Re: Clarification of close semantics

Linda Stanberry (linda@anduin.ocf.llnl.gov)
Fri, 28 Feb 1997 14:33:44 -0800 (PST)

When John and I were discussing this earlier today, we were debating
whether it was necessary to synchronize before the calls to MPI_CLOSE
on the separate processors, for example:

MPI_OPEN(comm, ..., &fh1) MPI_OPEN(comm, ..., &fh2)

MPI_BARRIER(comm) MPI_IREAD(fh2, ...)

MPI_CLOSE(fh1) MPI_BARRIER(comm)

MPI_CLOSE(fh2)

What we want to clarify is whether the user needs to guarantee that no
outstanding requests are active for the local file handle (fh1 or fh2)
or for the entire set of file handles ({fh1, fh2}) that was generated
by MPI_OPEN. If the former, then no explicit synchronization seems
needed, but if the latter, it does?

Linda Stanberry
lstanberry@llnl.gov