Add missing THCudaCheck around cudaMemcpy

Author: pavanky

THCReduceAll.cuh

@@ -331,7 +331,7 @@ bool THC_reduceAll(THCState* state,
  // If our destination is not on the device, copy the value back to
  // the host (synchronous!)
  if (!outOnDevice) {
- cudaMemcpy(out, devOut, sizeof(AccT), cudaMemcpyDeviceToHost);
+ THCudaCheck(cudaMemcpy(out, devOut, sizeof(AccT), cudaMemcpyDeviceToHost));
  }
 
  if (freeDevOut) {

generic/THCTensorMath.cu

@@ -223,7 +223,7 @@ void THCTensor_(catArray)(THCState *state, THCTensor *result,
  // update offset
  offset += dimSize;
  }
- cudaMemcpy(d_inputs, stackInputs, j * sizeof(CatArrInputTensor<real, unsigned int>), cudaMemcpyHostToDevice);
+ THCudaCheck(cudaMemcpy(d_inputs, stackInputs, j * sizeof(CatArrInputTensor<real, unsigned int>), cudaMemcpyHostToDevice));
 
  // Next, let's consider how we set our kernel launch parameters.
  // We borrow from THCApply, which the kernel's internal indexing