Software Coordinating Committee Conference Call
		       August 9, 2001 3:00 PM EDT
			  Recorder: C. DeTar

Present:

  Brower, Pochinski, DeTar, Holmgren, Edwards, Simone, Mendez, Watson

Absent: 

  Mawhinney

Signed off early:

  Watson

** = action required

Main agenda item: Linear Algebra API
    See Brower block chart http://buphy.bu.edu/~brower/SciDAC/doc/chart.msg

    Discussion of Level 2 combos of shifts and linear algebra

    DeTar: Worry about proliferation: 
         operations x shift types x numbers of operands x which
         operands we shift = big number
    Edwards: Found a few that involve matrix multiplication to be
         most useful.  Blocking helps.
    DeTar: Dslash would be built from very specialized routines -
         e.g. a 4-way shift and 4-way multiply.  But probably
         we want a parallel transport that acts on various
         color vectors.  So we'd supply only a few of the most useful combos.
    Pochinski: Would we change the geometry - say as in multigrid?
    Pochinski: Consider any permutation of sites in place of a "shift"?
    DeTar: Probably the combos would be restricted to a few shift-like
        operations.
    Edwards: Write a C-preprocessor macro to resolve proliferation.
        Macro would take operands, a shift specification, an operator
        specification.
    DeTar: Would be nice to see an example.
    Edwards: Lazy types in C++ might help.
    Brower: So how are we going to proceed, resolving details of the
        implementation?
    DeTar: We have a framework shaping up.  Let's write a few
        mockups to see how we like them.
    Brower: Make it short because Chip and Bob are absent.

Next meeting at Lattice '01, time to be announced.
Next call Aug 30.
Call concluded at 5:00 PM EDT.



_______________________________________________
QCDapi mailing list
QCDapi@physics.bu.edu
http://physics.bu.edu/mailman/listinfo/qcdapi
------- End of forwarded message -------