[Mpi3-hybridpm] Fwd: Re: First cut at slides
jhammond at alcf.anl.gov
Mon Jul 1 11:06:16 CDT 2013
My 10 am meeting is running over (again) so I will once again miss this or be extremely late.
----- Original Message -----
From: "Pavan Balaji" <balaji at mcs.anl.gov>
To: mpi3-hybridpm at lists.mpi-forum.org
Sent: Monday, July 1, 2013 11:03:49 AM
Subject: [Mpi3-hybridpm] Fwd: Re: First cut at slides
FYI, Here are the slides we are discussing.
-------- Original Message --------
Subject: Re: [Mpi3-hybridpm] First cut at slides
Date: Sat, 22 Jun 2013 13:27:59 +0000
From: Jeff Squyres (jsquyres) <jsquyres at cisco.com>
To: Pavan Balaji <balaji at mcs.anl.gov>
CC: <mpi3-hybridpm at lists.mpi-forum.org> <mpi3-hybridpm at lists.mpi-forum.org>
On Jun 22, 2013, at 9:03 AM, Pavan Balaji <balaji at mcs.anl.gov> wrote:
> Thanks. The slides seem to be confusing collective vs. synchronizing. I think you mean to say that INIT/FINALIZE are always collective, and they *might* be synchronizing (though we expect most implementations to not be synchronizing when they are not actually initializing or finalizing).
How so? I thought slide 13 was fairly explicit about that:
- INIT and FINALIZE still collective
- Continue to not specify if they synchronize or not
(although to be fair, I'm not sure if that is a local edit or not -- I
can't remember offhand what was in the last version of the PPTX I sent
out, so I've attached a current copy of the slides; see slide 13)
> Also, doesn't IS_THREAD_MAIN need to be thread safe as well, like QUERY_THREAD_LEVEL?
Yes, I missed that -- fixed.
> I agree with JeffH that CONNECT/DISCONNECT is not a problem. MPI cannot actually finalize in the first call to MPI_FINALIZE in your example, since the ref-count didn't reach zero.
I guess my point is that process A must wait for the *actual*
finalization in process B (which is the 2nd one), and if Finalize
synchronizes (which most.. but not all.. MPI implementations do), then A
will block until B calls the 2nd Finalize.
But I think I'm coming to the conclusion that if you want to not be
affected by Finalize-possibly-blocking semantics, then you should just
call MPI_COMM_DISCONNECT first.
jsquyres at cisco.com
For corporate legal information go to:
Mpi3-hybridpm mailing list
Mpi3-hybridpm at lists.mpi-forum.org
Argonne Leadership Computing Facility
University of Chicago Computation Institute
jhammond at alcf.anl.gov / (630) 252-5381
ALCF docs: http://www.alcf.anl.gov/user-guides
More information about the mpiwg-hybridpm