[Openmp-commits] [openmp] [OpenMP][AMDGPU] Adapt dynamic callstack sizes to HIP behavior (PR #74080)

[Joseph Huber via Openmp-commits]

jhuber6 wrote:

> > @jayfoad @b-sumner any comments on moving the duplicated code above to a header in llvm/lib/Target/AMDGPU?
> 
> I don't think the HIP/OpenCL runtime wants this dependence. So it would require a new API to be introduced in COMgr. "The llvm libs are already in scope for openmp runtime" is surprising to me.

The OpenMP libraries live in tree, so it's fairly trivial to include LLVM. OpenMP uses the LLVM backends for JIT functionality, as well as a load of ELF handling. We also use LLVM error handling and data structures almost everywhere. The LLVM libraries are generally statically linked unless shared libraries are explicitly requested by the user, so it introduces no extra dependencies in a release.

https://github.com/llvm/llvm-project/pull/74080