http://gcc.gnu.org/bugzilla/show_bug.cgi?id=48161
--- Comment #10 from Richard Guenther <rguenth at gcc dot gnu.org> 2011-03-17
15:37:24 UTC ---
(In reply to comment #9)
> Ah, it isn't any difference between --enable-checking=release and
> --enable-checking=yes behavior of cc1, but instead just that haifa-sched.i is
> different.
>
> We are expanding
> D.2070_29 = &MEM[(struct V *)0B].v[D.2034_30]{lb: 0 sz: 12};
> and expand that as (plus (const_int 0) (reg)).
Which doesn't sound wrong ;) Maybe simplification doesn't deal properly
with pointer constants? Well ... on RTL we can't distinguish pointer
constants from non-pointer constants ...
