(j3.2006) [ukfortran] (SC22WG5.5059) WG5 vote on draft TS on further coarray features

Bill Long longb
Tue Aug 6 09:20:15 EDT 2013



On 8/5/13 7:26 PM, Van Snyder wrote:

> It seems that if an image on an SMP node has several cores, between
> which work is shared using OpenMP, one needs an intrinsic, perhaps
> having nothing to do with coarrays, that shows how many cores are
> available for that image.  Is this already part of OpenMP?

Why is anything like that needed?  OpenMP directives are typically 
independent of the  number of threads.    The number is relevant when 
you launch the program.  OpenMP provides an environment variable to 
specify the number of threads.  Similarly, program launchers may specify 
this value.  The program can, and should be, written independent of how 
many threads/image and how many images are used for execution.

>
> Are coarrays really intended to be a replacement for both OpenMP and
> MPI, as originally sold at Tokyo (and earlier), or just for MPI?  It was

Coarrays are part of the Fortran 2008 parallel programming feature.  The 
feature is designed so that it can be implemented in a shared memory 
environment, a distributed memory environment, or a combination of both. 
   That's it.  What you decide to "replace" is up to you.

The names "MPI" and "OpenMP" are not part of the normative text in the 
standard.  There is an implicit assumption, for practical reasons, that 
implementations will "play nice" with these two models so that 
incremental code change is possible.   Formal specification of those 
interactions are the responsibilities of the MPI and OpenMP groups 
respectively.

> my understanding that the distinction between communication using an
> interconnect between nodes (whether or not using a global address
> space), and using shared memory on a single node, would be handled
> "under the covers" by a competent coarray implementation.  Is that no
> longer the expectation?

That has always been the expectation.  I have no idea why you imagine 
otherwise.

Cheers,
Bill

-- 
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





More information about the J3 mailing list