[PATCH] D40130: [InstSimplify] fold and/or of fcmp ord/uno when operand is known nnan

[Eli Friedman via Phabricator via llvm-commits]

efriedma added a comment.

>From the perspective of canonicalization, it seems kind of weird to have "%x uno %y" rather than "(%x uno 0) | (%y uno 0)".  I mean, the former is probably cheaper to lower, but it leads to a lot of special-case logic to handle the equivalency.


https://reviews.llvm.org/D40130