Question

I am calling cudaMemcpy and the copy returns successfully however the source values are not being copied to the destination. I wrote a similar piece using memcpy() and that works fine. What am I missing here?

// host externs
extern unsigned char landmask[DIMX * DIMY];

// use device constant memory for landmask
unsigned char *tempmask;
__device__ unsigned char *landmask_d;

..

void checkCUDAError(const char* msg) {
cudaError_t err = cudaGetLastError();
  if (cudaSuccess != err) {
    fprintf(stderr, "Cuda error: %s: %s.\n", msg, cudaGetErrorString(err));
    exit(EXIT_FAILURE);
  }
}

..

// try a memcpy
size_t landMemSize_t landMemSize = DIMX * DIMY * sizeof(char);
tempmask = (unsigned char*)malloc(landMemSize);
memcpy(tempmask, landmask, DIMX * DIMY);
if (landmask[0] != *tempmask) printf("FAILURE!\n");

// allocate device memory for landmask
cudaMalloc((void **)&landmask_d, landMemSize);
printf("allocating %ld Kbytes for landmask\n", landMemSize/1024);
checkCUDAError("memory allocation");

// copy landmask from host to device
cudaMemcpy(landmask_d, landmask, landMemSize, cudaMemcpyHostToDevice);
checkCUDAError("memory copy");
if (landmask[0] != *landmask_d) printf("FAILURE!\n");

None of the CUDAErrors report any problem yet the second FAILURE is called..?

Answer 1

 if (landmask[0] != *landmask_d) printf("FAILURE!\n");

you are comparing values in different memory domains.
You should first copy memory from device to cpu and then compare