Reimplement CmiNumNodes et al using hwloc
With hwloc in place, there is no need to maintain our custom querying of hardware PU counts.
#4 Updated by Eric Bohm about 2 months ago
The basic functionality of the current API needs to be maintained. Being able to get basic topology information with a single function call is incredibly convenient. HWLOC is much richer for the within node information, but requires the user to do a lot more work to obtain the same information.
The top level of the HWLOC ontology is "machine". We can make collections of those, but any questions about the collection would be answered by us, not HWLOC.
That said, the implementation of the within host aspects of the API using HWLOC would help clean up some redundant code and exposing more of the functionality of HWLOC would be welcome.