(j3.2006) Nick's concerns about coarrays
Wed Nov 5 14:17:42 EST 2008
On Wed, 2008-11-05 at 05:40 -0800, N.M. Maclaren wrote:
> In my view, the real concern is whether they will work RELIABLY on all
> platforms. This means without artificial deadlock, with only a
> bounded performance degradation, with specified and comprehansible
> behaviour, and without the code doing the wrong thing on occasion
> because the specification is essentially unimplementable.
> And that is why I am going to Tokyo! We need to ensure that is
> possible before including them.
Nick has hit the nail on the head here. Debates over whether Fortran
needs a parallel abstraction are too late by a decade or two. The
questions for me:
1. Are coarrays a superior abstraction to what we have already? I
haven't heard any arguments that PVM, MPI, SHMEM, OpenMP, HPF,
pthreads... are better. The question "In the fullness of time might we
come up with something better?" is pointless, because the answer is
almost certainly yes. Using that as an excuse not to proceed leads to
paralysis, and not just concerning coarrays. Thankfully, we got
external subprograms in Fortran II, which kept us going for thirty years
until we got modules in Fortran 90.
2. Can coarrays fit into current and expected architectures? Given that
processors can exploit subclause 1.3 to set the maximum number of images
they support to 1, I think the answer is yes. The effort to parse the
syntax and report violations of constraints is less than, say,
submodules but more than, say, allowing an empty CONTAINS part.
3. Can we preserve the traditional scalar/vector language and its
behavior when coarrays aren't used? We've been careful to try to
achieve this. Have we?
4. Have we gotten the specification right? I think Nick is saying "not
5. Have we gotten the words to implement the specification right?
Probably not; we haven't gotten that completely right for anything yet.
6. Assuming I'm wrong about 4 and 5, can the specification be
implemented? Cray says yes. Nick appears to have said "maybe only Cray
can do it."
Van Snyder | What fraction of Americans believe
Van.Snyder at jpl.nasa.gov | Wrestling is real and NASA is fake?
Any alleged opinions are my own and have not been approved or
disapproved by JPL, CalTech, NASA, the President, or anybody else.
More information about the J3