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

Michael Halkenhäuser via Openmp-commits openmp-commits at lists.llvm.org
Mon Dec 4 12:02:34 PST 2023


mhalk wrote:

> > > Jay's suggestion sounds good to me. HIP has different tradeoffs from living in a different repo - maybe the right thing to do there is include the header in comgr and re-export the function, but it's all very heavy handed relative to the #include and done approach in tree runtimes get to use.
> > 
> > 
> > On my first attempt to rename & move the corresponding `Generation` enum into `lib/Target/AMDGPU/Utils/AMDGPUBaseInfo.h` I would need to adapt (at least) ~ 25 other files within the backend; and TBH I've yet to get it to work as expected.
> > @jayfoad @JonChesterfield @jhuber6 Q: How should I go about this change to avoid introducing (lingering?) errors into the backend? Any suggestions welcome! (Is there a way to narrow that number? [I guess not.])
> 
> Can you just make an inline function somewhere that the current implementation also calls?

Another level of indirection sounds good to me, I'll check that.

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


More information about the Openmp-commits mailing list