[petsc-dev] MPI_Exscan

[Aron Ahmadia]

Lisandro raises an interesting point.  I just checked the MPI_VERSION
defined on the currently released IBM BlueGene/P driver, and I was surprised
to see 2.1 defined, given that the BlueGene doesn't support dynamic process
control (though I guess otherwise it is basically a full MPI-2), from the
developer's guide:

"The implementation of MPI on the Blue Gene/P system is the MPICH2 standard
that was developed by Argonne National Labs. For more information about
MPICH2, see the Message Passing Interface (MPI) standard Web site at:

http://www-unix.mcs.anl.gov/mpi/

A function of the MPI-2 standard that is not supported by Blue Gene/P is
dynamic process management (creating new MPI processes).16 However, the
various thread modes are supported".

I think this is worth considering carefully as MPI-3 is approved and we
start to see vendor implementations of some of its cooler features show up
on our systems.  Perhaps in addition to version numbers specific 'features'
could be macro-defined by the standard: MPI_HAS_EXSCAN, etc...?  This might
be a reasonable compromise between the two approaches.

A

On Tue, Aug 2, 2011 at 4:56 AM, Lisandro Dalcin <dalcinl at gmail.com> wrote:

> On 31 July 2011 14:34, William Gropp <wgropp at illinois.edu> wrote:
> > I'm referring to the MPI_VERSION and MPI_SUBVERSION.
> > I'm in general a supporter of feature checks, but in the case of MPI, the
> > documentation is very precise about the content of each version and there
> > are comprehensive tests for MPI implementations that have ensured
> compliance
> > with the spec.  The MPI Forum has repeatedly rejected any MPI subset,
> > ensuring that MPI_VERSION and MPI_SUBVERSION can be used.
>
> So, suppose an MPI implementation advertises 1.3 compliance (by
> defining  MPI_VERSION as 1 and MPI_SUBVERSION as 3), and implements
> some calls from MPI 2.0 like MPI_Exscan() ? Using the version macros
> would prevent PETSc to use the MPI_Exscan() available in the
> implementation as an extension... Or does "The MPI Forum has
> repeatedly rejected any MPI subset" means that such implementations
> are not conforming with the MPI std?
>
>
> --
> Lisandro Dalcin
> ---------------
> CIMEC (INTEC/CONICET-UNL)
> Predio CONICET-Santa Fe
> Colectora RN 168 Km 472, Paraje El Pozo
> 3000 Santa Fe, Argentina
> Tel: +54-342-4511594 (ext 1011)
> Tel/Fax: +54-342-4511169
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20110802/a60b4f94/attachment.html>