[Mpi3-hybridpm] Proposal for MPI + OpenMP coexistance

Anthony Skjellum tony at runtimecomputing.com
Wed Aug 12 19:27:14 CDT 2009


We had an explicit context model in Zipcode much for the same reasons... Allow communication structures to be built in several ways, not only with specific collective operation over a group... 

We went through some iterations together in MPI1 to make the context implicit/opaque :-). The concepts presented here are welcome generalizations...

Tony

Tony Skjellum, PhD
RunTime Computing Solutions, LLC
tony at runtimecomputing.com
direct: +1-205-314-3595
cell:    +1-205-807-4968

-----Original Message-----
From: Snir Marc <snir at illinois.edu>

Date: Wed, 12 Aug 2009 14:30:37 
To: mpi3-hybridpm at lists.mpi-forum.org<mpi3-hybridpm at lists.mpi-forum.org>
Subject: Re: [Mpi3-hybridpm] Proposal for MPI + OpenMP coexistance


The proposal for MPI extensions is unrelated to OpenMP -- it could be  
used to have MPI interoperate with Pthreads. As you observed. it is  
about separating the concept of an MPI endpoint -- a communication  
resource, and the thread that executes MPI calls -- a computation  
resource. So, the proposed MPI extensions provide an API to create MPI  
endpoints and to associate them with threads. This API is agnostic  
about the mechanisms used to create and  coordinate threads.

The proposal has three parts:
1, MPI extensions -- agnostic to the thread programming model
2. A binding for OpenMP
3. A description of a "static programming model" for MPI+OpenMP

I did 2 and 3 since MPI+OpenMP seems to be the most important  
combination of message-passing + shared memory.
...
>
> One question, though: is this proposal specifically OpenMP oriented?  
> I saw a notice about other threading models and languages closer to  
> the end, but I don't remember seeing Posix threads, etc., mentioned.  
> If I didn't simply miss this: is this a matter of principle  
> (pthreads can do more that OpenMP), an attempt to keep things simple  
> for starters, or a consequence of that OpenMP is considered as the  
> most popular threading layer used currently for hybrid programs?
>
> Best regards.
>
> Alexander
>

Marc Snir
Michael Faiman and Saburo Muroga Professor
Department of Computer Science, University of Illinois at Urbana  
Champaign
4323 Siebel Center, 201 N Goodwin, IL 61801
Tel (217) 244 6568
Web http://www.cs.uiuc.edu/homes/snir





_______________________________________________
Mpi3-hybridpm mailing list
Mpi3-hybridpm at lists.mpi-forum.org
http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-hybridpm




More information about the mpiwg-hybridpm mailing list