[PATCH] D134966: [DAG] Update foldSelectWithIdentityConstant to use llvm::isNeutralConstant

[Sanjay Patel via Phabricator via llvm-commits]

spatel added a comment.

In D134966#3833518 <https://reviews.llvm.org/D134966#3833518>, @goncharov wrote:

> as a headsup: I see this change causes failures in Halide https://github.com/halide/Halide/blob/main/test/correctness/div_round_to_zero.cpp on some architectures

Any chance you can post the IR for that code? My first guess is that we've hoisted a div/rem, and that's illegal for this transform because those are not speculatable ops. Looks like this would only happen with a 512-bit vector on an AVX512VL target if I'm seeing the target hook restriction correctly.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D134966/new/

https://reviews.llvm.org/D134966