Open MPI logo

Hardware Locality Development Mailing List Archives

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

Subject: Re: [hwloc-devel] asymmetric trees
From: Brice Goglin (Brice.Goglin_at_[hidden])
Date: 2011-04-15 00:35:26


"Group" too. When you have very hierarchy NUMA distances (Altix UV), you
create groups of groups of groups...

Brice

Le 15/04/2011 03:16, Jeff Squyres a écrit :
> Ok, I get it now -- cache is probably about the only object where this could occur, right?
>
>
> On Apr 14, 2011, at 9:08 PM, Jeff Squyres wrote:
>
>
>> On the subject of asymmetric trees, I see the description of hwloc_type_get_depth() in hwloc.h:
>>
>> /** \brief Returns the depth of objects of type \p type.
>> *
>> * ...snipped...
>> *
>> * If some objects of the given type exist in different levels, for instance
>> * L1 and L2 caches, the function returns HWLOC_TYPE_DEPTH_MULTIPLE.
>>
>> Am I reading the FAQ wrong? I thought the figure on page 21 (US letter) indicated that, for example, the depth of CORE objects would always be the same, even though one of them doesn't have the same caches as its siblings.
>>
>> --
>> Jeff Squyres
>> jsquyres_at_[hidden]
>> For corporate legal information go to:
>> http://www.cisco.com/web/about/doing_business/legal/cri/
>>
>>
>> _______________________________________________
>> hwloc-devel mailing list
>> hwloc-devel_at_[hidden]
>> http://www.open-mpi.org/mailman/listinfo.cgi/hwloc-devel
>>
>
>