[Mpi3-hybridpm] First cut at slides
sko at cray.com
Fri Jun 21 09:36:13 CDT 2013
Thanks for the use case info, that was really helpful.
Your response to the rest of the questions makes me realize
that I am reading too much into the proposal and not
catching where decisions are left to the implementation.
On 06/21/2013 07:35 AM, Jeff Squyres (jsquyres) wrote:
> On Jun 20, 2013, at 6:20 PM, Steve Oyanagi <sko at cray.com> wrote:
>> 1. Can slides be expanded or can some one explain why the items in
>> question are desirable? (i.e. use cases) One case is libraries,
>> but I suspect there are others.
> Sorry, yes, these slides assume that the reader has been attending the hybrid working group calls; they are not meant for the general forum. :-)
> The main motivating issue is the mixing of *multiple* MPI-aware libraries in a single application. Such libraries may or may not be aware of the other MPI-aware libraries.
>> 2. What does it mean if the first thread calls MPI_Init_thread with
>> required=MPI_THREAD_FUNNELD, then a later thread calls MPI_Init_thread
>> (MPI already initialized, now in "epoch" in slide-speak) with
>> required=MPI_THREAD_MULTIPLE? Does the second call see an error?
>> What does second call return for provided?
> That is up to the implementation, but I suspect that most will return in provided whatever was returned at the first call to MPI_INIT_THREAD.
>> 3. Slide 13 seems to imply that the implementation doesn't need to
>> actually implement an intermediate finalize/initialization, just
>> allow the calls. Is that correct? Does that satisfy the use
> Yes, that is one possible implementation.
More information about the mpiwg-hybridpm