https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117013
--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> --- Guess so, though here you have more code to consider, so propagate it back; plus the optab has changed meaning yesterday and again today (but I don't see further changes in the meaning on the table). For the 4 way (HONOR_NANS floating point case) the question is if it wouldn't be best to have conditional jump for the unlikely case (unordered) and handle the rest branchless.
