Bug#322218: marked as done (gcc -O2 optimization bug)
Your message dated Wed, 10 Aug 2005 00:40:08 +0200
with message-id <87ek92loon.fsf@debian.org>
and subject line Bug#322218: gcc -O2 optimization bug
has caused the attached Bug report to be marked as done.
This means that you claim that the problem has been dealt with.
If this is not the case it is now your responsibility to reopen the
Bug report if necessary, and/or fix the problem forthwith.
(NB: If you are a system administrator and have no idea what I am
talking about this indicates a serious mail system misconfiguration
somewhere. Please contact me immediately.)
Debian bug tracking system administrator
(administrator, Debian Bugs database)
--------------------------------------
Received: (at submit) by bugs.debian.org; 9 Aug 2005 19:21:26 +0000
>From houck@space.mit.edu Tue Aug 09 12:21:26 2005
Return-path: <houck@space.mit.edu>
Received: from space.mit.edu [18.75.0.10]
by spohr.debian.org with esmtp (Exim 3.36 1 (Debian))
id 1E2ZfC-0007UU-00; Tue, 09 Aug 2005 12:21:26 -0700
Received: from vex.mit.edu (vex [18.75.2.33])
by space.mit.edu (8.12.10/8.12.10/csr-3.2.5) with ESMTP id j79JLOvN018954
(version=TLSv1/SSLv3 cipher=DHE-RSA-AES256-SHA bits=256 verify=NOT)
for <submit@bugs.debian.org>; Tue, 9 Aug 2005 15:21:25 -0400 (EDT)
Received: from vex.mit.edu (localhost [127.0.0.1])
by vex.mit.edu (8.13.4/8.13.4/Debian-3) with ESMTP id j79JLOEx005737
(version=TLSv1/SSLv3 cipher=DHE-RSA-AES256-SHA bits=256 verify=NOT)
for <submit@bugs.debian.org>; Tue, 9 Aug 2005 15:21:24 -0400
Received: (from houck@localhost)
by vex.mit.edu (8.13.4/8.13.4/Submit) id j79JLO4T005735
for submit@bugs.debian.org; Tue, 9 Aug 2005 15:21:24 -0400
From: John Houck <houck@space.mit.edu>
Date: Tue, 9 Aug 2005 15:21:24 -0400
To: submit@bugs.debian.org
Subject: gcc -O2 optimization bug
Message-ID: <[🔎] 20050809192124.GA5710@vex.mit.edu>
Reply-To: houck@space.mit.edu
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
User-Agent: Mutt/1.5.9i
Delivered-To: submit@bugs.debian.org
X-Spam-Checker-Version: SpamAssassin 2.60-bugs.debian.org_2005_01_02
(1.212-2003-09-23-exp) on spohr.debian.org
X-Spam-Level:
X-Spam-Status: No, hits=-8.0 required=4.0 tests=BAYES_00,HAS_PACKAGE
autolearn=no version=2.60-bugs.debian.org_2005_01_02
Package: gcc
Version: 3.3.5-3
When compiled with -O2, the appended test program goes into an
infinite loop at run-time. When compiled with -O or with no
optimization, it runs correctly, exiting immediately and
generating no output.
I am using Debian GNU/Linux 3.1, with a 2.4.28 kernel and
libc6-2.3.2.ds1-22
#include <math.h>
#include <float.h>
#include <stdio.h>
/* gcc optimization bug
*
* Problem: Program goes into an infinite loop when
* compiled with -O2 optimization.
*
* compile: gcc $OPT bug.c -lm
* run: ./a.out
*
* gcc -O bug.c -lm => works with 2.95.4, 3.3.5, 3.4.4, 4.0.0
* gcc -O2 bug.c -lm => infinite loop with 3.3.5 and 3.4.4
* => works with 2.95.4 and 4.0.0
*/
static int bisection (double (*func)(double, void *), double a, double b, void *cd, double *xp)
{
unsigned int count = 1;
unsigned int bisect_count = 5;
double fa, fb;
double x = a;
if (a > b)
{
double tmp = a; a = b; b = tmp;
}
fa = (*func)(a, cd);
fb = (*func)(b, cd);
if (fa * fb > 0)
{
*xp = a;
return -1;
}
while (b > a)
{
double fx;
if (fb == 0.0)
{
x = b;
break;
}
if (fa == 0.0)
{
x = a;
break;
}
if (count % bisect_count)
{
x = (a*fb - b*fa) / (fb - fa);
if ((x <= a) || (x >= b))
x = 0.5 * (a + b);
}
else
x = 0.5 * (a + b);
if ((x <= a) || (x >= b))
break;
fx = (*func)(x, cd);
count++;
if (fx*fa < 0.0)
{
fb = fx;
b = x;
}
else
{
fa = fx;
a = x;
}
}
*xp = x;
return 0;
}
static double f (double x, void *cd)
{
(void) cd;
return x - cos(x);
}
int main (void)
{
double x;
(void) bisection (&f, 0.0, 1.0, NULL, &x);
if (fabs(f(x,NULL)) > DBL_EPSILON)
fprintf (stdout, "Error\n");
return 0;
}
---------------------------------------
Received: (at 322218-done) by bugs.debian.org; 9 Aug 2005 22:40:40 +0000
>From falk@debian.org Tue Aug 09 15:40:40 2005
Return-path: <falk@debian.org>
Received: from smtp06.web.de [217.72.192.224]
by spohr.debian.org with esmtp (Exim 3.36 1 (Debian))
id 1E2cm0-0007qM-00; Tue, 09 Aug 2005 15:40:40 -0700
Received: from [84.59.228.147] (helo=juist)
by smtp06.web.de with asmtp (TLSv1:DES-CBC3-SHA:168)
(WEB.DE 4.105 #314)
id 1E2clU-0002zX-00
for 322218-done@bugs.debian.org; Wed, 10 Aug 2005 00:40:08 +0200
Received: from falk by juist with local (Exim 4.52)
id 1E2clU-0007du-Sy
for 322218-done@bugs.debian.org; Wed, 10 Aug 2005 00:40:08 +0200
To: 322218-done@bugs.debian.org
Subject: Re: Bug#322218: gcc -O2 optimization bug
References: <[🔎] 20050809192124.GA5710@vex.mit.edu>
From: Falk Hueffner <falk@debian.org>
X-Face: "iUeUu$b*W_"w?tV83Y3*r:`rh&dRv}$YnZ3,LVeCZSYVuf[Gpo*5%_=/\_!gc_,SS}[~xZ
wY77I-M)xHIx:2f56g%/`SOw"Dx%4Xq0&f\Tj~>|QR|vGlU}TBYhiG(K:2<T^
Date: Wed, 10 Aug 2005 00:40:08 +0200
In-Reply-To: <[🔎] 20050809192124.GA5710@vex.mit.edu> (John Houck's message of
"Tue, 9 Aug 2005 15:21:24 -0400")
Message-ID: <87ek92loon.fsf@debian.org>
User-Agent: Gnus/5.1006 (Gnus v5.10.6) XEmacs/21.5 (cilantro, linux)
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Sender: falk@debian.org
X-Sender: falk.hueffner@web.de
Delivered-To: 322218-done@bugs.debian.org
X-Spam-Checker-Version: SpamAssassin 2.60-bugs.debian.org_2005_01_02
(1.212-2003-09-23-exp) on spohr.debian.org
X-Spam-Level:
X-Spam-Status: No, hits=-6.0 required=4.0 tests=BAYES_00,HAS_BUG_NUMBER
autolearn=no version=2.60-bugs.debian.org_2005_01_02
John Houck <houck@space.mit.edu> writes:
> When compiled with -O2, the appended test program goes into an
> infinite loop at run-time. When compiled with -O or with no
> optimization, it runs correctly, exiting immediately and generating
> no output.
This is caused by excess precision of the x87 FPU and therefore
slightly different rounding depending on optimizations. Your options
are:
* Write more robust code
* Use SSA floating point
* Use -ffloat-store
* Don't use x86
--
Falk
Reply to: