[Mpi3-hybridpm] MPI_Info argument in the thread start/end calls
balaji at mcs.anl.gov
Wed Jun 3 13:53:54 CDT 2009
I was discussing the thread start/end call proposal
(https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/161) with Alexander
Supalov (Intel) and Abhinav Vishnu (PNNL). One point that came up in the
discussion was with respect to the MPI_Info argument.
The first version of the API did not have this argument. However, a few
telecons back, there was a discussion on this and we decided to add the
MPI_Info argument. Is this argument needed? Alexander pointed out (and I
agree) that if there is a fixed set of hints that we are planning to
pass to the MPI implementation, then making them as true parameters
might be better. On the other hand, having MPI_Info as an argument would
allows us to later add implementation-specific extension hints as well;
but are there thoughts on what these hints might be?
While I don't strongly feel one way or the other, I'd like to reinitiate
this discussion and see if we can come to a consensus on this.
More information about the mpiwg-hybridpm