Search code examples
pythongpupycuda

Copying arrays with pyCuda from the GPU using pointers

I'm copying via pycuda some arrays on the GPU and then store the pointers to these arrays. How do I recuperate the data back?

dist = np.zeros(numPoints).astype(np.float32) distAddress = [gpuarray.to_gpu(dist).ptr for i in range(100)]

If I call the memcpy_dtoh function:

buf = np.zeros(400).astype(np.float32) cuda.memcpy_dtoh(buf,distAddress[0]), (where type(distAddress[0]) is long) I get the following error:

cuda.memcpy_dtoh(buf, distAddress[0]) LogicError: cuMemcpyDtoH failed: invalid argument

What am I doing wrong?

Thanks!

Solution

  • I think if you are using GPUArrays the way to copy from device to host is with the .get() method. For example

    dist = np.zeros(num_points).astype(np.float32)
dist_list = [gpuarray.to_gpu(dist) for i in range(100)]

buf = dist_list[0].get()