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("mem开发者_Go百科ory 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..?
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
精彩评论