[mpich-discuss] Indexed MPI_Reduce_local

[Dave Goodell]

On Dec 28, 2012, at 7:33 PM CST, Jed Brown wrote:

> The context is a comm library on top of MPI. I provide sparse communication with respect to a user-provided "unit" datatype (usually basic types or small contiguous structs).
[…]
> I could implement this operation in-place using MPI_Accumulate on COMM_SELF, but I'd rather avoid MPI_Accumulate because it doesn't work with quad-precision or with MINLOC on {long; long}. Related tickets include:
> https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/338

We definitely want to support this case (#338), but for a number of reasons haven't gotten around to implementing it yet.

> https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/318

I actually did a bunch of work in a private branch to support these types, but stopped short of finishing it when the overall Forum opinion on them turned rather sour.  Perhaps we should restart the effort on this (both standardization and implementation) now that MPI-3.0 is out.

> https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/319

I don't have any objection to these types, they just need to be implemented.  We can put them in as MPIX_ types for now.

> Any other suggestions?
> 
> In lieu of a more elegant solution, I'm going to match on the types and ops that I need to support and write the sparse reduction by hand.
> 
> In the longer term, I'm curious whether the MPICH developers, and eventually the Forum, may be interested in providing better support for higher level comm libraries like this.

Like motherhood and apple pie, of course the answer is yes :)  The devil is always in "to what extent".

-Dave