1. 17 Sep, 2009 1 commit
  2. 28 Aug, 2009 1 commit
  3. 12 Jun, 2009 1 commit
  4. 31 Mar, 2009 1 commit
  5. 08 Mar, 2009 1 commit
  6. 04 Mar, 2009 1 commit
    • Pavan Balaji's avatar
      [svn-r3938] Use MPI_Ssend instead of MPI_Send for Gatherv with large communicators · c96f0509
      Pavan Balaji authored
      (more than 32 processes for now). Just using plain MPI_Send was causing
      the non-root processes to run ahead and queue up a lot of unexpected
      messages on the root processes. This was: (i) causing performance loss,
      and (ii) causing some MPICH2 derivatives (such as BG/P) to abort due to
      lack of resources. We don't use MPI_Ssend for small communicators as it
      adds overhead.
      
      Reviewed by gropp.
      c96f0509
  7. 05 Jan, 2009 1 commit
  8. 25 Nov, 2008 1 commit
    • Darius Buntinas's avatar
      [svn-r3608] Reimplemented MPIR_Localcopy to copy noncontig data using... · 51b229cf
      Darius Buntinas authored
      [svn-r3608] Reimplemented MPIR_Localcopy to copy noncontig data using segment_pack and _unpack.  This fixes ticket #188 and avoids nested calls to progress.  Currently this copies noncontigs through a temporary buffer, but ideally this could be done with a single copy using a segment_transpack-type function.
      51b229cf
  9. 13 Oct, 2008 2 commits
  10. 11 Oct, 2008 1 commit
  11. 09 Oct, 2008 1 commit
  12. 24 Sep, 2008 1 commit
  13. 19 Jun, 2008 1 commit
  14. 18 Jun, 2008 1 commit
  15. 02 Nov, 2007 1 commit