[Bug optimization/12199] New: [3.3-hammer regression] long double miscompilation in gsl/amd64

[Bug optimization/12199] New: [3.3-hammer regression] long double miscompilation in gsl/amd64

[gbeauchesne at mandrakesoft dot com]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=12199

           Summary: [3.3-hammer regression] long double miscompilation in
                    gsl/amd64
           Product: gcc
           Version: 3.3.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: optimization
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: gbeauchesne at mandrakesoft dot com
                CC: aj at suse dot de,gcc-bugs at gcc dot gnu dot org,jh at
                    suse dot cz
  GCC host triplet: x86_64-unknown-linux-gnu

Hi,

The following testcase reduced from gsl is miscompiled only with 3.3-hammer branch as of today. 
This is a  regression from plain 3.3-branch. Sounds like conversion from double to long double 
delta variable gets mad.

The bug is exhausted at -O2 and vanishes at -O1 or with extra -fno-regmove.
---
extern void abort (void);

static double
fabs (double x)
{
  return x < 0 ? -x : x;
}

static void
check (double result, double expected, double relative_error)
{
  int status = -1;

  if (result < 0 || result > 0)
	status = (fabs(result - expected)/fabs(expected) > relative_error);

  if (status)
	abort ();
}

int
main (void)
{
  const int na = 14;

  const double rawa[] =
  {.0421, .0941, .1064, .0242, .1331,
   .0773, .0243, .0815, .1186, .0356,
   .0728, .0999, .0614, .0479};

  double rel = 1e-10;

  {
	const double mean = 0.0728;
	const double expected = 0.00113837428571429;
	long double variance = 0 ;
	int i;

	for (i = 0; i < na; i++)
	  {
		const long double delta = (rawa[i] - mean);
		variance += (delta * delta - variance) / (i + 1);
	  }

    check (variance, expected, rel);
  }

  return 0;
}

[Bug optimization/12199] [3.3-hammer regression] long double miscompilation in gsl/amd64

[pinskia at gcc dot gnu dot org]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=12199


pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   GCC host triplet|x86_64-unknown-linux-gnu    |
 GCC target triplet|                            |x86_64-unknown-linux-gnu
           Keywords|                            |wrong-code


------- Additional Comments From pinskia at gcc dot gnu dot org  2003-09-07 00:52 -------
I think the reason why with -O2, it is exhausted, is because check and (I think fabs, might have 
already) gets inlined because unit-at-a-time is enabled on the 3.3-hammer branch (and the 
mainline also) at -O2 and above.
It would be nice to know if this bug is also on the mainline.
Also what happens if you add __attribute__((__no_inline__)) to the function check, does it still 
create wrong code?

[Bug optimization/12199] [3.3-hammer regression] long double miscompilation in gsl/amd64

[gbeauchesne at mandrakesoft dot com]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=12199



------- Additional Comments From gbeauchesne at mandrakesoft dot com  2003-09-07 07:52 -------
Subject: Re:  [3.3-hammer regression] long double miscompilation in gsl/amd64

Hi,

> I think the reason why with -O2, it is exhausted, is because check and 
> (I think fabs, might have
> already) gets inlined because unit-at-a-time is enabled on the 
> 3.3-hammer branch (and the
> mainline also) at -O2 and above.

Actually it is also exhausted with a 3.3-hammer branch snapshot as far 
as 2003/05/27, i.e. without -funit-at-a-time support by default at -O2. 
I strongly believe fabs() & check() are not the culprit since in the 
original GSL test, they were part of other files I haven't rebuilt when 
reducing the testcase.

> It would be nice to know if this bug is also on the mainline.

It doesn't occur on mainline of yesterday.

> Also what happens if you add __attribute__((__no_inline__)) to the 
> function check, does it still
> create wrong code?

It does.

Bye,
Gwenole.

[Bug optimization/12199] [3.3-hammer regression] long double miscompilation in gsl/amd64

[gbeauchesne at mandrakesoft dot com]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=12199



------- Additional Comments From gbeauchesne at mandrakesoft dot com  2003-09-09 13:53 -------
Subject: Re:  [3.3-hammer regression] long double
 miscompilation in gsl/amd64

The testcase also fails on ia32 with SSE support that way:
gcc -m32 fp.c -O2 -mfpmath=sse -march=k8 (or -march=pentium4)

Actually, the following patch fixes it:
<http://gcc.gnu.org/ml/gcc-patches/2003-06/msg03278.html>

Jan, can you please commit the fix to 3.3-hammer along with the testcase?

Thanks,
Gwenole.

[Bug optimization/12199] [3.3-hammer regression] long double miscompilation in gsl/amd64

[jh at suse dot cz]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=12199



------- Additional Comments From jh at suse dot cz  2003-09-09 15:01 -------
Subject: Re:  [3.3-hammer regression] long double miscompilation in gsl/amd64

> The testcase also fails on ia32 with SSE support that way:
> gcc -m32 fp.c -O2 -mfpmath=sse -march=k8 (or -march=pentium4)
> 
> Actually, the following patch fixes it:
> <http://gcc.gnu.org/ml/gcc-patches/2003-06/msg03278.html>
> 
> Jan, can you please commit the fix to 3.3-hammer along with the testcase?
Sure, working on that.

Honza
> 
> Thanks,
> Gwenole.

[Bug optimization/12199] [3.3-hammer regression] long double miscompilation in gsl/amd64

[zlomek at gcc dot gnu dot org]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=12199


zlomek at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|                            |FIXED


------- Additional Comments From zlomek at gcc dot gnu dot org  2003-10-01 15:06 -------
Honza forgot to close bug report after comminting the fix.

[Bug optimization/12199] [3.3-hammer regression] long double miscompilation in gsl/amd64

[dhazeghi at yahoo dot com]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|---                         |3.4


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=12199

[Bug optimization/12199] [3.3-hammer regression] long double miscompilation in gsl/amd64

[pinskia at gcc dot gnu dot org]

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
   Target Milestone|3.4.0                       |---


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=12199