[ODK] Meeting 2018 11 21

LinBox meeting 2018-11-21

Winter meeting

• A Breust would prefer attending 5-6-7 (could come with train or car and return with the Grenoble group)

Done (or in progress) this week

• Zhu:

  • cra-omp: in a branch, increase test-solveCRA to test omp
    • current implem uses parallel regions only (do not need tasks)
      • surprisingly working and fastest
      • investigate if we need to port it in paladin/ or write a better
    • version with tasks
  • cra-mpi: cleanup and test coverage
    • bugs small sizes + Hadamard
    • inconsistent systems mod p sometimes -> real bug since Ax=b has a solution over QQ hence it must have at least one solution mod pi
    • TODO: identify where the bug is (either a wrong statement "inconsistent", or a bug in full-rank generator)
    • TODO: update usage of seeds and randomness, to allow user to reproduce a bug in the testsuite.
    • then it will make it easier to track the "inconsistency" bug

• Alexis:

  • [M2] almost done but still a bug in MPI detection on Clement's machine
  • [M3] cleanup solveCRA-mpi de Zhu en cours
  • [SV1] cleanup de solutions/solve.h -> pb de design de l'integration MPI et OMP à discuter
  • [SV2] test-suite:
    • Faut-il tester les matrices de rationnels ?
    • oui, faire le cas matrice rationnelle, mais ne pas passer de tps à le faire passer -> future work

Pb this week

• Hadamard and its usage in solve seems broken: eg 1x1 system -> refacto of Hadamard as a solution
• masked prime generator system used in solveCRA breaks: master process uses another generator (not masked).