[llvm] [SelectionDAG] Deal with POISON for INSERT_VECTOR_ELT/INSERT_SUBVECTOR (part 1) (PR #143102)
Björn Pettersson via llvm-commits
llvm-commits at lists.llvm.org
Fri Jul 4 02:06:09 PDT 2025
bjope wrote:
Yes, I was preparing a rebase on top of 145939 (thanks!). And I've updated this pull request now.
As shown in https://github.com/llvm/llvm-project/pull/143103 we can do some enhancements in isGuaranteedNotToBeUndefOrPoison to avoid some diffs otherwise introduced by this patch. The opcodes I identified in that patch was INSERT_VECTOR_ELT/INSERT_SUBVECTOR, but I have another patch here https://github.com/llvm/llvm-project/pull/147019 that I think would make isGuaranteedNotToBeUndefOrPoison treat demanded elts in even more situations (as a more general improvement).
I haven't gotten any feedback about https://github.com/llvm/llvm-project/pull/143105 which would avoid lots of the regressions around scalable vectors (I think you stumbled upon that problem in #146332). But that relies on that it would be OK to set the extended lanes to POISON in convertToScalableVector for AArch64 and RISC-V.
https://github.com/llvm/llvm-project/pull/143102
More information about the llvm-commits
mailing list