(j3.2006) (SC22WG5.5140) From WG9 minutes

[Bill Long]

What kind of parallelism do they mean?

If SIMD / vectorization, I don't really think there is an issue. 
Compilers have been automatically generating vector code for years, 
based on loop analysis or array expressions.

If shared-memory parallelism, we already have OpenMP that works across 
languages.  Do we really need to reinvent that wheel?    Compilers do 
have the opportunity to do automatic "threading" with constructs like DO 
CONCURRENT.   But the implementation of that is basically the same as 
for OpenMP.

If the discussion is about SPMD parallelism,  I do have some 
reservations.  Fortran is already covered.  I suspect the C++ people 
like Charm++, or maybe not.  If you have a Cray, you have coarrays for 
C++ available, which is a tiny step towards commonality.  I'm not keen 
on the idea (sometimes discussed) that all SPMD parallel implementation 
must be implemented on top of MPI, which (according to the same crowd) 
is the only, true, master of the parallel universe. This ends up 
requiring that native Fortran be never faster than MPI (a goal of some 
in the MPI community, unfortunately) but that would represent a 
degradation of current experience with coarrays.  I'm certainly opposed 
to movement in that direction.  (Of course, implementing MPI in Fortran 
would be more interesting :). )

I think they need to be more clear what is actually supposed to be 
studied here.  There are many meanings to "parallelism".

Cheers,
Bill


On 12/9/13 7:41 PM, Van Snyder wrote:
>>From Canada's national body report to the WG9 (Ada) meeting in
> Pittsburgh Friday 15 November 2013:
>
>          In Canada's previous national body report to WG9, it was
>          proposed that Canada would initiate a request to SC22 at its
>          September plenary to form a study group to determine the
>          feasibility of creating a parallelism model that could be shared
>          between programming languages. Such a model ideally would allow
>          for optimal use of the system processing resources and avoid
>          over-subscription of parallelism.
>
>          The request was made at the plenary, but SC22 declined to create
>          such a study group, and instead encouraged other languages to
>          participate in CPLEX.
>
> Incidentally, national bodies from Canada, France, Italy, Portugal,
> Spain, UK, and USA presented national body reports.  In addition to the
> Ada Rapporteur Group, which plays a similar role to WG9 as J3 does to
> WG5, WG9 maintains liaisons with Ada-Europe and SIGADA.
>
>
> _______________________________________________
> J3 mailing list
> J3 at mailman.j3-fortran.org
> http://mailman.j3-fortran.org/mailman/listinfo/j3
>

-- 
Bill Long                                           longb at cray.com
Fortran Technical Support    &                 voice: 651-605-9024
Bioinformatics Software Development            fax:   651-605-9142
Cray Inc./Cray Plaza, Suite 210/380 Jackson St./St. Paul, MN 55101