[Bug fortran/56800] New: [fortran-dev Regression] move_alloc_13.f90 failure

[Bug fortran/56800] New: [fortran-dev Regression] move_alloc_13.f90 failure

[tkoenig at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=56800

             Bug #: 56800
           Summary: [fortran-dev Regression] move_alloc_13.f90 failure
    Classification: Unclassified
           Product: gcc
           Version: fortran-dev
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: fortran
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: tkoenig@gcc.gnu.org


The move_alloc_13 failure on the fortran-dev branch can be reduced
to a wrong setting of sm to zero in:

ig25@linux-fd1f:~/Krempel/Branch> cat mv.f90
type t
end type t
type, extends(t) :: t2
  integer :: a
end type t2

class(t), allocatable :: y(:), z(:)

allocate(y(1), source=[t2(2)])

end
ig25@linux-fd1f:~/Krempel/Branch> gfortran -fdump-tree-original mv.f90

[...]

        __builtin_memset (y._data.base_addr, 0, 4);
        y._data.dtype = 41;
        y._data.dim[0].lower_bound = 1;
        y._data.dim[0].extent = 2 - y._data.dim[0].lower_bound;
        y._data.dim[0].sm = 0;
        y._data.offset = -1;

[Bug fortran/56800] [fortran-dev Regression] move_alloc_13.f90 failure

[tkoenig at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=56800

--- Comment #1 from Thomas Koenig <tkoenig at gcc dot gnu.org> 2013-04-01 17:05:35 UTC ---
The stride needs to be set from the source; it currently
is taken from y (which is an empty type, hence the 0
for sm).

[Bug fortran/56800] [fortran-dev Regression] move_alloc_13.f90 failure

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=56800

--- Comment #2 from Tobias Burnus <burnus at gcc dot gnu.org> 2013-04-01 18:01:03 UTC ---
The problem is in gfc_array_init_size. There, one should first obtain the
element size. And instead of gfc_conv_descriptor_stride_set one should use
gfc_conv_descriptor_sm_set.

Additionally, one could directly set 
gfc_conv_descriptor_extent_set (matching the current size variable) instead of
using gfc_conv_descriptor_ubound_set.

The only difficulty is to implement it such that the overflow algorithm works.

[Bug fortran/56800] [fortran-dev Regression] move_alloc_13.f90 failure

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=56800

--- Comment #3 from Tobias Burnus <burnus at gcc dot gnu.org> 2013-04-01 21:34:55 UTC ---
Created attachment 29769
  --> http://gcc.gnu.org/bugzilla/attachment.cgi?id=29769
Early draft patch

The patch mostly implements a fix for this bug - but it needs some clean up in
the bottom part. As I run out of time, I attach it here.

[Bug fortran/56800] [fortran-dev Regression] move_alloc_13.f90 failure

[dominiq at lps dot ens.fr]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=56800

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2013-04-27
     Ever Confirmed|0                           |1

--- Comment #4 from Dominique d'Humieres <dominiq at lps dot ens.fr> 2013-04-27 13:42:51 UTC ---
Confirmed for the fortran-dev branch r198346 with the patch at
http://gcc.gnu.org/ml/fortran/2013-04/msg00237.html.

[Bug fortran/56800] [fortran-dev Regression] move_alloc_13.f90 failure

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=56800

--- Comment #5 from Tobias Burnus <burnus at gcc dot gnu.org> 2013-04-30 21:57:28 UTC ---
For the test case in comment 0:
        __builtin_memset (y._data.base_addr, 0, 4);
        y._data.rank = 1;
        y._data.type = -1;
        y._data.dim[0].lower_bound = 1;
        y._data.dim[0].extent = 2 - y._data.dim[0].lower_bound;
        y._data.dim[0].sm = (integer(kind=8)) y._data.elem_len;

Here a line like:
  y._data.elem_len = 4;
is missing. elem_len needs always to be updated for BT_CLASS.

[Bug fortran/56800] [fortran-dev Regression] move_alloc_13.f90 failure

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=56800

--- Comment #6 from Tobias Burnus <burnus at gcc dot gnu.org> 2013-05-07 10:00:20 UTC ---
FIXED by the following commit (on the Fortran-dev branch).

Thanks for the report!


Author: burnus
Date: Tue May  7 09:37:19 2013
New Revision: 198669

URL: http://gcc.gnu.org/viewcvs?rev=198669&root=gcc&view=rev
Log:
2013-05-07  Tobias Burnus  <burnus@net-b.de>

        * trans-array.c (gfc_conv_descriptor_stride_get,
        gfc_conv_descriptor_stride_set): Use elem_len, unless it is
        a nonstring intrinsic type for which size_in_bytes is used.
        (gfc_array_init_size): Set elem_len before handling the
        strides.
        * trans-expr.c (gfc_conv_subref_array_arg): Remove no-op
        extent shifting code.


Modified:
    branches/fortran-dev/gcc/fortran/ChangeLog.fortran-dev
    branches/fortran-dev/gcc/fortran/trans-array.c
    branches/fortran-dev/gcc/fortran/trans-expr.c

[Bug fortran/56800] [fortran-dev Regression] move_alloc_13.f90 failure

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=56800

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

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

--- Comment #7 from Tobias Burnus <burnus at gcc dot gnu.org> 2013-05-07 10:01:47 UTC ---
really mark as FIXED