1. 23 Jan, 2015 2 commits
  2. 22 Jan, 2015 2 commits
    • Huiwei Lu's avatar
      FT: Fixes ref counts in shrink and agree · 93e816cc
      Huiwei Lu authored
      
      
      When process fails, fault tolerance scheme takes a different path to
      deal with MPI object reference counts than the existing one. Some
      reference counts were not properly set in FT path so when configured
      with --enable-g=all, some ft tests will show leaked context id, dirty
      COMM, GROUP and REQUEST objects and so on when exit.
      
      This patch fixes ft/shrink and ft/agree with "--enable-g=all". Stack
      allocated objects of requests, communicators and groups will be freed by
      FT.
      Signed-off-by: default avatarWesley Bland <wbland@anl.gov>
      93e816cc
    • Wesley Bland's avatar
      Fix for MPIX_COMM_AGREE to not return incorrect errors · a3dd5f40
      Wesley Bland authored
      
      
      MPIX_Comm_agree should not return errors if the failed processes have
      all been acknowledged. Previously, it was returning errors
      unnecessarily, but this makes sure that the errcode is MPI_SUCCESS when
      appropriate.
      Signed-off-by: default avatarHuiwei Lu <huiweilu@mcs.anl.gov>
      a3dd5f40
  3. 16 Jan, 2015 1 commit
  4. 15 Jan, 2015 1 commit
    • Wesley Bland's avatar
      Refactor MPI_Testall to have an MPIR version · 4be54219
      Wesley Bland authored and Kenneth Raffenetti's avatar Kenneth Raffenetti committed
      
      
      For some reason, there was no MPIR_Testall_impl as there is with many of
      the other MPI_* functions. This causes a linking problem when weak
      symbols are disabled and another MPI function needs to call MPI_*.
      
      This patch moves most of the MPI_Testall code into MPIR_Testall_impl and
      has MPI_Waitall call that function instead of MPI_Testall.
      Signed-off-by: Kenneth Raffenetti's avatarKen Raffenetti <raffenet@mcs.anl.gov>
      4be54219
  5. 14 Jan, 2015 3 commits
  6. 13 Jan, 2015 2 commits
    • Wesley Bland's avatar
      Remove ADI breakage introduced earlier · 6f646ca0
      Wesley Bland authored
      
      
      There was an accidental ADI breakage earlier when MPI level codes would
      query into the dev part of the MPID request object. This commit removes
      that breakage by adding a new macro into the mpiimpl.h file to portably
      check whether a request is anysource. For now, in pamid, this macro
      always evaluates to 0. This can easily be fixed by overwriting it in the
      pamid code, but since pamid doesn't support FT, it won't have any
      functional change either.
      Signed-off-by: default avatarHuiwei Lu <huiweilu@mcs.anl.gov>
      6f646ca0
    • Wesley Bland's avatar
      Move MPID_Comm_AS_enabled to MPID layer function · d1ab9e68
      Wesley Bland authored
      
      
      It was pointed out that by putting this in a macro and failing silently
      when unimplemented, this make things challenging for derivatives which
      will implement this function in the future. By moving this to an MPID
      level function, it becomes more obvious that the function should be
      implemented later.
      Signed-off-by: default avatarHuiwei Lu <huiweilu@mcs.anl.gov>
      d1ab9e68
  7. 12 Jan, 2015 9 commits
    • Wesley Bland's avatar
      Change MPIDI_CH3I_Comm_AS_enabled to be MPID level · 8cbbcae4
      Wesley Bland authored
      
      
      This macro was used inside CH3 to determine if the communicator could be
      used for anysource communication. With the rewrite of the anysource
      fault tolerance logic, it is now necessary to use it at the MPI level.
      Because it is a macro and not a function, the macro is defined in
      mpiimple.h as (1) and then overwritten in the ch3 device. Future devices
      can also overwrite it if desired.
      Signed-off-by: default avatarHuiwei Lu <huiweilu@mcs.anl.gov>
      8cbbcae4
    • Wesley Bland's avatar
      Handle anysource in blocking recv functions · 9f2db553
      Wesley Bland authored
      
      
      If a blocking recv function (MPI_Recv and MPI_Sendrecv) includes an
      MPI_ANY_SOURCE and there is a failure, handle it by cleaning up the
      request and returning MPIX_ERR_PROC_FAILED.
      Signed-off-by: default avatarHuiwei Lu <huiweilu@mcs.anl.gov>
      9f2db553
    • Wesley Bland's avatar
      Make anysource test more accurate · 648cd48c
      Wesley Bland authored
      
      
      Test for the specific error code so it doesn't accidentally catch
      MPI_ERR_OTHER.
      Signed-off-by: default avatarHuiwei Lu <huiweilu@mcs.anl.gov>
      648cd48c
    • Wesley Bland's avatar
      Allow MPIR_Request_complete to take a NULL request · f6cdb3c8
      Wesley Bland authored
      
      
      If the first argument is NULL, don't try to set it to MPI_REQUEST_NULL.
      For blocking functions that want to complete the MPID_Request object,
      this allows them to reuse the code.
      Signed-off-by: default avatarHuiwei Lu <huiweilu@mcs.anl.gov>
      f6cdb3c8
    • Wesley Bland's avatar
      Handle anysource in the wait* functions · 0418d495
      Wesley Bland authored
      
      
      If a wait operation involves an anysource, we need to first check to
      make sure that they haven't been disabled. If they have been, convert
      the wait* function to a test* function to prevent deadlocking inside the
      progress engine.
      Signed-off-by: default avatarHuiwei Lu <huiweilu@mcs.anl.gov>
      0418d495
    • Wesley Bland's avatar
      Break out of progress for anysource failures · 50d85e51
      Wesley Bland authored
      
      
      If a failure is detected, even if no request is actually complete, the
      completion counter will be incremented now as a way to give control back
      to the MPI layer to let it decide whether or not to continue.
      
      This gives the request completion functions a chance to see if they're
      waiting on an MPI_ANY_SOURCE request and if so, to return an error
      indicating that the completion function has a
      MPIX_ERR_PROC_FAILED_PENDING failure that the user needs to acknowledge.
      
      All of these functions should go into the progress engine at least once
      as a way to ensure that even if they will be returning an error, they'll
      at least give MPI a way to make progress and potentially still complete
      the request objects even if the user never acknowledges the failure.
      
      A follow on commit will add the functionality to keep the progress
      engine from getting stuck if a failure is discovered before entering the
      completion function.
      Signed-off-by: default avatarHuiwei Lu <huiweilu@mcs.anl.gov>
      50d85e51
    • Wesley Bland's avatar
      Strip out pending ANY_SOURCE request handling · 7a785c84
      Wesley Bland authored
      
      
      The existing way that we handle non-blocking requests involving wildcard
      receive operations is incorrect. We're cancelling request operations and
      trying to recreate them later. In the meantime, it's messing with
      matching and makes it possible (likely?) that some messages that arrive
      will never be matched. A new way of handling this is coming next.
      Signed-off-by: default avatarHuiwei Lu <huiweilu@mcs.anl.gov>
      7a785c84
    • Wesley Bland's avatar
      Don't free a request if it still pending · a96ac72e
      Wesley Bland authored
      
      
      If we had a failure that caused a request to be pending, we were freeing
      the request before calling the error handler. That caused segfaults. Now
      we switch the ordering of the two to avoid that.
      
      This also moves the assignment of the status_ptr to be a little earlier
      to avoid another segfault.
      Signed-off-by: default avatarHuiwei Lu <huiweilu@mcs.anl.gov>
      a96ac72e
    • Kenneth Raffenetti's avatar
      rptls: do not send pause/unpause messages to self · 47ac6f6b
      Kenneth Raffenetti authored
      
      
      CH3 ensures that self communication does not go through the netmod,
      so there is no need for a process to pause/unpause itself.
      Signed-off-by: default avatarAntonio J. Pena <apenya@mcs.anl.gov>
      47ac6f6b
  8. 08 Jan, 2015 2 commits
  9. 07 Jan, 2015 2 commits
  10. 06 Jan, 2015 1 commit
    • Kenneth Raffenetti's avatar
      Fortran profiling interface fix · 6ce38c27
      Kenneth Raffenetti authored
      
      
      Previous re-organization of the library symbols resulted in a
      situation where Fortran programs could no longer be profiled using
      tools written in C. Functions in libmpifort directly called the
      PMPI_* versions in libmpi.
      
      Now we always call the MPI_* versions from libmpifort. In the case
      where we are building a separate profiling library, we use a new
      preprocessor flag to ensure we call PMPI_* from inside libpmpi.
      
      Additional bug fix:
        - always define mpi_conversion_fn_null_, there is no pmpi version
      
      Fixes #2209
      Signed-off-by: default avatarJunchao Zhang <jczhang@mcs.anl.gov>
      6ce38c27
  11. 05 Jan, 2015 6 commits
  12. 04 Jan, 2015 2 commits
  13. 19 Dec, 2014 1 commit
    • Paul Coffman's avatar
      barrier in close whenever shared files supported · ef1cf141
      Paul Coffman authored and Rob Latham's avatar Rob Latham committed
      
      
      Currently in the MPI_File_close there is a barrier in place whenever the
      ADIO_SHARED_FP feature is enabled AND the ADIO_UNLINK_AFTER_CLOSE
      feature is disabled right before the code to close the shared file
      pointer and potentially unlink the shared file itself.  PE testing on
      GPFS revealed a situation using the non-collective
      MPI_File_read_shared/MPI_File_write_shared
      where based on this implementation all tasks needed to wait for all
      other tasks to complete processing before unlinking the shared file
      pointer or the open of the shared file pointer could fail.  This
      situation is illustrated as follows with the simplest example of 2 tasks
      that do this:
      MPI_File_Open
      MPI_File_set_view
      MPI_File_Read_shared
      MPI_File_close
      
      So both tasks call MPI_File_Read_shared at the same time which first
      does the ADIO_Get_shared_fp which does the file open with create mode on
      the shared file pointer.   Only 1 task can actually create the file, so
      there is a race to see who can get it done first.  If task 0 gets it
      created then he is the winner and goes on to use it, read the file and
      then MPI_File_close which then unlinks the shared file pointer first and
      then closes the output file.  Meanwhile, task 1 lost the race to create
      the file and is in error, the error handling in gpfs goes into effect
      and task 1 now just tries to open the file that task 0 created.  The
      problem is this error handling took longer that task 0 took to read and
      close the output file, so at the time when task 0 does the close he is
      the only process with a link since task 1 is still in the create file
      error handlilng code so therefore gpfs goes ahead and deletes the shared
      file pointer.  Then when the error handling code for task 1 does
      complete and he tries to do the open, the file is no longer there, so
      the open fails as does the subsequent read of the shared file pointer.
      Currently GPFS has the ADIO_UNLINK_AFTER_CLOSE  feature enabled, so the
      fix for this is to remove the additional condition of
      ADIO_UNLINK_AFTER_CLOSE  being disabled for the barrier in the close to
      be done.  Presumably this could be an issue for any parallel file system
      so this change is being done in the common code.
      
      See ticket #2214
      Signed-off-by: default avatarPaul Coffman <pkcoff@us.ibm.com>
      Signed-off-by: Rob Latham's avatarRob Latham <robl@mcs.anl.gov>
      ef1cf141
  14. 18 Dec, 2014 1 commit
  15. 17 Dec, 2014 2 commits
  16. 16 Dec, 2014 3 commits
    • Xin Zhao's avatar
      Simplify epoch checking in Win_lock · 52531f77
      Xin Zhao authored
      when lock_epoch_count != 0, we only need to check if access_state
      is PER_TARGET in Win_lock.
      
      No reviewer.
      52531f77
    • Xin Zhao's avatar
      Bug-fix: store lock entry pointer in request even data is dropped · 17202c7c
      Xin Zhao authored
      When data is dropped but lock is queued, we should still store
      the lock entry in current request, so that we can try to acquire
      the lock when we received and dropped all data.
      
      No reviewer.
      17202c7c
    • Xin Zhao's avatar
      Bug-fix: first dequeue lock entry from queue, then perform op in this entry · a44c53d6
      Xin Zhao authored
      Here we should first dequeue the current lock queue entry from
      lock queue then performing the operation in it. This is because
      when performing op in current lock entry, we may trigger
      release_lock() function, which go to check the lock queue again.
      If we did not remove current entry from the queue, release_lock()
      will try to process it for the second time, which leads to the wrong
      execution.
      
      No reviewer.
      a44c53d6