[Bug tree-optimization/64853] [5 Regression] wrong code at -Os and above on x86_64-linux-gnu

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

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64853

--- Comment #4 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
--- gcc/tree-vrp.c.jj    2015-01-28 08:39:51.000000000 +0100
+++ gcc/tree-vrp.c    2015-01-29 10:44:37.395688127 +0100
@@ -7093,14 +7093,14 @@ vrp_valueize_1 (tree name)
   if (TREE_CODE (name) == SSA_NAME)
     {
       value_range_t *vr = get_value_range (name);
-      if (range_int_cst_singleton_p (vr))
-    return vr->min;
       /* If the definition may be simulated again we cannot follow
          this SSA edge as the SSA propagator does not necessarily
      re-visit the use.  */
       gimple def_stmt = SSA_NAME_DEF_STMT (name);
       if (prop_simulate_again_p (def_stmt))
     return NULL_TREE;
+      if (range_int_cst_singleton_p (vr))
+    return vr->min;
     }
   return name;
 }

fixes this, even when a VR is singleton, it might be singleton just because it
is singleton only in the first iteration and not afterwards.
That said, I'd worry that this patch might degrade VRP folding too much.