Folks, Based on the discussion during the telecon as well as after the telecon with several folks, I've merged the proposal into the concept of MPI Windows. I've added discussion at several places where the current semantics made it difficult to add the new functionality (which was the original motivation to fork off). The latest draft is available here: http://www.mcs.anl.gov/~balaji/tmp/rma/rma-1.6.pdf Apart from the semantics from v1.5, I've incorporated an important interoperability issue with shared memory programming pointed out by Manoj @ PNNL during the telecon. That is, MPI_WIN_LOCK needs to block if a process can do load/store accesses on a window; note that this is different from MPI_WIN_LOCK needs to be lock for "local windows" in cases where a shared memory region is being added to a window. Thanks, -- Pavan On 03/01/2010 12:57 PM, Rajeev Thakur wrote:
We are having a phone call this Wednesday at 11am Central to discuss Pavan's proposal before the meeting next week. The time will not work for everyone, but whoever can participate, please do. The current version of the proposal is at www.mcs.anl.gov/~balaji/tmp/rma/rma-1.5.1.pdf .
Rajeev
Date: 2010-03-03 (Wednesday) Time: 11am Central, 9am Pacific, noon Eastern Passcode: 11797 Domestic dial number: 888 769 8760 International dial number: 212 287 1666
_______________________________________________ mpi3-rma mailing list [email protected] http://lists.mpi-forum.org/mailman/listinfo.cgi/mpi3-rma
-- Pavan Balaji http://www.mcs.anl.gov/~balaji