https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118384
--- Comment #13 from Andrew Pinski <pinskia at gcc dot gnu.org> --- Maybe the better check is if the sequence has a call in it, don't use that version unless the other one has a call in it. That should punish the call version since calls might cause caller saved registers needing to spill which would cause an increase in both speed and size ... Let me try that out as a heuristic improvement here.
