[Bug fortran/55207] New: Automatic deallocation of variables declared in the main program

[Bug fortran/55207] New: Automatic deallocation of variables declared in the main program

[janus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55207

             Bug #: 55207
           Summary: Automatic deallocation of variables declared in the
                    main program
    Classification: Unclassified
           Product: gcc
           Version: 4.8.0
            Status: UNCONFIRMED
          Keywords: wrong-code
          Severity: normal
          Priority: P3
         Component: fortran
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: janus@gcc.gnu.org


Although F95/F03/F08 only require automatic deallocation of local unsaved
variables upon termination of a procedure, gfortran currently also does it upon
termination of the main program.

Simple test case (for both scalars and arrays):

program main
  integer, allocatable :: i
  integer, allocatable, dimension(:) :: j
  allocate(i)
  allocate(j(1:5))
end program


The dump shows a block which automatically deallocates the variables at the end
of the main program:

  finally
    {
      if (j.data != 0B)
        {
          __builtin_free ((void *) j.data);
        }
      j.data = 0B;
      if (i != 0B)
        {
          __builtin_free ((void *) i);
        }
    }


>From F08:6.7.3.2:

"When the execution of a procedure is terminated by execution of a RETURN or
END statement, an unsaved allocatable local variable of the procedure retains
its allocation and definition status if it is a function result variable or a
subobject thereof; otherwise, it is deallocated."

This mentions only *procedures*, not the main program. Moreover, variables
declared in the main program are implicitly SAVEd in F08, cf. chapter 5.3.16:

"A variable, common block, or procedure pointer declared in the scoping unit of
a main program, module, or submodule implicitly has the SAVE attribute, which
may be confirmed by explicit specification. If a common block has the SAVE
attribute in any other kind of scoping unit, it shall have the SAVE attribute
in every scoping unit that is not of a main program, module, or submodule."

Currently we miss to flag this via attr.save = SAVE_IMPLICIT.

In F95 it is basically impossible to detect whether a compiler does
end-of-program auto-dealloc. However, in F03 it can be detected by using a
FINAL procedure, since deallocation (automatic or explicit) also implies
finalization (for finalizable variables).

[Bug fortran/55207] Automatic deallocation of variables declared in the main program

[janus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55207

--- Comment #1 from janus at gcc dot gnu.org 2012-11-04 18:32:29 UTC ---
Patch:

Index: gcc/fortran/trans-decl.c
===================================================================
--- gcc/fortran/trans-decl.c    (revision 193135)
+++ gcc/fortran/trans-decl.c    (working copy)
@@ -3771,9 +3771,10 @@ gfc_trans_deferred_vars (gfc_symbol * proc_sym, gf
           else
         gfc_restore_backend_locus (&loc);

-          /* Deallocate when leaving the scope. Nullifying is not
-         needed.  */
-          if (!sym->attr.result && !sym->attr.dummy)
+          /* Automatic deallocation when leaving the scope.
+         Nullifying is not needed.  */
+          if (!proc_sym->attr.is_main_program
+          && !sym->attr.result && !sym->attr.dummy)
         {
           if (sym->ts.type == BT_CLASS
               && CLASS_DATA (sym)->attr.codimension)



This regresses on:

FAIL: gfortran.dg/allocatable_scalar_9.f90  -O0   scan-tree-dump-times original
"__builtin_free" 32
FAIL: gfortran.dg/coarray_lib_alloc_2.f90  -O   scan-tree-dump-times original
"_gfortran_caf_deregister .&yy._data.token, 0B, 0B, 0.;" 1
FAIL: gfortran.dg/move_alloc_4.f90  -O0   scan-tree-dump-times original
"__builtin_free" 9

[Bug fortran/55207] Automatic deallocation of variables declared in the main program

[janus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55207

--- Comment #2 from janus at gcc dot gnu.org 2012-11-04 22:26:44 UTC ---
(In reply to comment #1)
> Patch:

Note: The patch in comment 1 only fixes the auto-deallocation for scalars.

[Bug fortran/55207] Automatic deallocation of variables declared in the main program

[janus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55207

--- Comment #3 from janus at gcc dot gnu.org 2012-11-04 22:48:37 UTC ---
The following patch applies the implicit SAVE attribute to variables declared
in the main program:

Index: gcc/fortran/decl.c
===================================================================
--- gcc/fortran/decl.c    (revision 193137)
+++ gcc/fortran/decl.c    (working copy)
@@ -3812,9 +3812,11 @@ match_attr_spec (void)
     }
     }

-  /* Since Fortran 2008 module variables implicitly have the SAVE attribute. 
*/
-  if (gfc_current_state () == COMP_MODULE && !current_attr.save
-      && (gfc_option.allow_std & GFC_STD_F2008) != 0)
+  /* Since Fortran 2008, variables declared in a MODULE or PROGRAM
+     implicitly have the SAVE attribute.  */
+  if ((gfc_current_state () == COMP_MODULE
+       || gfc_current_state () == COMP_PROGRAM)
+      && !current_attr.save && (gfc_option.allow_std & GFC_STD_F2008) != 0)
     current_attr.save = SAVE_IMPLICIT;

   colon_seen = 1;


This also removes the automatic allocation of both variables in the test case
in comment 0. Therefore it has the same testsuite failures as the patch in
comment 1 (possibly more?).

[Bug fortran/55207] Automatic deallocation of variables declared in the main program

[janus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55207

--- Comment #4 from janus at gcc dot gnu.org 2012-11-05 09:07:23 UTC ---
(In reply to comment #3)
> Therefore it has the same testsuite failures as the patch in
> comment 1 (possibly more?).

Indeed it has a few more ...

FAIL: gfortran.dg/alloc_comp_basics_1.f90  -O0   scan-tree-dump-times original
"builtin_free" 18
FAIL: gfortran.dg/alloc_comp_constructor_1.f90  -O0   scan-tree-dump-times
original "builtin_free" 19
FAIL: gfortran.dg/allocatable_scalar_9.f90  -O0   scan-tree-dump-times original
"__builtin_free" 32
FAIL: gfortran.dg/auto_dealloc_2.f90  -O   scan-tree-dump-times original
"__builtin_free" 3
FAIL: gfortran.dg/coarray_lib_alloc_2.f90  -O   scan-tree-dump-times original
"_gfortran_caf_deregister .&yy._data.token, 0B, 0B, 0.;" 1
FAIL: gfortran.dg/intent_optimize_1.f90  -O   scan-tree-dump-times optimized
"does_not_exist" 0
FAIL: gfortran.dg/storage_size_3.f08  -O0  execution test
FAIL: gfortran.dg/extends_14.f03  -O   scan-tree-dump-times original
"__builtin_free" 3
FAIL: gfortran.dg/volatile4.f90  -O   scan-tree-dump-not optimized
"NonVolatileNotOptimizedAway"
FAIL: gfortran.dg/volatile6.f90  -O   scan-tree-dump-not optimized
"NonVolatileNotOptimizedAway1"
FAIL: gfortran.dg/move_alloc_4.f90  -O0   scan-tree-dump-times original
"__builtin_free" 9
FAIL: gfortran.dg/typebound_proc_27.f03  -O0   scan-tree-dump-times original
"__builtin_free" 7


Most of them seem to be scan-tree-dump failures, except for:

FAIL: gfortran.dg/storage_size_3.f08  -O0  execution test

which apparently segfaults at runtime (invalid memory reference).

[Bug fortran/55207] Automatic deallocation of variables declared in the main program

[janus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55207

--- Comment #5 from janus at gcc dot gnu.org 2012-11-05 09:37:20 UTC ---
(In reply to comment #4)
> Most of them seem to be scan-tree-dump failures, except for:
> 
> FAIL: gfortran.dg/storage_size_3.f08  -O0  execution test
> 
> which apparently segfaults at runtime (invalid memory reference).


This is due to missing vptr initialization of the variable 'y':

class(t), allocatable :: y

[Bug fortran/55207] Automatic deallocation of variables declared in the main program

[janus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55207

--- Comment #6 from janus at gcc dot gnu.org 2012-11-05 17:45:23 UTC ---
Created attachment 28620
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=28620
patch

Here is an extended patch, based on comment 3, which fixes the storage_size_3
wrong-code problem (cf. gfc_trans_deferred_vars) and corrects the
__builtin_free counts in the test cases.

It still fails on the following:

FAIL: gfortran.dg/c_ptr_tests_16.f90  -O   scan-tree-dump-times optimized
"i_do_not_exist" 0
FAIL: gfortran.dg/coarray_lib_alloc_2.f90  -O   scan-tree-dump-times original
"_gfortran_caf_deregister .&xx._data.token, 0B, 0B, 0.;" 1
FAIL: gfortran.dg/intent_optimize_1.f90  -O   scan-tree-dump-times optimized
"does_not_exist" 0
FAIL: gfortran.dg/volatile4.f90  -O   scan-tree-dump-not optimized
"NonVolatileNotOptimizedAway"
FAIL: gfortran.dg/volatile6.f90  -O   scan-tree-dump-not optimized
"NonVolatileNotOptimizedAway1"

Except for one coarray issue, those seem to be problems with optimizing away
stuff in certain situations (I haven't checked any details yet).

[Bug fortran/55207] Automatic deallocation of variables declared in the main program

[janus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55207

janus at gcc dot gnu.org changed:

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

--- Comment #7 from janus at gcc dot gnu.org ---
I think this PR has been fixed by r199643.

[Bug fortran/55207] Automatic deallocation of variables declared in the main program

[janus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55207

janus at gcc dot gnu.org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|RESOLVED                    |REOPENED
   Last reconfirmed|                            |2013-07-31
         Resolution|FIXED                       |---
     Ever confirmed|0                           |1

--- Comment #8 from janus at gcc dot gnu.org ---
(In reply to janus from comment #7)
> I think this PR has been fixed by r199643.

Reopening. Auto-dealloc has indeed been fixed by the above commit. However,
variables in the main program are still not SAVE_IMPLICIT, which makes problems
e.g. in PR 57306.

I think we need the patch in comment 6 after all. But how do we get rid of the
remaining regressions?

[Bug fortran/55207] Automatic deallocation of variables declared in the main program

[janus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55207

--- Comment #9 from janus at gcc dot gnu.org ---
(In reply to janus from comment #8)
> I think we need the patch in comment 6 after all. But how do we get rid of
> the remaining regressions?

Simplest solution: Move the code in these test cases from the main program into
a subroutine (or similar). Then the variables will not be SAVEd and all
optimizations can be applied as before. (However, it's a bit unfortunate that
we lose the possibility to do these optimizations in the main program.)

[Bug fortran/55207] Automatic deallocation of variables declared in the main program

[janus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55207

janus at gcc dot gnu.org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
  Attachment #28620|0                           |1
        is obsolete|                            |

--- Comment #10 from janus at gcc dot gnu.org ---
Created attachment 30584
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=30584&action=edit
new patch

Here is a new version of the patch which regtests cleanly.

The trans-decl.c parts are not strictly necessary. They just do some clean-up.

[Bug fortran/55207] Automatic deallocation of variables declared in the main program

[janus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55207

--- Comment #13 from janus at gcc dot gnu.org ---
Test case from PR 57306 comment 7 (see also
http://gcc.gnu.org/ml/fortran/2013-07/msg00103.html):


  type :: c
  end type c

  type(c), target :: x
  class(c), pointer :: px => x

  if (.not. associated(px)) call abort()
end

[Bug fortran/55207] [F08] Variables declared in the main program should implicitly get the SAVE attribute

[janus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55207

janus at gcc dot gnu.org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|REOPENED                    |ASSIGNED
           Assignee|unassigned at gcc dot gnu.org      |janus at gcc dot gnu.org

[Bug fortran/55207] [F08] Variables declared in the main program should implicitly get the SAVE attribute

[janus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55207

--- Comment #14 from janus at gcc dot gnu.org ---
Author: janus
Date: Sat Mar 15 10:53:04 2014
New Revision: 208590

URL: http://gcc.gnu.org/viewcvs?rev=208590&root=gcc&view=rev
Log:
2014-03-15  Janus Weil  <janus@gcc.gnu.org>

    PR fortran/55207
    * decl.c (match_attr_spec): Variables in the main program implicitly
    get the SAVE attribute in Fortran 2008.


2014-03-15  Janus Weil  <janus@gcc.gnu.org>

    PR fortran/55207
    * gfortran.dg/assumed_rank_7.f90: Explicitly deallocate variables.
    * gfortran.dg/c_ptr_tests_16.f90: Put into subroutine.
    * gfortran.dg/inline_sum_bounds_check_1.f90: Add
    -Wno-aggressive-loop-optimizations and remove an unused variable.
    * gfortran.dg/intent_optimize_1.f90: Put into subroutine.
    * gfortran.dg/pointer_init_9.f90: New.
    * gfortran.dg/volatile4.f90: Put into subroutine.
    * gfortran.dg/volatile6.f90: Ditto.

Added:
    trunk/gcc/testsuite/gfortran.dg/pointer_init_9.f90
Modified:
    trunk/gcc/fortran/ChangeLog
    trunk/gcc/fortran/decl.c
    trunk/gcc/testsuite/ChangeLog
    trunk/gcc/testsuite/gfortran.dg/assumed_rank_7.f90
    trunk/gcc/testsuite/gfortran.dg/c_ptr_tests_16.f90
    trunk/gcc/testsuite/gfortran.dg/inline_sum_bounds_check_1.f90
    trunk/gcc/testsuite/gfortran.dg/intent_optimize_1.f90
    trunk/gcc/testsuite/gfortran.dg/volatile4.f90
    trunk/gcc/testsuite/gfortran.dg/volatile6.f90

[Bug fortran/55207] [F08] Variables declared in the main program should implicitly get the SAVE attribute

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55207

--- Comment #15 from Dominique d'Humieres <dominiq at lps dot ens.fr> ---
*** Bug 55887 has been marked as a duplicate of this bug. ***

[Bug fortran/55207] [F08] Variables declared in the main program should implicitly get the SAVE attribute

[janus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55207

janus at gcc dot gnu.org changed:

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

--- Comment #16 from janus at gcc dot gnu.org ---
Fixed with r208590. Closing.

[Bug fortran/55207] [F08] Variables declared in the main program should implicitly get the SAVE attribute

[janus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55207

--- Comment #17 from janus at gcc dot gnu.org ---
Author: janus
Date: Tue Mar 18 22:15:10 2014
New Revision: 208668

URL: http://gcc.gnu.org/viewcvs?rev=208668&root=gcc&view=rev
Log:
2014-03-18  Janus Weil  <janus@gcc.gnu.org>

    PR fortran/55207
    PR fortran/60549
    * decl.c (match_attr_spec): Revert r208590.

2014-03-18  Janus Weil  <janus@gcc.gnu.org>

    PR fortran/55207
    PR fortran/60549
    * gfortran.dg/assumed_rank_7.f90: Revert r208590.
    * gfortran.dg/c_ptr_tests_16.f90: Ditto.
    * gfortran.dg/inline_sum_bounds_check_1.f90: Ditto.
    * gfortran.dg/intent_optimize_1.f90: Ditto.
    * gfortran.dg/pointer_init_9.f90: Ditto.
    * gfortran.dg/volatile4.f90: Ditto.
    * gfortran.dg/volatile6.f90: Ditto.

Removed:
    trunk/gcc/testsuite/gfortran.dg/pointer_init_9.f90
Modified:
    trunk/gcc/fortran/ChangeLog
    trunk/gcc/fortran/decl.c
    trunk/gcc/testsuite/ChangeLog
    trunk/gcc/testsuite/gfortran.dg/assumed_rank_7.f90
    trunk/gcc/testsuite/gfortran.dg/c_ptr_tests_16.f90
    trunk/gcc/testsuite/gfortran.dg/inline_sum_bounds_check_1.f90
    trunk/gcc/testsuite/gfortran.dg/intent_optimize_1.f90
    trunk/gcc/testsuite/gfortran.dg/volatile4.f90
    trunk/gcc/testsuite/gfortran.dg/volatile6.f90

[Bug fortran/55207] [F08] Variables declared in the main program should implicitly get the SAVE attribute

[janus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=55207

janus at gcc dot gnu.org changed:

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

--- Comment #18 from janus at gcc dot gnu.org ---
Reopening. The patch has been reverted due to the problems mentioned in

http://gcc.gnu.org/ml/fortran/2014-03/msg00108.html

[Bug fortran/55207] [F08] Variables declared in the main program should implicitly get the SAVE attribute

[jb at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=55207

Janne Blomqvist <jb at gcc dot gnu.org> changed:

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

--- Comment #19 from Janne Blomqvist <jb at gcc dot gnu.org> ---
As was mentioned recently on comp.lang.fortran, this affects OpenMP. Consider

program teste_paralelo
  use omp_lib
  real :: v1(12,60,60,60,2)
  !$omp parallel do
  do kk=2,60
     v1(5,2,2,kk,1)=0
  enddo
  !$omp end parallel do
  !$omp parallel
  print *, 'Hello World!'
  !$omp end parallel
end program teste_paralelo

Compiling without OpenMP and checking the binary with size, the array v1 is
allocated statically in the bss section:

$ gfortran -g teste_paralelo.f90 -o test-serial
$ size test-serial 
   text    data     bss     dec     hex filename
   2147     608 20736032        20738787        13c72e3 test-serial

Enabling openmp, the array suddenly goes on the stack:

$ gfortran -g -fopenmp teste_paralelo.f90 -o test
$ size test
   text    data     bss     dec     hex filename
   2822     656       8    3486     d9e test


If one manually makes v1 saved, it naturally goes into the bss with openmp too:

$ gfortran -g -fopenmp teste_paralelo.f90 -o test-save 
$ size test-save 
   text    data     bss     dec     hex filename
   2790     656 20736032        20739478        13c7596 test-save

[Bug fortran/55207] [F08] Variables declared in the main program should implicitly get the SAVE attribute

[pault at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=55207

Paul Thomas <pault at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |pault at gcc dot gnu.org
             Blocks|37336, 78122                |

--- Comment #22 from Paul Thomas <pault at gcc dot gnu.org> ---
As far as I am aware, this bug has either been fixed or has been worked round
everywhere. However, it doesn't block PR37336 and so I am releasing that
dependency.

Cheers

Paul


Referenced Bugs:

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=37336
[Bug 37336] [F03] Finish derived-type finalization
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=78122
[Bug 78122] [5/6/7/8 Regression] [F08] ICE in get, at cgraph.h:395

[Bug fortran/55207] [F08] Variables declared in the main program should implicitly get the SAVE attribute

[pault at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=55207

--- Comment #23 from Paul Thomas <pault at gcc dot gnu.org> ---
As far as I am aware, this bug has either been fixed or has been worked round
everywhere. However, it doesn't block PR37336 and so I am releasing that
dependency.

Cheers

Paul

[Bug fortran/55207] [F08] Variables declared in the main program should implicitly get the SAVE attribute

[burnus at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=55207

--- Comment #24 from Tobias Burnus <burnus at gcc dot gnu.org> ---
Created attachment 54968
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=54968&action=edit
WIP patch – still many testcase fails

Patch to fixes some MODULE and PROGRAM issues with automatic save.
TODO:
(a) Fix testsuite fails (required)
[(b) Anything to do for COMMON or DATA blocks?]
(c) as follow up, cleanup code for no-longer needed checks (can be follow up)