[Openmp-commits] [llvm] [openmp] [OpenMP][offload] Inline target reductions (PR #196061)

[Robert Imschweiler via Openmp-commits]

ro-i wrote:

> If you use always_inline everywhere but the entry points (or you use flatten on the entry points), you get no speedup, correct?

yes

> What if you only use always_inline on the entry points?

yeah, that is actually sufficient, I removed the others

> Doesn't the first question imply that this is not about the autgenerated functions (copyFn, shflFn, ...)?

I don't think so. See https://github.com/llvm/llvm-project/pull/195940#issuecomment-4404570419

https://github.com/llvm/llvm-project/pull/196061