https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66489
--- Comment #2 from ktkachov at gcc dot gnu.org --- (In reply to Andrew Pinski from comment #1) > This was the reason why the pass on the tree level to fused multiply add was > added. Maybe we can do the same. > > Do you know how often this shows up? My understanding is that it occurs in SPEC2006 and leslie3d in particular, but with FP code there (just substitute int for double in the testcase and pick something other than eor). Don't have any numbers on how often or how much of a bottleneck it is to performance...
