[backport proposed but not yet applied 4.9/trunk patch for PR57214
 regression from PR56982 <mikpe@it.uu.se>]

List-Archive: <http://gcc.gnu.org/ml/gcc-patches/>
Date: Fri, 10 May 2013 10:34:37 +0200 (CEST)
From: Richard Biener <rguenther at suse dot de>
Subject: [PATCH] Fix PR57214

This fixes PR57214 - the simple constant propagation pass we perform
over unrolled loop bodies does not properly avoid propagating across
abnormal edges.

Bootstrap & regtest running on x86_64-unknown-linux-gnu, I'll install
this on trunk and the 4.8 branch where the bug is latent.

Richard.

2013-05-10  Richard Biener  <rguenther@suse.de>

	PR tree-optimization/57214
	* tree-ssa-loop-ivcanon.c (propagate_constants_for_unrolling): Do
	not propagate from SSA names that occur in abnormal PHI nodes.

--- gcc-4.8.0/gcc/testsuite/gcc.dg/torture/pr57214.c.~1~	1970-01-01 01:00:00.000000000 +0100
+++ gcc-4.8.0/gcc/testsuite/gcc.dg/torture/pr57214.c	2013-05-10 12:44:36.262956885 +0200
@@ -0,0 +1,25 @@
+/* { dg-do compile } */
+
+extern int baz (void);
+extern int foo (void) __attribute__ ((returns_twice));
+
+void
+bar (_Bool b)
+{
+  int buf[1];
+  while (1)
+    {
+      _Bool x = 1;
+      if (b)
+	baz ();
+      b = 1;
+      baz ();
+      x = 0;
+      int i;
+      while (buf[i] && i)
+	i++;
+      foo ();
+      if (!x)
+	b = 0;
+    }
+}
--- gcc-4.8.0/gcc/tree-ssa-loop-ivcanon.c.~1~	2013-03-01 23:56:18.000000000 +0100
+++ gcc-4.8.0/gcc/tree-ssa-loop-ivcanon.c	2013-05-10 12:44:36.262956885 +0200
@@ -1085,8 +1085,9 @@ propagate_constants_for_unrolling (basic
       tree lhs;
 
       if (is_gimple_assign (stmt)
+	  && gimple_assign_rhs_code (stmt) == INTEGER_CST
 	  && (lhs = gimple_assign_lhs (stmt), TREE_CODE (lhs) == SSA_NAME)
-	  && gimple_assign_rhs_code (stmt) == INTEGER_CST)
+	  && !SSA_NAME_OCCURS_IN_ABNORMAL_PHI (lhs))
 	{
 	  propagate_into_all_uses (lhs, gimple_assign_rhs1 (stmt));
 	  gsi_remove (&gsi, true);