[PATCH] D12096: [InstCombineAddSub opportunities]: More opportunities to factorize FAdd/FSub when unsafeAlgebra is present for Inst

[David Majnemer via llvm-commits]

majnemer added a comment.

In http://reviews.llvm.org/D12096#232730, @aditya_nandakumar wrote:

> Also out of general curiosity, shouldn't instcombine be capable of handling these patterns?(if reassociate is not run before instcombine in a pass pipeline).


IMO, we should endeavor to make InstCombine expect output canonicalized by reassociate (where possible/reasonable) and tell users to feed reassociate into InstCombine.


http://reviews.llvm.org/D12096