Hello,
On Aug 26, 2008, at 6:08 AM, Harvey Richardson wrote:
>
> Can you tell me how many production scientific/research or ISV
> applications
> are based on a co-array implementation? What platforms do these
> run on?
Obviously, only those applications that run exclusively on Cray
computers.
That's why coarrays should be standardized.
When the great effort of parallelizing an application is made,
many times the choice of paradigm must be one that is available
on many platforms. That means, today, MPI. OpenMP, in the past
at least, is a shared-memory scheme, so it is useless on DMP clusters.
Indeed the near-necessity of using MPI is the greatest hindrance
to portably parallelizing Fortran applications today.
When I parallelize an application, I choose the paradigm based upon what
I believe *some other user* of the application has available. That is,
I base the decision upon the widespread availability of the scheme,
not just what I have locally. So far, that is the greatest hindrance
to the use of coarrays.
Note that an efficient predictor of who supports coarrays is
who parallelizes applications. Those who do, support coarrays
in the standard. Those who don't parallelize applications try to find
any excuse at hand to denigrate coarrays. As uniprocessor chips fade
into history, the excuses become thinner and thinner.
OTOH, how many parallel scientific/research or ISV applications
are implemented using Fortress? How many compiler implementors
are implementing Fortress versus how many are implementing coarrays?
--
Cheers!
Dan Nagle
|