Open MPI logo

Open MPI Development Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Development mailing list

Subject: [OMPI devel] MPI 2.1
From: Jeff Squyres (jsquyres_at_[hidden])
Date: 2008-08-28 08:38:24


Short version:
--------------

Per tickets 1236 and 1237, the "MPI_SUBVERSION" #define in mpi.h will
soon be increasing from 0 to 1 (i.e., MPI 2.0 -> 2.1). The work is
currently pending in an hg tree; I'm just waiting for a review of my
changes before putting it all back to the SVN trunk.

Longer version:
---------------

In working on some MPI-2.1 items (cartesian and graph communicators in
tickets https://svn.open-mpi.org/trac/ompi/ticket/1236 and https://svn.open-mpi.org/trac/ompi/ticket/1237)
, I expanded the existing cartesian/graph tests in the ompi-tests IBM
test suite to test for the new behavior.

I used the MPI_VERSION and MPI_SUBVERSION #define's in mpi.h to
conditionally activate the new testing code. Otherwise, the newly-
expanded IBM tests would be unsuitable for testing older versions of
OMPI (e.g., the v1.2 series).

This meant that I needed to distinguish the newer OMPI from the older
OMPI -- so I bumped up the MPI_SUBVERSION from 0 to 1. So trunk/v1.3
mpi.h will shortly advertise:

#define MPI_VERSION 2
#define MPI_SUBVERSION 1

This is appropriate because we'll be conforming to MPI-2.1 as soon as
these two 2.1 tickets are reviewed/committed, and one more 2.1 ticket
is complete (https://svn.open-mpi.org/trac/ompi/ticket/1239).
Hopefully, that'll be RSN. Perhaps I'll dig into this on the plane
trip to Dublin so that we can report good 2.1 status at the Forum
meeting...

-- 
Jeff Squyres
Cisco Systems