[PATCH] D84309: [InstCombine] Preserve FMA opportunities when combining PHI nodes

Roman Lebedev via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Wed Jul 22 07:45:00 PDT 2020


lebedev.ri added a comment.

In D84309#2166548 <https://reviews.llvm.org/D84309#2166548>, @spatel wrote:

> Avoiding the canonicalization seems unlikely to be a universal win and won't solve the problem if the source was already in the form that we want to avoid.
>  I think we need to reverse the transform in CodeGenPrepare. There we can use TTI/TLI to decide if it is profitable (at the least, the target should have custom/legal FMA lowering for the value type in question?).


+1, this direction does not seem like the right one for instcombine.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D84309/new/

https://reviews.llvm.org/D84309





More information about the llvm-commits mailing list