmp9>& cmp,
+                                          uint tid, uint delta)
+    {
+        For<0, tuple_size<tuple<VP0, VP1, VP2, VP3, VP4, VP5, VP6, VP7, VP8, VP9> >::value>::merge(skeys, key, svals, val, cmp, tid, delta);
+    }
+
+    // Generic
+
+    template <int N> struct Generic
+    {
+        template <class KP, class KR, class VP, class VR, class Cmp>
+        __device__ static void reduce(KP skeys, KR key, VP svals, VR val, uint tid, Cmp cmp)
+        {
+            loadToSmem(skeys, key, tid);
+            loadValsToSmem(svals, val, tid);
+            if (N >= 32)
+                __syncthreads();
+
+            if (N >= 2048)
+            {
+                if (tid < 1024)
+                    merge(skeys, key, svals, val, cmp, tid, 1024);
+
+                __syncthreads();
+            }
+            if (N >= 1024)
+            {
+                if (tid < 512)
+                    merge(skeys, key, svals, val, cmp, tid, 512);
+
+                __syncthreads();
+            }
+            if (N >= 512)
+            {
+                if (tid < 256)
+                    merge(skeys, key, svals, val, cmp, tid, 256);
+
+                __syncthreads();
+            }
+            if (N >= 256)
+            {
+                if (tid < 128)
+                    merge(skeys, key, svals, val, cmp, tid, 128);
+
+                __syncthreads();
+            }
+            if (N >= 128)
+            {
+                if (tid < 64)
+                    merge(skeys, key, svals, val, cmp, tid, 64);
+
+                __syncthreads();
+            }
+            if (N >= 64)
+            {
+                if (tid < 32)
+                    merge(skeys, key, svals, val, cmp, tid, 32);
+            }
+
+            if (tid < 16)
+            {
+                merge(skeys, key, svals, val, cmp, tid, 16);
+                merge(skeys, key, svals, val, cmp, tid, 8);
+                merge(skeys, key, svals, val, cmp, tid, 4);
+                merge(skeys, key, svals, val, cmp, tid, 2);
+                merge(skeys, key, svals, val, cmp, tid, 1);
+            }
+        }
+    };
+
+    // Unroll
+
+    template <int I, class KP, class KR, class VP, class VR, class Cmp> struct Unroll
+    {
+        __device__ static void loop(KP skeys, KR key, VP svals, VR val, uint tid, Cmp cmp)
+        {
+            merge(skeys, key, svals, val, cmp, tid, I);
+            Unroll<I / 2, KP, KR, VP, VR, Cmp>::loop(skeys, key, svals, val, tid, cmp);
+        }
+    };
+
+    template <class KP, class KR, class VP, class VR, class Cmp> struct Unroll<0, KP, KR, VP, VR, Cmp>
+    {
+        __device__ __forceinline__ static void loop(KP, KR, VP, VR, uint, Cmp)
+        {
+        }
+    };
+
+    // WarpOptimized
+
+    template <int N> struct WarpOptimized
+    {
+        template <class KP, class KR, class VP, class VR, class Cmp>
+        __device__ static void reduce(KP skeys, KR key, VP svals, VR val, uint tid, Cmp cmp)
+        {
+            loadToSmem(skeys, key, tid);
+            loadToSmem(svals, val, tid);
+
+            if (tid < N / 2)
+                Unroll<N / 2, KP, KR, VP, VR, Cmp>::loop(skeys, key, svals, val, tid, cmp);
+        }
+    };
+
+    // GenericOptimized32
+
+    template <uint N> struct GenericOptimized32
+    {
+        enum { M = N / 32 };
+
+        template <class KP, class KR, class VP, class VR, class Cmp>
+        __device__ static void reduce(KP skeys, KR key, VP svals, VR val, uint tid, Cmp cmp)
+        {
+            const uint laneId = Warp::laneId();
+
+            loadToSmem(skeys, key, tid);
+            loadToSmem(svals, val, tid);
+
+            if (laneId < 16)
+                Unroll<16, KP, KR, VP, VR, Cmp>::loop(skeys, key, svals, val, tid, cmp);
+
+            __syncthreads();
+
+            if (laneId == 0)
+            {
+                loadToSmem(skeys, key, tid / 32);
+                loadToSmem(svals, val, tid / 32);
+            }
+
+            __syncthreads();
+
+            loadFromSmem(skeys, key, tid);
+
+            if (tid < 32)
+            {
+                Unroll<M / 2, KP, KR, VP, VR, Cmp>::loop(skeys, key, svals, val, tid, cmp);
+            }
+        }
+    };
+
+    template <int N> struct Dispatcher
+    {
+        typedef typename SelectIf<
+            (N <= 32) && IsPowerOf2<N>::value,
+            WarpOptimized<N>,
+            typename SelectIf<
+                (N <= 1024) && IsPowerOf2<N>::value,
+                GenericOptimized32<N>,
+                Generic<N>
+            >::type
+        >::type reductor;
+    };
+}
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/block/dynamic_smem.hpp

@@ -0,0 +1,86 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_BLOCK_DYNAMIC_SMEM_HPP__
+#define __OPENCV_CUDEV_BLOCK_DYNAMIC_SMEM_HPP__
+
+#include "../common.hpp"
+
+namespace cv { namespace cudev {
+
+template <class T> struct DynamicSharedMem
+{
+    __device__ __forceinline__ operator T*()
+    {
+        extern __shared__ int __smem[];
+        return (T*) __smem;
+    }
+
+    __device__ __forceinline__ operator const T*() const
+    {
+        extern __shared__ int __smem[];
+        return (T*) __smem;
+    }
+};
+
+// specialize for double to avoid unaligned memory access compile errors
+template <> struct DynamicSharedMem<double>
+{
+    __device__ __forceinline__ operator double*()
+    {
+        extern __shared__ double __smem_d[];
+        return (double*) __smem_d;
+    }
+
+    __device__ __forceinline__ operator const double*() const
+    {
+        extern __shared__ double __smem_d[];
+        return (double*) __smem_d;
+    }
+};
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/block/reduce.hpp

@@ -0,0 +1,128 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_BLOCK_REDUCE_HPP__
+#define __OPENCV_CUDEV_BLOCK_REDUCE_HPP__
+
+#include "../common.hpp"
+#include "../util/tuple.hpp"
+#include "../warp/reduce.hpp"
+#include "detail/reduce.hpp"
+#include "detail/reduce_key_val.hpp"
+
+namespace cv { namespace cudev {
+
+// blockReduce
+
+template <int N, typename T, class Op>
+__device__ __forceinline__ void blockReduce(volatile T* smem, T& val, uint tid, const Op& op)
+{
+    block_reduce_detail::Dispatcher<N>::reductor::template reduce<volatile T*, T&, const Op&>(smem, val, tid, op);
+}
+
+template <int N,
+          typename P0, typename P1, typename P2, typename P3, typename P4, typename P5, typename P6, typename P7, typename P8, typename P9,
+          typename R0, typename R1, typename R2, typename R3, typename R4, typename R5, typename R6, typename R7, typename R8, typename R9,
+          class Op0, class Op1, class Op2, class Op3, class Op4, class Op5, class Op6, class Op7, class Op8, class Op9>
+__device__ __forceinline__ void blockReduce(const tuple<P0, P1, P2, P3, P4, P5, P6, P7, P8, P9>& smem,
+                                            const tuple<R0, R1, R2, R3, R4, R5, R6, R7, R8, R9>& val,
+                                            uint tid,
+                                            const tuple<Op0, Op1, Op2, Op3, Op4, Op5, Op6, Op7, Op8, Op9>& op)
+{
+    block_reduce_detail::Dispatcher<N>::reductor::template reduce<
+            const tuple<P0, P1, P2, P3, P4, P5, P6, P7, P8, P9>&,
+            const tuple<R0, R1, R2, R3, R4, R5, R6, R7, R8, R9>&,
+            const tuple<Op0, Op1, Op2, Op3, Op4, Op5, Op6, Op7, Op8, Op9>&>(smem, val, tid, op);
+}
+
+// blockReduceKeyVal
+
+template <int N, typename K, typename V, class Cmp>
+__device__ __forceinline__ void blockReduceKeyVal(volatile K* skeys, K& key, volatile V* svals, V& val, uint tid, const Cmp& cmp)
+{
+    block_reduce_key_val_detail::Dispatcher<N>::reductor::template reduce<volatile K*, K&, volatile V*, V&, const Cmp&>(skeys, key, svals, val, tid, cmp);
+}
+
+template <int N,
+          typename K,
+          typename VP0, typename VP1, typename VP2, typename VP3, typename VP4, typename VP5, typename VP6, typename VP7, typename VP8, typename VP9,
+          typename VR0, typename VR1, typename VR2, typename VR3, typename VR4, typename VR5, typename VR6, typename VR7, typename VR8, typename VR9,
+          class Cmp>
+__device__ __forceinline__ void blockReduceKeyVal(volatile K* skeys, K& key,
+                                                  const tuple<VP0, VP1, VP2, VP3, VP4, VP5, VP6, VP7, VP8, VP9>& svals,
+                                                  const tuple<VR0, VR1, VR2, VR3, VR4, VR5, VR6, VR7, VR8, VR9>& val,
+                                                  uint tid, const Cmp& cmp)
+{
+    block_reduce_key_val_detail::Dispatcher<N>::reductor::template reduce<volatile K*, K&,
+            const tuple<VP0, VP1, VP2, VP3, VP4, VP5, VP6, VP7, VP8, VP9>&,
+            const tuple<VR0, VR1, VR2, VR3, VR4, VR5, VR6, VR7, VR8, VR9>&,
+            const Cmp&>(skeys, key, svals, val, tid, cmp);
+}
+
+template <int N,
+          typename KP0, typename KP1, typename KP2, typename KP3, typename KP4, typename KP5, typename KP6, typename KP7, typename KP8, typename KP9,
+          typename KR0, typename KR1, typename KR2, typename KR3, typename KR4, typename KR5, typename KR6, typename KR7, typename KR8, typename KR9,
+          typename VP0, typename VP1, typename VP2, typename VP3, typename VP4, typename VP5, typename VP6, typename VP7, typename VP8, typename VP9,
+          typename VR0, typename VR1, typename VR2, typename VR3, typename VR4, typename VR5, typename VR6, typename VR7, typename VR8, typename VR9,
+          class Cmp0, class Cmp1, class Cmp2, class Cmp3, class Cmp4, class Cmp5, class Cmp6, class Cmp7, class Cmp8, class Cmp9>
+__device__ __forceinline__ void blockReduceKeyVal(const tuple<KP0, KP1, KP2, KP3, KP4, KP5, KP6, KP7, KP8, KP9>& skeys,
+                                                  const tuple<KR0, KR1, KR2, KR3, KR4, KR5, KR6, KR7, KR8, KR9>& key,
+                                                  const tuple<VP0, VP1, VP2, VP3, VP4, VP5, VP6, VP7, VP8, VP9>& svals,
+                                                  const tuple<VR0, VR1, VR2, VR3, VR4, VR5, VR6, VR7, VR8, VR9>& val,
+                                                  uint tid,
+                                                  const tuple<Cmp0, Cmp1, Cmp2, Cmp3, Cmp4, Cmp5, Cmp6, Cmp7, Cmp8, Cmp9>& cmp)
+{
+    block_reduce_key_val_detail::Dispatcher<N>::reductor::template reduce<
+            const tuple<KP0, KP1, KP2, KP3, KP4, KP5, KP6, KP7, KP8, KP9>&,
+            const tuple<KR0, KR1, KR2, KR3, KR4, KR5, KR6, KR7, KR8, KR9>&,
+            const tuple<VP0, VP1, VP2, VP3, VP4, VP5, VP6, VP7, VP8, VP9>&,
+            const tuple<VR0, VR1, VR2, VR3, VR4, VR5, VR6, VR7, VR8, VR9>&,
+            const tuple<Cmp0, Cmp1, Cmp2, Cmp3, Cmp4, Cmp5, Cmp6, Cmp7, Cmp8, Cmp9>&
+            >(skeys, key, svals, val, tid, cmp);
+}
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/block/scan.hpp

@@ -0,0 +1,101 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_BLOCK_SCAN_HPP__
+#define __OPENCV_CUDEV_BLOCK_SCAN_HPP__
+
+#include "../common.hpp"
+#include "../warp/scan.hpp"
+
+namespace cv { namespace cudev {
+
+template <int THREADS_NUM, typename T>
+__device__ T blockScanInclusive(T data, volatile T* smem, uint tid)
+{
+    if (THREADS_NUM > WARP_SIZE)
+    {
+        // bottom-level inclusive warp scan
+        T warpResult = warpScanInclusive(data, smem, tid);
+
+        __syncthreads();
+
+        // save top elements of each warp for exclusive warp scan
+        // sync to wait for warp scans to complete (because s_Data is being overwritten)
+        if ((tid & (WARP_SIZE - 1)) == (WARP_SIZE - 1))
+        {
+            smem[tid >> LOG_WARP_SIZE] = warpResult;
+        }
+
+        __syncthreads();
+
+        if (tid < (THREADS_NUM / WARP_SIZE))
+        {
+            // grab top warp elements
+            T val = smem[tid];
+
+            // calculate exclusive scan and write back to shared memory
+            smem[tid] = warpScanExclusive(val, smem, tid);
+        }
+
+        __syncthreads();
+
+        // return updated warp scans with exclusive scan results
+        return warpResult + smem[tid >> LOG_WARP_SIZE];
+    }
+    else
+    {
+        return warpScanInclusive(data, smem, tid);
+    }
+}
+
+template <int THREADS_NUM, typename T>
+__device__ __forceinline__ T blockScanExclusive(T data, volatile T* smem, uint tid)
+{
+    return blockScanInclusive<THREADS_NUM>(data, smem, tid) - data;
+}
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/block/vec_distance.hpp

@@ -0,0 +1,184 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_BLOCK_VEC_DISTANCE_HPP__
+#define __OPENCV_CUDEV_BLOCK_VEC_DISTANCE_HPP__
+
+#include "../common.hpp"
+#include "../functional/functional.hpp"
+#include "../warp/reduce.hpp"
+#include "reduce.hpp"
+
+namespace cv { namespace cudev {
+
+// NormL1
+
+template <typename T> struct NormL1
+{
+    typedef int value_type;
+    typedef uint result_type;
+
+    result_type mySum;
+
+    __device__ __forceinline__ NormL1() : mySum(0) {}
+
+    __device__ __forceinline__ void reduceThread(value_type val1, value_type val2)
+    {
+        mySum = __sad(val1, val2, mySum);
+    }
+
+    __device__ __forceinline__ void reduceWarp(result_type* smem, uint tid)
+    {
+        warpReduce(smem, mySum, tid, plus<result_type>());
+    }
+
+    template <int THREAD_DIM> __device__ __forceinline__ void reduceBlock(result_type* smem, uint tid)
+    {
+        blockReduce<THREAD_DIM>(smem, mySum, tid, plus<result_type>());
+    }
+
+    __device__ __forceinline__ operator result_type() const
+    {
+        return mySum;
+    }
+};
+template <> struct NormL1<float>
+{
+    typedef float value_type;
+    typedef float result_type;
+
+    result_type mySum;
+
+    __device__ __forceinline__ NormL1() : mySum(0.0f) {}
+
+    __device__ __forceinline__ void reduceThread(value_type val1, value_type val2)
+    {
+        mySum += ::fabsf(val1 - val2);
+    }
+
+    __device__ __forceinline__ void reduceWarp(result_type* smem, uint tid)
+    {
+        warpReduce(smem, mySum, tid, plus<result_type>());
+    }
+
+    template <int THREAD_DIM> __device__ __forceinline__ void reduceBlock(result_type* smem, uint tid)
+    {
+        blockReduce<THREAD_DIM>(smem, mySum, tid, plus<result_type>());
+    }
+
+    __device__ __forceinline__ operator result_type() const
+    {
+        return mySum;
+    }
+};
+
+// NormL2
+
+struct NormL2
+{
+    typedef float value_type;
+    typedef float result_type;
+
+    result_type mySum;
+
+    __device__ __forceinline__ NormL2() : mySum(0.0f) {}
+
+    __device__ __forceinline__ void reduceThread(value_type val1, value_type val2)
+    {
+        const float diff = val1 - val2;
+        mySum += diff * diff;
+    }
+
+    __device__ __forceinline__ void reduceWarp(result_type* smem, uint tid)
+    {
+        warpReduce(smem, mySum, tid, plus<result_type>());
+    }
+
+    template <int THREAD_DIM> __device__ __forceinline__ void reduceBlock(result_type* smem, uint tid)
+    {
+        blockReduce<THREAD_DIM>(smem, mySum, tid, plus<result_type>());
+    }
+
+    __device__ __forceinline__ operator result_type() const
+    {
+        return ::sqrtf(mySum);
+    }
+};
+
+// NormHamming
+
+struct NormHamming
+{
+    typedef int value_type;
+    typedef int result_type;
+
+    result_type mySum;
+
+    __device__ __forceinline__ NormHamming() : mySum(0) {}
+
+    __device__ __forceinline__ void reduceThread(value_type val1, value_type val2)
+    {
+        mySum += __popc(val1 ^ val2);
+    }
+
+    __device__ __forceinline__ void reduceWarp(result_type* smem, uint tid)
+    {
+        warpReduce(smem, mySum, tid, plus<result_type>());
+    }
+
+    template <int THREAD_DIM> __device__ __forceinline__ void reduceBlock(result_type* smem, uint tid)
+    {
+        blockReduce<THREAD_DIM>(smem, mySum, tid, plus<result_type>());
+    }
+
+    __device__ __forceinline__ operator result_type() const
+    {
+        return mySum;
+    }
+};
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/common.hpp

@@ -0,0 +1,93 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_COMMON_HPP__
+#define __OPENCV_CUDEV_COMMON_HPP__
+
+#include <cuda_runtime.h>
+#include "opencv2/core/gpu.hpp"
+#include "opencv2/core/gpu_stream_accessor.hpp"
+
+namespace cv { namespace cudev {
+
+using namespace cv::gpu;
+
+// CV_CUDEV_ARCH
+
+#ifndef __CUDA_ARCH__
+#   define CV_CUDEV_ARCH 0
+#else
+#   define CV_CUDEV_ARCH __CUDA_ARCH__
+#endif
+
+// CV_CUDEV_SAFE_CALL
+
+__host__ __forceinline__ void checkCudaError(cudaError_t err, const char* file, const int line, const char* func)
+{
+    if (cudaSuccess != err)
+        cv::error(cv::Error::GpuApiCallError, cudaGetErrorString(err), func, file, line);
+}
+
+#ifdef __GNUC__
+#   define CV_CUDEV_SAFE_CALL(expr) cv::cudev::checkCudaError((expr), __FILE__, __LINE__, __func__)
+#else
+#   define CV_CUDEV_SAFE_CALL(expr) cv::cudev::checkCudaError((expr), __FILE__, __LINE__, "")
+#endif
+
+// divUp
+
+__host__ __device__ __forceinline__ int divUp(int total, int grain)
+{
+    return (total + grain - 1) / grain;
+}
+
+// math constants
+
+#define CV_PI_F   ((float)CV_PI)
+#define CV_LOG2_F ((float)CV_LOG2)
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/expr/binary_func.hpp

@@ -0,0 +1,75 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_EXPR_BINARY_FUNC_HPP__
+#define __OPENCV_CUDEV_EXPR_BINARY_FUNC_HPP__
+
+#include "../common.hpp"
+#include "../util/type_traits.hpp"
+#include "../ptr2d/traits.hpp"
+#include "../ptr2d/transform.hpp"
+#include "../functional/functional.hpp"
+#include "expr.hpp"
+
+namespace cv { namespace cudev {
+
+#define CV_CUDEV_EXPR_BINARY_FUNC(name) \
+    template <class SrcPtr1, class SrcPtr2> \
+    __host__ Expr<BinaryTransformPtrSz<typename PtrTraits<SrcPtr1>::ptr_type, typename PtrTraits<SrcPtr2>::ptr_type, name ## _func<typename LargerType<typename PtrTraits<SrcPtr1>::value_type, typename PtrTraits<SrcPtr2>::value_type>::type> > > \
+    name ## _(const SrcPtr1& src1, const SrcPtr2& src2) \
+    { \
+        return makeExpr(transformPtr(src1, src2, name ## _func<typename LargerType<typename PtrTraits<SrcPtr1>::value_type, typename PtrTraits<SrcPtr2>::value_type>::type>())); \
+    }
+
+CV_CUDEV_EXPR_BINARY_FUNC(hypot)
+CV_CUDEV_EXPR_BINARY_FUNC(magnitude)
+CV_CUDEV_EXPR_BINARY_FUNC(atan2)
+CV_CUDEV_EXPR_BINARY_FUNC(absdiff)
+
+#undef CV_CUDEV_EXPR_BINARY_FUNC
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/expr/binary_op.hpp

@@ -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.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_EXPR_BINARY_OP_HPP__
+#define __OPENCV_CUDEV_EXPR_BINARY_OP_HPP__
+
+#include "../common.hpp"
+#include "../util/type_traits.hpp"
+#include "../ptr2d/traits.hpp"
+#include "../ptr2d/transform.hpp"
+#include "../ptr2d/gpumat.hpp"
+#include "../ptr2d/texture.hpp"
+#include "../ptr2d/glob.hpp"
+#include "../functional/functional.hpp"
+#include "expr.hpp"
+
+namespace cv { namespace cudev {
+
+// Binary Operations
+
+#define CV_CUDEV_EXPR_BINOP_INST(op, functor) \
+    template <typename T> \
+    __host__ Expr<BinaryTransformPtrSz<typename PtrTraits<GpuMat_<T> >::ptr_type, typename PtrTraits<GpuMat_<T> >::ptr_type, functor<T> > > \
+    operator op(const GpuMat_<T>& src1, const GpuMat_<T>& src2) \
+    { \
+        return makeExpr(transformPtr(src1, src2, functor<T>())); \
+    } \
+    template <typename T> \
+    __host__ Expr<BinaryTransformPtrSz<typename PtrTraits<GpuMat_<T> >::ptr_type, typename PtrTraits<GlobPtrSz<T> >::ptr_type, functor<T> > > \
+    operator op(const GpuMat_<T>& src1, const GlobPtrSz<T>& src2) \
+    { \
+        return makeExpr(transformPtr(src1, src2, functor<T>())); \
+    } \
+    template <typename T> \
+    __host__ Expr<BinaryTransformPtrSz<typename PtrTraits<GlobPtrSz<T> >::ptr_type, typename PtrTraits<GpuMat_<T> >::ptr_type, functor<T> > > \
+    operator op(const GlobPtrSz<T>& src1, const GpuMat_<T>& src2) \
+    { \
+        return makeExpr(transformPtr(src1, src2, functor<T>())); \
+    } \
+    template <typename T> \
+    __host__ Expr<BinaryTransformPtrSz<typename PtrTraits<GpuMat_<T> >::ptr_type, typename PtrTraits<Texture<T> >::ptr_type, functor<T> > > \
+    operator op(const GpuMat_<T>& src1, const Texture<T>& src2) \
+    { \
+        return makeExpr(transformPtr(src1, src2, functor<T>())); \
+    } \
+    template <typename T> \
+    __host__ Expr<BinaryTransformPtrSz<typename PtrTraits<Texture<T> >::ptr_type, typename PtrTraits<GpuMat_<T> >::ptr_type, functor<T> > > \
+    operator op(const Texture<T>& src1, const GpuMat_<T>& src2) \
+    { \
+        return makeExpr(transformPtr(src1, src2, functor<T>())); \
+    } \
+    template <typename T, class Body> \
+    __host__ Expr<BinaryTransformPtrSz<typename PtrTraits<GpuMat_<T> >::ptr_type, typename PtrTraits<Body>::ptr_type, functor<typename LargerType<T, typename PtrTraits<Body>::value_type>::type> > > \
+    operator op(const GpuMat_<T>& src1, const Expr<Body>& src2) \
+    { \
+        return makeExpr(transformPtr(src1, src2.body, functor<typename LargerType<T, typename PtrTraits<Body>::value_type>::type>())); \
+    } \
+    template <typename T, class Body> \
+    __host__ Expr<BinaryTransformPtrSz<typename PtrTraits<Body>::ptr_type, typename PtrTraits<GpuMat_<T> >::ptr_type, functor<typename LargerType<T, typename PtrTraits<Body>::value_type>::type> > > \
+    operator op(const Expr<Body>& src1, const GpuMat_<T>& src2) \
+    { \
+        return makeExpr(transformPtr(src1.body, src2, functor<typename LargerType<T, typename PtrTraits<Body>::value_type>::type>())); \
+    } \
+    template <typename T> \
+    __host__ Expr<UnaryTransformPtrSz<typename PtrTraits<GpuMat_<T> >::ptr_type, Binder2nd< functor<T> > > > \
+    operator op(const GpuMat_<T>& src, T val) \
+    { \
+        return makeExpr(transformPtr(src, bind2nd(functor<T>(), val))); \
+    } \
+    template <typename T> \
+    __host__ Expr<UnaryTransformPtrSz<typename PtrTraits<GpuMat_<T> >::ptr_type, Binder1st< functor<T> > > > \
+    operator op(T val, const GpuMat_<T>& src) \
+    { \
+        return makeExpr(transformPtr(src, bind1st(functor<T>(), val))); \
+    } \
+    template <typename T> \
+    __host__ Expr<BinaryTransformPtrSz<typename PtrTraits<GlobPtrSz<T> >::ptr_type, typename PtrTraits<GlobPtrSz<T> >::ptr_type, functor<T> > > \
+    operator op(const GlobPtrSz<T>& src1, const GlobPtrSz<T>& src2) \
+    { \
+        return makeExpr(transformPtr(src1, src2, functor<T>())); \
+    } \
+    template <typename T> \
+    __host__ Expr<BinaryTransformPtrSz<typename PtrTraits<GlobPtrSz<T> >::ptr_type, typename PtrTraits<Texture<T> >::ptr_type, functor<T> > > \
+    operator op(const GlobPtrSz<T>& src1, const Texture<T>& src2) \
+    { \
+        return makeExpr(transformPtr(src1, src2, functor<T>())); \
+    } \
+    template <typename T> \
+    __host__ Expr<BinaryTransformPtrSz<typename PtrTraits<Texture<T> >::ptr_type, typename PtrTraits<GlobPtrSz<T> >::ptr_type, functor<T> > > \
+    operator op(const Texture<T>& src1, const GlobPtrSz<T>& src2) \
+    { \
+        return makeExpr(transformPtr(src1, src2, functor<T>())); \
+    } \
+    template <typename T, class Body> \
+    __host__ Expr<BinaryTransformPtrSz<typename PtrTraits<GlobPtrSz<T> >::ptr_type, typename PtrTraits<Body>::ptr_type, functor<typename LargerType<T, typename PtrTraits<Body>::value_type>::type> > > \
+    operator op(const GlobPtrSz<T>& src1, const Expr<Body>& src2) \
+    { \
+        return makeExpr(transformPtr(src1, src2.body, functor<typename LargerType<T, typename PtrTraits<Body>::value_type>::type>())); \
+    } \
+    template <typename T, class Body> \
+    __host__ Expr<BinaryTransformPtrSz<typename PtrTraits<Body>::ptr_type, typename PtrTraits<GlobPtrSz<T> >::ptr_type, functor<typename LargerType<T, typename PtrTraits<Body>::value_type>::type> > > \
+    operator op(const Expr<Body>& src1, const GlobPtrSz<T>& src2) \
+    { \
+        return makeExpr(transformPtr(src1.body, src2, functor<typename LargerType<T, typename PtrTraits<Body>::value_type>::type>())); \
+    } \
+    template <typename T> \
+    __host__ Expr<UnaryTransformPtrSz<typename PtrTraits<GlobPtrSz<T> >::ptr_type, Binder2nd< functor<T> > > > \
+    operator op(const GlobPtrSz<T>& src, T val) \
+    { \
+        return makeExpr(transformPtr(src, bind2nd(functor<T>(), val))); \
+    } \
+    template <typename T> \
+    __host__ Expr<UnaryTransformPtrSz<typename PtrTraits<GlobPtrSz<T> >::ptr_type, Binder1st< functor<T> > > > \
+    operator op(T val, const GlobPtrSz<T>& src) \
+    { \
+        return makeExpr(transformPtr(src, bind1st(functor<T>(), val))); \
+    } \
+    template <typename T> \
+    __host__ Expr<BinaryTransformPtrSz<typename PtrTraits<Texture<T> >::ptr_type, typename PtrTraits<Texture<T> >::ptr_type, functor<T> > > \
+    operator op(const Texture<T>& src1, const Texture<T>& src2) \
+    { \
+        return makeExpr(transformPtr(src1, src2, functor<T>())); \
+    } \
+    template <typename T, class Body> \
+    __host__ Expr<BinaryTransformPtrSz<typename PtrTraits<Texture<T> >::ptr_type, typename PtrTraits<Body>::ptr_type, functor<typename LargerType<T, typename PtrTraits<Body>::value_type>::type> > > \
+    operator op(const Texture<T>& src1, const Expr<Body>& src2) \
+    { \
+        return makeExpr(transformPtr(src1, src2.body, functor<typename LargerType<T, typename PtrTraits<Body>::value_type>::type>())); \
+    } \
+    template <typename T, class Body> \
+    __host__ Expr<BinaryTransformPtrSz<typename PtrTraits<Body>::ptr_type, typename PtrTraits<Texture<T> >::ptr_type, functor<typename LargerType<T, typename PtrTraits<Body>::value_type>::type> > > \
+    operator op(const Expr<Body>& src1, const Texture<T>& src2) \
+    { \
+        return makeExpr(transformPtr(src1.body, src2, functor<typename LargerType<T, typename PtrTraits<Body>::value_type>::type>())); \
+    } \
+    template <typename T> \
+    __host__ Expr<UnaryTransformPtrSz<typename PtrTraits<Texture<T> >::ptr_type, Binder2nd< functor<T> > > > \
+    operator op(const Texture<T>& src, T val) \
+    { \
+        return makeExpr(transformPtr(src, bind2nd(functor<T>(), val))); \
+    } \
+    template <typename T> \
+    __host__ Expr<UnaryTransformPtrSz<typename PtrTraits<Texture<T> >::ptr_type, Binder1st< functor<T> > > > \
+    operator op(T val, const Texture<T>& src) \
+    { \
+        return makeExpr(transformPtr(src, bind1st(functor<T>(), val))); \
+    } \
+    template <class Body1, class Body2> \
+    __host__ Expr<BinaryTransformPtrSz<typename PtrTraits<Body1>::ptr_type, typename PtrTraits<Body2>::ptr_type, functor<typename LargerType<typename PtrTraits<Body1>::value_type, typename PtrTraits<Body2>::value_type>::type> > > \
+    operator op(const Expr<Body1>& a, const Expr<Body2>& b) \
+    { \
+        return makeExpr(transformPtr(a.body, b.body, functor<typename LargerType<typename PtrTraits<Body1>::value_type, typename PtrTraits<Body2>::value_type>::type>())); \
+    } \
+    template <class Body> \
+    __host__ Expr<UnaryTransformPtrSz<typename PtrTraits<Body>::ptr_type, Binder2nd< functor<typename Body::value_type> > > > \
+    operator op(const Expr<Body>& a, typename Body::value_type val) \
+    { \
+        return makeExpr(transformPtr(a.body, bind2nd(functor<typename Body::value_type>(), val))); \
+    } \
+    template <class Body> \
+    __host__ Expr<UnaryTransformPtrSz<typename PtrTraits<Body>::ptr_type, Binder1st< functor<typename Body::value_type> > > > \
+    operator op(typename Body::value_type val, const Expr<Body>& a) \
+    { \
+        return makeExpr(transformPtr(a.body, bind1st(functor<typename Body::value_type>(), val))); \
+    }
+
+CV_CUDEV_EXPR_BINOP_INST(+, plus)
+CV_CUDEV_EXPR_BINOP_INST(-, minus)
+CV_CUDEV_EXPR_BINOP_INST(*, multiplies)
+CV_CUDEV_EXPR_BINOP_INST(/, divides)
+CV_CUDEV_EXPR_BINOP_INST(%, modulus)
+
+CV_CUDEV_EXPR_BINOP_INST(==, equal_to)
+CV_CUDEV_EXPR_BINOP_INST(!=, not_equal_to)
+CV_CUDEV_EXPR_BINOP_INST(>, greater)
+CV_CUDEV_EXPR_BINOP_INST(<, less)
+CV_CUDEV_EXPR_BINOP_INST(>=, greater_equal)
+CV_CUDEV_EXPR_BINOP_INST(<=, less_equal)
+
+CV_CUDEV_EXPR_BINOP_INST(&&, logical_and)
+CV_CUDEV_EXPR_BINOP_INST(||, logical_or)
+
+CV_CUDEV_EXPR_BINOP_INST(&, bit_and)
+CV_CUDEV_EXPR_BINOP_INST(|, bit_or)
+CV_CUDEV_EXPR_BINOP_INST(^, bit_xor)
+CV_CUDEV_EXPR_BINOP_INST(<<, bit_lshift)
+CV_CUDEV_EXPR_BINOP_INST(>>, bit_rshift)
+
+#undef CV_CUDEV_EXPR_BINOP_INST
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/expr/color.hpp

@@ -0,0 +1,282 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_EXPR_COLOR_HPP__
+#define __OPENCV_CUDEV_EXPR_COLOR_HPP__
+
+#include "../common.hpp"
+#include "../ptr2d/traits.hpp"
+#include "../ptr2d/transform.hpp"
+#include "../functional/color_cvt.hpp"
+#include "expr.hpp"
+
+namespace cv { namespace cudev {
+
+#define CV_CUDEV_EXPR_CVTCOLOR_INST(name) \
+    template <class SrcPtr> \
+    __host__ Expr<UnaryTransformPtrSz<typename PtrTraits<SrcPtr>::ptr_type, name ## _func<typename VecTraits<typename PtrTraits<SrcPtr>::value_type>::elem_type> > > \
+    name ## _(const SrcPtr& src) \
+    { \
+        return makeExpr(transformPtr(src, name ## _func<typename VecTraits<typename PtrTraits<SrcPtr>::value_type>::elem_type>())); \
+    }
+
+CV_CUDEV_EXPR_CVTCOLOR_INST(BGR_to_RGB)
+CV_CUDEV_EXPR_CVTCOLOR_INST(BGR_to_BGRA)
+CV_CUDEV_EXPR_CVTCOLOR_INST(BGR_to_RGBA)
+CV_CUDEV_EXPR_CVTCOLOR_INST(BGRA_to_BGR)
+CV_CUDEV_EXPR_CVTCOLOR_INST(BGRA_to_RGB)
+CV_CUDEV_EXPR_CVTCOLOR_INST(BGRA_to_RGBA)
+
+CV_CUDEV_EXPR_CVTCOLOR_INST(RGB_to_GRAY)
+CV_CUDEV_EXPR_CVTCOLOR_INST(BGR_to_GRAY)
+CV_CUDEV_EXPR_CVTCOLOR_INST(RGBA_to_GRAY)
+CV_CUDEV_EXPR_CVTCOLOR_INST(BGRA_to_GRAY)
+
+CV_CUDEV_EXPR_CVTCOLOR_INST(GRAY_to_BGR)
+CV_CUDEV_EXPR_CVTCOLOR_INST(GRAY_to_BGRA)
+
+CV_CUDEV_EXPR_CVTCOLOR_INST(RGB_to_YUV)
+CV_CUDEV_EXPR_CVTCOLOR_INST(RGBA_to_YUV)
+CV_CUDEV_EXPR_CVTCOLOR_INST(RGB_to_YUV4)
+CV_CUDEV_EXPR_CVTCOLOR_INST(RGBA_to_YUV4)
+CV_CUDEV_EXPR_CVTCOLOR_INST(BGR_to_YUV)
+CV_CUDEV_EXPR_CVTCOLOR_INST(BGRA_to_YUV)
+CV_CUDEV_EXPR_CVTCOLOR_INST(BGR_to_YUV4)
+CV_CUDEV_EXPR_CVTCOLOR_INST(BGRA_to_YUV4)
+
+CV_CUDEV_EXPR_CVTCOLOR_INST(YUV_to_RGB)
+CV_CUDEV_EXPR_CVTCOLOR_INST(YUV_to_RGBA)
+CV_CUDEV_EXPR_CVTCOLOR_INST(YUV4_to_RGB)
+CV_CUDEV_EXPR_CVTCOLOR_INST(YUV4_to_RGBA)
+CV_CUDEV_EXPR_CVTCOLOR_INST(YUV_to_BGR)
+CV_CUDEV_EXPR_CVTCOLOR_INST(YUV_to_BGRA)
+CV_CUDEV_EXPR_CVTCOLOR_INST(YUV4_to_BGR)
+CV_CUDEV_EXPR_CVTCOLOR_INST(YUV4_to_BGRA)
+
+CV_CUDEV_EXPR_CVTCOLOR_INST(RGB_to_YCrCb)
+CV_CUDEV_EXPR_CVTCOLOR_INST(RGBA_to_YCrCb)
+CV_CUDEV_EXPR_CVTCOLOR_INST(RGB_to_YCrCb4)
+CV_CUDEV_EXPR_CVTCOLOR_INST(RGBA_to_YCrCb4)
+CV_CUDEV_EXPR_CVTCOLOR_INST(BGR_to_YCrCb)
+CV_CUDEV_EXPR_CVTCOLOR_INST(BGRA_to_YCrCb)
+CV_CUDEV_EXPR_CVTCOLOR_INST(BGR_to_YCrCb4)
+CV_CUDEV_EXPR_CVTCOLOR_INST(BGRA_to_YCrCb4)
+
+CV_CUDEV_EXPR_CVTCOLOR_INST(YCrCb_to_RGB)
+CV_CUDEV_EXPR_CVTCOLOR_INST(YCrCb_to_RGBA)
+CV_CUDEV_EXPR_CVTCOLOR_INST(YCrCb4_to_RGB)
+CV_CUDEV_EXPR_CVTCOLOR_INST(YCrCb4_to_RGBA)
+CV_CUDEV_EXPR_CVTCOLOR_INST(YCrCb_to_BGR)
+CV_CUDEV_EXPR_CVTCOLOR_INST(YCrCb_to_BGRA)
+CV_CUDEV_EXPR_CVTCOLOR_INST(YCrCb4_to_BGR)
+CV_CUDEV_EXPR_CVTCOLOR_INST(YCrCb4_to_BGRA)
+
+CV_CUDEV_EXPR_CVTCOLOR_INST(RGB_to_XYZ)
+CV_CUDEV_EXPR_CVTCOLOR_INST(RGBA_to_XYZ)
+CV_CUDEV_EXPR_CVTCOLOR_INST(RGB_to_XYZ4)
+CV_CUDEV_EXPR_CVTCOLOR_INST(RGBA_to_XYZ4)
+CV_CUDEV_EXPR_CVTCOLOR_INST(BGR_to_XYZ)
+CV_CUDEV_EXPR_CVTCOLOR_INST(BGRA_to_XYZ)
+CV_CUDEV_EXPR_CVTCOLOR_INST(BGR_to_XYZ4)
+CV_CUDEV_EXPR_CVTCOLOR_INST(BGRA_to_XYZ4)
+
+CV_CUDEV_EXPR_CVTCOLOR_INST(XYZ_to_RGB)
+CV_CUDEV_EXPR_CVTCOLOR_INST(XYZ4_to_RGB)
+CV_CUDEV_EXPR_CVTCOLOR_INST(XYZ_to_RGBA)
+CV_CUDEV_EXPR_CVTCOLOR_INST(XYZ4_to_RGBA)
+CV_CUDEV_EXPR_CVTCOLOR_INST(XYZ_to_BGR)
+CV_CUDEV_EXPR_CVTCOLOR_INST(XYZ4_to_BGR)
+CV_CUDEV_EXPR_CVTCOLOR_INST(XYZ_to_BGRA)
+CV_CUDEV_EXPR_CVTCOLOR_INST(XYZ4_to_BGRA)
+
+CV_CUDEV_EXPR_CVTCOLOR_INST(RGB_to_HSV)
+CV_CUDEV_EXPR_CVTCOLOR_INST(RGBA_to_HSV)
+CV_CUDEV_EXPR_CVTCOLOR_INST(RGB_to_HSV4)
+CV_CUDEV_EXPR_CVTCOLOR_INST(RGBA_to_HSV4)
+CV_CUDEV_EXPR_CVTCOLOR_INST(BGR_to_HSV)
+CV_CUDEV_EXPR_CVTCOLOR_INST(BGRA_to_HSV)
+CV_CUDEV_EXPR_CVTCOLOR_INST(BGR_to_HSV4)
+CV_CUDEV_EXPR_CVTCOLOR_INST(BGRA_to_HSV4)
+
+CV_CUDEV_EXPR_CVTCOLOR_INST(RGB_to_HSV_FULL)
+CV_CUDEV_EXPR_CVTCOLOR_INST(RGBA_to_HSV_FULL)
+CV_CUDEV_EXPR_CVTCOLOR_INST(RGB_to_HSV4_FULL)
+CV_CUDEV_EXPR_CVTCOLOR_INST(RGBA_to_HSV4_FULL)
+CV_CUDEV_EXPR_CVTCOLOR_INST(BGR_to_HSV_FULL)
+CV_CUDEV_EXPR_CVTCOLOR_INST(BGRA_to_HSV_FULL)
+CV_CUDEV_EXPR_CVTCOLOR_INST(BGR_to_HSV4_FULL)
+CV_CUDEV_EXPR_CVTCOLOR_INST(BGRA_to_HSV4_FULL)
+
+CV_CUDEV_EXPR_CVTCOLOR_INST(HSV_to_RGB)
+CV_CUDEV_EXPR_CVTCOLOR_INST(HSV_to_RGBA)
+CV_CUDEV_EXPR_CVTCOLOR_INST(HSV4_to_RGB)
+CV_CUDEV_EXPR_CVTCOLOR_INST(HSV4_to_RGBA)
+CV_CUDEV_EXPR_CVTCOLOR_INST(HSV_to_BGR)
+CV_CUDEV_EXPR_CVTCOLOR_INST(HSV_to_BGRA)
+CV_CUDEV_EXPR_CVTCOLOR_INST(HSV4_to_BGR)
+CV_CUDEV_EXPR_CVTCOLOR_INST(HSV4_to_BGRA)
+
+CV_CUDEV_EXPR_CVTCOLOR_INST(HSV_to_RGB_FULL)
+CV_CUDEV_EXPR_CVTCOLOR_INST(HSV_to_RGBA_FULL)
+CV_CUDEV_EXPR_CVTCOLOR_INST(HSV4_to_RGB_FULL)
+CV_CUDEV_EXPR_CVTCOLOR_INST(HSV4_to_RGBA_FULL)
+CV_CUDEV_EXPR_CVTCOLOR_INST(HSV_to_BGR_FULL)
+CV_CUDEV_EXPR_CVTCOLOR_INST(HSV_to_BGRA_FULL)
+CV_CUDEV_EXPR_CVTCOLOR_INST(HSV4_to_BGR_FULL)
+CV_CUDEV_EXPR_CVTCOLOR_INST(HSV4_to_BGRA_FULL)
+
+CV_CUDEV_EXPR_CVTCOLOR_INST(RGB_to_HLS)
+CV_CUDEV_EXPR_CVTCOLOR_INST(RGBA_to_HLS)
+CV_CUDEV_EXPR_CVTCOLOR_INST(RGB_to_HLS4)
+CV_CUDEV_EXPR_CVTCOLOR_INST(RGBA_to_HLS4)
+CV_CUDEV_EXPR_CVTCOLOR_INST(BGR_to_HLS)
+CV_CUDEV_EXPR_CVTCOLOR_INST(BGRA_to_HLS)
+CV_CUDEV_EXPR_CVTCOLOR_INST(BGR_to_HLS4)
+CV_CUDEV_EXPR_CVTCOLOR_INST(BGRA_to_HLS4)
+
+CV_CUDEV_EXPR_CVTCOLOR_INST(RGB_to_HLS_FULL)
+CV_CUDEV_EXPR_CVTCOLOR_INST(RGBA_to_HLS_FULL)
+CV_CUDEV_EXPR_CVTCOLOR_INST(RGB_to_HLS4_FULL)
+CV_CUDEV_EXPR_CVTCOLOR_INST(RGBA_to_HLS4_FULL)
+CV_CUDEV_EXPR_CVTCOLOR_INST(BGR_to_HLS_FULL)
+CV_CUDEV_EXPR_CVTCOLOR_INST(BGRA_to_HLS_FULL)
+CV_CUDEV_EXPR_CVTCOLOR_INST(BGR_to_HLS4_FULL)
+CV_CUDEV_EXPR_CVTCOLOR_INST(BGRA_to_HLS4_FULL)
+
+CV_CUDEV_EXPR_CVTCOLOR_INST(HLS_to_RGB)
+CV_CUDEV_EXPR_CVTCOLOR_INST(HLS_to_RGBA)
+CV_CUDEV_EXPR_CVTCOLOR_INST(HLS4_to_RGB)
+CV_CUDEV_EXPR_CVTCOLOR_INST(HLS4_to_RGBA)
+CV_CUDEV_EXPR_CVTCOLOR_INST(HLS_to_BGR)
+CV_CUDEV_EXPR_CVTCOLOR_INST(HLS_to_BGRA)
+CV_CUDEV_EXPR_CVTCOLOR_INST(HLS4_to_BGR)
+CV_CUDEV_EXPR_CVTCOLOR_INST(HLS4_to_BGRA)
+
+CV_CUDEV_EXPR_CVTCOLOR_INST(HLS_to_RGB_FULL)
+CV_CUDEV_EXPR_CVTCOLOR_INST(HLS_to_RGBA_FULL)
+CV_CUDEV_EXPR_CVTCOLOR_INST(HLS4_to_RGB_FULL)
+CV_CUDEV_EXPR_CVTCOLOR_INST(HLS4_to_RGBA_FULL)
+CV_CUDEV_EXPR_CVTCOLOR_INST(HLS_to_BGR_FULL)
+CV_CUDEV_EXPR_CVTCOLOR_INST(HLS_to_BGRA_FULL)
+CV_CUDEV_EXPR_CVTCOLOR_INST(HLS4_to_BGR_FULL)
+CV_CUDEV_EXPR_CVTCOLOR_INST(HLS4_to_BGRA_FULL)
+
+CV_CUDEV_EXPR_CVTCOLOR_INST(RGB_to_Lab)
+CV_CUDEV_EXPR_CVTCOLOR_INST(RGBA_to_Lab)
+CV_CUDEV_EXPR_CVTCOLOR_INST(RGB_to_Lab4)
+CV_CUDEV_EXPR_CVTCOLOR_INST(RGBA_to_Lab4)
+CV_CUDEV_EXPR_CVTCOLOR_INST(BGR_to_Lab)
+CV_CUDEV_EXPR_CVTCOLOR_INST(BGRA_to_Lab)
+CV_CUDEV_EXPR_CVTCOLOR_INST(BGR_to_Lab4)
+CV_CUDEV_EXPR_CVTCOLOR_INST(BGRA_to_Lab4)
+
+CV_CUDEV_EXPR_CVTCOLOR_INST(LRGB_to_Lab)
+CV_CUDEV_EXPR_CVTCOLOR_INST(LRGBA_to_Lab)
+CV_CUDEV_EXPR_CVTCOLOR_INST(LRGB_to_Lab4)
+CV_CUDEV_EXPR_CVTCOLOR_INST(LRGBA_to_Lab4)
+CV_CUDEV_EXPR_CVTCOLOR_INST(LBGR_to_Lab)
+CV_CUDEV_EXPR_CVTCOLOR_INST(LBGRA_to_Lab)
+CV_CUDEV_EXPR_CVTCOLOR_INST(LBGR_to_Lab4)
+CV_CUDEV_EXPR_CVTCOLOR_INST(LBGRA_to_Lab4)
+
+CV_CUDEV_EXPR_CVTCOLOR_INST(Lab_to_RGB)
+CV_CUDEV_EXPR_CVTCOLOR_INST(Lab4_to_RGB)
+CV_CUDEV_EXPR_CVTCOLOR_INST(Lab_to_RGBA)
+CV_CUDEV_EXPR_CVTCOLOR_INST(Lab4_to_RGBA)
+CV_CUDEV_EXPR_CVTCOLOR_INST(Lab_to_BGR)
+CV_CUDEV_EXPR_CVTCOLOR_INST(Lab4_to_BGR)
+CV_CUDEV_EXPR_CVTCOLOR_INST(Lab_to_BGRA)
+CV_CUDEV_EXPR_CVTCOLOR_INST(Lab4_to_BGRA)
+
+CV_CUDEV_EXPR_CVTCOLOR_INST(Lab_to_LRGB)
+CV_CUDEV_EXPR_CVTCOLOR_INST(Lab4_to_LRGB)
+CV_CUDEV_EXPR_CVTCOLOR_INST(Lab_to_LRGBA)
+CV_CUDEV_EXPR_CVTCOLOR_INST(Lab4_to_LRGBA)
+CV_CUDEV_EXPR_CVTCOLOR_INST(Lab_to_LBGR)
+CV_CUDEV_EXPR_CVTCOLOR_INST(Lab4_to_LBGR)
+CV_CUDEV_EXPR_CVTCOLOR_INST(Lab_to_LBGRA)
+CV_CUDEV_EXPR_CVTCOLOR_INST(Lab4_to_LBGRA)
+
+CV_CUDEV_EXPR_CVTCOLOR_INST(RGB_to_Luv)
+CV_CUDEV_EXPR_CVTCOLOR_INST(RGBA_to_Luv)
+CV_CUDEV_EXPR_CVTCOLOR_INST(RGB_to_Luv4)
+CV_CUDEV_EXPR_CVTCOLOR_INST(RGBA_to_Luv4)
+CV_CUDEV_EXPR_CVTCOLOR_INST(BGR_to_Luv)
+CV_CUDEV_EXPR_CVTCOLOR_INST(BGRA_to_Luv)
+CV_CUDEV_EXPR_CVTCOLOR_INST(BGR_to_Luv4)
+CV_CUDEV_EXPR_CVTCOLOR_INST(BGRA_to_Luv4)
+
+CV_CUDEV_EXPR_CVTCOLOR_INST(LRGB_to_Luv)
+CV_CUDEV_EXPR_CVTCOLOR_INST(LRGBA_to_Luv)
+CV_CUDEV_EXPR_CVTCOLOR_INST(LRGB_to_Luv4)
+CV_CUDEV_EXPR_CVTCOLOR_INST(LRGBA_to_Luv4)
+CV_CUDEV_EXPR_CVTCOLOR_INST(LBGR_to_Luv)
+CV_CUDEV_EXPR_CVTCOLOR_INST(LBGRA_to_Luv)
+CV_CUDEV_EXPR_CVTCOLOR_INST(LBGR_to_Luv4)
+CV_CUDEV_EXPR_CVTCOLOR_INST(LBGRA_to_Luv4)
+
+CV_CUDEV_EXPR_CVTCOLOR_INST(Luv_to_RGB)
+CV_CUDEV_EXPR_CVTCOLOR_INST(Luv4_to_RGB)
+CV_CUDEV_EXPR_CVTCOLOR_INST(Luv_to_RGBA)
+CV_CUDEV_EXPR_CVTCOLOR_INST(Luv4_to_RGBA)
+CV_CUDEV_EXPR_CVTCOLOR_INST(Luv_to_BGR)
+CV_CUDEV_EXPR_CVTCOLOR_INST(Luv4_to_BGR)
+CV_CUDEV_EXPR_CVTCOLOR_INST(Luv_to_BGRA)
+CV_CUDEV_EXPR_CVTCOLOR_INST(Luv4_to_BGRA)
+
+CV_CUDEV_EXPR_CVTCOLOR_INST(Luv_to_LRGB)
+CV_CUDEV_EXPR_CVTCOLOR_INST(Luv4_to_LRGB)
+CV_CUDEV_EXPR_CVTCOLOR_INST(Luv_to_LRGBA)
+CV_CUDEV_EXPR_CVTCOLOR_INST(Luv4_to_LRGBA)
+CV_CUDEV_EXPR_CVTCOLOR_INST(Luv_to_LBGR)
+CV_CUDEV_EXPR_CVTCOLOR_INST(Luv4_to_LBGR)
+CV_CUDEV_EXPR_CVTCOLOR_INST(Luv_to_LBGRA)
+CV_CUDEV_EXPR_CVTCOLOR_INST(Luv4_to_LBGRA)
+
+#undef CV_CUDEV_EXPR_CVTCOLOR_INST
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/expr/deriv.hpp

@@ -0,0 +1,121 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_EXPR_DERIV_HPP__
+#define __OPENCV_CUDEV_EXPR_DERIV_HPP__
+
+#include "../common.hpp"
+#include "../ptr2d/traits.hpp"
+#include "../ptr2d/deriv.hpp"
+#include "expr.hpp"
+
+namespace cv { namespace cudev {
+
+// derivX
+
+template <class SrcPtr>
+__host__ Expr<DerivXPtrSz<typename PtrTraits<SrcPtr>::ptr_type> >
+derivX_(const SrcPtr& src)
+{
+    return makeExpr(derivXPtr(src));
+}
+
+// derivY
+
+template <class SrcPtr>
+__host__ Expr<DerivYPtrSz<typename PtrTraits<SrcPtr>::ptr_type> >
+derivY_(const SrcPtr& src)
+{
+    return makeExpr(derivYPtr(src));
+}
+
+// sobelX
+
+template <class SrcPtr>
+__host__ Expr<SobelXPtrSz<typename PtrTraits<SrcPtr>::ptr_type> >
+sobelX_(const SrcPtr& src)
+{
+    return makeExpr(sobelXPtr(src));
+}
+
+// sobelY
+
+template <class SrcPtr>
+__host__ Expr<SobelYPtrSz<typename PtrTraits<SrcPtr>::ptr_type> >
+sobelY_(const SrcPtr& src)
+{
+    return makeExpr(sobelYPtr(src));
+}
+
+// scharrX
+
+template <class SrcPtr>
+__host__ Expr<ScharrXPtrSz<typename PtrTraits<SrcPtr>::ptr_type> >
+scharrX_(const SrcPtr& src)
+{
+    return makeExpr(scharrXPtr(src));
+}
+
+// scharrY
+
+template <class SrcPtr>
+__host__ Expr<ScharrYPtrSz<typename PtrTraits<SrcPtr>::ptr_type> >
+scharrY_(const SrcPtr& src)
+{
+    return makeExpr(scharrYPtr(src));
+}
+
+// laplacian
+
+template <int ksize, class SrcPtr>
+__host__ Expr<LaplacianPtrSz<ksize, typename PtrTraits<SrcPtr>::ptr_type> >
+laplacian_(const SrcPtr& src)
+{
+    return makeExpr(laplacianPtr<ksize>(src));
+}
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/expr/expr.hpp

@@ -0,0 +1,92 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_EXPR_EXPR_HPP__
+#define __OPENCV_CUDEV_EXPR_EXPR_HPP__
+
+#include "../common.hpp"
+#include "../ptr2d/traits.hpp"
+
+namespace cv { namespace cudev {
+
+template <class Body> struct Expr
+{
+    Body body;
+};
+
+template <class Body>
+__host__ Expr<Body> makeExpr(const Body& body)
+{
+    Expr<Body> e;
+    e.body = body;
+    return e;
+}
+
+template <class Body> struct PtrTraits< Expr<Body> >
+{
+    typedef Expr<Body>                         ptr_sz_type;
+    typedef typename PtrTraits<Body>::ptr_type ptr_type;
+
+    typedef typename ptr_type::value_type value_type;
+
+    __host__ static ptr_type shrinkPtr(const Expr<Body>& expr)
+    {
+        return PtrTraits<Body>::shrinkPtr(expr.body);
+    }
+
+    __host__ static int getRows(const Expr<Body>& expr)
+    {
+        return PtrTraits<Body>::getRows(expr.body);
+    }
+
+    __host__ static int getCols(const Expr<Body>& expr)
+    {
+        return PtrTraits<Body>::getCols(expr.body);
+    }
+};
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/expr/per_element_func.hpp

@@ -0,0 +1,132 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_EXPR_PER_ELEMENT_FUNC_HPP__
+#define __OPENCV_CUDEV_EXPR_PER_ELEMENT_FUNC_HPP__
+
+#include "../common.hpp"
+#include "../util/type_traits.hpp"
+#include "../ptr2d/traits.hpp"
+#include "../ptr2d/transform.hpp"
+#include "../ptr2d/lut.hpp"
+#include "../functional/functional.hpp"
+#include "expr.hpp"
+
+namespace cv { namespace cudev {
+
+// min/max
+
+template <class SrcPtr1, class SrcPtr2>
+__host__ Expr<BinaryTransformPtrSz<typename PtrTraits<SrcPtr1>::ptr_type, typename PtrTraits<SrcPtr2>::ptr_type, minimum<typename LargerType<typename PtrTraits<SrcPtr1>::value_type, typename PtrTraits<SrcPtr2>::value_type>::type> > >
+min_(const SrcPtr1& src1, const SrcPtr2& src2)
+{
+    return makeExpr(transformPtr(src1, src2, minimum<typename LargerType<typename PtrTraits<SrcPtr1>::value_type, typename PtrTraits<SrcPtr2>::value_type>::type>()));
+}
+
+template <class SrcPtr1, class SrcPtr2>
+__host__ Expr<BinaryTransformPtrSz<typename PtrTraits<SrcPtr1>::ptr_type, typename PtrTraits<SrcPtr2>::ptr_type, maximum<typename LargerType<typename PtrTraits<SrcPtr1>::value_type, typename PtrTraits<SrcPtr2>::value_type>::type> > >
+max_(const SrcPtr1& src1, const SrcPtr2& src2)
+{
+    return makeExpr(transformPtr(src1, src2, maximum<typename LargerType<typename PtrTraits<SrcPtr1>::value_type, typename PtrTraits<SrcPtr2>::value_type>::type>()));
+}
+
+// threshold
+
+template <class SrcPtr>
+__host__ Expr<UnaryTransformPtrSz<typename PtrTraits<SrcPtr>::ptr_type, ThreshBinaryFunc<typename PtrTraits<SrcPtr>::value_type> > >
+threshBinary_(const SrcPtr& src, typename PtrTraits<SrcPtr>::value_type thresh, typename PtrTraits<SrcPtr>::value_type maxVal)
+{
+    return makeExpr(transformPtr(src, thresh_binary_func(thresh, maxVal)));
+}
+
+template <class SrcPtr>
+__host__ Expr<UnaryTransformPtrSz<typename PtrTraits<SrcPtr>::ptr_type, ThreshBinaryInvFunc<typename PtrTraits<SrcPtr>::value_type> > >
+threshBinaryInv_(const SrcPtr& src, typename PtrTraits<SrcPtr>::value_type thresh, typename PtrTraits<SrcPtr>::value_type maxVal)
+{
+    return makeExpr(transformPtr(src, thresh_binary_inv_func(thresh, maxVal)));
+}
+
+template <class SrcPtr>
+__host__ Expr<UnaryTransformPtrSz<typename PtrTraits<SrcPtr>::ptr_type, ThreshTruncFunc<typename PtrTraits<SrcPtr>::value_type> > >
+threshTrunc_(const SrcPtr& src, typename PtrTraits<SrcPtr>::value_type thresh)
+{
+    return makeExpr(transformPtr(src, thresh_trunc_func(thresh)));
+}
+
+template <class SrcPtr>
+__host__ Expr<UnaryTransformPtrSz<typename PtrTraits<SrcPtr>::ptr_type, ThreshToZeroFunc<typename PtrTraits<SrcPtr>::value_type> > >
+threshToZero_(const SrcPtr& src, typename PtrTraits<SrcPtr>::value_type thresh)
+{
+    return makeExpr(transformPtr(src, thresh_to_zero_func(thresh)));
+}
+
+template <class SrcPtr>
+__host__ Expr<UnaryTransformPtrSz<typename PtrTraits<SrcPtr>::ptr_type, ThreshToZeroInvFunc<typename PtrTraits<SrcPtr>::value_type> > >
+threshToZeroInv_(const SrcPtr& src, typename PtrTraits<SrcPtr>::value_type thresh)
+{
+    return makeExpr(transformPtr(src, thresh_to_zero_inv_func(thresh)));
+}
+
+// cvt
+
+template <typename D, class SrcPtr>
+__host__ Expr<UnaryTransformPtrSz<typename PtrTraits<SrcPtr>::ptr_type, saturate_cast_func<typename PtrTraits<SrcPtr>::value_type, D> > >
+cvt_(const SrcPtr& src)
+{
+    return makeExpr(transformPtr(src, saturate_cast_func<typename PtrTraits<SrcPtr>::value_type, D>()));
+}
+
+// lut
+
+template <class SrcPtr, class TablePtr>
+__host__ Expr<LutPtrSz<typename PtrTraits<SrcPtr>::ptr_type, typename PtrTraits<TablePtr>::ptr_type> >
+lut_(const SrcPtr& src, const TablePtr& tbl)
+{
+    return makeExpr(lutPtr(src, tbl));
+}
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/expr/reduction.hpp

@@ -0,0 +1,259 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_EXPR_REDUCTION_HPP__
+#define __OPENCV_CUDEV_EXPR_REDUCTION_HPP__
+
+#include "../common.hpp"
+#include "../grid/glob_reduce.hpp"
+#include "../grid/histogram.hpp"
+#include "../grid/integral.hpp"
+#include "../grid/reduce_to_vec.hpp"
+#include "../ptr2d/traits.hpp"
+#include "expr.hpp"
+
+namespace cv { namespace cudev {
+
+// sum
+
+template <class SrcPtr> struct SumExprBody
+{
+    SrcPtr src;
+
+    template <typename T>
+    __host__ void assignTo(GpuMat_<T>& dst, Stream& stream = Stream::Null()) const
+    {
+        gridCalcSum(src, dst, stream);
+    }
+};
+
+template <class SrcPtr>
+__host__ Expr<SumExprBody<SrcPtr> >
+sum_(const SrcPtr& src)
+{
+    SumExprBody<SrcPtr> body;
+    body.src = src;
+    return makeExpr(body);
+}
+
+// minVal
+
+template <class SrcPtr> struct FindMinValExprBody
+{
+    SrcPtr src;
+
+    template <typename T>
+    __host__ void assignTo(GpuMat_<T>& dst, Stream& stream = Stream::Null()) const
+    {
+        gridFindMinVal(src, dst, stream);
+    }
+};
+
+template <class SrcPtr>
+__host__ Expr<FindMinValExprBody<SrcPtr> >
+minVal_(const SrcPtr& src)
+{
+    FindMinValExprBody<SrcPtr> body;
+    body.src = src;
+    return makeExpr(body);
+}
+
+// maxVal
+
+template <class SrcPtr> struct FindMaxValExprBody
+{
+    SrcPtr src;
+
+    template <typename T>
+    __host__ void assignTo(GpuMat_<T>& dst, Stream& stream = Stream::Null()) const
+    {
+        gridFindMaxVal(src, dst, stream);
+    }
+};
+
+template <class SrcPtr>
+__host__ Expr<FindMaxValExprBody<SrcPtr> >
+maxVal_(const SrcPtr& src)
+{
+    FindMaxValExprBody<SrcPtr> body;
+    body.src = src;
+    return makeExpr(body);
+}
+
+// minMaxVal
+
+template <class SrcPtr> struct FindMinMaxValExprBody
+{
+    SrcPtr src;
+
+    template <typename T>
+    __host__ void assignTo(GpuMat_<T>& dst, Stream& stream = Stream::Null()) const
+    {
+        gridFindMinMaxVal(src, dst, stream);
+    }
+};
+
+template <class SrcPtr>
+__host__ Expr<FindMinMaxValExprBody<SrcPtr> >
+minMaxVal_(const SrcPtr& src)
+{
+    FindMinMaxValExprBody<SrcPtr> body;
+    body.src = src;
+    return makeExpr(body);
+}
+
+// countNonZero
+
+template <class SrcPtr> struct CountNonZeroExprBody
+{
+    SrcPtr src;
+
+    template <typename T>
+    __host__ void assignTo(GpuMat_<T>& dst, Stream& stream = Stream::Null()) const
+    {
+        gridCountNonZero(src, dst, stream);
+    }
+};
+
+template <class SrcPtr>
+__host__ Expr<CountNonZeroExprBody<SrcPtr> >
+countNonZero_(const SrcPtr& src)
+{
+    CountNonZeroExprBody<SrcPtr> body;
+    body.src = src;
+    return makeExpr(body);
+}
+
+// reduceToRow
+
+template <class Reductor, class SrcPtr> struct ReduceToRowBody
+{
+    SrcPtr src;
+
+    template <typename T>
+    __host__ void assignTo(GpuMat_<T>& dst, Stream& stream = Stream::Null()) const
+    {
+        gridReduceToRow<Reductor>(src, dst, stream);
+    }
+};
+
+template <class Reductor, class SrcPtr>
+__host__ Expr<ReduceToRowBody<Reductor, SrcPtr> >
+reduceToRow_(const SrcPtr& src)
+{
+    ReduceToRowBody<Reductor, SrcPtr> body;
+    body.src = src;
+    return makeExpr(body);
+}
+
+// reduceToColumn
+
+template <class Reductor, class SrcPtr> struct ReduceToColumnBody
+{
+    SrcPtr src;
+
+    template <typename T>
+    __host__ void assignTo(GpuMat_<T>& dst, Stream& stream = Stream::Null()) const
+    {
+        gridReduceToColumn<Reductor>(src, dst, stream);
+    }
+};
+
+template <class Reductor, class SrcPtr>
+__host__ Expr<ReduceToColumnBody<Reductor, SrcPtr> >
+reduceToColumn_(const SrcPtr& src)
+{
+    ReduceToColumnBody<Reductor, SrcPtr> body;
+    body.src = src;
+    return makeExpr(body);
+}
+
+// histogram
+
+template <int BIN_COUNT, class SrcPtr> struct HistogramBody
+{
+    SrcPtr src;
+
+    template <typename T>
+    __host__ void assignTo(GpuMat_<T>& dst, Stream& stream = Stream::Null()) const
+    {
+        gridHistogram<BIN_COUNT>(src, dst, stream);
+    }
+};
+
+template <int BIN_COUNT, class SrcPtr>
+__host__ Expr<HistogramBody<BIN_COUNT, SrcPtr> >
+histogram_(const SrcPtr& src)
+{
+    HistogramBody<BIN_COUNT, SrcPtr> body;
+    body.src = src;
+    return makeExpr(body);
+}
+
+// integral
+
+template <class SrcPtr> struct IntegralBody
+{
+    SrcPtr src;
+
+    template <typename T>
+    __host__ void assignTo(GpuMat_<T>& dst, Stream& stream = Stream::Null()) const
+    {
+        gridIntegral(src, dst, stream);
+    }
+};
+
+template <class SrcPtr>
+__host__ Expr<IntegralBody<SrcPtr> >
+integral_(const SrcPtr& src)
+{
+    IntegralBody<SrcPtr> body;
+    body.src = src;
+    return makeExpr(body);
+}
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/expr/unary_func.hpp

@@ -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.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_EXPR_UNARY_FUNC_HPP__
+#define __OPENCV_CUDEV_EXPR_UNARY_FUNC_HPP__
+
+#include "../common.hpp"
+#include "../ptr2d/traits.hpp"
+#include "../ptr2d/transform.hpp"
+#include "../functional/functional.hpp"
+#include "expr.hpp"
+
+namespace cv { namespace cudev {
+
+#define CV_CUDEV_EXPR_UNARY_FUNC(name) \
+    template <class SrcPtr> \
+    __host__ Expr<UnaryTransformPtrSz<typename PtrTraits<SrcPtr>::ptr_type, name ## _func<typename PtrTraits<SrcPtr>::value_type> > > \
+    name ## _(const SrcPtr& src) \
+    { \
+        return makeExpr(transformPtr(src, name ## _func<typename PtrTraits<SrcPtr>::value_type>())); \
+    }
+
+CV_CUDEV_EXPR_UNARY_FUNC(abs)
+CV_CUDEV_EXPR_UNARY_FUNC(sqr)
+CV_CUDEV_EXPR_UNARY_FUNC(sqrt)
+CV_CUDEV_EXPR_UNARY_FUNC(exp)
+CV_CUDEV_EXPR_UNARY_FUNC(exp2)
+CV_CUDEV_EXPR_UNARY_FUNC(exp10)
+CV_CUDEV_EXPR_UNARY_FUNC(log)
+CV_CUDEV_EXPR_UNARY_FUNC(log2)
+CV_CUDEV_EXPR_UNARY_FUNC(log10)
+CV_CUDEV_EXPR_UNARY_FUNC(sin)
+CV_CUDEV_EXPR_UNARY_FUNC(cos)
+CV_CUDEV_EXPR_UNARY_FUNC(tan)
+CV_CUDEV_EXPR_UNARY_FUNC(asin)
+CV_CUDEV_EXPR_UNARY_FUNC(acos)
+CV_CUDEV_EXPR_UNARY_FUNC(atan)
+CV_CUDEV_EXPR_UNARY_FUNC(sinh)
+CV_CUDEV_EXPR_UNARY_FUNC(cosh)
+CV_CUDEV_EXPR_UNARY_FUNC(tanh)
+CV_CUDEV_EXPR_UNARY_FUNC(asinh)
+CV_CUDEV_EXPR_UNARY_FUNC(acosh)
+CV_CUDEV_EXPR_UNARY_FUNC(atanh)
+
+#undef CV_CUDEV_EXPR_UNARY_FUNC
+
+template <class SrcPtr>
+__host__ Expr<UnaryTransformPtrSz<typename PtrTraits<SrcPtr>::ptr_type, Binder2nd<pow_func<typename PtrTraits<SrcPtr>::value_type> > > >
+pow_(const SrcPtr& src, float power)
+{
+    return makeExpr(transformPtr(src, bind2nd(pow_func<typename PtrTraits<SrcPtr>::value_type>(), power)));
+}
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/expr/unary_op.hpp

@@ -0,0 +1,94 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_EXPR_UNARY_OP_HPP__
+#define __OPENCV_CUDEV_EXPR_UNARY_OP_HPP__
+
+#include "../common.hpp"
+#include "../ptr2d/traits.hpp"
+#include "../ptr2d/transform.hpp"
+#include "../ptr2d/gpumat.hpp"
+#include "../ptr2d/texture.hpp"
+#include "../ptr2d/glob.hpp"
+#include "../functional/functional.hpp"
+#include "expr.hpp"
+
+namespace cv { namespace cudev {
+
+#define CV_CUDEV_EXPR_UNOP_INST(op, functor) \
+    template <typename T> \
+    __host__ Expr<UnaryTransformPtrSz<typename PtrTraits<GpuMat_<T> >::ptr_type, functor<T> > > \
+    operator op(const GpuMat_<T>& src) \
+    { \
+        return makeExpr(transformPtr(src, functor<T>())); \
+    } \
+    template <typename T> \
+    __host__ Expr<UnaryTransformPtrSz<typename PtrTraits<GlobPtrSz<T> >::ptr_type, functor<T> > > \
+    operator op(const GlobPtrSz<T>& src) \
+    { \
+        return makeExpr(transformPtr(src, functor<T>())); \
+    } \
+    template <typename T> \
+    __host__ Expr<UnaryTransformPtrSz<typename PtrTraits<Texture<T> >::ptr_type, functor<T> > > \
+    operator op(const Texture<T>& src) \
+    { \
+        return makeExpr(transformPtr(src, functor<T>())); \
+    } \
+    template <class Body> \
+    __host__ Expr<UnaryTransformPtrSz<typename PtrTraits<Body>::ptr_type, functor<typename Body::value_type> > > \
+    operator op(const Expr<Body>& src) \
+    { \
+        return makeExpr(transformPtr(src.body, functor<typename Body::value_type>())); \
+    }
+
+CV_CUDEV_EXPR_UNOP_INST(-, negate)
+CV_CUDEV_EXPR_UNOP_INST(!, logical_not)
+CV_CUDEV_EXPR_UNOP_INST(~, bit_not)
+
+#undef CV_CUDEV_EXPR_UNOP_INST
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/expr/warping.hpp

@@ -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.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_EXPR_WARPING_HPP__
+#define __OPENCV_CUDEV_EXPR_WARPING_HPP__
+
+#include "../common.hpp"
+#include "../ptr2d/traits.hpp"
+#include "../ptr2d/resize.hpp"
+#include "../ptr2d/remap.hpp"
+#include "../ptr2d/warping.hpp"
+#include "../grid/pyramids.hpp"
+#include "../grid/transpose.hpp"
+#include "expr.hpp"
+
+namespace cv { namespace cudev {
+
+// resize
+
+template <class SrcPtr>
+__host__ Expr<ResizePtrSz<typename PtrTraits<SrcPtr>::ptr_type> >
+resize_(const SrcPtr& src, float fx, float fy)
+{
+    return makeExpr(resizePtr(src, fx, fy));
+}
+
+// remap
+
+template <class SrcPtr, class MapPtr>
+__host__ Expr<RemapPtr1Sz<typename PtrTraits<SrcPtr>::ptr_type, typename PtrTraits<MapPtr>::ptr_type> >
+remap_(const SrcPtr& src, const MapPtr& map)
+{
+    return makeExpr(remapPtr(src, map));
+}
+
+template <class SrcPtr, class MapXPtr, class MapYPtr>
+__host__ Expr<RemapPtr2Sz<typename PtrTraits<SrcPtr>::ptr_type, typename PtrTraits<MapXPtr>::ptr_type, typename PtrTraits<MapYPtr>::ptr_type> >
+remap_(const SrcPtr& src, const MapXPtr& mapx, const MapYPtr& mapy)
+{
+    return makeExpr(remapPtr(src, mapx, mapy));
+}
+
+// warpAffine
+
+template <class SrcPtr>
+__host__ Expr<RemapPtr1Sz<typename PtrTraits<SrcPtr>::ptr_type, AffineMapPtr> >
+warpAffine_(const SrcPtr& src, Size dstSize, const GpuMat_<float>& warpMat)
+{
+    return makeExpr(warpAffinePtr(src, dstSize, warpMat));
+}
+
+// warpPerspective
+
+template <class SrcPtr>
+__host__ Expr<RemapPtr1Sz<typename PtrTraits<SrcPtr>::ptr_type, PerspectiveMapPtr> >
+warpPerspective_(const SrcPtr& src, Size dstSize, const GpuMat_<float>& warpMat)
+{
+    return makeExpr(warpPerspectivePtr(src, dstSize, warpMat));
+}
+
+// pyrDown
+
+template <class SrcPtr> struct PyrDownBody
+{
+    SrcPtr src;
+
+    template <typename T>
+    __host__ void assignTo(GpuMat_<T>& dst, Stream& stream = Stream::Null()) const
+    {
+        gridPyrDown(src, dst, stream);
+    }
+};
+
+template <class SrcPtr>
+__host__ Expr<PyrDownBody<SrcPtr> >
+pyrDown_(const SrcPtr& src)
+{
+    PyrDownBody<SrcPtr> body;
+    body.src = src;
+    return makeExpr(body);
+}
+
+// pyrUp
+
+template <class SrcPtr> struct PyrUpBody
+{
+    SrcPtr src;
+
+    template <typename T>
+    __host__ void assignTo(GpuMat_<T>& dst, Stream& stream = Stream::Null()) const
+    {
+        gridPyrUp(src, dst, stream);
+    }
+};
+
+template <class SrcPtr>
+__host__ Expr<PyrUpBody<SrcPtr> >
+pyrUp_(const SrcPtr& src)
+{
+    PyrUpBody<SrcPtr> body;
+    body.src = src;
+    return makeExpr(body);
+}
+
+// transpose
+
+template <class SrcPtr> struct TransposeBody
+{
+    SrcPtr src;
+
+    template <typename T>
+    __host__ void assignTo(GpuMat_<T>& dst, Stream& stream = Stream::Null()) const
+    {
+        gridTranspose(src, dst, stream);
+    }
+};
+
+template <class SrcPtr>
+__host__ Expr<TransposeBody<SrcPtr> >
+transpose_(const SrcPtr& src)
+{
+    TransposeBody<SrcPtr> body;
+    body.src = src;
+    return makeExpr(body);
+}
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/functional/color_cvt.hpp

@@ -0,0 +1,474 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_FUNCTIONAL_COLOR_CVT_HPP__
+#define __OPENCV_CUDEV_FUNCTIONAL_COLOR_CVT_HPP__
+
+#include "../common.hpp"
+#include "detail/color_cvt.hpp"
+
+namespace cv { namespace cudev {
+
+// Various 3/4-channel to 3/4-channel RGB transformations
+
+#define CV_CUDEV_RGB2RGB_INST(name, scn, dcn, bidx) \
+    template <typename SrcDepth> struct name ## _func : cv::cudev::color_cvt_detail::RGB2RGB<SrcDepth, scn, dcn, bidx> \
+    { \
+    };
+
+CV_CUDEV_RGB2RGB_INST(BGR_to_RGB, 3, 3, 2)
+CV_CUDEV_RGB2RGB_INST(BGR_to_BGRA, 3, 4, 0)
+CV_CUDEV_RGB2RGB_INST(BGR_to_RGBA, 3, 4, 2)
+CV_CUDEV_RGB2RGB_INST(BGRA_to_BGR, 4, 3, 0)
+CV_CUDEV_RGB2RGB_INST(BGRA_to_RGB, 4, 3, 2)
+CV_CUDEV_RGB2RGB_INST(BGRA_to_RGBA, 4, 4, 2)
+
+#undef CV_CUDEV_RGB2RGB_INST
+
+// RGB to Grayscale
+
+#define CV_CUDEV_RGB2GRAY_INST(name, scn, bidx) \
+    template <typename SrcDepth> struct name ## _func : cv::cudev::color_cvt_detail::RGB2Gray<SrcDepth, scn, bidx> \
+    { \
+    };
+
+CV_CUDEV_RGB2GRAY_INST(RGB_to_GRAY, 3, 2)
+CV_CUDEV_RGB2GRAY_INST(BGR_to_GRAY, 3, 0)
+CV_CUDEV_RGB2GRAY_INST(RGBA_to_GRAY, 4, 2)
+CV_CUDEV_RGB2GRAY_INST(BGRA_to_GRAY, 4, 0)
+
+#undef CV_CUDEV_RGB2GRAY_INST
+
+// Grayscale to RGB
+
+#define CV_CUDEV_GRAY2RGB_INST(name, dcn) \
+    template <typename SrcDepth> struct name ## _func : cv::cudev::color_cvt_detail::Gray2RGB<SrcDepth, dcn> \
+    { \
+    };
+
+CV_CUDEV_GRAY2RGB_INST(GRAY_to_BGR, 3)
+CV_CUDEV_GRAY2RGB_INST(GRAY_to_BGRA, 4)
+
+#undef CV_CUDEV_GRAY2RGB_INST
+
+// RGB to YUV
+
+#define CV_CUDEV_RGB2YUV_INST(name, scn, dcn, bidx) \
+    template <typename SrcDepth> struct name ## _func : cv::cudev::color_cvt_detail::RGB2YUV<SrcDepth, scn, dcn, bidx> \
+    { \
+    };
+
+CV_CUDEV_RGB2YUV_INST(RGB_to_YUV, 3, 3, 0)
+CV_CUDEV_RGB2YUV_INST(RGBA_to_YUV, 4, 3, 0)
+CV_CUDEV_RGB2YUV_INST(RGB_to_YUV4, 3, 4, 0)
+CV_CUDEV_RGB2YUV_INST(RGBA_to_YUV4, 4, 4, 0)
+CV_CUDEV_RGB2YUV_INST(BGR_to_YUV, 3, 3, 2)
+CV_CUDEV_RGB2YUV_INST(BGRA_to_YUV, 4, 3, 2)
+CV_CUDEV_RGB2YUV_INST(BGR_to_YUV4, 3, 4, 2)
+CV_CUDEV_RGB2YUV_INST(BGRA_to_YUV4, 4, 4, 2)
+
+#undef CV_CUDEV_RGB2YUV_INST
+
+// YUV to RGB
+
+#define CV_CUDEV_YUV2RGB_INST(name, scn, dcn, bidx) \
+    template <typename SrcDepth> struct name ## _func : cv::cudev::color_cvt_detail::YUV2RGB<SrcDepth, scn, dcn, bidx> \
+    { \
+    };
+
+CV_CUDEV_YUV2RGB_INST(YUV_to_RGB, 3, 3, 0)
+CV_CUDEV_YUV2RGB_INST(YUV_to_RGBA, 3, 4, 0)
+CV_CUDEV_YUV2RGB_INST(YUV4_to_RGB, 4, 3, 0)
+CV_CUDEV_YUV2RGB_INST(YUV4_to_RGBA, 4, 4, 0)
+CV_CUDEV_YUV2RGB_INST(YUV_to_BGR, 3, 3, 2)
+CV_CUDEV_YUV2RGB_INST(YUV_to_BGRA, 3, 4, 2)
+CV_CUDEV_YUV2RGB_INST(YUV4_to_BGR, 4, 3, 2)
+CV_CUDEV_YUV2RGB_INST(YUV4_to_BGRA, 4, 4, 2)
+
+#undef CV_CUDEV_YUV2RGB_INST
+
+// RGB to YCrCb
+
+#define CV_CUDEV_RGB2YCrCb_INST(name, scn, dcn, bidx) \
+    template <typename SrcDepth> struct name ## _func : cv::cudev::color_cvt_detail::RGB2YCrCb<SrcDepth, scn, dcn, bidx> \
+    { \
+    };
+
+CV_CUDEV_RGB2YCrCb_INST(RGB_to_YCrCb, 3, 3, 2)
+CV_CUDEV_RGB2YCrCb_INST(RGBA_to_YCrCb, 4, 3, 2)
+CV_CUDEV_RGB2YCrCb_INST(RGB_to_YCrCb4, 3, 4, 2)
+CV_CUDEV_RGB2YCrCb_INST(RGBA_to_YCrCb4, 4, 4, 2)
+CV_CUDEV_RGB2YCrCb_INST(BGR_to_YCrCb, 3, 3, 0)
+CV_CUDEV_RGB2YCrCb_INST(BGRA_to_YCrCb, 4, 3, 0)
+CV_CUDEV_RGB2YCrCb_INST(BGR_to_YCrCb4, 3, 4, 0)
+CV_CUDEV_RGB2YCrCb_INST(BGRA_to_YCrCb4, 4, 4, 0)
+
+#undef CV_CUDEV_RGB2YCrCb_INST
+
+// YCrCb to RGB
+
+#define CV_CUDEV_YCrCb2RGB_INST(name, scn, dcn, bidx) \
+    template <typename SrcDepth> struct name ## _func : cv::cudev::color_cvt_detail::YCrCb2RGB<SrcDepth, scn, dcn, bidx> \
+    { \
+    };
+
+CV_CUDEV_YCrCb2RGB_INST(YCrCb_to_RGB, 3, 3, 2)
+CV_CUDEV_YCrCb2RGB_INST(YCrCb_to_RGBA, 3, 4, 2)
+CV_CUDEV_YCrCb2RGB_INST(YCrCb4_to_RGB, 4, 3, 2)
+CV_CUDEV_YCrCb2RGB_INST(YCrCb4_to_RGBA, 4, 4, 2)
+CV_CUDEV_YCrCb2RGB_INST(YCrCb_to_BGR, 3, 3, 0)
+CV_CUDEV_YCrCb2RGB_INST(YCrCb_to_BGRA, 3, 4, 0)
+CV_CUDEV_YCrCb2RGB_INST(YCrCb4_to_BGR, 4, 3, 0)
+CV_CUDEV_YCrCb2RGB_INST(YCrCb4_to_BGRA, 4, 4, 0)
+
+#undef CV_CUDEV_YCrCb2RGB_INST
+
+// RGB to XYZ
+
+#define CV_CUDEV_RGB2XYZ_INST(name, scn, dcn, bidx) \
+    template <typename SrcDepth> struct name ## _func : cv::cudev::color_cvt_detail::RGB2XYZ<SrcDepth, scn, dcn, bidx> \
+    { \
+    };
+
+CV_CUDEV_RGB2XYZ_INST(RGB_to_XYZ, 3, 3, 2)
+CV_CUDEV_RGB2XYZ_INST(RGBA_to_XYZ, 4, 3, 2)
+CV_CUDEV_RGB2XYZ_INST(RGB_to_XYZ4, 3, 4, 2)
+CV_CUDEV_RGB2XYZ_INST(RGBA_to_XYZ4, 4, 4, 2)
+CV_CUDEV_RGB2XYZ_INST(BGR_to_XYZ, 3, 3, 0)
+CV_CUDEV_RGB2XYZ_INST(BGRA_to_XYZ, 4, 3, 0)
+CV_CUDEV_RGB2XYZ_INST(BGR_to_XYZ4, 3, 4, 0)
+CV_CUDEV_RGB2XYZ_INST(BGRA_to_XYZ4, 4, 4, 0)
+
+#undef CV_CUDEV_RGB2XYZ_INST
+
+// XYZ to RGB
+
+#define CV_CUDEV_XYZ2RGB_INST(name, scn, dcn, bidx) \
+    template <typename SrcDepth> struct name ## _func : cv::cudev::color_cvt_detail::XYZ2RGB<SrcDepth, scn, dcn, bidx> \
+    { \
+    };
+
+CV_CUDEV_XYZ2RGB_INST(XYZ_to_RGB, 3, 3, 2)
+CV_CUDEV_XYZ2RGB_INST(XYZ4_to_RGB, 4, 3, 2)
+CV_CUDEV_XYZ2RGB_INST(XYZ_to_RGBA, 3, 4, 2)
+CV_CUDEV_XYZ2RGB_INST(XYZ4_to_RGBA, 4, 4, 2)
+CV_CUDEV_XYZ2RGB_INST(XYZ_to_BGR, 3, 3, 0)
+CV_CUDEV_XYZ2RGB_INST(XYZ4_to_BGR, 4, 3, 0)
+CV_CUDEV_XYZ2RGB_INST(XYZ_to_BGRA, 3, 4, 0)
+CV_CUDEV_XYZ2RGB_INST(XYZ4_to_BGRA, 4, 4, 0)
+
+#undef CV_CUDEV_XYZ2RGB_INST
+
+// RGB to HSV
+
+#define CV_CUDEV_RGB2HSV_INST(name, scn, dcn, bidx) \
+    template <typename SrcDepth> struct name ## _func : cv::cudev::color_cvt_detail::RGB2HSV<SrcDepth, scn, dcn, bidx, 180> \
+    { \
+    }; \
+    template <typename SrcDepth> struct name ## _FULL ## _func : cv::cudev::color_cvt_detail::RGB2HSV<SrcDepth, scn, dcn, bidx, 256> \
+    { \
+    }; \
+    template <> struct name ## _func<float> : cv::cudev::color_cvt_detail::RGB2HSV<float, scn, dcn, bidx, 360> \
+    { \
+    }; \
+    template <> struct name ## _FULL ## _func<float> : cv::cudev::color_cvt_detail::RGB2HSV<float, scn, dcn, bidx, 360> \
+    { \
+    };
+
+CV_CUDEV_RGB2HSV_INST(RGB_to_HSV, 3, 3, 2)
+CV_CUDEV_RGB2HSV_INST(RGBA_to_HSV, 4, 3, 2)
+CV_CUDEV_RGB2HSV_INST(RGB_to_HSV4, 3, 4, 2)
+CV_CUDEV_RGB2HSV_INST(RGBA_to_HSV4, 4, 4, 2)
+CV_CUDEV_RGB2HSV_INST(BGR_to_HSV, 3, 3, 0)
+CV_CUDEV_RGB2HSV_INST(BGRA_to_HSV, 4, 3, 0)
+CV_CUDEV_RGB2HSV_INST(BGR_to_HSV4, 3, 4, 0)
+CV_CUDEV_RGB2HSV_INST(BGRA_to_HSV4, 4, 4, 0)
+
+#undef CV_CUDEV_RGB2HSV_INST
+
+// HSV to RGB
+
+#define CV_CUDEV_HSV2RGB_INST(name, scn, dcn, bidx) \
+    template <typename SrcDepth> struct name ## _func : cv::cudev::color_cvt_detail::HSV2RGB<SrcDepth, scn, dcn, bidx, 180> \
+    { \
+    }; \
+    template <typename SrcDepth> struct name ## _FULL ## _func : cv::cudev::color_cvt_detail::HSV2RGB<SrcDepth, scn, dcn, bidx, 255> \
+    { \
+    }; \
+    template <> struct name ## _func<float> : cv::cudev::color_cvt_detail::HSV2RGB<float, scn, dcn, bidx, 360> \
+    { \
+    }; \
+    template <> struct name ## _FULL ## _func<float> : cv::cudev::color_cvt_detail::HSV2RGB<float, scn, dcn, bidx, 360> \
+    { \
+    };
+
+CV_CUDEV_HSV2RGB_INST(HSV_to_RGB, 3, 3, 2)
+CV_CUDEV_HSV2RGB_INST(HSV_to_RGBA, 3, 4, 2)
+CV_CUDEV_HSV2RGB_INST(HSV4_to_RGB, 4, 3, 2)
+CV_CUDEV_HSV2RGB_INST(HSV4_to_RGBA, 4, 4, 2)
+CV_CUDEV_HSV2RGB_INST(HSV_to_BGR, 3, 3, 0)
+CV_CUDEV_HSV2RGB_INST(HSV_to_BGRA, 3, 4, 0)
+CV_CUDEV_HSV2RGB_INST(HSV4_to_BGR, 4, 3, 0)
+CV_CUDEV_HSV2RGB_INST(HSV4_to_BGRA, 4, 4, 0)
+
+#undef CV_CUDEV_HSV2RGB_INST
+
+// RGB to HLS
+
+#define CV_CUDEV_RGB2HLS_INST(name, scn, dcn, bidx) \
+    template <typename SrcDepth> struct name ## _func : cv::cudev::color_cvt_detail::RGB2HLS<SrcDepth, scn, dcn, bidx, 180> \
+    { \
+    }; \
+    template <typename SrcDepth> struct name ## _FULL ## _func : cv::cudev::color_cvt_detail::RGB2HLS<SrcDepth, scn, dcn, bidx, 256> \
+    { \
+    }; \
+    template <> struct name ## _func<float> : cv::cudev::color_cvt_detail::RGB2HLS<float, scn, dcn, bidx, 360> \
+    { \
+    }; \
+    template <> struct name ## _FULL ## _func<float> : cv::cudev::color_cvt_detail::RGB2HLS<float, scn, dcn, bidx, 360> \
+    { \
+    };
+
+CV_CUDEV_RGB2HLS_INST(RGB_to_HLS, 3, 3, 2)
+CV_CUDEV_RGB2HLS_INST(RGBA_to_HLS, 4, 3, 2)
+CV_CUDEV_RGB2HLS_INST(RGB_to_HLS4, 3, 4, 2)
+CV_CUDEV_RGB2HLS_INST(RGBA_to_HLS4, 4, 4, 2)
+CV_CUDEV_RGB2HLS_INST(BGR_to_HLS, 3, 3, 0)
+CV_CUDEV_RGB2HLS_INST(BGRA_to_HLS, 4, 3, 0)
+CV_CUDEV_RGB2HLS_INST(BGR_to_HLS4, 3, 4, 0)
+CV_CUDEV_RGB2HLS_INST(BGRA_to_HLS4, 4, 4, 0)
+
+#undef CV_CUDEV_RGB2HLS_INST
+
+// HLS to RGB
+
+#define CV_CUDEV_HLS2RGB_INST(name, scn, dcn, bidx) \
+    template <typename SrcDepth> struct name ## _func : cv::cudev::color_cvt_detail::HLS2RGB<SrcDepth, scn, dcn, bidx, 180> \
+    { \
+    }; \
+    template <typename SrcDepth> struct name ## _FULL ## _func : cv::cudev::color_cvt_detail::HLS2RGB<SrcDepth, scn, dcn, bidx, 255> \
+    { \
+    }; \
+    template <> struct name ## _func<float> : cv::cudev::color_cvt_detail::HLS2RGB<float, scn, dcn, bidx, 360> \
+    { \
+    }; \
+    template <> struct name ## _FULL ## _func<float> : cv::cudev::color_cvt_detail::HLS2RGB<float, scn, dcn, bidx, 360> \
+    { \
+    };
+
+CV_CUDEV_HLS2RGB_INST(HLS_to_RGB, 3, 3, 2)
+CV_CUDEV_HLS2RGB_INST(HLS_to_RGBA, 3, 4, 2)
+CV_CUDEV_HLS2RGB_INST(HLS4_to_RGB, 4, 3, 2)
+CV_CUDEV_HLS2RGB_INST(HLS4_to_RGBA, 4, 4, 2)
+CV_CUDEV_HLS2RGB_INST(HLS_to_BGR, 3, 3, 0)
+CV_CUDEV_HLS2RGB_INST(HLS_to_BGRA, 3, 4, 0)
+CV_CUDEV_HLS2RGB_INST(HLS4_to_BGR, 4, 3, 0)
+CV_CUDEV_HLS2RGB_INST(HLS4_to_BGRA, 4, 4, 0)
+
+#undef CV_CUDEV_HLS2RGB_INST
+
+// RGB to Lab
+
+#define CV_CUDEV_RGB2Lab_INST(name, scn, dcn, sRGB, blueIdx) \
+    template <typename SrcDepth> struct name ## _func : cv::cudev::color_cvt_detail::RGB2Lab<SrcDepth, scn, dcn, sRGB, blueIdx> \
+    { \
+    };
+
+CV_CUDEV_RGB2Lab_INST(RGB_to_Lab, 3, 3, true, 2)
+CV_CUDEV_RGB2Lab_INST(RGBA_to_Lab, 4, 3, true, 2)
+CV_CUDEV_RGB2Lab_INST(RGB_to_Lab4, 3, 4, true, 2)
+CV_CUDEV_RGB2Lab_INST(RGBA_to_Lab4, 4, 4, true, 2)
+CV_CUDEV_RGB2Lab_INST(BGR_to_Lab, 3, 3, true, 0)
+CV_CUDEV_RGB2Lab_INST(BGRA_to_Lab, 4, 3, true, 0)
+CV_CUDEV_RGB2Lab_INST(BGR_to_Lab4, 3, 4, true, 0)
+CV_CUDEV_RGB2Lab_INST(BGRA_to_Lab4, 4, 4, true, 0)
+
+CV_CUDEV_RGB2Lab_INST(LRGB_to_Lab, 3, 3, false, 2)
+CV_CUDEV_RGB2Lab_INST(LRGBA_to_Lab, 4, 3, false, 2)
+CV_CUDEV_RGB2Lab_INST(LRGB_to_Lab4, 3, 4, false, 2)
+CV_CUDEV_RGB2Lab_INST(LRGBA_to_Lab4, 4, 4, false, 2)
+CV_CUDEV_RGB2Lab_INST(LBGR_to_Lab, 3, 3, false, 0)
+CV_CUDEV_RGB2Lab_INST(LBGRA_to_Lab, 4, 3, false, 0)
+CV_CUDEV_RGB2Lab_INST(LBGR_to_Lab4, 3, 4, false, 0)
+CV_CUDEV_RGB2Lab_INST(LBGRA_to_Lab4, 4, 4, false, 0)
+
+#undef CV_CUDEV_RGB2Lab_INST
+
+// Lab to RGB
+
+#define CV_CUDEV_Lab2RGB_INST(name, scn, dcn, sRGB, blueIdx) \
+    template <typename SrcDepth> struct name ## _func : cv::cudev::color_cvt_detail::Lab2RGB<SrcDepth, scn, dcn, sRGB, blueIdx> \
+    { \
+    };
+
+CV_CUDEV_Lab2RGB_INST(Lab_to_RGB, 3, 3, true, 2)
+CV_CUDEV_Lab2RGB_INST(Lab4_to_RGB, 4, 3, true, 2)
+CV_CUDEV_Lab2RGB_INST(Lab_to_RGBA, 3, 4, true, 2)
+CV_CUDEV_Lab2RGB_INST(Lab4_to_RGBA, 4, 4, true, 2)
+CV_CUDEV_Lab2RGB_INST(Lab_to_BGR, 3, 3, true, 0)
+CV_CUDEV_Lab2RGB_INST(Lab4_to_BGR, 4, 3, true, 0)
+CV_CUDEV_Lab2RGB_INST(Lab_to_BGRA, 3, 4, true, 0)
+CV_CUDEV_Lab2RGB_INST(Lab4_to_BGRA, 4, 4, true, 0)
+
+CV_CUDEV_Lab2RGB_INST(Lab_to_LRGB, 3, 3, false, 2)
+CV_CUDEV_Lab2RGB_INST(Lab4_to_LRGB, 4, 3, false, 2)
+CV_CUDEV_Lab2RGB_INST(Lab_to_LRGBA, 3, 4, false, 2)
+CV_CUDEV_Lab2RGB_INST(Lab4_to_LRGBA, 4, 4, false, 2)
+CV_CUDEV_Lab2RGB_INST(Lab_to_LBGR, 3, 3, false, 0)
+CV_CUDEV_Lab2RGB_INST(Lab4_to_LBGR, 4, 3, false, 0)
+CV_CUDEV_Lab2RGB_INST(Lab_to_LBGRA, 3, 4, false, 0)
+CV_CUDEV_Lab2RGB_INST(Lab4_to_LBGRA, 4, 4, false, 0)
+
+#undef CV_CUDEV_Lab2RGB_INST
+
+// RGB to Luv
+
+#define CV_CUDEV_RGB2Luv_INST(name, scn, dcn, sRGB, blueIdx) \
+    template <typename SrcDepth> struct name ## _func : cv::cudev::color_cvt_detail::RGB2Luv<SrcDepth, scn, dcn, sRGB, blueIdx> \
+    { \
+    };
+
+CV_CUDEV_RGB2Luv_INST(RGB_to_Luv, 3, 3, true, 2)
+CV_CUDEV_RGB2Luv_INST(RGBA_to_Luv, 4, 3, true, 2)
+CV_CUDEV_RGB2Luv_INST(RGB_to_Luv4, 3, 4, true, 2)
+CV_CUDEV_RGB2Luv_INST(RGBA_to_Luv4, 4, 4, true, 2)
+CV_CUDEV_RGB2Luv_INST(BGR_to_Luv, 3, 3, true, 0)
+CV_CUDEV_RGB2Luv_INST(BGRA_to_Luv, 4, 3, true, 0)
+CV_CUDEV_RGB2Luv_INST(BGR_to_Luv4, 3, 4, true, 0)
+CV_CUDEV_RGB2Luv_INST(BGRA_to_Luv4, 4, 4, true, 0)
+
+CV_CUDEV_RGB2Luv_INST(LRGB_to_Luv, 3, 3, false, 2)
+CV_CUDEV_RGB2Luv_INST(LRGBA_to_Luv, 4, 3, false, 2)
+CV_CUDEV_RGB2Luv_INST(LRGB_to_Luv4, 3, 4, false, 2)
+CV_CUDEV_RGB2Luv_INST(LRGBA_to_Luv4, 4, 4, false, 2)
+CV_CUDEV_RGB2Luv_INST(LBGR_to_Luv, 3, 3, false, 0)
+CV_CUDEV_RGB2Luv_INST(LBGRA_to_Luv, 4, 3, false, 0)
+CV_CUDEV_RGB2Luv_INST(LBGR_to_Luv4, 3, 4, false, 0)
+CV_CUDEV_RGB2Luv_INST(LBGRA_to_Luv4, 4, 4, false, 0)
+
+#undef CV_CUDEV_RGB2Luv_INST
+
+// Luv to RGB
+
+#define CV_CUDEV_Luv2RGB_INST(name, scn, dcn, sRGB, blueIdx) \
+    template <typename SrcDepth> struct name ## _func : cv::cudev::color_cvt_detail::Luv2RGB<SrcDepth, scn, dcn, sRGB, blueIdx> \
+    { \
+    };
+
+CV_CUDEV_Luv2RGB_INST(Luv_to_RGB, 3, 3, true, 2)
+CV_CUDEV_Luv2RGB_INST(Luv4_to_RGB, 4, 3, true, 2)
+CV_CUDEV_Luv2RGB_INST(Luv_to_RGBA, 3, 4, true, 2)
+CV_CUDEV_Luv2RGB_INST(Luv4_to_RGBA, 4, 4, true, 2)
+CV_CUDEV_Luv2RGB_INST(Luv_to_BGR, 3, 3, true, 0)
+CV_CUDEV_Luv2RGB_INST(Luv4_to_BGR, 4, 3, true, 0)
+CV_CUDEV_Luv2RGB_INST(Luv_to_BGRA, 3, 4, true, 0)
+CV_CUDEV_Luv2RGB_INST(Luv4_to_BGRA, 4, 4, true, 0)
+
+CV_CUDEV_Luv2RGB_INST(Luv_to_LRGB, 3, 3, false, 2)
+CV_CUDEV_Luv2RGB_INST(Luv4_to_LRGB, 4, 3, false, 2)
+CV_CUDEV_Luv2RGB_INST(Luv_to_LRGBA, 3, 4, false, 2)
+CV_CUDEV_Luv2RGB_INST(Luv4_to_LRGBA, 4, 4, false, 2)
+CV_CUDEV_Luv2RGB_INST(Luv_to_LBGR, 3, 3, false, 0)
+CV_CUDEV_Luv2RGB_INST(Luv4_to_LBGR, 4, 3, false, 0)
+CV_CUDEV_Luv2RGB_INST(Luv_to_LBGRA, 3, 4, false, 0)
+CV_CUDEV_Luv2RGB_INST(Luv4_to_LBGRA, 4, 4, false, 0)
+
+#undef CV_CUDEV_Luv2RGB_INST
+
+// 24/32-bit RGB to 16-bit (565 or 555) RGB
+
+#define CV_CUDEV_RGB2RGB5x5_INST(name, scn, bidx, green_bits) \
+    typedef cv::cudev::color_cvt_detail::RGB2RGB5x5<scn, bidx, green_bits> name ## _func;
+
+CV_CUDEV_RGB2RGB5x5_INST(BGR_to_BGR555, 3, 0, 5)
+CV_CUDEV_RGB2RGB5x5_INST(BGR_to_BGR565, 3, 0, 6)
+CV_CUDEV_RGB2RGB5x5_INST(RGB_to_BGR555, 3, 2, 5)
+CV_CUDEV_RGB2RGB5x5_INST(RGB_to_BGR565, 3, 2, 6)
+CV_CUDEV_RGB2RGB5x5_INST(BGRA_to_BGR555, 4, 0, 5)
+CV_CUDEV_RGB2RGB5x5_INST(BGRA_to_BGR565, 4, 0, 6)
+CV_CUDEV_RGB2RGB5x5_INST(RGBA_to_BGR555, 4, 2, 5)
+CV_CUDEV_RGB2RGB5x5_INST(RGBA_to_BGR565, 4, 2, 6)
+
+#undef CV_CUDEV_RGB2RGB5x5_INST
+
+// 16-bit (565 or 555) RGB to 24/32-bit RGB
+
+#define CV_CUDEV_RGB5x52RGB_INST(name, dcn, bidx, green_bits) \
+    typedef cv::cudev::color_cvt_detail::RGB5x52RGB<dcn, bidx, green_bits> name ## _func;
+
+CV_CUDEV_RGB5x52RGB_INST(BGR555_to_RGB, 3, 2, 5)
+CV_CUDEV_RGB5x52RGB_INST(BGR565_to_RGB, 3, 2, 6)
+CV_CUDEV_RGB5x52RGB_INST(BGR555_to_BGR, 3, 0, 5)
+CV_CUDEV_RGB5x52RGB_INST(BGR565_to_BGR, 3, 0, 6)
+CV_CUDEV_RGB5x52RGB_INST(BGR555_to_RGBA, 4, 2, 5)
+CV_CUDEV_RGB5x52RGB_INST(BGR565_to_RGBA, 4, 2, 6)
+CV_CUDEV_RGB5x52RGB_INST(BGR555_to_BGRA, 4, 0, 5)
+CV_CUDEV_RGB5x52RGB_INST(BGR565_to_BGRA, 4, 0, 6)
+
+#undef CV_CUDEV_RGB5x52RGB_INST
+
+// Grayscale to 16-bit (565 or 555) RGB
+
+#define CV_CUDEV_GRAY2RGB5x5_INST(name, green_bits) \
+    typedef cv::cudev::color_cvt_detail::Gray2RGB5x5<green_bits> name ## _func;
+
+CV_CUDEV_GRAY2RGB5x5_INST(GRAY_to_BGR555, 5)
+CV_CUDEV_GRAY2RGB5x5_INST(GRAY_to_BGR565, 6)
+
+#undef CV_CUDEV_GRAY2RGB5x5_INST
+
+// 16-bit (565 or 555) RGB to Grayscale
+
+#define CV_CUDEV_RGB5x52GRAY_INST(name, green_bits) \
+    typedef cv::cudev::color_cvt_detail::RGB5x52Gray<green_bits> name ## _func;
+
+CV_CUDEV_RGB5x52GRAY_INST(BGR555_to_GRAY, 5)
+CV_CUDEV_RGB5x52GRAY_INST(BGR565_to_GRAY, 6)
+
+#undef CV_CUDEV_RGB5x52GRAY_INST
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/functional/functional.hpp

@@ -0,0 +1,854 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_FUNCTIONAL_FUNCTIONAL_HPP__
+#define __OPENCV_CUDEV_FUNCTIONAL_FUNCTIONAL_HPP__
+
+#include "../common.hpp"
+#include "../util/saturate_cast.hpp"
+#include "../util/vec_traits.hpp"
+#include "../util/vec_math.hpp"
+#include "../util/type_traits.hpp"
+
+namespace cv { namespace cudev {
+
+// Function Objects
+
+template <typename _Arg, typename _Result> struct unary_function
+{
+    typedef _Arg    argument_type;
+    typedef _Result result_type;
+};
+
+template <typename _Arg1, typename _Arg2, typename _Result> struct binary_function
+{
+    typedef _Arg1   first_argument_type;
+    typedef _Arg2   second_argument_type;
+    typedef _Result result_type;
+};
+
+// Arithmetic Operations
+
+template <typename T> struct plus : binary_function<T, T, T>
+{
+    __device__ __forceinline__ T operator ()(typename TypeTraits<T>::parameter_type a,
+                                             typename TypeTraits<T>::parameter_type b) const
+    {
+        return saturate_cast<T>(a + b);
+    }
+};
+
+template <typename T> struct minus : binary_function<T, T, T>
+{
+    __device__ __forceinline__ T operator ()(typename TypeTraits<T>::parameter_type a,
+                                             typename TypeTraits<T>::parameter_type b) const
+    {
+        return saturate_cast<T>(a - b);
+    }
+};
+
+template <typename T> struct multiplies : binary_function<T, T, T>
+{
+    __device__ __forceinline__ T operator ()(typename TypeTraits<T>::parameter_type a,
+                                             typename TypeTraits<T>::parameter_type b) const
+    {
+        return saturate_cast<T>(a * b);
+    }
+};
+
+template <typename T> struct divides : binary_function<T, T, T>
+{
+    __device__ __forceinline__ T operator ()(typename TypeTraits<T>::parameter_type a,
+                                             typename TypeTraits<T>::parameter_type b) const
+    {
+        return saturate_cast<T>(a / b);
+    }
+};
+
+template <typename T> struct modulus : binary_function<T, T, T>
+{
+    __device__ __forceinline__ T operator ()(typename TypeTraits<T>::parameter_type a,
+                                             typename TypeTraits<T>::parameter_type b) const
+    {
+        return saturate_cast<T>(a % b);
+    }
+};
+
+template <typename T> struct negate : unary_function<T, T>
+{
+    __device__ __forceinline__ T operator ()(typename TypeTraits<T>::parameter_type a) const
+    {
+        return saturate_cast<T>(-a);
+    }
+};
+
+// Comparison Operations
+
+template <typename T> struct equal_to : binary_function<T, T, typename MakeVec<uchar, VecTraits<T>::cn>::type>
+{
+    __device__ __forceinline__ typename MakeVec<uchar, VecTraits<T>::cn>::type
+                                    operator ()(typename TypeTraits<T>::parameter_type a,
+                                                typename TypeTraits<T>::parameter_type b) const
+    {
+        return a == b;
+    }
+};
+
+template <typename T> struct not_equal_to : binary_function<T, T, typename MakeVec<uchar, VecTraits<T>::cn>::type>
+{
+    __device__ __forceinline__ typename MakeVec<uchar, VecTraits<T>::cn>::type
+                                    operator ()(typename TypeTraits<T>::parameter_type a,
+                                                typename TypeTraits<T>::parameter_type b) const
+    {
+        return a != b;
+    }
+};
+
+template <typename T> struct greater : binary_function<T, T, typename MakeVec<uchar, VecTraits<T>::cn>::type>
+{
+    __device__ __forceinline__ typename MakeVec<uchar, VecTraits<T>::cn>::type
+                                    operator ()(typename TypeTraits<T>::parameter_type a,
+                                                typename TypeTraits<T>::parameter_type b) const
+    {
+        return a > b;
+    }
+};
+
+template <typename T> struct less : binary_function<T, T, typename MakeVec<uchar, VecTraits<T>::cn>::type>
+{
+    __device__ __forceinline__ typename MakeVec<uchar, VecTraits<T>::cn>::type
+                                    operator ()(typename TypeTraits<T>::parameter_type a,
+                                                typename TypeTraits<T>::parameter_type b) const
+    {
+        return a < b;
+    }
+};
+
+template <typename T> struct greater_equal : binary_function<T, T, typename MakeVec<uchar, VecTraits<T>::cn>::type>
+{
+    __device__ __forceinline__ typename MakeVec<uchar, VecTraits<T>::cn>::type
+                                    operator ()(typename TypeTraits<T>::parameter_type a,
+                                                typename TypeTraits<T>::parameter_type b) const
+    {
+        return a >= b;
+    }
+};
+
+template <typename T> struct less_equal : binary_function<T, T, typename MakeVec<uchar, VecTraits<T>::cn>::type>
+{
+    __device__ __forceinline__ typename MakeVec<uchar, VecTraits<T>::cn>::type
+                                    operator ()(typename TypeTraits<T>::parameter_type a,
+                                                typename TypeTraits<T>::parameter_type b) const
+    {
+        return a <= b;
+    }
+};
+
+// Logical Operations
+
+template <typename T> struct logical_and : binary_function<T, T, typename MakeVec<uchar, VecTraits<T>::cn>::type>
+{
+    __device__ __forceinline__ typename MakeVec<uchar, VecTraits<T>::cn>::type
+                                    operator ()(typename TypeTraits<T>::parameter_type a,
+                                                typename TypeTraits<T>::parameter_type b) const
+    {
+        return a && b;
+    }
+};
+
+template <typename T> struct logical_or : binary_function<T, T, typename MakeVec<uchar, VecTraits<T>::cn>::type>
+{
+    __device__ __forceinline__ typename MakeVec<uchar, VecTraits<T>::cn>::type
+                                    operator ()(typename TypeTraits<T>::parameter_type a,
+                                                typename TypeTraits<T>::parameter_type b) const
+    {
+        return a || b;
+    }
+};
+
+template <typename T> struct logical_not : unary_function<T, typename MakeVec<uchar, VecTraits<T>::cn>::type>
+{
+    __device__ __forceinline__ typename MakeVec<uchar, VecTraits<T>::cn>::type
+                                    operator ()(typename TypeTraits<T>::parameter_type a) const
+    {
+        return !a;
+    }
+};
+
+// Bitwise Operations
+
+template <typename T> struct bit_and : binary_function<T, T, T>
+{
+    __device__ __forceinline__ T operator ()(typename TypeTraits<T>::parameter_type a,
+                                             typename TypeTraits<T>::parameter_type b) const
+    {
+        return a & b;
+    }
+};
+
+template <typename T> struct bit_or : binary_function<T, T, T>
+{
+    __device__ __forceinline__ T operator ()(typename TypeTraits<T>::parameter_type a,
+                                             typename TypeTraits<T>::parameter_type b) const
+    {
+        return a | b;
+    }
+};
+
+template <typename T> struct bit_xor : binary_function<T, T, T>
+{
+    __device__ __forceinline__ T operator ()(typename TypeTraits<T>::parameter_type a,
+                                             typename TypeTraits<T>::parameter_type b) const
+    {
+        return a ^ b;
+    }
+};
+
+template <typename T> struct bit_not : unary_function<T, T>
+{
+    __device__ __forceinline__ T operator ()(typename TypeTraits<T>::parameter_type v) const
+    {
+        return ~v;
+    }
+};
+
+template <typename T> struct bit_lshift : binary_function<T, T, T>
+{
+    __device__ __forceinline__ T operator ()(typename TypeTraits<T>::parameter_type a,
+                                             typename TypeTraits<T>::parameter_type b) const
+    {
+        return a << b;
+    }
+};
+
+template <typename T> struct bit_rshift : binary_function<T, T, T>
+{
+    __device__ __forceinline__ T operator ()(typename TypeTraits<T>::parameter_type a,
+                                             typename TypeTraits<T>::parameter_type b) const
+    {
+        return a >> b;
+    }
+};
+
+// Generalized Identity Operations
+
+template <typename T> struct identity : unary_function<T, T>
+{
+    __device__ __forceinline__ T operator ()(typename TypeTraits<T>::parameter_type x) const
+    {
+        return x;
+    }
+};
+
+template <typename T1, typename T2> struct project1st : binary_function<T1, T2, T1>
+{
+    __device__ __forceinline__ T1
+                    operator ()(typename TypeTraits<T1>::parameter_type lhs,
+                                typename TypeTraits<T2>::parameter_type) const
+    {
+        return lhs;
+    }
+};
+
+template <typename T1, typename T2> struct project2nd : binary_function<T1, T2, T2>
+{
+    __device__ __forceinline__ T2
+                    operator ()(typename TypeTraits<T1>::parameter_type,
+                                typename TypeTraits<T2>::parameter_type rhs) const
+    {
+        return rhs;
+    }
+};
+
+// Min/Max Operations
+
+template <typename T> struct maximum : binary_function<T, T, T>
+{
+    __device__ __forceinline__ T operator ()(typename TypeTraits<T>::parameter_type a,
+                                             typename TypeTraits<T>::parameter_type b) const
+    {
+        return max(a, b);
+    }
+};
+
+template <typename T> struct minimum : binary_function<T, T, T>
+{
+    __device__ __forceinline__ T operator ()(typename TypeTraits<T>::parameter_type a,
+                                             typename TypeTraits<T>::parameter_type b) const
+    {
+        return min(a, b);
+    }
+};
+
+#define CV_CUDEV_MINMAX_INST(type, maxop, minop) \
+    template <> struct maximum<type> : binary_function<type, type, type> \
+    { \
+        __device__ __forceinline__ type operator ()(type a, type b) const {return maxop(a, b);} \
+    }; \
+    template <> struct minimum<type> : binary_function<type, type, type> \
+    { \
+        __device__ __forceinline__ type operator ()(type a, type b) const {return minop(a, b);} \
+    };
+
+
+CV_CUDEV_MINMAX_INST(uchar, ::max, ::min)
+CV_CUDEV_MINMAX_INST(schar, ::max, ::min)
+CV_CUDEV_MINMAX_INST(ushort, ::max, ::min)
+CV_CUDEV_MINMAX_INST(short, ::max, ::min)
+CV_CUDEV_MINMAX_INST(int, ::max, ::min)
+CV_CUDEV_MINMAX_INST(uint, ::max, ::min)
+CV_CUDEV_MINMAX_INST(float, ::fmaxf, ::fminf)
+CV_CUDEV_MINMAX_INST(double, ::fmax, ::fmin)
+
+#undef CV_CUDEV_MINMAX_INST
+
+// abs_func
+
+template <typename T> struct abs_func : unary_function<T, T>
+{
+    __device__ __forceinline__ T operator ()(typename TypeTraits<T>::parameter_type x) const
+    {
+        return abs(x);
+    }
+};
+
+template <> struct abs_func<uchar> : unary_function<uchar, uchar>
+{
+    __device__ __forceinline__ uchar operator ()(uchar x) const
+    {
+        return x;
+    }
+};
+
+template <> struct abs_func<schar> : unary_function<schar, schar>
+{
+    __device__ __forceinline__ schar operator ()(schar x) const
+    {
+        return ::abs((int) x);
+    }
+};
+
+template <> struct abs_func<ushort> : unary_function<ushort, ushort>
+{
+    __device__ __forceinline__ ushort operator ()(ushort x) const
+    {
+        return x;
+    }
+};
+
+template <> struct abs_func<short> : unary_function<short, short>
+{
+    __device__ __forceinline__ short operator ()(short x) const
+    {
+        return ::abs((int) x);
+    }
+};
+
+template <> struct abs_func<uint> : unary_function<uint, uint>
+{
+    __device__ __forceinline__ uint operator ()(uint x) const
+    {
+        return x;
+    }
+};
+
+template <> struct abs_func<int> : unary_function<int, int>
+{
+    __device__ __forceinline__ int operator ()(int x) const
+    {
+        return ::abs(x);
+    }
+};
+
+template <> struct abs_func<float> : unary_function<float, float>
+{
+    __device__ __forceinline__ float operator ()(float x) const
+    {
+        return ::fabsf(x);
+    }
+};
+
+template <> struct abs_func<double> : unary_function<double, double>
+{
+    __device__ __forceinline__ double operator ()(double x) const
+    {
+        return ::fabs(x);
+    }
+};
+
+// absdiff_func
+
+template <typename T> struct absdiff_func : binary_function<T, T, T>
+{
+    __device__ __forceinline__ T operator ()(typename TypeTraits<T>::parameter_type a, typename TypeTraits<T>::parameter_type b) const
+    {
+        abs_func<T> f;
+        return f(a - b);
+    }
+};
+
+// Math functions
+
+template <typename T> struct sqr_func : unary_function<T, T>
+{
+    __device__ __forceinline__ T operator ()(typename TypeTraits<T>::parameter_type x) const
+    {
+        return x * x;
+    }
+};
+
+namespace functional_detail
+{
+    template <typename T> struct FloatType
+    {
+        typedef typename MakeVec<
+            typename LargerType<float, typename VecTraits<T>::elem_type>::type,
+            VecTraits<T>::cn
+        >::type type;
+    };
+}
+
+#define CV_CUDEV_UNARY_FUNCTION_INST(name, func) \
+    template <typename T> struct name ## _func : unary_function<T, typename functional_detail::FloatType<T>::type> \
+    { \
+        __device__ __forceinline__ typename functional_detail::FloatType<T>::type operator ()(typename TypeTraits<T>::parameter_type a) const \
+        { \
+            return name(a); \
+        } \
+    }; \
+    template <> struct name ## _func<uchar> : unary_function<uchar, float> \
+    { \
+        __device__ __forceinline__ float operator ()(uchar a) const \
+        { \
+            return func ## f(a); \
+        } \
+    }; \
+    template <> struct name ## _func<schar> : unary_function<schar, float> \
+    { \
+        __device__ __forceinline__ float operator ()(schar a) const \
+        { \
+            return func ## f(a); \
+        } \
+    }; \
+    template <> struct name ## _func<ushort> : unary_function<ushort, float> \
+    { \
+        __device__ __forceinline__ float operator ()(ushort a) const \
+        { \
+            return func ## f(a); \
+        } \
+    }; \
+    template <> struct name ## _func<short> : unary_function<short, float> \
+    { \
+        __device__ __forceinline__ float operator ()(short a) const \
+        { \
+            return func ## f(a); \
+        } \
+    }; \
+    template <> struct name ## _func<uint> : unary_function<uint, float> \
+    { \
+        __device__ __forceinline__ float operator ()(uint a) const \
+        { \
+            return func ## f(a); \
+        } \
+    }; \
+    template <> struct name ## _func<int> : unary_function<int, float> \
+    { \
+        __device__ __forceinline__ float operator ()(int a) const \
+        { \
+            return func ## f(a); \
+        } \
+    }; \
+    template <> struct name ## _func<float> : unary_function<float, float> \
+    { \
+        __device__ __forceinline__ float operator ()(float a) const \
+        { \
+            return func ## f(a); \
+        } \
+    }; \
+    template <> struct name ## _func<double> : unary_function<double, double> \
+    { \
+        __device__ __forceinline__ double operator ()(double a) const \
+        { \
+            return func(a); \
+        } \
+    };
+
+CV_CUDEV_UNARY_FUNCTION_INST(sqrt, ::sqrt)
+CV_CUDEV_UNARY_FUNCTION_INST(exp, ::exp)
+CV_CUDEV_UNARY_FUNCTION_INST(exp2, ::exp2)
+CV_CUDEV_UNARY_FUNCTION_INST(exp10, ::exp10)
+CV_CUDEV_UNARY_FUNCTION_INST(log, ::log)
+CV_CUDEV_UNARY_FUNCTION_INST(log2, ::log2)
+CV_CUDEV_UNARY_FUNCTION_INST(log10, ::log10)
+CV_CUDEV_UNARY_FUNCTION_INST(sin, ::sin)
+CV_CUDEV_UNARY_FUNCTION_INST(cos, ::cos)
+CV_CUDEV_UNARY_FUNCTION_INST(tan, ::tan)
+CV_CUDEV_UNARY_FUNCTION_INST(asin, ::asin)
+CV_CUDEV_UNARY_FUNCTION_INST(acos, ::acos)
+CV_CUDEV_UNARY_FUNCTION_INST(atan, ::atan)
+CV_CUDEV_UNARY_FUNCTION_INST(sinh, ::sinh)
+CV_CUDEV_UNARY_FUNCTION_INST(cosh, ::cosh)
+CV_CUDEV_UNARY_FUNCTION_INST(tanh, ::tanh)
+CV_CUDEV_UNARY_FUNCTION_INST(asinh, ::asinh)
+CV_CUDEV_UNARY_FUNCTION_INST(acosh, ::acosh)
+CV_CUDEV_UNARY_FUNCTION_INST(atanh, ::atanh)
+
+#undef CV_CUDEV_UNARY_FUNCTION_INST
+
+#define CV_CUDEV_BINARY_FUNCTION_INST(name, func) \
+    template <typename T> struct name ## _func : binary_function<T, T, typename functional_detail::FloatType<T>::type> \
+    { \
+        __device__ __forceinline__ typename functional_detail::FloatType<T>::type operator ()(typename TypeTraits<T>::parameter_type a, typename TypeTraits<T>::parameter_type b) const \
+        { \
+            return name(a, b); \
+        } \
+    }; \
+    template <> struct name ## _func<uchar> : binary_function<uchar, uchar, float> \
+    { \
+        __device__ __forceinline__ float operator ()(uchar a, uchar b) const \
+        { \
+            return func ## f(a, b); \
+        } \
+    }; \
+    template <> struct name ## _func<schar> : binary_function<schar, schar, float> \
+    { \
+        __device__ __forceinline__ float operator ()(schar a, schar b) const \
+        { \
+            return func ## f(a, b); \
+        } \
+    }; \
+    template <> struct name ## _func<ushort> : binary_function<ushort, ushort, float> \
+    { \
+        __device__ __forceinline__ float operator ()(ushort a, ushort b) const \
+        { \
+            return func ## f(a, b); \
+        } \
+    }; \
+    template <> struct name ## _func<short> : binary_function<short, short, float> \
+    { \
+        __device__ __forceinline__ float operator ()(short a, short b) const \
+        { \
+            return func ## f(a, b); \
+        } \
+    }; \
+    template <> struct name ## _func<uint> : binary_function<uint, uint, float> \
+    { \
+        __device__ __forceinline__ float operator ()(uint a, uint b) const \
+        { \
+            return func ## f(a, b); \
+        } \
+    }; \
+    template <> struct name ## _func<int> : binary_function<int, int, float> \
+    { \
+        __device__ __forceinline__ float operator ()(int a, int b) const \
+        { \
+            return func ## f(a, b); \
+        } \
+    }; \
+    template <> struct name ## _func<float> : binary_function<float, float, float> \
+    { \
+        __device__ __forceinline__ float operator ()(float a, float b) const \
+        { \
+            return func ## f(a, b); \
+        } \
+    }; \
+    template <> struct name ## _func<double> : binary_function<double, double, double> \
+    { \
+        __device__ __forceinline__ double operator ()(double a, double b) const \
+        { \
+            return func(a, b); \
+        } \
+    };
+
+CV_CUDEV_BINARY_FUNCTION_INST(hypot, ::hypot)
+CV_CUDEV_BINARY_FUNCTION_INST(atan2, ::atan2)
+
+#undef CV_CUDEV_BINARY_FUNCTION_INST
+
+template <typename T> struct magnitude_func : binary_function<T, T, typename functional_detail::FloatType<T>::type>
+{
+    __device__ __forceinline__ typename functional_detail::FloatType<T>::type operator ()(typename TypeTraits<T>::parameter_type a, typename TypeTraits<T>::parameter_type b) const
+    {
+        sqrt_func<typename functional_detail::FloatType<T>::type> f;
+        return f(a * a + b * b);
+    }
+};
+
+template <typename T> struct pow_func : binary_function<T, float, float>
+{
+    __device__ __forceinline__ float operator ()(T val, float power) const
+    {
+        return ::powf(val, power);
+    }
+};
+template <> struct pow_func<double> : binary_function<double, double, double>
+{
+    __device__ __forceinline__ double operator ()(double val, double power) const
+    {
+        return ::pow(val, power);
+    }
+};
+
+// Saturate Cast Functor
+
+template <typename T, typename D> struct saturate_cast_func : unary_function<T, D>
+{
+    __device__ __forceinline__ D operator ()(typename TypeTraits<T>::parameter_type v) const
+    {
+        return saturate_cast<D>(v);
+    }
+};
+
+// Threshold Functors
+
+template <typename T> struct ThreshBinaryFunc : unary_function<T, T>
+{
+    T thresh;
+    T maxVal;
+
+    __device__ __forceinline__ T operator ()(typename TypeTraits<T>::parameter_type src) const
+    {
+        return saturate_cast<T>(src > thresh) * maxVal;
+    }
+};
+
+template <typename T>
+__host__ __device__ ThreshBinaryFunc<T> thresh_binary_func(T thresh, T maxVal)
+{
+    ThreshBinaryFunc<T> f;
+    f.thresh = thresh;
+    f.maxVal = maxVal;
+    return f;
+}
+
+template <typename T> struct ThreshBinaryInvFunc : unary_function<T, T>
+{
+    T thresh;
+    T maxVal;
+
+    __device__ __forceinline__ T operator ()(typename TypeTraits<T>::parameter_type src) const
+    {
+        return saturate_cast<T>(src <= thresh) * maxVal;
+    }
+};
+
+template <typename T>
+__host__ __device__ ThreshBinaryInvFunc<T> thresh_binary_inv_func(T thresh, T maxVal)
+{
+    ThreshBinaryInvFunc<T> f;
+    f.thresh = thresh;
+    f.maxVal = maxVal;
+    return f;
+}
+
+template <typename T> struct ThreshTruncFunc : unary_function<T, T>
+{
+    T thresh;
+
+    __device__ __forceinline__ T operator ()(typename TypeTraits<T>::parameter_type src) const
+    {
+        minimum<T> minOp;
+        return minOp(src, thresh);
+    }
+};
+
+template <typename T>
+__host__ __device__ ThreshTruncFunc<T> thresh_trunc_func(T thresh)
+{
+    ThreshTruncFunc<T> f;
+    f.thresh = thresh;
+    return f;
+}
+
+template <typename T> struct ThreshToZeroFunc : unary_function<T, T>
+{
+    T thresh;
+
+    __device__ __forceinline__ T operator ()(typename TypeTraits<T>::parameter_type src) const
+    {
+        return saturate_cast<T>(src > thresh) * src;
+    }
+};
+
+template <typename T>
+__host__ __device__ ThreshToZeroFunc<T> thresh_to_zero_func(T thresh)
+{
+    ThreshToZeroFunc<T> f;
+    f.thresh = thresh;
+    return f;
+}
+
+template <typename T> struct ThreshToZeroInvFunc : unary_function<T, T>
+{
+    T thresh;
+
+    __device__ __forceinline__ T operator ()(typename TypeTraits<T>::parameter_type src) const
+    {
+        return saturate_cast<T>(src <= thresh) * src;
+    }
+};
+
+template <typename T>
+__host__ __device__ ThreshToZeroInvFunc<T> thresh_to_zero_inv_func(T thresh)
+{
+    ThreshToZeroInvFunc<T> f;
+    f.thresh = thresh;
+    return f;
+}
+
+// Function Object Adaptors
+
+template <class Predicate> struct UnaryNegate : unary_function<typename Predicate::argument_type, typename Predicate::result_type>
+{
+    Predicate pred;
+
+    __device__ __forceinline__ typename Predicate::result_type operator ()(
+            typename TypeTraits<typename Predicate::argument_type>::parameter_type x) const
+    {
+        return !pred(x);
+    }
+};
+
+template <class Predicate>
+__host__ __device__ UnaryNegate<Predicate> not1(const Predicate& pred)
+{
+    UnaryNegate<Predicate> n;
+    n.pred = pred;
+    return n;
+}
+
+template <class Predicate> struct BinaryNegate : binary_function<typename Predicate::first_argument_type, typename Predicate::second_argument_type, typename Predicate::result_type>
+{
+    Predicate pred;
+
+    __device__ __forceinline__ typename Predicate::result_type operator ()(
+            typename TypeTraits<typename Predicate::first_argument_type>::parameter_type x,
+            typename TypeTraits<typename Predicate::second_argument_type>::parameter_type y) const
+    {
+        return !pred(x, y);
+    }
+};
+
+template <class Predicate>
+__host__ __device__ BinaryNegate<Predicate> not2(const Predicate& pred)
+{
+    BinaryNegate<Predicate> n;
+    n.pred = pred;
+    return n;
+}
+
+template <class Op> struct Binder1st : unary_function<typename Op::second_argument_type, typename Op::result_type>
+{
+    Op op;
+    typename Op::first_argument_type arg1;
+
+    __device__ __forceinline__ typename Op::result_type operator ()(
+            typename TypeTraits<typename Op::second_argument_type>::parameter_type a) const
+    {
+        return op(arg1, a);
+    }
+};
+
+template <class Op>
+__host__ __device__ Binder1st<Op> bind1st(const Op& op, const typename Op::first_argument_type& arg1)
+{
+    Binder1st<Op> b;
+    b.op = op;
+    b.arg1 = arg1;
+    return b;
+}
+
+template <class Op> struct Binder2nd : unary_function<typename Op::first_argument_type, typename Op::result_type>
+{
+    Op op;
+    typename Op::second_argument_type arg2;
+
+    __device__ __forceinline__ typename Op::result_type operator ()(
+            typename TypeTraits<typename Op::first_argument_type>::parameter_type a) const
+    {
+        return op(a, arg2);
+    }
+};
+
+template <class Op>
+__host__ __device__ Binder2nd<Op> bind2nd(const Op& op, const typename Op::second_argument_type& arg2)
+{
+    Binder2nd<Op> b;
+    b.op = op;
+    b.arg2 = arg2;
+    return b;
+}
+
+// Functor Traits
+
+template <typename F> struct IsUnaryFunction
+{
+    typedef char Yes;
+    struct No {Yes a[2];};
+
+    template <typename T, typename D> static Yes check(unary_function<T, D>);
+    static No check(...);
+
+    static F makeF();
+
+    enum { value = (sizeof(check(makeF())) == sizeof(Yes)) };
+};
+
+template <typename F> struct IsBinaryFunction
+{
+    typedef char Yes;
+    struct No {Yes a[2];};
+
+    template <typename T1, typename T2, typename D> static Yes check(binary_function<T1, T2, D>);
+    static No check(...);
+
+    static F makeF();
+
+    enum { value = (sizeof(check(makeF())) == sizeof(Yes)) };
+};
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/functional/tuple_adapter.hpp

@@ -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.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_FUNCTIONAL_TUPLE_ADAPTER_HPP__
+#define __OPENCV_CUDEV_FUNCTIONAL_TUPLE_ADAPTER_HPP__
+
+#include "../common.hpp"
+#include "../util/tuple.hpp"
+
+namespace cv { namespace cudev {
+
+template <class Op, int n> struct UnaryTupleAdapter
+{
+    typedef typename Op::result_type result_type;
+
+    Op op;
+
+    template <class Tuple>
+    __device__ __forceinline__ typename Op::result_type operator ()(const Tuple& t) const
+    {
+        return op(get<n>(t));
+    }
+};
+
+template <int n, class Op>
+__host__ __device__ UnaryTupleAdapter<Op, n> unaryTupleAdapter(const Op& op)
+{
+    UnaryTupleAdapter<Op, n> a;
+    a.op = op;
+    return a;
+}
+
+template <class Op, int n0, int n1> struct BinaryTupleAdapter
+{
+    typedef typename Op::result_type result_type;
+
+    Op op;
+
+    template <class Tuple>
+    __device__ __forceinline__ typename Op::result_type operator ()(const Tuple& t) const
+    {
+        return op(get<n0>(t), get<n1>(t));
+    }
+};
+
+template <int n0, int n1, class Op>
+__host__ __device__ BinaryTupleAdapter<Op, n0, n1> binaryTupleAdapter(const Op& op)
+{
+    BinaryTupleAdapter<Op, n0, n1> a;
+    a.op = op;
+    return a;
+}
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/grid/copy.hpp

@@ -0,0 +1,263 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_GRID_COPY_HPP__
+#define __OPENCV_CUDEV_GRID_COPY_HPP__
+
+#include "../common.hpp"
+#include "../util/tuple.hpp"
+#include "../ptr2d/traits.hpp"
+#include "../ptr2d/gpumat.hpp"
+#include "../ptr2d/mask.hpp"
+#include "../ptr2d/zip.hpp"
+#include "detail/copy.hpp"
+
+namespace cv { namespace cudev {
+
+template <class Policy, class SrcPtr, typename DstType, class MaskPtr>
+__host__ void gridCopy_(const SrcPtr& src, GpuMat_<DstType>& dst, const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    CV_Assert( getRows(mask) == rows && getCols(mask) == cols );
+
+    dst.create(rows, cols);
+
+    grid_copy_detail::copy<Policy>(shrinkPtr(src), shrinkPtr(dst), shrinkPtr(mask), rows, cols, StreamAccessor::getStream(stream));
+}
+
+template <class Policy, class SrcPtr, typename DstType>
+__host__ void gridCopy_(const SrcPtr& src, GpuMat_<DstType>& dst, Stream& stream = Stream::Null())
+{
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    dst.create(rows, cols);
+
+    grid_copy_detail::copy<Policy>(shrinkPtr(src), shrinkPtr(dst), WithOutMask(), rows, cols, StreamAccessor::getStream(stream));
+}
+
+template <class Policy, class SrcPtrTuple, typename D0, typename D1, class MaskPtr>
+__host__ void gridCopy_(const SrcPtrTuple& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>& >& dst, const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    CV_StaticAssert( tuple_size<SrcPtrTuple>::value == 2, "" );
+
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    CV_Assert( getRows(mask) == rows && getCols(mask) == cols );
+
+    get<0>(dst).create(rows, cols);
+    get<1>(dst).create(rows, cols);
+
+    grid_copy_detail::copy_tuple<Policy>(shrinkPtr(src),
+                                         shrinkPtr(zipPtr(get<0>(dst), get<1>(dst))),
+                                         shrinkPtr(mask),
+                                         rows, cols,
+                                         StreamAccessor::getStream(stream));
+}
+
+template <class Policy, class SrcPtrTuple, typename D0, typename D1>
+__host__ void gridCopy_(const SrcPtrTuple& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>& >& dst, Stream& stream = Stream::Null())
+{
+    CV_StaticAssert( tuple_size<SrcPtrTuple>::value == 2, "" );
+
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    get<0>(dst).create(rows, cols);
+    get<1>(dst).create(rows, cols);
+
+    grid_copy_detail::copy_tuple<Policy>(shrinkPtr(src),
+                                         shrinkPtr(zipPtr(get<0>(dst), get<1>(dst))),
+                                         WithOutMask(),
+                                         rows, cols,
+                                         StreamAccessor::getStream(stream));
+}
+
+template <class Policy, class SrcPtrTuple, typename D0, typename D1, typename D2, class MaskPtr>
+__host__ void gridCopy_(const SrcPtrTuple& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>&, GpuMat_<D2>& >& dst, const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    CV_StaticAssert( tuple_size<SrcPtrTuple>::value == 3, "" );
+
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    CV_Assert( getRows(mask) == rows && getCols(mask) == cols );
+
+    get<0>(dst).create(rows, cols);
+    get<1>(dst).create(rows, cols);
+    get<2>(dst).create(rows, cols);
+
+    grid_copy_detail::copy_tuple<Policy>(shrinkPtr(src),
+                                         shrinkPtr(zipPtr(get<0>(dst), get<1>(dst), get<2>(dst))),
+                                         shrinkPtr(mask),
+                                         rows, cols,
+                                         StreamAccessor::getStream(stream));
+}
+
+template <class Policy, class SrcPtrTuple, typename D0, typename D1, typename D2>
+__host__ void gridCopy_(const SrcPtrTuple& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>&, GpuMat_<D2>& >& dst, Stream& stream = Stream::Null())
+{
+    CV_StaticAssert( tuple_size<SrcPtrTuple>::value == 3, "" );
+
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    get<0>(dst).create(rows, cols);
+    get<1>(dst).create(rows, cols);
+    get<2>(dst).create(rows, cols);
+
+    grid_copy_detail::copy_tuple<Policy>(shrinkPtr(src),
+                                         shrinkPtr(zipPtr(get<0>(dst), get<1>(dst), get<2>(dst))),
+                                         WithOutMask(),
+                                         rows, cols,
+                                         StreamAccessor::getStream(stream));
+}
+
+template <class Policy, class SrcPtrTuple, typename D0, typename D1, typename D2, typename D3, class MaskPtr>
+__host__ void gridCopy_(const SrcPtrTuple& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>&, GpuMat_<D2>&, GpuMat_<D3>& >& dst, const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    CV_StaticAssert( tuple_size<SrcPtrTuple>::value == 4, "" );
+
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    CV_Assert( getRows(mask) == rows && getCols(mask) == cols );
+
+    get<0>(dst).create(rows, cols);
+    get<1>(dst).create(rows, cols);
+    get<2>(dst).create(rows, cols);
+    get<3>(dst).create(rows, cols);
+
+    grid_copy_detail::copy_tuple<Policy>(shrinkPtr(src),
+                                         shrinkPtr(zipPtr(get<0>(dst), get<1>(dst), get<2>(dst), get<3>(dst))),
+                                         shrinkPtr(mask),
+                                         rows, cols,
+                                         StreamAccessor::getStream(stream));
+}
+
+template <class Policy, class SrcPtrTuple, typename D0, typename D1, typename D2, typename D3>
+__host__ void gridCopy_(const SrcPtrTuple& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>&, GpuMat_<D2>&, GpuMat_<D3>& >& dst, Stream& stream = Stream::Null())
+{
+    CV_StaticAssert( tuple_size<SrcPtrTuple>::value == 4, "" );
+
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    get<0>(dst).create(rows, cols);
+    get<1>(dst).create(rows, cols);
+    get<2>(dst).create(rows, cols);
+    get<3>(dst).create(rows, cols);
+
+    grid_copy_detail::copy_tuple<Policy>(shrinkPtr(src),
+                                         shrinkPtr(zipPtr(get<0>(dst), get<1>(dst), get<2>(dst), get<3>(dst))),
+                                         WithOutMask(),
+                                         rows, cols,
+                                         StreamAccessor::getStream(stream));
+}
+
+// Default Policy
+
+struct DefaultCopyPolicy
+{
+    enum {
+        block_size_x = 32,
+        block_size_y = 8
+    };
+};
+
+template <class SrcPtr, typename DstType, class MaskPtr>
+__host__ void gridCopy(const SrcPtr& src, GpuMat_<DstType>& dst, const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    gridCopy_<DefaultCopyPolicy>(src, dst, mask, stream);
+}
+
+template <class SrcPtr, typename DstType>
+__host__ void gridCopy(const SrcPtr& src, GpuMat_<DstType>& dst, Stream& stream = Stream::Null())
+{
+    gridCopy_<DefaultCopyPolicy>(src, dst, stream);
+}
+
+template <class SrcPtrTuple, typename D0, typename D1, class MaskPtr>
+__host__ void gridCopy(const SrcPtrTuple& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>& >& dst, const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    gridCopy_<DefaultCopyPolicy>(src, dst, mask, stream);
+}
+
+template <class SrcPtrTuple, typename D0, typename D1>
+__host__ void gridCopy(const SrcPtrTuple& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>& >& dst, Stream& stream = Stream::Null())
+{
+    gridCopy_<DefaultCopyPolicy>(src, dst, stream);
+}
+
+template <class SrcPtrTuple, typename D0, typename D1, typename D2, class MaskPtr>
+__host__ void gridCopy(const SrcPtrTuple& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>&, GpuMat_<D2>& >& dst, const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    gridCopy_<DefaultCopyPolicy>(src, dst, mask, stream);
+}
+
+template <class SrcPtrTuple, typename D0, typename D1, typename D2>
+__host__ void gridCopy(const SrcPtrTuple& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>&, GpuMat_<D2>& >& dst, Stream& stream = Stream::Null())
+{
+    gridCopy_<DefaultCopyPolicy>(src, dst, stream);
+}
+
+template <class SrcPtrTuple, typename D0, typename D1, typename D2, typename D3, class MaskPtr>
+__host__ void gridCopy(const SrcPtrTuple& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>&, GpuMat_<D2>&, GpuMat_<D3>& >& dst, const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    gridCopy_<DefaultCopyPolicy>(src, dst, mask, stream);
+}
+
+template <class SrcPtrTuple, typename D0, typename D1, typename D2, typename D3>
+__host__ void gridCopy_(const SrcPtrTuple& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>&, GpuMat_<D2>&, GpuMat_<D3>& >& dst, Stream& stream = Stream::Null())
+{
+    gridCopy_<DefaultCopyPolicy>(src, dst, stream);
+}
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/grid/detail/copy.hpp

@@ -0,0 +1,132 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_GRID_COPY_DETAIL_HPP__
+#define __OPENCV_CUDEV_GRID_COPY_DETAIL_HPP__
+
+#include "../../common.hpp"
+#include "../../util/tuple.hpp"
+#include "../../util/saturate_cast.hpp"
+#include "../../ptr2d/glob.hpp"
+#include "../../ptr2d/traits.hpp"
+
+namespace cv { namespace cudev {
+
+namespace grid_copy_detail
+{
+    template <class SrcPtr, typename DstType, class MaskPtr>
+    __global__ void copy(const SrcPtr src, GlobPtr<DstType> dst, const MaskPtr mask, const int rows, const int cols)
+    {
+        const int x = blockIdx.x * blockDim.x + threadIdx.x;
+        const int y = blockIdx.y * blockDim.y + threadIdx.y;
+
+        if (x >= cols || y >= rows || !mask(y, x))
+            return;
+
+        dst(y, x) = saturate_cast<DstType>(src(y, x));
+    }
+
+    template <class Policy, class SrcPtr, typename DstType, class MaskPtr>
+    __host__ void copy(const SrcPtr& src, const GlobPtr<DstType>& dst, const MaskPtr& mask, int rows, int cols, cudaStream_t stream)
+    {
+        const dim3 block(Policy::block_size_x, Policy::block_size_y);
+        const dim3 grid(divUp(cols, block.x), divUp(rows, block.y));
+
+        copy<<<grid, block, 0, stream>>>(src, dst, mask, rows, cols);
+        CV_CUDEV_SAFE_CALL( cudaGetLastError() );
+
+        if (stream == 0)
+            CV_CUDEV_SAFE_CALL( cudaDeviceSynchronize() );
+    }
+
+    template <int count> struct Unroll
+    {
+        template <class SrcPtrTuple, class DstPtrTuple>
+        __device__ static void copy(const SrcPtrTuple& src, DstPtrTuple& dst, const int y, const int x)
+        {
+            typedef typename tuple_element<count - 1, DstPtrTuple>::type dst_ptr_type;
+            typedef typename PtrTraits<dst_ptr_type>::value_type dst_type;
+
+            get<count - 1>(dst)(y, x) = saturate_cast<dst_type>(get<count - 1>(src)(y, x));
+            Unroll<count - 1>::copy(src, dst, y, x);
+        }
+    };
+    template <> struct Unroll<0>
+    {
+        template <class SrcPtrTuple, class DstPtrTuple>
+        __device__ __forceinline__ static void copy(const SrcPtrTuple&, DstPtrTuple&, const int, const int)
+        {
+        }
+    };
+
+    template <class SrcPtrTuple, class DstPtrTuple, class MaskPtr>
+    __global__ void copy_tuple(const SrcPtrTuple src, DstPtrTuple dst, const MaskPtr mask, const int rows, const int cols)
+    {
+        const int x = blockIdx.x * blockDim.x + threadIdx.x;
+        const int y = blockIdx.y * blockDim.y + threadIdx.y;
+
+        if (x >= cols || y >= rows || !mask(y, x))
+            return;
+
+        Unroll<tuple_size<SrcPtrTuple>::value>::copy(src, dst, y, x);
+    }
+
+    template <class Policy, class SrcPtrTuple, class DstPtrTuple, class MaskPtr>
+    __host__ void copy_tuple(const SrcPtrTuple& src, const DstPtrTuple& dst, const MaskPtr& mask, int rows, int cols, cudaStream_t stream)
+    {
+        const dim3 block(Policy::block_size_x, Policy::block_size_y);
+        const dim3 grid(divUp(cols, block.x), divUp(rows, block.y));
+
+        copy_tuple<<<grid, block, 0, stream>>>(src, dst, mask, rows, cols);
+        CV_CUDEV_SAFE_CALL( cudaGetLastError() );
+
+        if (stream == 0)
+            CV_CUDEV_SAFE_CALL( cudaDeviceSynchronize() );
+    }
+}
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/grid/detail/glob_reduce.hpp

@@ -0,0 +1,475 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_GRID_GLOB_REDUCE_DETAIL_HPP__
+#define __OPENCV_CUDEV_GRID_GLOB_REDUCE_DETAIL_HPP__
+
+#include "../../common.hpp"
+#include "../../util/tuple.hpp"
+#include "../../util/saturate_cast.hpp"
+#include "../../util/atomic.hpp"
+#include "../../util/vec_traits.hpp"
+#include "../../util/type_traits.hpp"
+#include "../../util/limits.hpp"
+#include "../../block/reduce.hpp"
+#include "../../functional/functional.hpp"
+#include "../../ptr2d/traits.hpp"
+
+namespace cv { namespace cudev {
+
+namespace grid_glob_reduce_detail
+{
+    // Unroll
+
+    template <int cn> struct Unroll;
+
+    template <> struct Unroll<1>
+    {
+        template <int BLOCK_SIZE, typename R>
+        __device__ __forceinline__ static volatile R* smem(R* ptr)
+        {
+            return ptr;
+        }
+
+        template <typename R>
+        __device__ __forceinline__ static R& res(R& val)
+        {
+            return val;
+        }
+
+        template <class Op>
+        __device__ __forceinline__ static const Op& op(const Op& aop)
+        {
+            return aop;
+        }
+    };
+
+    template <> struct Unroll<2>
+    {
+        template <int BLOCK_SIZE, typename R>
+        __device__ __forceinline__ static tuple<volatile R*, volatile R*> smem(R* ptr)
+        {
+            return smem_tuple(ptr, ptr + BLOCK_SIZE);
+        }
+
+        template <typename R>
+        __device__ __forceinline__ static tuple<typename VecTraits<R>::elem_type&, typename VecTraits<R>::elem_type&> res(R& val)
+        {
+            return tie(val.x, val.y);
+        }
+
+        template <class Op>
+        __device__ __forceinline__ static tuple<Op, Op> op(const Op& aop)
+        {
+            return make_tuple(aop, aop);
+        }
+    };
+
+    template <> struct Unroll<3>
+    {
+        template <int BLOCK_SIZE, typename R>
+        __device__ __forceinline__ static tuple<volatile R*, volatile R*, volatile R*> smem(R* ptr)
+        {
+            return smem_tuple(ptr, ptr + BLOCK_SIZE, ptr + 2 * BLOCK_SIZE);
+        }
+
+        template <typename R>
+        __device__ __forceinline__ static tuple<typename VecTraits<R>::elem_type&,
+                                                typename VecTraits<R>::elem_type&,
+                                                typename VecTraits<R>::elem_type&> res(R& val)
+        {
+            return tie(val.x, val.y, val.z);
+        }
+
+        template <class Op>
+        __device__ __forceinline__ static tuple<Op, Op, Op> op(const Op& aop)
+        {
+            return make_tuple(aop, aop, aop);
+        }
+    };
+
+    template <> struct Unroll<4>
+    {
+        template <int BLOCK_SIZE, typename R>
+        __device__ __forceinline__ static tuple<volatile R*, volatile R*, volatile R*, volatile R*> smem(R* ptr)
+        {
+            return smem_tuple(ptr, ptr + BLOCK_SIZE, ptr + 2 * BLOCK_SIZE, ptr + 3 * BLOCK_SIZE);
+        }
+
+        template <typename R>
+        __device__ __forceinline__ static tuple<typename VecTraits<R>::elem_type&,
+                                                typename VecTraits<R>::elem_type&,
+                                                typename VecTraits<R>::elem_type&,
+                                                typename VecTraits<R>::elem_type&> res(R& val)
+        {
+            return tie(val.x, val.y, val.z, val.w);
+        }
+
+        template <class Op>
+        __device__ __forceinline__ static tuple<Op, Op, Op, Op> op(const Op& aop)
+        {
+            return make_tuple(aop, aop, aop, aop);
+        }
+    };
+
+    // AtomicUnroll
+
+    template <typename R, int cn> struct AtomicUnroll;
+
+    template <typename R> struct AtomicUnroll<R, 1>
+    {
+        __device__ __forceinline__ static void add(R* ptr, R val)
+        {
+            atomicAdd(ptr, val);
+        }
+
+        __device__ __forceinline__ static void min(R* ptr, R val)
+        {
+            atomicMin(ptr, val);
+        }
+
+        __device__ __forceinline__ static void max(R* ptr, R val)
+        {
+            atomicMax(ptr, val);
+        }
+    };
+
+    template <typename R> struct AtomicUnroll<R, 2>
+    {
+        typedef typename MakeVec<R, 2>::type val_type;
+
+        __device__ __forceinline__ static void add(R* ptr, val_type val)
+        {
+            atomicAdd(ptr, val.x);
+            atomicAdd(ptr + 1, val.y);
+        }
+
+        __device__ __forceinline__ static void min(R* ptr, val_type val)
+        {
+            atomicMin(ptr, val.x);
+            atomicMin(ptr + 1, val.y);
+        }
+
+        __device__ __forceinline__ static void max(R* ptr, val_type val)
+        {
+            atomicMax(ptr, val.x);
+            atomicMax(ptr + 1, val.y);
+        }
+    };
+
+    template <typename R> struct AtomicUnroll<R, 3>
+    {
+        typedef typename MakeVec<R, 3>::type val_type;
+
+        __device__ __forceinline__ static void add(R* ptr, val_type val)
+        {
+            atomicAdd(ptr, val.x);
+            atomicAdd(ptr + 1, val.y);
+            atomicAdd(ptr + 2, val.z);
+        }
+
+        __device__ __forceinline__ static void min(R* ptr, val_type val)
+        {
+            atomicMin(ptr, val.x);
+            atomicMin(ptr + 1, val.y);
+            atomicMin(ptr + 2, val.z);
+        }
+
+        __device__ __forceinline__ static void max(R* ptr, val_type val)
+        {
+            atomicMax(ptr, val.x);
+            atomicMax(ptr + 1, val.y);
+            atomicMax(ptr + 2, val.z);
+        }
+    };
+
+    template <typename R> struct AtomicUnroll<R, 4>
+    {
+        typedef typename MakeVec<R, 4>::type val_type;
+
+        __device__ __forceinline__ static void add(R* ptr, val_type val)
+        {
+            atomicAdd(ptr, val.x);
+            atomicAdd(ptr + 1, val.y);
+            atomicAdd(ptr + 2, val.z);
+            atomicAdd(ptr + 3, val.w);
+        }
+
+        __device__ __forceinline__ static void min(R* ptr, val_type val)
+        {
+            atomicMin(ptr, val.x);
+            atomicMin(ptr + 1, val.y);
+            atomicMin(ptr + 2, val.z);
+            atomicMin(ptr + 3, val.w);
+        }
+
+        __device__ __forceinline__ static void max(R* ptr, val_type val)
+        {
+            atomicMax(ptr, val.x);
+            atomicMax(ptr + 1, val.y);
+            atomicMax(ptr + 2, val.z);
+            atomicMax(ptr + 3, val.w);
+        }
+    };
+
+    // SumReductor
+
+    template <typename src_type, typename work_type> struct SumReductor
+    {
+        typedef typename VecTraits<work_type>::elem_type work_elem_type;
+        enum { cn = VecTraits<src_type>::cn };
+
+        work_type sum;
+
+        __device__ __forceinline__ SumReductor()
+        {
+            sum = VecTraits<work_type>::all(0);
+        }
+
+        __device__ __forceinline__ void reduceVal(typename TypeTraits<src_type>::parameter_type srcVal)
+        {
+            sum = sum + saturate_cast<work_type>(srcVal);
+        }
+
+        template <int BLOCK_SIZE>
+        __device__ void reduceGrid(work_elem_type* result, int tid)
+        {
+            __shared__ work_elem_type smem[BLOCK_SIZE * cn];
+
+            blockReduce<BLOCK_SIZE>(Unroll<cn>::template smem<BLOCK_SIZE>(smem), Unroll<cn>::res(sum), tid, Unroll<cn>::op(plus<work_elem_type>()));
+
+            if (tid == 0)
+                AtomicUnroll<work_elem_type, cn>::add(result, sum);
+        }
+    };
+
+    // MinMaxReductor
+
+    template <typename T> struct minop : minimum<T>
+    {
+        __device__ __forceinline__ static T initial()
+        {
+            return numeric_limits<T>::max();
+        }
+
+        __device__ __forceinline__ static void atomic(T* result, T myval)
+        {
+            atomicMin(result, myval);
+        }
+    };
+
+    template <typename T> struct maxop : maximum<T>
+    {
+        __device__ __forceinline__ static T initial()
+        {
+            return -numeric_limits<T>::max();
+        }
+
+        __device__ __forceinline__ static void atomic(T* result, T myval)
+        {
+            atomicMax(result, myval);
+        }
+    };
+
+    struct both
+    {
+    };
+
+    template <class Op, typename src_type, typename work_type> struct MinMaxReductor
+    {
+        work_type myval;
+
+        __device__ __forceinline__ MinMaxReductor()
+        {
+            myval = Op::initial();
+        }
+
+        __device__ __forceinline__ void reduceVal(typename TypeTraits<src_type>::parameter_type srcVal)
+        {
+            Op op;
+
+            myval = op(myval, srcVal);
+        }
+
+        template <int BLOCK_SIZE>
+        __device__ void reduceGrid(work_type* result, int tid)
+        {
+            __shared__ work_type smem[BLOCK_SIZE];
+
+            Op op;
+
+            blockReduce<BLOCK_SIZE>(smem, myval, tid, op);
+
+            if (tid == 0)
+                Op::atomic(result, myval);
+        }
+    };
+
+    template <typename src_type, typename work_type> struct MinMaxReductor<both, src_type, work_type>
+    {
+        work_type mymin;
+        work_type mymax;
+
+        __device__ __forceinline__ MinMaxReductor()
+        {
+            mymin = numeric_limits<work_type>::max();
+            mymax = -numeric_limits<work_type>::max();
+        }
+
+        __device__ __forceinline__ void reduceVal(typename TypeTraits<src_type>::parameter_type srcVal)
+        {
+            minimum<work_type> minOp;
+            maximum<work_type> maxOp;
+
+            mymin = minOp(mymin, srcVal);
+            mymax = maxOp(mymax, srcVal);
+        }
+
+        template <int BLOCK_SIZE>
+        __device__ void reduceGrid(work_type* result, int tid)
+        {
+            __shared__ work_type sminval[BLOCK_SIZE];
+            __shared__ work_type smaxval[BLOCK_SIZE];
+
+            minimum<work_type> minOp;
+            maximum<work_type> maxOp;
+
+            blockReduce<BLOCK_SIZE>(smem_tuple(sminval, smaxval), tie(mymin, mymax), tid, make_tuple(minOp, maxOp));
+
+            if (tid == 0)
+            {
+                atomicMin(result, mymin);
+                atomicMax(result + 1, mymax);
+            }
+        }
+    };
+
+    // glob_reduce
+
+    template <class Reductor, int BLOCK_SIZE, int PATCH_X, int PATCH_Y, class SrcPtr, typename ResType, class MaskPtr>
+    __global__ void glob_reduce(const SrcPtr src, ResType* result, const MaskPtr mask, const int rows, const int cols)
+    {
+        const int x0 = blockIdx.x * blockDim.x * PATCH_X + threadIdx.x;
+        const int y0 = blockIdx.y * blockDim.y * PATCH_Y + threadIdx.y;
+
+        Reductor reductor;
+
+        for (int i = 0, y = y0; i < PATCH_Y && y < rows; ++i, y += blockDim.y)
+        {
+            for (int j = 0, x = x0; j < PATCH_X && x < cols; ++j, x += blockDim.x)
+            {
+                if (mask(y, x))
+                {
+                    reductor.reduceVal(src(y, x));
+                }
+            }
+        }
+
+        const int tid = threadIdx.y * blockDim.x + threadIdx.x;
+
+        reductor.template reduceGrid<BLOCK_SIZE>(result, tid);
+    }
+
+    template <class Reductor, class Policy, class SrcPtr, typename ResType, class MaskPtr>
+    __host__ void glob_reduce(const SrcPtr& src, ResType* result, const MaskPtr& mask, int rows, int cols, cudaStream_t stream)
+    {
+        const dim3 block(Policy::block_size_x, Policy::block_size_y);
+        const dim3 grid(divUp(cols, block.x * Policy::patch_size_x), divUp(rows, block.y * Policy::patch_size_y));
+
+        const int BLOCK_SIZE = Policy::block_size_x * Policy::block_size_y;
+
+        glob_reduce<Reductor, BLOCK_SIZE, Policy::patch_size_x, Policy::patch_size_y><<<grid, block, 0, stream>>>(src, result, mask, rows, cols);
+        CV_CUDEV_SAFE_CALL( cudaGetLastError() );
+
+        if (stream == 0)
+            CV_CUDEV_SAFE_CALL( cudaDeviceSynchronize() );
+    }
+
+    // callers
+
+    template <class Policy, class SrcPtr, typename ResType, class MaskPtr>
+    __host__ void sum(const SrcPtr& src, ResType* result, const MaskPtr& mask, int rows, int cols, cudaStream_t stream)
+    {
+        typedef typename PtrTraits<SrcPtr>::value_type src_type;
+        const int cn = VecTraits<src_type>::cn;
+        typedef typename MakeVec<ResType, cn>::type work_type;
+
+        glob_reduce<SumReductor<src_type, work_type>, Policy>(src, result, mask, rows, cols, stream);
+    }
+
+    template <class Policy, class SrcPtr, typename ResType, class MaskPtr>
+    __host__ void minVal(const SrcPtr& src, ResType* result, const MaskPtr& mask, int rows, int cols, cudaStream_t stream)
+    {
+        typedef typename PtrTraits<SrcPtr>::value_type src_type;
+        const int cn = VecTraits<src_type>::cn;
+        typedef typename MakeVec<ResType, cn>::type work_type;
+
+        glob_reduce<MinMaxReductor<minop<work_type>, src_type, work_type>, Policy>(src, result, mask, rows, cols, stream);
+    }
+
+    template <class Policy, class SrcPtr, typename ResType, class MaskPtr>
+    __host__ void maxVal(const SrcPtr& src, ResType* result, const MaskPtr& mask, int rows, int cols, cudaStream_t stream)
+    {
+        typedef typename PtrTraits<SrcPtr>::value_type src_type;
+        const int cn = VecTraits<src_type>::cn;
+        typedef typename MakeVec<ResType, cn>::type work_type;
+
+        glob_reduce<MinMaxReductor<maxop<work_type>, src_type, work_type>, Policy>(src, result, mask, rows, cols, stream);
+    }
+
+    template <class Policy, class SrcPtr, typename ResType, class MaskPtr>
+    __host__ void minMaxVal(const SrcPtr& src, ResType* result, const MaskPtr& mask, int rows, int cols, cudaStream_t stream)
+    {
+        typedef typename PtrTraits<SrcPtr>::value_type src_type;
+        const int cn = VecTraits<src_type>::cn;
+        typedef typename MakeVec<ResType, cn>::type work_type;
+
+        glob_reduce<MinMaxReductor<both, src_type, work_type>, Policy>(src, result, mask, rows, cols, stream);
+    }
+}
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/grid/detail/histogram.hpp

@@ -0,0 +1,109 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_GRID_HISTOGRAM_DETAIL_HPP__
+#define __OPENCV_CUDEV_GRID_HISTOGRAM_DETAIL_HPP__
+
+#include "../../common.hpp"
+#include "../../util/atomic.hpp"
+
+namespace cv { namespace cudev {
+
+namespace grid_histogram_detail
+{
+    template <int BIN_COUNT, int BLOCK_SIZE, class SrcPtr, typename ResType, class MaskPtr>
+    __global__ void histogram(const SrcPtr src, ResType* hist, const MaskPtr mask, const int rows, const int cols)
+    {
+        __shared__ ResType smem[BIN_COUNT];
+
+        const int y = blockIdx.x * blockDim.y + threadIdx.y;
+        const int tid = threadIdx.y * blockDim.x + threadIdx.x;
+
+        for (int i = tid; i < BIN_COUNT; i += BLOCK_SIZE)
+            smem[i] = 0;
+
+        __syncthreads();
+
+        if (y < rows)
+        {
+            for (int x = threadIdx.x; x < cols; x += blockDim.x)
+            {
+                if (mask(y, x))
+                {
+                    const uint data = src(y, x);
+                    atomicAdd(&smem[data % BIN_COUNT], 1);
+                }
+            }
+        }
+
+        __syncthreads();
+
+        for (int i = tid; i < BIN_COUNT; i += BLOCK_SIZE)
+        {
+            const ResType histVal = smem[i];
+            if (histVal > 0)
+                atomicAdd(hist + i, histVal);
+        }
+    }
+
+    template <int BIN_COUNT, class Policy, class SrcPtr, typename ResType, class MaskPtr>
+    __host__ void histogram(const SrcPtr& src, ResType* hist, const MaskPtr& mask, int rows, int cols, cudaStream_t stream)
+    {
+        const dim3 block(Policy::block_size_x, Policy::block_size_y);
+        const dim3 grid(divUp(rows, block.y));
+
+        const int BLOCK_SIZE = Policy::block_size_x * Policy::block_size_y;
+
+        histogram<BIN_COUNT, BLOCK_SIZE><<<grid, block, 0, stream>>>(src, hist, mask, rows, cols);
+        CV_CUDEV_SAFE_CALL( cudaGetLastError() );
+
+        if (stream == 0)
+            CV_CUDEV_SAFE_CALL( cudaDeviceSynchronize() );
+    }
+}
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/grid/detail/integral.hpp

@@ -0,0 +1,626 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_GRID_INTEGRAL_DETAIL_HPP__
+#define __OPENCV_CUDEV_GRID_INTEGRAL_DETAIL_HPP__
+
+#include "../../common.hpp"
+#include "../../warp/shuffle.hpp"
+#include "../../block/scan.hpp"
+#include "../../ptr2d/glob.hpp"
+
+namespace cv { namespace cudev {
+
+namespace integral_detail
+{
+    // horizontal_pass
+
+    template <int NUM_SCAN_THREADS, class SrcPtr, typename D>
+    __global__ void horizontal_pass(const SrcPtr src, GlobPtr<D> dst, const int cols)
+    {
+        __shared__ D smem[NUM_SCAN_THREADS * 2];
+        __shared__ D carryElem;
+
+        carryElem = 0;
+
+        __syncthreads();
+
+        D* dst_row = dst.row(blockIdx.x);
+
+        int numBuckets = divUp(cols, NUM_SCAN_THREADS);
+        int offsetX = 0;
+
+        while (numBuckets--)
+        {
+            const int curElemOffs = offsetX + threadIdx.x;
+
+            D curElem = 0.0f;
+
+            if (curElemOffs < cols)
+                curElem = src(blockIdx.x, curElemOffs);
+
+            const D curScanElem = blockScanInclusive<NUM_SCAN_THREADS>(curElem, smem, threadIdx.x);
+
+            if (curElemOffs < cols)
+                dst_row[curElemOffs] = carryElem + curScanElem;
+
+            offsetX += NUM_SCAN_THREADS;
+
+            __syncthreads();
+
+            if (threadIdx.x == NUM_SCAN_THREADS - 1)
+            {
+                carryElem += curScanElem;
+            }
+
+            __syncthreads();
+        }
+    }
+
+    template <int NUM_SCAN_THREADS, typename T, typename D>
+    __global__ void horizontal_pass(const GlobPtr<T> src, GlobPtr<D> dst, const int cols)
+    {
+        __shared__ D smem[NUM_SCAN_THREADS * 2];
+        __shared__ D carryElem;
+
+        carryElem = 0;
+
+        __syncthreads();
+
+        const T* src_row = src.row(blockIdx.x);
+        D* dst_row = dst.row(blockIdx.x);
+
+        int numBuckets = divUp(cols, NUM_SCAN_THREADS);
+        int offsetX = 0;
+
+        while (numBuckets--)
+        {
+            const int curElemOffs = offsetX + threadIdx.x;
+
+            D curElem = 0.0f;
+
+            if (curElemOffs < cols)
+                curElem = src_row[curElemOffs];
+
+            const D curScanElem = blockScanInclusive<NUM_SCAN_THREADS>(curElem, smem, threadIdx.x);
+
+            if (curElemOffs < cols)
+                dst_row[curElemOffs] = carryElem + curScanElem;
+
+            offsetX += NUM_SCAN_THREADS;
+
+            __syncthreads();
+
+            if (threadIdx.x == NUM_SCAN_THREADS - 1)
+            {
+                carryElem += curScanElem;
+            }
+
+            __syncthreads();
+        }
+    }
+
+    template <class SrcPtr, typename D>
+    __host__ void horizontal_pass(const SrcPtr& src, const GlobPtr<D>& dst, int rows, int cols, cudaStream_t stream)
+    {
+        const int NUM_SCAN_THREADS = 256;
+
+        const dim3 block(NUM_SCAN_THREADS);
+        const dim3 grid(rows);
+
+        horizontal_pass<NUM_SCAN_THREADS><<<grid, block, 0, stream>>>(src, dst, cols);
+        CV_CUDEV_SAFE_CALL( cudaGetLastError() );
+    }
+
+    // horisontal_pass_8u_shfl
+
+    __device__ static uchar4 int_to_uchar4(unsigned int in)
+    {
+        uchar4 bytes;
+        bytes.x = (in & 0x000000ff) >>  0;
+        bytes.y = (in & 0x0000ff00) >>  8;
+        bytes.z = (in & 0x00ff0000) >> 16;
+        bytes.w = (in & 0xff000000) >> 24;
+        return bytes;
+    }
+
+    __global__ static void horisontal_pass_8u_shfl_kernel(const GlobPtr<uint4> img, GlobPtr<uint4> integral)
+    {
+    #if CV_CUDEV_ARCH >= 300
+        __shared__ int sums[128];
+
+        const int id = threadIdx.x;
+        const int lane_id = id % warpSize;
+        const int warp_id = id / warpSize;
+
+        const uint4 data = img(blockIdx.x, id);
+
+        const uchar4 a = int_to_uchar4(data.x);
+        const uchar4 b = int_to_uchar4(data.y);
+        const uchar4 c = int_to_uchar4(data.z);
+        const uchar4 d = int_to_uchar4(data.w);
+
+        int result[16];
+
+        result[0]  =              a.x;
+        result[1]  = result[0]  + a.y;
+        result[2]  = result[1]  + a.z;
+        result[3]  = result[2]  + a.w;
+
+        result[4]  = result[3]  + b.x;
+        result[5]  = result[4]  + b.y;
+        result[6]  = result[5]  + b.z;
+        result[7]  = result[6]  + b.w;
+
+        result[8]  = result[7]  + c.x;
+        result[9]  = result[8]  + c.y;
+        result[10] = result[9]  + c.z;
+        result[11] = result[10] + c.w;
+
+        result[12] = result[11] + d.x;
+        result[13] = result[12] + d.y;
+        result[14] = result[13] + d.z;
+        result[15] = result[14] + d.w;
+
+        int sum = result[15];
+
+        // the prefix sum for each thread's 16 value is computed,
+        // now the final sums (result[15]) need to be shared
+        // with the other threads and add.  To do this,
+        // the shfl_up() instruction is used and a shuffle scan
+        // operation is performed to distribute the sums to the correct
+        // threads
+        #pragma unroll
+        for (int i = 1; i < 32; i *= 2)
+        {
+            const int n = shfl_up(sum, i, 32);
+
+            if (lane_id >= i)
+            {
+                #pragma unroll
+                for (int k = 0; k < 16; ++k)
+                    result[k] += n;
+
+                sum += n;
+            }
+        }
+
+        // Now the final sum for the warp must be shared
+        // between warps.  This is done by each warp
+        // having a thread store to shared memory, then
+        // having some other warp load the values and
+        // compute a prefix sum, again by using shfl_up.
+        // The results are uniformly added back to the warps.
+        // last thread in the warp holding sum of the warp
+        // places that in shared
+        if (threadIdx.x % warpSize == warpSize - 1)
+            sums[warp_id] = result[15];
+
+        __syncthreads();
+
+        if (warp_id == 0)
+        {
+            int warp_sum = sums[lane_id];
+
+            #pragma unroll
+            for (int i = 1; i <= 32; i *= 2)
+            {
+                const int n = shfl_up(warp_sum, i, 32);
+
+                if (lane_id >= i)
+                    warp_sum += n;
+            }
+
+            sums[lane_id] = warp_sum;
+        }
+
+        __syncthreads();
+
+        int blockSum = 0;
+
+        // fold in unused warp
+        if (warp_id > 0)
+        {
+            blockSum = sums[warp_id - 1];
+
+            #pragma unroll
+            for (int k = 0; k < 16; ++k)
+                result[k] += blockSum;
+        }
+
+        // assemble result
+        // Each thread has 16 values to write, which are
+        // now integer data (to avoid overflow).  Instead of
+        // each thread writing consecutive uint4s, the
+        // approach shown here experiments using
+        // the shuffle command to reformat the data
+        // inside the registers so that each thread holds
+        // consecutive data to be written so larger contiguous
+        // segments can be assembled for writing.
+
+        /*
+            For example data that needs to be written as
+
+            GMEM[16] <- x0 x1 x2 x3 y0 y1 y2 y3 z0 z1 z2 z3 w0 w1 w2 w3
+            but is stored in registers (r0..r3), in four threads (0..3) as:
+
+            threadId   0  1  2  3
+              r0      x0 y0 z0 w0
+              r1      x1 y1 z1 w1
+              r2      x2 y2 z2 w2
+              r3      x3 y3 z3 w3
+
+              after apply shfl_xor operations to move data between registers r1..r3:
+
+            threadId  00 01 10 11
+                      x0 y0 z0 w0
+             xor(01)->y1 x1 w1 z1
+             xor(10)->z2 w2 x2 y2
+             xor(11)->w3 z3 y3 x3
+
+             and now x0..x3, and z0..z3 can be written out in order by all threads.
+
+             In the current code, each register above is actually representing
+             four integers to be written as uint4's to GMEM.
+        */
+
+        result[4]  = shfl_xor(result[4] , 1, 32);
+        result[5]  = shfl_xor(result[5] , 1, 32);
+        result[6]  = shfl_xor(result[6] , 1, 32);
+        result[7]  = shfl_xor(result[7] , 1, 32);
+
+        result[8]  = shfl_xor(result[8] , 2, 32);
+        result[9]  = shfl_xor(result[9] , 2, 32);
+        result[10] = shfl_xor(result[10], 2, 32);
+        result[11] = shfl_xor(result[11], 2, 32);
+
+        result[12] = shfl_xor(result[12], 3, 32);
+        result[13] = shfl_xor(result[13], 3, 32);
+        result[14] = shfl_xor(result[14], 3, 32);
+        result[15] = shfl_xor(result[15], 3, 32);
+
+        uint4* integral_row = integral.row(blockIdx.x);
+        uint4 output;
+
+        ///////
+
+        if (threadIdx.x % 4 == 0)
+            output = make_uint4(result[0], result[1], result[2], result[3]);
+
+        if (threadIdx.x % 4 == 1)
+            output = make_uint4(result[4], result[5], result[6], result[7]);
+
+        if (threadIdx.x % 4 == 2)
+            output = make_uint4(result[8], result[9], result[10], result[11]);
+
+        if (threadIdx.x % 4 == 3)
+            output = make_uint4(result[12], result[13], result[14], result[15]);
+
+        integral_row[threadIdx.x % 4 + (threadIdx.x / 4) * 16] = output;
+
+        ///////
+
+        if (threadIdx.x % 4 == 2)
+            output = make_uint4(result[0], result[1], result[2], result[3]);
+
+        if (threadIdx.x % 4 == 3)
+            output = make_uint4(result[4], result[5], result[6], result[7]);
+
+        if (threadIdx.x % 4 == 0)
+            output = make_uint4(result[8], result[9], result[10], result[11]);
+
+        if (threadIdx.x % 4 == 1)
+            output = make_uint4(result[12], result[13], result[14], result[15]);
+
+        integral_row[(threadIdx.x + 2) % 4 + (threadIdx.x / 4) * 16 + 8] = output;
+
+        // continuning from the above example,
+        // this use of shfl_xor() places the y0..y3 and w0..w3 data
+        // in order.
+
+        #pragma unroll
+        for (int i = 0; i < 16; ++i)
+            result[i] = shfl_xor(result[i], 1, 32);
+
+        if (threadIdx.x % 4 == 0)
+            output = make_uint4(result[0], result[1], result[2], result[3]);
+
+        if (threadIdx.x % 4 == 1)
+            output = make_uint4(result[4], result[5], result[6], result[7]);
+
+        if (threadIdx.x % 4 == 2)
+            output = make_uint4(result[8], result[9], result[10], result[11]);
+
+        if (threadIdx.x % 4 == 3)
+            output = make_uint4(result[12], result[13], result[14], result[15]);
+
+        integral_row[threadIdx.x % 4 + (threadIdx.x / 4) * 16 + 4] = output;
+
+        ///////
+
+        if (threadIdx.x % 4 == 2)
+            output = make_uint4(result[0], result[1], result[2], result[3]);
+
+        if (threadIdx.x % 4 == 3)
+            output = make_uint4(result[4], result[5], result[6], result[7]);
+
+        if (threadIdx.x % 4 == 0)
+            output = make_uint4(result[8], result[9], result[10], result[11]);
+
+        if (threadIdx.x % 4 == 1)
+            output = make_uint4(result[12], result[13], result[14], result[15]);
+
+        integral_row[(threadIdx.x + 2) % 4 + (threadIdx.x / 4) * 16 + 12] = output;
+    #endif
+    }
+
+    __host__ static void horisontal_pass_8u_shfl(const GlobPtr<uchar> src, GlobPtr<uint> integral, int rows, int cols, cudaStream_t stream)
+    {
+        // each thread handles 16 values, use 1 block/row
+        // save, because step is actually can't be less 512 bytes
+        const int block = cols / 16;
+
+        // launch 1 block / row
+        const int grid = rows;
+
+        CV_CUDEV_SAFE_CALL( cudaFuncSetCacheConfig(horisontal_pass_8u_shfl_kernel, cudaFuncCachePreferL1) );
+
+        GlobPtr<uint4> src4 = globPtr((uint4*) src.data, src.step);
+        GlobPtr<uint4> integral4 = globPtr((uint4*) integral.data, integral.step);
+
+        horisontal_pass_8u_shfl_kernel<<<grid, block, 0, stream>>>(src4, integral4);
+        CV_CUDEV_SAFE_CALL( cudaGetLastError() );
+    }
+
+    // vertical
+
+    template <typename T>
+    __global__ void vertical_pass(GlobPtr<T> integral, const int rows, const int cols)
+    {
+    #if CV_CUDEV_ARCH >= 300
+        __shared__ T sums[32][9];
+
+        const int tidx = blockIdx.x * blockDim.x + threadIdx.x;
+        const int lane_id = tidx % 8;
+
+        sums[threadIdx.x][threadIdx.y] = 0;
+        __syncthreads();
+
+        T stepSum = 0;
+
+        int numBuckets = divUp(rows, blockDim.y);
+        int y = threadIdx.y;
+
+        while (numBuckets--)
+        {
+            T* p = integral.row(y) + tidx;
+
+            T sum = (tidx < cols) && (y < rows) ? *p : 0;
+
+            y += blockDim.y;
+
+            sums[threadIdx.x][threadIdx.y] = sum;
+            __syncthreads();
+
+            // place into SMEM
+            // shfl scan reduce the SMEM, reformating so the column
+            // sums are computed in a warp
+            // then read out properly
+            const int j = threadIdx.x % 8;
+            const int k = threadIdx.x / 8 + threadIdx.y * 4;
+
+            T partial_sum = sums[k][j];
+
+            for (int i = 1; i <= 8; i *= 2)
+            {
+                T n = shfl_up(partial_sum, i, 32);
+
+                if (lane_id >= i)
+                    partial_sum += n;
+            }
+
+            sums[k][j] = partial_sum;
+            __syncthreads();
+
+            if (threadIdx.y > 0)
+                sum += sums[threadIdx.x][threadIdx.y - 1];
+
+            if (tidx < cols)
+            {
+                sum += stepSum;
+                stepSum += sums[threadIdx.x][blockDim.y - 1];
+                *p = sum;
+            }
+
+            __syncthreads();
+        }
+    #else
+        __shared__ T smem[32][32];
+        __shared__ T prevVals[32];
+
+        volatile T* smem_row = &smem[0][0] + 64 * threadIdx.y;
+
+        if (threadIdx.y == 0)
+            prevVals[threadIdx.x] = 0;
+
+        __syncthreads();
+
+        const int x = blockIdx.x * blockDim.x + threadIdx.x;
+
+        int numBuckets = divUp(rows, 8 * 4);
+        int offsetY = 0;
+
+        while (numBuckets--)
+        {
+            const int curRowOffs = offsetY + threadIdx.y;
+
+            T curElems[4];
+            T temp[4];
+
+            // load patch
+
+            smem[threadIdx.y +  0][threadIdx.x] = 0.0f;
+            smem[threadIdx.y +  8][threadIdx.x] = 0.0f;
+            smem[threadIdx.y + 16][threadIdx.x] = 0.0f;
+            smem[threadIdx.y + 24][threadIdx.x] = 0.0f;
+
+            if (x < cols)
+            {
+                for (int i = 0; i < 4; ++i)
+                {
+                    if (curRowOffs + i * 8 < rows)
+                        smem[threadIdx.y + i * 8][threadIdx.x] = integral(curRowOffs + i * 8, x);
+                }
+            }
+
+            __syncthreads();
+
+            // reduce
+
+            curElems[0] = smem[threadIdx.x][threadIdx.y     ];
+            curElems[1] = smem[threadIdx.x][threadIdx.y +  8];
+            curElems[2] = smem[threadIdx.x][threadIdx.y + 16];
+            curElems[3] = smem[threadIdx.x][threadIdx.y + 24];
+
+            __syncthreads();
+
+            temp[0] = curElems[0] = warpScanInclusive(curElems[0], smem_row, threadIdx.x);
+            temp[1] = curElems[1] = warpScanInclusive(curElems[1], smem_row, threadIdx.x);
+            temp[2] = curElems[2] = warpScanInclusive(curElems[2], smem_row, threadIdx.x);
+            temp[3] = curElems[3] = warpScanInclusive(curElems[3], smem_row, threadIdx.x);
+
+            curElems[0] += prevVals[threadIdx.y     ];
+            curElems[1] += prevVals[threadIdx.y +  8];
+            curElems[2] += prevVals[threadIdx.y + 16];
+            curElems[3] += prevVals[threadIdx.y + 24];
+
+            __syncthreads();
+
+            if (threadIdx.x == 31)
+            {
+                prevVals[threadIdx.y     ] += temp[0];
+                prevVals[threadIdx.y +  8] += temp[1];
+                prevVals[threadIdx.y + 16] += temp[2];
+                prevVals[threadIdx.y + 24] += temp[3];
+            }
+
+            smem[threadIdx.y     ][threadIdx.x] = curElems[0];
+            smem[threadIdx.y +  8][threadIdx.x] = curElems[1];
+            smem[threadIdx.y + 16][threadIdx.x] = curElems[2];
+            smem[threadIdx.y + 24][threadIdx.x] = curElems[3];
+
+            __syncthreads();
+
+            // store patch
+
+            if (x < cols)
+            {
+                // read 4 value from source
+                for (int i = 0; i < 4; ++i)
+                {
+                    if (curRowOffs + i * 8 < rows)
+                        integral(curRowOffs + i * 8, x) = smem[threadIdx.x][threadIdx.y + i * 8];
+                }
+            }
+
+            __syncthreads();
+
+            offsetY += 8 * 4;
+        }
+    #endif
+    }
+
+    template <typename T>
+    __host__ void vertical_pass(const GlobPtr<T>& integral, int rows, int cols, cudaStream_t stream)
+    {
+        const dim3 block(32, 8);
+        const dim3 grid(divUp(cols, block.x));
+
+        vertical_pass<<<grid, block, 0, stream>>>(integral, rows, cols);
+        CV_CUDEV_SAFE_CALL( cudaGetLastError() );
+    }
+
+    // integral
+
+    template <class SrcPtr, typename D>
+    __host__ void integral(const SrcPtr& src, const GlobPtr<D>& dst, int rows, int cols, cudaStream_t stream)
+    {
+        horizontal_pass(src, dst, rows, cols, stream);
+        vertical_pass(dst, rows, cols, stream);
+
+        if (stream == 0)
+            CV_CUDEV_SAFE_CALL( cudaDeviceSynchronize() );
+    }
+
+    __host__ static void integral(const GlobPtr<uchar> src, GlobPtr<uint> dst, int rows, int cols, cudaStream_t stream)
+    {
+        if (deviceSupports(FEATURE_SET_COMPUTE_30)
+            && (cols % 16 == 0)
+            && reinterpret_cast<intptr_t>(src.data) % 32 == 0
+            && reinterpret_cast<intptr_t>(dst.data) % 32 == 0)
+        {
+            horisontal_pass_8u_shfl(src, dst, rows, cols, stream);
+        }
+        else
+        {
+            horizontal_pass(src, dst, rows, cols, stream);
+        }
+
+        vertical_pass(dst, rows, cols, stream);
+
+        if (stream == 0)
+            CV_CUDEV_SAFE_CALL( cudaDeviceSynchronize() );
+    }
+
+    __host__ static void integral(const GlobPtr<uchar> src, GlobPtr<int> dst, int rows, int cols, cudaStream_t stream)
+    {
+        GlobPtr<uint> dstui = globPtr((uint*) dst.data, dst.step);
+        integral(src, dstui, rows, cols, stream);
+    }
+}
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/grid/detail/pyr_down.hpp

@@ -0,0 +1,201 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_GRID_PYR_DOWN_DETAIL_HPP__
+#define __OPENCV_CUDEV_GRID_PYR_DOWN_DETAIL_HPP__
+
+#include "../../common.hpp"
+#include "../../util/vec_traits.hpp"
+#include "../../util/saturate_cast.hpp"
+#include "../../util/type_traits.hpp"
+#include "../../ptr2d/glob.hpp"
+#include "../../ptr2d/traits.hpp"
+
+namespace cv { namespace cudev {
+
+namespace pyramids_detail
+{
+    template <class Brd, class SrcPtr, typename DstType>
+    __global__ void pyrDown(const SrcPtr src, GlobPtr<DstType> dst, const int src_rows, const int src_cols, const int dst_cols)
+    {
+        typedef typename PtrTraits<SrcPtr>::value_type src_type;
+        typedef typename VecTraits<src_type>::elem_type src_elem_type;
+        typedef typename LargerType<float, src_elem_type>::type work_elem_type;
+        typedef typename MakeVec<work_elem_type, VecTraits<src_type>::cn>::type work_type;
+
+        __shared__ work_type smem[256 + 4];
+
+        const int x = blockIdx.x * blockDim.x + threadIdx.x;
+        const int y = blockIdx.y;
+
+        const int src_y = 2 * y;
+
+        if (src_y >= 2 && src_y < src_rows - 2 && x >= 2 && x < src_cols - 2)
+        {
+            {
+                work_type sum;
+
+                sum =       0.0625f * src(src_y - 2, x);
+                sum = sum + 0.25f   * src(src_y - 1, x);
+                sum = sum + 0.375f  * src(src_y    , x);
+                sum = sum + 0.25f   * src(src_y + 1, x);
+                sum = sum + 0.0625f * src(src_y + 2, x);
+
+                smem[2 + threadIdx.x] = sum;
+            }
+
+            if (threadIdx.x < 2)
+            {
+                const int left_x = x - 2;
+
+                work_type sum;
+
+                sum =       0.0625f * src(src_y - 2, left_x);
+                sum = sum + 0.25f   * src(src_y - 1, left_x);
+                sum = sum + 0.375f  * src(src_y    , left_x);
+                sum = sum + 0.25f   * src(src_y + 1, left_x);
+                sum = sum + 0.0625f * src(src_y + 2, left_x);
+
+                smem[threadIdx.x] = sum;
+            }
+
+            if (threadIdx.x > 253)
+            {
+                const int right_x = x + 2;
+
+                work_type sum;
+
+                sum =       0.0625f * src(src_y - 2, right_x);
+                sum = sum + 0.25f   * src(src_y - 1, right_x);
+                sum = sum + 0.375f  * src(src_y    , right_x);
+                sum = sum + 0.25f   * src(src_y + 1, right_x);
+                sum = sum + 0.0625f * src(src_y + 2, right_x);
+
+                smem[4 + threadIdx.x] = sum;
+            }
+        }
+        else
+        {
+            {
+                work_type sum;
+
+                sum =       0.0625f * src(Brd::idx_low(src_y - 2, src_rows) , Brd::idx_high(x, src_cols));
+                sum = sum + 0.25f   * src(Brd::idx_low(src_y - 1, src_rows) , Brd::idx_high(x, src_cols));
+                sum = sum + 0.375f  * src(src_y                             , Brd::idx_high(x, src_cols));
+                sum = sum + 0.25f   * src(Brd::idx_high(src_y + 1, src_rows), Brd::idx_high(x, src_cols));
+                sum = sum + 0.0625f * src(Brd::idx_high(src_y + 2, src_rows), Brd::idx_high(x, src_cols));
+
+                smem[2 + threadIdx.x] = sum;
+            }
+
+            if (threadIdx.x < 2)
+            {
+                const int left_x = x - 2;
+
+                work_type sum;
+
+                sum =       0.0625f * src(Brd::idx_low(src_y - 2, src_rows) , Brd::idx_low(Brd::idx_high(left_x, src_cols), src_cols));
+                sum = sum + 0.25f   * src(Brd::idx_low(src_y - 1, src_rows) , Brd::idx_low(Brd::idx_high(left_x, src_cols), src_cols));
+                sum = sum + 0.375f  * src(src_y                             , Brd::idx_low(Brd::idx_high(left_x, src_cols), src_cols));
+                sum = sum + 0.25f   * src(Brd::idx_high(src_y + 1, src_rows), Brd::idx_low(Brd::idx_high(left_x, src_cols), src_cols));
+                sum = sum + 0.0625f * src(Brd::idx_high(src_y + 2, src_rows), Brd::idx_low(Brd::idx_high(left_x, src_cols), src_cols));
+
+                smem[threadIdx.x] = sum;
+            }
+
+            if (threadIdx.x > 253)
+            {
+                const int right_x = x + 2;
+
+                work_type sum;
+
+                sum =       0.0625f * src(Brd::idx_low(src_y - 2, src_rows) , Brd::idx_high(right_x, src_cols));
+                sum = sum + 0.25f   * src(Brd::idx_low(src_y - 1, src_rows) , Brd::idx_high(right_x, src_cols));
+                sum = sum + 0.375f  * src(src_y                             , Brd::idx_high(right_x, src_cols));
+                sum = sum + 0.25f   * src(Brd::idx_high(src_y + 1, src_rows), Brd::idx_high(right_x, src_cols));
+                sum = sum + 0.0625f * src(Brd::idx_high(src_y + 2, src_rows), Brd::idx_high(right_x, src_cols));
+
+                smem[4 + threadIdx.x] = sum;
+            }
+        }
+
+        __syncthreads();
+
+        if (threadIdx.x < 128)
+        {
+            const int tid2 = threadIdx.x * 2;
+
+            work_type sum;
+
+            sum =       0.0625f * smem[2 + tid2 - 2];
+            sum = sum + 0.25f   * smem[2 + tid2 - 1];
+            sum = sum + 0.375f  * smem[2 + tid2    ];
+            sum = sum + 0.25f   * smem[2 + tid2 + 1];
+            sum = sum + 0.0625f * smem[2 + tid2 + 2];
+
+            const int dst_x = (blockIdx.x * blockDim.x + tid2) / 2;
+
+            if (dst_x < dst_cols)
+                dst(y, dst_x) = saturate_cast<DstType>(sum);
+        }
+    }
+
+    template <class Brd, class SrcPtr, typename DstType>
+    __host__ void pyrDown(const SrcPtr& src, const GlobPtr<DstType>& dst, int src_rows, int src_cols, int dst_rows, int dst_cols, cudaStream_t stream)
+    {
+        const dim3 block(256);
+        const dim3 grid(divUp(src_cols, block.x), dst_rows);
+
+        pyrDown<Brd><<<grid, block, 0, stream>>>(src, dst, src_rows, src_cols, dst_cols);
+        CV_CUDEV_SAFE_CALL( cudaGetLastError() );
+
+        if (stream == 0)
+            CV_CUDEV_SAFE_CALL( cudaDeviceSynchronize() );
+    }
+}
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/grid/detail/pyr_up.hpp

@@ -0,0 +1,172 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_GRID_PYR_UP_DETAIL_HPP__
+#define __OPENCV_CUDEV_GRID_PYR_UP_DETAIL_HPP__
+
+#include "../../common.hpp"
+#include "../../util/vec_traits.hpp"
+#include "../../util/saturate_cast.hpp"
+#include "../../util/type_traits.hpp"
+#include "../../ptr2d/glob.hpp"
+#include "../../ptr2d/traits.hpp"
+
+namespace cv { namespace cudev {
+
+namespace pyramids_detail
+{
+    template <class SrcPtr, typename DstType>
+    __global__ void pyrUp(const SrcPtr src, GlobPtr<DstType> dst, const int src_rows, const int src_cols, const int dst_rows, const int dst_cols)
+    {
+        typedef typename PtrTraits<SrcPtr>::value_type src_type;
+        typedef typename VecTraits<src_type>::elem_type src_elem_type;
+        typedef typename LargerType<float, src_elem_type>::type work_elem_type;
+        typedef typename MakeVec<work_elem_type, VecTraits<src_type>::cn>::type work_type;
+
+        const int x = blockIdx.x * blockDim.x + threadIdx.x;
+        const int y = blockIdx.y * blockDim.y + threadIdx.y;
+
+        __shared__ work_type s_srcPatch[10][10];
+        __shared__ work_type s_dstPatch[20][16];
+
+        if (threadIdx.x < 10 && threadIdx.y < 10)
+        {
+            int srcx = static_cast<int>((blockIdx.x * blockDim.x) / 2 + threadIdx.x) - 1;
+            int srcy = static_cast<int>((blockIdx.y * blockDim.y) / 2 + threadIdx.y) - 1;
+
+            srcx = ::abs(srcx);
+            srcx = ::min(src_cols - 1, srcx);
+
+            srcy = ::abs(srcy);
+            srcy = ::min(src_rows - 1, srcy);
+
+            s_srcPatch[threadIdx.y][threadIdx.x] = saturate_cast<work_type>(src(srcy, srcx));
+        }
+
+        __syncthreads();
+
+        work_type sum = VecTraits<work_type>::all(0);
+
+        const int evenFlag = static_cast<int>((threadIdx.x & 1) == 0);
+        const int oddFlag  = static_cast<int>((threadIdx.x & 1) != 0);
+        const bool eveny = ((threadIdx.y & 1) == 0);
+        const int tidx = threadIdx.x;
+
+        if (eveny)
+        {
+            sum = sum + (evenFlag * 0.0625f) * s_srcPatch[1 + (threadIdx.y >> 1)][1 + ((tidx - 2) >> 1)];
+            sum = sum + ( oddFlag * 0.25f  ) * s_srcPatch[1 + (threadIdx.y >> 1)][1 + ((tidx - 1) >> 1)];
+            sum = sum + (evenFlag * 0.375f ) * s_srcPatch[1 + (threadIdx.y >> 1)][1 + ((tidx    ) >> 1)];
+            sum = sum + ( oddFlag * 0.25f  ) * s_srcPatch[1 + (threadIdx.y >> 1)][1 + ((tidx + 1) >> 1)];
+            sum = sum + (evenFlag * 0.0625f) * s_srcPatch[1 + (threadIdx.y >> 1)][1 + ((tidx + 2) >> 1)];
+        }
+
+        s_dstPatch[2 + threadIdx.y][threadIdx.x] = sum;
+
+        if (threadIdx.y < 2)
+        {
+            sum = VecTraits<work_type>::all(0);
+
+            if (eveny)
+            {
+                sum = sum + (evenFlag * 0.0625f) * s_srcPatch[0][1 + ((tidx - 2) >> 1)];
+                sum = sum + ( oddFlag * 0.25f  ) * s_srcPatch[0][1 + ((tidx - 1) >> 1)];
+                sum = sum + (evenFlag * 0.375f ) * s_srcPatch[0][1 + ((tidx    ) >> 1)];
+                sum = sum + ( oddFlag * 0.25f  ) * s_srcPatch[0][1 + ((tidx + 1) >> 1)];
+                sum = sum + (evenFlag * 0.0625f) * s_srcPatch[0][1 + ((tidx + 2) >> 1)];
+            }
+
+            s_dstPatch[threadIdx.y][threadIdx.x] = sum;
+        }
+
+        if (threadIdx.y > 13)
+        {
+            sum = VecTraits<work_type>::all(0);
+
+            if (eveny)
+            {
+                sum = sum + (evenFlag * 0.0625f) * s_srcPatch[9][1 + ((tidx - 2) >> 1)];
+                sum = sum + ( oddFlag * 0.25f  ) * s_srcPatch[9][1 + ((tidx - 1) >> 1)];
+                sum = sum + (evenFlag * 0.375f ) * s_srcPatch[9][1 + ((tidx    ) >> 1)];
+                sum = sum + ( oddFlag * 0.25f  ) * s_srcPatch[9][1 + ((tidx + 1) >> 1)];
+                sum = sum + (evenFlag * 0.0625f) * s_srcPatch[9][1 + ((tidx + 2) >> 1)];
+            }
+
+            s_dstPatch[4 + threadIdx.y][threadIdx.x] = sum;
+        }
+
+        __syncthreads();
+
+        sum = VecTraits<work_type>::all(0);
+
+        const int tidy = threadIdx.y;
+
+        sum = sum + 0.0625f * s_dstPatch[2 + tidy - 2][threadIdx.x];
+        sum = sum + 0.25f   * s_dstPatch[2 + tidy - 1][threadIdx.x];
+        sum = sum + 0.375f  * s_dstPatch[2 + tidy    ][threadIdx.x];
+        sum = sum + 0.25f   * s_dstPatch[2 + tidy + 1][threadIdx.x];
+        sum = sum + 0.0625f * s_dstPatch[2 + tidy + 2][threadIdx.x];
+
+        if (x < dst_cols && y < dst_rows)
+            dst(y, x) = saturate_cast<DstType>(4.0f * sum);
+    }
+
+    template <class SrcPtr, typename DstType>
+    __host__ void pyrUp(const SrcPtr& src, const GlobPtr<DstType>& dst, int src_rows, int src_cols, int dst_rows, int dst_cols, cudaStream_t stream)
+    {
+        const dim3 block(16, 16);
+        const dim3 grid(divUp(dst_cols, block.x), divUp(dst_rows, block.y));
+
+        pyrUp<<<grid, block, 0, stream>>>(src, dst, src_rows, src_cols, dst_rows, dst_cols);
+        CV_CUDEV_SAFE_CALL( cudaGetLastError() );
+
+        if (stream == 0)
+            CV_CUDEV_SAFE_CALL( cudaDeviceSynchronize() );
+    }
+}
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/grid/detail/reduce_to_column.hpp

@@ -0,0 +1,106 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_GRID_REDUCE_TO_COLUMN_DETAIL_HPP__
+#define __OPENCV_CUDEV_GRID_REDUCE_TO_COLUMN_DETAIL_HPP__
+
+#include "../../common.hpp"
+#include "../../util/saturate_cast.hpp"
+#include "../../block/reduce.hpp"
+
+namespace cv { namespace cudev {
+
+namespace grid_reduce_to_vec_detail
+{
+    template <class Reductor, int BLOCK_SIZE, class SrcPtr, typename ResType, class MaskPtr>
+    __global__ void reduceToColumn(const SrcPtr src, ResType* dst, const MaskPtr mask, const int cols)
+    {
+        typedef typename Reductor::work_type work_type;
+
+        __shared__ work_type smem[BLOCK_SIZE];
+
+        const int y = blockIdx.x;
+
+        work_type myVal = Reductor::initialValue();
+
+        Reductor op;
+
+        for (int x = threadIdx.x; x < cols; x += BLOCK_SIZE)
+        {
+            if (mask(y, x))
+            {
+                myVal = op(myVal, saturate_cast<work_type>(src(y, x)));
+            }
+        }
+
+        blockReduce<BLOCK_SIZE>(smem, myVal, threadIdx.x, op);
+
+        if (threadIdx.x == 0)
+            dst[y] = saturate_cast<ResType>(Reductor::result(myVal, cols));
+    }
+
+    template <class Reductor, class Policy, class SrcPtr, typename ResType, class MaskPtr>
+    __host__ void reduceToColumn(const SrcPtr& src, ResType* dst, const MaskPtr& mask, int rows, int cols, cudaStream_t stream)
+    {
+        const int BLOCK_SIZE_X = Policy::block_size_x;
+        const int BLOCK_SIZE_Y = Policy::block_size_y;
+
+        const int BLOCK_SIZE = BLOCK_SIZE_X * BLOCK_SIZE_Y;
+
+        const dim3 block(BLOCK_SIZE);
+        const dim3 grid(rows);
+
+        reduceToColumn<Reductor, BLOCK_SIZE><<<grid, block, 0, stream>>>(src, dst, mask, cols);
+        CV_CUDEV_SAFE_CALL( cudaGetLastError() );
+
+        if (stream == 0)
+            CV_CUDEV_SAFE_CALL( cudaDeviceSynchronize() );
+
+    }
+}
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/grid/detail/reduce_to_row.hpp

@@ -0,0 +1,118 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_GRID_REDUCE_TO_ROW_DETAIL_HPP__
+#define __OPENCV_CUDEV_GRID_REDUCE_TO_ROW_DETAIL_HPP__
+
+#include "../../common.hpp"
+#include "../../util/saturate_cast.hpp"
+#include "../../block/reduce.hpp"
+
+namespace cv { namespace cudev {
+
+namespace grid_reduce_to_vec_detail
+{
+    template <class Reductor, int BLOCK_SIZE_X, int BLOCK_SIZE_Y, class SrcPtr, typename ResType, class MaskPtr>
+    __global__ void reduceToRow(const SrcPtr src, ResType* dst, const MaskPtr mask, const int rows, const int cols)
+    {
+        typedef typename Reductor::work_type work_type;
+
+        __shared__ work_type smem[BLOCK_SIZE_X * BLOCK_SIZE_Y];
+
+        const int x = blockIdx.x * BLOCK_SIZE_X + threadIdx.x;
+
+        work_type myVal = Reductor::initialValue();
+
+        Reductor op;
+
+        if (x < cols)
+        {
+            for (int y = threadIdx.y; y < rows; y += BLOCK_SIZE_Y)
+            {
+                if (mask(y, x))
+                {
+                    myVal = op(myVal, saturate_cast<work_type>(src(y, x)));
+                }
+            }
+        }
+
+        smem[threadIdx.x * BLOCK_SIZE_Y + threadIdx.y] = myVal;
+
+        __syncthreads();
+
+        volatile work_type* srow = smem + threadIdx.y * BLOCK_SIZE_X;
+
+        myVal = srow[threadIdx.x];
+        blockReduce<BLOCK_SIZE_X>(srow, myVal, threadIdx.x, op);
+
+        if (threadIdx.x == 0)
+            srow[0] = myVal;
+
+        __syncthreads();
+
+        if (threadIdx.y == 0 && x < cols)
+            dst[x] = saturate_cast<ResType>(Reductor::result(smem[threadIdx.x * BLOCK_SIZE_X], rows));
+    }
+
+    template <class Reductor, class SrcPtr, typename ResType, class MaskPtr>
+    __host__ void reduceToRow(const SrcPtr& src, ResType* dst, const MaskPtr& mask, int rows, int cols, cudaStream_t stream)
+    {
+        const int BLOCK_SIZE_X = 16;
+        const int BLOCK_SIZE_Y = 16;
+
+        const dim3 block(BLOCK_SIZE_X, BLOCK_SIZE_Y);
+        const dim3 grid(divUp(cols, block.x));
+
+        reduceToRow<Reductor, BLOCK_SIZE_X, BLOCK_SIZE_Y><<<grid, block, 0, stream>>>(src, dst, mask, rows, cols);
+        CV_CUDEV_SAFE_CALL( cudaGetLastError() );
+
+        if (stream == 0)
+            CV_CUDEV_SAFE_CALL( cudaDeviceSynchronize() );
+    }
+}
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/grid/detail/split_merge.hpp

@@ -0,0 +1,282 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_GRID_SPLIT_MERGE_DETAIL_HPP__
+#define __OPENCV_CUDEV_GRID_SPLIT_MERGE_DETAIL_HPP__
+
+#include "../../common.hpp"
+#include "../../util/saturate_cast.hpp"
+#include "../../util/tuple.hpp"
+#include "../../util/vec_traits.hpp"
+#include "../../ptr2d/glob.hpp"
+#include "../../ptr2d/traits.hpp"
+
+namespace cv { namespace cudev {
+
+namespace grid_split_merge_detail
+{
+    // merge
+
+    template <class Src1Ptr, class Src2Ptr, typename DstType, class MaskPtr>
+    __global__ void mergeC2(const Src1Ptr src1, const Src2Ptr src2, GlobPtr<DstType> dst, const MaskPtr mask, const int rows, const int cols)
+    {
+        typedef typename VecTraits<DstType>::elem_type dst_elem_type;
+
+        const int x = blockIdx.x * blockDim.x + threadIdx.x;
+        const int y = blockIdx.y * blockDim.y + threadIdx.y;
+
+        if (x >= cols || y >= rows || !mask(y, x))
+            return;
+
+        dst(y, x) = VecTraits<DstType>::make(
+                    saturate_cast<dst_elem_type>(src1(y, x)),
+                    saturate_cast<dst_elem_type>(src2(y, x))
+                    );
+    }
+
+    template <class Policy, class Src1Ptr, class Src2Ptr, typename DstType, class MaskPtr>
+    __host__ void mergeC2(const Src1Ptr& src1, const Src2Ptr& src2, const GlobPtr<DstType>& dst, const MaskPtr& mask, int rows, int cols, cudaStream_t stream)
+    {
+        const dim3 block(Policy::block_size_x, Policy::block_size_y);
+        const dim3 grid(divUp(cols, block.x), divUp(rows, block.y));
+
+        mergeC2<<<grid, block, 0, stream>>>(src1, src2, dst, mask, rows, cols);
+        CV_CUDEV_SAFE_CALL( cudaGetLastError() );
+
+        if (stream == 0)
+            CV_CUDEV_SAFE_CALL(cudaDeviceSynchronize());
+    }
+
+    template <class Src1Ptr, class Src2Ptr, class Src3Ptr, typename DstType, class MaskPtr>
+    __global__ void mergeC3(const Src1Ptr src1, const Src2Ptr src2, const Src3Ptr src3, GlobPtr<DstType> dst, const MaskPtr mask, const int rows, const int cols)
+    {
+        typedef typename VecTraits<DstType>::elem_type dst_elem_type;
+
+        const int x = blockIdx.x * blockDim.x + threadIdx.x;
+        const int y = blockIdx.y * blockDim.y + threadIdx.y;
+
+        if (x >= cols || y >= rows || !mask(y, x))
+            return;
+
+        dst(y, x) = VecTraits<DstType>::make(
+                    saturate_cast<dst_elem_type>(src1(y, x)),
+                    saturate_cast<dst_elem_type>(src2(y, x)),
+                    saturate_cast<dst_elem_type>(src3(y, x))
+                    );
+    }
+
+    template <class Policy, class Src1Ptr, class Src2Ptr, class Src3Ptr, typename DstType, class MaskPtr>
+    __host__ void mergeC3(const Src1Ptr& src1, const Src2Ptr& src2, const Src3Ptr& src3, const GlobPtr<DstType>& dst, const MaskPtr& mask, int rows, int cols, cudaStream_t stream)
+    {
+        const dim3 block(Policy::block_size_x, Policy::block_size_y);
+        const dim3 grid(divUp(cols, block.x), divUp(rows, block.y));
+
+        mergeC3<<<grid, block, 0, stream>>>(src1, src2, src3, dst, mask, rows, cols);
+        CV_CUDEV_SAFE_CALL( cudaGetLastError() );
+
+        if (stream == 0)
+            CV_CUDEV_SAFE_CALL(cudaDeviceSynchronize());
+    }
+
+    template <class Src1Ptr, class Src2Ptr, class Src3Ptr, class Src4Ptr, typename DstType, class MaskPtr>
+    __global__ void mergeC4(const Src1Ptr src1, const Src2Ptr src2, const Src3Ptr src3, const Src4Ptr src4, GlobPtr<DstType> dst, const MaskPtr mask, const int rows, const int cols)
+    {
+        typedef typename VecTraits<DstType>::elem_type dst_elem_type;
+
+        const int x = blockIdx.x * blockDim.x + threadIdx.x;
+        const int y = blockIdx.y * blockDim.y + threadIdx.y;
+
+        if (x >= cols || y >= rows || !mask(y, x))
+            return;
+
+        dst(y, x) = VecTraits<DstType>::make(
+                    saturate_cast<dst_elem_type>(src1(y, x)),
+                    saturate_cast<dst_elem_type>(src2(y, x)),
+                    saturate_cast<dst_elem_type>(src3(y, x)),
+                    saturate_cast<dst_elem_type>(src4(y, x))
+                    );
+    }
+
+    template <class Policy, class Src1Ptr, class Src2Ptr, class Src3Ptr, class Src4Ptr, typename DstType, class MaskPtr>
+    __host__ void mergeC4(const Src1Ptr& src1, const Src2Ptr& src2, const Src3Ptr& src3, const Src4Ptr& src4, const GlobPtr<DstType>& dst, const MaskPtr& mask, int rows, int cols, cudaStream_t stream)
+    {
+        const dim3 block(Policy::block_size_x, Policy::block_size_y);
+        const dim3 grid(divUp(cols, block.x), divUp(rows, block.y));
+
+        mergeC4<<<grid, block, 0, stream>>>(src1, src2, src3, src4, dst, mask, rows, cols);
+        CV_CUDEV_SAFE_CALL( cudaGetLastError() );
+
+        if (stream == 0)
+            CV_CUDEV_SAFE_CALL(cudaDeviceSynchronize());
+    }
+
+    template <int cn, class Policy> struct MergeImpl;
+
+    template <class Policy> struct MergeImpl<2, Policy>
+    {
+        template <class SrcPtrTuple, typename DstType, class MaskPtr>
+        __host__ static void merge(const SrcPtrTuple& src, const GlobPtr<DstType>& dst, const MaskPtr& mask, int rows, int cols, cudaStream_t stream)
+        {
+            mergeC2<Policy>(get<0>(src), get<1>(src), dst, mask, rows, cols, stream);
+        }
+    };
+
+    template <class Policy> struct MergeImpl<3, Policy>
+    {
+        template <class SrcPtrTuple, typename DstType, class MaskPtr>
+        __host__ static void merge(const SrcPtrTuple& src, const GlobPtr<DstType>& dst, const MaskPtr& mask, int rows, int cols, cudaStream_t stream)
+        {
+            mergeC3<Policy>(get<0>(src), get<1>(src), get<2>(src), dst, mask, rows, cols, stream);
+        }
+    };
+
+    template <class Policy> struct MergeImpl<4, Policy>
+    {
+        template <class SrcPtrTuple, typename DstType, class MaskPtr>
+        __host__ static void merge(const SrcPtrTuple& src, const GlobPtr<DstType>& dst, const MaskPtr& mask, int rows, int cols, cudaStream_t stream)
+        {
+            mergeC4<Policy>(get<0>(src), get<1>(src), get<2>(src), get<3>(src), dst, mask, rows, cols, stream);
+        }
+    };
+
+    // split
+
+    template <class SrcPtr, typename DstType, class MaskPtr>
+    __global__ void split(const SrcPtr src, GlobPtr<DstType> dst1, GlobPtr<DstType> dst2, const MaskPtr mask, const int rows, const int cols)
+    {
+        typedef typename PtrTraits<SrcPtr>::value_type src_type;
+
+        const int x = blockIdx.x * blockDim.x + threadIdx.x;
+        const int y = blockIdx.y * blockDim.y + threadIdx.y;
+
+        if (x >= cols || y >= rows || !mask(y, x))
+            return;
+
+        const src_type src_value = src(y, x);
+
+        dst1(y, x) = src_value.x;
+        dst2(y, x) = src_value.y;
+    }
+
+    template <class Policy, class SrcPtr, typename DstType, class MaskPtr>
+    __host__ void split(const SrcPtr& src, const GlobPtr<DstType>& dst1, const GlobPtr<DstType>& dst2, const MaskPtr& mask, int rows, int cols, cudaStream_t stream)
+    {
+        const dim3 block(Policy::block_size_x, Policy::block_size_y);
+        const dim3 grid(divUp(cols, block.x), divUp(rows, block.y));
+
+        split<<<grid, block, 0, stream>>>(src, dst1, dst2, mask, rows, cols);
+        CV_CUDEV_SAFE_CALL( cudaGetLastError() );
+
+        if (stream == 0)
+            CV_CUDEV_SAFE_CALL(cudaDeviceSynchronize());
+    }
+
+    template <class SrcPtr, typename DstType, class MaskPtr>
+    __global__ void split(const SrcPtr src, GlobPtr<DstType> dst1, GlobPtr<DstType> dst2, GlobPtr<DstType> dst3, const MaskPtr mask, const int rows, const int cols)
+    {
+        typedef typename PtrTraits<SrcPtr>::value_type src_type;
+
+        const int x = blockIdx.x * blockDim.x + threadIdx.x;
+        const int y = blockIdx.y * blockDim.y + threadIdx.y;
+
+        if (x >= cols || y >= rows || !mask(y, x))
+            return;
+
+        const src_type src_value = src(y, x);
+
+        dst1(y, x) = src_value.x;
+        dst2(y, x) = src_value.y;
+        dst3(y, x) = src_value.z;
+    }
+
+    template <class Policy, class SrcPtr, typename DstType, class MaskPtr>
+    __host__ void split(const SrcPtr& src, const GlobPtr<DstType>& dst1, const GlobPtr<DstType>& dst2, const GlobPtr<DstType>& dst3, const MaskPtr& mask, int rows, int cols, cudaStream_t stream)
+    {
+        const dim3 block(Policy::block_size_x, Policy::block_size_y);
+        const dim3 grid(divUp(cols, block.x), divUp(rows, block.y));
+
+        split<<<grid, block, 0, stream>>>(src, dst1, dst2, dst3, mask, rows, cols);
+        CV_CUDEV_SAFE_CALL( cudaGetLastError() );
+
+        if (stream == 0)
+            CV_CUDEV_SAFE_CALL(cudaDeviceSynchronize());
+    }
+
+    template <class SrcPtr, typename DstType, class MaskPtr>
+    __global__ void split(const SrcPtr src, GlobPtr<DstType> dst1, GlobPtr<DstType> dst2, GlobPtr<DstType> dst3, GlobPtr<DstType> dst4, const MaskPtr mask, const int rows, const int cols)
+    {
+        typedef typename PtrTraits<SrcPtr>::value_type src_type;
+
+        const int x = blockIdx.x * blockDim.x + threadIdx.x;
+        const int y = blockIdx.y * blockDim.y + threadIdx.y;
+
+        if (x >= cols || y >= rows || !mask(y, x))
+            return;
+
+        const src_type src_value = src(y, x);
+
+        dst1(y, x) = src_value.x;
+        dst2(y, x) = src_value.y;
+        dst3(y, x) = src_value.z;
+        dst4(y, x) = src_value.w;
+    }
+
+    template <class Policy, class SrcPtr, typename DstType, class MaskPtr>
+    __host__ void split(const SrcPtr& src, const GlobPtr<DstType>& dst1, const GlobPtr<DstType>& dst2, const GlobPtr<DstType>& dst3, const GlobPtr<DstType>& dst4, const MaskPtr& mask, int rows, int cols, cudaStream_t stream)
+    {
+        const dim3 block(Policy::block_size_x, Policy::block_size_y);
+        const dim3 grid(divUp(cols, block.x), divUp(rows, block.y));
+
+        split<<<grid, block, 0, stream>>>(src, dst1, dst2, dst3, dst4, mask, rows, cols);
+        CV_CUDEV_SAFE_CALL( cudaGetLastError() );
+
+        if (stream == 0)
+            CV_CUDEV_SAFE_CALL(cudaDeviceSynchronize());
+    }
+}
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/grid/detail/transform.hpp

@@ -0,0 +1,423 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_GRID_TRANSFORM_DETAIL_HPP__
+#define __OPENCV_CUDEV_GRID_TRANSFORM_DETAIL_HPP__
+
+#include "../../common.hpp"
+#include "../../util/tuple.hpp"
+#include "../../util/saturate_cast.hpp"
+#include "../../util/vec_traits.hpp"
+#include "../../ptr2d/glob.hpp"
+#include "../../ptr2d/traits.hpp"
+
+namespace cv { namespace cudev {
+
+namespace grid_transform_detail
+{
+    // OpUnroller
+
+    template <int cn> struct OpUnroller;
+
+    template <> struct OpUnroller<1>
+    {
+        template <typename T, typename D, class UnOp, class MaskPtr>
+        __device__ __forceinline__ static void unroll(const T& src, D& dst, const UnOp& op, const MaskPtr& mask, int x_shifted, int y)
+        {
+            if (mask(y, x_shifted))
+                dst.x = op(src.x);
+        }
+
+        template <typename T1, typename T2, typename D, class BinOp, class MaskPtr>
+        __device__ __forceinline__ static void unroll(const T1& src1, const T2& src2, D& dst, const BinOp& op, const MaskPtr& mask, int x_shifted, int y)
+        {
+            if (mask(y, x_shifted))
+                dst.x = op(src1.x, src2.x);
+        }
+    };
+
+    template <> struct OpUnroller<2>
+    {
+        template <typename T, typename D, class UnOp, class MaskPtr>
+        __device__ __forceinline__ static void unroll(const T& src, D& dst, const UnOp& op, const MaskPtr& mask, int x_shifted, int y)
+        {
+            if (mask(y, x_shifted))
+                dst.x = op(src.x);
+            if (mask(y, x_shifted + 1))
+                dst.y = op(src.y);
+        }
+
+        template <typename T1, typename T2, typename D, class BinOp, class MaskPtr>
+        __device__ __forceinline__ static void unroll(const T1& src1, const T2& src2, D& dst, const BinOp& op, const MaskPtr& mask, int x_shifted, int y)
+        {
+            if (mask(y, x_shifted))
+                dst.x = op(src1.x, src2.x);
+            if (mask(y, x_shifted + 1))
+                dst.y = op(src1.y, src2.y);
+        }
+    };
+
+    template <> struct OpUnroller<3>
+    {
+        template <typename T, typename D, class UnOp, class MaskPtr>
+        __device__ __forceinline__ static void unroll(const T& src, D& dst, const UnOp& op, const MaskPtr& mask, int x_shifted, int y)
+        {
+            if (mask(y, x_shifted))
+                dst.x = op(src.x);
+            if (mask(y, x_shifted + 1))
+                dst.y = op(src.y);
+            if (mask(y, x_shifted + 2))
+                dst.z = op(src.z);
+        }
+
+        template <typename T1, typename T2, typename D, class BinOp, class MaskPtr>
+        __device__ __forceinline__ static void unroll(const T1& src1, const T2& src2, D& dst, const BinOp& op, const MaskPtr& mask, int x_shifted, int y)
+        {
+            if (mask(y, x_shifted))
+                dst.x = op(src1.x, src2.x);
+            if (mask(y, x_shifted + 1))
+                dst.y = op(src1.y, src2.y);
+            if (mask(y, x_shifted + 2))
+                dst.z = op(src1.z, src2.z);
+        }
+    };
+
+    template <> struct OpUnroller<4>
+    {
+        template <typename T, typename D, class UnOp, class MaskPtr>
+        __device__ __forceinline__ static void unroll(const T& src, D& dst, const UnOp& op, const MaskPtr& mask, int x_shifted, int y)
+        {
+            if (mask(y, x_shifted))
+                dst.x = op(src.x);
+            if (mask(y, x_shifted + 1))
+                dst.y = op(src.y);
+            if (mask(y, x_shifted + 2))
+                dst.z = op(src.z);
+            if (mask(y, x_shifted + 3))
+                dst.w = op(src.w);
+        }
+
+        template <typename T1, typename T2, typename D, class BinOp, class MaskPtr>
+        __device__ __forceinline__ static void unroll(const T1& src1, const T2& src2, D& dst, const BinOp& op, const MaskPtr& mask, int x_shifted, int y)
+        {
+            if (mask(y, x_shifted))
+                dst.x = op(src1.x, src2.x);
+            if (mask(y, x_shifted + 1))
+                dst.y = op(src1.y, src2.y);
+            if (mask(y, x_shifted + 2))
+                dst.z = op(src1.z, src2.z);
+            if (mask(y, x_shifted + 3))
+                dst.w = op(src1.w, src2.w);
+        }
+    };
+
+    // transformSimple
+
+    template <class SrcPtr, typename DstType, class UnOp, class MaskPtr>
+    __global__ void transformSimple(const SrcPtr src, GlobPtr<DstType> dst, const UnOp op, const MaskPtr mask, const int rows, const int cols)
+    {
+        const int x = blockIdx.x * blockDim.x + threadIdx.x;
+        const int y = blockIdx.y * blockDim.y + threadIdx.y;
+
+        if (x >= cols || y >= rows || !mask(y, x))
+            return;
+
+        dst(y, x) = saturate_cast<DstType>(op(src(y, x)));
+    }
+
+    template <class SrcPtr1, class SrcPtr2, typename DstType, class BinOp, class MaskPtr>
+    __global__ void transformSimple(const SrcPtr1 src1, const SrcPtr2 src2, GlobPtr<DstType> dst, const BinOp op, const MaskPtr mask, const int rows, const int cols)
+    {
+        const int x = blockIdx.x * blockDim.x + threadIdx.x;
+        const int y = blockIdx.y * blockDim.y + threadIdx.y;
+
+        if (x >= cols || y >= rows || !mask(y, x))
+            return;
+
+        dst(y, x) = saturate_cast<DstType>(op(src1(y, x), src2(y, x)));
+    }
+
+    // transformSmart
+
+    template <int SHIFT, typename SrcType, typename DstType, class UnOp, class MaskPtr>
+    __global__ void transformSmart(const GlobPtr<SrcType> src_, GlobPtr<DstType> dst_, const UnOp op, const MaskPtr mask, const int rows, const int cols)
+    {
+        typedef typename MakeVec<SrcType, SHIFT>::type read_type;
+        typedef typename MakeVec<DstType, SHIFT>::type write_type;
+
+        const int x = blockIdx.x * blockDim.x + threadIdx.x;
+        const int y = blockIdx.y * blockDim.y + threadIdx.y;
+        const int x_shifted = x * SHIFT;
+
+        if (y < rows)
+        {
+            const SrcType* src = src_.row(y);
+            DstType* dst = dst_.row(y);
+
+            if (x_shifted + SHIFT - 1 < cols)
+            {
+                const read_type src_n_el = ((const read_type*)src)[x];
+                write_type dst_n_el = ((const write_type*)dst)[x];
+
+                OpUnroller<SHIFT>::unroll(src_n_el, dst_n_el, op, mask, x_shifted, y);
+
+                ((write_type*)dst)[x] = dst_n_el;
+            }
+            else
+            {
+                for (int real_x = x_shifted; real_x < cols; ++real_x)
+                {
+                    if (mask(y, real_x))
+                        dst[real_x] = op(src[real_x]);
+                }
+            }
+        }
+    }
+
+    template <int SHIFT, typename SrcType1, typename SrcType2, typename DstType, class BinOp, class MaskPtr>
+    __global__ void transformSmart(const GlobPtr<SrcType1> src1_, const GlobPtr<SrcType2> src2_, PtrStep<DstType> dst_, const BinOp op, const MaskPtr mask, const int rows, const int cols)
+    {
+        typedef typename MakeVec<SrcType1, SHIFT>::type read_type1;
+        typedef typename MakeVec<SrcType2, SHIFT>::type read_type2;
+        typedef typename MakeVec<DstType, SHIFT>::type write_type;
+
+        const int x = blockIdx.x * blockDim.x + threadIdx.x;
+        const int y = blockIdx.y * blockDim.y + threadIdx.y;
+        const int x_shifted = x * SHIFT;
+
+        if (y < rows)
+        {
+            const SrcType1* src1 = src1_.row(y);
+            const SrcType2* src2 = src2_.row(y);
+            DstType* dst = dst_.row(y);
+
+            if (x_shifted + SHIFT - 1 < cols)
+            {
+                const read_type1 src1_n_el = ((const read_type1*)src1)[x];
+                const read_type2 src2_n_el = ((const read_type2*)src2)[x];
+                write_type dst_n_el = ((const write_type*)dst)[x];
+
+                OpUnroller<SHIFT>::unroll(src1_n_el, src2_n_el, dst_n_el, op, mask, x_shifted, y);
+
+                ((write_type*)dst)[x] = dst_n_el;
+            }
+            else
+            {
+                for (int real_x = x_shifted; real_x < cols; ++real_x)
+                {
+                    if (mask(y, real_x))
+                        dst[real_x] = op(src1[real_x], src2[real_x]);
+                }
+            }
+        }
+    }
+
+    // TransformDispatcher
+
+    template <bool UseSmart, class Policy> struct TransformDispatcher;
+
+    template <class Policy> struct TransformDispatcher<false, Policy>
+    {
+        template <class SrcPtr, typename DstType, class UnOp, class MaskPtr>
+        __host__ static void call(const SrcPtr& src, const GlobPtr<DstType>& dst, const UnOp& op, const MaskPtr& mask, int rows, int cols, cudaStream_t stream)
+        {
+            const dim3 block(Policy::block_size_x, Policy::block_size_y);
+            const dim3 grid(divUp(cols, block.x), divUp(rows, block.y));
+
+            transformSimple<<<grid, block, 0, stream>>>(src, dst, op, mask, rows, cols);
+            CV_CUDEV_SAFE_CALL( cudaGetLastError() );
+
+            if (stream == 0)
+                CV_CUDEV_SAFE_CALL( cudaDeviceSynchronize() );
+        }
+
+        template <class SrcPtr1, class SrcPtr2, typename DstType, class BinOp, class MaskPtr>
+        __host__ static void call(const SrcPtr1& src1, const SrcPtr2& src2, const GlobPtr<DstType>& dst, const BinOp& op, const MaskPtr& mask, int rows, int cols, cudaStream_t stream)
+        {
+            const dim3 block(Policy::block_size_x, Policy::block_size_y);
+            const dim3 grid(divUp(cols, block.x), divUp(rows, block.y));
+
+            transformSimple<<<grid, block, 0, stream>>>(src1, src2, dst, op, mask, rows, cols);
+            CV_CUDEV_SAFE_CALL( cudaGetLastError() );
+
+            if (stream == 0)
+                CV_CUDEV_SAFE_CALL( cudaDeviceSynchronize() );
+        }
+    };
+
+    template <class Policy> struct TransformDispatcher<true, Policy>
+    {
+        template <typename T>
+        __host__ static bool isAligned(const T* ptr, size_t size)
+        {
+            return reinterpret_cast<size_t>(ptr) % size == 0;
+        }
+
+        __host__ static bool isAligned(size_t step, size_t size)
+        {
+            return step % size == 0;
+        }
+
+        template <typename SrcType, typename DstType, class UnOp, class MaskPtr>
+        __host__ static void call(const GlobPtr<SrcType>& src, const GlobPtr<DstType>& dst, const UnOp& op, const MaskPtr& mask, int rows, int cols, cudaStream_t stream)
+        {
+            if (Policy::shift == 1 ||
+                !isAligned(src.data, Policy::shift * sizeof(SrcType)) || !isAligned(src.step, Policy::shift * sizeof(SrcType)) ||
+                !isAligned(dst.data, Policy::shift * sizeof(DstType)) || !isAligned(dst.step, Policy::shift * sizeof(DstType)))
+            {
+                TransformDispatcher<false, Policy>::call(src, dst, op, mask, rows, cols, stream);
+                return;
+            }
+
+            const dim3 block(Policy::block_size_x, Policy::block_size_y);
+            const dim3 grid(divUp(cols, block.x * Policy::shift), divUp(rows, block.y));
+
+            transformSmart<Policy::shift><<<grid, block, 0, stream>>>(src, dst, op, mask, rows, cols);
+            CV_CUDEV_SAFE_CALL( cudaGetLastError() );
+
+            if (stream == 0)
+                CV_CUDEV_SAFE_CALL( cudaDeviceSynchronize() );
+        }
+
+        template <typename SrcType1, typename SrcType2, typename DstType, class BinOp, class MaskPtr>
+        __host__ static void call(const GlobPtr<SrcType1>& src1, const GlobPtr<SrcType2>& src2, const GlobPtr<DstType>& dst, const BinOp& op, const MaskPtr& mask, int rows, int cols, cudaStream_t stream)
+        {
+            if (Policy::shift == 1 ||
+                !isAligned(src1.data, Policy::shift * sizeof(SrcType1)) || !isAligned(src1.step, Policy::shift * sizeof(SrcType1)) ||
+                !isAligned(src2.data, Policy::shift * sizeof(SrcType2)) || !isAligned(src2.step, Policy::shift * sizeof(SrcType2)) ||
+                !isAligned(dst.data,  Policy::shift * sizeof(DstType))  || !isAligned(dst.step,  Policy::shift * sizeof(DstType)))
+            {
+                TransformDispatcher<false, Policy>::call(src1, src2, dst, op, mask, rows, cols, stream);
+                return;
+            }
+
+            const dim3 block(Policy::block_size_x, Policy::block_size_y);
+            const dim3 grid(divUp(cols, block.x * Policy::shift), divUp(rows, block.y));
+
+            transformSmart<Policy::shift><<<grid, block, 0, stream>>>(src1, src2, dst, op, mask, rows, cols);
+            CV_CUDEV_SAFE_CALL( cudaGetLastError() );
+
+            if (stream == 0)
+                CV_CUDEV_SAFE_CALL( cudaDeviceSynchronize() );
+        }
+    };
+
+    template <class Policy, class SrcPtr, typename DstType, class UnOp, class MaskPtr>
+    __host__ void transform(const SrcPtr& src, const GlobPtr<DstType>& dst, const UnOp& op, const MaskPtr& mask, int rows, int cols, cudaStream_t stream)
+    {
+        TransformDispatcher<false, Policy>::call(src, dst, op, mask, rows, cols, stream);
+    }
+
+    template <class Policy, class SrcPtr1, class SrcPtr2, typename DstType, class BinOp, class MaskPtr>
+    __host__ void transform(const SrcPtr1& src1, const SrcPtr2& src2, const GlobPtr<DstType>& dst, const BinOp& op, const MaskPtr& mask, int rows, int cols, cudaStream_t stream)
+    {
+        TransformDispatcher<false, Policy>::call(src1, src2, dst, op, mask, rows, cols, stream);
+    }
+
+    template <class Policy, typename SrcType, typename DstType, class UnOp, class MaskPtr>
+    __host__ void transform(const GlobPtr<SrcType>& src, const GlobPtr<DstType>& dst, const UnOp& op, const MaskPtr& mask, int rows, int cols, cudaStream_t stream)
+    {
+        TransformDispatcher<VecTraits<SrcType>::cn == 1 && VecTraits<DstType>::cn == 1 && Policy::shift != 1, Policy>::call(src, dst, op, mask, rows, cols, stream);
+    }
+
+    template <class Policy, typename SrcType1, typename SrcType2, typename DstType, class BinOp, class MaskPtr>
+    __host__ void transform(const GlobPtr<SrcType1>& src1, const GlobPtr<SrcType2>& src2, const GlobPtr<DstType>& dst, const BinOp& op, const MaskPtr& mask, int rows, int cols, cudaStream_t stream)
+    {
+        TransformDispatcher<VecTraits<SrcType1>::cn == 1 && VecTraits<SrcType2>::cn == 1 && VecTraits<DstType>::cn == 1 && Policy::shift != 1, Policy>::call(src1, src2, dst, op, mask, rows, cols, stream);
+    }
+
+    // transform_tuple
+
+    template <int count> struct Unroll
+    {
+        template <class SrcVal, class DstPtrTuple, class OpTuple>
+        __device__ static void transform(const SrcVal& srcVal, DstPtrTuple& dst, const OpTuple& op, int y, int x)
+        {
+            typedef typename tuple_element<count - 1, DstPtrTuple>::type dst_ptr_type;
+            typedef typename PtrTraits<dst_ptr_type>::value_type dst_type;
+
+            get<count - 1>(dst)(y, x) = saturate_cast<dst_type>(get<count - 1>(op)(srcVal));
+            Unroll<count - 1>::transform(srcVal, dst, op, y, x);
+        }
+    };
+    template <> struct Unroll<0>
+    {
+        template <class SrcVal, class DstPtrTuple, class OpTuple>
+        __device__ __forceinline__ static void transform(const SrcVal&, DstPtrTuple&, const OpTuple&, int, int)
+        {
+        }
+    };
+
+    template <class SrcPtr, class DstPtrTuple, class OpTuple, class MaskPtr>
+    __global__ void transform_tuple(const SrcPtr src, DstPtrTuple dst, const OpTuple op, const MaskPtr mask, const int rows, const int cols)
+    {
+        const int x = blockIdx.x * blockDim.x + threadIdx.x;
+        const int y = blockIdx.y * blockDim.y + threadIdx.y;
+
+        if (x >= cols || y >= rows || !mask(y, x))
+            return;
+
+        typename PtrTraits<SrcPtr>::value_type srcVal = src(y, x);
+
+        Unroll<tuple_size<DstPtrTuple>::value>::transform(srcVal, dst, op, y, x);
+    }
+
+    template <class Policy, class SrcPtrTuple, class DstPtrTuple, class OpTuple, class MaskPtr>
+    __host__ void transform_tuple(const SrcPtrTuple& src, const DstPtrTuple& dst, const OpTuple& op, const MaskPtr& mask, int rows, int cols, cudaStream_t stream)
+    {
+        const dim3 block(Policy::block_size_x, Policy::block_size_y);
+        const dim3 grid(divUp(cols, block.x), divUp(rows, block.y));
+
+        transform_tuple<<<grid, block, 0, stream>>>(src, dst, op, mask, rows, cols);
+        CV_CUDEV_SAFE_CALL( cudaGetLastError() );
+
+        if (stream == 0)
+            CV_CUDEV_SAFE_CALL( cudaDeviceSynchronize() );
+    }
+}
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/grid/detail/transpose.hpp

@@ -0,0 +1,130 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_GRID_TRANSPOSE_DETAIL_HPP__
+#define __OPENCV_CUDEV_GRID_TRANSPOSE_DETAIL_HPP__
+
+#include "../../common.hpp"
+#include "../../util/saturate_cast.hpp"
+#include "../../ptr2d/glob.hpp"
+#include "../../ptr2d/traits.hpp"
+
+namespace cv { namespace cudev {
+
+namespace transpose_detail
+{
+    const int TRANSPOSE_TILE_DIM   = 16;
+    const int TRANSPOSE_BLOCK_ROWS = 16;
+
+    template <class SrcPtr, typename DstType>
+    __global__ void transpose(const SrcPtr src, GlobPtr<DstType> dst, const int rows, const int cols)
+    {
+        typedef typename PtrTraits<SrcPtr>::value_type src_type;
+
+        __shared__ src_type tile[TRANSPOSE_TILE_DIM][TRANSPOSE_TILE_DIM + 1];
+
+        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 < cols)
+        {
+            for (int i = 0; i < TRANSPOSE_TILE_DIM; i += TRANSPOSE_BLOCK_ROWS)
+            {
+                if (yIndex + i < 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 < rows)
+        {
+            for (int i = 0; i < TRANSPOSE_TILE_DIM; i += TRANSPOSE_BLOCK_ROWS)
+            {
+                if (yIndex + i < cols)
+                {
+                    dst(yIndex + i, xIndex) = saturate_cast<DstType>(tile[threadIdx.x][threadIdx.y + i]);
+                }
+            }
+        }
+    }
+
+    template <class SrcPtr, typename DstType>
+    __host__ void transpose(const SrcPtr& src, const GlobPtr<DstType>& dst, int rows, int cols, cudaStream_t stream)
+    {
+        const dim3 block(TRANSPOSE_TILE_DIM, TRANSPOSE_TILE_DIM);
+        const dim3 grid(divUp(cols, block.x), divUp(rows, block.y));
+
+        transpose<<<grid, block, 0, stream>>>(src, dst, rows, cols);
+        CV_CUDEV_SAFE_CALL( cudaGetLastError() );
+
+        if (stream == 0)
+            CV_CUDEV_SAFE_CALL( cudaDeviceSynchronize() );
+    }
+}
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/grid/glob_reduce.hpp

@@ -0,0 +1,314 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_GRID_GLOB_REDUCE_HPP__
+#define __OPENCV_CUDEV_GRID_GLOB_REDUCE_HPP__
+
+#include <limits>
+#include "../common.hpp"
+#include "../ptr2d/traits.hpp"
+#include "../ptr2d/gpumat.hpp"
+#include "../ptr2d/mask.hpp"
+#include "../ptr2d/transform.hpp"
+#include "detail/glob_reduce.hpp"
+
+namespace cv { namespace cudev {
+
+template <class Policy, class SrcPtr, typename ResType, class MaskPtr>
+__host__ void gridCalcSum_(const SrcPtr& src, GpuMat_<ResType>& dst, const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    dst.create(1, 1);
+    dst.setTo(0, stream);
+
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    CV_Assert( getRows(mask) == rows && getCols(mask) == cols );
+
+    grid_glob_reduce_detail::sum<Policy>(shrinkPtr(src),
+                                         dst[0],
+                                         shrinkPtr(mask),
+                                         rows, cols,
+                                         StreamAccessor::getStream(stream));
+}
+
+template <class Policy, class SrcPtr, typename ResType>
+__host__ void gridCalcSum_(const SrcPtr& src, GpuMat_<ResType>& dst, Stream& stream = Stream::Null())
+{
+    dst.create(1, 1);
+    dst.setTo(0, stream);
+
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    grid_glob_reduce_detail::sum<Policy>(shrinkPtr(src),
+                                         dst[0],
+                                         WithOutMask(),
+                                         rows, cols,
+                                         StreamAccessor::getStream(stream));
+}
+
+template <class Policy, class SrcPtr, typename ResType, class MaskPtr>
+__host__ void gridFindMinVal_(const SrcPtr& src, GpuMat_<ResType>& dst, const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    dst.create(1, 1);
+    dst.setTo(Scalar::all(std::numeric_limits<ResType>::max()), stream);
+
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    CV_Assert( getRows(mask) == rows && getCols(mask) == cols );
+
+    grid_glob_reduce_detail::minVal<Policy>(shrinkPtr(src),
+                                            dst[0],
+                                            shrinkPtr(mask),
+                                            rows, cols,
+                                            StreamAccessor::getStream(stream));
+}
+
+template <class Policy, class SrcPtr, typename ResType>
+__host__ void gridFindMinVal_(const SrcPtr& src, GpuMat_<ResType>& dst, Stream& stream = Stream::Null())
+{
+    dst.create(1, 1);
+    dst.setTo(Scalar::all(std::numeric_limits<ResType>::max()), stream);
+
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    grid_glob_reduce_detail::minVal<Policy>(shrinkPtr(src),
+                                            dst[0],
+                                            WithOutMask(),
+                                            rows, cols,
+                                            StreamAccessor::getStream(stream));
+}
+
+template <class Policy, class SrcPtr, typename ResType, class MaskPtr>
+__host__ void gridFindMaxVal_(const SrcPtr& src, GpuMat_<ResType>& dst, const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    dst.create(1, 1);
+    dst.setTo(Scalar::all(-std::numeric_limits<ResType>::max()), stream);
+
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    CV_Assert( getRows(mask) == rows && getCols(mask) == cols );
+
+    grid_glob_reduce_detail::maxVal<Policy>(shrinkPtr(src),
+                                            dst[0],
+                                            shrinkPtr(mask),
+                                            rows, cols,
+                                            StreamAccessor::getStream(stream));
+}
+
+template <class Policy, class SrcPtr, typename ResType>
+__host__ void gridFindMaxVal_(const SrcPtr& src, GpuMat_<ResType>& dst, Stream& stream = Stream::Null())
+{
+    dst.create(1, 1);
+    dst.setTo(Scalar::all(-std::numeric_limits<ResType>::max()), stream);
+
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    grid_glob_reduce_detail::maxVal<Policy>(shrinkPtr(src),
+                                            dst[0],
+                                            WithOutMask(),
+                                            rows, cols,
+                                            StreamAccessor::getStream(stream));
+}
+
+template <class Policy, class SrcPtr, typename ResType, class MaskPtr>
+__host__ void gridFindMinMaxVal_(const SrcPtr& src, GpuMat_<ResType>& dst, const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    dst.create(1, 2);
+    dst.col(0).setTo(Scalar::all(std::numeric_limits<ResType>::max()), stream);
+    dst.col(1).setTo(Scalar::all(-std::numeric_limits<ResType>::max()), stream);
+
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    CV_Assert( getRows(mask) == rows && getCols(mask) == cols );
+
+    grid_glob_reduce_detail::minMaxVal<Policy>(shrinkPtr(src),
+                                               dst[0],
+                                               shrinkPtr(mask),
+                                               rows, cols,
+                                               StreamAccessor::getStream(stream));
+}
+
+template <class Policy, class SrcPtr, typename ResType>
+__host__ void gridFindMinMaxVal_(const SrcPtr& src, GpuMat_<ResType>& dst, Stream& stream = Stream::Null())
+{
+    dst.create(1, 2);
+    dst.col(0).setTo(Scalar::all(std::numeric_limits<ResType>::max()), stream);
+    dst.col(1).setTo(Scalar::all(-std::numeric_limits<ResType>::max()), stream);
+
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    grid_glob_reduce_detail::minMaxVal<Policy>(shrinkPtr(src),
+                                               dst[0],
+                                               WithOutMask(),
+                                               rows, cols,
+                                               StreamAccessor::getStream(stream));
+}
+
+template <class Policy, class SrcPtr, typename ResType, class MaskPtr>
+__host__ void gridCountNonZero_(const SrcPtr& src, GpuMat_<ResType>& dst, const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    dst.create(1, 1);
+    dst.setTo(0, stream);
+
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    CV_Assert( getRows(mask) == rows && getCols(mask) == cols );
+
+    typedef typename PtrTraits<SrcPtr>::value_type src_type;
+    not_equal_to<src_type> ne_op;
+    const src_type zero = VecTraits<src_type>::all(0);
+
+    grid_glob_reduce_detail::sum<Policy>(shrinkPtr(transformPtr(src, bind2nd(ne_op, zero))),
+                                         dst[0],
+                                         shrinkPtr(mask),
+                                         rows, cols,
+                                         StreamAccessor::getStream(stream));
+}
+
+template <class Policy, class SrcPtr, typename ResType>
+__host__ void gridCountNonZero_(const SrcPtr& src, GpuMat_<ResType>& dst, Stream& stream = Stream::Null())
+{
+    dst.create(1, 1);
+    dst.setTo(0, stream);
+
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    typedef typename PtrTraits<SrcPtr>::value_type src_type;
+    not_equal_to<src_type> ne_op;
+    const src_type zero = VecTraits<src_type>::all(0);
+
+    grid_glob_reduce_detail::sum<Policy>(shrinkPtr(transformPtr(src, bind2nd(ne_op, zero))),
+                                         dst[0],
+                                         WithOutMask(),
+                                         rows, cols,
+                                         StreamAccessor::getStream(stream));
+}
+
+// default policy
+
+struct DefaultGlobReducePolicy
+{
+    enum {
+        block_size_x = 32,
+        block_size_y = 8,
+
+        patch_size_x = 4,
+        patch_size_y = 4
+    };
+};
+
+template <class SrcPtr, typename ResType, class MaskPtr>
+__host__ void gridCalcSum(const SrcPtr& src, GpuMat_<ResType>& dst, const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    gridCalcSum_<DefaultGlobReducePolicy>(src, dst, mask, stream);
+}
+
+template <class SrcPtr, typename ResType>
+__host__ void gridCalcSum(const SrcPtr& src, GpuMat_<ResType>& dst, Stream& stream = Stream::Null())
+{
+    gridCalcSum_<DefaultGlobReducePolicy>(src, dst, stream);
+}
+
+template <class SrcPtr, typename ResType, class MaskPtr>
+__host__ void gridFindMinVal(const SrcPtr& src, GpuMat_<ResType>& dst, const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    gridFindMinVal_<DefaultGlobReducePolicy>(src, dst, mask, stream);
+}
+
+template <class SrcPtr, typename ResType>
+__host__ void gridFindMinVal(const SrcPtr& src, GpuMat_<ResType>& dst, Stream& stream = Stream::Null())
+{
+    gridFindMinVal_<DefaultGlobReducePolicy>(src, dst, stream);
+}
+
+template <class SrcPtr, typename ResType, class MaskPtr>
+__host__ void gridFindMaxVal(const SrcPtr& src, GpuMat_<ResType>& dst, const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    gridFindMaxVal_<DefaultGlobReducePolicy>(src, dst, mask, stream);
+}
+
+template <class SrcPtr, typename ResType>
+__host__ void gridFindMaxVal(const SrcPtr& src, GpuMat_<ResType>& dst, Stream& stream = Stream::Null())
+{
+    gridFindMaxVal_<DefaultGlobReducePolicy>(src, dst, stream);
+}
+
+template <class SrcPtr, typename ResType, class MaskPtr>
+__host__ void gridFindMinMaxVal(const SrcPtr& src, GpuMat_<ResType>& dst, const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    gridFindMinMaxVal_<DefaultGlobReducePolicy>(src, dst, mask, stream);
+}
+
+template <class SrcPtr, typename ResType>
+__host__ void gridFindMinMaxVal(const SrcPtr& src, GpuMat_<ResType>& dst, Stream& stream = Stream::Null())
+{
+    gridFindMinMaxVal_<DefaultGlobReducePolicy>(src, dst, stream);
+}
+
+template <class SrcPtr, typename ResType, class MaskPtr>
+__host__ void gridCountNonZero(const SrcPtr& src, GpuMat_<ResType>& dst, const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    gridCountNonZero_<DefaultGlobReducePolicy>(src, dst, mask, stream);
+}
+
+template <class SrcPtr, typename ResType>
+__host__ void gridCountNonZero(const SrcPtr& src, GpuMat_<ResType>& dst, Stream& stream = Stream::Null())
+{
+    gridCountNonZero_<DefaultGlobReducePolicy>(src, dst, stream);
+}
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/grid/histogram.hpp

@@ -0,0 +1,115 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_GRID_HISTOGRAM_HPP__
+#define __OPENCV_CUDEV_GRID_HISTOGRAM_HPP__
+
+#include "../common.hpp"
+#include "../ptr2d/traits.hpp"
+#include "../ptr2d/gpumat.hpp"
+#include "../ptr2d/mask.hpp"
+#include "detail/histogram.hpp"
+
+namespace cv { namespace cudev {
+
+template <int BIN_COUNT, class Policy, class SrcPtr, typename ResType, class MaskPtr>
+__host__ void gridHistogram_(const SrcPtr& src, GpuMat_<ResType>& dst, const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    CV_Assert( getRows(mask) == rows && getCols(mask) == cols );
+
+    dst.create(1, BIN_COUNT);
+    dst.setTo(0, stream);
+
+    grid_histogram_detail::histogram<BIN_COUNT, Policy>(shrinkPtr(src),
+                                                        dst[0],
+                                                        shrinkPtr(mask),
+                                                        rows, cols,
+                                                        StreamAccessor::getStream(stream));
+}
+
+template <int BIN_COUNT, class Policy, class SrcPtr, typename ResType>
+__host__ void gridHistogram_(const SrcPtr& src, GpuMat_<ResType>& dst, Stream& stream = Stream::Null())
+{
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    dst.create(1, BIN_COUNT);
+    dst.setTo(0, stream);
+
+    grid_histogram_detail::histogram<BIN_COUNT, Policy>(shrinkPtr(src),
+                                                        dst[0],
+                                                        WithOutMask(),
+                                                        rows, cols,
+                                                        StreamAccessor::getStream(stream));
+}
+
+// default policy
+
+struct DefaultHistogramPolicy
+{
+    enum {
+        block_size_x = 32,
+        block_size_y = 8
+    };
+};
+
+template <int BIN_COUNT, class SrcPtr, typename ResType, class MaskPtr>
+__host__ void gridHistogram(const SrcPtr& src, GpuMat_<ResType>& dst, const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    gridHistogram_<BIN_COUNT, DefaultHistogramPolicy>(src, dst, mask, stream);
+}
+
+template <int BIN_COUNT, class SrcPtr, typename ResType>
+__host__ void gridHistogram(const SrcPtr& src, GpuMat_<ResType>& dst, Stream& stream = Stream::Null())
+{
+    gridHistogram_<BIN_COUNT, DefaultHistogramPolicy>(src, dst, stream);
+}
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/grid/integral.hpp

@@ -0,0 +1,69 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_GRID_INTEGRAL_HPP__
+#define __OPENCV_CUDEV_GRID_INTEGRAL_HPP__
+
+#include "../common.hpp"
+#include "../ptr2d/traits.hpp"
+#include "../ptr2d/gpumat.hpp"
+#include "detail/integral.hpp"
+
+namespace cv { namespace cudev {
+
+template <class SrcPtr, typename DstType>
+__host__ void gridIntegral(const SrcPtr& src, GpuMat_<DstType>& dst, Stream& stream = Stream::Null())
+{
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    dst.create(rows, cols);
+
+    integral_detail::integral(shrinkPtr(src), shrinkPtr(dst), rows, cols, StreamAccessor::getStream(stream));
+}
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/grid/pyramids.hpp

@@ -0,0 +1,88 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_GRID_PYRAMIDS_HPP__
+#define __OPENCV_CUDEV_GRID_PYRAMIDS_HPP__
+
+#include "../common.hpp"
+#include "../ptr2d/traits.hpp"
+#include "../ptr2d/gpumat.hpp"
+#include "../ptr2d/extrapolation.hpp"
+#include "detail/pyr_down.hpp"
+#include "detail/pyr_up.hpp"
+
+namespace cv { namespace cudev {
+
+template <class Brd, class SrcPtr, typename DstType>
+__host__ void gridPyrDown_(const SrcPtr& src, GpuMat_<DstType>& dst, Stream& stream = Stream::Null())
+{
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    dst.create(divUp(rows, 2), divUp(cols, 2));
+
+    pyramids_detail::pyrDown<Brd>(shrinkPtr(src), shrinkPtr(dst), rows, cols, dst.rows, dst.cols, StreamAccessor::getStream(stream));
+}
+
+template <class SrcPtr, typename DstType>
+__host__ void gridPyrDown(const SrcPtr& src, GpuMat_<DstType>& dst, Stream& stream = Stream::Null())
+{
+    gridPyrDown_<BrdReflect101>(src, dst, stream);
+}
+
+template <class SrcPtr, typename DstType>
+__host__ void gridPyrUp(const SrcPtr& src, GpuMat_<DstType>& dst, Stream& stream = Stream::Null())
+{
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    dst.create(rows * 2, cols * 2);
+
+    pyramids_detail::pyrUp(shrinkPtr(src), shrinkPtr(dst), rows, cols, dst.rows, dst.cols, StreamAccessor::getStream(stream));
+}
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/grid/reduce_to_vec.hpp

@@ -0,0 +1,209 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_GRID_REDUCE_TO_VEC_HPP__
+#define __OPENCV_CUDEV_GRID_REDUCE_TO_VEC_HPP__
+
+#include "../common.hpp"
+#include "../util/vec_traits.hpp"
+#include "../util/limits.hpp"
+#include "../ptr2d/traits.hpp"
+#include "../ptr2d/gpumat.hpp"
+#include "../ptr2d/mask.hpp"
+#include "../functional/functional.hpp"
+#include "detail/reduce_to_column.hpp"
+#include "detail/reduce_to_row.hpp"
+
+namespace cv { namespace cudev {
+
+template <typename T> struct Sum : plus<T>
+{
+    typedef T work_type;
+
+    __device__ __forceinline__ static T initialValue()
+    {
+        return VecTraits<T>::all(0);
+    }
+
+    __device__ __forceinline__ static T result(T r, int)
+    {
+        return r;
+    }
+};
+
+template <typename T> struct Avg : plus<T>
+{
+    typedef T work_type;
+
+    __device__ __forceinline__ static T initialValue()
+    {
+        return VecTraits<T>::all(0);
+    }
+
+    __device__ __forceinline__ static T result(T r, int sz)
+    {
+        return r / sz;
+    }
+};
+
+template <typename T> struct Min : minimum<T>
+{
+    typedef T work_type;
+
+    __device__ __forceinline__ static T initialValue()
+    {
+        return VecTraits<T>::all(numeric_limits<typename VecTraits<T>::elem_type>::max());
+    }
+
+    __device__ __forceinline__ static T result(T r, int)
+    {
+        return r;
+    }
+};
+
+template <typename T> struct Max : maximum<T>
+{
+    typedef T work_type;
+
+    __device__ __forceinline__ static T initialValue()
+    {
+        return VecTraits<T>::all(-numeric_limits<typename VecTraits<T>::elem_type>::max());
+    }
+
+    __device__ __forceinline__ static T result(T r, int)
+    {
+        return r;
+    }
+};
+
+template <class Reductor, class SrcPtr, typename ResType, class MaskPtr>
+__host__ void gridReduceToRow(const SrcPtr& src, GpuMat_<ResType>& dst, const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    CV_Assert( getRows(mask) == rows && getCols(mask) == cols );
+
+    dst.create(1, cols);
+
+    grid_reduce_to_vec_detail::reduceToRow<Reductor>(shrinkPtr(src),
+                                                     dst[0],
+                                                     shrinkPtr(mask),
+                                                     rows, cols,
+                                                     StreamAccessor::getStream(stream));
+}
+
+template <class Reductor, class SrcPtr, typename ResType>
+__host__ void gridReduceToRow(const SrcPtr& src, GpuMat_<ResType>& dst, Stream& stream = Stream::Null())
+{
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    dst.create(1, cols);
+
+    grid_reduce_to_vec_detail::reduceToRow<Reductor>(shrinkPtr(src),
+                                                     dst[0],
+                                                     WithOutMask(),
+                                                     rows, cols,
+                                                     StreamAccessor::getStream(stream));
+}
+
+template <class Reductor, class Policy, class SrcPtr, typename ResType, class MaskPtr>
+__host__ void gridReduceToColumn_(const SrcPtr& src, GpuMat_<ResType>& dst, const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    CV_Assert( getRows(mask) == rows && getCols(mask) == cols );
+
+    createContinuous(rows, 1, DataType<ResType>::type, dst);
+
+    grid_reduce_to_vec_detail::reduceToColumn<Reductor, Policy>(shrinkPtr(src),
+                                                                dst[0],
+                                                                shrinkPtr(mask),
+                                                                rows, cols,
+                                                                StreamAccessor::getStream(stream));
+}
+
+template <class Reductor, class Policy, class SrcPtr, typename ResType>
+__host__ void gridReduceToColumn_(const SrcPtr& src, GpuMat_<ResType>& dst, Stream& stream = Stream::Null())
+{
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    createContinuous(rows, 1, DataType<ResType>::type, dst);
+
+    grid_reduce_to_vec_detail::reduceToColumn<Reductor, Policy>(shrinkPtr(src),
+                                                                dst[0],
+                                                                WithOutMask(),
+                                                                rows, cols,
+                                                                StreamAccessor::getStream(stream));
+}
+
+// default policy
+
+struct DefaultReduceToVecPolicy
+{
+    enum {
+        block_size_x = 32,
+        block_size_y = 8
+    };
+};
+
+template <class Reductor, class SrcPtr, typename ResType, class MaskPtr>
+__host__ void gridReduceToColumn(const SrcPtr& src, GpuMat_<ResType>& dst, const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    gridReduceToColumn_<Reductor, DefaultReduceToVecPolicy>(src, dst, mask, stream);
+}
+
+template <class Reductor, class SrcPtr, typename ResType>
+__host__ void gridReduceToColumn(const SrcPtr& src, GpuMat_<ResType>& dst, Stream& stream = Stream::Null())
+{
+    gridReduceToColumn_<Reductor, DefaultReduceToVecPolicy>(src, dst, stream);
+}
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/grid/split_merge.hpp

@@ -0,0 +1,407 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_GRID_SPLIT_MERGE_HPP__
+#define __OPENCV_CUDEV_GRID_SPLIT_MERGE_HPP__
+
+#include "../common.hpp"
+#include "../util/tuple.hpp"
+#include "../util/vec_traits.hpp"
+#include "../ptr2d/traits.hpp"
+#include "../ptr2d/gpumat.hpp"
+#include "../ptr2d/mask.hpp"
+#include "detail/split_merge.hpp"
+
+namespace cv { namespace cudev {
+
+template <class Policy, class SrcPtrTuple, typename DstType, class MaskPtr>
+__host__ void gridMerge_(const SrcPtrTuple& src, GpuMat_<DstType>& dst, const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    CV_StaticAssert( VecTraits<DstType>::cn == tuple_size<SrcPtrTuple>::value, "" );
+
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    CV_Assert( getRows(mask) == rows && getCols(mask) == cols );
+
+    dst.create(rows, cols);
+
+    grid_split_merge_detail::MergeImpl<VecTraits<DstType>::cn, Policy>::merge(shrinkPtr(src),
+                                                                              shrinkPtr(dst),
+                                                                              shrinkPtr(mask),
+                                                                              rows, cols,
+                                                                              StreamAccessor::getStream(stream));
+}
+
+template <class Policy, class SrcPtrTuple, typename DstType>
+__host__ void gridMerge_(const SrcPtrTuple& src, GpuMat_<DstType>& dst, Stream& stream = Stream::Null())
+{
+    CV_StaticAssert( VecTraits<DstType>::cn == tuple_size<SrcPtrTuple>::value, "" );
+
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    dst.create(rows, cols);
+
+    grid_split_merge_detail::MergeImpl<VecTraits<DstType>::cn, Policy>::merge(shrinkPtr(src),
+                                                                              shrinkPtr(dst),
+                                                                              WithOutMask(),
+                                                                              rows, cols,
+                                                                              StreamAccessor::getStream(stream));
+}
+
+template <class Policy, class SrcPtr, typename DstType, class MaskPtr>
+__host__ void gridSplit_(const SrcPtr& src, const tuple< GpuMat_<DstType>&, GpuMat_<DstType>& >& dst, const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    CV_StaticAssert( VecTraits<typename PtrTraits<SrcPtr>::value_type>::cn == 2, "" );
+
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    CV_Assert( getRows(mask) == rows && getCols(mask) == cols );
+
+    get<0>(dst).create(rows, cols);
+    get<1>(dst).create(rows, cols);
+
+    grid_split_merge_detail::split<Policy>(shrinkPtr(src),
+                                           shrinkPtr(get<0>(dst)), shrinkPtr(get<1>(dst)),
+                                           shrinkPtr(mask),
+                                           rows, cols,
+                                           StreamAccessor::getStream(stream));
+}
+
+template <class Policy, class SrcPtr, typename DstType, class MaskPtr>
+__host__ void gridSplit_(const SrcPtr& src, GpuMat_<DstType> (&dst)[2], const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    CV_StaticAssert( VecTraits<typename PtrTraits<SrcPtr>::value_type>::cn == 2, "" );
+
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    CV_Assert( getRows(mask) == rows && getCols(mask) == cols );
+
+    dst[0].create(rows, cols);
+    dst[1].create(rows, cols);
+
+    grid_split_merge_detail::split<Policy>(shrinkPtr(src),
+                                           shrinkPtr(dst[0]), shrinkPtr(dst[1]),
+                                           shrinkPtr(mask),
+                                           rows, cols,
+                                           StreamAccessor::getStream(stream));
+}
+
+template <class Policy, class SrcPtr, typename DstType>
+__host__ void gridSplit_(const SrcPtr& src, const tuple< GpuMat_<DstType>&, GpuMat_<DstType>& >& dst, Stream& stream = Stream::Null())
+{
+    CV_StaticAssert( VecTraits<typename PtrTraits<SrcPtr>::value_type>::cn == 2, "" );
+
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    get<0>(dst).create(rows, cols);
+    get<1>(dst).create(rows, cols);
+
+    grid_split_merge_detail::split<Policy>(shrinkPtr(src),
+                                           shrinkPtr(get<0>(dst)), shrinkPtr(get<1>(dst)),
+                                           WithOutMask(),
+                                           rows, cols,
+                                           StreamAccessor::getStream(stream));
+}
+
+template <class Policy, class SrcPtr, typename DstType>
+__host__ void gridSplit_(const SrcPtr& src, GpuMat_<DstType> (&dst)[2], Stream& stream = Stream::Null())
+{
+    CV_StaticAssert( VecTraits<typename PtrTraits<SrcPtr>::value_type>::cn == 2, "" );
+
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    dst[0].create(rows, cols);
+    dst[1].create(rows, cols);
+
+    grid_split_merge_detail::split<Policy>(shrinkPtr(src),
+                                           shrinkPtr(dst[0]), shrinkPtr(dst[1]),
+                                           WithOutMask(),
+                                           rows, cols,
+                                           StreamAccessor::getStream(stream));
+}
+
+template <class Policy, class SrcPtr, typename DstType, class MaskPtr>
+__host__ void gridSplit_(const SrcPtr& src, const tuple< GpuMat_<DstType>&, GpuMat_<DstType>&, GpuMat_<DstType>& >& dst, const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    CV_StaticAssert( VecTraits<typename PtrTraits<SrcPtr>::value_type>::cn == 3, "" );
+
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    CV_Assert( getRows(mask) == rows && getCols(mask) == cols );
+
+    get<0>(dst).create(rows, cols);
+    get<1>(dst).create(rows, cols);
+    get<2>(dst).create(rows, cols);
+
+    grid_split_merge_detail::split<Policy>(shrinkPtr(src),
+                                           shrinkPtr(get<0>(dst)), shrinkPtr(get<1>(dst)), shrinkPtr(get<2>(dst)),
+                                           shrinkPtr(mask),
+                                           rows, cols,
+                                           StreamAccessor::getStream(stream));
+}
+
+template <class Policy, class SrcPtr, typename DstType, class MaskPtr>
+__host__ void gridSplit_(const SrcPtr& src, GpuMat_<DstType> (&dst)[3], const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    CV_StaticAssert( VecTraits<typename PtrTraits<SrcPtr>::value_type>::cn == 3, "" );
+
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    CV_Assert( getRows(mask) == rows && getCols(mask) == cols );
+
+    dst[0].create(rows, cols);
+    dst[1].create(rows, cols);
+    dst[2].create(rows, cols);
+
+    grid_split_merge_detail::split<Policy>(shrinkPtr(src),
+                                           shrinkPtr(dst[0]), shrinkPtr(dst[1]), shrinkPtr(dst[2]),
+                                           shrinkPtr(mask),
+                                           rows, cols,
+                                           StreamAccessor::getStream(stream));
+}
+
+template <class Policy, class SrcPtr, typename DstType>
+__host__ void gridSplit_(const SrcPtr& src, const tuple< GpuMat_<DstType>&, GpuMat_<DstType>&, GpuMat_<DstType>& >& dst, Stream& stream = Stream::Null())
+{
+    CV_StaticAssert( VecTraits<typename PtrTraits<SrcPtr>::value_type>::cn == 3, "" );
+
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    get<0>(dst).create(rows, cols);
+    get<1>(dst).create(rows, cols);
+    get<2>(dst).create(rows, cols);
+
+    grid_split_merge_detail::split<Policy>(shrinkPtr(src),
+                                           shrinkPtr(get<0>(dst)), shrinkPtr(get<1>(dst)), shrinkPtr(get<2>(dst)),
+                                           WithOutMask(),
+                                           rows, cols,
+                                           StreamAccessor::getStream(stream));
+}
+
+template <class Policy, class SrcPtr, typename DstType>
+__host__ void gridSplit_(const SrcPtr& src, GpuMat_<DstType> (&dst)[3], Stream& stream = Stream::Null())
+{
+    CV_StaticAssert( VecTraits<typename PtrTraits<SrcPtr>::value_type>::cn == 3, "" );
+
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    dst[0].create(rows, cols);
+    dst[1].create(rows, cols);
+    dst[2].create(rows, cols);
+
+    grid_split_merge_detail::split<Policy>(shrinkPtr(src),
+                                           shrinkPtr(dst[0]), shrinkPtr(dst[1]), shrinkPtr(dst[2]),
+                                           WithOutMask(),
+                                           rows, cols,
+                                           StreamAccessor::getStream(stream));
+}
+
+template <class Policy, class SrcPtr, typename DstType, class MaskPtr>
+__host__ void gridSplit_(const SrcPtr& src, const tuple< GpuMat_<DstType>&, GpuMat_<DstType>&, GpuMat_<DstType>&, GpuMat_<DstType>& >& dst, const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    CV_StaticAssert( VecTraits<typename PtrTraits<SrcPtr>::value_type>::cn == 4, "" );
+
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    CV_Assert( getRows(mask) == rows && getCols(mask) == cols );
+
+    get<0>(dst).create(rows, cols);
+    get<1>(dst).create(rows, cols);
+    get<2>(dst).create(rows, cols);
+    get<3>(dst).create(rows, cols);
+
+    grid_split_merge_detail::split<Policy>(shrinkPtr(src),
+                                           shrinkPtr(get<0>(dst)), shrinkPtr(get<1>(dst)), shrinkPtr(get<2>(dst)), shrinkPtr(get<3>(dst)),
+                                           shrinkPtr(mask),
+                                           rows, cols,
+                                           StreamAccessor::getStream(stream));
+}
+
+template <class Policy, class SrcPtr, typename DstType, class MaskPtr>
+__host__ void gridSplit_(const SrcPtr& src, GpuMat_<DstType> (&dst)[4], const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    CV_StaticAssert( VecTraits<typename PtrTraits<SrcPtr>::value_type>::cn == 4, "" );
+
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    CV_Assert( getRows(mask) == rows && getCols(mask) == cols );
+
+    dst[0].create(rows, cols);
+    dst[1].create(rows, cols);
+    dst[2].create(rows, cols);
+    dst[4].create(rows, cols);
+
+    grid_split_merge_detail::split<Policy>(shrinkPtr(src),
+                                           shrinkPtr(dst[0]), shrinkPtr(dst[1]), shrinkPtr(dst[2]), shrinkPtr(dst[4]),
+                                           shrinkPtr(mask),
+                                           rows, cols,
+                                           StreamAccessor::getStream(stream));
+}
+
+template <class Policy, class SrcPtr, typename DstType>
+__host__ void gridSplit_(const SrcPtr& src, const tuple< GpuMat_<DstType>&, GpuMat_<DstType>&, GpuMat_<DstType>&, GpuMat_<DstType>& >& dst, Stream& stream = Stream::Null())
+{
+    CV_StaticAssert( VecTraits<typename PtrTraits<SrcPtr>::value_type>::cn == 4, "" );
+
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    get<0>(dst).create(rows, cols);
+    get<1>(dst).create(rows, cols);
+    get<2>(dst).create(rows, cols);
+    get<3>(dst).create(rows, cols);
+
+    grid_split_merge_detail::split<Policy>(shrinkPtr(src),
+                                           shrinkPtr(get<0>(dst)), shrinkPtr(get<1>(dst)), shrinkPtr(get<2>(dst)), shrinkPtr(get<3>(dst)),
+                                           WithOutMask(),
+                                           rows, cols,
+                                           StreamAccessor::getStream(stream));
+}
+
+template <class Policy, class SrcPtr, typename DstType>
+__host__ void gridSplit_(const SrcPtr& src, GpuMat_<DstType> (&dst)[4], Stream& stream = Stream::Null())
+{
+    CV_StaticAssert( VecTraits<typename PtrTraits<SrcPtr>::value_type>::cn == 4, "" );
+
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    dst[0].create(rows, cols);
+    dst[1].create(rows, cols);
+    dst[2].create(rows, cols);
+    dst[4].create(rows, cols);
+
+    grid_split_merge_detail::split<Policy>(shrinkPtr(src),
+                                           shrinkPtr(dst[0]), shrinkPtr(dst[1]), shrinkPtr(dst[2]), shrinkPtr(dst[4]),
+                                           WithOutMask(),
+                                           rows, cols,
+                                           StreamAccessor::getStream(stream));
+}
+
+// Default Policy
+
+struct DefaultSplitMergePolicy
+{
+    enum {
+        block_size_x = 32,
+        block_size_y = 8
+    };
+};
+
+template <class SrcPtrTuple, typename DstType, class MaskPtr>
+__host__ void gridMerge(const SrcPtrTuple& src, GpuMat_<DstType>& dst, const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    gridMerge_<DefaultSplitMergePolicy>(src, dst, mask, stream);
+}
+
+template <class SrcPtrTuple, typename DstType>
+__host__ void gridMerge(const SrcPtrTuple& src, GpuMat_<DstType>& dst, Stream& stream = Stream::Null())
+{
+    gridMerge_<DefaultSplitMergePolicy>(src, dst, stream);
+}
+
+template <class SrcPtr, typename DstType, class MaskPtr>
+__host__ void gridSplit(const SrcPtr& src, const tuple< GpuMat_<DstType>&, GpuMat_<DstType>& >& dst, const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    gridSplit_<DefaultSplitMergePolicy>(src, dst, mask, stream);
+}
+
+template <class SrcPtr, typename DstType>
+__host__ void gridSplit(const SrcPtr& src, const tuple< GpuMat_<DstType>&, GpuMat_<DstType>& >& dst, Stream& stream = Stream::Null())
+{
+    gridSplit_<DefaultSplitMergePolicy>(src, dst, stream);
+}
+
+template <class SrcPtr, typename DstType, class MaskPtr>
+__host__ void gridSplit(const SrcPtr& src, const tuple< GpuMat_<DstType>&, GpuMat_<DstType>&, GpuMat_<DstType>& >& dst, const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    gridSplit_<DefaultSplitMergePolicy>(src, dst, mask, stream);
+}
+
+template <class SrcPtr, typename DstType>
+__host__ void gridSplit(const SrcPtr& src, const tuple< GpuMat_<DstType>&, GpuMat_<DstType>&, GpuMat_<DstType>& >& dst, Stream& stream = Stream::Null())
+{
+    gridSplit_<DefaultSplitMergePolicy>(src, dst, stream);
+}
+
+template <class SrcPtr, typename DstType, class MaskPtr>
+__host__ void gridSplit(const SrcPtr& src, const tuple< GpuMat_<DstType>&, GpuMat_<DstType>&, GpuMat_<DstType>&, GpuMat_<DstType>& >& dst, const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    gridSplit_<DefaultSplitMergePolicy>(src, dst, mask, stream);
+}
+
+template <class SrcPtr, typename DstType>
+__host__ void gridSplit(const SrcPtr& src, const tuple< GpuMat_<DstType>&, GpuMat_<DstType>&, GpuMat_<DstType>&, GpuMat_<DstType>& >& dst, Stream& stream = Stream::Null())
+{
+    gridSplit_<DefaultSplitMergePolicy>(src, dst, stream);
+}
+
+template <class SrcPtr, typename DstType, int COUNT, class MaskPtr>
+__host__ void gridSplit(const SrcPtr& src, GpuMat_<DstType> (&dst)[COUNT], const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    gridSplit_<DefaultSplitMergePolicy>(src, dst, mask, stream);
+}
+
+template <class SrcPtr, typename DstType, int COUNT>
+__host__ void gridSplit(const SrcPtr& src, GpuMat_<DstType> (&dst)[COUNT], Stream& stream = Stream::Null())
+{
+    gridSplit_<DefaultSplitMergePolicy>(src, dst, stream);
+}
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/grid/transform.hpp

@@ -0,0 +1,309 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_GRID_TRANSFORM_HPP__
+#define __OPENCV_CUDEV_GRID_TRANSFORM_HPP__
+
+#include "../common.hpp"
+#include "../util/tuple.hpp"
+#include "../ptr2d/traits.hpp"
+#include "../ptr2d/gpumat.hpp"
+#include "../ptr2d/mask.hpp"
+#include "../ptr2d/zip.hpp"
+#include "detail/transform.hpp"
+
+namespace cv { namespace cudev {
+
+template <class Policy, class SrcPtr, typename DstType, class UnOp, class MaskPtr>
+__host__ void gridTransform_(const SrcPtr& src, GpuMat_<DstType>& dst, const UnOp& op, const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    CV_Assert( getRows(mask) == rows && getCols(mask) == cols );
+
+    dst.create(rows, cols);
+
+    grid_transform_detail::transform<Policy>(shrinkPtr(src), shrinkPtr(dst), op, shrinkPtr(mask), rows, cols, StreamAccessor::getStream(stream));
+}
+
+template <class Policy, class SrcPtr, typename DstType, class UnOp>
+__host__ void gridTransform_(const SrcPtr& src, GpuMat_<DstType>& dst, const UnOp& op, Stream& stream = Stream::Null())
+{
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    dst.create(rows, cols);
+
+    grid_transform_detail::transform<Policy>(shrinkPtr(src), shrinkPtr(dst), op, WithOutMask(), rows, cols, StreamAccessor::getStream(stream));
+}
+
+template <class Policy, class SrcPtr1, class SrcPtr2, typename DstType, class BinOp, class MaskPtr>
+__host__ void gridTransform_(const SrcPtr1& src1, const SrcPtr2& src2, GpuMat_<DstType>& dst, const BinOp& op, const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    const int rows = getRows(src1);
+    const int cols = getCols(src1);
+
+    CV_Assert( getRows(src2) == rows && getCols(src2) == cols );
+    CV_Assert( getRows(mask) == rows && getCols(mask) == cols );
+
+    dst.create(rows, cols);
+
+    grid_transform_detail::transform<Policy>(shrinkPtr(src1), shrinkPtr(src2), shrinkPtr(dst), op, shrinkPtr(mask), rows, cols, StreamAccessor::getStream(stream));
+}
+
+template <class Policy, class SrcPtr1, class SrcPtr2, typename DstType, class BinOp>
+__host__ void gridTransform_(const SrcPtr1& src1, const SrcPtr2& src2, GpuMat_<DstType>& dst, const BinOp& op, Stream& stream = Stream::Null())
+{
+    const int rows = getRows(src1);
+    const int cols = getCols(src1);
+
+    CV_Assert( getRows(src2) == rows && getCols(src2) == cols );
+
+    dst.create(rows, cols);
+
+    grid_transform_detail::transform<Policy>(shrinkPtr(src1), shrinkPtr(src2), shrinkPtr(dst), op, WithOutMask(), rows, cols, StreamAccessor::getStream(stream));
+}
+
+template <class Policy, class SrcPtr, typename D0, typename D1, class OpTuple, class MaskPtr>
+__host__ void gridTransform_(const SrcPtr& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>& >& dst, const OpTuple& op, const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    CV_StaticAssert( tuple_size<OpTuple>::value == 2, "" );
+
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    CV_Assert( getRows(mask) == rows && getCols(mask) == cols );
+
+    get<0>(dst).create(rows, cols);
+    get<1>(dst).create(rows, cols);
+
+    grid_transform_detail::transform_tuple<Policy>(shrinkPtr(src),
+                                                   shrinkPtr(zipPtr(get<0>(dst), get<1>(dst))),
+                                                   op,
+                                                   shrinkPtr(mask),
+                                                   rows, cols,
+                                                   StreamAccessor::getStream(stream));
+}
+
+template <class Policy, class SrcPtr, typename D0, typename D1, class OpTuple>
+__host__ void gridTransform_(const SrcPtr& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>& >& dst, const OpTuple& op, Stream& stream = Stream::Null())
+{
+    CV_StaticAssert( tuple_size<OpTuple>::value == 2, "" );
+
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    get<0>(dst).create(rows, cols);
+    get<1>(dst).create(rows, cols);
+
+    grid_transform_detail::transform_tuple<Policy>(shrinkPtr(src),
+                                                   shrinkPtr(zipPtr(get<0>(dst), get<1>(dst))),
+                                                   op,
+                                                   WithOutMask(),
+                                                   rows, cols,
+                                                   StreamAccessor::getStream(stream));
+}
+
+template <class Policy, class SrcPtr, typename D0, typename D1, typename D2, class OpTuple, class MaskPtr>
+__host__ void gridTransform_(const SrcPtr& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>&, GpuMat_<D2>& >& dst, const OpTuple& op, const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    CV_StaticAssert( tuple_size<OpTuple>::value == 3, "" );
+
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    CV_Assert( getRows(mask) == rows && getCols(mask) == cols );
+
+    get<0>(dst).create(rows, cols);
+    get<1>(dst).create(rows, cols);
+    get<2>(dst).create(rows, cols);
+
+    grid_transform_detail::transform_tuple<Policy>(shrinkPtr(src),
+                                                   shrinkPtr(zipPtr(get<0>(dst), get<1>(dst), get<2>(dst))),
+                                                   op,
+                                                   shrinkPtr(mask),
+                                                   rows, cols,
+                                                   StreamAccessor::getStream(stream));
+}
+
+template <class Policy, class SrcPtr, typename D0, typename D1, typename D2, class OpTuple>
+__host__ void gridTransform_(const SrcPtr& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>&, GpuMat_<D2>& >& dst, const OpTuple& op, Stream& stream = Stream::Null())
+{
+    CV_StaticAssert( tuple_size<OpTuple>::value == 3, "" );
+
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    get<0>(dst).create(rows, cols);
+    get<1>(dst).create(rows, cols);
+    get<2>(dst).create(rows, cols);
+
+    grid_transform_detail::transform_tuple<Policy>(shrinkPtr(src),
+                                                   shrinkPtr(zipPtr(get<0>(dst), get<1>(dst), get<2>(dst))),
+                                                   op,
+                                                   WithOutMask(),
+                                                   rows, cols,
+                                                   StreamAccessor::getStream(stream));
+}
+
+template <class Policy, class SrcPtr, typename D0, typename D1, typename D2, typename D3, class OpTuple, class MaskPtr>
+__host__ void gridTransform_(const SrcPtr& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>&, GpuMat_<D2>&, GpuMat_<D3>& >& dst, const OpTuple& op, const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    CV_StaticAssert( tuple_size<OpTuple>::value == 4, "" );
+
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    CV_Assert( getRows(mask) == rows && getCols(mask) == cols );
+
+    get<0>(dst).create(rows, cols);
+    get<1>(dst).create(rows, cols);
+    get<2>(dst).create(rows, cols);
+    get<3>(dst).create(rows, cols);
+
+    grid_transform_detail::transform_tuple<Policy>(shrinkPtr(src),
+                                                   shrinkPtr(zipPtr(get<0>(dst), get<1>(dst), get<2>(dst), get<3>(dst))),
+                                                   op,
+                                                   shrinkPtr(mask),
+                                                   rows, cols,
+                                                   StreamAccessor::getStream(stream));
+}
+
+template <class Policy, class SrcPtr, typename D0, typename D1, typename D2, typename D3, class OpTuple>
+__host__ void gridTransform_(const SrcPtr& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>&, GpuMat_<D2>&, GpuMat_<D3>& >& dst, const OpTuple& op, Stream& stream = Stream::Null())
+{
+    CV_StaticAssert( tuple_size<OpTuple>::value == 4, "" );
+
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    get<0>(dst).create(rows, cols);
+    get<1>(dst).create(rows, cols);
+    get<2>(dst).create(rows, cols);
+    get<3>(dst).create(rows, cols);
+
+    grid_transform_detail::transform_tuple<Policy>(shrinkPtr(src),
+                                                   shrinkPtr(zipPtr(get<0>(dst), get<1>(dst), get<2>(dst), get<3>(dst))),
+                                                   op,
+                                                   WithOutMask(),
+                                                   rows, cols,
+                                                   StreamAccessor::getStream(stream));
+}
+
+// Default Policy
+
+struct DefaultTransformPolicy
+{
+    enum {
+        block_size_x = 32,
+        block_size_y = 8,
+        shift = 4
+    };
+};
+
+template <class SrcPtr, typename DstType, class Op, class MaskPtr>
+__host__ void gridTransform(const SrcPtr& src, GpuMat_<DstType>& dst, const Op& op, const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    gridTransform_<DefaultTransformPolicy>(src, dst, op, mask, stream);
+}
+
+template <class SrcPtr, typename DstType, class Op>
+__host__ void gridTransform(const SrcPtr& src, GpuMat_<DstType>& dst, const Op& op, Stream& stream = Stream::Null())
+{
+    gridTransform_<DefaultTransformPolicy>(src, dst, op, stream);
+}
+
+template <class SrcPtr1, class SrcPtr2, typename DstType, class Op, class MaskPtr>
+__host__ void gridTransform(const SrcPtr1& src1, const SrcPtr1& src2, GpuMat_<DstType>& dst, const Op& op, const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    gridTransform_<DefaultTransformPolicy>(src1, src2, dst, op, mask, stream);
+}
+
+template <class SrcPtr1, class SrcPtr2, typename DstType, class Op>
+__host__ void gridTransform(const SrcPtr1& src1, const SrcPtr1& src2, GpuMat_<DstType>& dst, const Op& op, Stream& stream = Stream::Null())
+{
+    gridTransform_<DefaultTransformPolicy>(src1, src2, dst, op, stream);
+}
+
+template <class SrcPtr, typename D0, typename D1, class OpTuple, class MaskPtr>
+__host__ void gridTransform(const SrcPtr& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>& >& dst, const OpTuple& op, const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    gridTransform_<DefaultTransformPolicy>(src, dst, op, mask, stream);
+}
+
+template <class SrcPtr, typename D0, typename D1, class OpTuple>
+__host__ void gridTransform(const SrcPtr& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>& >& dst, const OpTuple& op, Stream& stream = Stream::Null())
+{
+    gridTransform_<DefaultTransformPolicy>(src, dst, op, stream);
+}
+
+template <class SrcPtr, typename D0, typename D1, typename D2, class OpTuple, class MaskPtr>
+__host__ void gridTransform(const SrcPtr& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>&, GpuMat_<D2>& >& dst, const OpTuple& op, const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    gridTransform_<DefaultTransformPolicy>(src, dst, op, mask, stream);
+}
+
+template <class SrcPtr, typename D0, typename D1, typename D2, class OpTuple>
+__host__ void gridTransform(const SrcPtr& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>&, GpuMat_<D2>& >& dst, const OpTuple& op, Stream& stream = Stream::Null())
+{
+    gridTransform_<DefaultTransformPolicy>(src, dst, op, stream);
+}
+
+template <class SrcPtr, typename D0, typename D1, typename D2, typename D3, class OpTuple, class MaskPtr>
+__host__ void gridTransform(const SrcPtr& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>&, GpuMat_<D2>&, GpuMat_<D3>& >& dst, const OpTuple& op, const MaskPtr& mask, Stream& stream = Stream::Null())
+{
+    gridTransform_<DefaultTransformPolicy>(src, dst, op, mask, stream);
+}
+
+template <class SrcPtr, typename D0, typename D1, typename D2, typename D3, class OpTuple>
+__host__ void gridTransform(const SrcPtr& src, const tuple< GpuMat_<D0>&, GpuMat_<D1>&, GpuMat_<D2>&, GpuMat_<D3>& >& dst, const OpTuple& op, Stream& stream = Stream::Null())
+{
+    gridTransform_<DefaultTransformPolicy>(src, dst, op, stream);
+}
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/grid/transpose.hpp

@@ -0,0 +1,69 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_GRID_TRANSPOSE_HPP__
+#define __OPENCV_CUDEV_GRID_TRANSPOSE_HPP__
+
+#include "../common.hpp"
+#include "../ptr2d/traits.hpp"
+#include "../ptr2d/gpumat.hpp"
+#include "detail/transpose.hpp"
+
+namespace cv { namespace cudev {
+
+template <class SrcPtr, typename DstType>
+__host__ void gridTranspose(const SrcPtr& src, GpuMat_<DstType>& dst, Stream& stream = Stream::Null())
+{
+    const int rows = getRows(src);
+    const int cols = getCols(src);
+
+    dst.create(cols, rows);
+
+    transpose_detail::transpose(shrinkPtr(src), shrinkPtr(dst), rows, cols, StreamAccessor::getStream(stream));
+}
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/ptr2d/constant.hpp

@@ -0,0 +1,93 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_PTR2D_CONSTANT_HPP__
+#define __OPENCV_CUDEV_PTR2D_CONSTANT_HPP__
+
+#include "../common.hpp"
+#include "traits.hpp"
+
+namespace cv { namespace cudev {
+
+template <typename T> struct ConstantPtr
+{
+    typedef T   value_type;
+    typedef int index_type;
+
+    T value;
+
+    __device__ __forceinline__ T operator ()(int, int) const { return value; }
+};
+
+template <typename T> struct ConstantPtrSz : ConstantPtr<T>
+{
+    int rows, cols;
+};
+
+template <typename T>
+__host__ ConstantPtr<T> constantPtr(T value)
+{
+    ConstantPtr<T> p;
+    p.value = value;
+    return p;
+}
+
+template <typename T> ConstantPtrSz<T>
+__host__ constantPtr(T value, int rows, int cols)
+{
+    ConstantPtrSz<T> p;
+    p.value = value;
+    p.rows = rows;
+    p.cols = cols;
+    return p;
+}
+
+template <typename T> struct PtrTraits< ConstantPtrSz<T> > : PtrTraitsBase< ConstantPtrSz<T>, ConstantPtr<T> >
+{
+};
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/ptr2d/deriv.hpp

@@ -0,0 +1,393 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_PTR2D_DERIV_HPP__
+#define __OPENCV_CUDEV_PTR2D_DERIV_HPP__
+
+#include "../common.hpp"
+#include "../grid/copy.hpp"
+#include "traits.hpp"
+#include "gpumat.hpp"
+
+namespace cv { namespace cudev {
+
+// derivX
+
+template <class SrcPtr> struct DerivXPtr
+{
+    typedef typename PtrTraits<SrcPtr>::value_type value_type;
+    typedef int                                    index_type;
+
+    SrcPtr src;
+
+    __device__ __forceinline__ typename PtrTraits<SrcPtr>::value_type operator ()(int y, int x) const
+    {
+        return src(y, x + 1) - src(y, x - 1);
+    }
+};
+
+template <class SrcPtr> struct DerivXPtrSz : DerivXPtr<SrcPtr>
+{
+    int rows, cols;
+
+    template <typename T>
+    __host__ void assignTo(GpuMat_<T>& dst, Stream& stream = Stream::Null()) const
+    {
+        gridCopy(*this, dst, stream);
+    }
+};
+
+template <class SrcPtr>
+__host__ DerivXPtrSz<typename PtrTraits<SrcPtr>::ptr_type> derivXPtr(const SrcPtr& src)
+{
+    DerivXPtrSz<typename PtrTraits<SrcPtr>::ptr_type> s;
+    s.src = shrinkPtr(src);
+    s.rows = getRows(src);
+    s.cols = getCols(src);
+    return s;
+}
+
+template <class SrcPtr> struct PtrTraits< DerivXPtrSz<SrcPtr> > : PtrTraitsBase<DerivXPtrSz<SrcPtr>, DerivXPtr<SrcPtr> >
+{
+};
+
+// derivY
+
+template <class SrcPtr> struct DerivYPtr
+{
+    typedef typename PtrTraits<SrcPtr>::value_type value_type;
+    typedef int                                    index_type;
+
+    SrcPtr src;
+
+    __device__ __forceinline__ typename PtrTraits<SrcPtr>::value_type operator ()(int y, int x) const
+    {
+        return src(y + 1, x) - src(y - 1, x);
+    }
+};
+
+template <class SrcPtr> struct DerivYPtrSz : DerivYPtr<SrcPtr>
+{
+    int rows, cols;
+
+    template <typename T>
+    __host__ void assignTo(GpuMat_<T>& dst, Stream& stream = Stream::Null()) const
+    {
+        gridCopy(*this, dst, stream);
+    }
+};
+
+template <class SrcPtr>
+__host__ DerivYPtrSz<typename PtrTraits<SrcPtr>::ptr_type> derivYPtr(const SrcPtr& src)
+{
+    DerivYPtrSz<typename PtrTraits<SrcPtr>::ptr_type> s;
+    s.src = shrinkPtr(src);
+    s.rows = getRows(src);
+    s.cols = getCols(src);
+    return s;
+}
+
+template <class SrcPtr> struct PtrTraits< DerivYPtrSz<SrcPtr> > : PtrTraitsBase<DerivYPtrSz<SrcPtr>, DerivYPtr<SrcPtr> >
+{
+};
+
+// sobelX
+
+template <class SrcPtr> struct SobelXPtr
+{
+    typedef typename PtrTraits<SrcPtr>::value_type value_type;
+    typedef int                                    index_type;
+
+    SrcPtr src;
+
+    __device__ typename PtrTraits<SrcPtr>::value_type operator ()(int y, int x) const
+    {
+        typename PtrTraits<SrcPtr>::value_type vals[6] =
+        {
+            src(y - 1, x - 1), src(y - 1, x + 1),
+            src(y    , x - 1), src(y    , x + 1),
+            src(y + 1, x - 1), src(y + 1, x + 1),
+        };
+
+        return (vals[1] - vals[0]) + 2 * (vals[3] - vals[2]) + (vals[5] - vals[4]);
+    }
+};
+
+template <class SrcPtr> struct SobelXPtrSz : SobelXPtr<SrcPtr>
+{
+    int rows, cols;
+
+    template <typename T>
+    __host__ void assignTo(GpuMat_<T>& dst, Stream& stream = Stream::Null()) const
+    {
+        gridCopy(*this, dst, stream);
+    }
+};
+
+template <class SrcPtr>
+__host__ SobelXPtrSz<typename PtrTraits<SrcPtr>::ptr_type> sobelXPtr(const SrcPtr& src)
+{
+    SobelXPtrSz<typename PtrTraits<SrcPtr>::ptr_type> s;
+    s.src = shrinkPtr(src);
+    s.rows = getRows(src);
+    s.cols = getCols(src);
+    return s;
+}
+
+template <class SrcPtr> struct PtrTraits< SobelXPtrSz<SrcPtr> > : PtrTraitsBase<SobelXPtrSz<SrcPtr>, SobelXPtr<SrcPtr> >
+{
+};
+
+// sobelY
+
+template <class SrcPtr> struct SobelYPtr
+{
+    typedef typename PtrTraits<SrcPtr>::value_type value_type;
+    typedef int                                    index_type;
+
+    SrcPtr src;
+
+    __device__ typename PtrTraits<SrcPtr>::value_type operator ()(int y, int x) const
+    {
+        typename PtrTraits<SrcPtr>::value_type vals[6] =
+        {
+            src(y - 1, x - 1), src(y - 1, x), src(y - 1, x + 1),
+            src(y + 1, x - 1), src(y + 1, x), src(y + 1, x + 1)
+        };
+
+        return (vals[3] - vals[0]) + 2 * (vals[4] - vals[1]) + (vals[5] - vals[2]);
+    }
+};
+
+template <class SrcPtr> struct SobelYPtrSz : SobelYPtr<SrcPtr>
+{
+    int rows, cols;
+
+    template <typename T>
+    __host__ void assignTo(GpuMat_<T>& dst, Stream& stream = Stream::Null()) const
+    {
+        gridCopy(*this, dst, stream);
+    }
+};
+
+template <class SrcPtr>
+__host__ SobelYPtrSz<typename PtrTraits<SrcPtr>::ptr_type> sobelYPtr(const SrcPtr& src)
+{
+    SobelYPtrSz<typename PtrTraits<SrcPtr>::ptr_type> s;
+    s.src = shrinkPtr(src);
+    s.rows = getRows(src);
+    s.cols = getCols(src);
+    return s;
+}
+
+template <class SrcPtr> struct PtrTraits< SobelYPtrSz<SrcPtr> > : PtrTraitsBase<SobelYPtrSz<SrcPtr>, SobelYPtr<SrcPtr> >
+{
+};
+
+// scharrX
+
+template <class SrcPtr> struct ScharrXPtr
+{
+    typedef typename PtrTraits<SrcPtr>::value_type value_type;
+    typedef int                                    index_type;
+
+    SrcPtr src;
+
+    __device__ typename PtrTraits<SrcPtr>::value_type operator ()(int y, int x) const
+    {
+        typename PtrTraits<SrcPtr>::value_type vals[6] =
+        {
+            src(y - 1, x - 1), src(y - 1, x + 1),
+            src(y    , x - 1), src(y    , x + 1),
+            src(y + 1, x - 1), src(y + 1, x + 1),
+        };
+
+        return 3 * (vals[1] - vals[0]) + 10 * (vals[3] - vals[2]) + 3 * (vals[5] - vals[4]);
+    }
+};
+
+template <class SrcPtr> struct ScharrXPtrSz : ScharrXPtr<SrcPtr>
+{
+    int rows, cols;
+
+    template <typename T>
+    __host__ void assignTo(GpuMat_<T>& dst, Stream& stream = Stream::Null()) const
+    {
+        gridCopy(*this, dst, stream);
+    }
+};
+
+template <class SrcPtr>
+__host__ ScharrXPtrSz<typename PtrTraits<SrcPtr>::ptr_type> scharrXPtr(const SrcPtr& src)
+{
+    ScharrXPtrSz<typename PtrTraits<SrcPtr>::ptr_type> s;
+    s.src = shrinkPtr(src);
+    s.rows = getRows(src);
+    s.cols = getCols(src);
+    return s;
+}
+
+template <class SrcPtr> struct PtrTraits< ScharrXPtrSz<SrcPtr> > : PtrTraitsBase<ScharrXPtrSz<SrcPtr>, ScharrXPtr<SrcPtr> >
+{
+};
+
+// scharrY
+
+template <class SrcPtr> struct ScharrYPtr
+{
+    typedef typename PtrTraits<SrcPtr>::value_type value_type;
+    typedef int                                    index_type;
+
+    SrcPtr src;
+
+    __device__ typename PtrTraits<SrcPtr>::value_type operator ()(int y, int x) const
+    {
+        typename PtrTraits<SrcPtr>::value_type vals[6] =
+        {
+            src(y - 1, x - 1), src(y - 1, x), src(y - 1, x + 1),
+            src(y + 1, x - 1), src(y + 1, x), src(y + 1, x + 1)
+        };
+
+        return 3 * (vals[3] - vals[0]) + 10 * (vals[4] - vals[1]) + 3 * (vals[5] - vals[2]);
+    }
+};
+
+template <class SrcPtr> struct ScharrYPtrSz : ScharrYPtr<SrcPtr>
+{
+    int rows, cols;
+
+    template <typename T>
+    __host__ void assignTo(GpuMat_<T>& dst, Stream& stream = Stream::Null()) const
+    {
+        gridCopy(*this, dst, stream);
+    }
+};
+
+template <class SrcPtr>
+__host__ ScharrYPtrSz<typename PtrTraits<SrcPtr>::ptr_type> scharrYPtr(const SrcPtr& src)
+{
+    ScharrYPtrSz<typename PtrTraits<SrcPtr>::ptr_type> s;
+    s.src = shrinkPtr(src);
+    s.rows = getRows(src);
+    s.cols = getCols(src);
+    return s;
+}
+
+template <class SrcPtr> struct PtrTraits< ScharrYPtrSz<SrcPtr> > : PtrTraitsBase<ScharrYPtrSz<SrcPtr>, ScharrYPtr<SrcPtr> >
+{
+};
+
+// laplacian
+
+template <int ksize, class SrcPtr> struct LaplacianPtr;
+
+template <class SrcPtr> struct LaplacianPtr<1, SrcPtr>
+{
+    typedef typename PtrTraits<SrcPtr>::value_type value_type;
+    typedef int                                    index_type;
+
+    SrcPtr src;
+
+    __device__ typename PtrTraits<SrcPtr>::value_type operator ()(int y, int x) const
+    {
+        typename PtrTraits<SrcPtr>::value_type vals[5] =
+        {
+                           src(y - 1, x),
+            src(y, x - 1), src(y    , x), src(y, x + 1),
+                           src(y + 1, x)
+        };
+
+        return (vals[0] + vals[1] + vals[3] + vals[4]) - 4 * vals[2];
+    }
+};
+
+template <class SrcPtr> struct LaplacianPtr<3, SrcPtr>
+{
+    typedef typename PtrTraits<SrcPtr>::value_type value_type;
+    typedef int                                    index_type;
+
+   SrcPtr src;
+
+   __device__ typename PtrTraits<SrcPtr>::value_type operator ()(int y, int x) const
+   {
+       typename PtrTraits<SrcPtr>::value_type vals[5] =
+       {
+           src(y - 1, x - 1),            src(y - 1, x + 1),
+                              src(y, x),
+           src(y + 1, x - 1),            src(y + 1, x + 1)
+       };
+
+       return 2 * (vals[0] + vals[1] + vals[3] + vals[4]) - 8 * vals[2];
+   }
+};
+
+template <int ksize, class SrcPtr> struct LaplacianPtrSz : LaplacianPtr<ksize, SrcPtr>
+{
+    int rows, cols;
+
+    template <typename T>
+    __host__ void assignTo(GpuMat_<T>& dst, Stream& stream = Stream::Null()) const
+    {
+        gridCopy(*this, dst, stream);
+    }
+};
+
+template <int ksize, class SrcPtr>
+__host__ LaplacianPtrSz<ksize, typename PtrTraits<SrcPtr>::ptr_type> laplacianPtr(const SrcPtr& src)
+{
+    LaplacianPtrSz<ksize, typename PtrTraits<SrcPtr>::ptr_type> ptr;
+    ptr.src = shrinkPtr(src);
+    ptr.rows = getRows(src);
+    ptr.cols = getCols(src);
+    return ptr;
+}
+
+template <int ksize, class SrcPtr> struct PtrTraits< LaplacianPtrSz<ksize, SrcPtr> > : PtrTraitsBase<LaplacianPtrSz<ksize, SrcPtr>, LaplacianPtr<ksize, SrcPtr> >
+{
+};
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/ptr2d/detail/gpumat.hpp

@@ -0,0 +1,338 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_PTR2D_GPUMAT_DETAIL_HPP__
+#define __OPENCV_CUDEV_PTR2D_GPUMAT_DETAIL_HPP__
+
+#include "../gpumat.hpp"
+
+namespace cv { namespace cudev {
+
+template <typename T>
+__host__ GpuMat_<T>::GpuMat_()
+    : GpuMat()
+{
+    flags = (flags & ~CV_MAT_TYPE_MASK) | DataType<T>::type;
+}
+
+template <typename T>
+__host__ GpuMat_<T>::GpuMat_(int arows, int acols)
+    : GpuMat(arows, acols, DataType<T>::type)
+{
+}
+
+template <typename T>
+__host__ GpuMat_<T>::GpuMat_(Size asize)
+    : GpuMat(asize.height, asize.width, DataType<T>::type)
+{
+}
+
+template <typename T>
+__host__ GpuMat_<T>::GpuMat_(int arows, int acols, Scalar val)
+    : GpuMat(arows, acols, DataType<T>::type, val)
+{
+}
+
+template <typename T>
+__host__ GpuMat_<T>::GpuMat_(Size asize, Scalar val)
+    : GpuMat(asize.height, asize.width, DataType<T>::type, val)
+{
+}
+
+template <typename T>
+__host__ GpuMat_<T>::GpuMat_(const GpuMat_& m)
+    : GpuMat(m)
+{
+}
+
+template <typename T>
+__host__ GpuMat_<T>::GpuMat_(const GpuMat& m)
+    : GpuMat()
+{
+    flags = (flags & ~CV_MAT_TYPE_MASK) | DataType<T>::type;
+
+    if (DataType<T>::type == m.type())
+    {
+        GpuMat::operator =(m);
+        return;
+    }
+
+    if (DataType<T>::depth == m.depth())
+    {
+        GpuMat::operator =(m.reshape(DataType<T>::channels, m.rows));
+        return;
+    }
+
+    CV_Assert( DataType<T>::channels == m.channels() );
+    m.convertTo(*this, type());
+}
+
+template <typename T>
+__host__ GpuMat_<T>::GpuMat_(int arows, int acols, T* adata, size_t astep)
+    : GpuMat(arows, acols, DataType<T>::type, adata, astep)
+{
+}
+
+template <typename T>
+__host__ GpuMat_<T>::GpuMat_(Size asize, T* adata, size_t astep)
+    : GpuMat(asize.height, asize.width, DataType<T>::type, adata, astep)
+{
+}
+
+template <typename T>
+__host__ GpuMat_<T>::GpuMat_(const GpuMat_& m, Range arowRange, Range acolRange)
+    : GpuMat(m, arowRange, acolRange)
+{
+}
+
+template <typename T>
+__host__ GpuMat_<T>::GpuMat_(const GpuMat_& m, Rect roi)
+    : GpuMat(m, roi)
+{
+}
+
+template <typename T>
+__host__ GpuMat_<T>::GpuMat_(InputArray arr)
+    : GpuMat()
+{
+    flags = (flags & ~CV_MAT_TYPE_MASK) | DataType<T>::type;
+    upload(arr);
+}
+
+template <typename T>
+__host__ GpuMat_<T>& GpuMat_<T>::operator =(const GpuMat_& m)
+{
+    GpuMat::operator =(m);
+    return *this;
+}
+
+template <typename T>
+__host__ void GpuMat_<T>::create(int arows, int acols)
+{
+    GpuMat::create(arows, acols, DataType<T>::type);
+}
+
+template <typename T>
+__host__ void GpuMat_<T>::create(Size asize)
+{
+    GpuMat::create(asize, DataType<T>::type);
+}
+
+template <typename T>
+__host__ void GpuMat_<T>::swap(GpuMat_& mat)
+{
+    GpuMat::swap(mat);
+}
+
+template <typename T>
+__host__ void GpuMat_<T>::upload(InputArray arr)
+{
+    CV_Assert( arr.type() == DataType<T>::type );
+    GpuMat::upload(arr);
+}
+
+template <typename T>
+__host__ void GpuMat_<T>::upload(InputArray arr, Stream& stream)
+{
+    CV_Assert( arr.type() == DataType<T>::type );
+    GpuMat::upload(arr, stream);
+}
+
+template <typename T>
+__host__ GpuMat_<T>::operator GlobPtrSz<T>() const
+{
+    return globPtr((T*) data, step, rows, cols);
+}
+
+template <typename T>
+__host__ GpuMat_<T>::operator GlobPtr<T>() const
+{
+    return globPtr((T*) data, step);
+}
+
+template <typename T>
+__host__ GpuMat_<T> GpuMat_<T>::clone() const
+{
+    return GpuMat_(GpuMat::clone());
+}
+
+template <typename T>
+__host__ GpuMat_<T> GpuMat_<T>::row(int y) const
+{
+    return GpuMat_(*this, Range(y, y+1), Range::all());
+}
+
+template <typename T>
+__host__ GpuMat_<T> GpuMat_<T>::col(int x) const
+{
+    return GpuMat_(*this, Range::all(), Range(x, x+1));
+}
+
+template <typename T>
+__host__ GpuMat_<T> GpuMat_<T>::rowRange(int startrow, int endrow) const
+{
+    return GpuMat_(*this, Range(startrow, endrow), Range::all());
+}
+
+template <typename T>
+__host__ GpuMat_<T> GpuMat_<T>::rowRange(Range r) const
+{
+    return GpuMat_(*this, r, Range::all());
+}
+
+template <typename T>
+__host__ GpuMat_<T> GpuMat_<T>::colRange(int startcol, int endcol) const
+{
+    return GpuMat_(*this, Range::all(), Range(startcol, endcol));
+}
+
+template <typename T>
+__host__ GpuMat_<T> GpuMat_<T>::colRange(Range r) const
+{
+    return GpuMat_(*this, Range::all(), r);
+}
+
+template <typename T>
+__host__ GpuMat_<T> GpuMat_<T>::operator ()(Range _rowRange, Range _colRange) const
+{
+    return GpuMat_(*this, _rowRange, _colRange);
+}
+
+template <typename T>
+__host__ GpuMat_<T> GpuMat_<T>::operator ()(Rect roi) const
+{
+    return GpuMat_(*this, roi);
+}
+
+template <typename T>
+__host__ GpuMat_<T>& GpuMat_<T>::adjustROI(int dtop, int dbottom, int dleft, int dright)
+{
+    return (GpuMat_<T>&)(GpuMat::adjustROI(dtop, dbottom, dleft, dright));
+}
+
+template <typename T>
+__host__ size_t GpuMat_<T>::elemSize() const
+{
+    CV_DbgAssert( GpuMat::elemSize() == sizeof(T) );
+    return sizeof(T);
+}
+
+template <typename T>
+__host__ size_t GpuMat_<T>::elemSize1() const
+{
+    CV_DbgAssert( GpuMat::elemSize1() == sizeof(T) / DataType<T>::channels );
+    return sizeof(T) / DataType<T>::channels;
+}
+
+template <typename T>
+__host__ int GpuMat_<T>::type() const
+{
+    CV_DbgAssert( GpuMat::type() == DataType<T>::type );
+    return DataType<T>::type;
+}
+
+template <typename T>
+__host__ int GpuMat_<T>::depth() const
+{
+    CV_DbgAssert( GpuMat::depth() == DataType<T>::depth );
+    return DataType<T>::depth;
+}
+
+template <typename T>
+__host__ int GpuMat_<T>::channels() const
+{
+    CV_DbgAssert( GpuMat::channels() == DataType<T>::channels );
+    return DataType<T>::channels;
+}
+
+template <typename T>
+__host__ size_t GpuMat_<T>::stepT() const
+{
+    return step / elemSize();
+}
+
+template <typename T>
+__host__ size_t GpuMat_<T>::step1() const
+{
+    return step / elemSize1();
+}
+
+template <typename T>
+__host__ T* GpuMat_<T>::operator [](int y)
+{
+    return (T*)ptr(y);
+}
+
+template <typename T>
+__host__ const T* GpuMat_<T>::operator [](int y) const
+{
+    return (const T*)ptr(y);
+}
+
+template <typename T> template <class Body>
+__host__ GpuMat_<T>::GpuMat_(const Expr<Body>& expr)
+    : GpuMat()
+{
+    flags = (flags & ~CV_MAT_TYPE_MASK) | DataType<T>::type;
+    *this = expr;
+}
+
+template <typename T> template <class Body>
+__host__ GpuMat_<T>& GpuMat_<T>::operator =(const Expr<Body>& expr)
+{
+    expr.body.assignTo(*this);
+    return *this;
+}
+
+template <typename T> template <class Body>
+__host__ GpuMat_<T>& GpuMat_<T>::assign(const Expr<Body>& expr, Stream& stream)
+{
+    expr.body.assignTo(*this, stream);
+    return *this;
+}
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/ptr2d/extrapolation.hpp

@@ -0,0 +1,219 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_PTR2D_EXTRAPOLATION_HPP__
+#define __OPENCV_CUDEV_PTR2D_EXTRAPOLATION_HPP__
+
+#include "../common.hpp"
+#include "../util/vec_traits.hpp"
+#include "traits.hpp"
+
+namespace cv { namespace cudev {
+
+// BrdConstant
+
+template <class SrcPtr> struct BrdConstant
+{
+    typedef typename PtrTraits<SrcPtr>::value_type value_type;
+    typedef int                                    index_type;
+
+    SrcPtr src;
+    int rows, cols;
+    typename PtrTraits<SrcPtr>::value_type val;
+
+    __device__ __forceinline__ typename PtrTraits<SrcPtr>::value_type operator ()(int y, int x) const
+    {
+        return (x >= 0 && x < cols && y >= 0 && y < rows) ? src(y, x) : val;
+    }
+};
+
+template <class SrcPtr>
+__host__ BrdConstant<typename PtrTraits<SrcPtr>::ptr_type> brdConstant(const SrcPtr& src, typename PtrTraits<SrcPtr>::value_type val)
+{
+    BrdConstant<typename PtrTraits<SrcPtr>::ptr_type> b;
+    b.src = shrinkPtr(src);
+    b.rows = getRows(src);
+    b.cols = getCols(src);
+    b.val = val;
+    return b;
+}
+
+template <class SrcPtr>
+__host__ BrdConstant<typename PtrTraits<SrcPtr>::ptr_type> brdConstant(const SrcPtr& src)
+{
+    return brdConstant(src, VecTraits<typename PtrTraits<SrcPtr>::value_type>::all(0));
+}
+
+// BrdBase
+
+template <class BrdImpl, class SrcPtr> struct BrdBase
+{
+    typedef typename PtrTraits<SrcPtr>::value_type value_type;
+    typedef int                                    index_type;
+
+    SrcPtr src;
+    int rows, cols;
+
+    __device__ __forceinline__ int idx_row(int y) const
+    {
+        return BrdImpl::idx_low(BrdImpl::idx_high(y, rows), rows);
+    }
+
+    __device__ __forceinline__ int idx_col(int x) const
+    {
+        return BrdImpl::idx_low(BrdImpl::idx_high(x, cols), cols);
+    }
+
+    __device__ __forceinline__ typename PtrTraits<SrcPtr>::value_type operator ()(int y, int x) const
+    {
+        return src(idx_row(y), idx_col(x));
+    }
+};
+
+// BrdReplicate
+
+struct BrdReplicate
+{
+    __device__ __forceinline__ static int idx_low(int i, int len)
+    {
+        return ::max(i, 0);
+    }
+
+    __device__ __forceinline__ static int idx_high(int i, int len)
+    {
+        return ::min(i, len - 1);
+    }
+};
+
+template <class SrcPtr>
+__host__ BrdBase<BrdReplicate, typename PtrTraits<SrcPtr>::ptr_type> brdReplicate(const SrcPtr& src)
+{
+    BrdBase<BrdReplicate, typename PtrTraits<SrcPtr>::ptr_type> b;
+    b.src = shrinkPtr(src);
+    b.rows = getRows(src);
+    b.cols = getCols(src);
+    return b;
+}
+
+// BrdReflect101
+
+struct BrdReflect101
+{
+    __device__ __forceinline__ static int idx_low(int i, int len)
+    {
+        return ::abs(i) % len;
+    }
+
+    __device__ __forceinline__ static int idx_high(int i, int len)
+    {
+        const int last_ind = len - 1;
+        return ::abs(last_ind - ::abs(last_ind - i)) % len;
+    }
+};
+
+template <class SrcPtr>
+__host__ BrdBase<BrdReflect101, typename PtrTraits<SrcPtr>::ptr_type> brdReflect101(const SrcPtr& src)
+{
+    BrdBase<BrdReflect101, typename PtrTraits<SrcPtr>::ptr_type> b;
+    b.src = shrinkPtr(src);
+    b.rows = getRows(src);
+    b.cols = getCols(src);
+    return b;
+}
+
+// BrdReflect
+
+struct BrdReflect
+{
+    __device__ __forceinline__ static int idx_low(int i, int len)
+    {
+        return (::abs(i) - (i < 0)) % len;
+    }
+
+    __device__ __forceinline__ static int idx_high(int i, int len)
+    {
+        const int last_ind = len - 1;
+        return (last_ind - ::abs(last_ind - i) + (i > last_ind));
+    }
+};
+
+template <class SrcPtr>
+__host__ BrdBase<BrdReflect, typename PtrTraits<SrcPtr>::ptr_type> brdReflect(const SrcPtr& src)
+{
+    BrdBase<BrdReflect, typename PtrTraits<SrcPtr>::ptr_type> b;
+    b.src = shrinkPtr(src);
+    b.rows = getRows(src);
+    b.cols = getCols(src);
+    return b;
+}
+
+// BrdWrap
+
+struct BrdWrap
+{
+    __device__ __forceinline__ static int idx_low(int i, int len)
+    {
+        return (i >= 0) * i + (i < 0) * (i - ((i - len + 1) / len) * len);
+    }
+
+    __device__ __forceinline__ static int idx_high(int i, int len)
+    {
+        return (i < len) * i + (i >= len) * (i % len);
+    }
+};
+
+template <class SrcPtr>
+__host__ BrdBase<BrdWrap, typename PtrTraits<SrcPtr>::ptr_type> brdWrap(const SrcPtr& src)
+{
+    BrdBase<BrdWrap, typename PtrTraits<SrcPtr>::ptr_type> b;
+    b.src = shrinkPtr(src);
+    b.rows = getRows(src);
+    b.cols = getCols(src);
+    return b;
+}
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/ptr2d/glob.hpp

@@ -0,0 +1,100 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_PTR2D_GLOB_HPP__
+#define __OPENCV_CUDEV_PTR2D_GLOB_HPP__
+
+#include "../common.hpp"
+#include "traits.hpp"
+
+namespace cv { namespace cudev {
+
+template <typename T> struct GlobPtr
+{
+    typedef T   value_type;
+    typedef int index_type;
+
+    T* data;
+    size_t step;
+
+    __device__ __forceinline__       T* row(int y)       { return (      T*)( (      uchar*)data + y * step); }
+    __device__ __forceinline__ const T* row(int y) const { return (const T*)( (const uchar*)data + y * step); }
+
+    __device__ __forceinline__       T& operator ()(int y, int x)       { return row(y)[x]; }
+    __device__ __forceinline__ const T& operator ()(int y, int x) const { return row(y)[x]; }
+};
+
+template <typename T> struct GlobPtrSz : GlobPtr<T>
+{
+    int rows, cols;
+};
+
+template <typename T>
+__host__ GlobPtr<T> globPtr(T* data, size_t step)
+{
+    GlobPtr<T> p;
+    p.data = data;
+    p.step = step;
+    return p;
+}
+
+template <typename T>
+__host__ GlobPtrSz<T> globPtr(T* data, size_t step, int rows, int cols)
+{
+    GlobPtrSz<T> p;
+    p.data = data;
+    p.step = step;
+    p.rows = rows;
+    p.cols = cols;
+    return p;
+}
+
+template <typename T> struct PtrTraits< GlobPtrSz<T> > : PtrTraitsBase<GlobPtrSz<T>, GlobPtr<T> >
+{
+};
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/ptr2d/gpumat.hpp

@@ -0,0 +1,161 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_PTR2D_GPUMAT_HPP__
+#define __OPENCV_CUDEV_PTR2D_GPUMAT_HPP__
+
+#include "../common.hpp"
+#include "../util/vec_traits.hpp"
+#include "../expr/expr.hpp"
+#include "glob.hpp"
+
+namespace cv { namespace cudev {
+
+template <typename T>
+class GpuMat_ : public GpuMat
+{
+public:
+    typedef T value_type;
+
+    //! default constructor
+    __host__ GpuMat_();
+
+    //! constructs GpuMat of the specified size
+    __host__ GpuMat_(int arows, int acols);
+    __host__ explicit GpuMat_(Size asize);
+
+    //! constucts GpuMat and fills it with the specified value
+    __host__ GpuMat_(int arows, int acols, Scalar val);
+    __host__ GpuMat_(Size asize, Scalar val);
+
+    //! copy constructor
+    __host__ GpuMat_(const GpuMat_& m);
+
+    //! copy/conversion contructor. If m is of different type, it's converted
+    __host__ explicit GpuMat_(const GpuMat& m);
+
+    //! constructs a matrix on top of user-allocated data. step is in bytes(!!!), regardless of the type
+    __host__ GpuMat_(int arows, int acols, T* adata, size_t astep = Mat::AUTO_STEP);
+    __host__ GpuMat_(Size asize, T* adata, size_t astep = Mat::AUTO_STEP);
+
+    //! selects a submatrix
+    __host__ GpuMat_(const GpuMat_& m, Range arowRange, Range acolRange);
+    __host__ GpuMat_(const GpuMat_& m, Rect roi);
+
+    //! builds GpuMat from host memory (Blocking call)
+    __host__ explicit GpuMat_(InputArray arr);
+
+    //! assignment operators
+    __host__ GpuMat_& operator =(const GpuMat_& m);
+
+    //! allocates new GpuMat data unless the GpuMat already has specified size and type
+    __host__ void create(int arows, int acols);
+    __host__ void create(Size asize);
+
+    //! swaps with other smart pointer
+    __host__ void swap(GpuMat_& mat);
+
+    //! pefroms upload data to GpuMat (Blocking call)
+    __host__ void upload(InputArray arr);
+
+    //! pefroms upload data to GpuMat (Non-Blocking call)
+    __host__ void upload(InputArray arr, Stream& stream);
+
+    //! convert to GlobPtr
+    __host__ operator GlobPtrSz<T>() const;
+    __host__ operator GlobPtr<T>() const;
+
+    //! overridden forms of GpuMat::row() etc.
+    __host__ GpuMat_ clone() const;
+    __host__ GpuMat_ row(int y) const;
+    __host__ GpuMat_ col(int x) const;
+    __host__ GpuMat_ rowRange(int startrow, int endrow) const;
+    __host__ GpuMat_ rowRange(Range r) const;
+    __host__ GpuMat_ colRange(int startcol, int endcol) const;
+    __host__ GpuMat_ colRange(Range r) const;
+    __host__ GpuMat_ operator ()(Range rowRange, Range colRange) const;
+    __host__ GpuMat_ operator ()(Rect roi) const;
+    __host__ GpuMat_& adjustROI(int dtop, int dbottom, int dleft, int dright);
+
+    //! overridden forms of GpuMat::elemSize() etc.
+    __host__ size_t elemSize() const;
+    __host__ size_t elemSize1() const;
+    __host__ int type() const;
+    __host__ int depth() const;
+    __host__ int channels() const;
+    __host__ size_t step1() const;
+
+    //! returns step()/sizeof(T)
+    __host__ size_t stepT() const;
+
+    //! more convenient forms of row and element access operators
+    __host__ T* operator [](int y);
+    __host__ const T* operator [](int y) const;
+
+    //! expression templates
+    template <class Body> __host__ GpuMat_(const Expr<Body>& expr);
+    template <class Body> __host__ GpuMat_& operator =(const Expr<Body>& expr);
+    template <class Body> __host__ GpuMat_& assign(const Expr<Body>& expr, Stream& stream);
+};
+
+//! creates alternative GpuMat header for the same data, with different
+//! number of channels and/or different number of rows. see cvReshape.
+template <int cn, typename T>
+__host__ GpuMat_<typename MakeVec<typename VecTraits<T>::elem_type, cn>::type>
+reshape_(const GpuMat_<T>& mat, int rows = 0)
+{
+    GpuMat_<typename MakeVec<typename VecTraits<T>::elem_type, cn>::type> dst(mat.reshape(cn, rows));
+    return dst;
+}
+
+template <typename T> struct PtrTraits< GpuMat_<T> > : PtrTraitsBase<GpuMat_<T>, GlobPtr<T> >
+{
+};
+
+}}
+
+#include "detail/gpumat.hpp"
+
+#endif

modules/cudev/include/opencv2/cudev/ptr2d/interpolation.hpp

@@ -0,0 +1,385 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_PTR2D_INTERPOLATION_HPP__
+#define __OPENCV_CUDEV_PTR2D_INTERPOLATION_HPP__
+
+#include "../common.hpp"
+#include "../util/vec_traits.hpp"
+#include "../util/saturate_cast.hpp"
+#include "../util/type_traits.hpp"
+#include "../util/limits.hpp"
+#include "traits.hpp"
+
+namespace cv { namespace cudev {
+
+// Nearest
+
+template <class SrcPtr> struct NearestInterPtr
+{
+    typedef typename PtrTraits<SrcPtr>::value_type value_type;
+    typedef float                                  index_type;
+
+    SrcPtr src;
+
+    __device__ __forceinline__ typename PtrTraits<SrcPtr>::value_type operator ()(float y, float x) const
+    {
+        return src(__float2int_rn(y), __float2int_rn(x));
+    }
+};
+
+template <class SrcPtr> struct NearestInterPtrSz : NearestInterPtr<SrcPtr>
+{
+    int rows, cols;
+};
+
+template <class SrcPtr>
+__host__ NearestInterPtrSz<typename PtrTraits<SrcPtr>::ptr_type> interNearest(const SrcPtr& src)
+{
+    NearestInterPtrSz<typename PtrTraits<SrcPtr>::ptr_type> i;
+    i.src = shrinkPtr(src);
+    i.rows = getRows(src);
+    i.cols = getCols(src);
+    return i;
+}
+
+template <class SrcPtr> struct PtrTraits< NearestInterPtrSz<SrcPtr> > : PtrTraitsBase<NearestInterPtrSz<SrcPtr>, NearestInterPtr<SrcPtr> >
+{
+};
+
+// Linear
+
+template <typename SrcPtr> struct LinearInterPtr
+{
+    typedef typename PtrTraits<SrcPtr>::value_type value_type;
+    typedef float                                  index_type;
+
+    SrcPtr src;
+
+    __device__ typename PtrTraits<SrcPtr>::value_type operator ()(float y, float x) const
+    {
+        typedef typename PtrTraits<SrcPtr>::value_type src_type;
+        typedef typename VecTraits<src_type>::elem_type src_elem_type;
+        typedef typename LargerType<float, src_elem_type>::type work_elem_type;
+        typedef typename MakeVec<work_elem_type, VecTraits<src_type>::cn>::type work_type;
+
+        work_type out = VecTraits<work_type>::all(0);
+
+        const int x1 = __float2int_rd(x);
+        const int y1 = __float2int_rd(y);
+        const int x2 = x1 + 1;
+        const int y2 = y1 + 1;
+
+        typename PtrTraits<SrcPtr>::value_type src_reg = src(y1, x1);
+        out = out + src_reg * static_cast<work_elem_type>((x2 - x) * (y2 - y));
+
+        src_reg = src(y1, x2);
+        out = out + src_reg * static_cast<work_elem_type>((x - x1) * (y2 - y));
+
+        src_reg = src(y2, x1);
+        out = out + src_reg * static_cast<work_elem_type>((x2 - x) * (y - y1));
+
+        src_reg = src(y2, x2);
+        out = out + src_reg * static_cast<work_elem_type>((x - x1) * (y - y1));
+
+        return saturate_cast<typename PtrTraits<SrcPtr>::value_type>(out);
+    }
+};
+
+template <class SrcPtr> struct LinearInterPtrSz : LinearInterPtr<SrcPtr>
+{
+    int rows, cols;
+};
+
+template <class SrcPtr>
+__host__ LinearInterPtrSz<typename PtrTraits<SrcPtr>::ptr_type> interLinear(const SrcPtr& src)
+{
+    LinearInterPtrSz<typename PtrTraits<SrcPtr>::ptr_type> i;
+    i.src = shrinkPtr(src);
+    i.rows = getRows(src);
+    i.cols = getCols(src);
+    return i;
+}
+
+template <class SrcPtr> struct PtrTraits< LinearInterPtrSz<SrcPtr> > : PtrTraitsBase<LinearInterPtrSz<SrcPtr>, LinearInterPtr<SrcPtr> >
+{
+};
+
+// Cubic
+
+template <typename SrcPtr> struct CubicInterPtr
+{
+    typedef typename PtrTraits<SrcPtr>::value_type value_type;
+    typedef float                                  index_type;
+
+    SrcPtr src;
+
+    __device__ static float bicubicCoeff(float x_)
+    {
+        float x = ::fabsf(x_);
+        if (x <= 1.0f)
+        {
+            return x * x * (1.5f * x - 2.5f) + 1.0f;
+        }
+        else if (x < 2.0f)
+        {
+            return x * (x * (-0.5f * x + 2.5f) - 4.0f) + 2.0f;
+        }
+        else
+        {
+            return 0.0f;
+        }
+    }
+
+    __device__ typename PtrTraits<SrcPtr>::value_type operator ()(float y, float x) const
+    {
+        typedef typename PtrTraits<SrcPtr>::value_type src_type;
+        typedef typename VecTraits<src_type>::elem_type src_elem_type;
+        typedef typename LargerType<float, src_elem_type>::type work_elem_type;
+        typedef typename MakeVec<work_elem_type, VecTraits<src_type>::cn>::type work_type;
+
+        const float xmin = ::ceilf(x - 2.0f);
+        const float xmax = ::floorf(x + 2.0f);
+
+        const float ymin = ::ceilf(y - 2.0f);
+        const float ymax = ::floorf(y + 2.0f);
+
+        work_type sum = VecTraits<work_type>::all(0);
+        float wsum = 0.0f;
+
+        for (float cy = ymin; cy <= ymax; cy += 1.0f)
+        {
+            for (float cx = xmin; cx <= xmax; cx += 1.0f)
+            {
+                typename PtrTraits<SrcPtr>::value_type src_reg = src(__float2int_rd(cy), __float2int_rd(cx));
+                const float w = bicubicCoeff(x - cx) * bicubicCoeff(y - cy);
+
+                sum = sum + static_cast<work_elem_type>(w) * src_reg;
+                wsum += w;
+            }
+        }
+
+        work_type res = (wsum > numeric_limits<float>::epsilon()) ? VecTraits<work_type>::all(0) : sum / static_cast<work_elem_type>(wsum);
+
+        return saturate_cast<typename PtrTraits<SrcPtr>::value_type>(res);
+    }
+};
+
+template <class SrcPtr> struct CubicInterPtrSz : CubicInterPtr<SrcPtr>
+{
+    int rows, cols;
+};
+
+template <class SrcPtr>
+__host__ CubicInterPtrSz<typename PtrTraits<SrcPtr>::ptr_type> interCubic(const SrcPtr& src)
+{
+    CubicInterPtrSz<typename PtrTraits<SrcPtr>::ptr_type> i;
+    i.src = shrinkPtr(src);
+    i.rows = getRows(src);
+    i.cols = getCols(src);
+    return i;
+}
+
+template <class SrcPtr> struct PtrTraits< CubicInterPtrSz<SrcPtr> > : PtrTraitsBase<CubicInterPtrSz<SrcPtr>, CubicInterPtr<SrcPtr> >
+{
+};
+
+// IntegerArea
+
+template <typename SrcPtr> struct IntegerAreaInterPtr
+{
+    typedef typename PtrTraits<SrcPtr>::value_type value_type;
+    typedef float                                  index_type;
+
+    SrcPtr src;
+    int area_width, area_height;
+
+    __device__ typename PtrTraits<SrcPtr>::value_type operator ()(float y, float x) const
+    {
+        typedef typename PtrTraits<SrcPtr>::value_type src_type;
+        typedef typename VecTraits<src_type>::elem_type src_elem_type;
+        typedef typename LargerType<float, src_elem_type>::type work_elem_type;
+        typedef typename MakeVec<work_elem_type, VecTraits<src_type>::cn>::type work_type;
+
+        const int sx1 = __float2int_rd(x);
+        const int sx2 = sx1 + area_width;
+
+        const int sy1 = __float2int_rd(y);
+        const int sy2 = sy1 + area_height;
+
+        work_type out = VecTraits<work_type>::all(0);
+
+        for (int dy = sy1; dy < sy2; ++dy)
+        {
+            for (int dx = sx1; dx < sx2; ++dx)
+            {
+                out = out + saturate_cast<work_type>(src(dy, dx));
+            }
+        }
+
+        const work_elem_type scale = 1.0f / (area_width * area_height);
+
+        return saturate_cast<typename PtrTraits<SrcPtr>::value_type>(out * scale);
+    }
+};
+
+template <class SrcPtr> struct IntegerAreaInterPtrSz : IntegerAreaInterPtr<SrcPtr>
+{
+    int rows, cols;
+};
+
+template <class SrcPtr>
+__host__ IntegerAreaInterPtrSz<typename PtrTraits<SrcPtr>::ptr_type> interArea(const SrcPtr& src, Size areaSize)
+{
+    IntegerAreaInterPtrSz<typename PtrTraits<SrcPtr>::ptr_type> i;
+    i.src = shrinkPtr(src);
+    i.area_width = areaSize.width;
+    i.area_height = areaSize.height;
+    i.rows = getRows(src);
+    i.cols = getCols(src);
+    return i;
+}
+
+template <class SrcPtr> struct PtrTraits< IntegerAreaInterPtrSz<SrcPtr> > : PtrTraitsBase<IntegerAreaInterPtrSz<SrcPtr>, IntegerAreaInterPtr<SrcPtr> >
+{
+};
+
+// CommonArea
+
+template <typename SrcPtr> struct CommonAreaInterPtr
+{
+    typedef typename PtrTraits<SrcPtr>::value_type value_type;
+    typedef float                                  index_type;
+
+    SrcPtr src;
+    float area_width, area_height;
+
+    __device__ typename PtrTraits<SrcPtr>::value_type operator ()(float y, float x) const
+    {
+        typedef typename PtrTraits<SrcPtr>::value_type src_type;
+        typedef typename VecTraits<src_type>::elem_type src_elem_type;
+        typedef typename LargerType<float, src_elem_type>::type work_elem_type;
+        typedef typename MakeVec<work_elem_type, VecTraits<src_type>::cn>::type work_type;
+
+        const float fsx1 = x;
+        const float fsx2 = fsx1 + area_width;
+
+        const int sx1 = __float2int_rd(fsx1);
+        const int sx2 = __float2int_ru(fsx2);
+
+        const float fsy1 = y;
+        const float fsy2 = fsy1 + area_height;
+
+        const int sy1 = __float2int_rd(fsy1);
+        const int sy2 = __float2int_ru(fsy2);
+
+        work_type out = VecTraits<work_type>::all(0);
+
+        for (int dy = sy1; dy < sy2; ++dy)
+        {
+            for (int dx = sx1; dx < sx2; ++dx)
+                out = out + saturate_cast<work_type>(src(dy, dx));
+
+            if (sx1 > fsx1)
+                out = out + saturate_cast<work_type>(src(dy, sx1 - 1)) * static_cast<work_elem_type>(sx1 - fsx1);
+
+            if (sx2 < fsx2)
+                out = out + saturate_cast<work_type>(src(dy, sx2)) * static_cast<work_elem_type>(fsx2 - sx2);
+        }
+
+        if (sy1 > fsy1)
+        {
+            for (int dx = sx1; dx < sx2; ++dx)
+                out = out + saturate_cast<work_type>(src(sy1 - 1, dx)) * static_cast<work_elem_type>(sy1 - fsy1);
+        }
+
+        if (sy2 < fsy2)
+        {
+            for (int dx = sx1; dx < sx2; ++dx)
+                out = out + saturate_cast<work_type>(src(sy2, dx)) * static_cast<work_elem_type>(fsy2 - sy2);
+        }
+
+        if ((sy1 > fsy1) && (sx1 > fsx1))
+            out = out + saturate_cast<work_type>(src(sy1 - 1, sx1 - 1)) * static_cast<work_elem_type>((sy1 - fsy1) * (sx1 - fsx1));
+
+        if ((sy1 > fsy1) && (sx2 < fsx2))
+            out = out + saturate_cast<work_type>(src(sy1 - 1, sx2)) * static_cast<work_elem_type>((sy1 - fsy1) * (fsx2 - sx2));
+
+        if ((sy2 < fsy2) && (sx2 < fsx2))
+            out = out + saturate_cast<work_type>(src(sy2, sx2)) * static_cast<work_elem_type>((fsy2 - sy2) * (fsx2 - sx2));
+
+        if ((sy2 < fsy2) && (sx1 > fsx1))
+            out = out + saturate_cast<work_type>(src(sy2, sx1 - 1)) * static_cast<work_elem_type>((fsy2 - sy2) * (sx1 - fsx1));
+
+        const work_elem_type scale = 1.0f / (area_width * area_height);
+
+        return saturate_cast<typename PtrTraits<SrcPtr>::value_type>(out * scale);
+    }
+};
+
+template <class SrcPtr> struct CommonAreaInterPtrSz : CommonAreaInterPtr<SrcPtr>
+{
+    int rows, cols;
+};
+
+template <class SrcPtr>
+__host__ CommonAreaInterPtrSz<typename PtrTraits<SrcPtr>::ptr_type> interArea(const SrcPtr& src, Size2f areaSize)
+{
+    CommonAreaInterPtrSz<typename PtrTraits<SrcPtr>::ptr_type> i;
+    i.src = shrinkPtr(src);
+    i.area_width = areaSize.width;
+    i.area_height = areaSize.height;
+    i.rows = getRows(src);
+    i.cols = getCols(src);
+    return i;
+}
+
+template <class SrcPtr> struct PtrTraits< CommonAreaInterPtrSz<SrcPtr> > : PtrTraitsBase<CommonAreaInterPtrSz<SrcPtr>, CommonAreaInterPtr<SrcPtr> >
+{
+};
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/ptr2d/lut.hpp

@@ -0,0 +1,100 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_PTR2D_LUT_HPP__
+#define __OPENCV_CUDEV_PTR2D_LUT_HPP__
+
+#include "../common.hpp"
+#include "../grid/copy.hpp"
+#include "traits.hpp"
+#include "gpumat.hpp"
+
+namespace cv { namespace cudev {
+
+template <class SrcPtr, class TablePtr> struct LutPtr
+{
+    typedef typename PtrTraits<TablePtr>::value_type value_type;
+    typedef typename PtrTraits<SrcPtr>::index_type   index_type;
+
+    SrcPtr src;
+    TablePtr tbl;
+
+    __device__ __forceinline__ typename PtrTraits<TablePtr>::value_type operator ()(typename PtrTraits<SrcPtr>::index_type y, typename PtrTraits<SrcPtr>::index_type x) const
+    {
+        return tbl(0, src(y, x));
+    }
+};
+
+template <class SrcPtr, class TablePtr> struct LutPtrSz : LutPtr<SrcPtr, TablePtr>
+{
+    int rows, cols;
+
+    template <typename T>
+    __host__ void assignTo(GpuMat_<T>& dst, Stream& stream = Stream::Null()) const
+    {
+        gridCopy(*this, dst, stream);
+    }
+};
+
+template <class SrcPtr, class TablePtr>
+__host__ LutPtrSz<typename PtrTraits<SrcPtr>::ptr_type, typename PtrTraits<TablePtr>::ptr_type> lutPtr(const SrcPtr& src, const TablePtr& tbl)
+{
+    CV_Assert( getRows(tbl) == 1 );
+
+    LutPtrSz<typename PtrTraits<SrcPtr>::ptr_type, typename PtrTraits<TablePtr>::ptr_type> ptr;
+    ptr.src = shrinkPtr(src);
+    ptr.tbl = shrinkPtr(tbl);
+    ptr.rows = getRows(src);
+    ptr.cols = getCols(src);
+    return ptr;
+}
+
+template <class SrcPtr, class TablePtr> struct PtrTraits< LutPtrSz<SrcPtr, TablePtr> > : PtrTraitsBase<LutPtrSz<SrcPtr, TablePtr>, LutPtr<SrcPtr, TablePtr> >
+{
+};
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/ptr2d/mask.hpp

@@ -0,0 +1,67 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_PTR2D_MASK_HPP__
+#define __OPENCV_CUDEV_PTR2D_MASK_HPP__
+
+#include "../common.hpp"
+#include "traits.hpp"
+
+namespace cv { namespace cudev {
+
+struct WithOutMask
+{
+    typedef bool value_type;
+    typedef int  index_type;
+
+    __device__ __forceinline__ bool operator ()(int, int) const
+    {
+        return true;
+    }
+};
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/ptr2d/remap.hpp

@@ -0,0 +1,154 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_PTR2D_REMAP_HPP__
+#define __OPENCV_CUDEV_PTR2D_REMAP_HPP__
+
+#include "opencv2/core/base.hpp"
+#include "../common.hpp"
+#include "../grid/copy.hpp"
+#include "traits.hpp"
+#include "gpumat.hpp"
+
+namespace cv { namespace cudev {
+
+template <class SrcPtr, class MapPtr> struct RemapPtr1
+{
+    typedef typename PtrTraits<SrcPtr>::value_type value_type;
+    typedef typename PtrTraits<MapPtr>::index_type index_type;
+
+    SrcPtr src;
+    MapPtr map;
+
+    __device__ __forceinline__ typename PtrTraits<SrcPtr>::value_type operator ()(typename PtrTraits<MapPtr>::index_type y, typename PtrTraits<MapPtr>::index_type x) const
+    {
+        const typename PtrTraits<MapPtr>::value_type coord = map(y, x);
+        return src(coord.y, coord.x);
+    }
+};
+
+template <class SrcPtr, class MapXPtr, class MapYPtr> struct RemapPtr2
+{
+    typedef typename PtrTraits<SrcPtr>::value_type  value_type;
+    typedef typename PtrTraits<MapXPtr>::index_type index_type;
+
+    SrcPtr src;
+    MapXPtr mapx;
+    MapYPtr mapy;
+
+    __device__ __forceinline__ typename PtrTraits<SrcPtr>::value_type operator ()(typename PtrTraits<MapXPtr>::index_type y, typename PtrTraits<MapXPtr>::index_type x) const
+    {
+        const typename PtrTraits<MapXPtr>::value_type nx = mapx(y, x);
+        const typename PtrTraits<MapYPtr>::value_type ny = mapy(y, x);
+        return src(ny, nx);
+    }
+};
+
+template <class SrcPtr, class MapPtr> struct RemapPtr1Sz : RemapPtr1<SrcPtr, MapPtr>
+{
+    int rows, cols;
+
+    template <typename T>
+    __host__ void assignTo(GpuMat_<T>& dst, Stream& stream = Stream::Null()) const
+    {
+        gridCopy(*this, dst, stream);
+    }
+};
+
+template <class SrcPtr, class MapXPtr, class MapYPtr> struct RemapPtr2Sz : RemapPtr2<SrcPtr, MapXPtr, MapYPtr>
+{
+    int rows, cols;
+
+    template <typename T>
+    __host__ void assignTo(GpuMat_<T>& dst, Stream& stream = Stream::Null()) const
+    {
+        gridCopy(*this, dst, stream);
+    }
+};
+
+template <class SrcPtr, class MapPtr>
+__host__ RemapPtr1Sz<typename PtrTraits<SrcPtr>::ptr_type, typename PtrTraits<MapPtr>::ptr_type>
+remapPtr(const SrcPtr& src, const MapPtr& map)
+{
+    const int rows = getRows(map);
+    const int cols = getCols(map);
+
+    RemapPtr1Sz<typename PtrTraits<SrcPtr>::ptr_type, typename PtrTraits<MapPtr>::ptr_type> r;
+    r.src = shrinkPtr(src);
+    r.map = shrinkPtr(map);
+    r.rows = rows;
+    r.cols = cols;
+    return r;
+}
+
+template <class SrcPtr, class MapXPtr, class MapYPtr>
+__host__ RemapPtr2Sz<typename PtrTraits<SrcPtr>::ptr_type, typename PtrTraits<MapXPtr>::ptr_type, typename PtrTraits<MapYPtr>::ptr_type>
+remapPtr(const SrcPtr& src, const MapXPtr& mapx, const MapYPtr& mapy)
+{
+    const int rows = getRows(mapx);
+    const int cols = getCols(mapx);
+
+    CV_Assert( getRows(mapy) == rows && getCols(mapy) == cols );
+
+    RemapPtr2Sz<typename PtrTraits<SrcPtr>::ptr_type, typename PtrTraits<MapXPtr>::ptr_type, typename PtrTraits<MapYPtr>::ptr_type> r;
+    r.src = shrinkPtr(src);
+    r.mapx = shrinkPtr(mapx);
+    r.mapy = shrinkPtr(mapy);
+    r.rows = rows;
+    r.cols = cols;
+    return r;
+}
+
+template <class SrcPtr, class MapPtr> struct PtrTraits< RemapPtr1Sz<SrcPtr, MapPtr> > : PtrTraitsBase<RemapPtr1Sz<SrcPtr, MapPtr>, RemapPtr1<SrcPtr, MapPtr> >
+{
+};
+
+template <class SrcPtr, class MapXPtr, class MapYPtr> struct PtrTraits< RemapPtr2Sz<SrcPtr, MapXPtr, MapYPtr> > : PtrTraitsBase<RemapPtr2Sz<SrcPtr, MapXPtr, MapYPtr>, RemapPtr2<SrcPtr, MapXPtr, MapYPtr> >
+{
+};
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/ptr2d/resize.hpp

@@ -0,0 +1,103 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_PTR2D_RESIZE_HPP__
+#define __OPENCV_CUDEV_PTR2D_RESIZE_HPP__
+
+#include "opencv2/core/base.hpp"
+#include "../common.hpp"
+#include "../grid/copy.hpp"
+#include "traits.hpp"
+#include "gpumat.hpp"
+
+namespace cv { namespace cudev {
+
+template <class SrcPtr> struct ResizePtr
+{
+    typedef typename PtrTraits<SrcPtr>::value_type value_type;
+    typedef typename PtrTraits<SrcPtr>::index_type index_type;
+
+    SrcPtr src;
+    float fx, fy;
+
+    __device__ __forceinline__ typename PtrTraits<SrcPtr>::value_type operator ()(typename PtrTraits<SrcPtr>::index_type y, typename PtrTraits<SrcPtr>::index_type x) const
+    {
+        const float yn = static_cast<float>(y * fy);
+        const float xn = static_cast<float>(x * fx);
+
+        return src(yn, xn);
+    }
+};
+
+template <class SrcPtr> struct ResizePtrSz : ResizePtr<SrcPtr>
+{
+    int rows, cols;
+
+    template <typename T>
+    __host__ void assignTo(GpuMat_<T>& dst, Stream& stream = Stream::Null()) const
+    {
+        gridCopy(*this, dst, stream);
+    }
+};
+
+template <class SrcPtr>
+__host__ ResizePtrSz<typename PtrTraits<SrcPtr>::ptr_type> resizePtr(const SrcPtr& src, float fx, float fy)
+{
+    ResizePtrSz<typename PtrTraits<SrcPtr>::ptr_type> r;
+    r.src = shrinkPtr(src);
+    r.fx = 1.0f / fx;
+    r.fy = 1.0f / fy;
+    r.rows = cv::saturate_cast<int>(getRows(src) * fy);
+    r.cols = cv::saturate_cast<int>(getCols(src) * fx);
+    return r;
+}
+
+template <class SrcPtr> struct PtrTraits< ResizePtrSz<SrcPtr> > : PtrTraitsBase<ResizePtrSz<SrcPtr>, ResizePtr<SrcPtr> >
+{
+};
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/ptr2d/texture.hpp

@@ -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.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_PTR2D_TEXTURE_HPP__
+#define __OPENCV_CUDEV_PTR2D_TEXTURE_HPP__
+
+#include <cstring>
+#include "../common.hpp"
+#include "glob.hpp"
+#include "gpumat.hpp"
+#include "traits.hpp"
+
+namespace cv { namespace cudev {
+
+template <typename T> struct TexturePtr
+{
+    typedef T     value_type;
+    typedef float index_type;
+
+    cudaTextureObject_t texObj;
+
+    __device__ __forceinline__ T operator ()(float y, float x) const
+    {
+        return tex2D<T>(texObj, x, y);
+    }
+};
+
+template <typename T> struct Texture : TexturePtr<T>
+{
+    int rows, cols;
+
+    __host__ explicit Texture(const GlobPtrSz<T>& mat,
+                              bool normalizedCoords = false,
+                              cudaTextureFilterMode filterMode = cudaFilterModePoint,
+                              cudaTextureAddressMode addressMode = cudaAddressModeClamp)
+    {
+        CV_Assert( deviceSupports(FEATURE_SET_COMPUTE_30) );
+
+        rows = mat.rows;
+        cols = mat.cols;
+
+        cudaResourceDesc texRes;
+        std::memset(&texRes, 0, sizeof(texRes));
+        texRes.resType = cudaResourceTypePitch2D;
+        texRes.res.pitch2D.devPtr = mat.data;
+        texRes.res.pitch2D.height = mat.rows;
+        texRes.res.pitch2D.width = mat.cols;
+        texRes.res.pitch2D.pitchInBytes = mat.step;
+        texRes.res.pitch2D.desc = cudaCreateChannelDesc<T>();
+
+        cudaTextureDesc texDescr;
+        std::memset(&texDescr, 0, sizeof(texDescr));
+        texDescr.addressMode[0] = addressMode;
+        texDescr.addressMode[1] = addressMode;
+        texDescr.addressMode[2] = addressMode;
+        texDescr.filterMode = filterMode;
+        texDescr.readMode = cudaReadModeElementType;
+        texDescr.normalizedCoords = normalizedCoords;
+
+        CV_CUDEV_SAFE_CALL( cudaCreateTextureObject(&this->texObj, &texRes, &texDescr, 0) );
+    }
+
+    __host__ explicit Texture(const GpuMat_<T>& mat,
+                              bool normalizedCoords = false,
+                              cudaTextureFilterMode filterMode = cudaFilterModePoint,
+                              cudaTextureAddressMode addressMode = cudaAddressModeClamp)
+    {
+        CV_Assert( deviceSupports(FEATURE_SET_COMPUTE_30) );
+
+        rows = mat.rows;
+        cols = mat.cols;
+
+        cudaResourceDesc texRes;
+        std::memset(&texRes, 0, sizeof(texRes));
+        texRes.resType = cudaResourceTypePitch2D;
+        texRes.res.pitch2D.devPtr = mat.data;
+        texRes.res.pitch2D.height = mat.rows;
+        texRes.res.pitch2D.width = mat.cols;
+        texRes.res.pitch2D.pitchInBytes = mat.step;
+        texRes.res.pitch2D.desc = cudaCreateChannelDesc<T>();
+
+        cudaTextureDesc texDescr;
+        std::memset(&texDescr, 0, sizeof(texDescr));
+        texDescr.addressMode[0] = addressMode;
+        texDescr.addressMode[1] = addressMode;
+        texDescr.addressMode[2] = addressMode;
+        texDescr.filterMode = filterMode;
+        texDescr.readMode = cudaReadModeElementType;
+        texDescr.normalizedCoords = normalizedCoords;
+
+        CV_CUDEV_SAFE_CALL( cudaCreateTextureObject(&this->texObj, &texRes, &texDescr, 0) );
+    }
+
+    __host__ ~Texture()
+    {
+        cudaDestroyTextureObject(this->texObj);
+    }
+};
+
+template <typename T> struct PtrTraits< Texture<T> > : PtrTraitsBase<Texture<T>, TexturePtr<T> >
+{
+};
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/ptr2d/traits.hpp

@@ -0,0 +1,101 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_PTR2D_TRAITS_HPP__
+#define __OPENCV_CUDEV_PTR2D_TRAITS_HPP__
+
+#include "../common.hpp"
+
+namespace cv { namespace cudev {
+
+template <class Ptr2DSz, class Ptr2D> struct PtrTraitsBase
+{
+    typedef Ptr2DSz ptr_sz_type;
+    typedef Ptr2D   ptr_type;
+
+    typedef typename Ptr2D::value_type value_type;
+    typedef typename Ptr2D::index_type index_type;
+
+    __host__ static Ptr2D shrinkPtr(const Ptr2DSz& ptr)
+    {
+        return ptr;
+    }
+
+    __host__ static int getRows(const Ptr2DSz& ptr)
+    {
+        return ptr.rows;
+    }
+
+    __host__ static int getCols(const Ptr2DSz& ptr)
+    {
+        return ptr.cols;
+    }
+};
+
+template <class Ptr2DSz> struct PtrTraits : PtrTraitsBase<Ptr2DSz, Ptr2DSz>
+{
+};
+
+template <class Ptr2DSz>
+__host__ typename PtrTraits<Ptr2DSz>::ptr_type shrinkPtr(const Ptr2DSz& ptr)
+{
+    return PtrTraits<Ptr2DSz>::shrinkPtr(ptr);
+}
+
+template <class Ptr2DSz>
+__host__ int getRows(const Ptr2DSz& ptr)
+{
+    return PtrTraits<Ptr2DSz>::getRows(ptr);
+}
+
+template <class Ptr2DSz>
+__host__ int getCols(const Ptr2DSz& ptr)
+{
+    return PtrTraits<Ptr2DSz>::getCols(ptr);
+}
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/ptr2d/transform.hpp

@@ -0,0 +1,151 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_PTR2D_TRANSFORM_HPP__
+#define __OPENCV_CUDEV_PTR2D_TRANSFORM_HPP__
+
+#include "../common.hpp"
+#include "../grid/copy.hpp"
+#include "traits.hpp"
+#include "gpumat.hpp"
+
+namespace cv { namespace cudev {
+
+// UnaryTransformPtr
+
+template <class SrcPtr, class Op> struct UnaryTransformPtr
+{
+    typedef typename Op::result_type               value_type;
+    typedef typename PtrTraits<SrcPtr>::index_type index_type;
+
+    SrcPtr src;
+    Op op;
+
+    __device__ __forceinline__ typename Op::result_type operator ()(typename PtrTraits<SrcPtr>::index_type y, typename PtrTraits<SrcPtr>::index_type x) const
+    {
+        return op(src(y, x));
+    }
+};
+
+template <class SrcPtr, class Op> struct UnaryTransformPtrSz : UnaryTransformPtr<SrcPtr, Op>
+{
+    int rows, cols;
+
+    template <typename T>
+    __host__ void assignTo(GpuMat_<T>& dst, Stream& stream = Stream::Null()) const
+    {
+        gridCopy(*this, dst, stream);
+    }
+};
+
+template <class SrcPtr, class Op>
+__host__ UnaryTransformPtrSz<typename PtrTraits<SrcPtr>::ptr_type, Op>
+transformPtr(const SrcPtr& src, const Op& op)
+{
+    UnaryTransformPtrSz<typename PtrTraits<SrcPtr>::ptr_type, Op> ptr;
+    ptr.src = shrinkPtr(src);
+    ptr.op = op;
+    ptr.rows = getRows(src);
+    ptr.cols = getCols(src);
+    return ptr;
+}
+
+template <class SrcPtr, class Op> struct PtrTraits< UnaryTransformPtrSz<SrcPtr, Op> > : PtrTraitsBase<UnaryTransformPtrSz<SrcPtr, Op>, UnaryTransformPtr<SrcPtr, Op> >
+{
+};
+
+// BinaryTransformPtr
+
+template <class Src1Ptr, class Src2Ptr, class Op> struct BinaryTransformPtr
+{
+    typedef typename Op::result_type                value_type;
+    typedef typename PtrTraits<Src1Ptr>::index_type index_type;
+
+    Src1Ptr src1;
+    Src2Ptr src2;
+    Op op;
+
+    __device__ __forceinline__ typename Op::result_type operator ()(typename PtrTraits<Src1Ptr>::index_type y, typename PtrTraits<Src1Ptr>::index_type x) const
+    {
+        return op(src1(y, x), src2(y, x));
+    }
+};
+
+template <class Src1Ptr, class Src2Ptr, class Op> struct BinaryTransformPtrSz : BinaryTransformPtr<Src1Ptr, Src2Ptr, Op>
+{
+    int rows, cols;
+
+    template <typename T>
+    __host__ void assignTo(GpuMat_<T>& dst, Stream& stream = Stream::Null()) const
+    {
+        gridCopy(*this, dst, stream);
+    }
+};
+
+template <class Src1Ptr, class Src2Ptr, class Op>
+__host__ BinaryTransformPtrSz<typename PtrTraits<Src1Ptr>::ptr_type, typename PtrTraits<Src2Ptr>::ptr_type, Op>
+transformPtr(const Src1Ptr& src1, const Src2Ptr& src2, const Op& op)
+{
+    const int rows = getRows(src1);
+    const int cols = getCols(src1);
+
+    CV_Assert( getRows(src2) == rows && getCols(src2) == cols );
+
+    BinaryTransformPtrSz<typename PtrTraits<Src1Ptr>::ptr_type, typename PtrTraits<Src2Ptr>::ptr_type, Op> ptr;
+    ptr.src1 = shrinkPtr(src1);
+    ptr.src2 = shrinkPtr(src2);
+    ptr.op = op;
+    ptr.rows = rows;
+    ptr.cols = cols;
+    return ptr;
+}
+
+template <class Src1Ptr, class Src2Ptr, class Op> struct PtrTraits< BinaryTransformPtrSz<Src1Ptr, Src2Ptr, Op> > : PtrTraitsBase<BinaryTransformPtrSz<Src1Ptr, Src2Ptr, Op>, BinaryTransformPtr<Src1Ptr, Src2Ptr, Op> >
+{
+};
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/ptr2d/warping.hpp

@@ -0,0 +1,152 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_PTR2D_WARPING_HPP__
+#define __OPENCV_CUDEV_PTR2D_WARPING_HPP__
+
+#include "../common.hpp"
+#include "traits.hpp"
+#include "remap.hpp"
+#include "gpumat.hpp"
+
+namespace cv { namespace cudev {
+
+// affine
+
+struct AffineMapPtr
+{
+    typedef float2 value_type;
+    typedef float  index_type;
+
+    const float* warpMat;
+
+    __device__ __forceinline__ float2 operator ()(float y, float x) const
+    {
+        const float xcoo = warpMat[0] * x + warpMat[1] * y + warpMat[2];
+        const float ycoo = warpMat[3] * x + warpMat[4] * y + warpMat[5];
+
+        return make_float2(xcoo, ycoo);
+    }
+};
+
+struct AffineMapPtrSz : AffineMapPtr
+{
+    int rows, cols;
+};
+
+template <> struct PtrTraits<AffineMapPtrSz> : PtrTraitsBase<AffineMapPtrSz, AffineMapPtr>
+{
+};
+
+__host__ static AffineMapPtrSz affineMap(Size dstSize, const GpuMat_<float>& warpMat)
+{
+    CV_Assert( warpMat.rows == 2 && warpMat.cols == 3 );
+    CV_Assert( warpMat.isContinuous() );
+
+    AffineMapPtrSz map;
+    map.warpMat = warpMat[0];
+    map.rows = dstSize.height;
+    map.cols = dstSize.width;
+    return map;
+}
+
+template <class SrcPtr>
+__host__ RemapPtr1Sz<typename PtrTraits<SrcPtr>::ptr_type, AffineMapPtr>
+warpAffinePtr(const SrcPtr& src, Size dstSize, const GpuMat_<float>& warpMat)
+{
+    return remapPtr(src, affineMap(dstSize, warpMat));
+}
+
+// perspective
+
+struct PerspectiveMapPtr
+{
+    typedef float2 value_type;
+    typedef float  index_type;
+
+    const float* warpMat;
+
+    __device__ __forceinline__ float2 operator ()(float y, float x) const
+    {
+        const float coeff = 1.0f / (warpMat[6] * x + warpMat[7] * y + warpMat[8]);
+
+        const float xcoo = coeff * (warpMat[0] * x + warpMat[1] * y + warpMat[2]);
+        const float ycoo = coeff * (warpMat[3] * x + warpMat[4] * y + warpMat[5]);
+
+        return make_float2(xcoo, ycoo);
+    }
+};
+
+struct PerspectiveMapPtrSz : PerspectiveMapPtr
+{
+    int rows, cols;
+};
+
+template <> struct PtrTraits<PerspectiveMapPtrSz> : PtrTraitsBase<PerspectiveMapPtrSz, PerspectiveMapPtr>
+{
+};
+
+__host__ static PerspectiveMapPtrSz perspectiveMap(Size dstSize, const GpuMat_<float>& warpMat)
+{
+    CV_Assert( warpMat.rows == 3 && warpMat.cols == 3 );
+    CV_Assert( warpMat.isContinuous() );
+
+    PerspectiveMapPtrSz map;
+    map.warpMat = warpMat[0];
+    map.rows = dstSize.height;
+    map.cols = dstSize.width;
+    return map;
+}
+
+template <class SrcPtr>
+__host__ RemapPtr1Sz<typename PtrTraits<SrcPtr>::ptr_type, PerspectiveMapPtr>
+warpPerspectivePtr(const SrcPtr& src, Size dstSize, const GpuMat_<float>& warpMat)
+{
+    return remapPtr(src, perspectiveMap(dstSize, warpMat));
+}
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/ptr2d/zip.hpp

@@ -0,0 +1,173 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_PTR2D_ZIP_HPP__
+#define __OPENCV_CUDEV_PTR2D_ZIP_HPP__
+
+#include "../common.hpp"
+#include "../util/tuple.hpp"
+#include "traits.hpp"
+
+namespace cv { namespace cudev {
+
+template <class PtrTuple> struct ZipPtr;
+
+template <class Ptr0, class Ptr1> struct ZipPtr< tuple<Ptr0, Ptr1> > : tuple<Ptr0, Ptr1>
+{
+    typedef tuple<typename PtrTraits<Ptr0>::value_type,
+                  typename PtrTraits<Ptr1>::value_type> value_type;
+    typedef typename PtrTraits<Ptr0>::index_type        index_type;
+
+    __host__ __device__ __forceinline__ ZipPtr() {}
+    __host__ __device__ __forceinline__ ZipPtr(const tuple<Ptr0, Ptr1>& t) : tuple<Ptr0, Ptr1>(t) {}
+
+    __device__ __forceinline__ value_type operator ()(index_type y, index_type x) const
+    {
+        return make_tuple(cv::cudev::get<0>(*this)(y, x), cv::cudev::get<1>(*this)(y, x));
+    }
+};
+
+template <class Ptr0, class Ptr1, class Ptr2> struct ZipPtr< tuple<Ptr0, Ptr1, Ptr2> > : tuple<Ptr0, Ptr1, Ptr2>
+{
+    typedef tuple<typename PtrTraits<Ptr0>::value_type,
+                  typename PtrTraits<Ptr1>::value_type,
+                  typename PtrTraits<Ptr2>::value_type> value_type;
+    typedef typename PtrTraits<Ptr0>::index_type        index_type;
+
+    __host__ __device__ __forceinline__ ZipPtr() {}
+    __host__ __device__ __forceinline__ ZipPtr(const tuple<Ptr0, Ptr1, Ptr2>& t) : tuple<Ptr0, Ptr1, Ptr2>(t) {}
+
+    __device__ __forceinline__ value_type operator ()(index_type y, index_type x) const
+    {
+        return make_tuple(cv::cudev::get<0>(*this)(y, x), cv::cudev::get<1>(*this)(y, x), cv::cudev::get<2>(*this)(y, x));
+    }
+};
+
+template <class Ptr0, class Ptr1, class Ptr2, class Ptr3> struct ZipPtr< tuple<Ptr0, Ptr1, Ptr2, Ptr3> > : tuple<Ptr0, Ptr1, Ptr2, Ptr3>
+{
+    typedef tuple<typename PtrTraits<Ptr0>::value_type,
+                  typename PtrTraits<Ptr1>::value_type,
+                  typename PtrTraits<Ptr2>::value_type,
+                  typename PtrTraits<Ptr3>::value_type> value_type;
+    typedef typename PtrTraits<Ptr0>::index_type        index_type;
+
+    __host__ __device__ __forceinline__ ZipPtr() {}
+    __host__ __device__ __forceinline__ ZipPtr(const tuple<Ptr0, Ptr1, Ptr2, Ptr3>& t) : tuple<Ptr0, Ptr1, Ptr2, Ptr3>(t) {}
+
+    __device__ __forceinline__ value_type operator ()(index_type y, index_type x) const
+    {
+        return make_tuple(cv::cudev::get<0>(*this)(y, x), cv::cudev::get<1>(*this)(y, x), cv::cudev::get<2>(*this)(y, x), cv::cudev::get<3>(*this)(y, x));
+    }
+};
+
+template <class PtrTuple> struct ZipPtrSz : ZipPtr<PtrTuple>
+{
+    int rows, cols;
+
+    __host__ __device__ __forceinline__ ZipPtrSz() {}
+    __host__ __device__ __forceinline__ ZipPtrSz(const PtrTuple& t) : ZipPtr<PtrTuple>(t) {}
+};
+
+template <class Ptr0, class Ptr1>
+__host__ ZipPtrSz< tuple<typename PtrTraits<Ptr0>::ptr_type, typename PtrTraits<Ptr1>::ptr_type> >
+zipPtr(const Ptr0& ptr0, const Ptr1& ptr1)
+{
+    const int rows = getRows(ptr0);
+    const int cols = getCols(ptr0);
+
+    CV_Assert( getRows(ptr1) == rows && getCols(ptr1) == cols );
+
+    ZipPtrSz< tuple<typename PtrTraits<Ptr0>::ptr_type, typename PtrTraits<Ptr1>::ptr_type> >
+            z(make_tuple(shrinkPtr(ptr0), shrinkPtr(ptr1)));
+    z.rows = rows;
+    z.cols = cols;
+
+    return z;
+}
+
+template <class Ptr0, class Ptr1, class Ptr2>
+__host__ ZipPtrSz< tuple<typename PtrTraits<Ptr0>::ptr_type, typename PtrTraits<Ptr1>::ptr_type, typename PtrTraits<Ptr2>::ptr_type> >
+zipPtr(const Ptr0& ptr0, const Ptr1& ptr1, const Ptr2& ptr2)
+{
+    const int rows = getRows(ptr0);
+    const int cols = getCols(ptr0);
+
+    CV_Assert( getRows(ptr1) == rows && getCols(ptr1) == cols );
+    CV_Assert( getRows(ptr2) == rows && getCols(ptr2) == cols );
+
+    ZipPtrSz< tuple<typename PtrTraits<Ptr0>::ptr_type, typename PtrTraits<Ptr1>::ptr_type, typename PtrTraits<Ptr2>::ptr_type> >
+            z(make_tuple(shrinkPtr(ptr0), shrinkPtr(ptr1), shrinkPtr(ptr2)));
+    z.rows = rows;
+    z.cols = cols;
+
+    return z;
+}
+
+template <class Ptr0, class Ptr1, class Ptr2, class Ptr3>
+__host__ ZipPtrSz< tuple<typename PtrTraits<Ptr0>::ptr_type, typename PtrTraits<Ptr1>::ptr_type, typename PtrTraits<Ptr2>::ptr_type, typename PtrTraits<Ptr3>::ptr_type> >
+zipPtr(const Ptr0& ptr0, const Ptr1& ptr1, const Ptr2& ptr2, const Ptr3& ptr3)
+{
+    const int rows = getRows(ptr0);
+    const int cols = getCols(ptr0);
+
+    CV_Assert( getRows(ptr1) == rows && getCols(ptr1) == cols );
+    CV_Assert( getRows(ptr2) == rows && getCols(ptr2) == cols );
+    CV_Assert( getRows(ptr3) == rows && getCols(ptr3) == cols );
+
+    ZipPtrSz< tuple<typename PtrTraits<Ptr0>::ptr_type, typename PtrTraits<Ptr1>::ptr_type, typename PtrTraits<Ptr2>::ptr_type, typename PtrTraits<Ptr3>::ptr_type> >
+            z(make_tuple(shrinkPtr(ptr0), shrinkPtr(ptr1), shrinkPtr(ptr2), shrinkPtr(ptr3)));
+    z.rows = rows;
+    z.cols = cols;
+
+    return z;
+}
+
+template <class PtrTuple> struct PtrTraits< ZipPtrSz<PtrTuple> > : PtrTraitsBase<ZipPtrSz<PtrTuple>, ZipPtr<PtrTuple> >
+{
+};
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/util/atomic.hpp

@@ -0,0 +1,156 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_UTIL_ATOMIC_HPP__
+#define __OPENCV_CUDEV_UTIL_ATOMIC_HPP__
+
+#include "../common.hpp"
+
+namespace cv { namespace cudev {
+
+// atomicAdd
+
+__device__ __forceinline__ int atomicAdd(int* address, int val)
+{
+    return ::atomicAdd(address, val);
+}
+
+__device__ __forceinline__ uint atomicAdd(uint* address, uint val)
+{
+    return ::atomicAdd(address, val);
+}
+
+__device__ __forceinline__ float atomicAdd(float* address, float val)
+{
+    return ::atomicAdd(address, val);
+}
+
+__device__ static double atomicAdd(double* address, double val)
+{
+    unsigned long long int* address_as_ull = (unsigned long long int*) address;
+    unsigned long long int old = *address_as_ull, assumed;
+    do {
+        assumed = old;
+        old = ::atomicCAS(address_as_ull, assumed,
+            __double_as_longlong(val + __longlong_as_double(assumed)));
+    } while (assumed != old);
+    return __longlong_as_double(old);
+}
+
+// atomicMin
+
+__device__ __forceinline__ int atomicMin(int* address, int val)
+{
+    return ::atomicMin(address, val);
+}
+
+__device__ __forceinline__ uint atomicMin(uint* address, uint val)
+{
+    return ::atomicMin(address, val);
+}
+
+__device__ static float atomicMin(float* address, float val)
+{
+    int* address_as_i = (int*) address;
+    int old = *address_as_i, assumed;
+    do {
+        assumed = old;
+        old = ::atomicCAS(address_as_i, assumed,
+            __float_as_int(::fminf(val, __int_as_float(assumed))));
+    } while (assumed != old);
+    return __int_as_float(old);
+}
+
+__device__ static double atomicMin(double* address, double val)
+{
+    unsigned long long int* address_as_ull = (unsigned long long int*) address;
+    unsigned long long int old = *address_as_ull, assumed;
+    do {
+        assumed = old;
+        old = ::atomicCAS(address_as_ull, assumed,
+            __double_as_longlong(::fmin(val, __longlong_as_double(assumed))));
+    } while (assumed != old);
+    return __longlong_as_double(old);
+}
+
+// atomicMax
+
+__device__ __forceinline__ int atomicMax(int* address, int val)
+{
+    return ::atomicMax(address, val);
+}
+
+__device__ __forceinline__ uint atomicMax(uint* address, uint val)
+{
+    return ::atomicMax(address, val);
+}
+
+__device__ static float atomicMax(float* address, float val)
+{
+    int* address_as_i = (int*) address;
+    int old = *address_as_i, assumed;
+    do {
+        assumed = old;
+        old = ::atomicCAS(address_as_i, assumed,
+            __float_as_int(::fmaxf(val, __int_as_float(assumed))));
+    } while (assumed != old);
+    return __int_as_float(old);
+}
+
+__device__ static double atomicMax(double* address, double val)
+{
+    unsigned long long int* address_as_ull = (unsigned long long int*) address;
+    unsigned long long int old = *address_as_ull, assumed;
+    do {
+        assumed = old;
+        old = ::atomicCAS(address_as_ull, assumed,
+            __double_as_longlong(::fmax(val, __longlong_as_double(assumed))));
+    } while (assumed != old);
+    return __longlong_as_double(old);
+}
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/util/detail/tuple.hpp

@@ -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.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_UTIL_TUPLE_DETAIL_HPP__
+#define __OPENCV_CUDEV_UTIL_TUPLE_DETAIL_HPP__
+
+#include <thrust/tuple.h>
+
+namespace cv { namespace cudev {
+
+namespace tuple_detail
+{
+    using thrust::tuple;
+    using thrust::tuple_size;
+    using thrust::get;
+    using thrust::tuple_element;
+    using thrust::make_tuple;
+    using thrust::tie;
+
+    template <class Tuple, int SIZE, template <typename T> class CvtOp> struct ConvertTuple;
+
+    template <class Tuple, template <typename T> class CvtOp> struct ConvertTuple<Tuple, 2, CvtOp>
+    {
+        typedef tuple<
+            typename CvtOp<typename tuple_element<0, Tuple>::type>::type,
+            typename CvtOp<typename tuple_element<1, Tuple>::type>::type
+        > type;
+    };
+
+    template <class Tuple, template <typename T> class CvtOp> struct ConvertTuple<Tuple, 3, CvtOp>
+    {
+        typedef tuple<
+            typename CvtOp<typename tuple_element<0, Tuple>::type>::type,
+            typename CvtOp<typename tuple_element<1, Tuple>::type>::type,
+            typename CvtOp<typename tuple_element<2, Tuple>::type>::type
+        > type;
+    };
+
+    template <class Tuple, template <typename T> class CvtOp> struct ConvertTuple<Tuple, 4, CvtOp>
+    {
+        typedef tuple<
+            typename CvtOp<typename tuple_element<0, Tuple>::type>::type,
+            typename CvtOp<typename tuple_element<1, Tuple>::type>::type,
+            typename CvtOp<typename tuple_element<2, Tuple>::type>::type,
+            typename CvtOp<typename tuple_element<3, Tuple>::type>::type
+        > type;
+    };
+
+    template <class Tuple, template <typename T> class CvtOp> struct ConvertTuple<Tuple, 5, CvtOp>
+    {
+        typedef tuple<
+            typename CvtOp<typename tuple_element<0, Tuple>::type>::type,
+            typename CvtOp<typename tuple_element<1, Tuple>::type>::type,
+            typename CvtOp<typename tuple_element<2, Tuple>::type>::type,
+            typename CvtOp<typename tuple_element<3, Tuple>::type>::type,
+            typename CvtOp<typename tuple_element<4, Tuple>::type>::type
+        > type;
+    };
+
+    template <class Tuple, template <typename T> class CvtOp> struct ConvertTuple<Tuple, 6, CvtOp>
+    {
+        typedef tuple<
+            typename CvtOp<typename tuple_element<0, Tuple>::type>::type,
+            typename CvtOp<typename tuple_element<1, Tuple>::type>::type,
+            typename CvtOp<typename tuple_element<2, Tuple>::type>::type,
+            typename CvtOp<typename tuple_element<3, Tuple>::type>::type,
+            typename CvtOp<typename tuple_element<4, Tuple>::type>::type,
+            typename CvtOp<typename tuple_element<5, Tuple>::type>::type
+        > type;
+    };
+
+    template <class Tuple, template <typename T> class CvtOp> struct ConvertTuple<Tuple, 7, CvtOp>
+    {
+        typedef tuple<
+            typename CvtOp<typename tuple_element<0, Tuple>::type>::type,
+            typename CvtOp<typename tuple_element<1, Tuple>::type>::type,
+            typename CvtOp<typename tuple_element<2, Tuple>::type>::type,
+            typename CvtOp<typename tuple_element<3, Tuple>::type>::type,
+            typename CvtOp<typename tuple_element<4, Tuple>::type>::type,
+            typename CvtOp<typename tuple_element<5, Tuple>::type>::type,
+            typename CvtOp<typename tuple_element<6, Tuple>::type>::type
+        > type;
+    };
+
+    template <class Tuple, template <typename T> class CvtOp> struct ConvertTuple<Tuple, 8, CvtOp>
+    {
+        typedef tuple<
+            typename CvtOp<typename tuple_element<0, Tuple>::type>::type,
+            typename CvtOp<typename tuple_element<1, Tuple>::type>::type,
+            typename CvtOp<typename tuple_element<2, Tuple>::type>::type,
+            typename CvtOp<typename tuple_element<3, Tuple>::type>::type,
+            typename CvtOp<typename tuple_element<4, Tuple>::type>::type,
+            typename CvtOp<typename tuple_element<5, Tuple>::type>::type,
+            typename CvtOp<typename tuple_element<6, Tuple>::type>::type,
+            typename CvtOp<typename tuple_element<7, Tuple>::type>::type
+        > type;
+    };
+
+    template <class Tuple, template <typename T> class CvtOp> struct ConvertTuple<Tuple, 9, CvtOp>
+    {
+        typedef tuple<
+            typename CvtOp<typename tuple_element<0, Tuple>::type>::type,
+            typename CvtOp<typename tuple_element<1, Tuple>::type>::type,
+            typename CvtOp<typename tuple_element<2, Tuple>::type>::type,
+            typename CvtOp<typename tuple_element<3, Tuple>::type>::type,
+            typename CvtOp<typename tuple_element<4, Tuple>::type>::type,
+            typename CvtOp<typename tuple_element<5, Tuple>::type>::type,
+            typename CvtOp<typename tuple_element<6, Tuple>::type>::type,
+            typename CvtOp<typename tuple_element<7, Tuple>::type>::type,
+            typename CvtOp<typename tuple_element<8, Tuple>::type>::type
+        > type;
+    };
+
+    template <class Tuple, template <typename T> class CvtOp> struct ConvertTuple<Tuple, 10, CvtOp>
+    {
+        typedef tuple<
+            typename CvtOp<typename tuple_element<0, Tuple>::type>::type,
+            typename CvtOp<typename tuple_element<1, Tuple>::type>::type,
+            typename CvtOp<typename tuple_element<2, Tuple>::type>::type,
+            typename CvtOp<typename tuple_element<3, Tuple>::type>::type,
+            typename CvtOp<typename tuple_element<4, Tuple>::type>::type,
+            typename CvtOp<typename tuple_element<5, Tuple>::type>::type,
+            typename CvtOp<typename tuple_element<6, Tuple>::type>::type,
+            typename CvtOp<typename tuple_element<7, Tuple>::type>::type,
+            typename CvtOp<typename tuple_element<8, Tuple>::type>::type,
+            typename CvtOp<typename tuple_element<9, Tuple>::type>::type
+        > type;
+    };
+}
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/util/detail/type_traits.hpp

@@ -0,0 +1,238 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_UTIL_TYPE_TRAITS_DETAIL_HPP__
+#define __OPENCV_CUDEV_UTIL_TYPE_TRAITS_DETAIL_HPP__
+
+#include "../../common.hpp"
+
+namespace cv { namespace cudev {
+
+namespace type_traits_detail
+{
+    template <typename T> struct IsSignedIntergral { enum {value = 0}; };
+    template <> struct IsSignedIntergral<schar> { enum {value = 1}; };
+    template <> struct IsSignedIntergral<short> { enum {value = 1}; };
+    template <> struct IsSignedIntergral<int> { enum {value = 1}; };
+
+    template <typename T> struct IsUnsignedIntegral { enum {value = 0}; };
+    template <> struct IsUnsignedIntegral<uchar> { enum {value = 1}; };
+    template <> struct IsUnsignedIntegral<ushort> { enum {value = 1}; };
+    template <> struct IsUnsignedIntegral<uint> { enum {value = 1}; };
+
+    template <typename T> struct IsIntegral { enum {value = IsSignedIntergral<T>::value || IsUnsignedIntegral<T>::value}; };
+    template <> struct IsIntegral<char> { enum {value = 1}; };
+    template <> struct IsIntegral<bool> { enum {value = 1}; };
+
+    template <typename T> struct IsFloat { enum {value = 0}; };
+    template <> struct IsFloat<float> { enum {value = 1}; };
+    template <> struct IsFloat<double> { enum {value = 1}; };
+
+    template <typename T> struct IsVec { enum {value = 0}; };
+    template <> struct IsVec<uchar1> { enum {value = 1}; };
+    template <> struct IsVec<uchar2> { enum {value = 1}; };
+    template <> struct IsVec<uchar3> { enum {value = 1}; };
+    template <> struct IsVec<uchar4> { enum {value = 1}; };
+    template <> struct IsVec<char1> { enum {value = 1}; };
+    template <> struct IsVec<char2> { enum {value = 1}; };
+    template <> struct IsVec<char3> { enum {value = 1}; };
+    template <> struct IsVec<char4> { enum {value = 1}; };
+    template <> struct IsVec<ushort1> { enum {value = 1}; };
+    template <> struct IsVec<ushort2> { enum {value = 1}; };
+    template <> struct IsVec<ushort3> { enum {value = 1}; };
+    template <> struct IsVec<ushort4> { enum {value = 1}; };
+    template <> struct IsVec<short1> { enum {value = 1}; };
+    template <> struct IsVec<short2> { enum {value = 1}; };
+    template <> struct IsVec<short3> { enum {value = 1}; };
+    template <> struct IsVec<short4> { enum {value = 1}; };
+    template <> struct IsVec<uint1> { enum {value = 1}; };
+    template <> struct IsVec<uint2> { enum {value = 1}; };
+    template <> struct IsVec<uint3> { enum {value = 1}; };
+    template <> struct IsVec<uint4> { enum {value = 1}; };
+    template <> struct IsVec<int1> { enum {value = 1}; };
+    template <> struct IsVec<int2> { enum {value = 1}; };
+    template <> struct IsVec<int3> { enum {value = 1}; };
+    template <> struct IsVec<int4> { enum {value = 1}; };
+    template <> struct IsVec<float1> { enum {value = 1}; };
+    template <> struct IsVec<float2> { enum {value = 1}; };
+    template <> struct IsVec<float3> { enum {value = 1}; };
+    template <> struct IsVec<float4> { enum {value = 1}; };
+    template <> struct IsVec<double1> { enum {value = 1}; };
+    template <> struct IsVec<double2> { enum {value = 1}; };
+    template <> struct IsVec<double3> { enum {value = 1}; };
+    template <> struct IsVec<double4> { enum {value = 1}; };
+
+    template <class U> struct AddParameterType { typedef const U& type; };
+    template <class U> struct AddParameterType<U&> { typedef U& type; };
+    template <> struct AddParameterType<void> { typedef void type; };
+
+    // ReferenceTraits
+
+    template <class U> struct ReferenceTraits
+    {
+        enum { value = 0 };
+        typedef U type;
+    };
+    template <class U> struct ReferenceTraits<U&>
+    {
+        enum { value = 1 };
+        typedef U type;
+    };
+
+    // PointerTraits
+
+    template <class U> struct PointerTraits
+    {
+        enum { value = 0 };
+        typedef void type;
+    };
+    template <class U> struct PointerTraits<U*>
+    {
+        enum { value = 1 };
+        typedef U type;
+    };
+    template <class U> struct PointerTraits<U*&>
+    {
+        enum { value = 1 };
+        typedef U type;
+    };
+
+    // UnConst
+
+    template <class U> struct UnConst
+    {
+        typedef U type;
+        enum { value = 0 };
+    };
+    template <class U> struct UnConst<const U>
+    {
+        typedef U type;
+        enum { value = 1 };
+    };
+    template <class U> struct UnConst<const U&>
+    {
+        typedef U& type;
+        enum { value = 1 };
+    };
+
+    // UnVolatile
+
+    template <class U> struct UnVolatile
+    {
+        typedef U type;
+        enum { value = 0 };
+    };
+    template <class U> struct UnVolatile<volatile U>
+    {
+        typedef U type;
+        enum { value = 1 };
+    };
+    template <class U> struct UnVolatile<volatile U&>
+    {
+        typedef U& type;
+        enum { value = 1 };
+    };
+
+    // IsSimpleParameter
+
+    template <typename T> struct IsSimpleParameter
+    {
+        enum { value = IsIntegral<T>::value
+               || IsFloat<T>::value
+               || PointerTraits<typename ReferenceTraits<T>::type>::value};
+    };
+
+    // LargerDepth
+
+    template <bool, typename ThenType, typename ElseType> struct SelectIf
+    {
+        typedef ThenType type;
+    };
+    template <typename ThenType, typename ElseType> struct SelectIf<false, ThenType, ElseType>
+    {
+        typedef ElseType type;
+    };
+
+    template <typename A, typename B> struct LargerDepth
+    {
+        typedef typename SelectIf<sizeof(A) >= sizeof(B), A, B>::type type;
+    };
+    template <typename A> struct LargerDepth<A, float>
+    {
+        typedef float type;
+    };
+    template <typename A> struct LargerDepth<float, A>
+    {
+        typedef float type;
+    };
+    template <typename A> struct LargerDepth<A, double>
+    {
+        typedef double type;
+    };
+    template <typename A> struct LargerDepth<double, A>
+    {
+        typedef double type;
+    };
+    template <> struct LargerDepth<float, float>
+    {
+        typedef float type;
+    };
+    template <> struct LargerDepth<float, double>
+    {
+        typedef double type;
+    };
+    template <> struct LargerDepth<double, float>
+    {
+        typedef double type;
+    };
+    template <> struct LargerDepth<double, double>
+    {
+        typedef double type;
+    };
+}
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/util/limits.hpp

@@ -0,0 +1,124 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_UTIL_LIMITS_HPP__
+#define __OPENCV_CUDEV_UTIL_LIMITS_HPP__
+
+#include <limits.h>
+#include <float.h>
+#include "../common.hpp"
+
+namespace cv { namespace cudev {
+
+template <class T> struct numeric_limits;
+
+template <> struct numeric_limits<bool>
+{
+    __device__ __forceinline__ static bool min() { return false; }
+    __device__ __forceinline__ static bool max() { return true;  }
+    static const bool is_signed = false;
+};
+
+template <> struct numeric_limits<schar>
+{
+    __device__ __forceinline__ static schar min() { return SCHAR_MIN; }
+    __device__ __forceinline__ static schar max() { return SCHAR_MAX; }
+    static const bool is_signed = true;
+};
+
+template <> struct numeric_limits<uchar>
+{
+    __device__ __forceinline__ static uchar min() { return 0; }
+    __device__ __forceinline__ static uchar max() { return UCHAR_MAX; }
+    static const bool is_signed = false;
+};
+
+template <> struct numeric_limits<short>
+{
+    __device__ __forceinline__ static short min() { return SHRT_MIN; }
+    __device__ __forceinline__ static short max() { return SHRT_MAX; }
+    static const bool is_signed = true;
+};
+
+template <> struct numeric_limits<ushort>
+{
+    __device__ __forceinline__ static ushort min() { return 0; }
+    __device__ __forceinline__ static ushort max() { return USHRT_MAX; }
+    static const bool is_signed = false;
+};
+
+template <> struct numeric_limits<int>
+{
+    __device__ __forceinline__ static int min() { return INT_MIN; }
+    __device__ __forceinline__ static int max() { return INT_MAX; }
+    static const bool is_signed = true;
+};
+
+template <> struct numeric_limits<uint>
+{
+    __device__ __forceinline__ static uint min() { return 0; }
+    __device__ __forceinline__ static uint max() { return UINT_MAX; }
+    static const bool is_signed = false;
+};
+
+template <> struct numeric_limits<float>
+{
+    __device__ __forceinline__ static float min() { return FLT_MIN; }
+    __device__ __forceinline__ static float max() { return FLT_MAX; }
+    __device__ __forceinline__ static float epsilon() { return FLT_EPSILON; }
+    static const bool is_signed = true;
+};
+
+template <> struct numeric_limits<double>
+{
+    __device__ __forceinline__ static double min() { return DBL_MIN; }
+    __device__ __forceinline__ static double max() { return DBL_MAX; }
+    __device__ __forceinline__ static double epsilon() { return DBL_EPSILON; }
+    static const bool is_signed = true;
+};
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/util/saturate_cast.hpp

@@ -0,0 +1,264 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_UTIL_SATURATE_CAST_HPP__
+#define __OPENCV_CUDEV_UTIL_SATURATE_CAST_HPP__
+
+#include "../common.hpp"
+
+namespace cv { namespace cudev {
+
+template <typename T> __device__ __forceinline__ T saturate_cast(uchar v) { return T(v); }
+template <typename T> __device__ __forceinline__ T saturate_cast(schar v) { return T(v); }
+template <typename T> __device__ __forceinline__ T saturate_cast(ushort v) { return T(v); }
+template <typename T> __device__ __forceinline__ T saturate_cast(short v) { return T(v); }
+template <typename T> __device__ __forceinline__ T saturate_cast(uint v) { return T(v); }
+template <typename T> __device__ __forceinline__ T saturate_cast(int v) { return T(v); }
+template <typename T> __device__ __forceinline__ T saturate_cast(float v) { return T(v); }
+template <typename T> __device__ __forceinline__ T saturate_cast(double v) { return T(v); }
+
+template <> __device__ __forceinline__ uchar saturate_cast<uchar>(schar v)
+{
+    uint res = 0;
+    int vi = v;
+    asm("cvt.sat.u8.s8 %0, %1;" : "=r"(res) : "r"(vi));
+    return res;
+}
+template <> __device__ __forceinline__ uchar saturate_cast<uchar>(short v)
+{
+    uint res = 0;
+    asm("cvt.sat.u8.s16 %0, %1;" : "=r"(res) : "h"(v));
+    return res;
+}
+template <> __device__ __forceinline__ uchar saturate_cast<uchar>(ushort v)
+{
+    uint res = 0;
+    asm("cvt.sat.u8.u16 %0, %1;" : "=r"(res) : "h"(v));
+    return res;
+}
+template <> __device__ __forceinline__ uchar saturate_cast<uchar>(int v)
+{
+    uint res = 0;
+    asm("cvt.sat.u8.s32 %0, %1;" : "=r"(res) : "r"(v));
+    return res;
+}
+template <> __device__ __forceinline__ uchar saturate_cast<uchar>(uint v)
+{
+    uint res = 0;
+    asm("cvt.sat.u8.u32 %0, %1;" : "=r"(res) : "r"(v));
+    return res;
+}
+template <> __device__ __forceinline__ uchar saturate_cast<uchar>(float v)
+{
+    uint res = 0;
+    asm("cvt.rni.sat.u8.f32 %0, %1;" : "=r"(res) : "f"(v));
+    return res;
+}
+template <> __device__ __forceinline__ uchar saturate_cast<uchar>(double v)
+{
+    uint res = 0;
+    asm("cvt.rni.sat.u8.f64 %0, %1;" : "=r"(res) : "d"(v));
+    return res;
+}
+
+template <> __device__ __forceinline__ schar saturate_cast<schar>(uchar v)
+{
+    uint res = 0;
+    uint vi = v;
+    asm("cvt.sat.s8.u8 %0, %1;" : "=r"(res) : "r"(vi));
+    return res;
+}
+template <> __device__ __forceinline__ schar saturate_cast<schar>(short v)
+{
+    uint res = 0;
+    asm("cvt.sat.s8.s16 %0, %1;" : "=r"(res) : "h"(v));
+    return res;
+}
+template <> __device__ __forceinline__ schar saturate_cast<schar>(ushort v)
+{
+    uint res = 0;
+    asm("cvt.sat.s8.u16 %0, %1;" : "=r"(res) : "h"(v));
+    return res;
+}
+template <> __device__ __forceinline__ schar saturate_cast<schar>(int v)
+{
+    uint res = 0;
+    asm("cvt.sat.s8.s32 %0, %1;" : "=r"(res) : "r"(v));
+    return res;
+}
+template <> __device__ __forceinline__ schar saturate_cast<schar>(uint v)
+{
+    uint res = 0;
+    asm("cvt.sat.s8.u32 %0, %1;" : "=r"(res) : "r"(v));
+    return res;
+}
+template <> __device__ __forceinline__ schar saturate_cast<schar>(float v)
+{
+    uint res = 0;
+    asm("cvt.rni.sat.s8.f32 %0, %1;" : "=r"(res) : "f"(v));
+    return res;
+}
+template <> __device__ __forceinline__ schar saturate_cast<schar>(double v)
+{
+    uint res = 0;
+    asm("cvt.rni.sat.s8.f64 %0, %1;" : "=r"(res) : "d"(v));
+    return res;
+}
+
+template <> __device__ __forceinline__ ushort saturate_cast<ushort>(schar v)
+{
+    ushort res = 0;
+    int vi = v;
+    asm("cvt.sat.u16.s8 %0, %1;" : "=h"(res) : "r"(vi));
+    return res;
+}
+template <> __device__ __forceinline__ ushort saturate_cast<ushort>(short v)
+{
+    ushort res = 0;
+    asm("cvt.sat.u16.s16 %0, %1;" : "=h"(res) : "h"(v));
+    return res;
+}
+template <> __device__ __forceinline__ ushort saturate_cast<ushort>(int v)
+{
+    ushort res = 0;
+    asm("cvt.sat.u16.s32 %0, %1;" : "=h"(res) : "r"(v));
+    return res;
+}
+template <> __device__ __forceinline__ ushort saturate_cast<ushort>(uint v)
+{
+    ushort res = 0;
+    asm("cvt.sat.u16.u32 %0, %1;" : "=h"(res) : "r"(v));
+    return res;
+}
+template <> __device__ __forceinline__ ushort saturate_cast<ushort>(float v)
+{
+    ushort res = 0;
+    asm("cvt.rni.sat.u16.f32 %0, %1;" : "=h"(res) : "f"(v));
+    return res;
+}
+template <> __device__ __forceinline__ ushort saturate_cast<ushort>(double v)
+{
+    ushort res = 0;
+    asm("cvt.rni.sat.u16.f64 %0, %1;" : "=h"(res) : "d"(v));
+    return res;
+}
+
+template <> __device__ __forceinline__ short saturate_cast<short>(ushort v)
+{
+    short res = 0;
+    asm("cvt.sat.s16.u16 %0, %1;" : "=h"(res) : "h"(v));
+    return res;
+}
+template <> __device__ __forceinline__ short saturate_cast<short>(int v)
+{
+    short res = 0;
+    asm("cvt.sat.s16.s32 %0, %1;" : "=h"(res) : "r"(v));
+    return res;
+}
+template <> __device__ __forceinline__ short saturate_cast<short>(uint v)
+{
+    short res = 0;
+    asm("cvt.sat.s16.u32 %0, %1;" : "=h"(res) : "r"(v));
+    return res;
+}
+template <> __device__ __forceinline__ short saturate_cast<short>(float v)
+{
+    short res = 0;
+    asm("cvt.rni.sat.s16.f32 %0, %1;" : "=h"(res) : "f"(v));
+    return res;
+}
+template <> __device__ __forceinline__ short saturate_cast<short>(double v)
+{
+    short res = 0;
+    asm("cvt.rni.sat.s16.f64 %0, %1;" : "=h"(res) : "d"(v));
+    return res;
+}
+
+template <> __device__ __forceinline__ int saturate_cast<int>(uint v)
+{
+    int res = 0;
+    asm("cvt.sat.s32.u32 %0, %1;" : "=r"(res) : "r"(v));
+    return res;
+}
+template <> __device__ __forceinline__ int saturate_cast<int>(float v)
+{
+    return __float2int_rn(v);
+}
+template <> __device__ __forceinline__ int saturate_cast<int>(double v)
+{
+    return __double2int_rn(v);
+}
+
+template <> __device__ __forceinline__ uint saturate_cast<uint>(schar v)
+{
+    uint res = 0;
+    int vi = v;
+    asm("cvt.sat.u32.s8 %0, %1;" : "=r"(res) : "r"(vi));
+    return res;
+}
+template <> __device__ __forceinline__ uint saturate_cast<uint>(short v)
+{
+    uint res = 0;
+    asm("cvt.sat.u32.s16 %0, %1;" : "=r"(res) : "h"(v));
+    return res;
+}
+template <> __device__ __forceinline__ uint saturate_cast<uint>(int v)
+{
+    uint res = 0;
+    asm("cvt.sat.u32.s32 %0, %1;" : "=r"(res) : "r"(v));
+    return res;
+}
+template <> __device__ __forceinline__ uint saturate_cast<uint>(float v)
+{
+    return __float2uint_rn(v);
+}
+template <> __device__ __forceinline__ uint saturate_cast<uint>(double v)
+{
+    return __double2uint_rn(v);
+}
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/util/simd_functions.hpp

@@ -0,0 +1,913 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+/*
+ * Copyright (c) 2013 NVIDIA Corporation. All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *
+ *   Redistributions of source code must retain the above copyright notice,
+ *   this list of conditions and the following disclaimer.
+ *
+ *   Redistributions 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.
+ *
+ *   Neither the name of NVIDIA Corporation nor the names of its contributors
+ *   may 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 COPYRIGHT HOLDER 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.
+ */
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_UTIL_SIMD_FUNCTIONS_HPP__
+#define __OPENCV_CUDEV_UTIL_SIMD_FUNCTIONS_HPP__
+
+#include "../common.hpp"
+
+/*
+  This header file contains inline functions that implement intra-word SIMD
+  operations, that are hardware accelerated on sm_3x (Kepler) GPUs. Efficient
+  emulation code paths are provided for earlier architectures (sm_1x, sm_2x)
+  to make the code portable across all GPUs supported by CUDA. The following
+  functions are currently implemented:
+
+  vadd2(a,b)      per-halfword unsigned addition, with wrap-around: a + b
+  vsub2(a,b)      per-halfword unsigned subtraction, with wrap-around: a - b
+  vabsdiff2(a,b)  per-halfword unsigned absolute difference: |a - b|
+  vavg2(a,b)      per-halfword unsigned average: (a + b) / 2
+  vavrg2(a,b)     per-halfword unsigned rounded average: (a + b + 1) / 2
+  vseteq2(a,b)    per-halfword unsigned comparison: a == b ? 1 : 0
+  vcmpeq2(a,b)    per-halfword unsigned comparison: a == b ? 0xffff : 0
+  vsetge2(a,b)    per-halfword unsigned comparison: a >= b ? 1 : 0
+  vcmpge2(a,b)    per-halfword unsigned comparison: a >= b ? 0xffff : 0
+  vsetgt2(a,b)    per-halfword unsigned comparison: a > b ? 1 : 0
+  vcmpgt2(a,b)    per-halfword unsigned comparison: a > b ? 0xffff : 0
+  vsetle2(a,b)    per-halfword unsigned comparison: a <= b ? 1 : 0
+  vcmple2(a,b)    per-halfword unsigned comparison: a <= b ? 0xffff : 0
+  vsetlt2(a,b)    per-halfword unsigned comparison: a < b ? 1 : 0
+  vcmplt2(a,b)    per-halfword unsigned comparison: a < b ? 0xffff : 0
+  vsetne2(a,b)    per-halfword unsigned comparison: a != b ? 1 : 0
+  vcmpne2(a,b)    per-halfword unsigned comparison: a != b ? 0xffff : 0
+  vmax2(a,b)      per-halfword unsigned maximum: max(a, b)
+  vmin2(a,b)      per-halfword unsigned minimum: min(a, b)
+
+  vadd4(a,b)      per-byte unsigned addition, with wrap-around: a + b
+  vsub4(a,b)      per-byte unsigned subtraction, with wrap-around: a - b
+  vabsdiff4(a,b)  per-byte unsigned absolute difference: |a - b|
+  vavg4(a,b)      per-byte unsigned average: (a + b) / 2
+  vavrg4(a,b)     per-byte unsigned rounded average: (a + b + 1) / 2
+  vseteq4(a,b)    per-byte unsigned comparison: a == b ? 1 : 0
+  vcmpeq4(a,b)    per-byte unsigned comparison: a == b ? 0xff : 0
+  vsetge4(a,b)    per-byte unsigned comparison: a >= b ? 1 : 0
+  vcmpge4(a,b)    per-byte unsigned comparison: a >= b ? 0xff : 0
+  vsetgt4(a,b)    per-byte unsigned comparison: a > b ? 1 : 0
+  vcmpgt4(a,b)    per-byte unsigned comparison: a > b ? 0xff : 0
+  vsetle4(a,b)    per-byte unsigned comparison: a <= b ? 1 : 0
+  vcmple4(a,b)    per-byte unsigned comparison: a <= b ? 0xff : 0
+  vsetlt4(a,b)    per-byte unsigned comparison: a < b ? 1 : 0
+  vcmplt4(a,b)    per-byte unsigned comparison: a < b ? 0xff : 0
+  vsetne4(a,b)    per-byte unsigned comparison: a != b ? 1: 0
+  vcmpne4(a,b)    per-byte unsigned comparison: a != b ? 0xff: 0
+  vmax4(a,b)      per-byte unsigned maximum: max(a, b)
+  vmin4(a,b)      per-byte unsigned minimum: min(a, b)
+*/
+
+namespace cv { namespace cudev {
+
+// 2
+
+__device__ __forceinline__ uint vadd2(uint a, uint b)
+{
+    uint r = 0;
+
+#if CV_CUDEV_ARCH >= 300
+    asm("vadd2.u32.u32.u32.sat %0, %1, %2, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+#elif CV_CUDEV_ARCH >= 200
+    asm("vadd.u32.u32.u32.sat %0.h0, %1.h0, %2.h0, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+    asm("vadd.u32.u32.u32.sat %0.h1, %1.h1, %2.h1, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+#else
+    uint s;
+    s = a ^ b;          // sum bits
+    r = a + b;          // actual sum
+    s = s ^ r;          // determine carry-ins for each bit position
+    s = s & 0x00010000; // carry-in to high word (= carry-out from low word)
+    r = r - s;          // subtract out carry-out from low word
+#endif
+
+    return r;
+}
+
+__device__ __forceinline__ uint vsub2(uint a, uint b)
+{
+    uint r = 0;
+
+#if CV_CUDEV_ARCH >= 300
+    asm("vsub2.u32.u32.u32.sat %0, %1, %2, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+#elif CV_CUDEV_ARCH >= 200
+    asm("vsub.u32.u32.u32.sat %0.h0, %1.h0, %2.h0, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+    asm("vsub.u32.u32.u32.sat %0.h1, %1.h1, %2.h1, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+#else
+    uint s;
+    s = a ^ b;          // sum bits
+    r = a - b;          // actual sum
+    s = s ^ r;          // determine carry-ins for each bit position
+    s = s & 0x00010000; // borrow to high word
+    r = r + s;          // compensate for borrow from low word
+#endif
+
+    return r;
+}
+
+__device__ __forceinline__ uint vabsdiff2(uint a, uint b)
+{
+    uint r = 0;
+
+#if CV_CUDEV_ARCH >= 300
+    asm("vabsdiff2.u32.u32.u32.sat %0, %1, %2, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+#elif CV_CUDEV_ARCH >= 200
+    asm("vabsdiff.u32.u32.u32.sat %0.h0, %1.h0, %2.h0, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+    asm("vabsdiff.u32.u32.u32.sat %0.h1, %1.h1, %2.h1, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+#else
+    uint s, t, u, v;
+    s = a & 0x0000ffff; // extract low halfword
+    r = b & 0x0000ffff; // extract low halfword
+    u = ::max(r, s);    // maximum of low halfwords
+    v = ::min(r, s);    // minimum of low halfwords
+    s = a & 0xffff0000; // extract high halfword
+    r = b & 0xffff0000; // extract high halfword
+    t = ::max(r, s);    // maximum of high halfwords
+    s = ::min(r, s);    // minimum of high halfwords
+    r = u | t;          // maximum of both halfwords
+    s = v | s;          // minimum of both halfwords
+    r = r - s;          // |a - b| = max(a,b) - min(a,b);
+#endif
+
+    return r;
+}
+
+__device__ __forceinline__ uint vavg2(uint a, uint b)
+{
+    uint r, s;
+
+    // HAKMEM #23: a + b = 2 * (a & b) + (a ^ b) ==>
+    // (a + b) / 2 = (a & b) + ((a ^ b) >> 1)
+    s = a ^ b;
+    r = a & b;
+    s = s & 0xfffefffe; // ensure shift doesn't cross halfword boundaries
+    s = s >> 1;
+    s = r + s;
+
+    return s;
+}
+
+__device__ __forceinline__ uint vavrg2(uint a, uint b)
+{
+    uint r = 0;
+
+#if CV_CUDEV_ARCH >= 300
+    asm("vavrg2.u32.u32.u32 %0, %1, %2, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+#else
+    // HAKMEM #23: a + b = 2 * (a | b) - (a ^ b) ==>
+    // (a + b + 1) / 2 = (a | b) - ((a ^ b) >> 1)
+    uint s;
+    s = a ^ b;
+    r = a | b;
+    s = s & 0xfffefffe; // ensure shift doesn't cross half-word boundaries
+    s = s >> 1;
+    r = r - s;
+#endif
+
+    return r;
+}
+
+__device__ __forceinline__ uint vseteq2(uint a, uint b)
+{
+    uint r = 0;
+
+#if CV_CUDEV_ARCH >= 300
+    asm("vset2.u32.u32.eq %0, %1, %2, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+#else
+    // inspired by Alan Mycroft's null-byte detection algorithm:
+    // null_byte(x) = ((x - 0x01010101) & (~x & 0x80808080))
+    uint c;
+    r = a ^ b;          // 0x0000 if a == b
+    c = r | 0x80008000; // set msbs, to catch carry out
+    r = r ^ c;          // extract msbs, msb = 1 if r < 0x8000
+    c = c - 0x00010001; // msb = 0, if r was 0x0000 or 0x8000
+    c = r & ~c;         // msb = 1, if r was 0x0000
+    r = c >> 15;        // convert to bool
+#endif
+
+    return r;
+}
+
+__device__ __forceinline__ uint vcmpeq2(uint a, uint b)
+{
+    uint r, c;
+
+#if CV_CUDEV_ARCH >= 300
+    r = vseteq2(a, b);
+    c = r << 16;        // convert bool
+    r = c - r;          //  into mask
+#else
+    // inspired by Alan Mycroft's null-byte detection algorithm:
+    // null_byte(x) = ((x - 0x01010101) & (~x & 0x80808080))
+    r = a ^ b;          // 0x0000 if a == b
+    c = r | 0x80008000; // set msbs, to catch carry out
+    r = r ^ c;          // extract msbs, msb = 1 if r < 0x8000
+    c = c - 0x00010001; // msb = 0, if r was 0x0000 or 0x8000
+    c = r & ~c;         // msb = 1, if r was 0x0000
+    r = c >> 15;        // convert
+    r = c - r;          //  msbs to
+    r = c | r;          //   mask
+#endif
+
+    return r;
+}
+
+__device__ __forceinline__ uint vsetge2(uint a, uint b)
+{
+    uint r = 0;
+
+#if CV_CUDEV_ARCH >= 300
+    asm("vset2.u32.u32.ge %0, %1, %2, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+#else
+    uint c;
+    asm("not.b32 %0, %0;" : "+r"(b));
+    c = vavrg2(a, b);   // (a + ~b + 1) / 2 = (a - b) / 2
+    c = c & 0x80008000; // msb = carry-outs
+    r = c >> 15;        // convert to bool
+#endif
+
+    return r;
+}
+
+__device__ __forceinline__ uint vcmpge2(uint a, uint b)
+{
+    uint r, c;
+
+#if CV_CUDEV_ARCH >= 300
+    r = vsetge2(a, b);
+    c = r << 16;        // convert bool
+    r = c - r;          //  into mask
+#else
+    asm("not.b32 %0, %0;" : "+r"(b));
+    c = vavrg2(a, b);   // (a + ~b + 1) / 2 = (a - b) / 2
+    c = c & 0x80008000; // msb = carry-outs
+    r = c >> 15;        // convert
+    r = c - r;          //  msbs to
+    r = c | r;          //   mask
+#endif
+
+    return r;
+}
+
+__device__ __forceinline__ uint vsetgt2(uint a, uint b)
+{
+    uint r = 0;
+
+#if CV_CUDEV_ARCH >= 300
+    asm("vset2.u32.u32.gt %0, %1, %2, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+#else
+    uint c;
+    asm("not.b32 %0, %0;" : "+r"(b));
+    c = vavg2(a, b);    // (a + ~b) / 2 = (a - b) / 2 [rounded down]
+    c = c & 0x80008000; // msbs = carry-outs
+    r = c >> 15;        // convert to bool
+#endif
+
+    return r;
+}
+
+__device__ __forceinline__ uint vcmpgt2(uint a, uint b)
+{
+    uint r, c;
+
+#if CV_CUDEV_ARCH >= 300
+    r = vsetgt2(a, b);
+    c = r << 16;        // convert bool
+    r = c - r;          //  into mask
+#else
+    asm("not.b32 %0, %0;" : "+r"(b));
+    c = vavg2(a, b);    // (a + ~b) / 2 = (a - b) / 2 [rounded down]
+    c = c & 0x80008000; // msbs = carry-outs
+    r = c >> 15;        // convert
+    r = c - r;          //  msbs to
+    r = c | r;          //   mask
+#endif
+
+    return r;
+}
+
+__device__ __forceinline__ uint vsetle2(uint a, uint b)
+{
+    uint r = 0;
+
+#if CV_CUDEV_ARCH >= 300
+    asm("vset2.u32.u32.le %0, %1, %2, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+#else
+    uint c;
+    asm("not.b32 %0, %0;" : "+r"(a));
+    c = vavrg2(a, b);   // (b + ~a + 1) / 2 = (b - a) / 2
+    c = c & 0x80008000; // msb = carry-outs
+    r = c >> 15;        // convert to bool
+#endif
+
+    return r;
+}
+
+__device__ __forceinline__ uint vcmple2(uint a, uint b)
+{
+    uint r, c;
+
+#if CV_CUDEV_ARCH >= 300
+    r = vsetle2(a, b);
+    c = r << 16;        // convert bool
+    r = c - r;          //  into mask
+#else
+    asm("not.b32 %0, %0;" : "+r"(a));
+    c = vavrg2(a, b);   // (b + ~a + 1) / 2 = (b - a) / 2
+    c = c & 0x80008000; // msb = carry-outs
+    r = c >> 15;        // convert
+    r = c - r;          //  msbs to
+    r = c | r;          //   mask
+#endif
+
+    return r;
+}
+
+__device__ __forceinline__ uint vsetlt2(uint a, uint b)
+{
+    uint r = 0;
+
+#if CV_CUDEV_ARCH >= 300
+    asm("vset2.u32.u32.lt %0, %1, %2, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+#else
+    uint c;
+    asm("not.b32 %0, %0;" : "+r"(a));
+    c = vavg2(a, b);    // (b + ~a) / 2 = (b - a) / 2 [rounded down]
+    c = c & 0x80008000; // msb = carry-outs
+    r = c >> 15;        // convert to bool
+#endif
+
+    return r;
+}
+
+__device__ __forceinline__ uint vcmplt2(uint a, uint b)
+{
+    uint r, c;
+
+#if CV_CUDEV_ARCH >= 300
+    r = vsetlt2(a, b);
+    c = r << 16;        // convert bool
+    r = c - r;          //  into mask
+#else
+    asm("not.b32 %0, %0;" : "+r"(a));
+    c = vavg2(a, b);    // (b + ~a) / 2 = (b - a) / 2 [rounded down]
+    c = c & 0x80008000; // msb = carry-outs
+    r = c >> 15;        // convert
+    r = c - r;          //  msbs to
+    r = c | r;          //   mask
+#endif
+
+    return r;
+}
+
+__device__ __forceinline__ uint vsetne2(uint a, uint b)
+{
+    uint r = 0;
+
+#if CV_CUDEV_ARCH >= 300
+    asm ("vset2.u32.u32.ne %0, %1, %2, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+#else
+    // inspired by Alan Mycroft's null-byte detection algorithm:
+    // null_byte(x) = ((x - 0x01010101) & (~x & 0x80808080))
+    uint c;
+    r = a ^ b;          // 0x0000 if a == b
+    c = r | 0x80008000; // set msbs, to catch carry out
+    c = c - 0x00010001; // msb = 0, if r was 0x0000 or 0x8000
+    c = r | c;          // msb = 1, if r was not 0x0000
+    c = c & 0x80008000; // extract msbs
+    r = c >> 15;        // convert to bool
+#endif
+
+    return r;
+}
+
+__device__ __forceinline__ uint vcmpne2(uint a, uint b)
+{
+    uint r, c;
+
+#if CV_CUDEV_ARCH >= 300
+    r = vsetne2(a, b);
+    c = r << 16;        // convert bool
+    r = c - r;          //  into mask
+#else
+    // inspired by Alan Mycroft's null-byte detection algorithm:
+    // null_byte(x) = ((x - 0x01010101) & (~x & 0x80808080))
+    r = a ^ b;          // 0x0000 if a == b
+    c = r | 0x80008000; // set msbs, to catch carry out
+    c = c - 0x00010001; // msb = 0, if r was 0x0000 or 0x8000
+    c = r | c;          // msb = 1, if r was not 0x0000
+    c = c & 0x80008000; // extract msbs
+    r = c >> 15;        // convert
+    r = c - r;          //  msbs to
+    r = c | r;          //   mask
+#endif
+
+    return r;
+}
+
+__device__ __forceinline__ uint vmax2(uint a, uint b)
+{
+    uint r = 0;
+
+#if CV_CUDEV_ARCH >= 300
+    asm("vmax2.u32.u32.u32 %0, %1, %2, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+#elif CV_CUDEV_ARCH >= 200
+    asm("vmax.u32.u32.u32 %0.h0, %1.h0, %2.h0, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+    asm("vmax.u32.u32.u32 %0.h1, %1.h1, %2.h1, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+#else
+    uint s, t, u;
+    r = a & 0x0000ffff; // extract low halfword
+    s = b & 0x0000ffff; // extract low halfword
+    t = ::max(r, s);    // maximum of low halfwords
+    r = a & 0xffff0000; // extract high halfword
+    s = b & 0xffff0000; // extract high halfword
+    u = ::max(r, s);    // maximum of high halfwords
+    r = t | u;          // combine halfword maximums
+#endif
+
+    return r;
+}
+
+__device__ __forceinline__ uint vmin2(uint a, uint b)
+{
+    uint r = 0;
+
+#if CV_CUDEV_ARCH >= 300
+    asm("vmin2.u32.u32.u32 %0, %1, %2, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+#elif CV_CUDEV_ARCH >= 200
+    asm("vmin.u32.u32.u32 %0.h0, %1.h0, %2.h0, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+    asm("vmin.u32.u32.u32 %0.h1, %1.h1, %2.h1, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+#else
+    uint s, t, u;
+    r = a & 0x0000ffff; // extract low halfword
+    s = b & 0x0000ffff; // extract low halfword
+    t = ::min(r, s);    // minimum of low halfwords
+    r = a & 0xffff0000; // extract high halfword
+    s = b & 0xffff0000; // extract high halfword
+    u = ::min(r, s);    // minimum of high halfwords
+    r = t | u;          // combine halfword minimums
+#endif
+
+    return r;
+}
+
+// 4
+
+__device__ __forceinline__ uint vadd4(uint a, uint b)
+{
+    uint r = 0;
+
+#if CV_CUDEV_ARCH >= 300
+    asm("vadd4.u32.u32.u32.sat %0, %1, %2, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+#elif CV_CUDEV_ARCH >= 200
+    asm("vadd.u32.u32.u32.sat %0.b0, %1.b0, %2.b0, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+    asm("vadd.u32.u32.u32.sat %0.b1, %1.b1, %2.b1, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+    asm("vadd.u32.u32.u32.sat %0.b2, %1.b2, %2.b2, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+    asm("vadd.u32.u32.u32.sat %0.b3, %1.b3, %2.b3, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+#else
+    uint s, t;
+    s = a ^ b;          // sum bits
+    r = a & 0x7f7f7f7f; // clear msbs
+    t = b & 0x7f7f7f7f; // clear msbs
+    s = s & 0x80808080; // msb sum bits
+    r = r + t;          // add without msbs, record carry-out in msbs
+    r = r ^ s;          // sum of msb sum and carry-in bits, w/o carry-out
+#endif /* CV_CUDEV_ARCH >= 300 */
+
+    return r;
+}
+
+__device__ __forceinline__ uint vsub4(uint a, uint b)
+{
+    uint r = 0;
+
+#if CV_CUDEV_ARCH >= 300
+    asm("vsub4.u32.u32.u32.sat %0, %1, %2, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+#elif CV_CUDEV_ARCH >= 200
+    asm("vsub.u32.u32.u32.sat %0.b0, %1.b0, %2.b0, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+    asm("vsub.u32.u32.u32.sat %0.b1, %1.b1, %2.b1, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+    asm("vsub.u32.u32.u32.sat %0.b2, %1.b2, %2.b2, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+    asm("vsub.u32.u32.u32.sat %0.b3, %1.b3, %2.b3, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+#else
+    uint s, t;
+    s = a ^ ~b;         // inverted sum bits
+    r = a | 0x80808080; // set msbs
+    t = b & 0x7f7f7f7f; // clear msbs
+    s = s & 0x80808080; // inverted msb sum bits
+    r = r - t;          // subtract w/o msbs, record inverted borrows in msb
+    r = r ^ s;          // combine inverted msb sum bits and borrows
+#endif
+
+    return r;
+}
+
+__device__ __forceinline__ uint vavg4(uint a, uint b)
+{
+    uint r, s;
+
+    // HAKMEM #23: a + b = 2 * (a & b) + (a ^ b) ==>
+    // (a + b) / 2 = (a & b) + ((a ^ b) >> 1)
+    s = a ^ b;
+    r = a & b;
+    s = s & 0xfefefefe; // ensure following shift doesn't cross byte boundaries
+    s = s >> 1;
+    s = r + s;
+
+    return s;
+}
+
+__device__ __forceinline__ uint vavrg4(uint a, uint b)
+{
+    uint r = 0;
+
+#if CV_CUDEV_ARCH >= 300
+    asm("vavrg4.u32.u32.u32 %0, %1, %2, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+#else
+    // HAKMEM #23: a + b = 2 * (a | b) - (a ^ b) ==>
+    // (a + b + 1) / 2 = (a | b) - ((a ^ b) >> 1)
+    uint c;
+    c = a ^ b;
+    r = a | b;
+    c = c & 0xfefefefe; // ensure following shift doesn't cross byte boundaries
+    c = c >> 1;
+    r = r - c;
+#endif
+
+    return r;
+}
+
+__device__ __forceinline__ uint vseteq4(uint a, uint b)
+{
+    uint r = 0;
+
+#if CV_CUDEV_ARCH >= 300
+    asm("vset4.u32.u32.eq %0, %1, %2, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+#else
+    // inspired by Alan Mycroft's null-byte detection algorithm:
+    // null_byte(x) = ((x - 0x01010101) & (~x & 0x80808080))
+    uint c;
+    r = a ^ b;          // 0x00 if a == b
+    c = r | 0x80808080; // set msbs, to catch carry out
+    r = r ^ c;          // extract msbs, msb = 1 if r < 0x80
+    c = c - 0x01010101; // msb = 0, if r was 0x00 or 0x80
+    c = r & ~c;         // msb = 1, if r was 0x00
+    r = c >> 7;         // convert to bool
+#endif
+
+    return r;
+}
+
+__device__ __forceinline__ uint vcmpeq4(uint a, uint b)
+{
+    uint r, t;
+
+#if CV_CUDEV_ARCH >= 300
+    r = vseteq4(a, b);
+    t = r << 8;         // convert bool
+    r = t - r;          //  to mask
+#else
+    // inspired by Alan Mycroft's null-byte detection algorithm:
+    // null_byte(x) = ((x - 0x01010101) & (~x & 0x80808080))
+    t = a ^ b;          // 0x00 if a == b
+    r = t | 0x80808080; // set msbs, to catch carry out
+    t = t ^ r;          // extract msbs, msb = 1 if t < 0x80
+    r = r - 0x01010101; // msb = 0, if t was 0x00 or 0x80
+    r = t & ~r;         // msb = 1, if t was 0x00
+    t = r >> 7;         // build mask
+    t = r - t;          //  from
+    r = t | r;          //   msbs
+#endif
+
+    return r;
+}
+
+__device__ __forceinline__ uint vsetle4(uint a, uint b)
+{
+    uint r = 0;
+
+#if CV_CUDEV_ARCH >= 300
+    asm("vset4.u32.u32.le %0, %1, %2, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+#else
+    uint c;
+    asm("not.b32 %0, %0;" : "+r"(a));
+    c = vavrg4(a, b);   // (b + ~a + 1) / 2 = (b - a) / 2
+    c = c & 0x80808080; // msb = carry-outs
+    r = c >> 7;         // convert to bool
+#endif
+
+    return r;
+}
+
+__device__ __forceinline__ uint vcmple4(uint a, uint b)
+{
+    uint r, c;
+
+#if CV_CUDEV_ARCH >= 300
+    r = vsetle4(a, b);
+    c = r << 8;         // convert bool
+    r = c - r;          //  to mask
+#else
+    asm("not.b32 %0, %0;" : "+r"(a));
+    c = vavrg4(a, b);   // (b + ~a + 1) / 2 = (b - a) / 2
+    c = c & 0x80808080; // msbs = carry-outs
+    r = c >> 7;         // convert
+    r = c - r;          //  msbs to
+    r = c | r;          //   mask
+#endif
+
+    return r;
+}
+
+__device__ __forceinline__ uint vsetlt4(uint a, uint b)
+{
+    uint r = 0;
+
+#if CV_CUDEV_ARCH >= 300
+    asm("vset4.u32.u32.lt %0, %1, %2, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+#else
+    uint c;
+    asm("not.b32 %0, %0;" : "+r"(a));
+    c = vavg4(a, b);    // (b + ~a) / 2 = (b - a) / 2 [rounded down]
+    c = c & 0x80808080; // msb = carry-outs
+    r = c >> 7;         // convert to bool
+#endif
+
+    return r;
+}
+
+__device__ __forceinline__ uint vcmplt4(uint a, uint b)
+{
+    uint r, c;
+
+#if CV_CUDEV_ARCH >= 300
+    r = vsetlt4(a, b);
+    c = r << 8;         // convert bool
+    r = c - r;          //  to mask
+#else
+    asm("not.b32 %0, %0;" : "+r"(a));
+    c = vavg4(a, b);    // (b + ~a) / 2 = (b - a) / 2 [rounded down]
+    c = c & 0x80808080; // msbs = carry-outs
+    r = c >> 7;         // convert
+    r = c - r;          //  msbs to
+    r = c | r;          //   mask
+#endif
+
+    return r;
+}
+
+__device__ __forceinline__ uint vsetge4(uint a, uint b)
+{
+    uint r = 0;
+
+#if CV_CUDEV_ARCH >= 300
+    asm("vset4.u32.u32.ge %0, %1, %2, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+#else
+    uint c;
+    asm("not.b32 %0, %0;" : "+r"(b));
+    c = vavrg4(a, b);   // (a + ~b + 1) / 2 = (a - b) / 2
+    c = c & 0x80808080; // msb = carry-outs
+    r = c >> 7;         // convert to bool
+#endif
+
+    return r;
+}
+
+__device__ __forceinline__ uint vcmpge4(uint a, uint b)
+{
+    uint r, s;
+
+#if CV_CUDEV_ARCH >= 300
+    r = vsetge4(a, b);
+    s = r << 8;         // convert bool
+    r = s - r;          //  to mask
+#else
+    asm ("not.b32 %0,%0;" : "+r"(b));
+    r = vavrg4 (a, b);  // (a + ~b + 1) / 2 = (a - b) / 2
+    r = r & 0x80808080; // msb = carry-outs
+    s = r >> 7;         // build mask
+    s = r - s;          //  from
+    r = s | r;          //   msbs
+#endif
+
+    return r;
+}
+
+__device__ __forceinline__ uint vsetgt4(uint a, uint b)
+{
+    uint r = 0;
+
+#if CV_CUDEV_ARCH >= 300
+    asm("vset4.u32.u32.gt %0, %1, %2, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+#else
+    uint c;
+    asm("not.b32 %0, %0;" : "+r"(b));
+    c = vavg4(a, b);    // (a + ~b) / 2 = (a - b) / 2 [rounded down]
+    c = c & 0x80808080; // msb = carry-outs
+    r = c >> 7;         // convert to bool
+#endif
+
+    return r;
+}
+
+__device__ __forceinline__ uint vcmpgt4(uint a, uint b)
+{
+    uint r, c;
+
+#if CV_CUDEV_ARCH >= 300
+    r = vsetgt4(a, b);
+    c = r << 8;         // convert bool
+    r = c - r;          //  to mask
+#else
+    asm("not.b32 %0, %0;" : "+r"(b));
+    c = vavg4(a, b);    // (a + ~b) / 2 = (a - b) / 2 [rounded down]
+    c = c & 0x80808080; // msb = carry-outs
+    r = c >> 7;         // convert
+    r = c - r;          //  msbs to
+    r = c | r;          //   mask
+#endif
+
+    return r;
+}
+
+__device__ __forceinline__ uint vsetne4(uint a, uint b)
+{
+    uint r = 0;
+
+#if CV_CUDEV_ARCH >= 300
+    asm("vset4.u32.u32.ne %0, %1, %2, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+#else
+    // inspired by Alan Mycroft's null-byte detection algorithm:
+    // null_byte(x) = ((x - 0x01010101) & (~x & 0x80808080))
+    uint c;
+    r = a ^ b;          // 0x00 if a == b
+    c = r | 0x80808080; // set msbs, to catch carry out
+    c = c - 0x01010101; // msb = 0, if r was 0x00 or 0x80
+    c = r | c;          // msb = 1, if r was not 0x00
+    c = c & 0x80808080; // extract msbs
+    r = c >> 7;         // convert to bool
+#endif
+
+    return r;
+}
+
+__device__ __forceinline__ uint vcmpne4(uint a, uint b)
+{
+    uint r, c;
+
+#if CV_CUDEV_ARCH >= 300
+    r = vsetne4(a, b);
+    c = r << 8;         // convert bool
+    r = c - r;          //  to mask
+#else
+    // inspired by Alan Mycroft's null-byte detection algorithm:
+    // null_byte(x) = ((x - 0x01010101) & (~x & 0x80808080))
+    r = a ^ b;          // 0x00 if a == b
+    c = r | 0x80808080; // set msbs, to catch carry out
+    c = c - 0x01010101; // msb = 0, if r was 0x00 or 0x80
+    c = r | c;          // msb = 1, if r was not 0x00
+    c = c & 0x80808080; // extract msbs
+    r = c >> 7;         // convert
+    r = c - r;          //  msbs to
+    r = c | r;          //   mask
+#endif
+
+    return r;
+}
+
+__device__ __forceinline__ uint vabsdiff4(uint a, uint b)
+{
+    uint r = 0;
+
+#if CV_CUDEV_ARCH >= 300
+    asm("vabsdiff4.u32.u32.u32.sat %0, %1, %2, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+#elif CV_CUDEV_ARCH >= 200
+    asm("vabsdiff.u32.u32.u32.sat %0.b0, %1.b0, %2.b0, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+    asm("vabsdiff.u32.u32.u32.sat %0.b1, %1.b1, %2.b1, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+    asm("vabsdiff.u32.u32.u32.sat %0.b2, %1.b2, %2.b2, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+    asm("vabsdiff.u32.u32.u32.sat %0.b3, %1.b3, %2.b3, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+#else
+    uint s;
+    s = vcmpge4(a, b);  // mask = 0xff if a >= b
+    r = a ^ b;          //
+    s = (r &  s) ^ b;   // select a when a >= b, else select b => max(a,b)
+    r = s ^ r;          // select a when b >= a, else select b => min(a,b)
+    r = s - r;          // |a - b| = max(a,b) - min(a,b);
+#endif
+
+    return r;
+}
+
+__device__ __forceinline__ uint vmax4(uint a, uint b)
+{
+    uint r = 0;
+
+#if CV_CUDEV_ARCH >= 300
+    asm("vmax4.u32.u32.u32 %0, %1, %2, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+#elif CV_CUDEV_ARCH >= 200
+    asm("vmax.u32.u32.u32 %0.b0, %1.b0, %2.b0, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+    asm("vmax.u32.u32.u32 %0.b1, %1.b1, %2.b1, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+    asm("vmax.u32.u32.u32 %0.b2, %1.b2, %2.b2, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+    asm("vmax.u32.u32.u32 %0.b3, %1.b3, %2.b3, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+#else
+    uint s;
+    s = vcmpge4(a, b);  // mask = 0xff if a >= b
+    r = a & s;          // select a when b >= a
+    s = b & ~s;         // select b when b < a
+    r = r | s;          // combine byte selections
+#endif
+
+    return r;           // byte-wise unsigned maximum
+}
+
+__device__ __forceinline__ uint vmin4(uint a, uint b)
+{
+    uint r = 0;
+
+#if CV_CUDEV_ARCH >= 300
+    asm("vmin4.u32.u32.u32 %0, %1, %2, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+#elif CV_CUDEV_ARCH >= 200
+    asm("vmin.u32.u32.u32 %0.b0, %1.b0, %2.b0, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+    asm("vmin.u32.u32.u32 %0.b1, %1.b1, %2.b1, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+    asm("vmin.u32.u32.u32 %0.b2, %1.b2, %2.b2, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+    asm("vmin.u32.u32.u32 %0.b3, %1.b3, %2.b3, %3;" : "=r"(r) : "r"(a), "r"(b), "r"(r));
+#else
+    uint s;
+    s = vcmpge4(b, a);  // mask = 0xff if a >= b
+    r = a & s;          // select a when b >= a
+    s = b & ~s;         // select b when b < a
+    r = r | s;          // combine byte selections
+#endif
+
+    return r;
+}
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/util/tuple.hpp

@@ -0,0 +1,80 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_UTIL_TUPLE_HPP__
+#define __OPENCV_CUDEV_UTIL_TUPLE_HPP__
+
+#include "../common.hpp"
+#include "detail/tuple.hpp"
+
+namespace cv { namespace cudev {
+
+using tuple_detail::tuple;
+using tuple_detail::tuple_size;
+using tuple_detail::get;
+using tuple_detail::tuple_element;
+using tuple_detail::make_tuple;
+using tuple_detail::tie;
+
+template <typename T> struct TupleTraits
+{
+    enum { is_tuple = 0 };
+    enum { size = 1 };
+};
+template <typename P0, typename P1, typename P2, typename P3, typename P4, typename P5, typename P6, typename P7, typename P8, typename P9>
+struct TupleTraits< tuple<P0, P1, P2, P3, P4, P5, P6, P7, P8, P9> >
+{
+    enum { is_tuple = 1 };
+    enum { size = tuple_size< tuple<P0, P1, P2, P3, P4, P5, P6, P7, P8, P9> >::value };
+};
+
+template <class Tuple, template <typename T> class CvtOp> struct ConvertTuple
+{
+    typedef typename tuple_detail::ConvertTuple<Tuple, tuple_size<Tuple>::value, CvtOp>::type type;
+};
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/util/type_traits.hpp

@@ -0,0 +1,169 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_UTIL_TYPE_TRAITS_HPP__
+#define __OPENCV_CUDEV_UTIL_TYPE_TRAITS_HPP__
+
+#include "../common.hpp"
+#include "vec_traits.hpp"
+#include "detail/type_traits.hpp"
+
+namespace cv { namespace cudev {
+
+// NullType
+
+struct NullType {};
+
+// Int2Type
+
+template <int A> struct Int2Type
+{
+   enum { value = A };
+};
+
+// ArrayWrapper
+
+template <typename T, int COUNT> struct ArrayWrapper
+{
+    T array[COUNT];
+};
+
+// Log2 (compile time calculation)
+
+template <int N, int CURRENT_VAL = N, int COUNT = 0> struct Log2
+{
+    enum { value = Log2<N, (CURRENT_VAL >> 1), COUNT + 1>::VALUE };
+};
+template <int N, int COUNT> struct Log2<N, 0, COUNT>
+{
+    enum { value = (1 << (COUNT - 1) < N) ? COUNT : COUNT - 1 };
+};
+
+// IsPowerOf2
+
+template <int N> struct IsPowerOf2
+{
+    enum { value = ((N != 0) && !(N & (N - 1))) };
+};
+
+// SelectIf
+
+template <bool, typename ThenType, typename ElseType> struct SelectIf
+{
+    typedef ThenType type;
+};
+template <typename ThenType, typename ElseType> struct SelectIf<false, ThenType, ElseType>
+{
+    typedef ElseType type;
+};
+
+// EnableIf
+
+template <bool, typename T = void> struct EnableIf {};
+template <typename T> struct EnableIf<true, T> { typedef T type; };
+
+// DisableIf
+
+template <bool, typename T = void> struct DisableIf {};
+template <typename T> struct DisableIf<false, T> { typedef T type; };
+
+// TypesEquals
+
+template <typename A, typename B> struct TypesEquals
+{
+    enum { value = 0 };
+};
+template <typename A> struct TypesEquals<A, A>
+{
+    enum { value = 1 };
+};
+
+// TypeTraits
+
+template <typename T> struct TypeTraits
+{
+    typedef typename type_traits_detail::UnConst<T>::type                                                non_const_type;
+    typedef typename type_traits_detail::UnVolatile<T>::type                                             non_volatile_type;
+    typedef typename type_traits_detail::UnVolatile<typename type_traits_detail::UnConst<T>::type>::type unqualified_type;
+    typedef typename type_traits_detail::PointerTraits<unqualified_type>::type                           pointee_type;
+    typedef typename type_traits_detail::ReferenceTraits<T>::type                                        referred_type;
+
+    enum { is_const          = type_traits_detail::UnConst<T>::value };
+    enum { is_volatile       = type_traits_detail::UnVolatile<T>::value };
+
+    enum { is_reference      = type_traits_detail::ReferenceTraits<unqualified_type>::value };
+    enum { is_pointer        = type_traits_detail::PointerTraits<typename type_traits_detail::ReferenceTraits<unqualified_type>::type>::value };
+
+    enum { is_unsigned_int   = type_traits_detail::IsUnsignedIntegral<unqualified_type>::value };
+    enum { is_signed_int     = type_traits_detail::IsSignedIntergral<unqualified_type>::value };
+    enum { is_integral       = type_traits_detail::IsIntegral<unqualified_type>::value };
+    enum { is_float          = type_traits_detail::IsFloat<unqualified_type>::value };
+    enum { is_scalar         = is_integral || is_float };
+    enum { is_vec            = type_traits_detail::IsVec<unqualified_type>::value };
+
+    typedef typename SelectIf<type_traits_detail::IsSimpleParameter<unqualified_type>::value,
+        T, typename type_traits_detail::AddParameterType<T>::type>::type parameter_type;
+};
+
+// LargerType
+
+template <typename A, typename B> struct LargerType
+{
+    typedef typename SelectIf<
+        VecTraits<A>::cn != VecTraits<B>::cn,
+        void,
+        typename MakeVec<
+            typename type_traits_detail::LargerDepth<
+                typename VecTraits<A>::elem_type,
+                typename VecTraits<B>::elem_type
+            >::type,
+            VecTraits<A>::cn
+        >::type
+    >::type type;
+};
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/util/vec_math.hpp

@@ -0,0 +1,923 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_UTIL_VEC_MATH_HPP__
+#define __OPENCV_CUDEV_UTIL_VEC_MATH_HPP__
+
+#include "vec_traits.hpp"
+#include "saturate_cast.hpp"
+
+namespace cv { namespace cudev {
+
+// saturate_cast
+
+namespace vec_math_detail
+{
+    template <int cn, typename VecD> struct SatCastHelper;
+
+    template <typename VecD> struct SatCastHelper<1, VecD>
+    {
+        template <typename VecS> __device__ __forceinline__ static VecD cast(const VecS& v)
+        {
+            typedef typename VecTraits<VecD>::elem_type D;
+            return VecTraits<VecD>::make(saturate_cast<D>(v.x));
+        }
+    };
+
+    template <typename VecD> struct SatCastHelper<2, VecD>
+    {
+        template <typename VecS> __device__ __forceinline__ static VecD cast(const VecS& v)
+        {
+            typedef typename VecTraits<VecD>::elem_type D;
+            return VecTraits<VecD>::make(saturate_cast<D>(v.x), saturate_cast<D>(v.y));
+        }
+    };
+
+    template <typename VecD> struct SatCastHelper<3, VecD>
+    {
+        template <typename VecS> __device__ __forceinline__ static VecD cast(const VecS& v)
+        {
+            typedef typename VecTraits<VecD>::elem_type D;
+            return VecTraits<VecD>::make(saturate_cast<D>(v.x), saturate_cast<D>(v.y), saturate_cast<D>(v.z));
+        }
+    };
+
+    template <typename VecD> struct SatCastHelper<4, VecD>
+    {
+        template <typename VecS> __device__ __forceinline__ static VecD cast(const VecS& v)
+        {
+            typedef typename VecTraits<VecD>::elem_type D;
+            return VecTraits<VecD>::make(saturate_cast<D>(v.x), saturate_cast<D>(v.y), saturate_cast<D>(v.z), saturate_cast<D>(v.w));
+        }
+    };
+}
+
+template<typename T> __device__ __forceinline__ T saturate_cast(const uchar1& v) { return vec_math_detail::SatCastHelper<VecTraits<T>::cn, T>::cast(v); }
+template<typename T> __device__ __forceinline__ T saturate_cast(const char1& v) { return vec_math_detail::SatCastHelper<VecTraits<T>::cn, T>::cast(v); }
+template<typename T> __device__ __forceinline__ T saturate_cast(const ushort1& v) { return vec_math_detail::SatCastHelper<VecTraits<T>::cn, T>::cast(v); }
+template<typename T> __device__ __forceinline__ T saturate_cast(const short1& v) { return vec_math_detail::SatCastHelper<VecTraits<T>::cn, T>::cast(v); }
+template<typename T> __device__ __forceinline__ T saturate_cast(const uint1& v) { return vec_math_detail::SatCastHelper<VecTraits<T>::cn, T>::cast(v); }
+template<typename T> __device__ __forceinline__ T saturate_cast(const int1& v) { return vec_math_detail::SatCastHelper<VecTraits<T>::cn, T>::cast(v); }
+template<typename T> __device__ __forceinline__ T saturate_cast(const float1& v) { return vec_math_detail::SatCastHelper<VecTraits<T>::cn, T>::cast(v); }
+template<typename T> __device__ __forceinline__ T saturate_cast(const double1& v) { return vec_math_detail::SatCastHelper<VecTraits<T>::cn, T>::cast(v); }
+
+template<typename T> __device__ __forceinline__ T saturate_cast(const uchar2& v) { return vec_math_detail::SatCastHelper<VecTraits<T>::cn, T>::cast(v); }
+template<typename T> __device__ __forceinline__ T saturate_cast(const char2& v) { return vec_math_detail::SatCastHelper<VecTraits<T>::cn, T>::cast(v); }
+template<typename T> __device__ __forceinline__ T saturate_cast(const ushort2& v) { return vec_math_detail::SatCastHelper<VecTraits<T>::cn, T>::cast(v); }
+template<typename T> __device__ __forceinline__ T saturate_cast(const short2& v) { return vec_math_detail::SatCastHelper<VecTraits<T>::cn, T>::cast(v); }
+template<typename T> __device__ __forceinline__ T saturate_cast(const uint2& v) { return vec_math_detail::SatCastHelper<VecTraits<T>::cn, T>::cast(v); }
+template<typename T> __device__ __forceinline__ T saturate_cast(const int2& v) { return vec_math_detail::SatCastHelper<VecTraits<T>::cn, T>::cast(v); }
+template<typename T> __device__ __forceinline__ T saturate_cast(const float2& v) { return vec_math_detail::SatCastHelper<VecTraits<T>::cn, T>::cast(v); }
+template<typename T> __device__ __forceinline__ T saturate_cast(const double2& v) { return vec_math_detail::SatCastHelper<VecTraits<T>::cn, T>::cast(v); }
+
+template<typename T> __device__ __forceinline__ T saturate_cast(const uchar3& v) { return vec_math_detail::SatCastHelper<VecTraits<T>::cn, T>::cast(v); }
+template<typename T> __device__ __forceinline__ T saturate_cast(const char3& v) { return vec_math_detail::SatCastHelper<VecTraits<T>::cn, T>::cast(v); }
+template<typename T> __device__ __forceinline__ T saturate_cast(const ushort3& v) { return vec_math_detail::SatCastHelper<VecTraits<T>::cn, T>::cast(v); }
+template<typename T> __device__ __forceinline__ T saturate_cast(const short3& v) { return vec_math_detail::SatCastHelper<VecTraits<T>::cn, T>::cast(v); }
+template<typename T> __device__ __forceinline__ T saturate_cast(const uint3& v) { return vec_math_detail::SatCastHelper<VecTraits<T>::cn, T>::cast(v); }
+template<typename T> __device__ __forceinline__ T saturate_cast(const int3& v) { return vec_math_detail::SatCastHelper<VecTraits<T>::cn, T>::cast(v); }
+template<typename T> __device__ __forceinline__ T saturate_cast(const float3& v) { return vec_math_detail::SatCastHelper<VecTraits<T>::cn, T>::cast(v); }
+template<typename T> __device__ __forceinline__ T saturate_cast(const double3& v) { return vec_math_detail::SatCastHelper<VecTraits<T>::cn, T>::cast(v); }
+
+template<typename T> __device__ __forceinline__ T saturate_cast(const uchar4& v) { return vec_math_detail::SatCastHelper<VecTraits<T>::cn, T>::cast(v); }
+template<typename T> __device__ __forceinline__ T saturate_cast(const char4& v) { return vec_math_detail::SatCastHelper<VecTraits<T>::cn, T>::cast(v); }
+template<typename T> __device__ __forceinline__ T saturate_cast(const ushort4& v) { return vec_math_detail::SatCastHelper<VecTraits<T>::cn, T>::cast(v); }
+template<typename T> __device__ __forceinline__ T saturate_cast(const short4& v) { return vec_math_detail::SatCastHelper<VecTraits<T>::cn, T>::cast(v); }
+template<typename T> __device__ __forceinline__ T saturate_cast(const uint4& v) { return vec_math_detail::SatCastHelper<VecTraits<T>::cn, T>::cast(v); }
+template<typename T> __device__ __forceinline__ T saturate_cast(const int4& v) { return vec_math_detail::SatCastHelper<VecTraits<T>::cn, T>::cast(v); }
+template<typename T> __device__ __forceinline__ T saturate_cast(const float4& v) { return vec_math_detail::SatCastHelper<VecTraits<T>::cn, T>::cast(v); }
+template<typename T> __device__ __forceinline__ T saturate_cast(const double4& v) { return vec_math_detail::SatCastHelper<VecTraits<T>::cn, T>::cast(v); }
+
+// unary operators
+
+#define CV_CUDEV_IMPLEMENT_VEC_UNARY_OP(op, input_type, output_type) \
+    __device__ __forceinline__ output_type ## 1 operator op(const input_type ## 1 & a) \
+    { \
+        return VecTraits<output_type ## 1>::make(op (a.x)); \
+    } \
+    __device__ __forceinline__ output_type ## 2 operator op(const input_type ## 2 & a) \
+    { \
+        return VecTraits<output_type ## 2>::make(op (a.x), op (a.y)); \
+    } \
+    __device__ __forceinline__ output_type ## 3 operator op(const input_type ## 3 & a) \
+    { \
+        return VecTraits<output_type ## 3>::make(op (a.x), op (a.y), op (a.z)); \
+    } \
+    __device__ __forceinline__ output_type ## 4 operator op(const input_type ## 4 & a) \
+    { \
+        return VecTraits<output_type ## 4>::make(op (a.x), op (a.y), op (a.z), op (a.w)); \
+    }
+
+CV_CUDEV_IMPLEMENT_VEC_UNARY_OP(-, char, char)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_OP(-, short, short)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_OP(-, int, int)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_OP(-, float, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_OP(-, double, double)
+
+CV_CUDEV_IMPLEMENT_VEC_UNARY_OP(!, uchar, uchar)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_OP(!, char, uchar)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_OP(!, ushort, uchar)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_OP(!, short, uchar)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_OP(!, int, uchar)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_OP(!, uint, uchar)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_OP(!, float, uchar)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_OP(!, double, uchar)
+
+CV_CUDEV_IMPLEMENT_VEC_UNARY_OP(~, uchar, uchar)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_OP(~, char, char)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_OP(~, ushort, ushort)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_OP(~, short, short)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_OP(~, int, int)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_OP(~, uint, uint)
+
+#undef CV_CUDEV_IMPLEMENT_VEC_UNARY_OP
+
+// unary functions
+
+#define CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(func_name, func, input_type, output_type) \
+    __device__ __forceinline__ output_type ## 1 func_name(const input_type ## 1 & a) \
+    { \
+        return VecTraits<output_type ## 1>::make(func (a.x)); \
+    } \
+    __device__ __forceinline__ output_type ## 2 func_name(const input_type ## 2 & a) \
+    { \
+        return VecTraits<output_type ## 2>::make(func (a.x), func (a.y)); \
+    } \
+    __device__ __forceinline__ output_type ## 3 func_name(const input_type ## 3 & a) \
+    { \
+        return VecTraits<output_type ## 3>::make(func (a.x), func (a.y), func (a.z)); \
+    } \
+    __device__ __forceinline__ output_type ## 4 func_name(const input_type ## 4 & a) \
+    { \
+        return VecTraits<output_type ## 4>::make(func (a.x), func (a.y), func (a.z), func (a.w)); \
+    }
+
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(abs, /*::abs*/, uchar, uchar)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(abs, ::abs, char, char)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(abs, /*::abs*/, ushort, ushort)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(abs, ::abs, short, short)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(abs, ::abs, int, int)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(abs, /*::abs*/, uint, uint)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(abs, ::fabsf, float, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(abs, ::fabs, double, double)
+
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(sqrt, ::sqrtf, uchar, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(sqrt, ::sqrtf, char, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(sqrt, ::sqrtf, ushort, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(sqrt, ::sqrtf, short, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(sqrt, ::sqrtf, int, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(sqrt, ::sqrtf, uint, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(sqrt, ::sqrtf, float, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(sqrt, ::sqrt, double, double)
+
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(exp, ::expf, uchar, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(exp, ::expf, char, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(exp, ::expf, ushort, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(exp, ::expf, short, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(exp, ::expf, int, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(exp, ::expf, uint, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(exp, ::expf, float, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(exp, ::exp, double, double)
+
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(exp2, ::exp2f, uchar, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(exp2, ::exp2f, char, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(exp2, ::exp2f, ushort, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(exp2, ::exp2f, short, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(exp2, ::exp2f, int, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(exp2, ::exp2f, uint, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(exp2, ::exp2f, float, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(exp2, ::exp2, double, double)
+
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(exp10, ::exp10f, uchar, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(exp10, ::exp10f, char, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(exp10, ::exp10f, ushort, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(exp10, ::exp10f, short, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(exp10, ::exp10f, int, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(exp10, ::exp10f, uint, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(exp10, ::exp10f, float, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(exp10, ::exp10, double, double)
+
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(log, ::logf, uchar, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(log, ::logf, char, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(log, ::logf, ushort, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(log, ::logf, short, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(log, ::logf, int, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(log, ::logf, uint, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(log, ::logf, float, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(log, ::log, double, double)
+
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(log2, ::log2f, uchar, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(log2, ::log2f, char, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(log2, ::log2f, ushort, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(log2, ::log2f, short, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(log2, ::log2f, int, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(log2, ::log2f, uint, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(log2, ::log2f, float, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(log2, ::log2, double, double)
+
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(log10, ::log10f, uchar, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(log10, ::log10f, char, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(log10, ::log10f, ushort, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(log10, ::log10f, short, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(log10, ::log10f, int, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(log10, ::log10f, uint, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(log10, ::log10f, float, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(log10, ::log10, double, double)
+
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(sin, ::sinf, uchar, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(sin, ::sinf, char, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(sin, ::sinf, ushort, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(sin, ::sinf, short, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(sin, ::sinf, int, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(sin, ::sinf, uint, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(sin, ::sinf, float, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(sin, ::sin, double, double)
+
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(cos, ::cosf, uchar, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(cos, ::cosf, char, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(cos, ::cosf, ushort, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(cos, ::cosf, short, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(cos, ::cosf, int, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(cos, ::cosf, uint, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(cos, ::cosf, float, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(cos, ::cos, double, double)
+
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(tan, ::tanf, uchar, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(tan, ::tanf, char, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(tan, ::tanf, ushort, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(tan, ::tanf, short, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(tan, ::tanf, int, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(tan, ::tanf, uint, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(tan, ::tanf, float, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(tan, ::tan, double, double)
+
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(asin, ::asinf, uchar, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(asin, ::asinf, char, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(asin, ::asinf, ushort, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(asin, ::asinf, short, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(asin, ::asinf, int, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(asin, ::asinf, uint, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(asin, ::asinf, float, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(asin, ::asin, double, double)
+
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(acos, ::acosf, uchar, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(acos, ::acosf, char, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(acos, ::acosf, ushort, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(acos, ::acosf, short, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(acos, ::acosf, int, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(acos, ::acosf, uint, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(acos, ::acosf, float, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(acos, ::acos, double, double)
+
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(atan, ::atanf, uchar, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(atan, ::atanf, char, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(atan, ::atanf, ushort, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(atan, ::atanf, short, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(atan, ::atanf, int, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(atan, ::atanf, uint, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(atan, ::atanf, float, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(atan, ::atan, double, double)
+
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(sinh, ::sinhf, uchar, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(sinh, ::sinhf, char, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(sinh, ::sinhf, ushort, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(sinh, ::sinhf, short, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(sinh, ::sinhf, int, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(sinh, ::sinhf, uint, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(sinh, ::sinhf, float, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(sinh, ::sinh, double, double)
+
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(cosh, ::coshf, uchar, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(cosh, ::coshf, char, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(cosh, ::coshf, ushort, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(cosh, ::coshf, short, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(cosh, ::coshf, int, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(cosh, ::coshf, uint, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(cosh, ::coshf, float, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(cosh, ::cosh, double, double)
+
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(tanh, ::tanhf, uchar, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(tanh, ::tanhf, char, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(tanh, ::tanhf, ushort, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(tanh, ::tanhf, short, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(tanh, ::tanhf, int, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(tanh, ::tanhf, uint, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(tanh, ::tanhf, float, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(tanh, ::tanh, double, double)
+
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(asinh, ::asinhf, uchar, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(asinh, ::asinhf, char, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(asinh, ::asinhf, ushort, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(asinh, ::asinhf, short, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(asinh, ::asinhf, int, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(asinh, ::asinhf, uint, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(asinh, ::asinhf, float, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(asinh, ::asinh, double, double)
+
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(acosh, ::acoshf, uchar, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(acosh, ::acoshf, char, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(acosh, ::acoshf, ushort, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(acosh, ::acoshf, short, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(acosh, ::acoshf, int, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(acosh, ::acoshf, uint, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(acosh, ::acoshf, float, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(acosh, ::acosh, double, double)
+
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(atanh, ::atanhf, uchar, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(atanh, ::atanhf, char, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(atanh, ::atanhf, ushort, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(atanh, ::atanhf, short, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(atanh, ::atanhf, int, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(atanh, ::atanhf, uint, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(atanh, ::atanhf, float, float)
+CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC(atanh, ::atanh, double, double)
+
+#undef CV_CUDEV_IMPLEMENT_VEC_UNARY_FUNC
+
+// binary operators (vec & vec)
+
+#define CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(op, input_type, output_type) \
+    __device__ __forceinline__ output_type ## 1 operator op(const input_type ## 1 & a, const input_type ## 1 & b) \
+    { \
+        return VecTraits<output_type ## 1>::make(a.x op b.x); \
+    } \
+    __device__ __forceinline__ output_type ## 2 operator op(const input_type ## 2 & a, const input_type ## 2 & b) \
+    { \
+        return VecTraits<output_type ## 2>::make(a.x op b.x, a.y op b.y); \
+    } \
+    __device__ __forceinline__ output_type ## 3 operator op(const input_type ## 3 & a, const input_type ## 3 & b) \
+    { \
+        return VecTraits<output_type ## 3>::make(a.x op b.x, a.y op b.y, a.z op b.z); \
+    } \
+    __device__ __forceinline__ output_type ## 4 operator op(const input_type ## 4 & a, const input_type ## 4 & b) \
+    { \
+        return VecTraits<output_type ## 4>::make(a.x op b.x, a.y op b.y, a.z op b.z, a.w op b.w); \
+    }
+
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(+, uchar, int)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(+, char, int)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(+, ushort, int)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(+, short, int)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(+, int, int)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(+, uint, uint)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(+, float, float)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(+, double, double)
+
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(-, uchar, int)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(-, char, int)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(-, ushort, int)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(-, short, int)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(-, int, int)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(-, uint, uint)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(-, float, float)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(-, double, double)
+
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(*, uchar, int)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(*, char, int)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(*, ushort, int)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(*, short, int)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(*, int, int)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(*, uint, uint)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(*, float, float)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(*, double, double)
+
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(/, uchar, int)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(/, char, int)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(/, ushort, int)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(/, short, int)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(/, int, int)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(/, uint, uint)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(/, float, float)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(/, double, double)
+
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(==, uchar, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(==, char, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(==, ushort, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(==, short, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(==, int, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(==, uint, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(==, float, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(==, double, uchar)
+
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(!=, uchar, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(!=, char, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(!=, ushort, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(!=, short, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(!=, int, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(!=, uint, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(!=, float, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(!=, double, uchar)
+
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(>, uchar, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(>, char, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(>, ushort, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(>, short, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(>, int, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(>, uint, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(>, float, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(>, double, uchar)
+
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(<, uchar, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(<, char, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(<, ushort, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(<, short, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(<, int, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(<, uint, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(<, float, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(<, double, uchar)
+
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(>=, uchar, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(>=, char, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(>=, ushort, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(>=, short, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(>=, int, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(>=, uint, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(>=, float, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(>=, double, uchar)
+
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(<=, uchar, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(<=, char, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(<=, ushort, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(<=, short, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(<=, int, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(<=, uint, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(<=, float, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(<=, double, uchar)
+
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(&&, uchar, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(&&, char, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(&&, ushort, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(&&, short, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(&&, int, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(&&, uint, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(&&, float, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(&&, double, uchar)
+
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(||, uchar, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(||, char, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(||, ushort, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(||, short, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(||, int, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(||, uint, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(||, float, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(||, double, uchar)
+
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(&, uchar, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(&, char, char)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(&, ushort, ushort)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(&, short, short)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(&, int, int)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(&, uint, uint)
+
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(|, uchar, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(|, char, char)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(|, ushort, ushort)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(|, short, short)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(|, int, int)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(|, uint, uint)
+
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(^, uchar, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(^, char, char)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(^, ushort, ushort)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(^, short, short)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(^, int, int)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_OP(^, uint, uint)
+
+#undef CV_CUDEV_IMPLEMENT_VEC_BINARY_OP
+
+// binary operators (vec & scalar)
+
+#define CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(op, input_type, scalar_type, output_type) \
+    __device__ __forceinline__ output_type ## 1 operator op(const input_type ## 1 & a, scalar_type s) \
+    { \
+        return VecTraits<output_type ## 1>::make(a.x op s); \
+    } \
+    __device__ __forceinline__ output_type ## 1 operator op(scalar_type s, const input_type ## 1 & b) \
+    { \
+        return VecTraits<output_type ## 1>::make(s op b.x); \
+    } \
+    __device__ __forceinline__ output_type ## 2 operator op(const input_type ## 2 & a, scalar_type s) \
+    { \
+        return VecTraits<output_type ## 2>::make(a.x op s, a.y op s); \
+    } \
+    __device__ __forceinline__ output_type ## 2 operator op(scalar_type s, const input_type ## 2 & b) \
+    { \
+        return VecTraits<output_type ## 2>::make(s op b.x, s op b.y); \
+    } \
+    __device__ __forceinline__ output_type ## 3 operator op(const input_type ## 3 & a, scalar_type s) \
+    { \
+        return VecTraits<output_type ## 3>::make(a.x op s, a.y op s, a.z op s); \
+    } \
+    __device__ __forceinline__ output_type ## 3 operator op(scalar_type s, const input_type ## 3 & b) \
+    { \
+        return VecTraits<output_type ## 3>::make(s op b.x, s op b.y, s op b.z); \
+    } \
+    __device__ __forceinline__ output_type ## 4 operator op(const input_type ## 4 & a, scalar_type s) \
+    { \
+        return VecTraits<output_type ## 4>::make(a.x op s, a.y op s, a.z op s, a.w op s); \
+    } \
+    __device__ __forceinline__ output_type ## 4 operator op(scalar_type s, const input_type ## 4 & b) \
+    { \
+        return VecTraits<output_type ## 4>::make(s op b.x, s op b.y, s op b.z, s op b.w); \
+    }
+
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(+, uchar, int, int)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(+, uchar, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(+, uchar, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(+, char, int, int)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(+, char, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(+, char, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(+, ushort, int, int)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(+, ushort, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(+, ushort, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(+, short, int, int)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(+, short, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(+, short, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(+, int, int, int)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(+, int, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(+, int, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(+, uint, uint, uint)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(+, uint, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(+, uint, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(+, float, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(+, float, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(+, double, double, double)
+
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(-, uchar, int, int)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(-, uchar, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(-, uchar, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(-, char, int, int)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(-, char, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(-, char, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(-, ushort, int, int)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(-, ushort, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(-, ushort, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(-, short, int, int)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(-, short, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(-, short, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(-, int, int, int)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(-, int, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(-, int, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(-, uint, uint, uint)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(-, uint, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(-, uint, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(-, float, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(-, float, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(-, double, double, double)
+
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(*, uchar, int, int)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(*, uchar, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(*, uchar, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(*, char, int, int)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(*, char, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(*, char, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(*, ushort, int, int)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(*, ushort, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(*, ushort, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(*, short, int, int)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(*, short, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(*, short, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(*, int, int, int)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(*, int, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(*, int, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(*, uint, uint, uint)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(*, uint, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(*, uint, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(*, float, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(*, float, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(*, double, double, double)
+
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(/, uchar, int, int)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(/, uchar, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(/, uchar, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(/, char, int, int)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(/, char, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(/, char, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(/, ushort, int, int)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(/, ushort, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(/, ushort, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(/, short, int, int)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(/, short, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(/, short, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(/, int, int, int)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(/, int, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(/, int, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(/, uint, uint, uint)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(/, uint, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(/, uint, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(/, float, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(/, float, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(/, double, double, double)
+
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(==, uchar, uchar, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(==, char, char, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(==, ushort, ushort, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(==, short, short, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(==, int, int, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(==, uint, uint, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(==, float, float, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(==, double, double, uchar)
+
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(!=, uchar, uchar, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(!=, char, char, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(!=, ushort, ushort, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(!=, short, short, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(!=, int, int, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(!=, uint, uint, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(!=, float, float, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(!=, double, double, uchar)
+
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(>, uchar, uchar, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(>, char, char, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(>, ushort, ushort, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(>, short, short, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(>, int, int, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(>, uint, uint, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(>, float, float, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(>, double, double, uchar)
+
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(<, uchar, uchar, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(<, char, char, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(<, ushort, ushort, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(<, short, short, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(<, int, int, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(<, uint, uint, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(<, float, float, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(<, double, double, uchar)
+
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(>=, uchar, uchar, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(>=, char, char, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(>=, ushort, ushort, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(>=, short, short, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(>=, int, int, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(>=, uint, uint, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(>=, float, float, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(>=, double, double, uchar)
+
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(<=, uchar, uchar, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(<=, char, char, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(<=, ushort, ushort, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(<=, short, short, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(<=, int, int, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(<=, uint, uint, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(<=, float, float, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(<=, double, double, uchar)
+
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(&&, uchar, uchar, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(&&, char, char, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(&&, ushort, ushort, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(&&, short, short, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(&&, int, int, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(&&, uint, uint, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(&&, float, float, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(&&, double, double, uchar)
+
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(||, uchar, uchar, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(||, char, char, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(||, ushort, ushort, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(||, short, short, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(||, int, int, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(||, uint, uint, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(||, float, float, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(||, double, double, uchar)
+
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(&, uchar, uchar, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(&, char, char, char)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(&, ushort, ushort, ushort)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(&, short, short, short)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(&, int, int, int)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(&, uint, uint, uint)
+
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(|, uchar, uchar, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(|, char, char, char)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(|, ushort, ushort, ushort)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(|, short, short, short)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(|, int, int, int)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(|, uint, uint, uint)
+
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(^, uchar, uchar, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(^, char, char, char)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(^, ushort, ushort, ushort)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(^, short, short, short)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(^, int, int, int)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP(^, uint, uint, uint)
+
+#undef CV_CUDEV_IMPLEMENT_SCALAR_BINARY_OP
+
+// binary function (vec & vec)
+
+#define CV_CUDEV_IMPLEMENT_VEC_BINARY_FUNC(func_name, func, input_type, output_type) \
+    __device__ __forceinline__ output_type ## 1 func_name(const input_type ## 1 & a, const input_type ## 1 & b) \
+    { \
+        return VecTraits<output_type ## 1>::make(func (a.x, b.x)); \
+    } \
+    __device__ __forceinline__ output_type ## 2 func_name(const input_type ## 2 & a, const input_type ## 2 & b) \
+    { \
+        return VecTraits<output_type ## 2>::make(func (a.x, b.x), func (a.y, b.y)); \
+    } \
+    __device__ __forceinline__ output_type ## 3 func_name(const input_type ## 3 & a, const input_type ## 3 & b) \
+    { \
+        return VecTraits<output_type ## 3>::make(func (a.x, b.x), func (a.y, b.y), func (a.z, b.z)); \
+    } \
+    __device__ __forceinline__ output_type ## 4 func_name(const input_type ## 4 & a, const input_type ## 4 & b) \
+    { \
+        return VecTraits<output_type ## 4>::make(func (a.x, b.x), func (a.y, b.y), func (a.z, b.z), func (a.w, b.w)); \
+    }
+
+CV_CUDEV_IMPLEMENT_VEC_BINARY_FUNC(max, ::max, uchar, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_FUNC(max, ::max, char, char)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_FUNC(max, ::max, ushort, ushort)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_FUNC(max, ::max, short, short)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_FUNC(max, ::max, uint, uint)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_FUNC(max, ::max, int, int)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_FUNC(max, ::fmaxf, float, float)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_FUNC(max, ::fmax, double, double)
+
+CV_CUDEV_IMPLEMENT_VEC_BINARY_FUNC(min, ::min, uchar, uchar)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_FUNC(min, ::min, char, char)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_FUNC(min, ::min, ushort, ushort)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_FUNC(min, ::min, short, short)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_FUNC(min, ::min, uint, uint)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_FUNC(min, ::min, int, int)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_FUNC(min, ::fminf, float, float)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_FUNC(min, ::fmin, double, double)
+
+CV_CUDEV_IMPLEMENT_VEC_BINARY_FUNC(hypot, ::hypotf, uchar, float)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_FUNC(hypot, ::hypotf, char, float)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_FUNC(hypot, ::hypotf, ushort, float)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_FUNC(hypot, ::hypotf, short, float)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_FUNC(hypot, ::hypotf, uint, float)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_FUNC(hypot, ::hypotf, int, float)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_FUNC(hypot, ::hypotf, float, float)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_FUNC(hypot, ::hypot, double, double)
+
+CV_CUDEV_IMPLEMENT_VEC_BINARY_FUNC(atan2, ::atan2f, uchar, float)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_FUNC(atan2, ::atan2f, char, float)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_FUNC(atan2, ::atan2f, ushort, float)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_FUNC(atan2, ::atan2f, short, float)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_FUNC(atan2, ::atan2f, uint, float)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_FUNC(atan2, ::atan2f, int, float)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_FUNC(atan2, ::atan2f, float, float)
+CV_CUDEV_IMPLEMENT_VEC_BINARY_FUNC(atan2, ::atan2, double, double)
+
+#undef CV_CUDEV_IMPLEMENT_VEC_BINARY_FUNC
+
+// binary function (vec & scalar)
+
+#define CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(func_name, func, input_type, scalar_type, output_type) \
+    __device__ __forceinline__ output_type ## 1 func_name(const input_type ## 1 & a, scalar_type s) \
+    { \
+        return VecTraits<output_type ## 1>::make(func ((output_type) a.x, (output_type) s)); \
+    } \
+    __device__ __forceinline__ output_type ## 1 func_name(scalar_type s, const input_type ## 1 & b) \
+    { \
+        return VecTraits<output_type ## 1>::make(func ((output_type) s, (output_type) b.x)); \
+    } \
+    __device__ __forceinline__ output_type ## 2 func_name(const input_type ## 2 & a, scalar_type s) \
+    { \
+        return VecTraits<output_type ## 2>::make(func ((output_type) a.x, (output_type) s), func ((output_type) a.y, (output_type) s)); \
+    } \
+    __device__ __forceinline__ output_type ## 2 func_name(scalar_type s, const input_type ## 2 & b) \
+    { \
+        return VecTraits<output_type ## 2>::make(func ((output_type) s, (output_type) b.x), func ((output_type) s, (output_type) b.y)); \
+    } \
+    __device__ __forceinline__ output_type ## 3 func_name(const input_type ## 3 & a, scalar_type s) \
+    { \
+        return VecTraits<output_type ## 3>::make(func ((output_type) a.x, (output_type) s), func ((output_type) a.y, (output_type) s), func ((output_type) a.z, (output_type) s)); \
+    } \
+    __device__ __forceinline__ output_type ## 3 func_name(scalar_type s, const input_type ## 3 & b) \
+    { \
+        return VecTraits<output_type ## 3>::make(func ((output_type) s, (output_type) b.x), func ((output_type) s, (output_type) b.y), func ((output_type) s, (output_type) b.z)); \
+    } \
+    __device__ __forceinline__ output_type ## 4 func_name(const input_type ## 4 & a, scalar_type s) \
+    { \
+        return VecTraits<output_type ## 4>::make(func ((output_type) a.x, (output_type) s), func ((output_type) a.y, (output_type) s), func ((output_type) a.z, (output_type) s), func ((output_type) a.w, (output_type) s)); \
+    } \
+    __device__ __forceinline__ output_type ## 4 func_name(scalar_type s, const input_type ## 4 & b) \
+    { \
+        return VecTraits<output_type ## 4>::make(func ((output_type) s, (output_type) b.x), func ((output_type) s, (output_type) b.y), func ((output_type) s, (output_type) b.z), func ((output_type) s, (output_type) b.w)); \
+    }
+
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(max, ::max, uchar, uchar, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(max, ::fmaxf, uchar, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(max, ::fmax, uchar, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(max, ::max, char, char, char)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(max, ::fmaxf, char, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(max, ::fmax, char, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(max, ::max, ushort, ushort, ushort)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(max, ::fmaxf, ushort, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(max, ::fmax, ushort, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(max, ::max, short, short, short)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(max, ::fmaxf, short, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(max, ::fmax, short, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(max, ::max, uint, uint, uint)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(max, ::fmaxf, uint, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(max, ::fmax, uint, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(max, ::max, int, int, int)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(max, ::fmaxf, int, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(max, ::fmax, int, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(max, ::fmaxf, float, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(max, ::fmax, float, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(max, ::fmax, double, double, double)
+
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(min, ::min, uchar, uchar, uchar)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(min, ::fminf, uchar, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(min, ::fmin, uchar, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(min, ::min, char, char, char)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(min, ::fminf, char, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(min, ::fmin, char, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(min, ::min, ushort, ushort, ushort)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(min, ::fminf, ushort, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(min, ::fmin, ushort, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(min, ::min, short, short, short)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(min, ::fminf, short, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(min, ::fmin, short, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(min, ::min, uint, uint, uint)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(min, ::fminf, uint, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(min, ::fmin, uint, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(min, ::min, int, int, int)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(min, ::fminf, int, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(min, ::fmin, int, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(min, ::fminf, float, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(min, ::fmin, float, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(min, ::fmin, double, double, double)
+
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(hypot, ::hypotf, uchar, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(hypot, ::hypot, uchar, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(hypot, ::hypotf, char, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(hypot, ::hypot, char, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(hypot, ::hypotf, ushort, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(hypot, ::hypot, ushort, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(hypot, ::hypotf, short, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(hypot, ::hypot, short, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(hypot, ::hypotf, uint, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(hypot, ::hypot, uint, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(hypot, ::hypotf, int, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(hypot, ::hypot, int, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(hypot, ::hypotf, float, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(hypot, ::hypot, float, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(hypot, ::hypot, double, double, double)
+
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(atan2, ::atan2f, uchar, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(atan2, ::atan2, uchar, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(atan2, ::atan2f, char, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(atan2, ::atan2, char, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(atan2, ::atan2f, ushort, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(atan2, ::atan2, ushort, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(atan2, ::atan2f, short, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(atan2, ::atan2, short, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(atan2, ::atan2f, uint, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(atan2, ::atan2, uint, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(atan2, ::atan2f, int, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(atan2, ::atan2, int, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(atan2, ::atan2f, float, float, float)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(atan2, ::atan2, float, double, double)
+CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC(atan2, ::atan2, double, double, double)
+
+#undef CV_CUDEV_IMPLEMENT_SCALAR_BINARY_FUNC
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/util/vec_traits.hpp

@@ -0,0 +1,320 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_UTIL_VEC_TRAITS_HPP__
+#define __OPENCV_CUDEV_UTIL_VEC_TRAITS_HPP__
+
+#include "../common.hpp"
+
+namespace cv { namespace cudev {
+
+// MakeVec
+
+template<typename T, int CN> struct MakeVec;
+
+#define CV_CUDEV_MAKE_VEC_INST(elem_type) \
+    template<> struct MakeVec<elem_type, 1> { typedef elem_type      type; }; \
+    template<> struct MakeVec<elem_type, 2> { typedef elem_type ## 2 type; }; \
+    template<> struct MakeVec<elem_type, 3> { typedef elem_type ## 3 type; }; \
+    template<> struct MakeVec<elem_type, 4> { typedef elem_type ## 4 type; };
+
+CV_CUDEV_MAKE_VEC_INST(uchar)
+CV_CUDEV_MAKE_VEC_INST(ushort)
+CV_CUDEV_MAKE_VEC_INST(short)
+CV_CUDEV_MAKE_VEC_INST(int)
+CV_CUDEV_MAKE_VEC_INST(uint)
+CV_CUDEV_MAKE_VEC_INST(float)
+CV_CUDEV_MAKE_VEC_INST(double)
+
+#undef CV_CUDEV_MAKE_VEC_INST
+
+template<> struct MakeVec<schar, 1> { typedef char  type; };
+template<> struct MakeVec<schar, 2> { typedef char2 type; };
+template<> struct MakeVec<schar, 3> { typedef char3 type; };
+template<> struct MakeVec<schar, 4> { typedef char4 type; };
+
+template<> struct MakeVec<bool, 1> { typedef uchar  type; };
+template<> struct MakeVec<bool, 2> { typedef uchar2 type; };
+template<> struct MakeVec<bool, 3> { typedef uchar3 type; };
+template<> struct MakeVec<bool, 4> { typedef uchar4 type; };
+
+// VecTraits
+
+template<typename T> struct VecTraits;
+
+#define CV_CUDEV_VEC_TRAITS_INST(type) \
+    template <> struct VecTraits<type> \
+    { \
+        typedef type elem_type; \
+        enum {cn=1}; \
+        __host__ __device__ __forceinline__ static type all(type v) {return v;} \
+        __host__ __device__ __forceinline__ static type make(type x) {return x;} \
+        __host__ __device__ __forceinline__ static type make(const type* v) {return *v;} \
+    }; \
+    template <> struct VecTraits<type ## 1> \
+    { \
+        typedef type elem_type; \
+        enum {cn=1}; \
+        __host__ __device__ __forceinline__ static type ## 1 all(type v) {return make_ ## type ## 1(v);} \
+        __host__ __device__ __forceinline__ static type ## 1 make(type x) {return make_ ## type ## 1(x);} \
+        __host__ __device__ __forceinline__ static type ## 1 make(const type* v) {return make_ ## type ## 1(*v);} \
+    }; \
+    template <> struct VecTraits<type ## 2> \
+    { \
+        typedef type elem_type; \
+        enum {cn=2}; \
+        __host__ __device__ __forceinline__ static type ## 2 all(type v) {return make_ ## type ## 2(v, v);} \
+        __host__ __device__ __forceinline__ static type ## 2 make(type x, type y) {return make_ ## type ## 2(x, y);} \
+        __host__ __device__ __forceinline__ static type ## 2 make(const type* v) {return make_ ## type ## 2(v[0], v[1]);} \
+    }; \
+    template <> struct VecTraits<type ## 3> \
+    { \
+        typedef type elem_type; \
+        enum {cn=3}; \
+        __host__ __device__ __forceinline__ static type ## 3 all(type v) {return make_ ## type ## 3(v, v, v);} \
+        __host__ __device__ __forceinline__ static type ## 3 make(type x, type y, type z) {return make_ ## type ## 3(x, y, z);} \
+        __host__ __device__ __forceinline__ static type ## 3 make(const type* v) {return make_ ## type ## 3(v[0], v[1], v[2]);} \
+    }; \
+    template <> struct VecTraits<type ## 4> \
+    { \
+        typedef type elem_type; \
+        enum {cn=4}; \
+        __host__ __device__ __forceinline__ static type ## 4 all(type v) {return make_ ## type ## 4(v, v, v, v);} \
+        __host__ __device__ __forceinline__ static type ## 4 make(type x, type y, type z, type w) {return make_ ## type ## 4(x, y, z, w);} \
+        __host__ __device__ __forceinline__ static type ## 4 make(const type* v) {return make_ ## type ## 4(v[0], v[1], v[2], v[3]);} \
+    };
+
+CV_CUDEV_VEC_TRAITS_INST(uchar)
+CV_CUDEV_VEC_TRAITS_INST(ushort)
+CV_CUDEV_VEC_TRAITS_INST(short)
+CV_CUDEV_VEC_TRAITS_INST(int)
+CV_CUDEV_VEC_TRAITS_INST(uint)
+CV_CUDEV_VEC_TRAITS_INST(float)
+CV_CUDEV_VEC_TRAITS_INST(double)
+
+#undef CV_CUDEV_VEC_TRAITS_INST
+
+template<> struct VecTraits<schar>
+{
+    typedef schar elem_type;
+    enum {cn=1};
+    __host__ __device__ __forceinline__ static schar all(schar v) {return v;}
+    __host__ __device__ __forceinline__ static schar make(schar x) {return x;}
+    __host__ __device__ __forceinline__ static schar make(const schar* x) {return *x;}
+};
+template<> struct VecTraits<char1>
+{
+    typedef schar elem_type;
+    enum {cn=1};
+    __host__ __device__ __forceinline__ static char1 all(schar v) {return make_char1(v);}
+    __host__ __device__ __forceinline__ static char1 make(schar x) {return make_char1(x);}
+    __host__ __device__ __forceinline__ static char1 make(const schar* v) {return make_char1(v[0]);}
+};
+template<> struct VecTraits<char2>
+{
+    typedef schar elem_type;
+    enum {cn=2};
+    __host__ __device__ __forceinline__ static char2 all(schar v) {return make_char2(v, v);}
+    __host__ __device__ __forceinline__ static char2 make(schar x, schar y) {return make_char2(x, y);}
+    __host__ __device__ __forceinline__ static char2 make(const schar* v) {return make_char2(v[0], v[1]);}
+};
+template<> struct VecTraits<char3>
+{
+    typedef schar elem_type;
+    enum {cn=3};
+    __host__ __device__ __forceinline__ static char3 all(schar v) {return make_char3(v, v, v);}
+    __host__ __device__ __forceinline__ static char3 make(schar x, schar y, schar z) {return make_char3(x, y, z);}
+    __host__ __device__ __forceinline__ static char3 make(const schar* v) {return make_char3(v[0], v[1], v[2]);}
+};
+template<> struct VecTraits<char4>
+{
+    typedef schar elem_type;
+    enum {cn=4};
+    __host__ __device__ __forceinline__ static char4 all(schar v) {return make_char4(v, v, v, v);}
+    __host__ __device__ __forceinline__ static char4 make(schar x, schar y, schar z, schar w) {return make_char4(x, y, z, w);}
+    __host__ __device__ __forceinline__ static char4 make(const schar* v) {return make_char4(v[0], v[1], v[2], v[3]);}
+};
+
+}}
+
+// DataType
+
+namespace cv {
+
+template <> class DataType<uint>
+{
+public:
+    typedef uint         value_type;
+    typedef value_type   work_type;
+    typedef value_type   channel_type;
+    typedef value_type   vec_type;
+    enum { generic_type = 0,
+           depth        = CV_32S,
+           channels     = 1,
+           fmt          = (int)'i',
+           type         = CV_MAKE_TYPE(depth, channels)
+         };
+};
+
+#define CV_CUDEV_DATA_TYPE_INST(_depth_type, _channel_num) \
+    template <> class DataType< _depth_type ## _channel_num > \
+    { \
+    public: \
+        typedef _depth_type ## _channel_num     value_type; \
+        typedef value_type                      work_type; \
+        typedef _depth_type                     channel_type; \
+        typedef value_type                      vec_type; \
+        enum { generic_type = 0, \
+               depth        = DataType<channel_type>::depth, \
+               channels     = _channel_num, \
+               fmt          = DataType<channel_type>::fmt + ((channels - 1) << 8), \
+               type         = CV_MAKE_TYPE(depth, channels) \
+             }; \
+    };
+
+CV_CUDEV_DATA_TYPE_INST(uchar, 1)
+CV_CUDEV_DATA_TYPE_INST(uchar, 2)
+CV_CUDEV_DATA_TYPE_INST(uchar, 3)
+CV_CUDEV_DATA_TYPE_INST(uchar, 4)
+
+CV_CUDEV_DATA_TYPE_INST(ushort, 1)
+CV_CUDEV_DATA_TYPE_INST(ushort, 2)
+CV_CUDEV_DATA_TYPE_INST(ushort, 3)
+CV_CUDEV_DATA_TYPE_INST(ushort, 4)
+
+CV_CUDEV_DATA_TYPE_INST(short, 1)
+CV_CUDEV_DATA_TYPE_INST(short, 2)
+CV_CUDEV_DATA_TYPE_INST(short, 3)
+CV_CUDEV_DATA_TYPE_INST(short, 4)
+
+CV_CUDEV_DATA_TYPE_INST(int, 1)
+CV_CUDEV_DATA_TYPE_INST(int, 2)
+CV_CUDEV_DATA_TYPE_INST(int, 3)
+CV_CUDEV_DATA_TYPE_INST(int, 4)
+
+CV_CUDEV_DATA_TYPE_INST(uint, 1)
+CV_CUDEV_DATA_TYPE_INST(uint, 2)
+CV_CUDEV_DATA_TYPE_INST(uint, 3)
+CV_CUDEV_DATA_TYPE_INST(uint, 4)
+
+CV_CUDEV_DATA_TYPE_INST(float, 1)
+CV_CUDEV_DATA_TYPE_INST(float, 2)
+CV_CUDEV_DATA_TYPE_INST(float, 3)
+CV_CUDEV_DATA_TYPE_INST(float, 4)
+
+CV_CUDEV_DATA_TYPE_INST(double, 1)
+CV_CUDEV_DATA_TYPE_INST(double, 2)
+CV_CUDEV_DATA_TYPE_INST(double, 3)
+CV_CUDEV_DATA_TYPE_INST(double, 4)
+
+#undef CV_CUDEV_DATA_TYPE_INST
+
+template<> class DataType<char1>
+{
+public:
+    typedef char1      value_type;
+    typedef value_type work_type;
+    typedef schar      channel_type;
+    typedef value_type vec_type;
+
+    enum { generic_type = 0,
+           depth        = DataType<channel_type>::depth,
+           channels     = 1,
+           fmt          = DataType<channel_type>::fmt + ((channels - 1) << 8),
+           type         = CV_MAKE_TYPE(depth, channels)
+         };
+};
+
+template<> class DataType<char2>
+{
+public:
+    typedef char2      value_type;
+    typedef value_type work_type;
+    typedef schar      channel_type;
+    typedef value_type vec_type;
+
+    enum { generic_type = 0,
+           depth        = DataType<channel_type>::depth,
+           channels     = 2,
+           fmt          = DataType<channel_type>::fmt + ((channels - 1) << 8),
+           type         = CV_MAKE_TYPE(depth, channels)
+         };
+};
+
+template<> class DataType<char3>
+{
+public:
+    typedef char3      value_type;
+    typedef value_type work_type;
+    typedef schar      channel_type;
+    typedef value_type vec_type;
+
+    enum { generic_type = 0,
+           depth        = DataType<channel_type>::depth,
+           channels     = 3,
+           fmt          = DataType<channel_type>::fmt + ((channels - 1) << 8),
+           type         = CV_MAKE_TYPE(depth, channels)
+         };
+};
+
+template<> class DataType<char4>
+{
+public:
+    typedef char4      value_type;
+    typedef value_type work_type;
+    typedef schar      channel_type;
+    typedef value_type vec_type;
+
+    enum { generic_type = 0,
+           depth        = DataType<channel_type>::depth,
+           channels     = 4,
+           fmt          = DataType<channel_type>::fmt + ((channels - 1) << 8),
+           type         = CV_MAKE_TYPE(depth, channels)
+         };
+};
+
+}
+
+#endif

modules/cudev/include/opencv2/cudev/warp/detail/reduce.hpp

@@ -0,0 +1,222 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_WARP_REDUCE_DETAIL_HPP__
+#define __OPENCV_CUDEV_WARP_REDUCE_DETAIL_HPP__
+
+#include "../../common.hpp"
+#include "../../util/tuple.hpp"
+#include "../../warp/shuffle.hpp"
+
+namespace cv { namespace cudev {
+
+namespace warp_reduce_detail
+{
+    // GetType
+
+    template <typename T> struct GetType;
+
+    template <typename T> struct GetType<T*>
+    {
+        typedef T type;
+    };
+
+    template <typename T> struct GetType<volatile T*>
+    {
+        typedef T type;
+    };
+
+    template <typename T> struct GetType<T&>
+    {
+        typedef T type;
+    };
+
+    // For
+
+    template <int I, int N> struct For
+    {
+        template <class PointerTuple, class ValTuple>
+        __device__ static void loadToSmem(const PointerTuple& smem, const ValTuple& val, uint tid)
+        {
+            get<I>(smem)[tid] = get<I>(val);
+
+            For<I + 1, N>::loadToSmem(smem, val, tid);
+        }
+
+        template <class PointerTuple, class ValTuple, class OpTuple>
+        __device__ static void merge(const PointerTuple& smem, const ValTuple& val, uint tid, uint delta, const OpTuple& op)
+        {
+            typename GetType<typename tuple_element<I, PointerTuple>::type>::type reg = get<I>(smem)[tid + delta];
+            get<I>(smem)[tid] = get<I>(val) = get<I>(op)(get<I>(val), reg);
+
+            For<I + 1, N>::merge(smem, val, tid, delta, op);
+        }
+
+#if CV_CUDEV_ARCH >= 300
+        template <class ValTuple, class OpTuple>
+        __device__ static void mergeShfl(const ValTuple& val, uint delta, uint width, const OpTuple& op)
+        {
+            typename GetType<typename tuple_element<I, ValTuple>::type>::type reg = shfl_down(get<I>(val), delta, width);
+            get<I>(val) = get<I>(op)(get<I>(val), reg);
+
+            For<I + 1, N>::mergeShfl(val, delta, width, op);
+        }
+#endif
+    };
+
+    template <int N> struct For<N, N>
+    {
+        template <class PointerTuple, class ValTuple>
+        __device__ __forceinline__ static void loadToSmem(const PointerTuple&, const ValTuple&, uint)
+        {
+        }
+
+        template <class PointerTuple, class ValTuple, class OpTuple>
+        __device__ __forceinline__ static void merge(const PointerTuple&, const ValTuple&, uint, uint, const OpTuple&)
+        {
+        }
+
+#if CV_CUDEV_ARCH >= 300
+        template <class ValTuple, class OpTuple>
+        __device__ __forceinline__ static void mergeShfl(const ValTuple&, uint, uint, const OpTuple&)
+        {
+        }
+#endif
+    };
+
+    // loadToSmem
+
+    template <typename T>
+    __device__ __forceinline__ void loadToSmem(volatile T* smem, T& val, uint tid)
+    {
+        smem[tid] = val;
+    }
+
+    template <typename P0, typename P1, typename P2, typename P3, typename P4, typename P5, typename P6, typename P7, typename P8, typename P9,
+              typename R0, typename R1, typename R2, typename R3, typename R4, typename R5, typename R6, typename R7, typename R8, typename R9>
+    __device__ __forceinline__ void loadToSmem(const tuple<P0, P1, P2, P3, P4, P5, P6, P7, P8, P9>& smem,
+                                               const tuple<R0, R1, R2, R3, R4, R5, R6, R7, R8, R9>& val,
+                                               uint tid)
+    {
+        For<0, tuple_size<tuple<P0, P1, P2, P3, P4, P5, P6, P7, P8, P9> >::value>::loadToSmem(smem, val, tid);
+    }
+
+    // merge
+
+    template <typename T, class Op>
+    __device__ __forceinline__ void merge(volatile T* smem, T& val, uint tid, uint delta, const Op& op)
+    {
+        T reg = smem[tid + delta];
+        smem[tid] = val = op(val, reg);
+    }
+
+    template <typename P0, typename P1, typename P2, typename P3, typename P4, typename P5, typename P6, typename P7, typename P8, typename P9,
+              typename R0, typename R1, typename R2, typename R3, typename R4, typename R5, typename R6, typename R7, typename R8, typename R9,
+              class Op0, class Op1, class Op2, class Op3, class Op4, class Op5, class Op6, class Op7, class Op8, class Op9>
+    __device__ __forceinline__ void merge(const tuple<P0, P1, P2, P3, P4, P5, P6, P7, P8, P9>& smem,
+                                          const tuple<R0, R1, R2, R3, R4, R5, R6, R7, R8, R9>& val,
+                                          uint tid,
+                                          uint delta,
+                                          const tuple<Op0, Op1, Op2, Op3, Op4, Op5, Op6, Op7, Op8, Op9>& op)
+    {
+        For<0, tuple_size<tuple<P0, P1, P2, P3, P4, P5, P6, P7, P8, P9> >::value>::merge(smem, val, tid, delta, op);
+    }
+
+    // mergeShfl
+
+#if CV_CUDEV_ARCH >= 300
+    template <typename T, class Op>
+    __device__ __forceinline__ void mergeShfl(T& val, uint delta, uint width, const Op& op)
+    {
+        T reg = shfl_down(val, delta, width);
+        val = op(val, reg);
+    }
+
+    template <typename R0, typename R1, typename R2, typename R3, typename R4, typename R5, typename R6, typename R7, typename R8, typename R9,
+              class Op0, class Op1, class Op2, class Op3, class Op4, class Op5, class Op6, class Op7, class Op8, class Op9>
+    __device__ __forceinline__ void mergeShfl(const tuple<R0, R1, R2, R3, R4, R5, R6, R7, R8, R9>& val,
+                                              uint delta,
+                                              uint width,
+                                              const tuple<Op0, Op1, Op2, Op3, Op4, Op5, Op6, Op7, Op8, Op9>& op)
+    {
+        For<0, tuple_size<tuple<R0, R1, R2, R3, R4, R5, R6, R7, R8, R9> >::value>::mergeShfl(val, delta, width, op);
+    }
+#endif
+
+    // WarpReductor
+
+    struct WarpReductor
+    {
+        template <typename Pointer, typename Reference, class Op>
+        __device__ static void reduce(Pointer smem, Reference val, uint tid, Op op)
+        {
+        #if CV_CUDEV_ARCH >= 300
+            (void) smem;
+            (void) tid;
+
+            mergeShfl(val, 16, 32, op);
+            mergeShfl(val, 8, 32, op);
+            mergeShfl(val, 4, 32, op);
+            mergeShfl(val, 2, 32, op);
+            mergeShfl(val, 1, 32, op);
+        #else
+            loadToSmem(smem, val, tid);
+
+            if (tid < 16)
+            {
+                merge(smem, val, tid, 16, op);
+                merge(smem, val, tid, 8, op);
+                merge(smem, val, tid, 4, op);
+                merge(smem, val, tid, 2, op);
+                merge(smem, val, tid, 1, op);
+            }
+        #endif
+        }
+    };
+}
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/warp/detail/reduce_key_val.hpp

@@ -0,0 +1,239 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_WARP_REDUCE_KEY_VAL_DETAIL_HPP__
+#define __OPENCV_CUDEV_WARP_REDUCE_KEY_VAL_DETAIL_HPP__
+
+#include "../../common.hpp"
+#include "../../util/tuple.hpp"
+
+namespace cv { namespace cudev {
+
+namespace warp_reduce_key_val_detail
+{
+    // GetType
+
+    template <typename T> struct GetType;
+
+    template <typename T> struct GetType<T*>
+    {
+        typedef T type;
+    };
+
+    template <typename T> struct GetType<volatile T*>
+    {
+        typedef T type;
+    };
+
+    template <typename T> struct GetType<T&>
+    {
+        typedef T type;
+    };
+
+    // For
+
+    template <int I, int N> struct For
+    {
+        template <class PointerTuple, class ReferenceTuple>
+        __device__ static void loadToSmem(const PointerTuple& smem, const ReferenceTuple& data, uint tid)
+        {
+            get<I>(smem)[tid] = get<I>(data);
+
+            For<I + 1, N>::loadToSmem(smem, data, tid);
+        }
+
+        template <class PointerTuple, class ReferenceTuple>
+        __device__ static void copy(const PointerTuple& svals, const ReferenceTuple& val, uint tid, uint delta)
+        {
+            get<I>(svals)[tid] = get<I>(val) = get<I>(svals)[tid + delta];
+
+            For<I + 1, N>::copy(svals, val, tid, delta);
+        }
+
+        template <class KeyPointerTuple, class KeyReferenceTuple, class ValPointerTuple, class ValReferenceTuple, class CmpTuple>
+        __device__ static void merge(const KeyPointerTuple& skeys, const KeyReferenceTuple& key,
+                                     const ValPointerTuple& svals, const ValReferenceTuple& val,
+                                     const CmpTuple& cmp,
+                                     uint tid, uint delta)
+        {
+            typename GetType<typename tuple_element<I, KeyPointerTuple>::type>::type reg = get<I>(skeys)[tid + delta];
+
+            if (get<I>(cmp)(reg, get<I>(key)))
+            {
+                get<I>(skeys)[tid] = get<I>(key) = reg;
+                get<I>(svals)[tid] = get<I>(val) = get<I>(svals)[tid + delta];
+            }
+
+            For<I + 1, N>::merge(skeys, key, svals, val, cmp, tid, delta);
+        }
+    };
+
+    template <int N> struct For<N, N>
+    {
+        template <class PointerTuple, class ReferenceTuple>
+        __device__ __forceinline__ static void loadToSmem(const PointerTuple&, const ReferenceTuple&, uint)
+        {
+        }
+
+        template <class PointerTuple, class ReferenceTuple>
+        __device__ __forceinline__ static void copy(const PointerTuple&, const ReferenceTuple&, uint, uint)
+        {
+        }
+
+        template <class KeyPointerTuple, class KeyReferenceTuple, class ValPointerTuple, class ValReferenceTuple, class CmpTuple>
+        __device__ __forceinline__ static void merge(const KeyPointerTuple&, const KeyReferenceTuple&,
+                                                     const ValPointerTuple&, const ValReferenceTuple&,
+                                                     const CmpTuple&,
+                                                     uint, uint)
+        {
+        }
+    };
+
+    // loadToSmem
+
+    template <typename T>
+    __device__ __forceinline__ void loadToSmem(volatile T* smem, T& data, uint tid)
+    {
+        smem[tid] = data;
+    }
+
+    template <typename VP0, typename VP1, typename VP2, typename VP3, typename VP4, typename VP5, typename VP6, typename VP7, typename VP8, typename VP9,
+              typename VR0, typename VR1, typename VR2, typename VR3, typename VR4, typename VR5, typename VR6, typename VR7, typename VR8, typename VR9>
+    __device__ __forceinline__ void loadToSmem(const tuple<VP0, VP1, VP2, VP3, VP4, VP5, VP6, VP7, VP8, VP9>& smem,
+                                               const tuple<VR0, VR1, VR2, VR3, VR4, VR5, VR6, VR7, VR8, VR9>& data,
+                                               uint tid)
+    {
+        For<0, tuple_size<tuple<VP0, VP1, VP2, VP3, VP4, VP5, VP6, VP7, VP8, VP9> >::value>::loadToSmem(smem, data, tid);
+    }
+
+    // copyVals
+
+    template <typename V>
+    __device__ __forceinline__ void copyVals(volatile V* svals, V& val, uint tid, uint delta)
+    {
+        svals[tid] = val = svals[tid + delta];
+    }
+
+    template <typename VP0, typename VP1, typename VP2, typename VP3, typename VP4, typename VP5, typename VP6, typename VP7, typename VP8, typename VP9,
+              typename VR0, typename VR1, typename VR2, typename VR3, typename VR4, typename VR5, typename VR6, typename VR7, typename VR8, typename VR9>
+    __device__ __forceinline__ void copyVals(const tuple<VP0, VP1, VP2, VP3, VP4, VP5, VP6, VP7, VP8, VP9>& svals,
+                                             const tuple<VR0, VR1, VR2, VR3, VR4, VR5, VR6, VR7, VR8, VR9>& val,
+                                             uint tid, uint delta)
+    {
+        For<0, tuple_size<tuple<VP0, VP1, VP2, VP3, VP4, VP5, VP6, VP7, VP8, VP9> >::value>::copy(svals, val, tid, delta);
+    }
+
+    // merge
+
+    template <typename K, typename V, class Cmp>
+    __device__ void merge(volatile K* skeys, K& key, volatile V* svals, V& val, const Cmp& cmp, uint tid, uint delta)
+    {
+        K reg = skeys[tid + delta];
+
+        if (cmp(reg, key))
+        {
+            skeys[tid] = key = reg;
+            copyVals(svals, val, tid, delta);
+        }
+    }
+
+    template <typename K,
+              typename VP0, typename VP1, typename VP2, typename VP3, typename VP4, typename VP5, typename VP6, typename VP7, typename VP8, typename VP9,
+              typename VR0, typename VR1, typename VR2, typename VR3, typename VR4, typename VR5, typename VR6, typename VR7, typename VR8, typename VR9,
+              class Cmp>
+    __device__ void merge(volatile K* skeys, K& key,
+                          const tuple<VP0, VP1, VP2, VP3, VP4, VP5, VP6, VP7, VP8, VP9>& svals,
+                          const tuple<VR0, VR1, VR2, VR3, VR4, VR5, VR6, VR7, VR8, VR9>& val,
+                          const Cmp& cmp, uint tid, uint delta)
+    {
+        K reg = skeys[tid + delta];
+
+        if (cmp(reg, key))
+        {
+            skeys[tid] = key = reg;
+            copyVals(svals, val, tid, delta);
+        }
+    }
+
+    template <typename KP0, typename KP1, typename KP2, typename KP3, typename KP4, typename KP5, typename KP6, typename KP7, typename KP8, typename KP9,
+              typename KR0, typename KR1, typename KR2, typename KR3, typename KR4, typename KR5, typename KR6, typename KR7, typename KR8, typename KR9,
+              typename VP0, typename VP1, typename VP2, typename VP3, typename VP4, typename VP5, typename VP6, typename VP7, typename VP8, typename VP9,
+              typename VR0, typename VR1, typename VR2, typename VR3, typename VR4, typename VR5, typename VR6, typename VR7, typename VR8, typename VR9,
+              class Cmp0, class Cmp1, class Cmp2, class Cmp3, class Cmp4, class Cmp5, class Cmp6, class Cmp7, class Cmp8, class Cmp9>
+    __device__ __forceinline__ void merge(const tuple<KP0, KP1, KP2, KP3, KP4, KP5, KP6, KP7, KP8, KP9>& skeys,
+                                          const tuple<KR0, KR1, KR2, KR3, KR4, KR5, KR6, KR7, KR8, KR9>& key,
+                                          const tuple<VP0, VP1, VP2, VP3, VP4, VP5, VP6, VP7, VP8, VP9>& svals,
+                                          const tuple<VR0, VR1, VR2, VR3, VR4, VR5, VR6, VR7, VR8, VR9>& val,
+                                          const tuple<Cmp0, Cmp1, Cmp2, Cmp3, Cmp4, Cmp5, Cmp6, Cmp7, Cmp8, Cmp9>& cmp,
+                                          uint tid, uint delta)
+    {
+        For<0, tuple_size<tuple<VP0, VP1, VP2, VP3, VP4, VP5, VP6, VP7, VP8, VP9> >::value>::merge(skeys, key, svals, val, cmp, tid, delta);
+    }
+
+    // WarpReductor
+
+    struct WarpReductor
+    {
+        template <class KP, class KR, class VP, class VR, class Cmp>
+        __device__ static void reduce(KP skeys, KR key, VP svals, VR val, uint tid, Cmp cmp)
+        {
+            loadToSmem(skeys, key, tid);
+            loadToSmem(svals, val, tid);
+
+            if (tid < 16)
+            {
+                merge(skeys, key, svals, val, cmp, tid, 16);
+                merge(skeys, key, svals, val, cmp, tid, 8);
+                merge(skeys, key, svals, val, cmp, tid, 4);
+                merge(skeys, key, svals, val, cmp, tid, 2);
+                merge(skeys, key, svals, val, cmp, tid, 1);
+            }
+        }
+    };
+}
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/warp/reduce.hpp

@@ -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.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_WARP_REDUCE_HPP__
+#define __OPENCV_CUDEV_WARP_REDUCE_HPP__
+
+#include "../common.hpp"
+#include "../util/tuple.hpp"
+#include "detail/reduce.hpp"
+#include "detail/reduce_key_val.hpp"
+
+namespace cv { namespace cudev {
+
+// warpReduce
+
+template <typename T, class Op>
+__device__ __forceinline__ void warpReduce(volatile T* smem, T& val, uint tid, const Op& op)
+{
+    warp_reduce_detail::WarpReductor::template reduce<volatile T*, T&, const Op&>(smem, val, tid, op);
+}
+
+template <typename P0, typename P1, typename P2, typename P3, typename P4, typename P5, typename P6, typename P7, typename P8, typename P9,
+          typename R0, typename R1, typename R2, typename R3, typename R4, typename R5, typename R6, typename R7, typename R8, typename R9,
+          class Op0, class Op1, class Op2, class Op3, class Op4, class Op5, class Op6, class Op7, class Op8, class Op9>
+__device__ __forceinline__ void warpReduce(const tuple<P0, P1, P2, P3, P4, P5, P6, P7, P8, P9>& smem,
+                                           const tuple<R0, R1, R2, R3, R4, R5, R6, R7, R8, R9>& val,
+                                           uint tid,
+                                           const tuple<Op0, Op1, Op2, Op3, Op4, Op5, Op6, Op7, Op8, Op9>& op)
+{
+    warp_reduce_detail::WarpReductor::template reduce<
+            const tuple<P0, P1, P2, P3, P4, P5, P6, P7, P8, P9>&,
+            const tuple<R0, R1, R2, R3, R4, R5, R6, R7, R8, R9>&,
+            const tuple<Op0, Op1, Op2, Op3, Op4, Op5, Op6, Op7, Op8, Op9>&>(smem, val, tid, op);
+}
+
+// warpReduceKeyVal
+
+template <typename K, typename V, class Cmp>
+__device__ __forceinline__ void warpReduceKeyVal(volatile K* skeys, K& key, volatile V* svals, V& val, uint tid, const Cmp& cmp)
+{
+    warp_reduce_key_val_detail::WarpReductor::template reduce<volatile K*, K&, volatile V*, V&, const Cmp&>(skeys, key, svals, val, tid, cmp);
+}
+
+template <typename K,
+          typename VP0, typename VP1, typename VP2, typename VP3, typename VP4, typename VP5, typename VP6, typename VP7, typename VP8, typename VP9,
+          typename VR0, typename VR1, typename VR2, typename VR3, typename VR4, typename VR5, typename VR6, typename VR7, typename VR8, typename VR9,
+          class Cmp>
+__device__ __forceinline__ void warpReduceKeyVal(volatile K* skeys, K& key,
+                                                 const tuple<VP0, VP1, VP2, VP3, VP4, VP5, VP6, VP7, VP8, VP9>& svals,
+                                                 const tuple<VR0, VR1, VR2, VR3, VR4, VR5, VR6, VR7, VR8, VR9>& val,
+                                                 uint tid, const Cmp& cmp)
+{
+    warp_reduce_key_val_detail::WarpReductor::template reduce<volatile K*, K&,
+            const tuple<VP0, VP1, VP2, VP3, VP4, VP5, VP6, VP7, VP8, VP9>&,
+            const tuple<VR0, VR1, VR2, VR3, VR4, VR5, VR6, VR7, VR8, VR9>&,
+            const Cmp&>(skeys, key, svals, val, tid, cmp);
+}
+
+template <typename KP0, typename KP1, typename KP2, typename KP3, typename KP4, typename KP5, typename KP6, typename KP7, typename KP8, typename KP9,
+          typename KR0, typename KR1, typename KR2, typename KR3, typename KR4, typename KR5, typename KR6, typename KR7, typename KR8, typename KR9,
+          typename VP0, typename VP1, typename VP2, typename VP3, typename VP4, typename VP5, typename VP6, typename VP7, typename VP8, typename VP9,
+          typename VR0, typename VR1, typename VR2, typename VR3, typename VR4, typename VR5, typename VR6, typename VR7, typename VR8, typename VR9,
+          class Cmp0, class Cmp1, class Cmp2, class Cmp3, class Cmp4, class Cmp5, class Cmp6, class Cmp7, class Cmp8, class Cmp9>
+__device__ __forceinline__ void warpReduceKeyVal(const tuple<KP0, KP1, KP2, KP3, KP4, KP5, KP6, KP7, KP8, KP9>& skeys,
+                                                 const tuple<KR0, KR1, KR2, KR3, KR4, KR5, KR6, KR7, KR8, KR9>& key,
+                                                 const tuple<VP0, VP1, VP2, VP3, VP4, VP5, VP6, VP7, VP8, VP9>& svals,
+                                                 const tuple<VR0, VR1, VR2, VR3, VR4, VR5, VR6, VR7, VR8, VR9>& val,
+                                                 uint tid,
+                                                 const tuple<Cmp0, Cmp1, Cmp2, Cmp3, Cmp4, Cmp5, Cmp6, Cmp7, Cmp8, Cmp9>& cmp)
+{
+    warp_reduce_key_val_detail::WarpReductor::template reduce<
+            const tuple<KP0, KP1, KP2, KP3, KP4, KP5, KP6, KP7, KP8, KP9>&,
+            const tuple<KR0, KR1, KR2, KR3, KR4, KR5, KR6, KR7, KR8, KR9>&,
+            const tuple<VP0, VP1, VP2, VP3, VP4, VP5, VP6, VP7, VP8, VP9>&,
+            const tuple<VR0, VR1, VR2, VR3, VR4, VR5, VR6, VR7, VR8, VR9>&,
+            const tuple<Cmp0, Cmp1, Cmp2, Cmp3, Cmp4, Cmp5, Cmp6, Cmp7, Cmp8, Cmp9>&
+            >(skeys, key, svals, val, tid, cmp);
+}
+
+// smem_tuple
+
+template <typename T0>
+__device__ __forceinline__
+tuple<volatile T0*>
+smem_tuple(T0* t0)
+{
+    return make_tuple((volatile T0*) t0);
+}
+
+template <typename T0, typename T1>
+__device__ __forceinline__
+tuple<volatile T0*, volatile T1*>
+smem_tuple(T0* t0, T1* t1)
+{
+    return make_tuple((volatile T0*) t0, (volatile T1*) t1);
+}
+
+template <typename T0, typename T1, typename T2>
+__device__ __forceinline__
+tuple<volatile T0*, volatile T1*, volatile T2*>
+smem_tuple(T0* t0, T1* t1, T2* t2)
+{
+    return make_tuple((volatile T0*) t0, (volatile T1*) t1, (volatile T2*) t2);
+}
+
+template <typename T0, typename T1, typename T2, typename T3>
+__device__ __forceinline__
+tuple<volatile T0*, volatile T1*, volatile T2*, volatile T3*>
+smem_tuple(T0* t0, T1* t1, T2* t2, T3* t3)
+{
+    return make_tuple((volatile T0*) t0, (volatile T1*) t1, (volatile T2*) t2, (volatile T3*) t3);
+}
+
+template <typename T0, typename T1, typename T2, typename T3, typename T4>
+__device__ __forceinline__
+tuple<volatile T0*, volatile T1*, volatile T2*, volatile T3*, volatile T4*>
+smem_tuple(T0* t0, T1* t1, T2* t2, T3* t3, T4* t4)
+{
+    return make_tuple((volatile T0*) t0, (volatile T1*) t1, (volatile T2*) t2, (volatile T3*) t3, (volatile T4*) t4);
+}
+
+template <typename T0, typename T1, typename T2, typename T3, typename T4, typename T5>
+__device__ __forceinline__
+tuple<volatile T0*, volatile T1*, volatile T2*, volatile T3*, volatile T4*, volatile T5*>
+smem_tuple(T0* t0, T1* t1, T2* t2, T3* t3, T4* t4, T5* t5)
+{
+    return make_tuple((volatile T0*) t0, (volatile T1*) t1, (volatile T2*) t2, (volatile T3*) t3, (volatile T4*) t4, (volatile T5*) t5);
+}
+
+template <typename T0, typename T1, typename T2, typename T3, typename T4, typename T5, typename T6>
+__device__ __forceinline__
+tuple<volatile T0*, volatile T1*, volatile T2*, volatile T3*, volatile T4*, volatile T5*, volatile T6*>
+smem_tuple(T0* t0, T1* t1, T2* t2, T3* t3, T4* t4, T5* t5, T6* t6)
+{
+    return make_tuple((volatile T0*) t0, (volatile T1*) t1, (volatile T2*) t2, (volatile T3*) t3, (volatile T4*) t4, (volatile T5*) t5, (volatile T6*) t6);
+}
+
+template <typename T0, typename T1, typename T2, typename T3, typename T4, typename T5, typename T6, typename T7>
+__device__ __forceinline__
+tuple<volatile T0*, volatile T1*, volatile T2*, volatile T3*, volatile T4*, volatile T5*, volatile T6*, volatile T7*>
+smem_tuple(T0* t0, T1* t1, T2* t2, T3* t3, T4* t4, T5* t5, T6* t6, T7* t7)
+{
+    return make_tuple((volatile T0*) t0, (volatile T1*) t1, (volatile T2*) t2, (volatile T3*) t3, (volatile T4*) t4, (volatile T5*) t5, (volatile T6*) t6, (volatile T7*) t7);
+}
+
+template <typename T0, typename T1, typename T2, typename T3, typename T4, typename T5, typename T6, typename T7, typename T8>
+__device__ __forceinline__
+tuple<volatile T0*, volatile T1*, volatile T2*, volatile T3*, volatile T4*, volatile T5*, volatile T6*, volatile T7*, volatile T8*>
+smem_tuple(T0* t0, T1* t1, T2* t2, T3* t3, T4* t4, T5* t5, T6* t6, T7* t7, T8* t8)
+{
+    return make_tuple((volatile T0*) t0, (volatile T1*) t1, (volatile T2*) t2, (volatile T3*) t3, (volatile T4*) t4, (volatile T5*) t5, (volatile T6*) t6, (volatile T7*) t7, (volatile T8*) t8);
+}
+
+template <typename T0, typename T1, typename T2, typename T3, typename T4, typename T5, typename T6, typename T7, typename T8, typename T9>
+__device__ __forceinline__
+tuple<volatile T0*, volatile T1*, volatile T2*, volatile T3*, volatile T4*, volatile T5*, volatile T6*, volatile T7*, volatile T8*, volatile T9*>
+smem_tuple(T0* t0, T1* t1, T2* t2, T3* t3, T4* t4, T5* t5, T6* t6, T7* t7, T8* t8, T9* t9)
+{
+    return make_tuple((volatile T0*) t0, (volatile T1*) t1, (volatile T2*) t2, (volatile T3*) t3, (volatile T4*) t4, (volatile T5*) t5, (volatile T6*) t6, (volatile T7*) t7, (volatile T8*) t8, (volatile T9*) t9);
+}
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/warp/scan.hpp

@@ -0,0 +1,99 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_WARP_SCAN_HPP__
+#define __OPENCV_CUDEV_WARP_SCAN_HPP__
+
+#include "../common.hpp"
+#include "warp.hpp"
+#include "shuffle.hpp"
+
+namespace cv { namespace cudev {
+
+template <typename T>
+__device__ T warpScanInclusive(T data, volatile T* smem, uint tid)
+{
+#if CV_CUDEV_ARCH >= 300
+    (void) smem;
+    (void) tid;
+
+    const uint laneId = Warp::laneId();
+
+    // scan on shufl functions
+    #pragma unroll
+    for (int i = 1; i <= (WARP_SIZE / 2); i *= 2)
+    {
+        const T val = shfl_up(data, i);
+        if (laneId >= i)
+              data += val;
+    }
+
+    return data;
+#else
+    uint pos = 2 * tid - (tid & (WARP_SIZE - 1));
+    smem[pos] = 0;
+
+    pos += WARP_SIZE;
+    smem[pos] = data;
+
+    smem[pos] += smem[pos - 1];
+    smem[pos] += smem[pos - 2];
+    smem[pos] += smem[pos - 4];
+    smem[pos] += smem[pos - 8];
+    smem[pos] += smem[pos - 16];
+
+    return smem[pos];
+#endif
+}
+
+template <typename T>
+__device__ __forceinline__ T warpScanExclusive(T data, volatile T* smem, uint tid)
+{
+    return warpScanInclusive(data, smem, tid) - data;
+}
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/warp/shuffle.hpp

@@ -0,0 +1,424 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_WARP_SHUFFLE_HPP__
+#define __OPENCV_CUDEV_WARP_SHUFFLE_HPP__
+
+#include "../common.hpp"
+#include "../util/vec_traits.hpp"
+
+namespace cv { namespace cudev {
+
+#if CV_CUDEV_ARCH >= 300
+
+// shfl
+
+__device__ __forceinline__ uchar shfl(uchar val, int srcLane, int width = warpSize)
+{
+    return (uchar) __shfl((int) val, srcLane, width);
+}
+
+__device__ __forceinline__ schar shfl(schar val, int srcLane, int width = warpSize)
+{
+    return (schar) __shfl((int) val, srcLane, width);
+}
+
+__device__ __forceinline__ ushort shfl(ushort val, int srcLane, int width = warpSize)
+{
+    return (ushort) __shfl((int) val, srcLane, width);
+}
+
+__device__ __forceinline__ short shfl(short val, int srcLane, int width = warpSize)
+{
+    return (short) __shfl((int) val, srcLane, width);
+}
+
+__device__ __forceinline__ int shfl(int val, int srcLane, int width = warpSize)
+{
+    return __shfl(val, srcLane, width);
+}
+
+__device__ __forceinline__ uint shfl(uint val, int srcLane, int width = warpSize)
+{
+    return (uint) __shfl((int) val, srcLane, width);
+}
+
+__device__ __forceinline__ float shfl(float val, int srcLane, int width = warpSize)
+{
+    return __shfl(val, srcLane, width);
+}
+
+__device__ double shfl(double val, int srcLane, int width = warpSize)
+{
+    int lo = __double2loint(val);
+    int hi = __double2hiint(val);
+
+    lo = __shfl(lo, srcLane, width);
+    hi = __shfl(hi, srcLane, width);
+
+    return __hiloint2double(hi, lo);
+}
+
+#define CV_CUDEV_SHFL_VEC_INST(input_type) \
+    __device__ __forceinline__ input_type ## 1 shfl(const input_type ## 1 & val, int srcLane, int width = warpSize) \
+    { \
+        return VecTraits<input_type ## 1>::make( \
+                        shfl(val.x, srcLane, width) \
+                    ); \
+    } \
+    __device__ __forceinline__ input_type ## 2 shfl(const input_type ## 2 & val, int srcLane, int width = warpSize) \
+    { \
+        return VecTraits<input_type ## 2>::make( \
+                        shfl(val.x, srcLane, width), \
+                        shfl(val.y, srcLane, width) \
+                    ); \
+    } \
+    __device__ __forceinline__ input_type ## 3 shfl(const input_type ## 3 & val, int srcLane, int width = warpSize) \
+    { \
+        return VecTraits<input_type ## 3>::make( \
+                        shfl(val.x, srcLane, width), \
+                        shfl(val.y, srcLane, width), \
+                        shfl(val.z, srcLane, width) \
+                    ); \
+    } \
+    __device__ __forceinline__ input_type ## 4 shfl(const input_type ## 4 & val, int srcLane, int width = warpSize) \
+    { \
+        return VecTraits<input_type ## 4>::make( \
+                        shfl(val.x, srcLane, width), \
+                        shfl(val.y, srcLane, width), \
+                        shfl(val.z, srcLane, width), \
+                        shfl(val.w, srcLane, width) \
+                    ); \
+    }
+
+CV_CUDEV_SHFL_VEC_INST(uchar)
+CV_CUDEV_SHFL_VEC_INST(char)
+CV_CUDEV_SHFL_VEC_INST(ushort)
+CV_CUDEV_SHFL_VEC_INST(short)
+CV_CUDEV_SHFL_VEC_INST(uint)
+CV_CUDEV_SHFL_VEC_INST(int)
+CV_CUDEV_SHFL_VEC_INST(float)
+CV_CUDEV_SHFL_VEC_INST(double)
+
+#undef CV_CUDEV_SHFL_VEC_INST
+
+// shfl_up
+
+__device__ __forceinline__ uchar shfl_up(uchar val, uint delta, int width = warpSize)
+{
+    return (uchar) __shfl_up((int) val, delta, width);
+}
+
+__device__ __forceinline__ schar shfl_up(schar val, uint delta, int width = warpSize)
+{
+    return (schar) __shfl_up((int) val, delta, width);
+}
+
+__device__ __forceinline__ ushort shfl_up(ushort val, uint delta, int width = warpSize)
+{
+    return (ushort) __shfl_up((int) val, delta, width);
+}
+
+__device__ __forceinline__ short shfl_up(short val, uint delta, int width = warpSize)
+{
+    return (short) __shfl_up((int) val, delta, width);
+}
+
+__device__ __forceinline__ int shfl_up(int val, uint delta, int width = warpSize)
+{
+    return __shfl_up(val, delta, width);
+}
+
+__device__ __forceinline__ uint shfl_up(uint val, uint delta, int width = warpSize)
+{
+    return (uint) __shfl_up((int) val, delta, width);
+}
+
+__device__ __forceinline__ float shfl_up(float val, uint delta, int width = warpSize)
+{
+    return __shfl_up(val, delta, width);
+}
+
+__device__ double shfl_up(double val, uint delta, int width = warpSize)
+{
+    int lo = __double2loint(val);
+    int hi = __double2hiint(val);
+
+    lo = __shfl_up(lo, delta, width);
+    hi = __shfl_up(hi, delta, width);
+
+    return __hiloint2double(hi, lo);
+}
+
+#define CV_CUDEV_SHFL_UP_VEC_INST(input_type) \
+    __device__ __forceinline__ input_type ## 1 shfl_up(const input_type ## 1 & val, uint delta, int width = warpSize) \
+    { \
+        return VecTraits<input_type ## 1>::make( \
+                        shfl_up(val.x, delta, width) \
+                    ); \
+    } \
+    __device__ __forceinline__ input_type ## 2 shfl_up(const input_type ## 2 & val, uint delta, int width = warpSize) \
+    { \
+        return VecTraits<input_type ## 2>::make( \
+                        shfl_up(val.x, delta, width), \
+                        shfl_up(val.y, delta, width) \
+                    ); \
+    } \
+    __device__ __forceinline__ input_type ## 3 shfl_up(const input_type ## 3 & val, uint delta, int width = warpSize) \
+    { \
+        return VecTraits<input_type ## 3>::make( \
+                        shfl_up(val.x, delta, width), \
+                        shfl_up(val.y, delta, width), \
+                        shfl_up(val.z, delta, width) \
+                    ); \
+    } \
+    __device__ __forceinline__ input_type ## 4 shfl_up(const input_type ## 4 & val, uint delta, int width = warpSize) \
+    { \
+        return VecTraits<input_type ## 4>::make( \
+                        shfl_up(val.x, delta, width), \
+                        shfl_up(val.y, delta, width), \
+                        shfl_up(val.z, delta, width), \
+                        shfl_up(val.w, delta, width) \
+                    ); \
+    }
+
+CV_CUDEV_SHFL_UP_VEC_INST(uchar)
+CV_CUDEV_SHFL_UP_VEC_INST(char)
+CV_CUDEV_SHFL_UP_VEC_INST(ushort)
+CV_CUDEV_SHFL_UP_VEC_INST(short)
+CV_CUDEV_SHFL_UP_VEC_INST(uint)
+CV_CUDEV_SHFL_UP_VEC_INST(int)
+CV_CUDEV_SHFL_UP_VEC_INST(float)
+CV_CUDEV_SHFL_UP_VEC_INST(double)
+
+#undef CV_CUDEV_SHFL_UP_VEC_INST
+
+// shfl_down
+
+__device__ __forceinline__ uchar shfl_down(uchar val, uint delta, int width = warpSize)
+{
+    return (uchar) __shfl_down((int) val, delta, width);
+}
+
+__device__ __forceinline__ schar shfl_down(schar val, uint delta, int width = warpSize)
+{
+    return (schar) __shfl_down((int) val, delta, width);
+}
+
+__device__ __forceinline__ ushort shfl_down(ushort val, uint delta, int width = warpSize)
+{
+    return (ushort) __shfl_down((int) val, delta, width);
+}
+
+__device__ __forceinline__ short shfl_down(short val, uint delta, int width = warpSize)
+{
+    return (short) __shfl_down((int) val, delta, width);
+}
+
+__device__ __forceinline__ int shfl_down(int val, uint delta, int width = warpSize)
+{
+    return __shfl_down(val, delta, width);
+}
+
+__device__ __forceinline__ uint shfl_down(uint val, uint delta, int width = warpSize)
+{
+    return (uint) __shfl_down((int) val, delta, width);
+}
+
+__device__ __forceinline__ float shfl_down(float val, uint delta, int width = warpSize)
+{
+    return __shfl_down(val, delta, width);
+}
+
+__device__ double shfl_down(double val, uint delta, int width = warpSize)
+{
+    int lo = __double2loint(val);
+    int hi = __double2hiint(val);
+
+    lo = __shfl_down(lo, delta, width);
+    hi = __shfl_down(hi, delta, width);
+
+    return __hiloint2double(hi, lo);
+}
+
+#define CV_CUDEV_SHFL_DOWN_VEC_INST(input_type) \
+    __device__ __forceinline__ input_type ## 1 shfl_down(const input_type ## 1 & val, uint delta, int width = warpSize) \
+    { \
+        return VecTraits<input_type ## 1>::make( \
+                        shfl_down(val.x, delta, width) \
+                    ); \
+    } \
+    __device__ __forceinline__ input_type ## 2 shfl_down(const input_type ## 2 & val, uint delta, int width = warpSize) \
+    { \
+        return VecTraits<input_type ## 2>::make( \
+                        shfl_down(val.x, delta, width), \
+                        shfl_down(val.y, delta, width) \
+                    ); \
+    } \
+    __device__ __forceinline__ input_type ## 3 shfl_down(const input_type ## 3 & val, uint delta, int width = warpSize) \
+    { \
+        return VecTraits<input_type ## 3>::make( \
+                        shfl_down(val.x, delta, width), \
+                        shfl_down(val.y, delta, width), \
+                        shfl_down(val.z, delta, width) \
+                    ); \
+    } \
+    __device__ __forceinline__ input_type ## 4 shfl_down(const input_type ## 4 & val, uint delta, int width = warpSize) \
+    { \
+        return VecTraits<input_type ## 4>::make( \
+                        shfl_down(val.x, delta, width), \
+                        shfl_down(val.y, delta, width), \
+                        shfl_down(val.z, delta, width), \
+                        shfl_down(val.w, delta, width) \
+                    ); \
+    }
+
+CV_CUDEV_SHFL_DOWN_VEC_INST(uchar)
+CV_CUDEV_SHFL_DOWN_VEC_INST(char)
+CV_CUDEV_SHFL_DOWN_VEC_INST(ushort)
+CV_CUDEV_SHFL_DOWN_VEC_INST(short)
+CV_CUDEV_SHFL_DOWN_VEC_INST(uint)
+CV_CUDEV_SHFL_DOWN_VEC_INST(int)
+CV_CUDEV_SHFL_DOWN_VEC_INST(float)
+CV_CUDEV_SHFL_DOWN_VEC_INST(double)
+
+#undef CV_CUDEV_SHFL_DOWN_VEC_INST
+
+// shfl_xor
+
+__device__ __forceinline__ uchar shfl_xor(uchar val, int laneMask, int width = warpSize)
+{
+    return (uchar) __shfl_xor((int) val, laneMask, width);
+}
+
+__device__ __forceinline__ schar shfl_xor(schar val, int laneMask, int width = warpSize)
+{
+    return (schar) __shfl_xor((int) val, laneMask, width);
+}
+
+__device__ __forceinline__ ushort shfl_xor(ushort val, int laneMask, int width = warpSize)
+{
+    return (ushort) __shfl_xor((int) val, laneMask, width);
+}
+
+__device__ __forceinline__ short shfl_xor(short val, int laneMask, int width = warpSize)
+{
+    return (short) __shfl_xor((int) val, laneMask, width);
+}
+
+__device__ __forceinline__ int shfl_xor(int val, int laneMask, int width = warpSize)
+{
+    return __shfl_xor(val, laneMask, width);
+}
+
+__device__ __forceinline__ uint shfl_xor(uint val, int laneMask, int width = warpSize)
+{
+    return (uint) __shfl_xor((int) val, laneMask, width);
+}
+
+__device__ __forceinline__ float shfl_xor(float val, int laneMask, int width = warpSize)
+{
+    return __shfl_xor(val, laneMask, width);
+}
+
+__device__ double shfl_xor(double val, int laneMask, int width = warpSize)
+{
+    int lo = __double2loint(val);
+    int hi = __double2hiint(val);
+
+    lo = __shfl_xor(lo, laneMask, width);
+    hi = __shfl_xor(hi, laneMask, width);
+
+    return __hiloint2double(hi, lo);
+}
+
+#define CV_CUDEV_SHFL_XOR_VEC_INST(input_type) \
+    __device__ __forceinline__ input_type ## 1 shfl_xor(const input_type ## 1 & val, int laneMask, int width = warpSize) \
+    { \
+        return VecTraits<input_type ## 1>::make( \
+                        shfl_xor(val.x, laneMask, width) \
+                    ); \
+    } \
+    __device__ __forceinline__ input_type ## 2 shfl_xor(const input_type ## 2 & val, int laneMask, int width = warpSize) \
+    { \
+        return VecTraits<input_type ## 2>::make( \
+                        shfl_xor(val.x, laneMask, width), \
+                        shfl_xor(val.y, laneMask, width) \
+                    ); \
+    } \
+    __device__ __forceinline__ input_type ## 3 shfl_xor(const input_type ## 3 & val, int laneMask, int width = warpSize) \
+    { \
+        return VecTraits<input_type ## 3>::make( \
+                        shfl_xor(val.x, laneMask, width), \
+                        shfl_xor(val.y, laneMask, width), \
+                        shfl_xor(val.z, laneMask, width) \
+                    ); \
+    } \
+    __device__ __forceinline__ input_type ## 4 shfl_xor(const input_type ## 4 & val, int laneMask, int width = warpSize) \
+    { \
+        return VecTraits<input_type ## 4>::make( \
+                        shfl_xor(val.x, laneMask, width), \
+                        shfl_xor(val.y, laneMask, width), \
+                        shfl_xor(val.z, laneMask, width), \
+                        shfl_xor(val.w, laneMask, width) \
+                    ); \
+    }
+
+CV_CUDEV_SHFL_XOR_VEC_INST(uchar)
+CV_CUDEV_SHFL_XOR_VEC_INST(char)
+CV_CUDEV_SHFL_XOR_VEC_INST(ushort)
+CV_CUDEV_SHFL_XOR_VEC_INST(short)
+CV_CUDEV_SHFL_XOR_VEC_INST(uint)
+CV_CUDEV_SHFL_XOR_VEC_INST(int)
+CV_CUDEV_SHFL_XOR_VEC_INST(float)
+CV_CUDEV_SHFL_XOR_VEC_INST(double)
+
+#undef CV_CUDEV_SHFL_XOR_VEC_INST
+
+#endif // CV_CUDEV_ARCH >= 300
+
+}}
+
+#endif

modules/cudev/include/opencv2/cudev/warp/warp.hpp

@@ -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.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#pragma once
+
+#ifndef __OPENCV_CUDEV_WARP_WARP_HPP__
+#define __OPENCV_CUDEV_WARP_WARP_HPP__
+
+#include "../common.hpp"
+
+namespace cv { namespace cudev {
+
+enum
+{
+    LOG_WARP_SIZE = 5,
+    WARP_SIZE     = 1 << LOG_WARP_SIZE
+};
+
+struct Warp
+{
+    __device__ __forceinline__ static uint laneId()
+    {
+        uint ret;
+        asm("mov.u32 %0, %laneid;" : "=r"(ret));
+        return ret;
+    }
+
+    __device__ __forceinline__ static uint warpId()
+    {
+        const uint tid = (threadIdx.z * blockDim.y + threadIdx.y) * blockDim.x + threadIdx.x;
+        return tid / WARP_SIZE;
+    }
+};
+
+template <class It, typename T>
+__device__ __forceinline__ void warpFill(It beg, It end, const T& value)
+{
+    for(It t = beg + Warp::laneId(); t < end; t += WARP_SIZE)
+        *t = value;
+}
+
+template <class InIt, class OutIt>
+__device__ __forceinline__ OutIt warpCopy(InIt beg, InIt end, OutIt out)
+{
+    for(InIt t = beg + Warp::laneId(); t < end; t += WARP_SIZE, out += WARP_SIZE)
+        *out = *t;
+    return out;
+}
+
+template <class InIt, class OutIt, class UnOp>
+__device__ __forceinline__ OutIt warpTransform(InIt beg, InIt end, OutIt out, const UnOp& op)
+{
+    for(InIt t = beg + Warp::laneId(); t < end; t += WARP_SIZE, out += WARP_SIZE)
+        *out = op(*t);
+    return out;
+}
+
+template <class InIt1, class InIt2, class OutIt, class BinOp>
+__device__ __forceinline__ OutIt warpTransform(InIt1 beg1, InIt1 end1, InIt2 beg2, OutIt out, const BinOp& op)
+{
+    uint lane = Warp::laneId();
+
+    InIt1 t1 = beg1 + lane;
+    InIt2 t2 = beg2 + lane;
+    for(; t1 < end1; t1 += WARP_SIZE, t2 += WARP_SIZE, out += WARP_SIZE)
+        *out = op(*t1, *t2);
+    return out;
+}
+
+template<typename OutIt, typename T>
+__device__ __forceinline__ void warpYota(OutIt beg, OutIt end, T value)
+{
+    uint lane = Warp::laneId();
+    value += lane;
+
+    for(OutIt t = beg + lane; t < end; t += WARP_SIZE, value += WARP_SIZE)
+        *t = value;
+}
+
+}}
+
+#endif

modules/cudev/test/CMakeLists.txt

@@ -0,0 +1,33 @@
+set(test_deps opencv_core opencv_imgproc opencv_highgui opencv_ts ${OPENCV_MODULE_opencv_ts_DEPS})
+
+ocv_check_dependencies(${test_deps})
+
+if(OCV_DEPENDENCIES_FOUND)
+  set(the_target "opencv_test_${name}")
+
+  ocv_module_include_directories("${test_deps}" "${the_module}")
+
+  file(GLOB test_srcs "${CMAKE_CURRENT_SOURCE_DIR}/*.cpp" "${CMAKE_CURRENT_SOURCE_DIR}/*.cu")
+  file(GLOB test_hdrs "${CMAKE_CURRENT_SOURCE_DIR}/*.hpp")
+  source_group("Src" FILES ${test_srcs})
+  source_group("Include" FILES ${test_hdrs})
+  set(OPENCV_TEST_${the_module}_SOURCES ${test_srcs} ${test_hdrs})
+
+  CUDA_ADD_EXECUTABLE(${the_target} ${OPENCV_TEST_${the_module}_SOURCES})
+  target_link_libraries(${the_target} ${test_deps} ${OPENCV_LINKER_LIBS} ${CUDA_LIBRARIES})
+  add_dependencies(opencv_tests ${the_target})
+
+  # Additional target properties
+  set_target_properties(${the_target} PROPERTIES
+    DEBUG_POSTFIX "${OPENCV_DEBUG_POSTFIX}"
+    RUNTIME_OUTPUT_DIRECTORY "${EXECUTABLE_OUTPUT_PATH}"
+  )
+
+  if(ENABLE_SOLUTION_FOLDERS)
+    set_target_properties(${the_target} PROPERTIES FOLDER "tests accuracy")
+  endif()
+
+  enable_testing()
+  get_target_property(LOC ${the_target} LOCATION)
+  add_test(${the_target} "${LOC}")
+endif()

modules/cudev/test/test_arithm_func.cu

@@ -0,0 +1,168 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#include "test_precomp.hpp"
+
+using namespace cv;
+using namespace cv::gpu;
+using namespace cv::cudev;
+using namespace cvtest;
+
+////////////////////////////////////////////////////////////////////////////////
+// SqrtTest
+
+template <typename T>
+class SqrtTest : public ::testing::Test
+{
+public:
+    void test_gpumat()
+    {
+        const Size size = randomSize(100, 400);
+        const int type = DataType<T>::type;
+
+        Mat src = randomMat(size, type);
+
+        GpuMat_<T> d_src(src);
+
+        GpuMat_<T> dst = sqrt_(d_src);
+
+        Mat dst_gold;
+        cv::sqrt(src, dst_gold);
+
+        EXPECT_MAT_NEAR(dst_gold, dst, 0.0);
+    }
+
+    void test_expr()
+    {
+        const Size size = randomSize(100, 400);
+        const int type = DataType<T>::type;
+
+        Mat src1 = randomMat(size, type);
+        Mat src2 = randomMat(size, type);
+
+        GpuMat_<T> d_src1(src1), d_src2(src2);
+
+        GpuMat_<T> dst = sqrt_(d_src1 * d_src2);
+
+        Mat dst_gold;
+        cv::multiply(src1, src2, dst_gold);
+        cv::sqrt(dst_gold, dst_gold);
+
+        EXPECT_MAT_NEAR(dst_gold, dst, 0.0);
+    }
+};
+
+TYPED_TEST_CASE(SqrtTest, float);
+
+TYPED_TEST(SqrtTest, GpuMat)
+{
+    SqrtTest<TypeParam>::test_gpumat();
+}
+
+TYPED_TEST(SqrtTest, Expr)
+{
+    SqrtTest<TypeParam>::test_expr();
+}
+
+////////////////////////////////////////////////////////////////////////////////
+// MagnitudeTest
+
+template <typename T>
+class MagnitudeTest : public ::testing::Test
+{
+public:
+    void test_accuracy()
+    {
+        const Size size = randomSize(100, 400);
+        const int type = DataType<T>::type;
+
+        Mat src1 = randomMat(size, type);
+        Mat src2 = randomMat(size, type);
+
+        GpuMat_<T> d_src1(src1), d_src2(src2);
+
+        GpuMat_<T> dst1 = hypot_(d_src1, d_src2);
+        GpuMat_<T> dst2 = magnitude_(d_src1, d_src2);
+        GpuMat_<T> dst3 = sqrt_(sqr_(d_src1) + sqr_(d_src2));
+
+        EXPECT_MAT_NEAR(dst1, dst2, 1e-4);
+        EXPECT_MAT_NEAR(dst2, dst3, 0.0);
+    }
+};
+
+TYPED_TEST_CASE(MagnitudeTest, float);
+
+TYPED_TEST(MagnitudeTest, Accuracy)
+{
+    MagnitudeTest<TypeParam>::test_accuracy();
+}
+
+////////////////////////////////////////////////////////////////////////////////
+// PowTest
+
+template <typename T>
+class PowTest : public ::testing::Test
+{
+public:
+    void test_accuracy()
+    {
+        const Size size = randomSize(100, 400);
+        const int type = DataType<T>::type;
+
+        Mat src = randomMat(size, type);
+
+        GpuMat_<T> d_src(src);
+
+        GpuMat_<T> dst1 = pow_(d_src, 0.5);
+        GpuMat_<T> dst2 = sqrt_(d_src);
+
+        EXPECT_MAT_NEAR(dst1, dst2, 1e-5);
+    }
+};
+
+TYPED_TEST_CASE(PowTest, float);
+
+TYPED_TEST(PowTest, Accuracy)
+{
+    PowTest<TypeParam>::test_accuracy();
+}

modules/cudev/test/test_arithm_op.cu

@@ -0,0 +1,395 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#include "test_precomp.hpp"
+
+using namespace cv;
+using namespace cv::gpu;
+using namespace cv::cudev;
+using namespace cvtest;
+
+typedef ::testing::Types<uchar, ushort, short, int, float> AllTypes;
+typedef ::testing::Types<short, int, float> SignedTypes;
+
+////////////////////////////////////////////////////////////////////////////////
+// UnaryMinusTest
+
+template <typename T>
+class UnaryMinusTest : public ::testing::Test
+{
+public:
+    void test_gpumat()
+    {
+        const Size size = randomSize(100, 400);
+        const int type = DataType<T>::type;
+
+        Mat src = randomMat(size, type);
+
+        GpuMat_<T> d_src(src);
+
+        GpuMat_<T> dst = -d_src;
+
+        Mat dst_gold;
+        src.convertTo(dst_gold, src.depth(), -1);
+
+        EXPECT_MAT_NEAR(dst_gold, dst, 0.0);
+    }
+
+    void test_globptr()
+    {
+        const Size size = randomSize(100, 400);
+        const int type = DataType<T>::type;
+
+        Mat src = randomMat(size, type);
+
+        GpuMat_<T> d_src(src);
+        GlobPtrSz<T> d_src_ptr = d_src;
+
+        GpuMat_<T> dst = -d_src_ptr;
+
+        Mat dst_gold;
+        src.convertTo(dst_gold, src.depth(), -1);
+
+        EXPECT_MAT_NEAR(dst_gold, dst, 0.0);
+    }
+
+    void test_texptr()
+    {
+        const Size size = randomSize(100, 400);
+        const int type = DataType<T>::type;
+
+        Mat src = randomMat(size, type);
+
+        GpuMat_<T> d_src(src);
+        Texture<T> tex_src(d_src);
+
+        GpuMat_<T> dst = -tex_src;
+
+        Mat dst_gold;
+        src.convertTo(dst_gold, src.depth(), -1);
+
+        EXPECT_MAT_NEAR(dst_gold, dst, 0.0);
+    }
+
+    void test_expr()
+    {
+        const Size size = randomSize(100, 400);
+        const int type = DataType<T>::type;
+
+        Mat src1 = randomMat(size, type);
+        Mat src2 = randomMat(size, type);
+
+        GpuMat_<T> d_src1(src1), d_src2(src2);
+
+        GpuMat_<T> dst = -(d_src1 + d_src2);
+
+        Mat dst_gold;
+        cv::add(src1, src2, dst_gold);
+        dst_gold.convertTo(dst_gold, dst_gold.depth(), -1);
+
+        EXPECT_MAT_NEAR(dst_gold, dst, 0.0);
+    }
+};
+
+TYPED_TEST_CASE(UnaryMinusTest, SignedTypes);
+
+TYPED_TEST(UnaryMinusTest, GpuMat)
+{
+    UnaryMinusTest<TypeParam>::test_gpumat();
+}
+
+TYPED_TEST(UnaryMinusTest, GlobPtrSz)
+{
+    UnaryMinusTest<TypeParam>::test_globptr();
+}
+
+TYPED_TEST(UnaryMinusTest, TexturePtr)
+{
+    UnaryMinusTest<TypeParam>::test_texptr();
+}
+
+TYPED_TEST(UnaryMinusTest, Expr)
+{
+    UnaryMinusTest<TypeParam>::test_expr();
+}
+
+////////////////////////////////////////////////////////////////////////////////
+// PlusTest
+
+template <typename T>
+class PlusTest : public ::testing::Test
+{
+public:
+    void test_gpumat_gpumat()
+    {
+        const Size size = randomSize(100, 400);
+        const int type = DataType<T>::type;
+
+        Mat src1 = randomMat(size, type);
+        Mat src2 = randomMat(size, type);
+
+        GpuMat_<T> d_src1(src1), d_src2(src2);
+
+        GpuMat_<T> dst = d_src1 + d_src2;
+
+        Mat dst_gold;
+        cv::add(src1, src2, dst_gold);
+
+        EXPECT_MAT_NEAR(dst_gold, dst, 0.0);
+    }
+
+    void test_texptr_scalar()
+    {
+        const Size size = randomSize(100, 400);
+        const int type = DataType<T>::type;
+
+        Mat src = randomMat(size, type);
+
+        GpuMat_<T> d_src(src);
+        Texture<T> tex_src(d_src);
+
+        GpuMat_<T> dst = tex_src + static_cast<T>(5);
+
+        Mat dst_gold;
+        cv::add(src, 5, dst_gold);
+
+        EXPECT_MAT_NEAR(dst_gold, dst, 0.0);
+    }
+
+    void test_expr_gpumat()
+    {
+        const Size size = randomSize(100, 400);
+        const int type = DataType<T>::type;
+
+        Mat src1 = randomMat(size, type);
+        Mat src2 = randomMat(size, type);
+        Mat src3 = randomMat(size, type);
+
+        GpuMat_<T> d_src1(src1), d_src2(src2), d_src3(src3);
+
+        GpuMat_<T> dst = d_src1 + d_src2 + d_src3;
+
+        Mat dst_gold;
+        cv::add(src1, src2, dst_gold);
+        cv::add(dst_gold, src3, dst_gold);
+
+        EXPECT_MAT_NEAR(dst_gold, dst, 0.0);
+    }
+
+    void test_scalar_expr()
+    {
+        const Size size = randomSize(100, 400);
+        const int type = DataType<T>::type;
+
+        Mat src1 = randomMat(size, type);
+        Mat src2 = randomMat(size, type);
+
+        GpuMat_<T> d_src1(src1), d_src2(src2);
+
+        GpuMat_<T> dst = static_cast<T>(5) + (d_src1 + d_src2);
+
+        Mat dst_gold;
+        cv::add(src1, src2, dst_gold);
+        cv::add(dst_gold, 5, dst_gold);
+
+        EXPECT_MAT_NEAR(dst_gold, dst, 0.0);
+    }
+};
+
+TYPED_TEST_CASE(PlusTest, AllTypes);
+
+TYPED_TEST(PlusTest, GpuMat_GpuMat)
+{
+    PlusTest<TypeParam>::test_gpumat_gpumat();
+}
+
+TYPED_TEST(PlusTest, TexturePtr_Scalar)
+{
+    PlusTest<TypeParam>::test_texptr_scalar();
+}
+
+TYPED_TEST(PlusTest, Expr_GpuMat)
+{
+    PlusTest<TypeParam>::test_expr_gpumat();
+}
+
+TYPED_TEST(PlusTest, Scalar_Expr)
+{
+    PlusTest<TypeParam>::test_scalar_expr();
+}
+
+////////////////////////////////////////////////////////////////////////////////
+// MinusTest
+
+template <typename T>
+class MinusTest : public ::testing::Test
+{
+public:
+    void test_gpumat_gpumat()
+    {
+        const Size size = randomSize(100, 400);
+        const int type = DataType<T>::type;
+
+        Mat src1 = randomMat(size, type);
+        Mat src2 = randomMat(size, type);
+
+        GpuMat_<T> d_src1(src1), d_src2(src2);
+
+        GpuMat_<T> dst = d_src1 - d_src2;
+
+        Mat dst_gold;
+        cv::subtract(src1, src2, dst_gold);
+
+        EXPECT_MAT_NEAR(dst_gold, dst, 0.0);
+    }
+
+    void test_texptr_scalar()
+    {
+        const Size size = randomSize(100, 400);
+        const int type = DataType<T>::type;
+
+        Mat src = randomMat(size, type);
+
+        GpuMat_<T> d_src(src);
+        Texture<T> tex_src(d_src);
+
+        GpuMat_<T> dst = tex_src - static_cast<T>(5);
+
+        Mat dst_gold;
+        cv::subtract(src, 5, dst_gold);
+
+        EXPECT_MAT_NEAR(dst_gold, dst, 0.0);
+    }
+
+    void test_expr_gpumat()
+    {
+        const Size size = randomSize(100, 400);
+        const int type = DataType<T>::type;
+
+        Mat src1 = randomMat(size, type);
+        Mat src2 = randomMat(size, type);
+        Mat src3 = randomMat(size, type);
+
+        GpuMat_<T> d_src1(src1), d_src2(src2), d_src3(src3);
+
+        GpuMat_<T> dst = (d_src1 + d_src2) - d_src3;
+
+        Mat dst_gold;
+        cv::add(src1, src2, dst_gold);
+        cv::subtract(dst_gold, src3, dst_gold);
+
+        EXPECT_MAT_NEAR(dst_gold, dst, 0.0);
+    }
+
+    void test_scalar_expr()
+    {
+        const Size size = randomSize(100, 400);
+        const int type = DataType<T>::type;
+
+        Mat src1 = randomMat(size, type);
+        Mat src2 = randomMat(size, type);
+
+        GpuMat_<T> d_src1(src1), d_src2(src2);
+
+        GpuMat_<T> dst = static_cast<T>(5) - (d_src1 + d_src2);
+
+        Mat dst_gold;
+        cv::add(src1, src2, dst_gold);
+        cv::subtract(5, dst_gold, dst_gold);
+
+        EXPECT_MAT_NEAR(dst_gold, dst, 0.0);
+    }
+};
+
+TYPED_TEST_CASE(MinusTest, SignedTypes);
+
+TYPED_TEST(MinusTest, GpuMat_GpuMat)
+{
+    MinusTest<TypeParam>::test_gpumat_gpumat();
+}
+
+TYPED_TEST(MinusTest, TexturePtr_Scalar)
+{
+    MinusTest<TypeParam>::test_texptr_scalar();
+}
+
+TYPED_TEST(MinusTest, Expr_GpuMat)
+{
+    MinusTest<TypeParam>::test_expr_gpumat();
+}
+
+TYPED_TEST(MinusTest, Scalar_Expr)
+{
+    MinusTest<TypeParam>::test_scalar_expr();
+}
+
+////////////////////////////////////////////////////////////////////////////////
+// AbsDiffTest
+
+template <typename T>
+class AbsDiffTest : public ::testing::Test
+{
+public:
+    void test_accuracy()
+    {
+        const Size size = randomSize(100, 400);
+        const int type = DataType<T>::type;
+
+        Mat src1 = randomMat(size, type);
+        Mat src2 = randomMat(size, type);
+
+        GpuMat_<T> d_src1(src1), d_src2(src2);
+
+        GpuMat_<T> dst1 = absdiff_(d_src1, d_src2);
+        GpuMat_<T> dst2 = abs_(d_src1 - d_src2);
+
+        EXPECT_MAT_NEAR(dst1, dst2, 0.0);
+    }
+};
+
+TYPED_TEST_CASE(AbsDiffTest, SignedTypes);
+
+TYPED_TEST(AbsDiffTest, Accuracy)
+{
+    AbsDiffTest<TypeParam>::test_accuracy();
+}

modules/cudev/test/test_bitwize_op.cu

@@ -0,0 +1,146 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#include "test_precomp.hpp"
+
+using namespace cv;
+using namespace cv::gpu;
+using namespace cv::cudev;
+using namespace cvtest;
+
+typedef ::testing::Types<uchar, ushort, short, int> IntTypes;
+
+////////////////////////////////////////////////////////////////////////////////
+// BitNotTest
+
+template <typename T>
+class BitNotTest : public ::testing::Test
+{
+public:
+    void test_gpumat()
+    {
+        const Size size = randomSize(100, 400);
+        const int type = DataType<T>::type;
+
+        Mat src = randomMat(size, type);
+
+        GpuMat_<T> d_src(src);
+
+        GpuMat_<T> dst = ~d_src;
+
+        Mat dst_gold;
+        cv::bitwise_not(src, dst_gold);
+
+        EXPECT_MAT_NEAR(dst_gold, dst, 0.0);
+    }
+};
+
+TYPED_TEST_CASE(BitNotTest, IntTypes);
+
+TYPED_TEST(BitNotTest, GpuMat)
+{
+    BitNotTest<TypeParam>::test_gpumat();
+}
+
+////////////////////////////////////////////////////////////////////////////////
+// BitAndTest
+
+template <typename T>
+class BitAndTest : public ::testing::Test
+{
+public:
+    void test_gpumat_gpumat()
+    {
+        const Size size = randomSize(100, 400);
+        const int type = DataType<T>::type;
+
+        Mat src1 = randomMat(size, type);
+        Mat src2 = randomMat(size, type);
+
+        GpuMat_<T> d_src1(src1), d_src2(src2);
+
+        GpuMat_<T> dst = d_src1 & d_src2;
+
+        Mat dst_gold;
+        cv::bitwise_and(src1, src2, dst_gold);
+
+        EXPECT_MAT_NEAR(dst_gold, dst, 0.0);
+    }
+};
+
+TYPED_TEST_CASE(BitAndTest, IntTypes);
+
+TYPED_TEST(BitAndTest, GpuMat_GpuMat)
+{
+    BitAndTest<TypeParam>::test_gpumat_gpumat();
+}
+
+////////////////////////////////////////////////////////////////////////////////
+// LShiftTest
+
+template <typename T>
+class LShiftTest : public ::testing::Test
+{
+public:
+    void test_accuracy()
+    {
+        const Size size = randomSize(100, 400);
+        const int type = DataType<T>::type;
+
+        Mat src = randomMat(size, type);
+
+        GpuMat_<T> d_src(src);
+
+        GpuMat_<T> dst1 = d_src << 2;
+        GpuMat_<T> dst2 = d_src * 4;
+
+        EXPECT_MAT_NEAR(dst1, dst2, 0.0);
+    }
+};
+
+TYPED_TEST_CASE(LShiftTest, int);
+
+TYPED_TEST(LShiftTest, Accuracy)
+{
+    LShiftTest<TypeParam>::test_accuracy();
+}

modules/cudev/test/test_cmp_op.cu

@@ -0,0 +1,151 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#include "test_precomp.hpp"
+
+using namespace cv;
+using namespace cv::gpu;
+using namespace cv::cudev;
+using namespace cvtest;
+
+typedef ::testing::Types<uchar, ushort, short, int, float> AllTypes;
+
+////////////////////////////////////////////////////////////////////////////////
+// LessTest
+
+template <typename T>
+class LessTest : public ::testing::Test
+{
+public:
+    void test_gpumat_gpumat()
+    {
+        const Size size = randomSize(100, 400);
+        const int type = DataType<T>::type;
+
+        Mat src1 = randomMat(size, type);
+        Mat src2 = randomMat(size, type);
+
+        GpuMat_<T> d_src1(src1), d_src2(src2);
+
+        GpuMat_<uchar> dst = (d_src1 < d_src2) * 255;
+
+        Mat dst_gold;
+        cv::compare(src1, src2, dst_gold, CMP_LT);
+
+        EXPECT_MAT_NEAR(dst_gold, dst, 0.0);
+    }
+};
+
+TYPED_TEST_CASE(LessTest, AllTypes);
+
+TYPED_TEST(LessTest, GpuMat_GpuMat)
+{
+    LessTest<TypeParam>::test_gpumat_gpumat();
+}
+
+////////////////////////////////////////////////////////////////////////////////
+// MinTest
+
+template <typename T>
+class MinTest : public ::testing::Test
+{
+public:
+    void test_gpumat_gpumat()
+    {
+        const Size size = randomSize(100, 400);
+        const int type = DataType<T>::type;
+
+        Mat src1 = randomMat(size, type);
+        Mat src2 = randomMat(size, type);
+
+        GpuMat_<T> d_src1(src1), d_src2(src2);
+
+        GpuMat_<T> dst = min_(d_src1, d_src2);
+
+        Mat dst_gold;
+        cv::min(src1, src2, dst_gold);
+
+        EXPECT_MAT_NEAR(dst_gold, dst, 0.0);
+    }
+};
+
+TYPED_TEST_CASE(MinTest, AllTypes);
+
+TYPED_TEST(MinTest, GpuMat_GpuMat)
+{
+    MinTest<TypeParam>::test_gpumat_gpumat();
+}
+
+////////////////////////////////////////////////////////////////////////////////
+// ThreshBinaryTest
+
+typedef ::testing::Types<uchar, short, float> ThreshTypes;
+
+template <typename T>
+class ThreshBinaryTest : public ::testing::Test
+{
+public:
+    void test_gpumat()
+    {
+        const Size size = randomSize(100, 400);
+        const int type = DataType<T>::type;
+
+        Mat src = randomMat(size, type);
+
+        GpuMat_<T> d_src(src);
+
+        GpuMat_<T> dst = threshBinary_(d_src, 128, 0);
+
+        Mat dst_gold;
+        cv::threshold(src, dst_gold, 128, 0, THRESH_BINARY);
+
+        EXPECT_MAT_NEAR(dst_gold, dst, 0.0);
+    }
+};
+
+TYPED_TEST_CASE(ThreshBinaryTest, ThreshTypes);
+
+TYPED_TEST(ThreshBinaryTest, GpuMat)
+{
+    ThreshBinaryTest<TypeParam>::test_gpumat();
+}

modules/cudev/test/test_color_cvt.cu

@@ -0,0 +1,180 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#include "test_precomp.hpp"
+
+using namespace cv;
+using namespace cv::gpu;
+using namespace cv::cudev;
+using namespace cvtest;
+
+namespace cv {
+
+enum {
+    COLOR_BGR2BGR = COLOR_BGR2RGB,
+    COLOR_BGR2LRGB = COLOR_BGR2RGB,
+    COLOR_BGR2LBGR = COLOR_BGR2RGB
+};
+
+}
+
+#define CVT_COLOR_TEST(src_space, dst_space, src_cn, dst_cn) \
+    TEST(CvtColor, src_space ## _to_ ## dst_space) \
+    { \
+        const Size size = randomSize(100, 400); \
+        Mat bgrb = randomMat(size, CV_8UC3); \
+        Mat srcb; \
+        cv::cvtColor(bgrb, srcb, COLOR_BGR ## 2 ## src_space, src_cn); \
+        GpuMat_<SelectIf<src_cn == 1, uchar, uchar ## src_cn>::type> d_srcb(srcb); \
+        GpuMat_<SelectIf<dst_cn == 1, uchar, uchar ## dst_cn>::type> dstb = src_space ## _to_ ## dst_space ## _(d_srcb); \
+        Mat dstb_gold; \
+        cv::cvtColor(srcb, dstb_gold, COLOR_ ## src_space ## 2 ## dst_space); \
+        EXPECT_MAT_NEAR(dstb_gold, dstb, 1.0); \
+        Mat bgrf = randomMat(size, CV_32FC3, 0, 1); \
+        Mat srcf; \
+        cv::cvtColor(bgrf, srcf, COLOR_BGR ## 2 ## src_space, src_cn); \
+        GpuMat_<SelectIf<src_cn == 1, float, float ## src_cn>::type> d_srcf(srcf); \
+        GpuMat_<SelectIf<dst_cn == 1, float, float ## dst_cn>::type> dstf = src_space ## _to_ ## dst_space ## _(d_srcf); \
+        Mat dstf_gold; \
+        cv::cvtColor(srcf, dstf_gold, COLOR_ ## src_space ## 2 ## dst_space); \
+        EXPECT_MAT_NEAR(dstf_gold, dstf, 1.0); \
+    }
+
+// RGB <-> BGR
+
+CVT_COLOR_TEST(BGR, RGB, 3, 3)
+CVT_COLOR_TEST(BGR, BGRA, 3, 4)
+CVT_COLOR_TEST(BGR, RGBA, 3, 4)
+CVT_COLOR_TEST(BGRA, BGR, 4, 3)
+CVT_COLOR_TEST(BGRA, RGB, 4, 3)
+CVT_COLOR_TEST(BGRA, RGBA, 4, 4)
+
+// RGB <-> Gray
+
+CVT_COLOR_TEST(BGR, GRAY, 3, 1)
+CVT_COLOR_TEST(RGB, GRAY, 3, 1)
+CVT_COLOR_TEST(BGRA, GRAY, 4, 1)
+CVT_COLOR_TEST(RGBA, GRAY, 4, 1)
+
+CVT_COLOR_TEST(GRAY, BGR, 1, 3)
+CVT_COLOR_TEST(GRAY, BGRA, 1, 4)
+
+// RGB <-> YUV
+
+CVT_COLOR_TEST(RGB, YUV, 3, 3)
+CVT_COLOR_TEST(BGR, YUV, 3, 3)
+
+CVT_COLOR_TEST(YUV, RGB, 3, 3)
+CVT_COLOR_TEST(YUV, BGR, 3, 3)
+
+// RGB <-> YCrCb
+
+CVT_COLOR_TEST(RGB, YCrCb, 3, 3)
+CVT_COLOR_TEST(BGR, YCrCb, 3, 3)
+
+CVT_COLOR_TEST(YCrCb, RGB, 3, 3)
+CVT_COLOR_TEST(YCrCb, BGR, 3, 3)
+
+// RGB <-> XYZ
+
+CVT_COLOR_TEST(RGB, XYZ, 3, 3)
+CVT_COLOR_TEST(BGR, XYZ, 3, 3)
+
+CVT_COLOR_TEST(XYZ, RGB, 3, 3)
+CVT_COLOR_TEST(XYZ, BGR, 3, 3)
+
+// RGB <-> HSV
+
+CVT_COLOR_TEST(RGB, HSV, 3, 3)
+CVT_COLOR_TEST(BGR, HSV, 3, 3)
+
+CVT_COLOR_TEST(HSV, RGB, 3, 3)
+CVT_COLOR_TEST(HSV, BGR, 3, 3)
+
+CVT_COLOR_TEST(RGB, HSV_FULL, 3, 3)
+CVT_COLOR_TEST(BGR, HSV_FULL, 3, 3)
+
+CVT_COLOR_TEST(HSV, RGB_FULL, 3, 3)
+CVT_COLOR_TEST(HSV, BGR_FULL, 3, 3)
+
+// RGB <-> HLS
+
+CVT_COLOR_TEST(RGB, HLS, 3, 3)
+CVT_COLOR_TEST(BGR, HLS, 3, 3)
+
+CVT_COLOR_TEST(HLS, RGB, 3, 3)
+CVT_COLOR_TEST(HLS, BGR, 3, 3)
+
+CVT_COLOR_TEST(RGB, HLS_FULL, 3, 3)
+CVT_COLOR_TEST(BGR, HLS_FULL, 3, 3)
+
+CVT_COLOR_TEST(HLS, RGB_FULL, 3, 3)
+CVT_COLOR_TEST(HLS, BGR_FULL, 3, 3)
+
+// RGB <-> Lab
+
+CVT_COLOR_TEST(RGB, Lab, 3, 3)
+CVT_COLOR_TEST(BGR, Lab, 3, 3)
+
+CVT_COLOR_TEST(Lab, RGB, 3, 3)
+CVT_COLOR_TEST(Lab, BGR, 3, 3)
+
+CVT_COLOR_TEST(LRGB, Lab, 3, 3)
+CVT_COLOR_TEST(LBGR, Lab, 3, 3)
+
+CVT_COLOR_TEST(Lab, LRGB, 3, 3)
+CVT_COLOR_TEST(Lab, LBGR, 3, 3)
+
+// RGB <-> Luv
+
+CVT_COLOR_TEST(RGB, Luv, 3, 3)
+CVT_COLOR_TEST(BGR, Luv, 3, 3)
+
+CVT_COLOR_TEST(Luv, RGB, 3, 3)
+CVT_COLOR_TEST(Luv, BGR, 3, 3)
+
+CVT_COLOR_TEST(LRGB, Luv, 3, 3)
+CVT_COLOR_TEST(LBGR, Luv, 3, 3)
+
+CVT_COLOR_TEST(Luv, LRGB, 3, 3)
+CVT_COLOR_TEST(Luv, LBGR, 3, 3)

modules/cudev/test/test_cvt.cu

@@ -0,0 +1,83 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#include "test_precomp.hpp"
+
+using namespace cv;
+using namespace cv::gpu;
+using namespace cv::cudev;
+using namespace cvtest;
+
+typedef ::testing::Types<uchar, ushort, short, int, float> AllTypes;
+
+////////////////////////////////////////////////////////////////////////////////
+// CvtTest
+
+template <typename T>
+class CvtTest : public ::testing::Test
+{
+public:
+    void test_gpumat()
+    {
+        const Size size = randomSize(100, 400);
+        const int type = DataType<T>::type;
+
+        Mat src = randomMat(size, type);
+
+        GpuMat_<T> d_src(src);
+
+        GpuMat_<T> dst = cvt_<T>(cvt_<float>(d_src) * 2.0f - 10.0f);
+
+        Mat dst_gold;
+        src.convertTo(dst_gold, src.depth(), 2, -10);
+
+        EXPECT_MAT_NEAR(dst_gold, dst, 0.0);
+    }
+};
+
+TYPED_TEST_CASE(CvtTest, AllTypes);
+
+TYPED_TEST(CvtTest, GpuMat)
+{
+    CvtTest<TypeParam>::test_gpumat();
+}

modules/cudev/test/test_deriv.cu

@@ -0,0 +1,109 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#include "test_precomp.hpp"
+
+using namespace cv;
+using namespace cv::gpu;
+using namespace cv::cudev;
+using namespace cvtest;
+
+TEST(Sobel, Accuracy)
+{
+    const Size size = randomSize(100, 400);
+
+    Mat src = randomMat(size, CV_8UC1);
+
+    GpuMat_<uchar> d_src(src);
+    Texture<uchar> tex_src(d_src);
+
+    GpuMat_<short> dx = sobelX_(cvt_<int>(tex_src));
+    GpuMat_<short> dy = sobelY_(cvt_<int>(tex_src));
+
+    Mat dx_gold, dy_gold;
+    cv::Sobel(src, dx_gold, CV_16S, 1, 0, 3, 1, 0, BORDER_REPLICATE);
+    cv::Sobel(src, dy_gold, CV_16S, 0, 1, 3, 1, 0, BORDER_REPLICATE);
+
+    EXPECT_MAT_NEAR(dx_gold, dx, 0.0);
+    EXPECT_MAT_NEAR(dy_gold, dy, 0.0);
+}
+
+TEST(Scharr, Accuracy)
+{
+    const Size size = randomSize(100, 400);
+
+    Mat src = randomMat(size, CV_8UC1);
+
+    GpuMat_<uchar> d_src(src);
+    Texture<uchar> tex_src(d_src);
+
+    GpuMat_<short> dx = scharrX_(cvt_<int>(tex_src));
+    GpuMat_<short> dy = scharrY_(cvt_<int>(tex_src));
+
+    Mat dx_gold, dy_gold;
+    cv::Scharr(src, dx_gold, CV_16S, 1, 0, 1, 0, BORDER_REPLICATE);
+    cv::Scharr(src, dy_gold, CV_16S, 0, 1, 1, 0, BORDER_REPLICATE);
+
+    EXPECT_MAT_NEAR(dx_gold, dx, 0.0);
+    EXPECT_MAT_NEAR(dy_gold, dy, 0.0);
+}
+
+TEST(Laplacian, Accuracy)
+{
+    const Size size = randomSize(100, 400);
+
+    Mat src = randomMat(size, CV_8UC1);
+
+    GpuMat_<uchar> d_src(src);
+    Texture<uchar> tex_src(d_src);
+
+    GpuMat_<short> dst1 = laplacian_<1>(cvt_<int>(tex_src));
+    GpuMat_<short> dst3 = laplacian_<3>(cvt_<int>(tex_src));
+
+    Mat dst1_gold, dst3_gold;
+    cv::Laplacian(src, dst1_gold, CV_16S, 1, 1, 0, BORDER_REPLICATE);
+    cv::Laplacian(src, dst3_gold, CV_16S, 3, 1, 0, BORDER_REPLICATE);
+
+    EXPECT_MAT_NEAR(dst1_gold, dst1, 0.0);
+    EXPECT_MAT_NEAR(dst3_gold, dst3, 0.0);
+}

modules/cudev/test/test_integral.cu

@@ -0,0 +1,103 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#include "test_precomp.hpp"
+
+using namespace cv;
+using namespace cv::gpu;
+using namespace cv::cudev;
+using namespace cvtest;
+
+TEST(Integral, _8u)
+{
+    const Size size = randomSize(100, 400);
+
+    Mat src = randomMat(size, CV_8UC1);
+
+    GpuMat_<uchar> d_src(src);
+
+    GpuMat_<uint> dst = integral_(d_src);
+
+    Mat dst_gold;
+    cv::integral(src, dst_gold);
+
+    dst_gold = dst_gold(Rect(1, 1, size.width, size.height));
+
+    ASSERT_MAT_NEAR(dst_gold, dst, 0.0);
+}
+
+TEST(Integral, _32f)
+{
+    const Size size = randomSize(100, 400);
+
+    Mat src = randomMat(size, CV_32FC1, 0, 1);
+
+    GpuMat_<float> d_src(src);
+
+    GpuMat_<float> dst = integral_(d_src);
+
+    Mat dst_gold;
+    cv::integral(src, dst_gold, CV_32F);
+
+    dst_gold = dst_gold(Rect(1, 1, size.width, size.height));
+
+    ASSERT_PRED_FORMAT2(cvtest::MatComparator(1e-5, 0), dst_gold, Mat(dst));
+}
+
+TEST(Integral, _8u_opt)
+{
+    const Size size(640, 480);
+
+    Mat src = randomMat(size, CV_8UC1);
+
+    GpuMat_<uchar> d_src(src);
+
+    GpuMat_<uint> dst = integral_(d_src);
+
+    Mat dst_gold;
+    cv::integral(src, dst_gold);
+
+    dst_gold = dst_gold(Rect(1, 1, size.width, size.height));
+
+    ASSERT_MAT_NEAR(dst_gold, dst, 0.0);
+}

modules/cudev/test/test_lut.cu

@@ -0,0 +1,82 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#include "test_precomp.hpp"
+
+using namespace cv;
+using namespace cv::gpu;
+using namespace cv::cudev;
+using namespace cvtest;
+
+////////////////////////////////////////////////////////////////////////////////
+// LutTest
+
+template <typename T>
+class LutTest : public ::testing::Test
+{
+public:
+    void test_gpumat()
+    {
+        const Size size = randomSize(100, 400);
+        const int type = DataType<T>::type;
+
+        Mat src = randomMat(size, type);
+        Mat tbl = randomMat(Size(256, 1), type);
+
+        GpuMat_<T> d_src(src), d_tbl(tbl);
+
+        GpuMat_<T> dst = lut_(d_src, d_tbl);
+
+        Mat dst_gold;
+        cv::LUT(src, tbl, dst_gold);
+
+        EXPECT_MAT_NEAR(dst_gold, dst, 0.0);
+    }
+};
+
+TYPED_TEST_CASE(LutTest, uchar);
+
+TYPED_TEST(LutTest, GpuMat)
+{
+    LutTest<TypeParam>::test_gpumat();
+}

modules/cudev/test/test_main.cpp

@@ -0,0 +1,46 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#include "opencv2/ts.hpp"
+
+CV_TEST_MAIN("cv")

modules/cudev/test/test_precomp.hpp

@@ -0,0 +1,55 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#ifndef __OPENCV_TEST_PRECOMP_HPP__
+#define __OPENCV_TEST_PRECOMP_HPP__
+
+#include "opencv2/core.hpp"
+#include "opencv2/imgproc.hpp"
+#include "opencv2/highgui.hpp"
+#include "opencv2/cudev.hpp"
+
+#include "opencv2/ts.hpp"
+#include "opencv2/ts/gpu_test.hpp"
+
+#endif

modules/cudev/test/test_pyramids.cu

@@ -0,0 +1,81 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#include "test_precomp.hpp"
+
+using namespace cv;
+using namespace cv::gpu;
+using namespace cv::cudev;
+using namespace cvtest;
+
+TEST(PyrDown, _8uc1)
+{
+    const Size size = randomSize(100, 400);
+
+    Mat src = randomMat(size, CV_8UC1);
+
+    GpuMat_<uchar> d_src(src);
+
+    GpuMat_<uchar> dst = pyrDown_(d_src);
+
+    Mat dst_gold;
+    cv::pyrDown(src, dst_gold);
+
+    ASSERT_MAT_NEAR(dst_gold, dst, 1.0);
+}
+
+TEST(PyrUp, _32fc4)
+{
+    const Size size = randomSize(100, 400);
+
+    Mat src = randomMat(size, CV_32FC4);
+
+    GpuMat_<float4> d_src(src);
+
+    GpuMat_<float4> dst = pyrDown_(d_src);
+
+    Mat dst_gold;
+    cv::pyrDown(src, dst_gold);
+
+    ASSERT_MAT_NEAR(dst_gold, dst, 1e-4);
+}

modules/cudev/test/test_reduction.cu

@@ -0,0 +1,312 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#include "test_precomp.hpp"
+
+using namespace cv;
+using namespace cv::gpu;
+using namespace cv::cudev;
+using namespace cvtest;
+
+TEST(Sum, GpuMat)
+{
+    const Size size = randomSize(100, 400);
+
+    Mat src = randomMat(size, CV_8UC1);
+
+    GpuMat_<uchar> d_src(src);
+
+    GpuMat_<float> dst = sum_(d_src);
+    float res;
+    dst.download(_OutputArray(&res, 1));
+
+    Scalar dst_gold = cv::sum(src);
+
+    ASSERT_FLOAT_EQ(static_cast<float>(dst_gold[0]), res);
+}
+
+TEST(Sum, Expr)
+{
+    const Size size = randomSize(100, 400);
+
+    Mat src1 = randomMat(size, CV_32FC1, 0, 1);
+    Mat src2 = randomMat(size, CV_32FC1, 0, 1);
+
+    GpuMat_<float> d_src1(src1), d_src2(src2);
+
+    GpuMat_<float> dst = sum_(abs_(d_src1 - d_src2));
+    float res;
+    dst.download(_OutputArray(&res, 1));
+
+    Scalar dst_gold = cv::norm(src1, src2, NORM_L1);
+
+    ASSERT_FLOAT_EQ(static_cast<float>(dst_gold[0]), res);
+}
+
+TEST(MinVal, GpuMat)
+{
+    const Size size = randomSize(100, 400);
+
+    Mat src = randomMat(size, CV_8UC1);
+
+    GpuMat_<uchar> d_src(src);
+
+    GpuMat_<float> dst = minVal_(d_src);
+    float res;
+    dst.download(_OutputArray(&res, 1));
+
+    double res_gold;
+    cv::minMaxLoc(src, &res_gold, 0);
+
+    ASSERT_FLOAT_EQ(static_cast<float>(res_gold), res);
+}
+
+TEST(MaxVal, Expr)
+{
+    const Size size = randomSize(100, 400);
+
+    Mat src1 = randomMat(size, CV_32SC1);
+    Mat src2 = randomMat(size, CV_32SC1);
+
+    GpuMat_<int> d_src1(src1), d_src2(src2);
+
+    GpuMat_<float> dst = maxVal_(abs_(d_src1 - d_src2));
+    float res;
+    dst.download(_OutputArray(&res, 1));
+
+    double res_gold = cv::norm(src1, src2, NORM_INF);
+
+    ASSERT_FLOAT_EQ(static_cast<float>(res_gold), res);
+}
+
+TEST(MinMaxVal, GpuMat)
+{
+    const Size size = randomSize(100, 400);
+
+    Mat src = randomMat(size, CV_8UC1);
+
+    GpuMat_<uchar> d_src(src);
+
+    GpuMat_<float> dst = minMaxVal_(d_src);
+    float res[2];
+    dst.download(Mat(1, 2, CV_32FC1, res));
+
+    double res_gold[2];
+    cv::minMaxLoc(src, &res_gold[0], &res_gold[1]);
+
+    ASSERT_FLOAT_EQ(static_cast<float>(res_gold[0]), res[0]);
+    ASSERT_FLOAT_EQ(static_cast<float>(res_gold[1]), res[1]);
+}
+
+TEST(NonZeroCount, Accuracy)
+{
+    const Size size = randomSize(100, 400);
+
+    Mat src = randomMat(size, CV_8UC1, 0, 5);
+
+    GpuMat_<uchar> d_src(src);
+
+    GpuMat_<int> dst1 = countNonZero_(d_src);
+    GpuMat_<int> dst2 = sum_(cvt_<int>(d_src) != 0);
+
+    EXPECT_MAT_NEAR(dst1, dst2, 0.0);
+}
+
+TEST(ReduceToRow, Sum)
+{
+    const Size size = randomSize(100, 400);
+
+    Mat src = randomMat(size, CV_8UC1);
+
+    GpuMat_<uchar> d_src(src);
+
+    GpuMat_<int> dst = reduceToRow_<Sum<int> >(d_src);
+
+    Mat dst_gold;
+    cv::reduce(src, dst_gold, 0, REDUCE_SUM, CV_32S);
+
+    EXPECT_MAT_NEAR(dst_gold, dst, 0.0);
+}
+
+TEST(ReduceToRow, Avg)
+{
+    const Size size = randomSize(100, 400);
+
+    Mat src = randomMat(size, CV_8UC1);
+
+    GpuMat_<uchar> d_src(src);
+
+    GpuMat_<float> dst = reduceToRow_<Avg<float> >(d_src);
+
+    Mat dst_gold;
+    cv::reduce(src, dst_gold, 0, REDUCE_AVG, CV_32F);
+
+    EXPECT_MAT_NEAR(dst_gold, dst, 1e-4);
+}
+
+TEST(ReduceToRow, Min)
+{
+    const Size size = randomSize(100, 400);
+
+    Mat src = randomMat(size, CV_8UC1);
+
+    GpuMat_<uchar> d_src(src);
+
+    GpuMat_<uchar> dst = reduceToRow_<Min<uchar> >(d_src);
+
+    Mat dst_gold;
+    cv::reduce(src, dst_gold, 0, REDUCE_MIN);
+
+    EXPECT_MAT_NEAR(dst_gold, dst, 0.0);
+}
+
+TEST(ReduceToRow, Max)
+{
+    const Size size = randomSize(100, 400);
+
+    Mat src = randomMat(size, CV_8UC1);
+
+    GpuMat_<uchar> d_src(src);
+
+    GpuMat_<uchar> dst = reduceToRow_<Max<uchar> >(d_src);
+
+    Mat dst_gold;
+    cv::reduce(src, dst_gold, 0, REDUCE_MAX);
+
+    EXPECT_MAT_NEAR(dst_gold, dst, 0.0);
+}
+
+TEST(ReduceToColumn, Sum)
+{
+    const Size size = randomSize(100, 400);
+
+    Mat src = randomMat(size, CV_8UC1);
+
+    GpuMat_<uchar> d_src(src);
+
+    GpuMat_<int> dst = reduceToColumn_<Sum<int> >(d_src);
+
+    Mat dst_gold;
+    cv::reduce(src, dst_gold, 1, REDUCE_SUM, CV_32S);
+
+    EXPECT_MAT_NEAR(dst_gold, dst, 0.0);
+}
+
+TEST(ReduceToColumn, Avg)
+{
+    const Size size = randomSize(100, 400);
+
+    Mat src = randomMat(size, CV_8UC1);
+
+    GpuMat_<uchar> d_src(src);
+
+    GpuMat_<float> dst = reduceToColumn_<Avg<float> >(d_src);
+
+    Mat dst_gold;
+    cv::reduce(src, dst_gold, 1, REDUCE_AVG, CV_32F);
+
+    EXPECT_MAT_NEAR(dst_gold, dst, 1e-4);
+}
+
+TEST(ReduceToColumn, Min)
+{
+    const Size size = randomSize(100, 400);
+
+    Mat src = randomMat(size, CV_8UC1);
+
+    GpuMat_<uchar> d_src(src);
+
+    GpuMat_<uchar> dst = reduceToColumn_<Min<uchar> >(d_src);
+
+    Mat dst_gold;
+    cv::reduce(src, dst_gold, 1, REDUCE_MIN);
+
+    EXPECT_MAT_NEAR(dst_gold, dst, 0.0);
+}
+
+TEST(ReduceToColumn, Max)
+{
+    const Size size = randomSize(100, 400);
+
+    Mat src = randomMat(size, CV_8UC1);
+
+    GpuMat_<uchar> d_src(src);
+
+    GpuMat_<uchar> dst = reduceToColumn_<Max<uchar> >(d_src);
+
+    Mat dst_gold;
+    cv::reduce(src, dst_gold, 1, REDUCE_MAX);
+
+    EXPECT_MAT_NEAR(dst_gold, dst, 0.0);
+}
+
+static void calcHistGold(const cv::Mat& src, cv::Mat& hist)
+{
+    hist.create(1, 256, CV_32SC1);
+    hist.setTo(cv::Scalar::all(0));
+
+    int* hist_row = hist.ptr<int>();
+    for (int y = 0; y < src.rows; ++y)
+    {
+        const uchar* src_row = src.ptr(y);
+
+        for (int x = 0; x < src.cols; ++x)
+            ++hist_row[src_row[x]];
+    }
+}
+
+TEST(Histogram, GpuMat)
+{
+    const Size size = randomSize(100, 400);
+
+    Mat src = randomMat(size, CV_8UC1);
+
+    GpuMat_<uchar> d_src(src);
+
+    GpuMat_<int> dst = histogram_<256>(d_src);
+
+    Mat dst_gold;
+    calcHistGold(src, dst_gold);
+
+    EXPECT_MAT_NEAR(dst_gold, dst, 0.0);
+}

modules/cudev/test/test_split_merge.cu

@@ -0,0 +1,181 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#include "test_precomp.hpp"
+
+using namespace cv;
+using namespace cv::gpu;
+using namespace cv::cudev;
+using namespace cvtest;
+
+typedef ::testing::Types<uchar, ushort, short, int, float> AllTypes;
+
+////////////////////////////////////////////////////////////////////////////////
+// MergeTest
+
+template <typename T>
+class MergeTest : public ::testing::Test
+{
+public:
+    void test_c2()
+    {
+        const Size size = randomSize(100, 400);
+
+        const int src_type = DataType<T>::type;
+
+        Mat src1 = randomMat(size, src_type);
+        Mat src2 = randomMat(size, src_type);
+
+        GpuMat_<T> d_src1(src1);
+        GpuMat_<T> d_src2(src2);
+
+        GpuMat_<typename MakeVec<T, 2>::type> dst;
+        gridMerge(zipPtr(d_src1, d_src2), dst);
+
+        Mat dst_gold;
+        Mat srcs[] = {src1, src2};
+        cv::merge(srcs, 2, dst_gold);
+
+        EXPECT_MAT_NEAR(dst_gold, dst, 0.0);
+    }
+
+    void test_c3()
+    {
+        const Size size = randomSize(100, 400);
+
+        const int src_type = DataType<T>::type;
+
+        Mat src1 = randomMat(size, src_type);
+        Mat src2 = randomMat(size, src_type);
+        Mat src3 = randomMat(size, src_type);
+
+        GpuMat_<T> d_src1(src1);
+        GpuMat_<T> d_src2(src2);
+        GpuMat_<T> d_src3(src3);
+
+        GpuMat_<typename MakeVec<T, 3>::type> dst;
+        gridMerge(zipPtr(d_src1, d_src2, d_src3), dst);
+
+        Mat dst_gold;
+        Mat srcs[] = {src1, src2, src3};
+        cv::merge(srcs, 3, dst_gold);
+
+        ASSERT_MAT_NEAR(dst_gold, dst, 0.0);
+    }
+};
+
+TYPED_TEST_CASE(MergeTest, AllTypes);
+
+TYPED_TEST(MergeTest, C2)
+{
+    MergeTest<TypeParam>::test_c2();
+}
+
+TYPED_TEST(MergeTest, C3)
+{
+    MergeTest<TypeParam>::test_c3();
+}
+
+////////////////////////////////////////////////////////////////////////////////
+// SplitTest
+
+template <typename T>
+class SplitTest : public ::testing::Test
+{
+public:
+    void test_c3()
+    {
+        const Size size = randomSize(100, 400);
+
+        const int src_type = CV_MAKE_TYPE(DataType<T>::depth, 3);
+
+        Mat src = randomMat(size, src_type);
+
+        GpuMat_<typename MakeVec<T, 3>::type> d_src(src);
+
+        GpuMat_<T> dst1, dst2, dst3;
+        gridSplit(d_src, tie(dst1, dst2, dst3));
+
+        std::vector<Mat> dst;
+        cv::split(src, dst);
+
+        ASSERT_MAT_NEAR(dst[0], dst1, 0.0);
+        ASSERT_MAT_NEAR(dst[1], dst2, 0.0);
+        ASSERT_MAT_NEAR(dst[2], dst3, 0.0);
+    }
+
+    void test_c4()
+    {
+        const Size size = randomSize(100, 400);
+
+        const int src_type = CV_MAKE_TYPE(DataType<T>::depth, 4);
+
+        Mat src = randomMat(size, src_type);
+
+        GpuMat_<typename MakeVec<T, 4>::type> d_src(src);
+
+        GpuMat_<T> dst1, dst2, dst3, dst4;
+        gridSplit(d_src, tie(dst1, dst2, dst3, dst4));
+
+        std::vector<Mat> dst;
+        cv::split(src, dst);
+
+        ASSERT_MAT_NEAR(dst[0], dst1, 0.0);
+        ASSERT_MAT_NEAR(dst[1], dst2, 0.0);
+        ASSERT_MAT_NEAR(dst[2], dst3, 0.0);
+        ASSERT_MAT_NEAR(dst[3], dst4, 0.0);
+    }
+};
+
+TYPED_TEST_CASE(SplitTest, AllTypes);
+
+TYPED_TEST(SplitTest, C3)
+{
+    SplitTest<TypeParam>::test_c3();
+}
+
+TYPED_TEST(SplitTest, C4)
+{
+    SplitTest<TypeParam>::test_c4();
+}
+

modules/cudev/test/test_warp.cu

@@ -0,0 +1,256 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#include "test_precomp.hpp"
+
+using namespace cv;
+using namespace cv::gpu;
+using namespace cv::cudev;
+using namespace cvtest;
+
+// remap
+
+enum { HALF_SIZE=0, UPSIDE_DOWN, REFLECTION_X, REFLECTION_BOTH };
+
+static void generateMap(Mat& mapx, Mat& mapy, int remapMode)
+{
+    for (int j = 0; j < mapx.rows; ++j)
+    {
+        for (int i = 0; i < mapx.cols; ++i)
+        {
+            switch (remapMode)
+            {
+            case HALF_SIZE:
+                if (i > mapx.cols*0.25 && i < mapx.cols*0.75 && j > mapx.rows*0.25 && j < mapx.rows*0.75)
+                {
+                    mapx.at<float>(j,i) = 2.f * (i - mapx.cols * 0.25f) + 0.5f;
+                    mapy.at<float>(j,i) = 2.f * (j - mapx.rows * 0.25f) + 0.5f;
+                }
+                else
+                {
+                    mapx.at<float>(j,i) = 0.f;
+                    mapy.at<float>(j,i) = 0.f;
+                }
+                break;
+            case UPSIDE_DOWN:
+                mapx.at<float>(j,i) = static_cast<float>(i);
+                mapy.at<float>(j,i) = static_cast<float>(mapx.rows - j);
+                break;
+            case REFLECTION_X:
+                mapx.at<float>(j,i) = static_cast<float>(mapx.cols - i);
+                mapy.at<float>(j,i) = static_cast<float>(j);
+                break;
+            case REFLECTION_BOTH:
+                mapx.at<float>(j,i) = static_cast<float>(mapx.cols - i);
+                mapy.at<float>(j,i) = static_cast<float>(mapx.rows - j);
+                break;
+            } // end of switch
+        }
+    }
+}
+
+static void test_remap(int remapMode)
+{
+    const Size size = randomSize(100, 400);
+
+    Mat src = randomMat(size, CV_32FC1, 0, 1);
+
+    Mat mapx(size, CV_32FC1);
+    Mat mapy(size, CV_32FC1);
+    generateMap(mapx, mapy, remapMode);
+
+    GpuMat_<float> d_src(src);
+    GpuMat_<float> d_mapx(mapx);
+    GpuMat_<float> d_mapy(mapy);
+
+    GpuMat_<float> dst = remap_(interNearest(brdReplicate(d_src)), d_mapx, d_mapy);
+
+    Mat dst_gold;
+    cv::remap(src, dst_gold, mapx, mapy, INTER_NEAREST, BORDER_REPLICATE);
+
+    EXPECT_MAT_NEAR(dst_gold, dst, 0.0);
+}
+
+TEST(Remap, HALF_SIZE)
+{
+    test_remap(HALF_SIZE);
+}
+
+TEST(Remap, UPSIDE_DOWN)
+{
+    test_remap(UPSIDE_DOWN);
+}
+
+TEST(Remap, REFLECTION_X)
+{
+    test_remap(REFLECTION_X);
+}
+
+TEST(Remap, REFLECTION_BOTH)
+{
+    test_remap(REFLECTION_BOTH);
+}
+
+// resize
+
+TEST(Resize, Upscale)
+{
+    const Size size = randomSize(100, 400);
+
+    Mat src = randomMat(size, CV_32FC1, 0, 1);
+
+    GpuMat_<float> d_src(src);
+    Texture<float> tex_src(d_src);
+
+    GpuMat_<float> dst1 = resize_(interCubic(tex_src), 2, 2);
+
+    Mat mapx(size.height * 2, size.width * 2, CV_32FC1);
+    Mat mapy(size.height * 2, size.width * 2, CV_32FC1);
+
+    for (int y = 0; y < mapx.rows; ++y)
+    {
+        for (int x = 0; x < mapx.cols; ++x)
+        {
+            mapx.at<float>(y, x) = static_cast<float>(x / 2);
+            mapy.at<float>(y, x) = static_cast<float>(y / 2);
+        }
+    }
+
+    GpuMat_<float> d_mapx(mapx);
+    GpuMat_<float> d_mapy(mapy);
+
+    GpuMat_<float> dst2 = remap_(interCubic(brdReplicate(d_src)), d_mapx, d_mapy);
+
+    EXPECT_MAT_NEAR(dst1, dst2, 0.0);
+}
+
+TEST(Resize, Downscale)
+{
+    const Size size = randomSize(100, 400);
+
+    Mat src = randomMat(size, CV_32FC1, 0, 1);
+    const float fx = 1.0f / 3.0f;
+    const float fy = 1.0f / 3.0f;
+
+    GpuMat_<float> d_src(src);
+    Texture<float> tex_src(d_src);
+
+    GpuMat_<float> dst1 = resize_(interArea(tex_src, Size(3, 3)), fx, fy);
+
+    Mat mapx(cv::saturate_cast<int>(size.height * fy), cv::saturate_cast<int>(size.width * fx), CV_32FC1);
+    Mat mapy(cv::saturate_cast<int>(size.height * fy), cv::saturate_cast<int>(size.width * fx), CV_32FC1);
+
+    for (int y = 0; y < mapx.rows; ++y)
+    {
+        for (int x = 0; x < mapx.cols; ++x)
+        {
+            mapx.at<float>(y, x) = x / fx;
+            mapy.at<float>(y, x) = y / fy;
+        }
+    }
+
+    GpuMat_<float> d_mapx(mapx);
+    GpuMat_<float> d_mapy(mapy);
+
+    GpuMat_<float> dst2 = remap_(interArea(brdReplicate(d_src), Size(3, 3)), d_mapx, d_mapy);
+
+    EXPECT_MAT_NEAR(dst1, dst2, 0.0);
+}
+
+// warpAffine & warpPerspective
+
+Mat createAffineTransfomMatrix(Size srcSize, float angle, bool perspective)
+{
+    cv::Mat M(perspective ? 3 : 2, 3, CV_32FC1);
+
+    {
+        M.at<float>(0, 0) = std::cos(angle); M.at<float>(0, 1) = -std::sin(angle); M.at<float>(0, 2) = static_cast<float>(srcSize.width / 2);
+        M.at<float>(1, 0) = std::sin(angle); M.at<float>(1, 1) =  std::cos(angle); M.at<float>(1, 2) = 0.0f;
+    }
+    if (perspective)
+    {
+        M.at<float>(2, 0) = 0.0f           ; M.at<float>(2, 1) =  0.0f           ; M.at<float>(2, 2) = 1.0f;
+    }
+
+    return M;
+}
+
+TEST(WarpAffine, Rotation)
+{
+    const Size size = randomSize(100, 400);
+
+    Mat src = randomMat(size, CV_32FC1, 0, 1);
+    Mat M = createAffineTransfomMatrix(size, static_cast<float>(CV_PI / 4), false);
+
+    GpuMat_<float> d_src(src);
+    GpuMat_<float> d_M;
+    createContinuous(M.size(), M.type(), d_M);
+    d_M.upload(M);
+
+    GpuMat_<float> dst = warpAffine_(interNearest(brdConstant(d_src)), size, d_M);
+
+    Mat dst_gold;
+    cv::warpAffine(src, dst_gold, M, size, INTER_NEAREST | WARP_INVERSE_MAP);
+
+    EXPECT_MAT_SIMILAR(dst_gold, dst, 1e-3);
+}
+
+TEST(WarpPerspective, Rotation)
+{
+    const Size size = randomSize(100, 400);
+
+    Mat src = randomMat(size, CV_32FC1, 0, 1);
+    Mat M = createAffineTransfomMatrix(size, static_cast<float>(CV_PI / 4), true);
+
+    GpuMat_<float> d_src(src);
+    GpuMat_<float> d_M;
+    createContinuous(M.size(), M.type(), d_M);
+    d_M.upload(M);
+
+    GpuMat_<float> dst = warpPerspective_(interNearest(brdConstant(d_src)), size, d_M);
+
+    Mat dst_gold;
+    cv::warpPerspective(src, dst_gold, M, size, INTER_NEAREST | WARP_INVERSE_MAP);
+
+    EXPECT_MAT_SIMILAR(dst_gold, dst, 1e-3);
+}

modules/cudev/test/transpose.cu

@@ -0,0 +1,81 @@
+/*M///////////////////////////////////////////////////////////////////////////////////////
+//
+//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
+//
+//  By downloading, copying, installing or using the software you agree to this license.
+//  If you do not agree to this license, do not download, install,
+//  copy or use the software.
+//
+//
+//                          License Agreement
+//                For Open Source Computer Vision Library
+//
+// Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
+// Copyright (C) 2009, Willow Garage Inc., all rights reserved.
+// Copyright (C) 2013, OpenCV Foundation, all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//   * Redistribution's of source code must retain the above copyright notice,
+//     this list of conditions and the following disclaimer.
+//
+//   * Redistribution's in binary form must reproduce the above copyright notice,
+//     this list of conditions and the following disclaimer in the documentation
+//     and/or other materials provided with the distribution.
+//
+//   * The name of the copyright holders may not be used to endorse or promote products
+//     derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//
+//M*/
+
+#include "test_precomp.hpp"
+
+using namespace cv;
+using namespace cv::gpu;
+using namespace cv::cudev;
+using namespace cvtest;
+
+TEST(Transpose, _8uc1)
+{
+    const Size size = randomSize(100, 400);
+
+    Mat src = randomMat(size, CV_8UC1);
+
+    GpuMat_<uchar> d_src(src);
+
+    GpuMat_<uchar> dst = transpose_(d_src);
+
+    Mat dst_gold;
+    cv::transpose(src, dst_gold);
+
+    ASSERT_MAT_NEAR(dst_gold, dst, 0.0);
+}
+
+TEST(Transpose, _32fc3)
+{
+    const Size size = randomSize(100, 400);
+
+    Mat src = randomMat(size, CV_32FC3);
+
+    GpuMat_<float3> d_src(src);
+
+    GpuMat_<float3> dst = transpose_(d_src);
+
+    Mat dst_gold;
+    cv::transpose(src, dst_gold);
+
+    ASSERT_MAT_NEAR(dst_gold, dst, 0.0);
+}