[Mpi3-hybridpm] Reminder for the hybrid telecon tomorrow

[Jeff Squyres (jsquyres)]

On Apr 24, 2013, at 10:46 AM, Jeff Hammond <jhammond at alcf.anl.gov> wrote:

>> a) it could solve a lot of Jeff Hammond's originally-cited problems
> 
> I think I said already that MPI_INIT* cannot be a solution to my
> problem because it is collective and I might not want to use the
> theoretical threaded library described in my ticket on every MPI
> process.

Rain on my parade.  Matches the weather here, today, too.

But I guess this brings up another subtle point (if no one brought it up already): since INIT is collective, if we have multiple threads invoking INIT simultaneously in each MPI process, we effectively have a matching problem -- how do you match each INIT (or even just the first/actually-initializing INIT) from each process?

The obvious answer is to add a tag to INIT.  But I will shoot anyone who suggests such a monstrosity (that doesn't solve all matching issues, anyway).  And then I will deny that I ever mentioned it.

-- 
Jeff Squyres
jsquyres at cisco.com
For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/