[Bug tree-optimization/67915] ICE on valid code at -O2 and -O3 on x86_64-linux-gnu

["rguenth at gcc dot gnu.org" <gcc-bugzilla at gcc dot gnu dot org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=67915

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2015-10-12
                 CC|                            |rsandifo at gcc dot gnu.org
     Ever confirmed|0                           |1

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
Confirmed.  Awww - that code again :/

The issue is that cleanup_control_expr_graph doing

        case GIMPLE_COND:
          val = fold_binary_loc (loc, gimple_cond_code (stmt),
                                 boolean_type_node,
                                 gimple_cond_lhs (stmt),
                                 gimple_cond_rhs (stmt));
          break;

now walks the SSA def chain but those defs may be released (were in a CFG
cleanup which can end up removing BBs and releasing SSA names).

My plan for the above was to remove the fold_binary and just detect whether
this is a true/false condition (see in gimple-fold how we canonicalize those).
Last time I tried this there were some missed CFG cleanups because of that
(unfolded GIMPLE_CONDs).

But of course the fallback was always to do this cleanup in some better
order (though CFG cleanup cannot compute sth like dominators for the obvious
reason - unreachable blocks).