[Bug debug/105630] [10/11/12/13 Regression] '-fcompare-debug' failure w/ -O1 -ftree-parallelize-loops=2

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

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105630

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jakub at gcc dot gnu.org

--- Comment #1 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Ah, the difference between this and PR105415 is that here the VAR_DECL n has
DECL_RTL set, so we don't go through
      /* This decl was probably optimized away.  */
      if (!op0
          /* At least label RTXen are sometimes replaced by
             NOTE_INSN_DELETED_LABEL.  Any notes here are not
             handled by copy_rtx.  */
          || NOTE_P (op0))
        {
          if (!VAR_P (exp)
              || DECL_EXTERNAL (exp)
              || !TREE_STATIC (exp)
              || !DECL_NAME (exp)
              || DECL_HARD_REGISTER (exp)
              || DECL_IN_CONSTANT_POOL (exp)
              || mode == VOIDmode
              || symtab_node::get (exp) == NULL)
            return NULL;

          op0 = make_decl_rtl_for_debug (exp);
          if (!MEM_P (op0)
              || GET_CODE (XEXP (op0, 0)) != SYMBOL_REF
              || SYMBOL_REF_DECL (XEXP (op0, 0)) != exp)
            return NULL;
        }
because op0 is non-NULL, so we go with
      else
        op0 = copy_rtx (op0);
instead.
So perhaps something like:
      else if (VAR_P (exp)
               && is_global_var (exp)
               && symtab_node::get (exp) == NULL)
        return NULL;
before the else or so?
But, we'll need to compare the effects on debug info with that change too.