On Fri, Aug 15, 2008 at 20:42, Jeff Squyres <jsquyres_at_[hidden]> wrote:
> On Aug 15, 2008, at 1:50 PM, Bert Wesarg wrote:
>
>> Its better to handle the default values, so that the code don't need
>> to changed again if powerpc is up-to-date.
>
> Just to make sure I understand -- you're saying that if we read <0 from
> phsyical_package_id, we should treat it as a non-fatal error (i.e., ignore
> that processor), right?
Hmm, good question.
>
> Come to think of it, I can't imagine a system with both an AMD/Intel
> processor *and* a POWER processor, so we could probably dump out of the loop
> at the first <0 physical_package_id (or any error) and assume that topology
> isn't supported at all.
Me too. I think its safest to error out fatal, in case of any <0 p_p_id.
>
> But I guess the cost is extremely low, and this isn't performance critical
> code, so having it check all the processors anyway is a safe thing to do (in
> case some machine someday *does* contain heterogeneous chips, all running
> Linux, which, with the advent of accelerators, I guess isn't *too* far
> fetched...). PLPA *almost* does this today, but could probably be a bit
> better about error checking.
In the end, we need a valid mapping, i.e. processor_id <-> (socket_id,
core_id). If we can't guarantee this, we should tell this.
Bert
>
> --
> Jeff Squyres
> Cisco Systems
>
> _______________________________________________
> plpa-users mailing list
> plpa-users_at_[hidden]
> http://www.open-mpi.org/mailman/listinfo.cgi/plpa-users
>
|
![[Flag]](../../../../../community/mirrors/flag-us.gif){kind=small}