[Bug fortran/46703] New: Wrong I/O output (only) when running under valgrind

[Bug fortran/46703] New: Wrong I/O output (only) when running under valgrind

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46703

           Summary: Wrong I/O output (only) when running under valgrind
           Product: gcc
           Version: 4.6.0
            Status: UNCONFIRMED
          Keywords: wrong-code
          Severity: normal
          Priority: P3
         Component: fortran
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: burnus@gcc.gnu.org
                CC: jvdelisle@gcc.gnu.org


There is something odd with the following program. It shows (here) always the
correct output -- except I run it through valgrind.

(The program is from
http://groups.google.com/group/comp.lang.fortran/msg/cb8e13e6226bc6c5 (part of
the c.l.f thread
http://groups.google.com/group/comp.lang.fortran/browse_thread/thread/1b907e3b7b6f3461).
James reports that it fails on MinGW with the www.equation.com version but not
with the MinGW version.)

Without valgrind, the program prints (with GCC 4.6 and 4.5 and any optimization
level):
  3.36210314311209350626E-4912
  ...
  3.36210314311209350626E-4912
  1.18973149535723176502E+4932
  ...
  1.18973149535723176502E+4932

However, using valgrind I get:
   0.0000000000000000000
   ...
   0.0000000000000000000
  i.f                   E+0000
   ...
  i.f                   E+0000

But otherwise no warning and no leakage.

Using GCC 4.5, I get with valgrind "0.0000" and "+Infinity".

The issue seems to be in libgfortran as the result depends on the library and
not on the FE compiler. I am also not sure whether there is a real issue or
whether it is just a side effect of valgrind.


program bugs2
   implicit none
   integer, parameter :: ep = 10
   real(ep) x
   integer i

   x = tiny(x)
   write(*,*) x
   do i = 1, 10
      x = 100*x
      write(*,*) x
   end do
   x = huge(x)
   write(*,*) x
   do i = 1, 10
      x = x/1.0e100_ep
      write(*,*) x
   end do
end program bugs2

[Bug fortran/46703] Wrong I/O output (only) when running under valgrind

[mikael at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46703

Mikael Morin <mikael at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |mikael at gcc dot gnu.org

--- Comment #1 from Mikael Morin <mikael at gcc dot gnu.org> 2010-11-29 19:26:37 UTC ---
Maybe yet another of x86's extra floating point precision effects.
I suppose where valgrind needs to do its verifications, it interrupts the
program and saves/restores the registers (to memory), thus losing the extra
precision.

Try to ask to the valgrind developers, the problems looks very
valgrind-related.

[Bug fortran/46703] Wrong I/O output (only) when running under valgrind

[dfranke at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46703

Daniel Franke <dfranke at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |WAITING
   Last reconfirmed|                            |2010.12.27 00:23:27
                 CC|                            |dfranke at gcc dot gnu.org
     Ever Confirmed|0                           |1

--- Comment #2 from Daniel Franke <dfranke at gcc dot gnu.org> 2010-12-27 00:23:27 UTC ---
(In reply to comment #1)
> Maybe yet another of x86's extra floating point precision effects.
> I suppose where valgrind needs to do its verifications, it interrupts the
> program and saves/restores the registers (to memory), thus losing the extra
> precision.

Seconded. Simple test: does -ffloat-store change the output as well?

[Bug fortran/46703] Wrong I/O output (only) when running under valgrind

[jvdelisle at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46703

--- Comment #3 from Jerry DeLisle <jvdelisle at gcc dot gnu.org> 2010-12-27 01:03:55 UTC ---
-ffloat-store is the first thing I tried and no change.  However, the run-time
side, libgfortran, is not compiled normally with -ffloat-store so maybe we
should try that.

[Bug fortran/46703] Wrong I/O output (only) when running under valgrind

[dfranke at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46703

--- Comment #4 from Daniel Franke <dfranke at gcc dot gnu.org> 2011-07-24 18:53:21 UTC ---
Any news here? May this report be closed?

[Bug fortran/46703] Wrong I/O output (only) when running under valgrind

[bdavis at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46703

--- Comment #6 from Bud Davis <bdavis at gcc dot gnu.org> ---
It is a problem with Valgrind.
One that is even mentioned in the (valgrind) manual.

https://bugs.kde.org/show_bug.cgi?id=197915

It has been open for about 4 years, not fixed yet.

Short summary.  Don't use valgrind on real's larger than 64 bits.

--bud davis

[Bug fortran/46703] Wrong I/O output (only) when running under valgrind

[jvdelisle at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46703

Jerry DeLisle <jvdelisle at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|WAITING                     |RESOLVED
         Resolution|---                         |INVALID

--- Comment #7 from Jerry DeLisle <jvdelisle at gcc dot gnu.org> ---
Closing as invalid.