svn commit: r260361 - head/contrib/gcc

Pedro F. Giffuni pfg at FreeBSD.org
Mon Jan 6 00:52:40 UTC 2014


Author: pfg
Date: Mon Jan  6 00:52:39 2014
New Revision: 260361
URL: http://svnweb.freebsd.org/changeset/base/260361

Log:
  Fix optimization bug.
  
  GCC-PR rtl-optimization/34628
  	* combine.c (try_combine): Stop and undo after the first combination
  	if an autoincrement side-effect on the first insn has effectively
  	been lost.
  
  The issue was detected in OpenBSD but their fix was not very good. Huge
  thanks to the upstream author, Eric Botcazou, for permitting the use of
  this patch under GPLv2.
  
  MFC after:	5 days

Modified:
  head/contrib/gcc/combine.c

Modified: head/contrib/gcc/combine.c
==============================================================================
--- head/contrib/gcc/combine.c	Mon Jan  6 00:15:19 2014	(r260360)
+++ head/contrib/gcc/combine.c	Mon Jan  6 00:52:39 2014	(r260361)
@@ -2370,12 +2370,17 @@ try_combine (rtx i3, rtx i2, rtx i1, int
 
   if (i1 && GET_CODE (newpat) != CLOBBER)
     {
-      /* Before we can do this substitution, we must redo the test done
-	 above (see detailed comments there) that ensures  that I1DEST
-	 isn't mentioned in any SETs in NEWPAT that are field assignments.  */
-
-      if (! combinable_i3pat (NULL_RTX, &newpat, i1dest, NULL_RTX,
-			      0, (rtx*) 0))
+      /* Check that an autoincrement side-effect on I1 has not been lost.
+	 This happens if I1DEST is mentioned in I2 and dies there, and
+	 has disappeared from the new pattern.  */
+      if ((FIND_REG_INC_NOTE (i1, NULL_RTX) != 0
+	   && !i1_feeds_i3
+	   && dead_or_set_p (i2, i1dest)
+	   && !reg_overlap_mentioned_p (i1dest, newpat))
+	  /* Before we can do this substitution, we must redo the test done
+	     above (see detailed comments there) that ensures  that I1DEST
+	     isn't mentioned in any SETs in NEWPAT that are field assignments.  */
+          || !combinable_i3pat (NULL_RTX, &newpat, i1dest, NULL_RTX, 0, 0))
 	{
 	  undo_all ();
 	  return 0;


More information about the svn-src-all mailing list