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

Sun Feb 16 12:19:19 CST 2014

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" <wgropp at illinois.edu>
> To: "Rolf Rabenseifner" <rabenseifner at hlrs.de>
> Cc: "Pavan Balaji" <balaji at anl.gov>, "Rajeev Thakur" <thakur at anl.gov>
> 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" < wgropp at illinois.edu >
> 
> 
> To: "Pavan Balaji" < balaji at anl.gov >
> 
> 
> Cc: "Rajeev Thakur" < thakur at anl.gov >, "Rolf Rabenseifner" <
> rabenseifner at hlrs.de >
> 
> 
> 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" < thakur at anl.gov > 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 < wgropp at illinois.edu >
> 
> 
> 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" < rabenseifner at hlrs.de >
> 
> 
> 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 rabenseifner at hlrs.de
> 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 rabenseifner at hlrs.de
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)