openmpi

Ralph Castain a3060cdd15 Fix the bind_downward code - it was incorrectly looking across the entire node instead of only looking below the locale to which the proc had been assigned. In other words, if the proc was mapped to a core, then the only hwthreads that should be considered for binding are those directly below that core. The binding algo was incorrectly looking at ALL hwthreads in that scenario, causing the proc to be bound to an HT outside of the mapped location.

This now results in the procs being bound within their assigned location. It also causes us to use only the 0th HT on a core unless --use-hwthread-cpus has been specified (in which case, we use all the HTs in a core). Bind to core binds you to all HTs regardless - the --use-hwthread-cpus only impacts the oversubscribed determination and when binding to HT.

cmr:v1.7

This commit was SVN r27342.

2012-09-14 22:01:19 +00:00

base.h

Sorry for mid-day commit, but I had promised on the call to do this upon my return.

2012-04-06 14:23:13 +00:00

help-orte-rmaps-base.txt

Cleanly error out when someone asks to map-to <object> if that object doesn't exist on a node.

2012-08-04 21:52:36 +00:00

Makefile.am

Update the hwloc build on Windows and related files.

2012-07-20 12:14:28 +00:00

rmaps_base_binding.c

Fix the bind_downward code - it was incorrectly looking across the entire node instead of only looking below the locale to which the proc had been assigned. In other words, if the proc was mapped to a core, then the only hwthreads that should be considered for binding are those directly below that core. The binding algo was incorrectly looking at ALL hwthreads in that scenario, causing the proc to be bound to an HT outside of the mapped location.