[PATCH] D42806: [AArch64] optionally filter out denorms when using frsqrte to calculate sqrt

[Eli Friedman via Phabricator via llvm-commits]

efriedma added a comment.

> Should still change the select operand to FPZero to produce the 'bic' rather than 'bsl' though?

I think that changes the result for sqrt(-0.0)?  I guess we could still do it under appropriate fast-math flags.


https://reviews.llvm.org/D42806