cudamemcpyasync code example

Example: cudamemcpyasync

cudaStream_t streams[3];

for (int i = 0; i < 3; i++) {
	cudaStreamCreate(&streams[i]);
}

cudaMemcpyAsync(h_array_1, d_array_1, size_1 * sizeof(int), cudaMemcpyDeviceToHost, streams[0]);
cudaMemcpyAsync(h_array_2, d_array_2, size_2 * sizeof(int), cudaMemcpyDeviceToHost, streams[1]);
cudaMemcpyAsync(h_array_3, d_array_3, size_3 * sizeof(int), cudaMemcpyDeviceToHost, streams[2]);

cudaDeviceSynchronize();

for (int i = 0; i < 3; i++) {
	cudaStreamDestroy(streams[i]);
}