Software Coordinating Committee Conference Call
		       January 3, 2002 1:00 PM EST
			  Recorder: C. DeTar

Present:

  Brower, DeTar, Mawhinney, Mendez, Holmgren, Watson (snowbound), 
      Edwards (snowbound)

Absent:
   
  Pochinski

** Action items

     1. Still need help in getting together the Software Slides for
        the poster for the SciDAC PI meeting next week.

        Maybe create a single large poster - easy to do after the fact
        Mendez: svPablo graphics
        Robert: P4 Benchmarks for bar chart graphics Chris McLendon
        Holmgren: P4 data
        Watson: MP-API graphic
**      Any other suggestions:

     2. Vicky White and Robin Staffin from the DOE are visiting BU
        next Monday Jan 7 and MIT on Tuesday Jan 8.

        Specifically here are the questions Vicky sent to Brower:

       Brower;  Vicky White and Robin Staffin want to know
   >      discussion of participation - who, where;
   >      discussion of problem areas - what, why;
   >      and is there anything we can do to help.

**     Give me your input. Especially helpful to me (if not in the
       meeting in an Email message) is an eplicit list of all
       supported SciDAC personnel and their assigned tasks. I want to
       make an accurate slide for Vicky's visit which I can defend.
       Not that she asks what we need to succeed. So if there are
       obvious unmeet needs I won't hesitate to pass this along.

    3. Quarterly report at all hands meeting 

**     All to contribute

    4. All hands Jlab meeting date Feb 1, Feb 2
       Claudio Rebbi is organizing content of all hands mtg
**     Brower: Try for software meeting Saturday afternoon
        and evening.

    5. Performance tools
       Brower: What are tradeoffs between ease-of-use coding and 
       more difficult, but optimized coding?
       Can we use performance tools to give programmer good advice?
       Can use Celso's svPablo? How much does it need to be
       changed?
       Celso: Can't see many changes required.  Can add mirror library
       for function passing (same routines that incorporate
       profiling).
       Brower: Since MILC code has been used as a guinea pig for
       svPablo, perhaps what we need is documentation to show how it
       can be applied to QCD code.
       Celso: P4 not supported by PAPI - perhaps in a few months,
       it will.
       Brower: What abou QCDOC?  
       Mawhinney: Simulator works at gate level, so gives all the
       information needed.  But difficult to use.  So community
       resource would be to provide optimized staggered and Wilson
       inverters as models for writing optimized code.  Current
       performance: Staggered single processor 70% (Christian), Wilson
       80% (Boyle).

    6. Continuing saga level 2 design and data layout
         Edwards: Subsets of sites could be implemented via
          a copy with mask.
         Brower: Robert, Bob, and Carleton to work further
          on details.
         DeTar: We need to hear from rest of committee whether the
          general direction makes sense.
         Bob: Level 3 should operate on a predefined generic
          layout for the purpose of writing optimized QCDOC Level 3 code.
         Bob: The Level 3 routine would be required to do remapping of
          input and output to achieve compatibility with a Level 2
          layout.
         Brower: Should we publish this generic layout as a default
          standard for Level 2?
         DeTar: What does publish mean?  I would hope it means merely
          advising the QCDOC user to link with "layout_qcdoc.o" to
          avoid remapping overhead, but allowing the user to make any
          other choice of layout on a different platform.  It
          shouldn't mean the user can make additional assumptions
          about where data is without breaking the Level 2 protocol.
         Robert: By agreeing to remap in going from Level 2 to Level 3
          and back, we preserve the data parallel design and hide the
          layout details from the user.

Next call Thursday, January 10, 3 PM EST

Conference concluded at 3 PM EST.
- ------- End of forwarded message -------
------- End of forwarded message -------