undefined | Better HN

0 pointscoffeebeqn2y ago0 comments

Could there be a compiler/transpiler from CUSA to whatever AMD is pushing ?

0 comments

2 comments · 2 top-level

That's what https://github.com/ROCm/HIPIFY is (as a part of ROCm)

I think it's llama.cpp that simply #defines all cuda_ functions to rocm_ (99% name-name). Porting seems to be that trivial.

j / k navigate · click thread line to collapse