ICE in gfc_conv_array_index_offset

[Maik Beckmann]

Hello List,

I'm pretty new to fortran and don't know if the code below is even valid. Thus 
to I'm posting this report to this list rather than adding it to via 
bugzilla.  Sorry for the inconvenience.

I'm want to port a bunch of code of my research group from intel to gfortran.  
I've encountered an ICE.  Several hours later I've reduced the code to
<foo.f90>
MODULE foo

  TYPE, PUBLIC :: bar
    PRIVATE
      REAL :: value
  END TYPE bar

  INTERFACE ASSIGNMENT (=)
    MODULE PROCEDURE assign_bar
  END INTERFACE ASSIGNMENT (=)

CONTAINS

  ELEMENTAL SUBROUTINE assign_bar (to, from)
    TYPE(bar), INTENT(OUT) :: to
    TYPE(bar), INTENT(IN) :: from
    to%value= from%value
  END SUBROUTINE

  SUBROUTINE my_sub (in, out)
    IMPLICIT NONE
    TYPE(bar), DIMENSION(:,:), POINTER :: in
    TYPE(bar), DIMENSION(:,:), POINTER :: out

    ALLOCATE( out(1:42, 1:42) )
    out(1, 1:42) = in(1, 1:42)

  END SUBROUTINE

END MODULE foo
</foo.f90>

The compiler output is attached.

Is this code valid?  If ELEMENTAL is omitted or replaced by PURE (I don't 
really know the semantics of these keyword, google says they're about 
parallelism) gfortran compiles fine.

Best,
 -- Maik
-------------- next part --------------
A non-text attachment was scrubbed...
Name: compiler_output
Type: text/x-diff
Size: 767 bytes
Desc: not available
URL: <http://gcc.gnu.org/pipermail/fortran/attachments/20080213/7809a309/attachment.bin>