[PATCH] D145278: Attributor: Start looking at uses when inferring nofpclass

[Matt Arsenault via Phabricator via llvm-commits]

arsenm added inline comments.


================
Comment at: llvm/test/Transforms/Attributor/nofpclass.ll:267
   %phi = phi float [ %call, %bb0 ], [ %arg, %entry ]
   ret float %phi
 }
----------------
jdoerfert wrote:
> If you introduce a block for the non-nan case you can use the value and instruction in the VisitValueCB to determine its non-nan. Only need to do that if the CtxI of the AA for the value is not the CtxI of the VisitValueCB though.
I missed this comment initially and don't quite follow it


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D145278/new/

https://reviews.llvm.org/D145278