diff --git a/modules/gpu/src/cuda/imgproc.cu b/modules/gpu/src/cuda/imgproc.cu
index 537364172..44a54ca47 100644
--- a/modules/gpu/src/cuda/imgproc.cu
+++ b/modules/gpu/src/cuda/imgproc.cu
@@ -66,6 +66,24 @@ namespace cv { namespace gpu { namespace imgproc
             dst.ptr(y)[x] = saturate_cast<T>(src(ycoo, xcoo));
         }
     }
+
+    template <template <typename> class Filter, template <typename> class B, typename T> struct RemapDispatcherStream
+    {
+        static void call(const DevMem2D_<T>& src, const DevMem2Df& mapx, const DevMem2Df& mapy, const DevMem2D_<T>& dst, const float* borderValue, cudaStream_t stream)
+        {
+            typedef typename TypeVec<float, VecTraits<T>::cn>::vec_type work_type; 
+            
+            dim3 block(32, 8);
+            dim3 grid(divUp(dst.cols, block.x), divUp(dst.rows, block.y));
+
+            B<work_type> brd(src.rows, src.cols, VecTraits<work_type>::make(borderValue));
+            BorderReader< PtrStep_<T>, B<work_type> > brdSrc(src, brd);
+            Filter< BorderReader< PtrStep_<T>, B<work_type> > > filter_src(brdSrc);
+
+            remap<<<grid, block, 0, stream>>>(filter_src, mapx, mapy, dst);
+            cudaSafeCall( cudaGetLastError() );
+        }
+    };
     
     template <template <typename> class Filter, template <typename> class B, typename T> struct RemapDispatcherNonStream
     {
@@ -163,22 +181,7 @@ namespace cv { namespace gpu { namespace imgproc
             if (stream == 0)
                 RemapDispatcherNonStream<Filter, B, T>::call(src, mapx, mapy, dst, borderValue);
             else
-                callStream(src, mapx, mapy, dst, borderValue, stream);
-        }
-        
-        static void callStream(const DevMem2D_<T>& src, const DevMem2Df& mapx, const DevMem2Df& mapy, const DevMem2D_<T>& dst, const float* borderValue, cudaStream_t stream)
-        {
-            typedef typename TypeVec<float, VecTraits<T>::cn>::vec_type work_type; 
-            
-            dim3 block(32, 8);
-            dim3 grid(divUp(dst.cols, block.x), divUp(dst.rows, block.y));
-
-            B<work_type> brd(src.rows, src.cols, VecTraits<work_type>::make(borderValue));
-            BorderReader< PtrStep_<T>, B<work_type> > brd_src(src, brd);
-            Filter< BorderReader< PtrStep_<T>, B<work_type> > > filter_src(brd_src);
-
-            remap<<<grid, block, 0, stream>>>(filter_src, mapx, mapy, dst);
-            cudaSafeCall( cudaGetLastError() );
+                RemapDispatcherStream<Filter, B, T>::call(src, mapx, mapy, dst, borderValue, stream);
         }
     };
 
diff --git a/modules/gpu/src/opencv2/gpu/device/border_interpolate.hpp b/modules/gpu/src/opencv2/gpu/device/border_interpolate.hpp
index 50c929a4a..1f226b98a 100644
--- a/modules/gpu/src/opencv2/gpu/device/border_interpolate.hpp
+++ b/modules/gpu/src/opencv2/gpu/device/border_interpolate.hpp
@@ -758,6 +758,29 @@ namespace cv { namespace gpu { namespace device
         const Ptr2D ptr;
         const B b;
     };
+
+    // under win32 there is some bug with templated types that passed as kernel parameters
+    // with this specialization all works fine
+    template <typename Ptr2D, typename D> struct BorderReader< Ptr2D, BrdConstant<D> >
+    {
+        typedef typename BrdConstant<D>::result_type elem_type;
+        typedef typename Ptr2D::index_type index_type;
+
+        __host__ __device__ __forceinline__ BorderReader(const Ptr2D& src_, const BrdConstant<D>& b) : 
+            src(src_), height(b.height), width(b.width), val(b.val) 
+        {
+        }
+
+        __device__ __forceinline__ D operator ()(index_type y, index_type x) const
+        {
+            return (x >= 0 && x < width && y >= 0 && y < height) ? saturate_cast<D>(src(y, x)) : val;
+        }
+
+        const Ptr2D src;
+        const int height;
+        const int width;
+        const D val;
+    };
 }}}
 
 #endif // __OPENCV_GPU_BORDER_INTERPOLATE_HPP__
diff --git a/samples/gpu/performance/tests.cpp b/samples/gpu/performance/tests.cpp
index e03988be3..5f70236af 100644
--- a/samples/gpu/performance/tests.cpp
+++ b/samples/gpu/performance/tests.cpp
@@ -80,7 +80,7 @@ TEST(remap)
     gpu::GpuMat d_src, d_dst, d_xmap, d_ymap;
     
     int interpolation = INTER_LINEAR;
-    int borderMode = BORDER_CONSTANT;
+    int borderMode = BORDER_REPLICATE;
 
     for (int size = 1000; size <= 4000; size *= 2)
     {