xiaoxi
February 10, 2011, 2:12am
1
I have tried to use function cudaMemcpy2D, code as below, but it doesn’t work. Anyone can help me? Thank you in advance.
BYTE gpu_h; //hashtable ) &gpu_h, sizeof(BYTE*) * curlen);h_q ; )malloc(sizeof(BYTE*) * 10);
…
int nx = 4, ny = 10; // lengght of sizeof(BYTE*) is 4, so I set nx(width) as 4.
size_t d_pitchBytes;
size_t h_pitchBytes = ny*sizeof(BYTE*);
cudaMallocPitch( (void**) &gpu_h, &d_pitchBytes, nx*sizeof(byte), ny);
cudaMemcpy2D( h_q, h_pitchBytes, gpu_h, d_pitchBytes, nx * sizeof(BYTE), ny, cudaMemcpyDeviceToHost);
Thanks
You’re using cudaMemcpy2D on a buffer you allocated using cudaMalloc, you should just use cudaMemcpy () instead. It looks like you might have meant to allocate an array using cudaMallocPitch though. Change one or the other?
curlen must == 10 (otherwise sizeof ( gpu_h ) != sizeof ( h_q ))
size_t h_pitchBytes = nysizeof(BYTE );
The pitch is the WIDTH in bytes of one row, so nx * sizeof(byte), make sure d_pitchBytes is correct in this respect.
I have tried to use function cudaMemcpy2D, code as below, but it doesn’t work. Anyone can help me? Thank you in advance.
BYTE **gpu_h; //hashtable
cudaMalloc((void**) &gpu_h, sizeof(BYTE*) * curlen);
BYTE **h_q ;
h_q = (BYTE**)malloc(sizeof(BYTE*) * 10);
…
find_matchlen_test <<<grid, THREAD_NUM>>>(gpu_out, gpu_p, gpu_h);
…
int nx = 4, ny = 10; // lengght of sizeof(BYTE*) is 4, so I set nx(width) as 4.
size_t d_pitchBytes;
size_t h_pitchBytes = ny*sizeof(BYTE*);
cudaMallocPitch( (void**) &gpu_h, &d_pitchBytes, nx*sizeof(byte), ny);
cudaMemcpy2D( h_q, h_pitchBytes, gpu_h, d_pitchBytes, nx * sizeof(BYTE), ny, cudaMemcpyDeviceToHost);
Thanks
Rock