1. 11 Oct, 2016 1 commit
  2. 12 Nov, 2015 1 commit
  3. 10 Oct, 2015 2 commits
  4. 09 Oct, 2015 1 commit
  5. 14 Sep, 2015 1 commit
  6. 31 Aug, 2015 2 commits
    • Iulian Grindeanu's avatar
      wrong use for MPI_Gatherv in the test · 83a8c65d
      Iulian Grindeanu authored
      counts and displacements for gatherv should be initialized
      they are not output
      also, checking the ownership test was wrong
      the formula vtxdata[2*i  ] = (x << 2) | y; did not ensure uniqueness
      change it to vtxdata[2*i  ] = (x << 3) | y;
      
      this test never passed correctly
      83a8c65d
    • Dan Wu's avatar
      Fixed some memory leak issues found by valgrind on MOAB_iMeshP_unit_tests.... · d6837535
      Dan Wu authored
      Fixed some memory leak issues found by valgrind on MOAB_iMeshP_unit_tests. There are still some leaks from test_entity_owner() and test_create_ghost_ent() when this test is run in parallel (it is also noticed that make check runs it in serial).
      d6837535
  7. 13 Feb, 2014 1 commit
  8. 06 Sep, 2013 1 commit
  9. 12 Jun, 2012 1 commit
  10. 25 Jan, 2012 1 commit
  11. 30 Jul, 2011 1 commit
  12. 20 Apr, 2011 1 commit
  13. 14 Jan, 2011 1 commit
  14. 23 Nov, 2010 1 commit
    • Tim Tautges's avatar
      - changing names of a few more gs functions to avoid name conflicts with Nek · d0ca3f60
      Tim Tautges authored
      - fixing bug resolving parallel interface for cases where a face's vertices are
      shared across a number of procs but the face itself isn't shared
      - adding some additional debug output in ParallelComm used to diagnose this problem
      - fixing MOAB_iMeshP_unit_tests for when run on one proc, which was dying inside
      MPI_Gatherv in the single-proc case; shouldn't really die there, but couldn't figure out
      what else to do
      - adding ParallelComm::get_pstatus
      
      Passes make check in serial and parallel.
      
      
      
      git-svn-id: https://svn.mcs.anl.gov/repos/ITAPS/MOAB/trunk@4298 6162379e-bd28-0410-9a7d-b7f4dcfcad3c
      d0ca3f60
  15. 22 Nov, 2010 1 commit
  16. 22 Sep, 2010 1 commit
  17. 29 Mar, 2010 1 commit
  18. 22 Mar, 2010 1 commit
  19. 12 Mar, 2010 1 commit
  20. 10 Mar, 2010 1 commit
  21. 09 Mar, 2010 2 commits
  22. 30 Nov, 2009 1 commit
    • Tim Tautges's avatar
      Fairly major changes to parallel ghosting. First, implemented a handshaking approach for · 1274295f
      Tim Tautges authored
      exchanging large messages; this guarantees that a buffer will be allocated and ready
      when a large message arrives.  Second, implementing an abstract Buffer object (in this case, 
      a class in MBParallelComm) for buffering.  This provides more explicit control over memory
      management in buffers, and makes it easier to debug memory access problems.  One shortcoming 
      of the current code is that there's an MPI_Barrier call between sending/receiving entities and
      sending/receiving remote handles.  I'll try and eliminate this after parallel testing.  I'm also
      sure there are plenty of places where this code can be cleaned up and made more bulletproof,
      and where comments need to be added/improved.
      
      With this commit, parallel ghost exchange now works for all the serial tests I've run so far.
      Next up is to start running parallel and timing tests.  A couple of the iMeshP tests still fail also.
      
      
      Specific changes:
      MBSkinner: fix problem causing scdtest to die
      MBParallelConventions.h: inserted clarifying comment
      pcomm_serial: requesting ghosting using faces as bridge entities
      pcomm_unit, mbparallelcomm_test: updated to use new buffering scheme
      MBParallelComm: many changes summarized in first part of message
      iMeshP_unit_tests.cpp: adding PARTITION option
      
      
      
      
      git-svn-id: https://svn.mcs.anl.gov/repos/ITAPS/MOAB/trunk@3389 6162379e-bd28-0410-9a7d-b7f4dcfcad3c
      1274295f
  23. 02 Nov, 2009 1 commit
    • 's avatar
      o Move MPI configuration macros to a separate file · d2bdfdd2
      authored
      o Change all #includes of mpi.h to include MBmpi.h instead
      o Put better configure logic about handling of conflicts between symbols
        in the MPI C++ API and the standard C IO headers into MBmpi.h:
          - If not C++, just include mpi.h
          - If it works, just include mpi.h
          - Otherwise define MPICH_CXX_IGNORE_SEEK, if that helps
          - Otherwise do old #undef games so that at least the C
              API is usable.
      
      
      git-svn-id: https://svn.mcs.anl.gov/repos/ITAPS/MOAB/trunk@3255 6162379e-bd28-0410-9a7d-b7f4dcfcad3c
      d2bdfdd2
  24. 16 Apr, 2009 1 commit
  25. 19 Nov, 2008 1 commit
  26. 18 Nov, 2008 1 commit
  27. 15 Nov, 2008 1 commit
  28. 14 Nov, 2008 1 commit
  29. 13 Nov, 2008 6 commits
  30. 12 Nov, 2008 1 commit