In lam_basic, they're all O(N) if there are <=4 MPI processes in the
communicator; otherwise, O(log N).
The smp algorithms are based on the MagPIe algorithms. Those are a two
step process; a selection is used for the on-node processes and
off-node processes (i.e., either lam_basic or shmem will be used).
The shmem algorithms are essentially O(N).
Look in share/ssi/coll for the actual code.
Hope that helps.
On Nov 14, 2004, at 2:55 AM, Elankovan wrote:
> Hi,
>
> Can someone please tell me where I can get information on the
> algorithm used for all the available collective communication routines
> in LAM/MPI?
>
> Thanks.
>
> Regards,
> Elan.
>
> Do you Yahoo!?
> Check out the new Yahoo! Front Page.
> www.yahoo.com_______________________________________________
> This list is archived at http://www.lam-mpi.org/MailArchives/lam/
--
{+} Jeff Squyres
{+} jsquyres_at_[hidden]
{+} http://www.lam-mpi.org/
|
