Open MPI logo

Open MPI Development Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Development mailing list

Subject: Re: [OMPI devel] About Marshalling and Umarshalling
From: N.M. Maclaren (nmm1_at_[hidden])
Date: 2012-11-05 10:34:52

On Nov 5 2012, Ralph Castain wrote:
> We adhere to the MPI standard, so we expect the user in such an instance
> to define a datatype that reflects the structure they are trying to send.
> We will then do the voodoo to correctly send that data in a heterogeneous
> environment, and pass the data back (in the defined datatype) to the user
> on the remote end. They can then put the various items back into their
> structure.

There is actually a serious issue, but it is about much more advanced
aspects than the OP was asking about, and the MPI standard does not
address the general problem, as far as I know. I do NOT think that
Open MPI would be wise to add it as an extension, of course :-)

Specifically, when you have a structure that is not simple, logically
process-independent, data, you have to use explicit marshalling and
unmarshalling (which are the terms Python uses, but there have been a
zillion others over the years, including export and import). That
can be as simple as a structure (not a C/C++/Fortran one) with internal
pointers, but can include cases where some objects are 'active' and
can be modified only by calling an auxiliary procedure.

If the OP needs to do that, that's what he needs to do, and should
transfer the marshalled form using MPI_BYTE or MPI_PACKED. But,
whether or not that is the case, he does need to look at MPI's
facilities in more detail.

Nick Maclaren.