I forget who asked about this at the OMPI developer's conference this week, but I misspoke the answer. The question was if the user could specify (via an MCA parameter) which cores were to be used by mpirun for an application, and then let the mapper assign the procs to those cores in a round-robin fashion. The answer is (of course) "yes".
All you have to do is specify "mpirun --cpu-list 0,2,4" to tell us to use the 0, 2, and 4 logical cpus for this application. In the absence of any other directives, we will map the procs round-robin (one per core) to the cores in that list on each node, binding as directed. This is different from --slot-list 0,2,4 as that would tell us to bind each process to the 0,2,4 logical cpus (i.e., bind the proc to all three of those cores) HTH - and sorry for the misspeak. Ralph
