Re: c/5277: NaN unexpectedly

Re: c/5277: NaN unexpectedly

[bangerth]

Synopsis: NaN unexpectedly

State-Changed-From-To: open->closed
State-Changed-By: bangerth
State-Changed-When: Tue Nov  5 07:59:10 2002
State-Changed-Why:
    Reporter states that this was his error.

http://gcc.gnu.org/cgi-bin/gnatsweb.pl?cmd=view%20audit-trail&database=gcc&pr=5277

Re: c/5277: NaN unexpectedly

[Knox S. Long]

The following reply was made to PR c/5277; it has been noted by GNATS.

From: "Knox S. Long" <long@stsci.edu>
To: gcc-gnats@gcc.gnu.org, gcc-prs@gcc.gnu.org, long@stsci.edu,
	gcc-bugs@gcc.gnu.org
Cc:  
Subject: Re: c/5277: NaN unexpectedly
Date: Sun, 06 Jan 2002 13:33:49 -0500

 http://gcc.gnu.org/cgi-bin/gnatsweb.pl?cmd=view%20audit-trail&database=gcc&pr=5277
 
 Mea culpa!  I now understand the cause of this problem and I don't know whether it 
   is a bug or not.  In all of the cases in which this problem appeared, it was in
 a function which had been declared as a double, e.g.
 
 
 double
 thierry_velocity (x, v)
       double x[];
       double v[];
 {
    int n; ...
 
 
 In this particular case what was being calculated was a velocity v from a position 
 x.  The routine was returning the speed, which is simply a double precision number.
 
 
 In this and all of the other cases where I observed this problem, the routine had 
 not been declared as a double in the calling routine (since I was not actually 
 interested in the returned value).  Once this was done the program, all spurious 
 instances on Nan disappeared.
 
 I have no idea of course whether this is the expected performance of gcc, though 
 certainly the problem is not dire for me if it is. It was lesson to me to use 
 -Wall on anything I really seriously care about.
 
 I apologize to anyone this has inconvenienced.
 
 Sincerely,
 Knox Long
 
 

c/5277: NaN unexpectedly

[long]

>Number:         5277
>Category:       c
>Synopsis:       NaN unexpectedly
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    unassigned
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Fri Jan 04 11:06:00 PST 2002
>Closed-Date:
>Last-Modified:
>Originator:     Knox S. Long
>Release:        gcc-2.96-81
>Organization:
>Environment:
Redhat 7.1  (on a Dell Dimension 1 Ghz machine)
glibc-2.2.4-19.3
>Description:
I am seeing a problem in which simple double precision math
occasionally generates NAN problems.  The probalem is odd
because if on printf's the offending variable, and reprints 
the result, the NAN problem goes away

Here is an example code fragement where this can occur.
double
thierry_velocity (x, v)
     double x[];
     double v[];
{
  int n;
  double r, speed;
  double frac,z;
  double length ();

z=x[0]*x[0]+x[1]*x[1]+x[2]*x[2];
if(sane_check(z)){
        printf("%f\n",z);
        z=x[0]*x[0]+x[1]*x[1]+x[2]*x[2];
        printf("%f\n",z);
}
  r = length (x);

...

where sane_check simply checks to see whether x is finite, e.g.
int
sane_check (x)
     double x;
{
  int i;
  if ((i = finite (x)) == 0)
    {
      Error ("sane_check: %d %e\n", i, x);
      return (-1);
    }
  return (0);
}


There was some discussion of a similar problem on the web 
associated with gcc 2.95.1. and even a reference to a bug
report which I could not locate.

I ran the identical program on a Sun (with gcc 2.95) and
did not see the problem.




>How-To-Repeat:

>Fix:

>Release-Note:
>Audit-Trail:
>Unformatted: