Open MPI logo

Hardware Locality Development Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Hardware Locality Development mailing list

Subject: Re: [hwloc-devel] 1.3.1 and 1.4rc1
From: Jeff Squyres (jsquyres_at_[hidden])
Date: 2011-12-09 07:47:57


Two other enhancements that would be good to get (sometime -- not necessarily for 1.4):

1. Will there ever be any differentiation between cache levels in hwloc_obj.type? I ask because in OMPI, we found that the various counting routines were not helpful because they only search by *type*, not by (obj.type, obj.attr->cache.depth). This was somewhat of a bummer; we basically ended up writing our own traversal helpers in OMPI because when searching for an OBJ_CACHE, we had to search on the tuple -- not just the type.

(I think I might have brought this up before, but I don't remember where that discussion ended up...)

2. It would be helpful to have a member in the obj that represents the logical AND of online_cpuset and allowed_cpuset. This is the set of CPUs that you can actually *use*. Admittedly, this is purely a convenience; I mention it because we found ourselves always computing this value when deciding which PU's to bind to. So it would be nice if it was pre-computed by hwloc and stashed on the obj. (I'm not saying to remove the online_cpuset and allowed_cpuset fields -- they're useful information, too; I'm just saying that the AND of these values is also a useful piece of information)

On Dec 9, 2011, at 3:18 AM, Brice Goglin wrote:

> I want to release 1.3.1 for christmas, so I'll do a 1.3.1rc1 on monday.
> There are no critical fixes in there, but several changes that are
> useful, so it's better to finally flush all these (some have been
> waiting almost since the 1.3.1 two months ago).
>
> Additionally, some people want a multinode support release soon too and
> I kind of promised it would arrive by the end of the year. So I'll start
> polishing the v1.4 branch and do a first 1.4rc next week or so, so that
> people can start playing with it.
>
> There are currently 3 tickets against v1.4:
> #50 is about adding instruction caches
> #41 wants conversion function between "L2cache"-like string and
> (type,depth) and actual depth within the tree
> #43 wants throughput matrices on the side of the existing latency matrices
> None of this is ready, and there are still open questions in #50 and #43
> so I intend to defer all these to v1.5 so that multinode support doesn't
> wait too long.
>
> Brice
>
> _______________________________________________
> hwloc-devel mailing list
> hwloc-devel_at_[hidden]
> http://www.open-mpi.org/mailman/listinfo.cgi/hwloc-devel

-- 
Jeff Squyres
jsquyres_at_[hidden]
For corporate legal information go to:
http://www.cisco.com/web/about/doing_business/legal/cri/