Updates as of 17:30 US Pacific time

On 2/1/2012 1:45 PM, Paul H. Hargrove wrote:

The following two remain, as far as I know, UNRESOLVED:

+ hwloc-1.3.1 "gmake check" failure on Solaris-10/SPARC/gccfss
http://www.open-mpi.org/community/lists/hwloc-devel/2012/01/2729.php
For this compiler the ffs() implementation is missing from libgcc (https://forums.oracle.com/forums/thread.jspa?threadID=1997328)
Hwloc could either:
1) Detect this compiler at configure time and modify misc.h to pretend ffs() is missing
2) Document this compiler as broken/unsupported

I've posted a patch to implement approach #1.
Others will need to decide if the approach is acceptable.



+ hwloc-1.3.1 assertion failures on Linux/POWER7
http://www.open-mpi.org/community/lists/hwloc-devel/2012/01/2733.php
Brice has requested more info off-list, which I have provided.
There are actually 2 issues in that thread:
1) XLC appears to be doing bad things w/ the set_affinity code (also seen by Chris Samuel).

I tracked this down to xlc passing a configure test that should have failed.
The work-around is "CFLAGS=-qhalt=e", and there have been discussions about how to enforce that.
Approaches range from just documentation to various levels of "enforcement".
The discussion has not moved toward a specific resolution.

2) Assertion failures, possibly related to IBM's virtualization, and present w/ both gcc and xlc.

Brice has stated that use of Linux cgroups can reproduce something similar to my virtual node.
He says that has exposed problems that may or may not be related to mine.


-Paul
-- 
Paul H. Hargrove                          PHHargrove@lbl.gov
Future Technologies Group
HPC Research Department                   Tel: +1-510-495-2352
Lawrence Berkeley National Laboratory     Fax: +1-510-486-6900