re PR tree-optimization/58223 (wrong code at -O3 on x86_64-linux-gnu)

2013-08-30  Richard Biener  <rguenther@suse.de>

	PR tree-optimization/58223
	* tree-loop-distribution.c (has_anti_dependence): Rename to ...
	(has_anti_or_output_dependence): ... this and adjust to also
	look for output dependences.
	(mark_nodes_having_upstream_mem_writes): Adjust.
	(rdg_flag_uses): Likewise.

	* gcc.dg/torture/pr58223.c: New testcase.
	* gcc.dg/tree-ssa/ldist-16.c: Flip expected behavior.

From-SVN: r202096

gcc/ChangeLog

 2013-08-30  Richard Biener  <rguenther@suse.de>
 
+	PR tree-optimization/58223
+	* tree-loop-distribution.c (has_anti_dependence): Rename to ...
+	(has_anti_or_output_dependence): ... this and adjust to also
+	look for output dependences.
+	(mark_nodes_having_upstream_mem_writes): Adjust.
+	(rdg_flag_uses): Likewise.
+
+2013-08-30  Richard Biener  <rguenther@suse.de>
+
 	PR tree-optimization/58010
 	* tree-vect-loop.c (vect_create_epilog_for_reduction): Remove
 	assert that we have a loop-closed PHI.

gcc/testsuite/ChangeLog

 2013-08-30  Richard Biener  <rguenther@suse.de>
 
+	PR tree-optimization/58223
+	* gcc.dg/torture/pr58223.c: New testcase.
+	* gcc.dg/tree-ssa/ldist-16.c: Flip expected behavior.
+
+2013-08-30  Richard Biener  <rguenther@suse.de>
+
 	PR tree-optimization/58010
 	* gcc.dg/pr58010.c: New testcase.
 

gcc/testsuite/gcc.dg/torture/pr58223.c

+/* { dg-do run } */
+
+extern void abort (void);
+int a[2], b;
+
+int main ()
+{
+  for (b = 0; b < 2; b++)
+    {
+      a[0] = 1;
+      a[b] = 0;
+    }
+  if (a[0] != 1)
+    abort ();
+  return 0;
+}

gcc/testsuite/gcc.dg/tree-ssa/ldist-16.c

@@ -14,8 +14,8 @@ void foo (int n)
     }
 }
 
-/* We should apply loop distribution and generate a memset (0).  */
+/* We should not apply loop distribution and not generate a memset (0).  */
 
-/* { dg-final { scan-tree-dump "distributed: split to 2" "ldist" } } */
-/* { dg-final { scan-tree-dump-times "generated memset zero" 1 "ldist" } } */
+/* { dg-final { scan-tree-dump "Loop 1 is the same" "ldist" } } */
+/* { dg-final { scan-tree-dump-times "generated memset zero" 0 "ldist" } } */
 /* { dg-final { cleanup-tree-dump "ldist" } } */

gcc/tree-loop-distribution.c

@@ -542,17 +542,19 @@ already_processed_vertex_p (bitmap processed, int v)
 	  || !bitmap_bit_p (remaining_stmts, v));
 }
 
-/* Returns NULL when there is no anti-dependence among the successors
-   of vertex V, otherwise returns the edge with the anti-dep.  */
+/* Returns NULL when there is no anti-dependence or output-dependence
+   among the successors of vertex V, otherwise returns the edge with the
+   dependency.  */
 
 static struct graph_edge *
-has_anti_dependence (struct vertex *v)
+has_anti_or_output_dependence (struct vertex *v)
 {
   struct graph_edge *e;
 
   if (v->succ)
     for (e = v->succ; e; e = e->succ_next)
-      if (RDGE_TYPE (e) == anti_dd)
+      if (RDGE_TYPE (e) == anti_dd
+	  || RDGE_TYPE (e) == output_dd)
 	return e;
 
   return NULL;
@@ -604,11 +606,10 @@ mark_nodes_having_upstream_mem_writes (struct graph *rdg)
 		|| predecessor_has_mem_write (rdg, &(rdg->vertices[x]))
 		/* In anti dependences the read should occur before
 		   the write, this is why both the read and the write
-		   should be placed in the same partition.  */
-		|| has_anti_dependence (&(rdg->vertices[x])))
-	      {
-		bitmap_set_bit (upstream_mem_writes, x);
-	      }
+		   should be placed in the same partition.  In output
+		   dependences the writes order need to be preserved.  */
+		|| has_anti_or_output_dependence (&(rdg->vertices[x])))
+	      bitmap_set_bit (upstream_mem_writes, x);
 	  }
 
 	nodes.release ();
@@ -637,7 +638,7 @@ rdg_flag_uses (struct graph *rdg, int u, partition_t partition, bitmap loops,
   use_operand_p use_p;
   struct vertex *x = &(rdg->vertices[u]);
   gimple stmt = RDGV_STMT (x);
-  struct graph_edge *anti_dep = has_anti_dependence (x);
+  struct graph_edge *anti_dep = has_anti_or_output_dependence (x);
 
   /* Keep in the same partition the destination of an antidependence,
      because this is a store to the exact same location.  Putting this