[PATCH] D155437: ValueTracking: Fix computeKnownFPClass canonicalize handling

[Matt Arsenault via Phabricator via llvm-commits]

arsenm added a comment.

In D155437#4509006 <https://reviews.llvm.org/D155437#4509006>, @foad wrote:

>> It was also mishandling literal signaling nans as always poison.
>
> I don't see anything in this patch related to constants or poison.

Constants were just demonstrating the issue, with output that could only be poison since the returned constant violated the reported class


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D155437/new/

https://reviews.llvm.org/D155437