[Bug tree-optimization/23297] immediate uses hosed after CCP

["amacleod at redhat dot com" <gcc-bugzilla@gcc.gnu.org>]

------- Additional Comments From amacleod at redhat dot com  2005-08-10 15:55 -------
A quick glance shows that parse_ssa_operands() does not expect to receive a tree
that looks like:

b.c[d_5].i

get_expr_operands is called on this, and  when processing COMPONENT_REF:
       ref = okay_component_ref_for_subvars (expr, &offset, &size);
        if (ref)
          {
            subvar_t svars = get_subvars_for_var (ref);
            subvar_t sv;
            for (sv = svars; sv; sv = sv->next)
              {
                bool exact;
                if (overlap_subvar (offset, size, sv, &exact))
                  {
                    int subvar_flags = flags;
                    if (!exact)
                      subvar_flags &= ~opf_kill_def;
                    add_stmt_operand (&sv->var, s_ann, subvar_flags);
                  }
              }
          }
        else
          get_expr_operands (stmt, &TREE_OPERAND (expr, 0),
                             flags & ~opf_kill_def);

if okay_component_ref_for_subvars() is true (which it is in this case), we never
call get_expr_operands on the rest of the expression, which contains the array
ref. Therefore the operand builder never sees the use of d_5 in the expression,
and chaos breaks out as you have observed.

I dont pay much attention to the semantics of gimple, but either b.c[d_5].i is
not valid gimple, or you have to be prepared to explore the component ref deeper
in get_expr_operands.  I would have expected to trip over this earlier if it was
valid, but what do I know :-)

 

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23297