This Week's Schedule

[Jeffrey Oldham]

Here are excerpts from the schedule constructed by Scott, Stephen,
Jim, and I for Pooma optimization work.

**** This Week 30Sep-06Oct: Correctness and More Measurements

Tau: parallel profiling code
  JDO will ensure installation and integration with Pooma on Nirvana and Linux
  JDO and Stephen will determine places in Pooma code to measure

correctness of regression tests, examples, benchmarks
  Using QMTest testsuites to automate testing, we need to ensure that
  the invariant that

    All Pooma regression tests, benchmarks, and examples compile and
    correctly execute.

  JDO: distribute QMTest testsuites for regression tests, benchmarks, and examples
    statigraphic flow and possibly Caramana hydrodynamics revisions
  Stephen: Doof2dField corrections
  Julian: particle regression tests, benchmarks, and examples
  Scott: examples except Caramana, statigraphic flow, and examples
  unassigned: other benchmarks (is this set empty?)

correctness: memory leaks
  Jim will use check for memory leaks in regression tests, benchmarks,
  and examples using Purify or other memory checking tools on nirvana
  and linux.  Presumably memory leaks are independent of the compiler.

benchmark creation
  ABCTest: JDO: ensure there exists a multipatch compressible version with
    nontrivial initial conditions
  Doof2d: merge various codes into one program
    ensure field version runs
    write version using relations
    mostly finished week of 24->29Oct by Stephen and JDO
  Krylov solver: JDO
    ensure field version runs
    write version using relations
  scalar advection:
    Scott: convert to new field abstraction
  statigraphic flow
    JDO: ensure compilation and execution
    JDO: add loops and input parameters for various problem sizes, etc.

measurements:
  JDO, Stephen, Scott, Jim
  The program configurations to measure include:
    brick, 1-processor
    multipatch uncompressed, 1-processor, various patch sizes
    multipatch compressed, 1-processor, various patch sizes
    multipatch uncompressed, multiprocessor, various patch sizes
    multipatch compressed, multiprocessor, various patch sizes
  Hopefully, measuring Doof2d will occur during the week of 24->29Sep
  and extending this work to other executables will be trivial.

  Most of the time will be refining what to measure and understanding
  the results.

  Jim will specialize in measuring programs that are running cross-box.


Thanks,
Jeffrey D. Oldham
oldham at codesourcery.com