I want to render some slices of a 3d surface.
cudaArray* surfArray;
cudaChannelFormatDesc channelDesc = cudaCreateChannelDesc(8, 8, 8, 8, cudaChannelFormatKindUnsigned);
cudaExtent surfSize = make_cudaExtent(640,480,2);
cudaMalloc3DArray(&surfArray,&channelDesc,surfSize,cudaArraySurfaceLoadStore);
Therefore, I tried to get the subsets via MemcpyArrayToArray() and an offset.
cudaGraphicsSubResourceGetMappedArray(&cu_rgbArray,tex_rgb,0,0);
cudaMemcpyArrayToArray(cu_rgbArray,0,0,surfArray,0,0,640*480*sizeof(uchar4),cudaMemcpyDeviceToDevice);
cudaGraphicsSubResourceGetMappedArray(&cu_depthArray,tex_depth,0,0);
cudaMemcpyArrayToArray(cu_depthArray,0,0,surfArray,640,480,640*480*sizeof(uchar4),cudaMemcpyDeviceToDevice);
However, while the first memcpy succeeds, the second one fails. (It does succeed if the offset is 0,0).
Any thoughts how to get around this problem without an extra kernel?
I believe you should use
cudaMemcpy3D
, notcudaMemcpyArrayToArray
.