[Bug fortran/53310] New: [4.6/4.7/4.8 Regression] EOSHIFT leaks memory

[Bug fortran/53310] New: [4.6/4.7/4.8 Regression] EOSHIFT leaks memory

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53310

             Bug #: 53310
           Summary: [4.6/4.7/4.8 Regression] EOSHIFT leaks memory
    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: burnus@gcc.gnu.org
                CC: tkoenig@gcc.gnu.org


https://groups.google.com/forum/?fromgroups#!topic/comp.lang.fortran/DAzRG_a6wkg
shows that gfortran leaks memory.

The following program leaks 6 MB per iteration. It shows no valgrind failues
with gfortran 4.1.2 but with 4.3 and later it leaks memory. For 4.8, valgrind
shows that the memory is allocated in the library.

I do not have GCC 4.3's libgfortran at hand - only the compiler itself. But I
can re-produce the leakage with GCC 4.6/4.7/4.8's libgfortran. I don't know
whether 4.3 to 4.5 are affected or not.


Valgrind shows:

2,097,152 bytes in 1 blocks are definitely lost in loss record 1 of 2
  at 0x4C2ABED: malloc (in
/usr/lib64/valgrind/vgpreload_memcheck-amd64-linux.so)
  by 0x4E4ECE4: _gfortrani_xmalloc (memory.c:39)
  by 0x4F1DEE4: eoshift2 (eoshift2.c:95)
  by 0x4F1E0F1: _gfortran_eoshift2_4 (in
/projects/tob/gcc-git/gcc-trunk/lib64/libgfortran.so.3.0.0)
  by 0x4014D9: MAIN__ (gh444.f90:21)
  by 0x401910: main (gh444.f90:25)

4,194,304 bytes in 2 blocks are possibly lost in loss record 2 of 2
  at 0x4C2ABED: malloc (in
/usr/lib64/valgrind/vgpreload_memcheck-amd64-linux.so)
  by 0x4E4ECE4: _gfortrani_xmalloc (memory.c:39)
  by 0x4F1DEE4: eoshift2 (eoshift2.c:95)
  by 0x4F1E0F1: _gfortran_eoshift2_4 (in
/projects/tob/gcc-git/gcc-trunk/lib64/libgfortran.so.3.0.0)
  by 0x4014D9: MAIN__ (gh444.f90:21)
  by 0x401910: main (gh444.f90:25)

LEAK SUMMARY:
  definitely lost: 2,097,152 bytes in 1 blocks
  indirectly lost: 0 bytes in 0 blocks
    possibly lost: 4,194,304 bytes in 2 blocks



PROGRAM eo_test

  INTEGER, PARAMETER :: pp=SELECTED_REAL_KIND(6)
  INTEGER :: x, y, z, dmn, step
  REAL(pp), ALLOCATABLE, DIMENSION(:,:,:,:) :: D
  REAL(pp), DIMENSION(:,:,:,:), ALLOCATABLE, TARGET :: buffer
  REAL(pp), DIMENSION(:,:,:), POINTER :: lower
  REAL :: tt

  y=1024; x=8; z=8; dmn=8
  ALLOCATE( buffer(y,z,dmn,6) )
  ALLOCATE( D(y,z,x,dmn) )
  lower => buffer(:,:,:,3)

  D(:,:,:,:) = 1.0_pp

step = 1
!DO step = 1, 1 !1000  
  tt=REAL(step)
  buffer(:,:,:,:) = tt*SIN(tt)
  D(:,:,:,:) = EOSHIFT(D(:,:,:,:),-1, lower(:,:,:), DIM=3)
  WRITE(*,*) 'Step, max(D):', tt,MAXVAL( D(:,:,:,:) )
!END DO

END PROGRAM eo_test

[Bug fortran/53310] [4.5/4.6/4.7/4.8 Regression] EOSHIFT leaks memory

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53310

Dominique d'Humieres <dominiq at lps dot ens.fr> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2012-05-10
            Summary|[4.6/4.7/4.8 Regression]    |[4.5/4.6/4.7/4.8
                   |EOSHIFT leaks memory        |Regression] EOSHIFT leaks
                   |                            |memory
     Ever Confirmed|0                           |1

--- Comment #1 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2012-05-10 21:03:45 UTC ---
I don't see any memory leak with 4.4.6, but I see it with 4.5.3.

==6547== LEAK SUMMARY:
==6547==    definitely lost: 6,291,456 bytes in 3 blocks
==6547==    indirectly lost: 0 bytes in 0 blocks
==6547==      possibly lost: 0 bytes in 0 blocks
==6547==    still reachable: 196 bytes in 3 blocks
==6547==         suppressed: 88 bytes in 1 blocks

[Bug fortran/53310] [4.5/4.6/4.7/4.8 Regression] EOSHIFT leaks memory

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53310

Tobias Burnus <burnus at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |burnus at gcc dot gnu.org
   Target Milestone|---                         |4.6.4

--- Comment #2 from Tobias Burnus <burnus at gcc dot gnu.org> 2012-05-10 21:27:52 UTC ---
Draft patch. I think it makes sense to move the allocation out of the loop:


--- a/libgfortran/intrinsics/eoshift2.c
+++ b/libgfortran/intrinsics/eoshift2.c
@@ -79,2 +79,6 @@ eoshift2 (gfc_array_char *ret, const gfc_array_char *array,
       ret->dtype = array->dtype;
+
+      /* xmalloc allocates a single byte for zero size.  */
+      ret->base_addr = xmalloc (size * arraysize);
+
       for (i = 0; i < GFC_DESCRIPTOR_RANK (array); i++)
@@ -92,6 +96,2 @@ eoshift2 (gfc_array_char *ret, const gfc_array_char *array,
          GFC_DIMENSION_SET(ret->dim[i], 0, ub, str);
-
-          /* xmalloc allocates a single byte for zero size.  */
-         ret->base_addr = xmalloc (size * arraysize);
-
         }

[Bug fortran/53310] [4.5/4.6/4.7/4.8 Regression] EOSHIFT leaks memory

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53310

--- Comment #3 from Tobias Burnus <burnus at gcc dot gnu.org> 2012-05-11 08:15:05 UTC ---
Author: burnus
Date: Fri May 11 08:14:56 2012
New Revision: 187395

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=187395
Log:
2012-05-11  Tobias Burnus  <burnus@net-b.de>

        PR fortran/53310
        * intrinsics/eoshift2.c (eoshift2): Do not leak
        memory by allocating it in the loop.


Modified:
    trunk/libgfortran/ChangeLog
    trunk/libgfortran/intrinsics/eoshift2.c

[Bug fortran/53310] [4.5/4.6/4.7/4.8 Regression] EOSHIFT leaks memory

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53310

--- Comment #4 from Tobias Burnus <burnus at gcc dot gnu.org> 2012-05-11 22:32:36 UTC ---
Author: burnus
Date: Fri May 11 22:32:27 2012
New Revision: 187417

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=187417
Log:
2012-05-12  Tobias Burnus  <burnus@net-b.de>

        PR fortran/53310
        * intrinsics/eoshift2.c (eoshift2): Do not leak
        memory by allocating it in the loop.


Modified:
    branches/gcc-4_7-branch/libgfortran/ChangeLog
    branches/gcc-4_7-branch/libgfortran/intrinsics/eoshift2.c

[Bug fortran/53310] [4.5/4.6/4.7/4.8 Regression] EOSHIFT leaks memory

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53310

--- Comment #5 from Tobias Burnus <burnus at gcc dot gnu.org> 2012-05-11 22:33:30 UTC ---
Author: burnus
Date: Fri May 11 22:33:21 2012
New Revision: 187418

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=187418
Log:
2012-05-12  Tobias Burnus  <burnus@net-b.de>

        PR fortran/53310
        * intrinsics/eoshift2.c (eoshift2): Do not leak
        memory by allocating it in the loop.


Modified:
    branches/gcc-4_6-branch/libgfortran/ChangeLog
    branches/gcc-4_6-branch/libgfortran/intrinsics/eoshift2.c

[Bug fortran/53310] [4.5/4.6/4.7/4.8 Regression] EOSHIFT leaks memory

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53310

--- Comment #6 from Tobias Burnus <burnus at gcc dot gnu.org> 2012-05-11 23:09:35 UTC ---
Author: burnus
Date: Fri May 11 23:09:30 2012
New Revision: 187419

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=187419
Log:
2012-05-12  Tobias Burnus  <burnus@net-b.de>

        PR fortran/53310
        * intrinsics/eoshift2.c (eoshift2): Do not leak
        memory by allocating it in the loop.


Modified:
    branches/gcc-4_5-branch/libgfortran/ChangeLog
    branches/gcc-4_5-branch/libgfortran/intrinsics/eoshift2.c

[Bug fortran/53310] [4.5/4.6/4.7/4.8 Regression] EOSHIFT leaks memory

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53310

Tobias Burnus <burnus at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|                            |FIXED
   Target Milestone|4.6.4                       |4.5.4

--- Comment #7 from Tobias Burnus <burnus at gcc dot gnu.org> 2012-05-11 23:11:16 UTC ---
FIXED on the 4.8 trunk and backported to 4.5-4.7.

[Bug fortran/53310] [4.5/4.6/4.7/4.8 Regression] EOSHIFT leaks memory

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53310

Dominique d'Humieres <dominiq at lps dot ens.fr> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |herve.manche at libertysurf
                   |                            |dot fr

--- Comment #8 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2013-02-27 11:09:50 UTC ---
*** Bug 51487 has been marked as a duplicate of this bug. ***