[Mpi3-hybridpm] Hybrid WG meeting at the Forum

[Douglas Miller]

Re: SHMEM ALLOC: I thought we were considering adding a "const char *key"
param so that shared memory allocations better match OS (e.g. POSIX) shmem
allocations and also to allow the call to be non-synchronizing.

Re: HELPER THREADS:

A) I don't see mention of the exception to MPI threading rules for the
MPI_HELPER_* calls. Was that intentionally removed? Did I forget a
discussion of that? I was thinking we need to state to implementers that
MPI_HELPER_* must be thread safe (in all modes except MPI_THREAD_SINGLE)
and to users that these calls may be used by any/all threads in any mode
(except MPI_THREAD_SINGLE). Do we need to discuss this more?

B) We did not finish discussion on whether all communications that are
started after JOIN must be completed (MPI_WAIT et al.) before JOIN. I think
this is important, if not for correctness then because it allows the
implementation to be more optimal and so far I cannot think of a reason one
would need to make communcations span the LEAVE.

C) here is an example use:
(if this is not an interesting enough example, let me know what you'd like
to see)

Example 1: OpenMP program with distinct compute/communicate phases

A simple example where one thread performs a blocking allreduce but the
rest of the threads lend themselves as helpers. The omp barrier is to
ensure that all endpoints are available when the allreduce begins.

MPI_Init_thread(&argc, &argv, MPI_THREAD_FUNNELED, &threading);
/* more program setup... */
tid = gettid();
#pragma omp parallel num_threads(N) {

	tno = omp_get_thread_num();
	MPI_Helper_team team;
	MPI_Helper_team_create(tid, omp_get_num_threads(), &team);

	/*
	 * some parallel computation may occur here...
	 */

	/***** communications phase begins: *****/
	MPI_Helper_join(team);
	#pragma omp barrier
	if (tno == 0) {
		MPI_Allreduce(...);
	}
	MPI_Helper_leave(team);
	/***** communications phase ends. *****/

	/*
	 * more computation and/or communication phases
	 */

	MPI_Helper_team_free(&team);
}



_______________________________________________
Douglas Miller                  BlueGene Messaging Development
IBM Corp., Rochester, MN USA                     Bldg 030-2 A410
dougmill at us.ibm.com               Douglas Miller/Rochester/IBM


                                                                           
             Pavan Balaji                                                  
             <balaji at mcs.anl.g                                             
             ov>                                                        To 
             Sent by:                  mpi3-hybridpm at lists.mpi-forum.org   
             mpi3-hybridpm-bou                                          cc 
             nces at lists.mpi-fo                                             
             rum.org                                               Subject 
                                       [Mpi3-hybridpm] Hybrid WG meeting   
                                       at the Forum                        
             12/05/2010 09:31                                              
             PM                                                            
                                                                           
                                                                           
             Please respond to                                             
             mpi3-hybridpm at lis                                             
             ts.mpi-forum.org                                              
                                                                           
                                                                           




Folks,

I've uploaded a new draft of the EI chapter on the wiki. This is some
edits based on the discussion during the previous call. Things left to
be done:

1. Examples

2. Discussion of persistent threads vs. non-persistent threads (in the
context of OpenMP).

3. Additions from the endpoints proposal

I'll be adding these this week. However, for the Forum, I think we can
get started with presenting what we have.

Unfortunately, I'll not be attending the Forum meeting this time. Is
anyone from the working group going to be there? Will you be able to
bring up the edited chapter and read out the changes to make sure there
are no major complaints from the Forum?

Thanks,

  -- Pavan

--
Pavan Balaji
http://www.mcs.anl.gov/~balaji
_______________________________________________
Mpi3-hybridpm mailing list
Mpi3-hybridpm at lists.mpi-forum.org
http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-hybridpm
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mpi-forum.org/pipermail/mpiwg-hybridpm/attachments/20101206/cd95485b/attachment-0001.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: graycol.gif
Type: image/gif
Size: 105 bytes
Desc: not available
URL: <http://lists.mpi-forum.org/pipermail/mpiwg-hybridpm/attachments/20101206/cd95485b/attachment-0003.gif>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: pic12955.gif
Type: image/gif
Size: 1255 bytes
Desc: not available
URL: <http://lists.mpi-forum.org/pipermail/mpiwg-hybridpm/attachments/20101206/cd95485b/attachment-0004.gif>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: ecblank.gif
Type: image/gif
Size: 45 bytes
Desc: not available
URL: <http://lists.mpi-forum.org/pipermail/mpiwg-hybridpm/attachments/20101206/cd95485b/attachment-0005.gif>