[PATCH] D84516: [AMDGPU] Use fma for lowering frem

[Jay Foad via Phabricator via llvm-commits]

foad added a comment.

> If it doesn’t actually help accuracy, maybe it should check for fast FMA

I don't have any evidence that it improves accuracy but I'm not sure what the conditions should be for using FMA (I looked at SITargetLowering::isFMAFasterThanFMulAndFAdd but it is complicated) so I left it as a TODO.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D84516/new/

https://reviews.llvm.org/D84516