Debian Bug report logs - #36876
[PR c/6900] Floating point behaviour of gcc on pentium

Toggle useless messages

Message #5 received at submit@bugs.debian.org (full text, mbox, reply):

From: Ghanashyam Date <shyam@imsc.ernet.in>

To: submit@bugs.debian.org

Cc: sure@imsc.ernet.in

Subject: Erratic behaviour of gcc/egcc on pentium

Date: Thu, 29 Apr 1999 13:04:16 +0530 (IST)

The following program gives erratic/correct answers under different
conditions. The program is self explanatory.

The expected behaviour is to get the value of the variable 'q' to equal the
value of the variable 'shift' . Erratic behaviour refers to deviation from
this equality. The deviation can be controlled by increasing/decreasing the
value of 'factor' . The variable take GENERIC values.

The expected behaviour is seen on SGI and SPARC machines with their native
compilers and also with gcc (2.7x), with or without optimization.

On pentium with gcc/egcs the following happens:

Without optimization:	only those x which are powers of 2 give q = shift.
With optimization:	all give q = shift.

The deviations are NOT an artifact of printf formats. If subsequents
iterations are done with further code, these deviations actually get
amplified, and thus are genuine. In a non trivial code, this uncontrolled
error accumulation is a disaster.

While it is true that beyond 15 decimal places, one has garbage bits in the
internal representation, a suitable round-off/truncation etc should discard
these bits correctly and consistantly for generic values. (Other platforms
mentioned above seem to take care of this so also optimization on pentium
platform.)

In another context, with a somewhat different program, the expected behaviour
resulted WITH-OUT optimization but WITH optimization, the behaviour was
erratic.

Details of versions etc:
------------------------

$ uname -a ;
Linux aspc30 2.0.36 #1 Mon Mar 29 12:46:14 IST 1999 i586 unknown

$ egcc -v ;
Reading specs from /usr/lib/gcc-lib/i486-linux/egcs-2.90.29/specs
gcc version egcs-2.90.29 980515 (egcs-1.0.3 release)

$ gcc -v ;
Reading specs from /usr/lib/gcc-lib/i486-linux/2.7.2.3/specs
gcc version 2.7.2.3

$ ld -v ;
GNU ld version 2.9.1 (with BFD 2.9.1)

$ ldd /usr/bin/gcc
	/lib/nfslock.so.0 => /lib/nfslock.so.0 (0x4000d000)
	libc.so.6 => /lib/libc.so.6 (0x40013000)
	/lib/ld-linux.so.2 => /lib/ld-linux.so.2 (0x40000000)

$ ldd /usr/bin/egcc
	/lib/nfslock.so.0 => /lib/nfslock.so.0 (0x4000d000)
	libc.so.6 => /lib/libc.so.6 (0x40013000)
	/lib/ld-linux.so.2 => /lib/ld-linux.so.2 (0x40000000)

--------------------------------------------------------
PLEASE ADVISE.
	-G. Date (shyam@imsc.ernet.in)
--------------------------------------------------------

/*---------------------------------------------------------------------*/
# include <stdio.h>

main(argc, argv)
int argc;
char **argv ;

{
	double q, q0, x, factor, shift;

	factor = 100000000.000001L ; 	/* Choose any typical fraction */
	shift = 0.0L ;

	for(x = 0.0L; x < 257.0L; x = x + 1.0L )	{
		q0 = (factor)*x ;
		q =  q0 - factor*x + shift ;
		if( q == shift )
			printf(" TRUE:\t%20.8E%20.8E%20.8E\n", x, q0, q);
		else		
			printf("FALSE:\t%20.8E%20.8E%20.8E\n", x, q0, q);
	}
}

/*---------------------------------------------------------------------*/

Message #12 received at 36876@bugs.debian.org (full text, mbox, reply):

From: Matthias Klose <doko@cs.tu-berlin.de>

To: egcs-bugs@egcs.cygnus.com

Cc: 36876@bugs.debian.org, sure@imsc.ernet.in

Subject: Erratic behaviour of gcc/egcc on pentium

Date: Sun, 16 May 1999 17:36:48 +0200 (MET DST)

[This bug was reported to me as a co-maintainer of the Debian GNU/Linux EGCS
packages. Please Cc: 36876@bugs.debian.org in your response so it gets
archived in the Debian bugtracking system (http://www.debian.org/Bugs/).
Please Cc me as I'm not on the egcs-bugs list.]

The following program gives erratic/correct answers under different
conditions. The program is self explanatory.

The expected behaviour is to get the value of the variable 'q' to equal the
value of the variable 'shift' . Erratic behaviour refers to deviation from
this equality. The deviation can be controlled by increasing/decreasing the
value of 'factor' . The variable take GENERIC values.

The expected behaviour is seen on SGI and SPARC machines with their native
compilers and also with gcc (2.7x), with or without optimization.

On pentium with gcc/egcs the following happens:

Without optimization:   only those x which are powers of 2 give q = shift.
With optimization:      all give q = shift.

The deviations are NOT an artifact of printf formats. If subsequents
iterations are done with further code, these deviations actually get
amplified, and thus are genuine. In a non trivial code, this uncontrolled
error accumulation is a disaster.

While it is true that beyond 15 decimal places, one has garbage bits in the
internal representation, a suitable round-off/truncation etc should discard
these bits correctly and consistantly for generic values. (Other platforms
mentioned above seem to take care of this so also optimization on pentium
platform.)

In another context, with a somewhat different program, the expected behaviour
resulted WITH-OUT optimization but WITH optimization, the behaviour was
erratic.

Details of versions etc:
------------------------

$ uname -a ;
Linux aspc30 2.0.36 #1 Mon Mar 29 12:46:14 IST 1999 i586 unknown

$ egcc -v ;
Reading specs from /usr/lib/gcc-lib/i486-linux/egcs-2.90.29/specs
gcc version egcs-2.90.29 980515 (egcs-1.0.3 release)

$ gcc -v ;
Reading specs from /usr/lib/gcc-lib/i486-linux/2.7.2.3/specs
gcc version 2.7.2.3

$ ld -v ;
GNU ld version 2.9.1 (with BFD 2.9.1)

$ ldd /usr/bin/gcc
        /lib/nfslock.so.0 => /lib/nfslock.so.0 (0x4000d000)
        libc.so.6 => /lib/libc.so.6 (0x40013000)
        /lib/ld-linux.so.2 => /lib/ld-linux.so.2 (0x40000000)

$ ldd /usr/bin/egcc
        /lib/nfslock.so.0 => /lib/nfslock.so.0 (0x4000d000)
        libc.so.6 => /lib/libc.so.6 (0x40013000)
        /lib/ld-linux.so.2 => /lib/ld-linux.so.2 (0x40000000)

--------------------------------------------------------
PLEASE ADVISE.
        -G. Date (shyam@imsc.ernet.in)
--------------------------------------------------------

/*---------------------------------------------------------------------*/
# include <stdio.h>

main(argc, argv)
int argc;
char **argv ;

{
        double q, q0, x, factor, shift;

        factor = 100000000.000001L ;    /* Choose any typical fraction */
        shift = 0.0L ;

        for(x = 0.0L; x < 257.0L; x = x + 1.0L )        {
                q0 = (factor)*x ;
                q =  q0 - factor*x + shift ;
                if( q == shift )
                        printf(" TRUE:\t%20.8E%20.8E%20.8E\n", x, q0, q);
                else            
                        printf("FALSE:\t%20.8E%20.8E%20.8E\n", x, q0, q);
        }
}

/*---------------------------------------------------------------------*/

Message #17 received at 36876@bugs.debian.org (full text, mbox, reply):

From: Matthias Klose <doko@cs.tu-berlin.de>

To: 36876@bugs.debian.org, Ghanashyam Date <shyam@imsc.ernet.in>

Subject: forwarded message from Tim Schmielau

Date: Sun, 16 May 1999 19:25:56 +0200 (MET DST)

[Message part 1 (message/rfc822, inline)]

From: Tim Schmielau <tim@physik3.uni-rostock.de>

To: Matthias Klose <doko@cs.tu-berlin.de>

Cc: egcs-bugs@egcs.cygnus.com

Subject: Re: Erratic behaviour of gcc/egcc on pentium

Date: Sun, 16 May 1999 19:05:07 +0200 (MEST)

This is not a bug in egcs, but a general problem of floating point
arithmetics. If you print out (q/q0), you see that the difference of
"q" and "shift" is always zero, to within machine precision.

Tim Schmielau (tim@physik3.uni-rostock.de)

Message #24 received at 36876@bugs.debian.org (full text, mbox, reply):

From: Matthias Klose <doko@cs.tu-berlin.de>

To: gcc@gcc.gnu.org

Cc: 36876@bugs.debian.org, 40189@bugs.debian.org

Subject: floating point arithmetics on ix86

Date: Fri, 5 Nov 1999 15:54:58 +0100 (MET)

I am looking for some pointers / FAQ's how to use fp arithmetics on
the ix86 architecture. The background are "bug" reports as found on
http://www.debian.org/Bugs/db/36/36876.html and
http://www.debian.org/Bugs/db/40/40189.html (or appended here).
AFAIK these are issues with the machine precision, which could be
described in a FAQ added to the binary packages.

First Report:
=============

The following program inconsistently chooses minimal
array elements in case of equal elements being present.

Here is the code (ttr.c):
------------------------
#include <stdio.h>
#include <math.h>


int
main() {

  double
a[4]={0.006776,2.5,3.0,0.006776},w[4]={0.000030,3.7,3.0,0.000030},t,amin;
  int ir,i,nr=4;
  double tol=9.31322574615479e-10*3.0;

  for (i=0,ir=-1,amin=DBL_MAX;i<nr;i++) 
    if (a[i]>tol) {
      t=w[i]/a[i];
      if (t<amin || ir==-1) {
        amin=t;
        ir=i;
      }
    }
  
  printf("ir=%d\n",ir);

  for (i=0,ir=-1,amin=DBL_MAX;i<nr;i++) 
    if (a[i]>tol && ((t=w[i]/a[i])<amin || ir==-1)) {
        amin=t;
        ir=i;
      }
  
  printf("ir=%d\n",ir);

  exit(0);

}
---------------  
and here is the execution:
---------------
intech19:~$ egcc -g -Wall ttr.c -o ttr -lm
intech19:~$ ./ttr
ir=0
ir=3
-----------------

Second report:
==============

The following program gives erratic/correct answers under different
conditions. The program is self explanatory.

The expected behaviour is to get the value of the variable 'q' to
equal the value of the variable 'shift' . Erratic behaviour refers to
deviation from this equality. The deviation can be controlled by
increasing/decreasing the value of 'factor' . The variable take
GENERIC values.

The expected behaviour is seen on SGI and SPARC machines with their
native compilers and also with gcc (2.7x), with or without
optimization.

On pentium with gcc/egcs the following happens:

Without optimization:   only those x which are powers of 2 give q = shift.
With optimization:      all give q = shift.

The deviations are NOT an artifact of printf formats. If subsequents
iterations are done with further code, these deviations actually get
amplified, and thus are genuine. In a non trivial code, this
uncontrolled error accumulation is a disaster.

While it is true that beyond 15 decimal places, one has garbage bits
in the internal representation, a suitable round-off/truncation etc
should discard these bits correctly and consistantly for generic
values. (Other platforms mentioned above seem to take care of this so
also optimization on pentium platform.)

In another context, with a somewhat different program, the expected
behaviour resulted WITH-OUT optimization but WITH optimization, the
behaviour was erratic.

/*---------------------------------------------------------------------*/
# include <stdio.h>

main(argc, argv)
int argc;
char **argv ;

{
        double q, q0, x, factor, shift;

        factor = 100000000.000001L ;    /* Choose any typical fraction */
        shift = 0.0L ;

        for(x = 0.0L; x < 257.0L; x = x + 1.0L )        {
                q0 = (factor)*x ;
                q =  q0 - factor*x + shift ;
                if( q == shift )
                        printf(" TRUE:\t%20.8E%20.8E%20.8E\n", x, q0, q);
                else            
                        printf("FALSE:\t%20.8E%20.8E%20.8E\n", x, q0, q);
        }
}

/*---------------------------------------------------------------------*/

Message #29 received at 36876@bugs.debian.org (full text, mbox, reply):

From: N8TM@aol.com

To: doko@cs.tu-berlin.de, gcc@gcc.gnu.org

Cc: 36876@bugs.debian.org, 40189@bugs.debian.org

Subject: Re: floating point arithmetics on ix86

Date: Fri, 5 Nov 1999 21:13:43 EST

re "First Report" [reported cases where compiler appears to be reversing 
sense of comparison between otherwise equivalent test cases]

I assumed that #include <float.h> is intended.

If so, gcc -Os -march=pentiumpro -V2.95.2 compiliation produces at execution:

ir=0
ir=0

Same with -O2.  Same with egcs-19991102, even with -ffast-math added, which 
would allow for the reported variation.  This result appears to be the only 
correct one, unless specifying a relaxation of standards, as -ffast-math 
apparently does.

re: Second report [case where an assignment sometimes rounds off to declared 
precision, sometimes not, according to chosen level of optimization]

               q0 = (factor)*x ;
                q =  q0 - factor*x + shift ;


Both egcs-2.96 and gcc-2.95.2 exhibit the reported behavior, where q0 is 
rounded to double when not optimized, so that q0 - factor*x is non-zero.  
When optimized, the same expression turns up zero.  From my personal point of 
view, I might expect the small change

                q =  q0 - (double)(factor*x) + shift ;

to make a difference, but it does not.

BTW, it is even more difficult to predict the result of this when running on 
machines without normal extended precision but with a multiply-accumulate 
instruction which does not round off between multiplication and subtraction.  
I assume that the C9X draft shows the committee's consensus that should be 
modified with the new math library functions if one wishes to produce 
predictable results in future versions of C in such a case.  

Tim
tprince@computer.org

Message #34 received at 36876@bugs.debian.org (full text, mbox, reply):

From: gcc-gnats@gcc.gnu.org

To: 36876@bugs.debian.org

Subject: Re: c/6900: Floating point behaviour of gcc on pentium

Date: 1 Jun 2002 21:06:00 -0000

Thank you very much for your problem report.
It has the internal identification `c/6900'.
The individual assigned to look at your
report is: unassigned. 

>Category:       c
>Responsible:    unassigned
>Synopsis:       Floating point behaviour of gcc on pentium
>Arrival-Date:   Sat Jun 01 14:06:00 PDT 2002

Message #43 received at 36876-quiet@bugs.debian.org (full text, mbox, reply):

From: Matthias Klose <doko@klose.in-berlin.de>

To: control@bugs.debian.org

Cc: 36876-quiet@bugs.debian.org

Subject: gcc: submitted Debian report #36876 to gcc-gnats as PR c/6900

Date: Sat, 01 Jun 2002 23:16:08 +0200

# submitted Debian report #36876 to gcc-gnats as PR c/6900
# http://gcc.gnu.org/cgi-bin/gnatsweb.pl?cmd=view&pr=c/6900&database=gcc

forwarded 36876 gcc-gnats@gcc.gnu.org
retitle 36876 [PR c/6900] Floating point behaviour of gcc on pentium
thanks

Message #48 received at 36876@bugs.debian.org (full text, mbox, reply):

From: "Christian Ehrhardt" <ehrhardt@mathematik.uni-ulm.de>

To: gcc-gnats@gcc.gnu.org, gcc-prs@gcc.gnu.org, tprince@computer.org, sure@imsc.ernet.in, gcc-bugs@gcc.gnu.org, nobody@gcc.gnu.org, 36876@bugs.debian.org, shyam@imsc.ernet.in

Subject: Re: c/6900: Floating point behaviour of gcc on pentium

Date: Wed, 4 Dec 2002 15:51:45 +0100

Hi,

This is not a bug. Comparing floating point numbers with == may yield
unexpected results due to rounding errors. Someone please close this PR.

   regards  Christian

http://gcc.gnu.org/cgi-bin/gnatsweb.pl?cmd=view%20audit-trail&database=gcc&pr=6900

Message #53 received at 36876@bugs.debian.org (full text, mbox, reply):

From: bangerth@dealii.org

To: 36876@bugs.debian.org, gcc-bugs@gcc.gnu.org, gcc-prs@gcc.gnu.org, nobody@gcc.gnu.org, shyam@imsc.ernet.in, sure@imsc.ernet.in, tprince@computer.org

Subject: Re: c/6900: Floating point behaviour of gcc on pentium

Date: 4 Dec 2002 15:02:02 -0000

Synopsis: Floating point behaviour of gcc on pentium

State-Changed-From-To: open->closed
State-Changed-By: bangerth
State-Changed-When: Wed Dec  4 07:01:59 2002
State-Changed-Why:
    Christian explained why.

http://gcc.gnu.org/cgi-bin/gnatsweb.pl?cmd=view%20audit-trail&database=gcc&pr=6900

Message #58 received at 36876-done@bugs.debian.org (full text, mbox, reply):

From: Matthias Klose <doko@cs.tu-berlin.de>

To: 36876-done@bugs.debian.org

Subject: Re: Bug#36876: c/6900: Floating point behaviour of gcc on pentium

Date: Fri, 27 Dec 2002 14:40:31 +0100

closing the Debian report as well.

Christian Ehrhardt writes:
> 
> Hi,
> 
> This is not a bug. Comparing floating point numbers with == may yield
> unexpected results due to rounding errors. Someone please close this PR.
> 
>    regards  Christian
> 
> http://gcc.gnu.org/cgi-bin/gnatsweb.pl?cmd=view%20audit-trail&database=gcc&pr=6900

Send a report that this bug log contains spam.