QPyTorch
QPyTorch copied to clipboard
Enablement on ROCm
This PR contains the below two changes
- forceinline needs inline and always_inline on ROCm
- extra_include_paths is required for the hipification of quant_cuda header files.