c/7857: Different behaviour if compiled with optimization!

c/7857: Different behaviour if compiled with optimization!

[jcas]

>Number:         7857
>Category:       c
>Synopsis:       Different behaviour if compiled with optimization!
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    unassigned
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Sun Sep 08 07:46:00 PDT 2002
>Closed-Date:
>Last-Modified:
>Originator:     Jose Santos
>Release:        3.10
>Organization:
>Environment:
Windows XP
>Description:
If code is compiled with any optimization (-O1 to -O3) it has a different behaviour than with compiled with no optimizations.
The code is related with precision errors.With optimizations there are no precision errors, without optimization there's a precision error!
>How-To-Repeat:
Just run the file with -O3 and without -O3 and see that in first case it runs fine(requires a couple of seconds, eg: 15s), in second case it enters an infinite loop(and just ends after looping all integers back to 0)
>Fix:
Don't known...perhaps don't use the sqrt function ?
>Release-Note:
>Audit-Trail:
>Unformatted:
----gnatsweb-attachment----
Content-Type: text/plain; name="bug.c"
Content-Disposition: inline; filename="bug.c"

#include <stdio.h>
#include <math.h>

/*
Compile with -O3 and runs fine
without there's a precision error.
SQ's value has the correct value with -O3 (+1 than without -O3 for n==46611179)

E-mail me at jcas@netcabo.pt
*/
#define NUMPAIRS 10

int main(){
  long long n=2, count = 0;
  for(n=2; count<NUMPAIRS; n++){
    long long n2 =n*n;
    //double sq = sqrt(1+8.0*n2);
    long long sq = sqrt(1+8*n2);
    long long k = -(1 - sq)/2;    
    long long k2 = k*k;  	
    if(n==46611179)
      printf("%lld, %lld\n", sq, k);      
    if(2*n2==k2+k){
      printf("%10lld %10lld\n", n, k);
      count++;
    }
  }
  return 0;
}

Re: c/7857: Different behaviour if compiled with optimization!

[Pop Sébastian]

The following reply was made to PR c/7857; it has been noted by GNATS.

From: =?iso-8859-1?Q?Pop_S=E9bastian?= <pop@gauvain.u-strasbg.fr>
To: jcas@netcabo.pt
Cc: gcc-gnats@gcc.gnu.org, austern@apple.com
Subject: Re: c/7857: Different behaviour if compiled with optimization!
Date: Wed, 9 Oct 2002 19:15:06 +0200

 On Sun, Sep 08, 2002 at 02:36:46PM -0000, jcas@netcabo.pt wrote:
 > /*
 > Compile with -O3 and runs fine
 > without there's a precision error.
 > SQ's value has the correct value with -O3 (+1 than without -O3 for n==46611179)
 > 
 
 Reproduced with 
 gcc version 2.95.4 20011002 (Debian prerelease)
 version gcc 3.3 20021007 (experimental)
 
 
 Compiling the bug report without -O3 gives an error: "Undefined reference to 'sqrt'".
 For completing the link we have to specify the -lm.
 Thus with no optimization, the 'sqrt' function is the one defined in the math library,
 whereas with -O3 the 'sqrt' function is the one defined by GCC... (I think)
 
 I hope that this helps,
 Sebastian.
 
 
 pop@gauvain:~/gcc/main/test$ ../usr/bin/gcc -o pr7857 ./pr7857.c -O3
 pop@gauvain:~/gcc/main/test$ ../usr/bin/gcc -o pr7857 ./pr7857.c
 /tmp/emacs-09607/cc4vzVim.o: dans la fonction `main':
 /tmp/emacs-09607/cc4vzVim.o(.text+0xaf): r\351f\351rence ind\351finie vers `sqrt'
 collect2: ld a retourn\351 1 code d'\351tat d'ex\351cution
 pop@gauvain:~/gcc/main/test$ ../usr/bin/gcc -o pr7857 ./pr7857.c -lm
 pop@gauvain:~/gcc/main/test$