Improve issues with CUDA interacting with std flags