[Bug fortran/15273] New: Wrong output from a pure subroutine

[Bug fortran/15273] New: Wrong output from a pure subroutine

[tcc at sentex dot net]

The code below gives wrong output, but is ok if "pure" is dropped. 
 
program testp 
integer, parameter :: intp = 3 
integer :: j 
call subx(intp,j) 
print *, intp, j 
contains 
pure subroutine subx(i,iout) 
integer, intent(in) :: i 
integer, intent(out) :: iout 
iout = i + 1 
end subroutine subx 
end program testpar 
 
$ gfortran -o testp testp.f90 
$ ./testp 
           3   1073873990 
$

-- 
           Summary: Wrong output from a pure subroutine
           Product: gcc
           Version: tree-ssa
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: fortran
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: tcc at sentex dot net
                CC: gcc-bugs at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15273

[Bug fortran/15273] [gfortran] Wrong output from a pure subroutine

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-05-04 02:25 -------
I think the issue is that the function is not really pure in the GCC sense.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
     Ever Confirmed|                            |1
           Keywords|                            |wrong-code
   Last reconfirmed|0000-00-00 00:00:00         |2004-05-04 02:25:02
               date|                            |
            Summary|Wrong output from a pure    |[gfortran] Wrong output from
                   |subroutine                  |a pure subroutine
   Target Milestone|---                         |tree-ssa


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15273

[Bug fortran/15273] [gfortran] Wrong output from a pure subroutine

[Tobias dot Schlueter at physik dot uni-muenchen dot de]

------- Additional Comments From Tobias dot Schlueter at physik dot uni-muenchen dot de  2004-05-04 13:30 -------
(In reply to comment #0)
> The code below gives wrong output, but is ok if "pure" is dropped. 

That code should not compile.

> pure subroutine subx(i,iout) 
> integer, intent(in) :: i 
> integer, intent(out) :: iout 

the first Constraint from 12.6. in the (draft) standard reads:
---
The /specification-part/ of a pure subroutine subprogram shall specify that all
dummy arguments have INTENT(IN) except procedure arguments and arguments with
the POINTER attribute.
---
This is violated here.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15273

[Bug fortran/15273] [gfortran] Wrong output from a pure subroutine

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-05-04 13:31 -------
That is what I thought, it is invalid code so marking as such.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|wrong-code                  |accepts-invalid
   Last reconfirmed|2004-05-04 02:25:02         |2004-05-04 13:31:48
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15273

[Bug fortran/15273] [gfortran] Wrong output from a pure subroutine

[Tobias dot Schlueter at physik dot uni-muenchen dot de]

------- Additional Comments From Tobias dot Schlueter at physik dot uni-muenchen dot de  2004-05-04 13:44 -------
It's quite interesting to see how this problem comes to happen: 
I used this slightly revised test program: 
[tobi@marktplatz tests]$ cat pr15273.f90
program testp
integer, parameter :: intp = 3
integer :: j
call subx(intp,j)
print *, intp, j
call suby(intp,j)
print *, intp, j
contains
pure subroutine subx(i,iout)
integer, intent(in) :: i
integer, intent(out) :: iout
iout = i + 1
end subroutine subx
  
subroutine suby(i,iout)
integer, intent(in) :: i
integer, intent(out) :: iout
iout = i + 1
end subroutine suby
end program testp

compiling with "gfortran pr15273.f90 -fdump-tree-all" yields these dumps:
[tobi@marktplatz tests]$ cat pr15273.f90.t03.original
suby (i, iout)
{
  *iout = *i + 1;
 
 
subx (i, iout)
{
  *iout = *i + 1;
 
 
MAIN__ ()
{
  int4 j;
  static  subx;
  static  suby;
 
  {
    int4 T.4;
 
    T.4 = 3;
    subx (&T.4, &j);;
  }
  _gfortran_filename = "pr15273.f90";
  _gfortran_line = 5;
  _gfortran_ioparm.unit = 6;
  _gfortran_ioparm.list_format = 1;
  _gfortran_st_write ();
  {
    int4 T.5;
 
    T.5 = 3;
    _gfortran_transfer_integer (&T.5, 4);;
  }
  _gfortran_transfer_integer (&j, 4);
  _gfortran_st_write_done ();
  {
    int4 T.6;
 
    T.6 = 3;
    suby (&T.6, &j);;
  }
  _gfortran_filename = "pr15273.f90";
  _gfortran_line = 7;
  _gfortran_ioparm.unit = 6;
  _gfortran_ioparm.list_format = 1;
  _gfortran_st_write ();
  {
    int4 T.7;
 
    T.7 = 3;
    _gfortran_transfer_integer (&T.7, 4);;
  }
  _gfortran_transfer_integer (&j, 4);
  _gfortran_st_write_done ();;
}
 
 
[tobi@marktplatz tests]$ cat pr15273.f90.t13.lower
 
;; Function subx (subx.1)
 
subx (i, iout)
{
  int4 T.3;
  int4 T.2;
 
  T.2 = *i;
  T.3 = T.2 + 1;
  *iout = T.3;
}
 
 
 
;; Function suby (suby.0)
 
suby (i, iout)
{
  int4 T.1;
  int4 T.0;
 
  T.0 = *i;
  T.1 = T.0 + 1;
  *iout = T.1;
}
 
 
 
;; Function MAIN__ (MAIN__)
 
MAIN__ ()
{
  int4 T.7;
  int4 T.6;
  int4 T.5;
  int4 T.4;
  int4 j;
 
  T.4 = 3;
  _gfortran_filename = "pr15273.f90";
  _gfortran_line = 5;
  _gfortran_ioparm.unit = 6;
  _gfortran_ioparm.list_format = 1;
  _gfortran_st_write ();
  T.5 = 3;
  _gfortran_transfer_integer (&T.5, 4);
  _gfortran_transfer_integer (&j, 4);
  _gfortran_st_write_done ();
  T.6 = 3;
  suby (&T.6, &j);
  _gfortran_filename = "pr15273.f90";
  _gfortran_line = 7;
  _gfortran_ioparm.unit = 6;
  _gfortran_ioparm.list_format = 1;
  _gfortran_st_write ();
  T.7 = 3;
  _gfortran_transfer_integer (&T.7, 4);
  _gfortran_transfer_integer (&j, 4);
  _gfortran_st_write_done ();
}

the interesting part is that in the lowered tree, the call to the pure
subroutine has disappeared. That's why j is uninitialized. Now I don't know if
that's a legal transformation, but I think it is.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15273

[Bug fortran/15273] [gfortran] Wrong output from a pure subroutine

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-05-04 13:46 -------
It is legal because the return value for subx is ignored since subx is pure the call can be removed.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|2004-05-04 13:31:48         |2004-05-04 13:46:14
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15273

[Bug fortran/15273] [gfortran] Wrong output from a pure subroutine

[Tobias dot Schlueter at physik dot uni-muenchen dot de]

------- Additional Comments From Tobias dot Schlueter at physik dot uni-muenchen dot de  2004-05-04 14:00 -------
Sorry, it looks like I switched paragraphs in the middle of reading the
standard. The code is LEGAL.

The first two constraints read correctly:
----
The /specification-part/ of a pure *function* subprogram shall specify that all
dummy arguments have INTENT(IN) except procedure arguments and arguments with
the POINTER attribute.
----
and
----
The /specification-part/ of a pure *subroutine* subprogram shall specify the
intents of all dummy arguments except procedure arguments, alternate return
indicators, and arguments with the POINTER attribute.
----
Come to think about it , this makes sense, as otherwise a pure subroutine
couldn't be very meaningful (and optimized away at will, as in this example).
For the curious, note 12.34 gives some insight into the logic behind this.

So it looks like the actual bug is that the middle-end is overly effective in
eliminating calls to pure functions.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15273

[Bug fortran/15273] [gfortran] Wrong output from a pure subroutine

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-05-04 14:03 -------
The real issue is the front-end is marking this function call as pure in the sense if the return value is 
ignored then it can be removed so the bug is in the front-end marking it as pure when it is not really 
that kind of pure.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|accepts-invalid             |wrong-code
   Last reconfirmed|2004-05-04 13:46:14         |2004-05-04 14:03:51
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15273

[Bug fortran/15273] [gfortran] Wrong output from a pure subroutine

[Tobias dot Schlueter at physik dot uni-muenchen dot de]

------- Additional Comments From Tobias dot Schlueter at physik dot uni-muenchen dot de  2004-05-04 19:53 -------
Patch here: http://gcc.gnu.org/ml/gcc-patches/2004-05/msg00195.html

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15273

[Bug fortran/15273] [gfortran] Wrong output from a pure subroutine

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-05-22 14:27 -------
Fixed.

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


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=15273