[backport from gcc-4.7/trunk <mikpe@it.uu.se>]

From: "Andreas Krebbel" <krebbel at linux dot vnet dot ibm dot com>
Date: Fri, 4 Mar 2011 13:04:53 +0100
Subject: [PATCH] recog.c: Fix RTX unsharing in change groups
List-Archive: <http://gcc.gnu.org/ml/gcc-patches/>

Hi,

we currently miss to do RTX unsharing in some cases when mixing
validate_change requests with and without rtx unsharing in the same
change group.

Debugging a setjmp testcase on s390 I've seen the following behaviour:

In an insn like A = B + C B is replaced with D in fold_rtx. Unsharing is
requested for D in this replacement (A = D + C). Later on fold_rtx
invokes canonicalize_change_group which decides to swap D and C
without requesting unsharing again (A = C + D). Since the unsharing
for a change group is delayed until confirm_change_group and is bound
to a location rather then the expression C is unshared instead of D
then.

The attached patch prevents that by doing RTX unsharing for all
subsequent changes after finding a single change where unsharing has
been explicitly requested.

Ok for mainline?

From: "Andreas Krebbel" <krebbel at linux dot vnet dot ibm dot com>
Date: Tue, 22 Mar 2011 08:46:32 +0100
Subject: [PATCH] recog.c: Fix RTX unsharing in change groups - v2
List-Archive: <http://gcc.gnu.org/ml/gcc-patches/>

Hi,

this is an alternate solution to the problem described in:
http://gcc.gnu.org/ml/gcc-patches/2011-03/msg00187.html

Although I still prefer the patch above since it would make
validate_change easier to use I understand that it might add
unnecessary unsharing. The unsharing for subsequent changes is only
needed if these changes re-use parts of previous changes and insert it
at a different location in the insn and that's probably a rare case.
Swapping operands like in canonicalize_change_group is one example
which can be easily fixed.

Done with the attached patch.

Ok for mainline?

gcc/

2011-03-22  Andreas Krebbel  <Andreas.Krebbel@de.ibm.com>

	* recog.c (canonicalize_change_group): Use validate_unshare_change.

--- gcc-4.6.0/gcc/recog.c.~1~	2010-11-30 17:36:19.000000000 +0100
+++ gcc-4.6.0/gcc/recog.c	2011-05-14 15:58:02.000000000 +0200
@@ -277,8 +277,8 @@ canonicalize_change_group (rtx insn, rtx
       /* Oops, the caller has made X no longer canonical.
 	 Let's redo the changes in the correct order.  */
       rtx tem = XEXP (x, 0);
-      validate_change (insn, &XEXP (x, 0), XEXP (x, 1), 1);
-      validate_change (insn, &XEXP (x, 1), tem, 1);
+      validate_unshare_change (insn, &XEXP (x, 0), XEXP (x, 1), 1);
+      validate_unshare_change (insn, &XEXP (x, 1), tem, 1);
       return true;
     }
   else