cudaGetLastError() for asynchronous calls

Accelerated Computing CUDA CUDA Programming and Performance
1

I have a typical for loop that asynchronously copies data to device and calls (also asynchronously) kernels that process those chunks of data.

My question is: how to control that kernels executed successfully?

For the synchronous case we call:

cudaDeviceSynchronize(); cudaGetLastError();

But, is the cudaGetLastError() a proper choice for the async case? Currently, I do the following:

for( size_t si = 0; si < streams_num; si++ ) {
cudaMemcpyAsync(..., streams[si]);
kernel<<<....,streams[si]>>>();
checkCudaErrors(cudaStreamSynchronize(streams[si]));
checkCudaErrors(cudaGetLastError());
}

checkCudaErrors is an auxiliary macros that handles the return code.

2

google “proper cuda error checking”

take the first hit

it explains how to handle the API call case, and the kernel call case