Software Coordinating Committee Conference Call
		       August 15, 2002 1:00 PM EDT
			  Recorder: C. DeTar

Present:
  
  Holmgren, DeTar, Osborn, Mendez, Mawhinney, Edwards, Gregory,
  Kostya Petrov, Toussaint, Chulwoo Jun, Brower, Simone

Absent:

  Pochinsky, Watson (China)

** Action items

Agenda:

	#1. Test of QMP in MILC on the QCDOC ASIC simulator.

        Brower: Where do these things stand?  Issues that need to
           be completed this Fall.
        Jun:  MILC code with QMP from Eric - testing on emulator
           Not quite able to test in multinode form, yet.
        Brower: But aren't there tricks to get around this?
        Mawhinney: Can create multiple instances of the MILC code
           on the simulator.  But MILC code doesn't communicate
           in this mode.  Could put two QCDOC chips on the simulator
           to get around it.  Some problems arose with IBM upgrades
           to simulator that prevent this.  Not sorted out with
           IBM yet.
        Toussaint: Have successfully run a single iteration of cg on
           2^4 on one node.  
        Gregory: MILC on QMP on MPI is working.
        Mawhinney: Before the upgrades we ran a successful ethernet
           test.  Could revert to old version of 440 core to run
           the test, if necessary.
        Brower: At some point in the near future we need this test 
           in some way or another.
        Mawhinney: Need a test suite for QMP (see agenda #3 below)
        Brower: We need someone to construct a test suite for QMP with Jie
           Chen's help.
**      Simone: I could take a crack at it.  We need a test suite.
**      Mendez: I would also be willing to help.

        #2. Asqtad Dirac inverter and Force term for QCDOC

        Brower: Naive inverter exists, correct?  What about Naik?
        Mawhinney: Naive works in high performance mode.  A new 
           student, Huei Wen, is working on the Naik term.
           Calin Christian wrote naive one (he has left).
           Not time critical -- needs hardware first.
        Brower: Certainly needed before the 128 node prototype is ready.
        Toussaint: Part of the exercise is to learn what is involved
           in creating and optimizing the software.
        Mawhinney: Expect ready some time in mid-Fall.
        Toussaint: Can we test the Naik-style communication
           on the emulator to estimate the performance?
        Mawhinney: Too late to alter the chip design.
        Brower: But need to know the numbers for release of funds.
        Toussaint: We haven't learned anything about the eventual
           performance of the MILC code on the QCDOC, yet.
        Mawhinney: Write force term in QDP.
        DeTar: First we need QLA - then QDP - later in the Fall
        Mawhinney: Focus on a subset of QDP needed for the force term.
        DeTar: So we complete the definition of QDP, write
           the Asqtad force term in QDP and see what QDP routines we need.
        Brower: Agree with this approach.
        Mawhinney: Wants to see flop count schedules for force term
           to see where communication and computation can be overlapped.
        Toussaint: Total overhead of force terms plus fat links 
           in 2+1 simulation is equal to about 500 cg iterations.
           Costs about an additional factor of 2 for ~0.4m_s.
           Less of a penalty for lighter quark masses.
**      DeTar: Between Utah and Arizona, we should be able at least to
           come up with a list of QDP routines we think are needed
           to do the Asqtad force.
        Mawhinney: Also would like the op counts per routine.
        Edwards: Need to nail down the C binding for QDP.
**      DeTar: We are very close. Will send around my tweaks.
        Brower: How soon to get a QDP C implementation that works.
        Osborn: Should be fairly soon - couple of months.

        #3. QMP Test Software.

        Discussed above.

	#4. I/O routines.
	#5. Data File formats.

        Brower: Andrew - Writing I/O routines for QDP.
        DeTar: Why not the NERSC archive format?
        Edwards: Need to define the QDP interface for I/O.
**      Brower: Meet in two weeks to discuss this.
        
        #6. Current status and schedule from the QLA and QDP++ working groups.

        Brower: Discussion Aug 22 - to be convened by Robert.


- ----------------------------------------------------------------------
Conference concluded at 2:35 PM EDT.
Next plenary call Aug 29 3 PM EDT.