[*] Fixed #1464

[~] NCVPyramid uses tr1 and thus can be compiled with CL, commented out on linux
[+] Moved reduction functors to NCVAlg
[*] Warnings in NCV

modules/gpu/src/nvidia/NCVHaarObjectDetection.cu

@@ -565,14 +565,7 @@ __global__ void applyHaarClassifierClassifierParallel(Ncv32u *d_IImg, Ncv32u IIm
             curRootNodeOffset += NUM_THREADS_CLASSIFIERPARALLEL;
         }
 
-        struct functorAddValues
+        Ncv32f finalStageSum = subReduce<Ncv32f, functorAddValues<Ncv32f>, NUM_THREADS_CLASSIFIERPARALLEL>(curStageSum);
-        {
-            __device__ void reduce(Ncv32f &in1out, Ncv32f &in2)
-            {
-                in1out += in2;
-            }
-        };
-        Ncv32f finalStageSum = subReduce<Ncv32f, functorAddValues, NUM_THREADS_CLASSIFIERPARALLEL>(curStageSum);
 
         if (finalStageSum < stageThreshold)
         {

modules/gpu/src/nvidia/core/NCV.cu

@@ -278,6 +278,7 @@ NCVMemStackAllocator::NCVMemStackAllocator(NCVMemoryType memT, size_t capacity,
 {
     NcvBool bProperAlignment = (alignment & (alignment-1)) == 0;
     ncvAssertPrintCheck(bProperAlignment, "NCVMemStackAllocator ctor:: _alignment not power of 2");
+    ncvAssertPrintCheck(memT != NCVMemoryTypeNone, "NCVMemStackAllocator ctor:: Incorrect allocator type");
 
     allocBegin = NULL;
 
@@ -295,6 +296,7 @@ NCVMemStackAllocator::NCVMemStackAllocator(NCVMemoryType memT, size_t capacity,
         case NCVMemoryTypeHostPageable:
             allocBegin = (Ncv8u *)malloc(capacity);
             break;
+        default:;
         }
     }
     else
@@ -335,6 +337,7 @@ NCVMemStackAllocator::~NCVMemStackAllocator()
             case NCVMemoryTypeHostPageable:
                 free(allocBegin);
                 break;
+            default:;
             }
         }
 
@@ -455,6 +458,7 @@ NCVStatus NCVMemNativeAllocator::alloc(NCVMemSegment &seg, size_t size)
     case NCVMemoryTypeHostPageable:
         seg.begin.ptr = (Ncv8u *)malloc(size);
         break;
+    default:;
     }
 
     this->currentSize += alignUp(size, this->_alignment);
@@ -487,6 +491,7 @@ NCVStatus NCVMemNativeAllocator::dealloc(NCVMemSegment &seg)
     case NCVMemoryTypeHostPageable:
         free(seg.begin.ptr);
         break;
+    default:;
     }
 
     seg.clear();

modules/gpu/src/nvidia/core/NCVAlg.hpp

@@ -52,6 +52,36 @@ static T divUp(T a, T b)
 }
 
 
+template<typename T>
+struct functorAddValues
+{
+    static __device__ __inline__ void reduce(T &in1out, T &in2)
+    {
+        in1out += in2;
+    }
+};
+
+
+template<typename T>
+struct functorMinValues
+{
+    static __device__ __inline__ void reduce(T &in1out, T &in2)
+    {
+        in1out = in1out > in2 ? in2 : in1out;
+    }
+};
+
+
+template<typename T>
+struct functorMaxValues
+{
+    static __device__ __inline__ void reduce(T &in1out, T &in2)
+    {
+        in1out = in1out > in2 ? in1out : in2;
+    }
+};
+
+
 template<typename Tdata, class Tfunc, Ncv32u nThreads>
 static __device__ Tdata subReduce(Tdata threadElem)
 {

modules/gpu/src/nvidia/core/NCVPyramid.cu

@@ -46,6 +46,7 @@
 #include "NCVPyramid.hpp"
 #include "NCVPixelOperations.hpp"
 
+#ifdef _WIN32
 
 template<typename T, Ncv32u CN> struct __average4_CN {static T _average4_CN(const T &p00, const T &p01, const T &p10, const T &p11);};
 
@@ -395,3 +396,5 @@ template class NCVImagePyramid<uint4>;
 template class NCVImagePyramid<float1>;
 template class NCVImagePyramid<float3>;
 template class NCVImagePyramid<float4>;
+
+#endif //_WIN32

modules/gpu/src/nvidia/core/NCVPyramid.hpp

@@ -47,6 +47,7 @@
 #include <vector>
 #include "NCV.hpp"
 
+#ifdef _WIN32
 
 template <class T>
 class NCV_EXPORTS NCVMatrixStack
@@ -93,5 +94,6 @@ private:
     Ncv32u nLayers;
 };
 
+#endif //_WIN32
 
 #endif //_ncvpyramid_hpp_