Currently, GCC only supports OpenMP on shared-memory systems.
It would be nice if it could be extended to support OpenMP via the ethernet,
infiniband etc. (distributed-memory systems).

There exists (at least) one implementation of distributed-memory OpenMP: The
Intel Compilers 9.1's Cluster OpenMP.

Intel's Whitepaper:
http://www.intel.com/cd/software/products/asmo-na/eng/compilers/285865.htm

Intel's Cluster OpenMP User manual:
http://softwareforums.intel.com/attachments/ids/11/3796/1/UsersGuide.pdf

Intel's Extentions to OpenMP:
"– the sharable directive. The sharable directive identifies
variables that are referenced by more than one thread. These
variables are the ones that are managed by the DSM system."
(Some are automatically marked shareable, see whitepaper, p. 5)

The whitepaper also describes some implemenation details to deal with global
memory.

Running such a Intel-Cluster-OpenMP program: ./a.out, which first reads the
config file then starts the program on the remote computer via rsh or ssh. The
data transfer via TCP or Direct Access Provider Library (DAPL.sf.net).

 * * *

Besides OpenMP there exists another multi-processor extention which works at
compile time: Cray's co-arrays, scheduled (in modified form) to be included
into the Fortran 2008 standard (see http://www.co-array.org/). Having a look at
the co-array specs shouldn't harm, when implementing this. Maybe some different
or further things should be considered.


-- 
           Summary: Support Cluster OpenMP (distributed-memory OpenMP)
           Product: gcc
           Version: 4.2.0
            Status: UNCONFIRMED
          Severity: enhancement
          Priority: P3
         Component: other
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: tobias dot burnus at physik dot fu-berlin dot de


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27541