http://gcc.gnu.org/bugzilla/show_bug.cgi?id=35290

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED
         AssignedTo|unassigned at gcc dot       |rguenth at gcc dot gnu.org
                   |gnu.org                     |

--- Comment #4 from Richard Guenther <rguenth at gcc dot gnu.org> 2010-11-10 
12:32:16 UTC ---
Well, it's not working on the tree level to make it cheaper:

  /* We deliberately stop on clobbering statements and not only on
     killing ones to make walking cheaper.  Otherwise we can just
     continue walking until both stores have equal reference trees.  */
  while (!stmt_may_clobber_ref_p (temp, gimple_assign_lhs (stmt)));

changing that to

  while (!stmt_kills_ref_p (temp, gimple_assign_lhs (stmt)));

will make it work (and slower).

Also needs (possibly unsafe in general if used cross loop iteration)

Index: gcc/tree-ssa-alias.c
===================================================================
--- gcc/tree-ssa-alias.c        (revision 166526)
+++ gcc/tree-ssa-alias.c        (working copy)
@@ -1577,7 +1577,12 @@ stmt_kills_ref_p_1 (gimple stmt, ao_ref
       base = get_ref_base_and_extent (lhs, &offset, &size, &max_size);
       /* We can get MEM[symbol: sZ, index: D.8862_1] here,
         so base == ref->base does not always hold.  */
-      if (base == ref->base)
+      if (base == ref->base
+         || (TREE_CODE (base) == MEM_REF
+             && TREE_CODE (ref->base) == MEM_REF
+             && TREE_OPERAND (base, 0) == TREE_OPERAND (ref->base, 0)
+             && tree_int_cst_equal (TREE_OPERAND (base, 1),
+                                    TREE_OPERAND (ref->base, 1))))
        {
          /* For a must-alias check we need to be able to constrain
             the accesses properly.  */

Reply via email to