https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122307
--- Comment #4 from Igor Shevlyakov <igor at tachyum dot com> --- Sorry, -fsignalling-nans is not related to the subject. half of IEEE comparisons are "signaling" on qNaNs. I think only -ffinite-math-only and -fno-traping-math shodul affect this check.
