normalize Sync node by sorting their inputs

This also makes the compiler alot more deterministic avoiding, lots of
"random" jumping around in series of stores and loads.

diff --git a/ir/ir/iropt.c b/ir/ir/iropt.c
index 44c40d4..bbcfc0d 100644 (file)
--- a/ir/ir/iropt.c
+++ b/ir/ir/iropt.c
@@ -6194,7 +6194,8 @@ static ir_node *transform_node_Sync(ir_node *n)
                                        ++arity;
                                        break;
                                }
-                               if (get_Sync_pred(n, k) == pred_pred) break;
+                               if (get_Sync_pred(n, k) == pred_pred)
+                                       break;
                        }
                }
        }
@@ -6820,6 +6821,15 @@ void del_identities(ir_graph *irg)
                del_pset(irg->value_table);
 }
 
+static int cmp_node_nr(const void *a, const void *b)
+{
+       ir_node **p1 = (ir_node**)a;
+       ir_node **p2 = (ir_node**)b;
+       long      n1 = get_irn_node_nr(*p1);
+       long      n2 = get_irn_node_nr(*p2);
+       return (n1>n2) - (n1<n2);
+}
+
 void ir_normalize_node(ir_node *n)
 {
        if (is_op_commutative(get_irn_op(n))) {
@@ -6835,6 +6845,30 @@ void ir_normalize_node(ir_node *n)
                        set_binop_right(n, l);
                        hook_normalize(n);
                }
+       } else if (is_Sync(n)) {
+               /* we assume that most of the time the inputs of a Sync node are already
+                * sorted, so check this first as a shortcut */
+               bool           ins_sorted = true;
+               int            arity      = get_irn_arity(n);
+               const ir_node *last       = get_irn_n(n, 0);
+               int      i;
+               for (i = 1; i < arity; ++i) {
+                       const ir_node *node = get_irn_n(n, i);
+                       if (get_irn_node_nr(node) < get_irn_node_nr(last)) {
+                               ins_sorted = false;
+                               break;
+                       }
+                       last = node;
+               }
+
+               if (!ins_sorted) {
+                       ir_node **ins     = get_irn_in(n)+1;
+                       ir_node **new_ins = XMALLOCN(ir_node*, arity);
+                       memcpy(new_ins, ins, arity*sizeof(ins[0]));
+                       qsort(new_ins, arity, sizeof(new_ins[0]), cmp_node_nr);
+                       set_irn_in(n, arity, new_ins);
+                       xfree(new_ins);
+               }
        }
 }