1.x related fixes

40c76b9d · marina.kolpakova · 3f68e5bb · 40c76b9d · 40c76b9d · 40c76b9d
Commit 40c76b9d authored Aug 09, 2012 by marina.kolpakova
6 changed files
--- a/modules/gpu/perf/perf_labeling.cpp
+++ b/modules/gpu/perf/perf_labeling.cpp
@@ -48,7 +48,9 @@ GPU_PERF_TEST(ConnectedComponents, cv::gpu::DeviceInfo, cv::Size)
    cv::gpu::DeviceInfo devInfo = GET_PARAM(0);
    cv::gpu::setDevice(devInfo.deviceID());
-    cv::Mat image = readImage("gpu/labeling/label.png", cv::IMREAD_GRAYSCALE);
+    cv::Mat image = readImage("gpu/labeling/aloe-disp.png", cv::IMREAD_GRAYSCALE);
+    cv::threshold(image, image, 150, 255, CV_THRESH_BINARY);
    cv::gpu::GpuMat mask;
    mask.create(image.rows, image.cols, CV_8UC1);

--- a/modules/gpu/perf_cpu/perf_labeling.cpp
+++ b/modules/gpu/perf_cpu/perf_labeling.cpp
+/*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) 2008-2011, Willow Garage Inc., all rights reserved.
+// Third party copyrights are property of their respective owners.
+//
+// Redistribution and use in source and binary forms, with or without modification,
+// are permitted provided that the following conditions are met:
+//
+//  * 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.
+//
+//  * The name of the copyright holders may not be used to endorse or promote products
+//    derived from this software without specific prior written permission.
+//
+// This software is provided by the copyright holders and contributors "as is" and
+// any express or implied warranties, including, but not limited to, the implied
+// warranties of merchantability and fitness for a particular purpose are disclaimed.
+// In no event shall the Intel Corporation or contributors be liable for any direct,
+// indirect, incidental, special, exemplary, or consequential damages
+// (including, but not limited to, procurement of substitute goods or services;
+// loss of use, data, or profits; or business interruption) however caused
+// and on any theory of liability, whether in contract, strict liability,
+// or tort (including negligence or otherwise) arising in any way out of
+// the use of this software, even if advised of the possibility of such damage.
+//M*/
+#include "perf_precomp.hpp"
+#ifdef HAVE_CUDA
+namespace {
+    struct GreedyLabeling
+    {
+        struct dot
+        {
+            int x;
+            int y;
+            static dot make(int i, int j)
+            {
+                dot d; d.x = i; d.y = j;
+                return d;
+            }
+        };
+        struct InInterval
+        {
+            InInterval(const int& _lo, const int& _hi) : lo(-_lo), hi(_hi) {};
+            const int lo, hi;
+            bool operator() (const unsigned char a, const unsigned char b) const
+            {
+                int d = a - b;
+                return lo <= d && d <= hi;
+            }
+        };
+        GreedyLabeling(cv::Mat img)
+        : image(img), _labels(image.size(), CV_32SC1, cv::Scalar::all(-1)) {stack = new dot[image.cols * image.rows];}
+        ~GreedyLabeling(){delete[] stack;}
+        void operator() (cv::Mat labels) const
+        {
+            InInterval inInt(0, 2);
+            int cc = -1;
+            int* dist_labels = (int*)labels.data;
+            int pitch = labels.step1();
+            unsigned char* source = (unsigned char*)image.data;
+            int width = image.cols;
+            int height = image.rows;
+            for (int j = 0; j < image.rows; ++j)
+                for (int i = 0; i < image.cols; ++i)
+                {
+                    if (dist_labels[j * pitch + i] != -1) continue;
+                    dot* top = stack;
+                    dot p = dot::make(i, j);
+                    cc++;
+                    dist_labels[j * pitch + i] = cc;
+                    while (top >= stack)
+                    {
+                        int*  dl = &dist_labels[p.y * pitch + p.x];
+                        unsigned char* sp = &source[p.y * image.step1() + p.x];
+                        dl[0] = cc;
+                        //right
+                        if( p.x < (width - 1) && dl[ +1] == -1 && inInt(sp[0], sp[+1]))
+                            *top++ = dot::make(p.x + 1, p.y);
+                        //left
+                        if( p.x > 0 && dl[-1] == -1 && inInt(sp[0], sp[-1]))
+                            *top++ = dot::make(p.x - 1, p.y);
+                        //bottom
+                        if( p.y < (height - 1) && dl[+pitch] == -1 && inInt(sp[0], sp[+image.step1()]))
+                            *top++ = dot::make(p.x, p.y + 1);
+                        //top
+                        if( p.y > 0 && dl[-pitch] == -1 && inInt(sp[0], sp[-image.step1()]))
+                            *top++ = dot::make(p.x, p.y - 1);
+                        p = *--top;
+                    }
+                }
+        }
+        cv::Mat image;
+        cv::Mat _labels;
+        dot* stack;
+    };
+}
+GPU_PERF_TEST(ConnectedComponents, cv::gpu::DeviceInfo, cv::Size)
+{
+    cv::gpu::DeviceInfo devInfo = GET_PARAM(0);
+    cv::gpu::setDevice(devInfo.deviceID());
+    cv::Mat image = readImage("gpu/labeling/aloe-disp.png", cv::IMREAD_GRAYSCALE);
+    GreedyLabeling host(image);
+    host(host._labels);
+    declare.time(1.0);
+    TEST_CYCLE()
+    {
+        host(host._labels);
+    }
+}
+INSTANTIATE_TEST_CASE_P(Labeling, ConnectedComponents, testing::Combine(ALL_DEVICES, testing::Values(cv::Size(261, 262))));
+#endif
\ No newline at end of file
--- a/modules/gpu/src/cuda/ccomponetns.cu
+++ b/modules/gpu/src/cuda/ccomponetns.cu
@@ -42,6 +42,7 @@
 #include <opencv2/gpu/device/common.hpp>
 #include <opencv2/gpu/device/vec_traits.hpp>
 #include <opencv2/gpu/device/vec_math.hpp>
+#include <opencv2/gpu/device/emulation.hpp>
 #include <iostream>
 #include <stdio.h>
@@ -255,8 +256,7 @@ namespace cv { namespace gpu { namespace device
                    edgesTile[yloc][xloc] = c;
                }
+            for (int k = 0; ;++k)
-            for (int i = 0; ; ++i)
            {
                //1. backup
                #pragma unroll
@@ -312,11 +312,12 @@ namespace cv { namespace gpu { namespace device
                        if (new_labels[i][j] < old_labels[i][j])
                        {
                            changed = 1;
-                            atomicMin(&labelsTile[0][0] + old_labels[i][j], new_labels[i][j]);
+                            Emulation::smem::atomicMin(&labelsTile[0][0] + old_labels[i][j], new_labels[i][j]);
                        }
                    }
-                changed = __syncthreads_or(changed);
+                changed = Emulation::sycthOr(changed);
                if (!changed)
                    break;

--- a/modules/gpu/src/graphcuts.cpp
+++ b/modules/gpu/src/graphcuts.cpp
 /*M///////////////////////////////////////////////////////////////////////////////////////
 //
 //  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
 //
 //  By downloading, copying, installing or using the software you agree to this license.
 //  If you do not agree to this license, do not download, install,
 //  copy or use the software.
 //
 //
 //                           License Agreement
 //                For Open Source Computer Vision Library
 //
 // Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
 // Copyright (C) 2009, Willow Garage Inc., all rights reserved.
 // Third party copyrights are property of their respective owners.
 //
 // Redistribution and use in source and binary forms, with or without modification,
 // are permitted provided that the following conditions are met:
 //
 //   * Redistribution's of source code must retain the above copyright notice,
 //     this list of conditions and the following disclaimer.
 //
 //   * Redistribution's in binary form must reproduce the above copyright notice,
 //     this list of conditions and the following disclaimer in the documentation
 //     and/or other GpuMaterials 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 bpied warranties, including, but not limited to, the bpied
 // warranties of merchantability and fitness for a particular purpose are disclaimed.
 // In no event shall the Intel Corporation or contributors be liable for any direct,
 // indirect, incidental, special, exemplary, or consequential damages
 // (including, but not limited to, procurement of substitute goods or services;
 // loss of use, data, or profits; or business interruption) however caused
 // and on any theory of liability, whether in contract, strict liability,
 // or tort (including negligence or otherwise) arising in any way out of
 // the use of this software, even if advised of the possibility of such damage.
 //
 //M*/
 #include "precomp.hpp"
 #if !defined (HAVE_CUDA)
 void cv::gpu::graphcut(GpuMat&, GpuMat&, GpuMat&, GpuMat&, GpuMat&, GpuMat&, GpuMat&, Stream&) { throw_nogpu(); }
 void cv::gpu::graphcut(GpuMat&, GpuMat&, GpuMat&, GpuMat&, GpuMat&, GpuMat&, GpuMat&, GpuMat&, GpuMat&, GpuMat&, GpuMat&, Stream&) { throw_nogpu(); }
 void cv::gpu::connectivityMask(const GpuMat&, GpuMat&, const cv::Scalar&, const cv::Scalar&, Stream&) { throw_nogpu(); }
 void cv::gpu::labelComponents(const GpuMat& mask, GpuMat& components, int, Stream& stream) { throw_nogpu(); }
 #else /* !defined (HAVE_CUDA) */
 namespace cv { namespace gpu { namespace device
 {
    namespace ccl
    {
        void labelComponents(const DevMem2D& edges, DevMem2Di comps, int flags, cudaStream_t stream);
        template<typename T>
        void computeEdges(const DevMem2D& image, DevMem2D edges, const float4& lo, const float4& hi, cudaStream_t stream);
    }
 }}}
 float4 scalarToCudaType(const cv::Scalar& in)
 {
    float4 res;
    res.x = in[0]; res.y = in[1]; res.z = in[2]; res.w = in[3];
    return res;
 }
 void cv::gpu::connectivityMask(const GpuMat& image, GpuMat& mask, const cv::Scalar& lo, const cv::Scalar& hi, Stream& s)
 {
    CV_Assert(!image.empty());
    int ch = image.channels();
    CV_Assert(ch <= 4);
    int depth = image.depth();
    typedef void (*func_t)(const DevMem2D& image, DevMem2D edges, const float4& lo, const float4& hi, cudaStream_t stream);
    static const func_t suppotLookup[8][4] =
    {   //    1,    2,     3,     4
        { device::ccl::computeEdges<uchar>,  0,  device::ccl::computeEdges<uchar3>,  device::ccl::computeEdges<uchar4>  },// CV_8U
        { 0,                                 0,  0,                                  0                                  },// CV_16U
        { device::ccl::computeEdges<ushort>, 0,  device::ccl::computeEdges<ushort3>, device::ccl::computeEdges<ushort4> },// CV_8S
        { 0,                                 0,  0,                                  0                                  },// CV_16S
        { device::ccl::computeEdges<int>,    0,  0,                                  0                                  },// CV_32S
        { device::ccl::computeEdges<float>,  0,  0,                                  0                                  },// CV_32F
        { 0,                                 0,  0,                                  0                                  },// CV_64F
        { 0,                                 0,  0,                                  0                                  } // CV_USRTYPE1
    };
    func_t f = suppotLookup[depth][ch - 1];
    CV_Assert(f);
    if (image.size() != mask.size() || mask.type() != CV_8UC1)
        mask.create(image.size(), CV_8UC1);
    cudaStream_t stream = StreamAccessor::getStream(s);
    float4 culo = scalarToCudaType(lo), cuhi = scalarToCudaType(hi);
    f(image, mask, culo, cuhi, stream);
 }
 void cv::gpu::labelComponents(const GpuMat& mask, GpuMat& components, int flags, Stream& s)
 {
-    CV_Assert(!mask.empty() && mask.type() == CV_8U);
+    if (!TargetArchs::builtWith(SHARED_ATOMICS) || !DeviceInfo().supports(SHARED_ATOMICS))
+        CV_Error(CV_StsNotImplemented, "The device doesn't support shared atomics and communicative synchronization!");
-    if (mask.size() != components.size() || components.type() != CV_32SC1)
+    CV_Assert(!mask.empty() && mask.type() == CV_8U);
-        components.create(mask.size(), CV_32SC1);
+    if (mask.size() != components.size() || components.type() != CV_32SC1)
-    cudaStream_t stream = StreamAccessor::getStream(s);
+        components.create(mask.size(), CV_32SC1);
-    device::ccl::labelComponents(mask, components, flags, stream);
-}
+    cudaStream_t stream = StreamAccessor::getStream(s);
+    device::ccl::labelComponents(mask, components, flags, stream);
-namespace
+}
-{
-    typedef NppStatus (*init_func_t)(NppiSize oSize, NppiGraphcutState** ppState, Npp8u* pDeviceMem);
+namespace
+{
-    class NppiGraphcutStateHandler
+    typedef NppStatus (*init_func_t)(NppiSize oSize, NppiGraphcutState** ppState, Npp8u* pDeviceMem);
-    {
-    public:
+    class NppiGraphcutStateHandler
-        NppiGraphcutStateHandler(NppiSize sznpp, Npp8u* pDeviceMem, const init_func_t func)
+    {
-        {
+    public:
-            nppSafeCall( func(sznpp, &pState, pDeviceMem) );
+        NppiGraphcutStateHandler(NppiSize sznpp, Npp8u* pDeviceMem, const init_func_t func)
-        }
+        {
+            nppSafeCall( func(sznpp, &pState, pDeviceMem) );
-        ~NppiGraphcutStateHandler()
+        }
-        {
-            nppSafeCall( nppiGraphcutFree(pState) );
+        ~NppiGraphcutStateHandler()
-        }
+        {
+            nppSafeCall( nppiGraphcutFree(pState) );
-        operator NppiGraphcutState*()
+        }
-        {
-            return pState;
+        operator NppiGraphcutState*()
-        }
+        {
+            return pState;
-    private:
+        }
-        NppiGraphcutState* pState;
-    };
+    private:
-}
+        NppiGraphcutState* pState;
+    };
-void cv::gpu::graphcut(GpuMat& terminals, GpuMat& leftTransp, GpuMat& rightTransp, GpuMat& top, GpuMat& bottom, GpuMat& labels, GpuMat& buf, Stream& s)
+}
-{
-#if (CUDA_VERSION < 5000)
+void cv::gpu::graphcut(GpuMat& terminals, GpuMat& leftTransp, GpuMat& rightTransp, GpuMat& top, GpuMat& bottom, GpuMat& labels, GpuMat& buf, Stream& s)
-    CV_Assert(terminals.type() == CV_32S);
+{
-#else
+#if (CUDA_VERSION < 5000)
-    CV_Assert(terminals.type() == CV_32S || terminals.type() == CV_32F);
+    CV_Assert(terminals.type() == CV_32S);
-#endif
+#else
+    CV_Assert(terminals.type() == CV_32S || terminals.type() == CV_32F);
-    Size src_size = terminals.size();
+#endif
-    CV_Assert(leftTransp.size() == Size(src_size.height, src_size.width));
+    Size src_size = terminals.size();
-    CV_Assert(leftTransp.type() == terminals.type());
+    CV_Assert(leftTransp.size() == Size(src_size.height, src_size.width));
-    CV_Assert(rightTransp.size() == Size(src_size.height, src_size.width));
+    CV_Assert(leftTransp.type() == terminals.type());
-    CV_Assert(rightTransp.type() == terminals.type());
+    CV_Assert(rightTransp.size() == Size(src_size.height, src_size.width));
-    CV_Assert(top.size() == src_size);
+    CV_Assert(rightTransp.type() == terminals.type());
-    CV_Assert(top.type() == terminals.type());
+    CV_Assert(top.size() == src_size);
-    CV_Assert(bottom.size() == src_size);
+    CV_Assert(top.type() == terminals.type());
-    CV_Assert(bottom.type() == terminals.type());
+    CV_Assert(bottom.size() == src_size);
-    labels.create(src_size, CV_8U);
+    CV_Assert(bottom.type() == terminals.type());
-    NppiSize sznpp;
+    labels.create(src_size, CV_8U);
-    sznpp.width = src_size.width;
-    sznpp.height = src_size.height;
+    NppiSize sznpp;
+    sznpp.width = src_size.width;
-    int bufsz;
+    sznpp.height = src_size.height;
-    nppSafeCall( nppiGraphcutGetSize(sznpp, &bufsz) );
+    int bufsz;
-    ensureSizeIsEnough(1, bufsz, CV_8U, buf);
+    nppSafeCall( nppiGraphcutGetSize(sznpp, &bufsz) );
-    cudaStream_t stream = StreamAccessor::getStream(s);
+    ensureSizeIsEnough(1, bufsz, CV_8U, buf);
-    NppStreamHandler h(stream);
+    cudaStream_t stream = StreamAccessor::getStream(s);
-    NppiGraphcutStateHandler state(sznpp, buf.ptr<Npp8u>(), nppiGraphcutInitAlloc);
+    NppStreamHandler h(stream);
-#if (CUDA_VERSION < 5000)
+    NppiGraphcutStateHandler state(sznpp, buf.ptr<Npp8u>(), nppiGraphcutInitAlloc);
-    nppSafeCall( nppiGraphcut_32s8u(terminals.ptr<Npp32s>(), leftTransp.ptr<Npp32s>(), rightTransp.ptr<Npp32s>(), top.ptr<Npp32s>(), bottom.ptr<Npp32s>(),
-        static_cast<int>(terminals.step), static_cast<int>(leftTransp.step), sznpp, labels.ptr<Npp8u>(), static_cast<int>(labels.step), state) );
+#if (CUDA_VERSION < 5000)
-#else
+    nppSafeCall( nppiGraphcut_32s8u(terminals.ptr<Npp32s>(), leftTransp.ptr<Npp32s>(), rightTransp.ptr<Npp32s>(), top.ptr<Npp32s>(), bottom.ptr<Npp32s>(),
-    if (terminals.type() == CV_32S)
+        static_cast<int>(terminals.step), static_cast<int>(leftTransp.step), sznpp, labels.ptr<Npp8u>(), static_cast<int>(labels.step), state) );
-    {
+#else
-        nppSafeCall( nppiGraphcut_32s8u(terminals.ptr<Npp32s>(), leftTransp.ptr<Npp32s>(), rightTransp.ptr<Npp32s>(), top.ptr<Npp32s>(), bottom.ptr<Npp32s>(),
+    if (terminals.type() == CV_32S)
-            static_cast<int>(terminals.step), static_cast<int>(leftTransp.step), sznpp, labels.ptr<Npp8u>(), static_cast<int>(labels.step), state) );
+    {
-    }
+        nppSafeCall( nppiGraphcut_32s8u(terminals.ptr<Npp32s>(), leftTransp.ptr<Npp32s>(), rightTransp.ptr<Npp32s>(), top.ptr<Npp32s>(), bottom.ptr<Npp32s>(),
-    else
+            static_cast<int>(terminals.step), static_cast<int>(leftTransp.step), sznpp, labels.ptr<Npp8u>(), static_cast<int>(labels.step), state) );
-    {
+    }
-        nppSafeCall( nppiGraphcut_32f8u(terminals.ptr<Npp32f>(), leftTransp.ptr<Npp32f>(), rightTransp.ptr<Npp32f>(), top.ptr<Npp32f>(), bottom.ptr<Npp32f>(),
+    else
-            static_cast<int>(terminals.step), static_cast<int>(leftTransp.step), sznpp, labels.ptr<Npp8u>(), static_cast<int>(labels.step), state) );
+    {
-    }
+        nppSafeCall( nppiGraphcut_32f8u(terminals.ptr<Npp32f>(), leftTransp.ptr<Npp32f>(), rightTransp.ptr<Npp32f>(), top.ptr<Npp32f>(), bottom.ptr<Npp32f>(),
-#endif
+            static_cast<int>(terminals.step), static_cast<int>(leftTransp.step), sznpp, labels.ptr<Npp8u>(), static_cast<int>(labels.step), state) );
+    }
-    if (stream == 0)
+#endif
-        cudaSafeCall( cudaDeviceSynchronize() );
-}
+    if (stream == 0)
+        cudaSafeCall( cudaDeviceSynchronize() );
-void cv::gpu::graphcut(GpuMat& terminals, GpuMat& leftTransp, GpuMat& rightTransp, GpuMat& top, GpuMat& topLeft, GpuMat& topRight,
+}
-              GpuMat& bottom, GpuMat& bottomLeft, GpuMat& bottomRight, GpuMat& labels, GpuMat& buf, Stream& s)
-{
+void cv::gpu::graphcut(GpuMat& terminals, GpuMat& leftTransp, GpuMat& rightTransp, GpuMat& top, GpuMat& topLeft, GpuMat& topRight,
-#if (CUDA_VERSION < 5000)
+              GpuMat& bottom, GpuMat& bottomLeft, GpuMat& bottomRight, GpuMat& labels, GpuMat& buf, Stream& s)
-    CV_Assert(terminals.type() == CV_32S);
+{
-#else
+#if (CUDA_VERSION < 5000)
-    CV_Assert(terminals.type() == CV_32S || terminals.type() == CV_32F);
+    CV_Assert(terminals.type() == CV_32S);
-#endif
+#else
+    CV_Assert(terminals.type() == CV_32S || terminals.type() == CV_32F);
-    Size src_size = terminals.size();
+#endif
-    CV_Assert(leftTransp.size() == Size(src_size.height, src_size.width));
+    Size src_size = terminals.size();
-    CV_Assert(leftTransp.type() == terminals.type());
+    CV_Assert(leftTransp.size() == Size(src_size.height, src_size.width));
-    CV_Assert(rightTransp.size() == Size(src_size.height, src_size.width));
+    CV_Assert(leftTransp.type() == terminals.type());
-    CV_Assert(rightTransp.type() == terminals.type());
+    CV_Assert(rightTransp.size() == Size(src_size.height, src_size.width));
-    CV_Assert(top.size() == src_size);
+    CV_Assert(rightTransp.type() == terminals.type());
-    CV_Assert(top.type() == terminals.type());
+    CV_Assert(top.size() == src_size);
-    CV_Assert(topLeft.size() == src_size);
+    CV_Assert(top.type() == terminals.type());
-    CV_Assert(topLeft.type() == terminals.type());
+    CV_Assert(topLeft.size() == src_size);
-    CV_Assert(topRight.size() == src_size);
+    CV_Assert(topLeft.type() == terminals.type());
-    CV_Assert(topRight.type() == terminals.type());
+    CV_Assert(topRight.size() == src_size);
-    CV_Assert(bottom.size() == src_size);
+    CV_Assert(topRight.type() == terminals.type());
-    CV_Assert(bottom.type() == terminals.type());
+    CV_Assert(bottom.size() == src_size);
-    CV_Assert(bottomLeft.size() == src_size);
+    CV_Assert(bottom.type() == terminals.type());
-    CV_Assert(bottomLeft.type() == terminals.type());
+    CV_Assert(bottomLeft.size() == src_size);
-    CV_Assert(bottomRight.size() == src_size);
+    CV_Assert(bottomLeft.type() == terminals.type());
-    CV_Assert(bottomRight.type() == terminals.type());
+    CV_Assert(bottomRight.size() == src_size);
-    labels.create(src_size, CV_8U);
+    CV_Assert(bottomRight.type() == terminals.type());
-    NppiSize sznpp;
+    labels.create(src_size, CV_8U);
-    sznpp.width = src_size.width;
-    sznpp.height = src_size.height;
+    NppiSize sznpp;
+    sznpp.width = src_size.width;
-    int bufsz;
+    sznpp.height = src_size.height;
-    nppSafeCall( nppiGraphcut8GetSize(sznpp, &bufsz) );
+    int bufsz;
-    ensureSizeIsEnough(1, bufsz, CV_8U, buf);
+    nppSafeCall( nppiGraphcut8GetSize(sznpp, &bufsz) );
-    cudaStream_t stream = StreamAccessor::getStream(s);
+    ensureSizeIsEnough(1, bufsz, CV_8U, buf);
-    NppStreamHandler h(stream);
+    cudaStream_t stream = StreamAccessor::getStream(s);
-    NppiGraphcutStateHandler state(sznpp, buf.ptr<Npp8u>(), nppiGraphcut8InitAlloc);
+    NppStreamHandler h(stream);
-#if (CUDA_VERSION < 5000)
+    NppiGraphcutStateHandler state(sznpp, buf.ptr<Npp8u>(), nppiGraphcut8InitAlloc);
-    nppSafeCall( nppiGraphcut8_32s8u(terminals.ptr<Npp32s>(), leftTransp.ptr<Npp32s>(), rightTransp.ptr<Npp32s>(),
-        top.ptr<Npp32s>(), topLeft.ptr<Npp32s>(), topRight.ptr<Npp32s>(),
+#if (CUDA_VERSION < 5000)
-        bottom.ptr<Npp32s>(), bottomLeft.ptr<Npp32s>(), bottomRight.ptr<Npp32s>(),
+    nppSafeCall( nppiGraphcut8_32s8u(terminals.ptr<Npp32s>(), leftTransp.ptr<Npp32s>(), rightTransp.ptr<Npp32s>(),
-        static_cast<int>(terminals.step), static_cast<int>(leftTransp.step), sznpp, labels.ptr<Npp8u>(), static_cast<int>(labels.step), state) );
+        top.ptr<Npp32s>(), topLeft.ptr<Npp32s>(), topRight.ptr<Npp32s>(),
-#else
+        bottom.ptr<Npp32s>(), bottomLeft.ptr<Npp32s>(), bottomRight.ptr<Npp32s>(),
-    if (terminals.type() == CV_32S)
+        static_cast<int>(terminals.step), static_cast<int>(leftTransp.step), sznpp, labels.ptr<Npp8u>(), static_cast<int>(labels.step), state) );
-    {
+#else
-        nppSafeCall( nppiGraphcut8_32s8u(terminals.ptr<Npp32s>(), leftTransp.ptr<Npp32s>(), rightTransp.ptr<Npp32s>(),
+    if (terminals.type() == CV_32S)
-            top.ptr<Npp32s>(), topLeft.ptr<Npp32s>(), topRight.ptr<Npp32s>(),
+    {
-            bottom.ptr<Npp32s>(), bottomLeft.ptr<Npp32s>(), bottomRight.ptr<Npp32s>(),
+        nppSafeCall( nppiGraphcut8_32s8u(terminals.ptr<Npp32s>(), leftTransp.ptr<Npp32s>(), rightTransp.ptr<Npp32s>(),
-            static_cast<int>(terminals.step), static_cast<int>(leftTransp.step), sznpp, labels.ptr<Npp8u>(), static_cast<int>(labels.step), state) );
+            top.ptr<Npp32s>(), topLeft.ptr<Npp32s>(), topRight.ptr<Npp32s>(),
-    }
+            bottom.ptr<Npp32s>(), bottomLeft.ptr<Npp32s>(), bottomRight.ptr<Npp32s>(),
-    else
+            static_cast<int>(terminals.step), static_cast<int>(leftTransp.step), sznpp, labels.ptr<Npp8u>(), static_cast<int>(labels.step), state) );
-    {
+    }
-        nppSafeCall( nppiGraphcut8_32f8u(terminals.ptr<Npp32f>(), leftTransp.ptr<Npp32f>(), rightTransp.ptr<Npp32f>(),
+    else
-            top.ptr<Npp32f>(), topLeft.ptr<Npp32f>(), topRight.ptr<Npp32f>(),
+    {
-            bottom.ptr<Npp32f>(), bottomLeft.ptr<Npp32f>(), bottomRight.ptr<Npp32f>(),
+        nppSafeCall( nppiGraphcut8_32f8u(terminals.ptr<Npp32f>(), leftTransp.ptr<Npp32f>(), rightTransp.ptr<Npp32f>(),
-            static_cast<int>(terminals.step), static_cast<int>(leftTransp.step), sznpp, labels.ptr<Npp8u>(), static_cast<int>(labels.step), state) );
+            top.ptr<Npp32f>(), topLeft.ptr<Npp32f>(), topRight.ptr<Npp32f>(),
-    }
+            bottom.ptr<Npp32f>(), bottomLeft.ptr<Npp32f>(), bottomRight.ptr<Npp32f>(),
-#endif
+            static_cast<int>(terminals.step), static_cast<int>(leftTransp.step), sznpp, labels.ptr<Npp8u>(), static_cast<int>(labels.step), state) );
+    }
-    if (stream == 0)
+#endif
-        cudaSafeCall( cudaDeviceSynchronize() );
-}
+    if (stream == 0)
+        cudaSafeCall( cudaDeviceSynchronize() );
-#endif /* !defined (HAVE_CUDA) */
+}
+#endif /* !defined (HAVE_CUDA) */
--- a/modules/gpu/src/opencv2/gpu/device/emulation.hpp
+++ b/modules/gpu/src/opencv2/gpu/device/emulation.hpp
 /*M///////////////////////////////////////////////////////////////////////////////////////
 //
 //  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
 //
 //  By downloading, copying, installing or using the software you agree to this license.
 //  If you do not agree to this license, do not download, install,
 //  copy or use the software.
 //
 //
 //                           License Agreement
 //                For Open Source Computer Vision Library
 //
 // Copyright (C) 2000-2008, Intel Corporation, all rights reserved.
 // Copyright (C) 2009, Willow Garage Inc., all rights reserved.
 // Third party copyrights are property of their respective owners.
 //
 // Redistribution and use in source and binary forms, with or without modification,
 // are permitted provided that the following conditions are met:
 //
 //   * Redistribution's of source code must retain the above copyright notice,
 //     this list of conditions and the following disclaimer.
 //
 //   * Redistribution's in binary form must reproduce the above copyright notice,
 //     this list of conditions and the following disclaimer in the documentation
 //     and/or other materials provided with the distribution.
 //
 //   * The name of the copyright holders may not be used to endorse or promote products
 //     derived from this software without specific prior written permission.
 //
 // This software is provided by the copyright holders and contributors "as is" and
 // any express or bpied warranties, including, but not limited to, the bpied
 // 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_GPU_EMULATION_HPP_
 #define OPENCV_GPU_EMULATION_HPP_
 #include "warp_reduce.hpp"
 #include <stdio.h>
 namespace cv { namespace gpu { namespace device
 {
    struct Emulation
    {
-        template<int CTA_SIZE>
-        static __forceinline__ __device__ int Ballot(int predicate)
+        static __device__ __forceinline__ int sycthOr(int pred)
        {
-#if defined (__CUDA_ARCH__) && (__CUDA_ARCH__ >= 200)
+#if defined (__CUDA_ARCH__) && (__CUDA_ARCH__ < 120)
-            return __ballot(predicate);
+                // just campilation stab
-#else
+                return false;
-            __shared__ volatile int cta_buffer[CTA_SIZE];
+#else
+                return __syncthreads_or(pred);
-            int tid = threadIdx.x;
+#endif
-            cta_buffer[tid] = predicate ? (1 << (tid & 31)) : 0;
+        }
-            return warp_reduce(cta_buffer);
-#endif
+        template<int CTA_SIZE>
-        }
+        static __forceinline__ __device__ int Ballot(int predicate)
+        {
-        struct smem
+#if defined (__CUDA_ARCH__) && (__CUDA_ARCH__ >= 200)
-        {
+            return __ballot(predicate);
-            enum { TAG_MASK = (1U << ( (sizeof(unsigned int) << 3) - 5U)) - 1U };
+#else
+            __shared__ volatile int cta_buffer[CTA_SIZE];
-            template<typename T>
-            static __device__ __forceinline__ T atomicInc(T* address, T val)
+            int tid = threadIdx.x;
-            {
+            cta_buffer[tid] = predicate ? (1 << (tid & 31)) : 0;
-#if defined (__CUDA_ARCH__) && (__CUDA_ARCH__ < 120)
+            return warp_reduce(cta_buffer);
-                T count;
+#endif
-                unsigned int tag = threadIdx.x << ( (sizeof(unsigned int) << 3) - 5U);
+        }
-                do
-                {
+        struct smem
-                    count = *address & TAG_MASK;
+        {
-                    count = tag | (count + 1);
+            enum { TAG_MASK = (1U << ( (sizeof(unsigned int) << 3) - 5U)) - 1U };
-                    *address = count;
-                } while (*address != count);
+            template<typename T>
+            static __device__ __forceinline__ T atomicInc(T* address, T val)
-                return (count & TAG_MASK) - 1;
+            {
-#else
+#if defined (__CUDA_ARCH__) && (__CUDA_ARCH__ < 120)
-                return ::atomicInc(address, val);
+                T count;
-#endif
+                unsigned int tag = threadIdx.x << ( (sizeof(unsigned int) << 3) - 5U);
-            }
+                do
+                {
-            template<typename T>
+                    count = *address & TAG_MASK;
-            static __device__ __forceinline__ void atomicAdd(T* address, T val)
+                    count = tag | (count + 1);
-            {
+                    *address = count;
-#if defined (__CUDA_ARCH__) && (__CUDA_ARCH__ < 120)
+                } while (*address != count);
-                T count;
-                unsigned int tag = threadIdx.x << ( (sizeof(unsigned int) << 3) - 5U);
+                return (count & TAG_MASK) - 1;
-                do
+#else
-                {
+                return ::atomicInc(address, val);
-                    count = *address & TAG_MASK;
+#endif
-                    count = tag | (count + val);
+            }
-                    *address = count;
-                } while (*address != count);
+            template<typename T>
-#else
+            static __device__ __forceinline__ void atomicAdd(T* address, T val)
-                ::atomicAdd(address, val);
+            {
-#endif
+#if defined (__CUDA_ARCH__) && (__CUDA_ARCH__ < 120)
-            }
+                T count;
+                unsigned int tag = threadIdx.x << ( (sizeof(unsigned int) << 3) - 5U);
-            template<typename T>
+                do
-            static __device__ __forceinline__ T atomicMin(T* address, T val)
+                {
-            {
+                    count = *address & TAG_MASK;
-#if defined (__CUDA_ARCH__) && (__CUDA_ARCH__ < 120)
+                    count = tag | (count + val);
-                T count = min(*address, val);
+                    *address = count;
-                do
+                } while (*address != count);
-                {
+#else
-                    *address = count;
+                ::atomicAdd(address, val);
-                } while (*address > count);
+#endif
+            }
-                return count;
-#else
+            template<typename T>
-                return ::atomicMin(address, val);
+            static __device__ __forceinline__ T atomicMin(T* address, T val)
-#endif
+            {
-            }
+#if defined (__CUDA_ARCH__) && (__CUDA_ARCH__ < 120)
-        };
+                T count = min(*address, val);
-    };
+                do
-}}} // namespace cv { namespace gpu { namespace device
+                {
+                    *address = count;
+                } while (*address > count);
+                return count;
+#else
+                return ::atomicMin(address, val);
+#endif
+            }
+        };
+    };
+}}} // namespace cv { namespace gpu { namespace device
 #endif /* OPENCV_GPU_EMULATION_HPP_ */
\ No newline at end of file
--- a/modules/gpu/test/test_labeling.cpp
+++ b/modules/gpu/test/test_labeling.cpp
@@ -164,7 +164,7 @@ struct Labeling : testing::TestWithParam<cv::gpu::DeviceInfo>
    cv::Mat loat_image()
    {
-        return cv::imread(std::string( cvtest::TS::ptr()->get_data_path() ) + "labeling/label.png");
+        return cv::imread(std::string( cvtest::TS::ptr()->get_data_path() ) + "labeling/IMG_0727.JPG");
    }
 };
@@ -173,6 +173,8 @@ TEST_P(Labeling, ConnectedComponents)
    cv::Mat image;
    cvtColor(loat_image(), image, CV_BGR2GRAY);
+    cv::threshold(image, image, 150, 255, CV_THRESH_BINARY);
    ASSERT_TRUE(image.type() == CV_8UC1);
    GreedyLabeling host(image);
@@ -189,6 +191,10 @@ TEST_P(Labeling, ConnectedComponents)
    ASSERT_NO_THROW(cv::gpu::labelComponents(mask, components));
    host.checkCorrectness(cv::Mat(components));
+    cv::imshow("test", image);
+    cv::waitKey(0);
+    cv::imshow("test", host._labels);
+    cv::waitKey(0);
 }
 INSTANTIATE_TEST_CASE_P(ConnectedComponents, Labeling, ALL_DEVICES);