Open
Description
Reported by jhammond on 22 Dec 2013 20:05 UTC
In addition to #5, we should optimize intranode communication via shared memory windows.
Until https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/397 is resolved, we can only implement this using overlapping windows, one for internode and the other for intranode.