[Bug tree-optimization/57124] New: 254.gap@spec2000 got miscompare after r198413

[Bug tree-optimization/57124] New: 254.gap@spec2000 got miscompare after r198413

[izamyatin at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=57124

             Bug #: 57124
           Summary: 254.gap@spec2000 got miscompare after r198413
    Classification: Unclassified
           Product: gcc
           Version: 4.9.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: izamyatin@gmail.com


Happens for example for option set "-O2 -ffast-math -march=corei7".

Unfortunately, now have no possibility to make some reproducer

[Bug tree-optimization/57124] 254.gap@spec2000 got miscompare after r198413

[law at redhat dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=57124

Jeffrey A. Law <law at redhat dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |ASSIGNED
   Last reconfirmed|                            |2013-04-30
     Ever Confirmed|0                           |1

[Bug tree-optimization/57124] 254.gap@spec2000 got miscompare after r198413

[law at redhat dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=57124

--- Comment #1 from Jeffrey A. Law <law at redhat dot com> 2013-04-30 19:20:19 UTC ---
It looks like range_fits_type_p may not be handling overflows correctly. 
Investigating.

[Bug tree-optimization/57124] 254.gap@spec2000 got miscompare after r198413

[law at redhat dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=57124

--- Comment #2 from Jeffrey A. Law <law at redhat dot com> 2013-05-01 05:13:38 UTC ---
__attribute__ ((noinline))
foo(short unsigned int *p1, short unsigned int *p2)
{
  short unsigned int x1, x4;
  int x2, x3, x5, x6;
  unsigned int x7;

  x1 = *p1;
  x2 = (int) x1;
  x3 = x2 * 65536;
  x4 = *p2;
  x5 = (int) x4;
  x6 = x3 + x4;
  x7 = (unsigned int) x6;
  if (x7 <= 268435455U)
    abort ();
  exit (0);
}

main()
{
  short unsigned int x, y;
  x = -5;
  y = -10;
  foo (&x, &y);
}


Is a much reduced testcase.  Ultimately we have a range for x6 that's [0,
+INF(OVF)] at which point it's not safe to replace x7 with x6 in the
conditional.

[Bug tree-optimization/57124] 254.gap@spec2000 got miscompare after r198413

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=57124

--- Comment #3 from Richard Biener <rguenth at gcc dot gnu.org> 2013-05-02 08:01:39 UTC ---
At least the reduced testcase contains undefined signed integer overflow.
It multiplies 65531 by 65536 here:

  x3 = x2 * 65536;

it's still "miscompiled" with -fno-strict-overflow though, a QOI issue
(it works correctly with -fwrapv).  The docs trying to point out the
difference between -fno-strict-overflow and -fwrapv are especially poor
(ISTR strongly objecting the notion of -f[no-]strict-overflow ...)

[Bug tree-optimization/57124] 254.gap@spec2000 got miscompare after r198413

[law at redhat dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=57124

--- Comment #4 from Jeffrey A. Law <law at redhat dot com> 2013-05-07 04:25:27 UTC ---
Yea, 254.gap is definitely overflowing signed types.  I've got changes to make
the warnings and -fno-strict-overflow work that I'll put through their paces
tomorrow.

I think we should twiddle the testcase to verify we don't muck it up when 
-fno-strict-overflow is active.

[Bug tree-optimization/57124] 254.gap@spec2000 got miscompare after r198413

[ysrumyan at gmail dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=57124

Yuri Rumyantsev <ysrumyan at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |ysrumyan at gmail dot com

--- Comment #6 from Yuri Rumyantsev <ysrumyan at gmail dot com> ---
We prepared a patch for 254.gap which resolves issue. So this bug can be
closed.

[Bug tree-optimization/57124] 254.gap@spec2000 got miscompare after r198413

[rguenth at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=57124

--- Comment #7 from Richard Biener <rguenth at gcc dot gnu.org> ---
(In reply to Jeffrey A. Law from comment #4)
> Yea, 254.gap is definitely overflowing signed types.  I've got changes to
> make the warnings and -fno-strict-overflow work that I'll put through their
> paces tomorrow.
> 
> I think we should twiddle the testcase to verify we don't muck it up when 
> -fno-strict-overflow is active.

Yes, that would be nice.

[Bug tree-optimization/57124] 254.gap@spec2000 got miscompare after r198413

[law at redhat dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=57124

Jeffrey A. Law <law at redhat dot com> changed:

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

--- Comment #8 from Jeffrey A. Law <law at redhat dot com> ---
Trunk updated to only perform optimization when overflow semantics allow it. 
254.gap should be compiling with -fno-strict-overflow to avoid the undefined
behaviour.