(j3.2006) UPC Liaision
Dan Nagle
dannagle
Tue Sep 23 19:44:54 EDT 2008
Hello,
I spent two days at the UPC Workshop, where I was
very well received. Bob Numrich also attended.
UPC wants to become an International Standard, they
are getting serious attention from industry and government.
They will pursue a Category A Liaison with JTC1 as
the first step in the process. There was general agreement
that becoming an IS is the next logical step in their growth.
UPC will make some minor last-minute adjustments to their
specification while the Liaison application is processed.
Otherwise, they are ready to fast-track their specification.
UPC had great interest in working with Fortran to define
the interoperability of UPC and CAF. By happy coincidence,
our HPC subgroup is also our Interop subgroup. :-)
Cray and IBM (see below) will likely lead the effort anyway.
IBM is building a single RTL to support UPC, CAF and X10.
They expect to release sometime next year. The compilers
will then support UPC and CAF.
The HP representative told me that the HP RTL could support
CAF easily. However, HP is not currently pursuing CAF.
HP is funding support for UPC in gdb, and moving the GCC UPC
compiler into the mainline of GCC (so it will be built
as a normal part of building GCC). This means that
the gfortran coarrays effort will have the GCC UPC support
as a basis.
There is an interesting parallel performance tool being built
at U of Florida, see http://ppw.hcs.ufl.edu They are
integrating the tool into Eclipse, see http://eclipse.org/ptp
This tool does not now support CAF, but the student presenting
the paper told me he had investigated doing so, and thought
it was not too hard.
UPC is unsure how to pursue interoperability with MPI.
I suggested they liaise with MPIF.
J3/WG5 should be prepared to discuss these questions:
Do we want to pursue Interoperability of CAF & UPC?
What do we mean by Interoperability of CAF & UPC?
What is the right vehicle for doing what we want to do?
--
Cheers!
Dan Nagle
More information about the J3
mailing list