I have uploaded a new version of the document on the ticket. It corrects the info argument problem and lists an item for discussion for the next telecon. https://svn.mpi-forum.org/trac/mpi-forum-web/ticket/287 -- Pavan On 08/26/2011 02:18 AM, Pavan Balaji wrote:
On 08/26/2011 02:12 AM, Supalov, Alexander wrote:
Thanks. The advice to implementers mentioned an Info argument (see 23/6). There appears to be no MPI_Info argument in the syntax (see 22/26). Is this a remnant of the ol'times, or a glint of the future?
Actually, having an MPI_Info here would be very tempting, as one could, e.g., ask the new communicator ranks to be reordered to honour cache hierarchy, or latency, etc. With the time we could then also agree on some keys/effects common to all implementations.
Oops, the info argument was always the plan. I somehow messed up the prototype. Thanks for catching it.
We did have a detailed discussion on whether we can predefine some info arguments, but there was no consensus. So we decided to leave it as implementation specific. If you have any ideas, please feel free to propose them.
-- Pavan
-- Pavan Balaji http://www.mcs.anl.gov/~balaji