Bug status and Stage 3

[jdelia@intec.unl.edu.ar]

Dear all,

Salvatore Filippone <salvatore.filippone@uniroma2.it> ha escrito:

>> Dear All,
>>
>> I had a dip into the co-array proposal today.  This looks like quite a
>> lot of work, although it would be very worthwhile.
>>
>> Sometime soon we are going to have to have a #gfortranconference to
>> decide our priorities for gcc-4.5.
>
> If my user perspective is of any help, I would say that jumping onto
> this part of the Fortran 2008  standard BEFORE having the bulk of
> Fortran 2003 done, makes no sense.

That depends. I am also a Fortran user. I am interested in both "full  
POO" with Fortran as well as in coarrays. In the last case as an  
alternative to MPI or OMP (omitting HPF or PVM).

It is known that the inclusion of corrays  is a controversial subject  
(e.g how many parallel models should be in the standard and why).  
Nevertheless its inclusion would be specially welcome for cluster  
programming with single or multicore nodes. The lack of an option for  
"distributed arrays" in the standard is rather a problematic missing  
feature for cluster programming.

In fact, I am greatly suprised with the new inclusion of corray in the  
g95 compiler as an extension. The package was installed in our cluster  
without any trouble and, at least, simple tests with images  
distributed across the cluster nodes were OK, a feature that, for  
instance, with OMP would be not possible.

> Indeed I am rather surprised that the committee included substantial new
> features at a time when no fully compliant F2003 compiler is available
> (IIRC we now have XLF 12 essentially complete, but it's only been out
> for what? A couple months?); had it been me, I would only have
> sanctioned submodules (which where a TR and an extension of concepts
> already in).

There are several reasons by which J3 had included substantial new  
features as well as, at the present time, there is not a fully  
compliant F2003 compiler available.

> BTW, this is not to say that co-array are not interesting; they are, but
> I have seen too many attempts at defining "THE" parallel extension for
> "THE" programming language to take them at face value, much less
> throwing substantial resources at them right now.

Yes, again, corrays  is a controversial subject, e.g. how many  
parallel models should be in the standard and which would be the best  
one.

Regards,
Jorge.
--

----------------------------------------------------------------
This message was sent using IMP, the Internet Messaging Program.