[mpich-discuss] broadcast and reduce mechanism
Gra zeus
gra_zeus at yahoo.com
Wed Aug 12 18:31:10 CDT 2009
thank you very much. This paper and source codes help me alot.
--- On Wed, 8/12/09, Rajeev Thakur <thakur at mcs.anl.gov> wrote:
From: Rajeev Thakur <thakur at mcs.anl.gov>
Subject: Re: [mpich-discuss] broadcast and reduce mechanism
To: mpich-discuss at mcs.anl.gov
Date: Wednesday, August 12, 2009, 8:59 AM
The code is in the src/mpi/coll directory. See bcast.c and reduce.c.
Rajeev
> -----Original Message-----
> From: mpich-discuss-bounces at mcs.anl.gov
> [mailto:mpich-discuss-bounces at mcs.anl.gov] On Behalf Of Dave Goodell
> Sent: Wednesday, August 12, 2009 9:01 AM
> To: mpich-discuss at mcs.anl.gov
> Subject: Re: [mpich-discuss] broadcast and reduce mechanism
>
> You can find some information in this paper:
> http://www.mcs.anl.gov/~thakur/papers/mpi-coll.pdf
>
> However since that was written there have been a few changes, some
> major and some minor. One difference is that we now perform those
> collective operations hierarchically on SMP systems (intranode,
> internode, then intranode again). Also, additional algorithms might
> have been selected and the cutoff points are almost certainly
> different. So the best way to figure out what's going on in
> there is
> to read the code.
>
> As far as I know we don't use sequential communication to implement
> any of our collective operations.
>
> -Dave
>
> On Aug 11, 2009, at 9:34 PM, Gra zeus wrote:
>
> > Hello,
> >
> > Can any one comfirm or link me to documents that explain
> > communication mechanism of MPI_Bcast and MPI_Reduce?
> >
> > I searched in google and It said Bcast,Reduce can use tree
> algorithm
> > or sequential communication.
> > However, I can't find any specification that indicate
> communication
> > mechanism of MPI_Bcast and MPI_Reduce in MPICH2.
> >
> > thx,
> > gra
> >
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/mpich-discuss/attachments/20090812/e4b68f00/attachment.htm>
More information about the mpich-discuss
mailing list