[MPI3 Fortran] Results of San Jose Forum meeting

[Rolf Rabenseifner]

Hi all,

I expect, that in our special case, the alignment problems cannot
cause problems, because all is Fortran INTEGER and therefore
at least 4-byte-aligned.
With intentionally misaligned handles, the new derived type handle
seems to work better (but not worse) than integer handles.

I've attached a small test code to solve the question, whether 
 - INTEGER and
 - TYPE handle; SEQUENCE INTEGER :: val; END TYPE handle
is passed to a C back-end in the same way.

Does it produce the expected results on all compilers:

    IBM (several compilers)
    Cray (ditto?)
    NAG Fortran
    Intel
    gfortran
    Oracle/Sun
    Pathscale
    Portland Group
    Possibly g95

I checked with Intel:

  cc -c handle_test_c.c
  ifort -o handle_test handle_test.f90 handle_test_c.o

  fortcom: Warning: handle_test.f90, line 42: The structure contains one or more misaligned fields.   [SEQUENCE_A]
    TYPE sequence_A
  -------^
  fortcom: Warning: handle_test.f90, line 49: The structure contains one or more misaligned fields.   [SEQUENCE_B]
    TYPE sequence_B
  -------^

  ./handle_test

  hd_1 = 12, hd_2 = 13  (expected: 12, 13)
  hd_i=112, hd_t%val=113 (expected: 112, 113)
 
  hd_1 = 14, hd_2 = 15  (expected: 14, 15)
  strA%handle=114, strB%handle%val=115 (expected: 114, 115)
 
  hd_1 = 24, hd_2 = 75  (expected: 25, 75)
  strA%handle=225, seqB%handle%val=775 (expected: 225, 775)
 
  handle_test(12868): unaligned access to 0x600000000000cf29, ip=0x4000000000003530
  hd_1 = 84, hd_2 = 35  (expected: 84, 35)
  handle_test(12868): unaligned access to 0x600000000000cf29, ip=0x4000000000003571
  seqA%handle=884, strB%handle%val=335 (expected: 884, 335)
 
  handle_test(12868): unaligned access to 0x600000000000cf29, ip=0x4000000000003530
  hd_1 = 94, hd_2 = 95  (expected: 94, 95)
  handle_test(12868): unaligned access to 0x600000000000cf29, ip=0x4000000000003571
  seqA%handle=994, seqB%handle%val=995 (expected: 994, 995)

Result:
All works, but there is really a small difference:
 - With misaligned INTEGER handle seqA, some runtime warnings occur
 - With new derive-type handle seqB, **NO** runtime warnings occur

It would be helpful, if you can check with some other compilers.

If you find another alignment problem with some other tests for
exactly these two types (no other derived types),
then please can you report.

The test program should test, whether internally the Fortran
compiler produces a subroutine call with an argument list
for both handle types that fits to the same C back-end: int*.

Rolf

----- "N.M. Maclaren" <nmm1 at cam.ac.uk> wrote:

> Sorry.  Posted in haste, and I forgot a critical paragraph.
> 
> With those structures, also try seeing what they do when you remove
> each of 'x' and 'y' in turn.  Specifically, does 'z' continue to be
> in the same location?
> 
> In the SPECIFIC example of a single default integer in a sequence
> type, it will a rare compiler that lays it out differently.  But
> all of the INTEGER, INTEGER BIND(C), sequence type and derived type
> without the SEQUENCE may have different sizes and alignments.  And,
> on SOME systems, a pointer to one of those may not be usable as a
> pointer to another.
> 
> 
> 
> _______________________________________________
> mpi3-fortran mailing list
> mpi3-fortran at lists.mpi-forum.org
> http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-fortran

-- 
Dr. Rolf Rabenseifner . . . . . . . . . .. email rabenseifner at hlrs.de
High Performance Computing Center (HLRS) . phone ++49(0)711/685-65530
University of Stuttgart . . . . . . . . .. fax ++49(0)711 / 685-65832
Head of Dpmt Parallel Computing . . . www.hlrs.de/people/rabenseifner
Nobelstr. 19, D-70550 Stuttgart, Germany . (Office: Allmandring 30)
-------------- next part --------------
A non-text attachment was scrubbed...
Name: handle_test.f90
Type: text/x-fortran
Size: 2284 bytes
Desc: not available
URL: <http://lists.mpi-forum.org/pipermail/mpiwg-fortran/attachments/20100313/2d997e06/attachment-0001.bin>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: handle_test_c.c
Type: text/x-csrc
Size: 291 bytes
Desc: not available
URL: <http://lists.mpi-forum.org/pipermail/mpiwg-fortran/attachments/20100313/2d997e06/attachment-0001.c>