Hacker News new | ask | show | jobs
by hzwanip 128 days ago
> No LLVM. No HIP translation layer. No "convert your CUDA to something else first."

What is the problem with such approaches?