Re: A proposal to change the direction of 1-sided comm

Eric Salo (salo@mrjones.engr.sgi.com)
Tue, 25 Jun 1996 13:48:35 -0700

Just to follow up a bit on David's message...

The "Reader's Digest" version is that we may be able to get away with having
nothing more than GET, PUT, BARRIER, and maybe RMW/ACCUMULATE. All GET/PUT
calls become essentially asynchronous even though they don't create requests;
the contents of the buffers are undefined until the next barrier, even for
GETs. If we do this, then the concept of GET/PUT windows vanishes entirely; we
don't need windows, we don't need window counters, we don't need WINDOW_IN or
WINDOW_OUT, and we don't even need IGET or IPUT since they would be redundant.

This is in fact quite similar (I believe) to the Cray SHMEM model, expect that
MPI barriers are a bit different from SHMEM barriers and we would probably need
to expand our barrier semantics in some way.

David, it would probably be useful to post some concrete examples for the
proposed interface...

-- 
Eric Salo         Silicon Graphics Inc.             "Do you know what the
(415)933-2998     2011 N. Shoreline Blvd, 8U-808     last Xon said, just
salo@sgi.com      Mountain View, CA   94043-1389     before he died?"