https://gcc.gnu.org/bugzilla/show_bug.cgi?id=83497
--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> --- The initial bisection rev. is indeed odd. I'm testing a fix for the inliner issue btw, but I've not seen the mgrid fail on any of our x86/ia64 configs. As far as I see the miscompare is -0.8 vs. 0.18 so it doesn't look like a precision issue to me. Does it only happen for power6 / bigendian? Maybe you can narrow down the root-cause somewhat - mgrid is a very small benchmark. power6 always hints at scheduling.
