From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <gcc-return-106398-listarch-gcc=gcc.gnu.org@gcc.gnu.org>
Received: (qmail 7188 invoked by alias); 10 Dec 2004 15:36:46 -0000
Mailing-List: contact gcc-help@gcc.gnu.org; run by ezmlm
Precedence: bulk
List-Archive: <http://gcc.gnu.org/ml/gcc/>
List-Post: <mailto:gcc@gcc.gnu.org>
List-Help: <http://gcc.gnu.org/ml/>
Sender: gcc-owner@gcc.gnu.org
Received: (qmail 6717 invoked from network); 10 Dec 2004 15:36:36 -0000
Received: from unknown (HELO alwazn.tat.physik.uni-tuebingen.de) (134.2.170.157)
  by sourceware.org with SMTP; 10 Dec 2004 15:36:36 -0000
Received: by alwazn.tat.physik.uni-tuebingen.de (Postfix, from userid 30023)
	id E02133FA1; Fri, 10 Dec 2004 16:36:35 +0100 (CET)
Received: from localhost (localhost [127.0.0.1])
	by alwazn.tat.physik.uni-tuebingen.de (Postfix) with ESMTP id D7F3A76988
	for <gcc@gcc.gnu.org>; Fri, 10 Dec 2004 16:36:35 +0100 (CET)
Date: Fri, 10 Dec 2004 15:36:00 -0000
From: Richard Guenther <rguenth@tat.physik.uni-tuebingen.de>
To: gcc@gcc.gnu.org
Subject: Simple loops not interchanged?
Message-ID: <Pine.LNX.4.44.0412101616530.4584-100000@alwazn.tat.physik.uni-tuebingen.de>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII
X-SW-Source: 2004-12/txt/msg00360.txt.bz2

Hi!

I expected -ftree-loop-linear to exchange loops in

double foo(double *a)
{
        int i,j;
        double r = 0.0;
	for (i=0; i<8; ++i)
                for (j=0; j<8; ++j)
			r += a[j*8+i];
        return r;
}

but it tells me (regardless of loop order) that
"Won't transform loop. Optimal transform is the identity transform"
which I cannot believe, obviously.

>From further experiments, if I don't pass a as a function parameter,
but make its declaration globally available, the transform succeeds.

The main difference in tree dumps before the transformation attempt
is the data access:

 <L1>:;
   D.1129_12 = j_4 * 8;
   D.1130_13 = i_3 + D.1129_12;
-  D.1131_14 = (unsigned int) D.1130_13;
-  D.1132_15 = D.1131_14 * 8;
-  D.1133_16 = (double *) D.1132_15;
-  D.1134_18 = D.1133_16 + a_17;
-  D.1135_19 = *D.1134_18;
+  D.1131_15 = a[D.1130_13];
   r_6 = r_5 + D.1131_15;
   j_17 = j_4 + 1;
   if (N_7 > j_17) goto <L16>; else goto <L3>;

which is just D.1131_15 = a[D.1130_13] in case of the global.  Note
that is the difference starting from the generic tree dump.

What's going wrong here?

Richard.

--
Richard Guenther <richard dot guenther at uni-tuebingen dot de>
WWW: http://www.tat.physik.uni-tuebingen.de/~rguenth/