(j3.2006) (SC22WG5.3668) [ukfortran] N1755: Request for new features from MPI Forum

[N.M. Maclaren]

On Nov 11 2008, Aleksandar Donev wrote:
>
>> This one is very simple, anyway.  As far as the program is concerned, MPI
>> transfers are semantically just a specialised form of I/O.  Does anyone
>> seriously dispute that?
>
> I am not, obviously I proposed to extend the already existing mechanisms 
> we have for async Fortran I/O to cover user-defined data transfers.

Upon thinking it over, I think that it's even simpler than I thought.  And
I am even more firmly of the view that the right attribute is ASYNCHRONOUS
and not VOLATILE - the latter is toxic in all languages that have it.

All that really needs saying (in Fortran) is that a companion processor can 
set and clear the pending attribute and create an ID in a processor 
dependent fashion. MPI then needs to say that it does precisely that - MPI 
is a perfectly respectable companion processor, after all! If an 
implementor can't get it right after that, he or she is beyond redemption.

Now, tying that up in standardsese is not not easy :-(


Regards,
Nick Maclaren,
University of Cambridge Computing Service,
New Museums Site, Pembroke Street, Cambridge CB2 3QH, England.
Email:  nmm1 at cam.ac.uk
Tel.:  +44 1223 334761    Fax:  +44 1223 334679