http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50190
Richard Guenther <rguenth at gcc dot gnu.org> changed: What |Removed |Added ---------------------------------------------------------------------------- Target|x86-64 |x86_64-*-* Component|rtl-optimization |fortran Host|x86-64 | Build|x86-64 | --- Comment #1 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-08-26 07:43:35 UTC --- I think it's not a "bug" in the sense that the documentation states that this can happen. Whether Fortran people want to have a working polyhedron with -Ofast is of course their decision. For x86_64 we want a working SPEC CPU 2006 with -Ofast for example, just as a measure as what is reasonable to allow with -Ofast.