Changed "NeutralValue" to function templates

Source/Math/GPUTensor.cu

@@ -264,16 +264,12 @@ struct TensorOps
 //----------------------------------------------------------------------------
 // For reductions we need the neutral elements of the corresponding binary ops
 //----------------------------------------------------------------------------
-class BinaryOpConstants
+template <typename ElemType> __device__ ElemType NeutralValue(ElementWiseOperator op)
 {
-public:
-    template <typename ElemType> __device__ static ElemType NeutralValue(ElementWiseOperator op)
-    {
-        return 0; // error, only the explicit instantiations below should be used.
-    }
+    return 0; // error, only the explicit instantiations below should be used.
 };
 
-template<> __device__ static float BinaryOpConstants::NeutralValue<float>(ElementWiseOperator op)
+template<> __device__ float NeutralValue<float>(ElementWiseOperator op)
 {
     switch (op)
     {
@@ -282,9 +278,9 @@ template<> __device__ static float BinaryOpConstants::NeutralValue<float>(Elemen
     case ElementWiseOperator::opSum: return 0;
     default:                         return 0; // error
     }
-}
+};
 
-template<> __device__ static double BinaryOpConstants::NeutralValue<double>(ElementWiseOperator op)
+template<> __device__ double NeutralValue<double>(ElementWiseOperator op)
 {
     switch (op)
     {
@@ -293,7 +289,7 @@ template<> __device__ static double BinaryOpConstants::NeutralValue<double>(Elem
     case ElementWiseOperator::opSum: return 0;
     default:                         return 0; // error
     }
-}
+};
 
 
 // ----------------------------------------------------------------------------
@@ -505,7 +501,7 @@ struct TensorOpElement<ElemType, N, M, K, /*parallelReduce=*/true, /*k=*/-1>
         CUDA_LONG reductionEnd = min(reductionBegin + reductionChunkSize, reductionDim);
 
         // compute the operation for this input coordinate
-        ReduceElemType aggregate = BinaryOpConstants::NeutralValue<ReduceElemType>(reductionOp);
+        ReduceElemType aggregate = NeutralValue<ReduceElemType>(reductionOp);
 
         for (CUDA_LONG redId = reductionBegin + tid; redId < reductionEnd; redId += tids)
         {