MPI Forum Meetings logo

MPI Forum: mpi-21 Mailing List Archives

all MPI Forum: mpi-21 mailing list

Subject: [mpi-21] Ballot 4 - User defined datarep - was Re: Question to MPI/IO
From: Rolf Rabenseifner (rabenseifner_at_[hidden])
Date: 2008-01-31 11:48:57


This is a proposal for MPI 2.1, Ballot 4.

I'm asking especially
  Hubert Ritzdorf, Jean-Pierre Prost, John May, Bill Nitzberg,
the participants of the email-discussion in 1999, to review this proposal.

This is a follow up to:
  Interpretation of user defined datarep
  in http://www.cs.uiuc.edu/homes/wgropp/projects/parallel/MPI/mpi-errata/index.html
with mail discussion in
  http://www.cs.uiuc.edu/homes/wgropp/projects/parallel/MPI/mpi-errata/discuss/datarep/
___________________________________

Proposal:
MPI-2.0 Sect. 9.5.3 User-defined Data Representations, page 254,
lines 13-15 read:
  Then in subsequent calls to the conversion function,
  MPI will increment the value in position by the count of items
  converted in the previous call.
 
but should read:
  Then in subsequent calls to the conversion function,
  MPI will increment the value in position by the count of items
  converted in the previous call, and userbuf is kept unchanged.
  
___________________________________
Rationale for this clarification:
  It was not clear, whether the userbuf pointer must also be moved
  in the subsequent calls. This clarification was already done in
  1999 and should already be implemented in existing implementations
  of user-defined data representations.
___________________________________
Total text page 254 lines 8-15:
  If MPI cannot allocate a buffer large enough to hold all the
  data to be converted from a read operation, it may call the
  conversion function repeatedly using the same datatype and
  userbuf, and reading successive chunks of data to be converted
  in filebuf. For the first call (and in the case when all the data
  to be converted fits into filebuf), MPI will call the function
  with position set to zero. Data converted during this call will
  be stored in the userbuf according to the first count data items
  in datatype. Then in subsequent calls to the conversion function,
  MPI will increment the value in position by the count of items
  converted in the previous call,
  (new text added:)
  and userbuf is kept unchanged.
___________________________________

Best regards
Rolf

Dr. Rolf Rabenseifner . . . . . . . . . .. email rabenseifner_at_[hidden]
High Performance Computing Center (HLRS) . phone ++49(0)711/685-65530
University of Stuttgart . . . . . . . . .. fax ++49(0)711 / 685-65832
Head of Dpmt Parallel Computing . . . www.hlrs.de/people/rabenseifner
Nobelstr. 19, D-70550 Stuttgart, Germany . (Office: Allmandring 30)