Bill, Pavan, Rajeev, Brian, Dave and Jeff, Bill, thank you for the ticket. Pavan, thank you for supporting it in the March meeting (Rajeev will not be there). Is there a RMA WG Meeting at the March Forum Meeting? Will you do an MPI-3.0 errata plenary reading or will you put it into the errata by WG dicision, because it is only an example? In both cases #413 should be latest tomorrow on the agenda. Because it is one block of text at one precise location, the ticket format may be enough formalism, i.e., no extra pdf. Brian, Dave, Jeff, please can you confirm on ticket 413 https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/413 Best regards and thank you all for your support on this question Rolf ----- Original Message -----

From: "William Gropp" <[email protected]> To: "Rolf Rabenseifner" <[email protected]> Cc: "Pavan Balaji" <[email protected]>, "Rajeev Thakur" <[email protected]> Sent: Saturday, February 15, 2014 2:59:56 PM Subject: Re: [mpiwg-rma] [EXTERNAL] Re: Synchronization on shared memory windows

I've created #413 to help us keep track of this, and I'll add it to the list of open RMA tickets.

Bill

William Gropp Director, Parallel Computing Institute Deputy Director for Research Institute for Advanced Computing Applications and Technologies Thomas M. Siebel Chair in Computer Science

University of Illinois Urbana-Champaign

On Feb 14, 2014, at 5:33 PM, Rolf Rabenseifner wrote:

Bill, Pavan, and Rajeev,

I would add the following example at the end of MPI-3.0 Sect. 11.7, i.e., after page 461 line 20:

-------------------- Example 11.13 The following example demonstrates the rules for the unified memory model if the data transfer is implemented only with load and store (instead of MPI_PUT or MPI_GET) and the synchronization between the processes is done with MPI communication (instead of RMA synchronization routines). The used synchronization must be supplemented with MPI_WIN_SYNC which acts only local as a processor-memory-barrier. For MPI_WIN_SYNC, a passive target epoch is established with MPI_WIN_LOCK_ALL.   X is part of a shared memory window and should be the same memory location in both processes. The first synchronization finishes the load/store access epoch in A and starts the access epoch in B. The second synchronization transfers the access epoch back to A.

Process A               Process B

MPI_WIN_LOCK_ALL(       MPI_WIN_LOCK_ALL( MPI_MODE_NOCHECK,win)   MPI_MODE_NOCHECK,win)

DO ...                  DO ...  X=...

 MPI_F_SYNC_REG(X)  MPI_WIN_SYNC(win)      MPI_Send                MPI_Recv                          MPI_WIN_SYNC(win)                          MPI_F_SYNC_REG(X)

                         print X

                         MPI_F_SYNC_REG(X)                          MPI_WIN_SYNC(win)      MPI_Recv                MPI_Send  MPI_WIN_SYNC(win)  MPI_F_SYNC_REG(X)                        END DO                  END DO

MPI_WIN_UNLOCK_ALL(win) MPI_WIN_UNLOCK_ALL(win) --------------------

This is the result of the discussion and identical to the implemented example. It is tested with mpich.

Best regards Rolf

----- Original Message -----

From: "William Gropp" < [email protected] >

To: "Pavan Balaji" < [email protected] >

Cc: "Rajeev Thakur" < [email protected] >, "Rolf Rabenseifner" < [email protected] >

Sent: Thursday, February 13, 2014 10:31:45 PM

Subject: Re: [mpiwg-rma] [EXTERNAL] Re: Synchronization on shared memory windows

More precisely, we need exactly the text that would be inserted.  I

do view this as a minor change; as an example, it cannot change the

standard, even if we get it wrong.

Bill

William Gropp

Director, Parallel Computing Institute

Deputy Director for Research

Institute for Advanced Computing Applications and Technologies Thomas

M. Siebel Chair in Computer Science

University of Illinois Urbana-Champaign

On Feb 13, 2014, at 2:00 PM, Balaji, Pavan wrote:

If this is for 3.1, it needs to be read this time.  We need more than

a ticket.  We need a full draft.

— Pavan

On 2/13/14, 1:59 PM, "Thakur, Rajeev" < [email protected] > wrote:

Pavan,

Would you be able to create the ticket? I won't be at the

next Forum mtg either.

Rajeev

On Feb 13, 2014, at 10:08 AM, William Gropp < [email protected] >

wrote:

Yes, examples should be the responsibility of the chapter committee.

However, a ticket for something as large of this would help the

chapter committee with deciding exactly what to do.

Bill

William Gropp

On Feb 13, 2014, at 6:44 AM, Balaji, Pavan wrote:

Actually Bill and Rajeev are the RMA WG leads.  cc’ed Bill

I’ll be happy to help, of course.  Aren’t examples chapter committee

prerogative anyway (so they don’t need a vote)?

— Pavan

On 2/13/14, 4:09 AM, "Rolf Rabenseifner" < [email protected] >

wrote:

Pavan and Rajeev (head of mpiwg-rma?),

could you please file a ticket and convince the Forum

at the meetings that it should go into MPI-3.1

because it is a weak (because it is only an example)

clarification on why and when MPI_WIN_SYNC is needed

on shared memory windows.

Therefore, it should not wait until MPI-4.0.

-- Dr. Rolf Rabenseifner . . . . . . . . . .. email [email protected] High Performance Computing Center (HLRS) . phone ++49(0)711/685-65530 University of Stuttgart . . . . . . . . .. fax ++49(0)711 / 685-65832 Head of Dpmt Parallel Computing . . . www.hlrs.de/people/rabenseifner Nobelstr. 19, D-70550 Stuttgart, Germany . . . . (Office: Room 1.307) <halo_1sided_store_win_alloc_shared_othersync.c> <halo_1sided_store_win_alloc_shared_othersync_20.f90> <halo_1sided_store_win_alloc_shared_othersync_30.f90>

-- Dr. Rolf Rabenseifner . . . . . . . . . .. email [email protected] High Performance Computing Center (HLRS) . phone ++49(0)711/685-65530 University of Stuttgart . . . . . . . . .. fax ++49(0)711 / 685-65832 Head of Dpmt Parallel Computing . . . www.hlrs.de/people/rabenseifner Nobelstr. 19, D-70550 Stuttgart, Germany . . . . (Office: Room 1.307)