We should port these, which will show gaps in our support: https://github.com/NVIDIA/cuda-samples/