Software Coordinating Committee Conference Call
		       June 27, 2001 1:00 EDT
			  Recorder: C. DeTar

Present:

Celso, Holmgren, Mawhinney, DeTar, Watson, Edwards, Brower

** = action required

1.) Future meeting times:

    **Reply about proposed dates

2.) Subcommittees, one messaging, one linear algebra
    We agreed to Rich's suggested membership:

    Messaging:  Chip, Robert, Celso, Rich
    Linear Alg: Bob, Carleton, Don

3.) Agenda item du jour: Messaging Passing API
    
    We discussed Chip's and Robert's version 0.4

    Initialization of layout: 

    We assume that the operating system imposes constraints on the
    layout, depending on the machine.  So the generateTopology layout
    routine would interrogate the system to learn the constraints and
    complete the job modulo the remaining flexibility.

    Message Declarations:

    CD advocates allowing message passing across arbitrary
    permutations of sites.  Do we want this?

    Bob mentioned QCDOC capability of ganging multiple sends.
    Can this be done?

    Reductions should be deterministic.
    Would be nice to also have global max and reduction according
    to a user-supplied two-argument function.

    Bob: What about sums over a single time slice?  Perhaps OK to do
    this with global sums for now.


    Rich: We need to devise intuitive naming conventions for these
    routines.

    Bob: Let's use symbolic notation for the shift directions.

    Rich: Future evolution messaging API software should include a 
	  "run everywhere" MPI implementation with capable of generating 
          performance data for the target Terascale architectures.
          (Celso will work on this.)

    ** Messaging subcommittee to carry on

Next call on July 9 at 1 PM EDT to continue API discussion.

Call concluded at 2:55 PM EDT.