[llvm] [AMDGPU] Improve selection of ballot.i64 intrinsic in wave32 mode in SelectionDAG. (PR #71556)

Matt Arsenault via llvm-commits llvm-commits at lists.llvm.org
Tue Nov 14 23:25:27 PST 2023


arsenm wrote:

> > I'm confused, InstCombine can't be too late, it runs many times after all IR producers. There's no practical path between the sanitizer passes and the backend where it won't run
> 
> I'm sorry, I thought it's machine IR pass. I don't see wavesize type correction for _ballot_ in _GCNTTIImpl::instCombineIntrinsic_, do you beleive it's better to add it there?

Oh yes, it is missing. I recently added it for amdgcn_mbcnt_hi above. We should do the same for ballot 

https://github.com/llvm/llvm-project/pull/71556


More information about the llvm-commits mailing list