On the contrary, it's great. Cuda is the single sane compute API and system, so I'll use it even if it means being vendor-locked. If my CUDA programs start running elsewhere without much intervention, that'd be amazing