https://gcc.gnu.org/bugzilla/show_bug.cgi?id=123111
--- Comment #5 from Andrew Pinski <pinskia at gcc dot gnu.org> --- Created attachment 63053 --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=63053&action=edit testcase that shows the problem before This a slightly modifed version of the original one which shows the problem even in GCC 5. The reason why it shows up there (even without my cleanupcfg patches) is because critial edges splits up the phi and nothing removes the forwarder block because of the phi in the fowarder block. I didn't expect all of this performance fallout But it definitely showcases how much things were just by accidently working (or in this testcase not working).
