[Bug tree-optimization/46879] [4.6 Regression] ICE: in separate_decls_in_region_debug_bind, at tree-parloops.c:778 with -flto -ftree-parallelize-loops -gdwarf-3

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46879

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2010.12.10 15:58:39
     Ever Confirmed|0                           |1

--- Comment #4 from Richard Guenther <rguenth at gcc dot gnu.org> 2010-12-10 15:58:39 UTC ---
(In reply to comment #3)
> (In reply to comment #2)
> > Hm, well.  Then the debug machinery has to deal with type mismatches I guess.
> 
> Well, all the debug machinery cares is that the var has the original, declared
> type.
> 
> > But I wonder why we have debug-stmts for global variables anyway?  I thought
> > they are only used for registers.
> 
> It is not a global variable, it is a VAR_DECL with DECL_ABSTRACT_ORIGIN
> PARM_DECL in bar, i.e. optimized away parameter/automatic variable.  It is a
> non-localized var in main, because it is never referenced there, except for
> GIMPLE_DEBUG.

Ah, ok.  The fix is then to amend the

1768              if (TREE_CODE (*basep) == VAR_DECL
1769              && !auto_var_in_fn_p (*basep, current_function_decl))

check properly.  The easiest check would be && !is_gimple_debug, or
a auto_var_in_fn_p variant without a specific function, thus auto_var_p.