Re: relaxed 1-sided model

Raja Daoud (raja@tbag.rsn.hp.com)
Fri, 14 Jun 1996 19:13:13 CDT

> Hmm...if we do this, do we even need WINDOW_IN and WINDOW_OUT at all anymore?
> Meaning, could we get away with just using MPI_Barrier() as our progress/sync
> point?

I don't think so: Window_in allows the implementation to do a Test on
the hidden Irecv for the put/get-service, thus maintaining the pt-to-pt
nature of the transfer. Requiring Barrier would unnecessarily make it a
global operation.

--Raja

-=-
Raja Daoud Hewlett-Packard Co.
raja@rsn.hp.com Convex Division