Sounds great!
Were you able to test these on the Altix?
I'm having great difficulty applying these patches, probably because
of various stupid email issues. Can you re-send them as attachments?
Or do you want to commit them directly?
On Jul 23, 2009, at 5:47 PM, Bert Wesarg wrote:
> I just got access to an SGI Altix machine with 192 dual core
> Itaniums and I
> noticed the vary high socket id numbers, they range from 262144 to
> 359936.
> For the (temporary) cores_on_sockets array that would be around 43MB.
>
> This patch series addresses this problem.
>
> The first two patches will abstract between the socket_ids and the
> socket
> arrays, to make the actual changes less intrusive.
>
> Patch 1/5 will introduce a mapping function from socket_id to
> indexes into the
> socket arrays and back. There is no functional change.
>
> Patch 2/5 will introduce a variable which holds the extent of the
> indexes of
> the socket arrays (i.e. the actual size of these arrays). There is no
> functional change.
>
> Patch 3/5 will maintain the minimum socket_id, so that we can
> subtract this
> minimum from all the array extent.
>
> Patch 4/5 will calculate the length of common trailing zeros in all
> socket_ids,
> so that we can safely shift away these zeros.
>
> (Optional) Patch 5/5 will introduce get_socket_info3(), so that the
> user can have access
> to the min_socket_id.
>
> After this patch the range of socket indexes is down to 0..191,
> which gives
> nice small arrays.
>
> Bye,
> Bert
>
> PS: I will test this on the SGI machine tomorrow.
>
> ---
> src/libplpa/plpa.h.in | 6 ++
> src/libplpa/plpa_map.c | 131 ++++++++++++++++++++++++++++++++++
> +------------
> 2 files changed, 103 insertions(+), 34 deletions(-)
>
> _______________________________________________
> plpa-users mailing list
> plpa-users_at_[hidden]
> http://www.open-mpi.org/mailman/listinfo.cgi/plpa-users
>
--
Jeff Squyres
jsquyres_at_[hidden]
|
![[Flag]](../../../../../community/mirrors/flag-us.gif){kind=small}