Add support for HIP via a lightweight header wrapper
This adds support for HIP via a lightweight wrapper that #define
s the appropriate cuda*
APIs to hip*
. This is sufficient to get a basic build working on Spock.
Status as of this MR:
- The code builds on Spock with HIP versions of orientation matching and phasing (CuPy)
- When running, I see a freeze at generation 1, regardless of whether GPU support is enabled or not
- However, I can comfirm that CuPy at least is running as intended, because phasing time is in fact reduced, even in generation 0