This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: Fwd: C++/Fortran Interop for Assumed Shape Arrays
- From: Tobias Burnus <burnus at net-b dot de>
- To: gfortran <fortran at gcc dot gnu dot org>
- Cc: GCC Mailing List <gcc at gcc dot gnu dot org>, Bob Fischer <rpf2116 at columbia dot edu>
- Date: Fri, 25 Oct 2013 23:30:41 +0200
- Subject: Re: Fwd: C++/Fortran Interop for Assumed Shape Arrays
- Authentication-results: sourceware.org; auth=none
- References: <90A023E3-7C7D-45EA-9490-3205F595B441 at columbia dot edu> <526ADD0A dot 4000802 at net-b dot de>
Bob Fischer wrote:
It would be useful if one could declare assumed shape arrays in Fortran, and then pass them to C/C++. The GNU Fortran documentation claims this is not normally possible in Fortran 2003. However, I have developed (and use daily) a standards-compliant way to do it.
Well, you do not directly use the array descriptor of Fortran – you
convert it into your own array descriptor which you pass around. That's
neat and quite useful but still something different.
"The function c_loc_double(x) is required because c_loc() does not seem
to work directly on elements of arrays (at least not in GNU's gfortran)."
That bug has finally be fixed this spring for GCC 4.9 – together with
lifting many other restrictions. For instance, Fortran 2008 or TS29113
(I forgot which one) now permits to take the C_LOC / C_FUNLOC of
noninteroperable variables/procedures.
"The deltas() array contains the address of the second element in each
dimension. True skip offsets for each dimension can be computed by
spec%deltas(i) - spec%base. This is not done because Fortran does not
have pointer arithmetic."
Well, in Fortran you can do pointer arithmetics - but with more effort:
You have to convert the pointer to an of type integer(c_ptrdiff) or
(c_intptr) with TRANSFER. Then you can add/subtract - and convert it
back. That works relatively well, even if it is done in a round-about
fashion and uses "processor-defined" behaviour.
Please feel free to add this to the gfortran manual, improve it, etc. This would be an amendment to section 7.1.4 of the manual at:
http://gcc.gnu.org/onlinedocs/gfortran/Interoperable-Subroutines-and-Functions.html
Well, I think it would better fit to
http://gcc.gnu.org/onlinedocs/gfortran/Further-Interoperability-of-Fortran-with-C.html,
which already refers to Chasm.
Still, I think TS29113 is more convenient. One just includes on the C
(C++, ...) side one header file:
http://gcc.gnu.org/viewcvs/gcc/branches/fortran-dev/libgfortran/ISO_Fortran_binding.h.tmpl?&view=markup#l1
And one has then full access to Fortran's array descriptor, e.g.
http://gcc.gnu.org/viewcvs/gcc/branches/fortran-dev/gcc/testsuite/gfortran.dg/iso-ts-29113_1_c.c?view=markup#l1
http://gcc.gnu.org/viewcvs/gcc/branches/fortran-dev/gcc/testsuite/gfortran.dg/iso-ts-29113_2_c.c?view=markup#l1
http://gcc.gnu.org/viewcvs/gcc/branches/fortran-dev/gcc/testsuite/gfortran.dg/iso-ts-29113_3_c.c?view=markup#l1
(Links refer to the Fortran-dev branch, which is used to define a new
array descriptor; it already works quite well, but has several known
issues. Hence, it shouldn't be used in production code.)
See here for description of what I've done:
http://jerseybiker.blogspot.com/2013/10/passing-assumed-shape-arrays-between.html
Does anyone know how to contact the people involved with the relevant ISO spec?
ISO/IEC TS 29113:2012
I would start by visiting http://j3-fortran.org/.
Curious question: Why do you want to contact the committee, which
developed TS29113? Do you miss some functionality? (Which ones?) I think
they do plan to add some features for the next Fortran standard.
(Schedule: ftp://ftp.nag.co.uk/sc22wg5/N1951-N2000/N1979.txt )
Tobias