DubLinBox 2010

Agenda

Topics

• Start filling git repository
  • When do we switch working on linbox-1.1.x with svn to linbox-2.0 on git ?

• Parallelism
  • (LinBox-1.8 and ParLinBox?) or (LinBox-2.0 is parallel)
  • Separate compilation
  • experiments with OMP show that OMP default is good
  • extend LBA usage
  • Define the overall macros and strategy for handling seq., OMP, kaapi, etc.
  • LinBox-1.7 is NOT thread safe

• Documentation
  • Doxygen ?
  • Specify the semantics: what is allowed, when will a generic algorithm be able to work on this, etc.
  • At least good documentation on the 10 solutions

• Design
  • MatrixDesign
  • MatrixAllocation
  • Clean-up and simplify
  • Dave's Domain design (domains handle dense matrices. Sparse matrices take care of themselves)
  • FieldAXPY sum/sumandprodcut
  • Reference counting ??? Brice sparse matrices (seq/multicore/GPU)
  • Configuration scripts

• Tests
  • Testing framework, regression tests
  • Every class MyFoo, in a "myfoofile.h" file, must have
    • a function "bool MyFooTest(int argc, char argv);" which tests its class,
    • inside a myfoofile_test.h
  • Default macros for .h single inclusions

    #ifndef __LINBOX__MyFoo_H
    #define __LINBOX__MyFoo_H
     
      ...
    #endif
    

• Separate compilation
  • Need to add a new template layer with specializations calling generic algorithms specialized and separately compiled.
  • See TestSeparate.tgz attached.

Goals (TODO before Friday night)

• Put in the git directory: Dense matrices, ffield implementations