From patchwork Wed Aug  4 14:52:55 2010
Content-Type: text/plain; charset="utf-8"
MIME-Version: 1.0
Content-Transfer-Encoding: 7bit
Subject: Use rtx_refs_may_alias_p instead of alias_sets_conflict_p in
Date: Wed, 04 Aug 2010 04:52:55 -0000
From: Bingfeng Mei <bmei@broadcom.com>
X-Patchwork-Id: 60857
Message-Id: <7FB04A5C213E9943A72EE127DB74F0ADA6899756DA@SJEXCHCCR02.corp.ad.broadcom.com>
To: "gcc-patches@gcc.gnu.org" <gcc-patches@gcc.gnu.org>
Cc: "Richard Guenther" <richard.guenther@gmail.com>

Hi,
This patch uses rtx_refs_may_alias_p instead of alias_sets_conflict_p to
get more accurate alias set information in walk_mems_2, which is used
in building DDG. In the following example, no false cross-iteration 
dependence is drawn between writing to a[i] and reading b[i] in 
next iteration after the patch. 

Impact for compile-time is minimal because it only applies before modulo
scheduling. Bootstrapped and tested on x86_64-unknown-linux-gnu. 
OK for trunk? 

void foo(int * restrict a, int * restrict b, int n)
{
  int i;
  for(i = 0; i < n; i++)
  {
    a[i] = b[i] * 100;
  }
}


Cheers,
Bingfeng Mei


2010-08-04  Bingfeng Mei  <bmei@broadcom.com>

	* alias.c (walk_mems_2): Call rtx_refs_may_alias_p
	instead of alias_sets_conflict_p to get more accurate 
	alias set information.


Index: alias.c
===================================================================
--- alias.c     (revision 162821)
+++ alias.c     (working copy)
@@ -454,7 +454,7 @@
 {
   if (MEM_P (*x))
     {
-      if (alias_sets_conflict_p (MEM_ALIAS_SET(*x), MEM_ALIAS_SET(mem)))
+      if (rtx_refs_may_alias_p (*x, mem, true))
         return 1;

       return -1;