1. 03 Apr, 2015 3 commits
    • Paul Coffman's avatar
      ROMIO One-sided Collective IO Aggregation base code · 0e4dcc43
      Paul Coffman authored and Rob Latham's avatar Rob Latham committed
      
      
      Optimized collective IO aggregation algorithm added for GPFS which replaces the
      existing two-phase aggregation with one-sided MPI_Put and MPI_Get for writing and
      reading respectively.  Significant performance and memory optimization possible
      for many workloads.  Guarded by the GPFSMPIO_AGGMETHOD environment variable --
      see ad_gpfs_tuning.c for details.
      Signed-off-by: Rob Latham's avatarRob Latham <robl@mcs.anl.gov>
      0e4dcc43
    • Kenneth Raffenetti's avatar
      portals4: increase max_entries limit · c6c0d6f6
      Kenneth Raffenetti authored
      Increases the number of matching list entries we can append. The netmod
      would bump into this limit on the Get portal when doing many sends
      above the large threshold.
      
      No reviewer.
      c6c0d6f6
    • Kenneth Raffenetti's avatar
      portals4: fixup for [7d1fd292] · d1f533e6
      Kenneth Raffenetti authored
      When cleaning up uses of the max_eqs limit, I missed where we were passing
      it to the Rportals layer. Use EVENT_COUNT instead, same as the PtlEQAlloc
      calls.
      
      No reviewer.
      d1f533e6
  2. 02 Apr, 2015 2 commits
  3. 01 Apr, 2015 1 commit
  4. 31 Mar, 2015 4 commits
  5. 30 Mar, 2015 1 commit
  6. 28 Mar, 2015 1 commit
  7. 27 Mar, 2015 3 commits
    • Pavan Balaji's avatar
      Bump up the bcast2/bcast3 timeouts a bit more. · 5538b7ce
      Pavan Balaji authored
      FreeBSD seems to take slightly more than 20 minutes to run these tests
      primarily because the poll() call is somehow superslow on that OS.
      This new timeout should be sufficient to cover that case.
      
      No reviewer.
      5538b7ce
    • Huiwei Lu's avatar
      Fixes build failure with per-object thread-cs · 983a60e8
      Huiwei Lu authored
      When configured with --enable-thread-cs=per-object, MPICH build will go
      wrong because MPID_cc_t will be defined as OPA_int_t, which can not be
      directly compared with or assigned with int.
      
      This patch adds one macro,  MPID_cc_get, for correctly reading
      MPID_cc_t. Also, change writing to MPID_cc_t to use MPID_cc_set.
      
      Fixes #2250
      983a60e8
    • Kenneth Raffenetti's avatar
      testsuite: Makefile cleanup · ed383176
      Kenneth Raffenetti authored
      The nonblocking4 test was renamed and moved to the errors directory
      in [a0ebb441].
      
      No reviewer.
      ed383176
  8. 26 Mar, 2015 1 commit
    • Kenneth Raffenetti's avatar
      gforker: fixup for [7bb8bd59] · b6a5b179
      Kenneth Raffenetti authored
      
      
      The abort fix for gforker was not agressive enough about cleaning up
      MPI processes. An abort could lead to a situation where some MPI
      processes were left running on the node after mpiexec has exited. Now
      we force kill all processes before exit. Refs #2249
      Signed-off-by: default avatarHuiwei Lu <huiweilu@mcs.anl.gov>
      b6a5b179
  9. 25 Mar, 2015 1 commit
    • Kenneth Raffenetti's avatar
      gforker: implement PMI_Abort handler · 7bb8bd59
      Kenneth Raffenetti authored
      Since [b1e89abf
      
      ], PMI_Abort sends an abort command to the PMI server
      and relies on it to cleanup the application. In the case of gforker,
      nothing was done with this command, leaving the PMI client hanging
      waiting for a response. Gforker will now exit with the requested abort
      code upon receiving the command. Fixes #2249
      Signed-off-by: default avatarHuiwei Lu <huiweilu@mcs.anl.gov>
      7bb8bd59
  10. 20 Mar, 2015 2 commits
    • Pavan Balaji's avatar
      Reduce the number of processes in the commcreatep test. · cc663e76
      Pavan Balaji authored
      
      
      This test is run with 64 processes.  On platforms that do not schedule
      that many processes well (such as FreeBSD), this can slow down the
      test a lot and cause timeouts.
      Signed-off-by: Rob Latham's avatarRob Latham <robl@mcs.anl.gov>
      cc663e76
    • Wesley Bland's avatar
      Extend timeouts for putfence1/getfence1 · 5c060dfd
      Wesley Bland authored
      These tests take a longer time on OSX due to the poor performance of
      malloc/calloc on OSX platforms. The extended timeout gives them a chance
      to complete.
      
      This could be better fixed in the future by only allocating the memory
      for the window once instead of over and over (~5000 times).
      
      No reviewer
      5c060dfd
  11. 16 Mar, 2015 11 commits
  12. 15 Mar, 2015 1 commit
  13. 13 Mar, 2015 1 commit
  14. 10 Mar, 2015 1 commit
  15. 09 Mar, 2015 3 commits
  16. 06 Mar, 2015 3 commits
  17. 05 Mar, 2015 1 commit
    • Sangmin Seo's avatar
      Do not reference a CH3 variable in ROMIO glue. · c75f7cda
      Sangmin Seo authored and Rob Latham's avatar Rob Latham committed
      
      
      Although MPIDI_CH3I_progress_blocked is a variable only used in CH3,
      it was referenced in the ROMIO glue code. This caused a build problem
      when pamid is used as a device. This patch removed the reference to
      MPIDI_CH3I_progress_blocked, but it degrades the efficiency of
      MPIR_Ext_cs_yield_allfunc_if_progress_blocked() since we do not have
      a way to check if the progress engine is blocked for now (related to
      ticket #2202). For a better solution for ticket #2202, we need to
      fix a wait function of the extended generalized request.
      
      Fixes #2242
      Signed-off-by: Rob Latham's avatarRob Latham <robl@mcs.anl.gov>
      c75f7cda