[Mpi3-rma] RMA chapter for quick review

Torsten Hoefler htor at illinois.edu
Fri Mar 11 13:10:08 CST 2011

Hi Brian,

Thanks for you review!

> >An updated version of the RMA chapter is attached to this mail and on the
> >wiki. We would like to make it available to the Forum by tomorrow
> >afternoon in order to meet the two-week window for the first reading at
> >the March meeting. If you have a chance, please take a look at it and let
> >us know your comments before tomorrow afternoon.
> >
> >In particular:
> > - We fixed some issues with the rules on pg 47 and 48. Please take a
> >look at those rules in their entirety.
> > - pg 3, ln 43, has been updated to specify that default value for the
> >accumulate_ops key is same_op_no_op.
> Page and line numbers are from the marked-up copy, not the clean copy.
> Pg 1, ln 45/46 & pg 2, ln 1&2: The discussion of communication operations
> being delayed for the separate model and immediate for the unified model
> is a bit confusing.  When the implementation *starts* the communication is
> the same for both models, depending on the synchronization calls.  The
> only difference between the two models is when the private and public
> copies on the target must be synchronized.  As I read the standard, an
> implantation is free to start a put immediately upon calling MPI_PUT in
> the separate model and also free to start a put during MPI_UNLOCK in the
> unified model.
It's a very high-level view - we don't separate between completion and
start at this point (in the intro). Both statements are "can" statements
and not wrong. Do you have a concrete wording proposal that makes it
less confusing?

> Pg 2, ln 3&4: I think the last sentence might read a bit better as "A
> large number of synchronization calls, which support both models, are
> provided to support difference synchronization styles."  But I'm open to
> better wording :).

> Pg 26, ln 47: There's a word missing in the final sentence of the
> paragraph.  Perhaps: "The stronger semantics of the RMA unified model
> allow the user to omit some synchronization calls and potentially improve
> performance."

> Pg 50, ln 2: "synchronization in Process B", the p in process should be
> lowercase.
> Pg 50, ln 28: p in Process A should be lowercase
> Pg 50, ln 29: p in Process B should be lowercase (twice)
> Pg 51, ln 15: p in Process A should be lowercase

> Pg 51, ln 25: Should be "local store by process B before the lock" (insert
> word process)
> Pg 52, ln 21: insert word "process" at "allow B to read", "stored by A"
> Pg 52, ln 24: reword sentence, should be "in the private copy of process A"
> Pg 52, ln 24: "the update to Y", not "the update on Y"
> Pg 52, ln 27: insert work process in "the value stored by A"
> Pg 52, ln 32: insert "process" at "private copy of X at process B"
> Pg 53, ln 13: can remove the comment, someone fixed the latex foo
All fixed.

> Pg 53, ln 42: The citation for the algorithm is missing (Torsten?)
I removed the reference because the algorithm is very well known

I will put it up at the wiki soon!

All the Best,

 bash$ :(){ :|:&};: --------------------- http://www.unixer.de/ -----
Torsten Hoefler         | Performance Modeling and Simulation Lead
Blue Waters Directorate | University of Illinois (UIUC)
1205 W Clark Street     | Urbana, IL, 61801
NCSA Building           | +01 (217) 244-7736

More information about the mpiwg-rma mailing list