Feature Request: HIP Support

[ghost]

HIP is a sort-of subset of the CUDA API with a few trivial name-changes, which compiles to either CUDA on NVidia platforms, or to compatible code on AMD cards. It's open-source AFAIK, and integrated into AMD's ROCm stack. AMD have been porting some of the big ML frameworks to ROCm using HIP, including Caffe.

It would be pretty cool if Rust's CUDA ecosystem could be grown to include AMD devices using HIP. But, being unable to contribute much to the effort, it's only a suggestion. :)