[Bug rtl-optimization/25654] [4.0/4.1/4.2 Regression] RTL alias analysis unprepared to handle stack slot sharing

[steven at gcc dot gnu dot org]

------- Comment #2 from steven at gcc dot gnu dot org  2006-01-13 21:30 -------
I wonder if this problem can also be triggered without using two variables of
the same union type.  There is code in add_alias_set_conflicts to avoid the
situation we're running into:

static void
add_alias_set_conflicts (void)
{
  size_t i, j, n = stack_vars_num;

  for (i = 0; i < n; ++i)
    {
      tree type_i = TREE_TYPE (stack_vars[i].decl);
      bool aggr_i = AGGREGATE_TYPE_P (type_i);

      for (j = 0; j < i; ++j)
        {
          tree type_j = TREE_TYPE (stack_vars[j].decl);
          bool aggr_j = AGGREGATE_TYPE_P (type_j);
          if (aggr_i != aggr_j || !objects_must_conflict_p (type_i, type_j))
            add_stack_var_conflict (i, j);
        }
    }
}

but if you have two stack variables of the same union type, aggr_i == aggr_j,
and type_i == type_j so objects_must_conflict_p returns true.  So perhaps if
type_i and type_j are unions and type_i == type_j, we should also do an
add_stack_var_conflict...?


-- 

steven at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |hubicka at gcc dot gnu dot
                   |                            |org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=25654