Open MPI logo

Hardware Locality Development Mailing List Archives

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

Subject: Re: [hwloc-devel] hwloc powerpc rhel5 and power7 patch
From: Brice Goglin (Brice.Goglin_at_[hidden])
Date: 2010-09-16 11:03:48


(replying to Alexey's mail which got rejected from the list)

Le 16/09/2010 15:58, Alexey Kardashevskiy a écrit :
> On 16/09/10 21:58, Brice Goglin wrote:
> > Le 16/09/2010 13:52, Alexey Kardashevskiy a écrit :
> >
> >> On 16/09/10 20:50, Brice Goglin wrote:
> >>
> >>> We'll likely apply it, we just need to figure out where to put it if
> >>> it's reusable for AIX.
> >>>
> >>>
> >>>
> >> Good! And what is about messages coming twice in this maillist? Am I
> >> the only person who experiences that? :)
> >>
> > I thought it was my mailer replying to both you and the mailing list,
> > but you were actually subscribed twice, I just fixed this.
> >
> >
>
> Thank you!
>
> > The attached patch should do what you need for sparse numa number
> > properly. I had to rework the code a bit because my local tests with
> > fake topologies reported numa nodes out of order, causing the distance
> > matrix to be misordered. So I am now placing all node indexes in the
> > cpuset, then I fill the index array by reading the cpuset in order, and
> > then I read distances from sysfs. Could you confirm that it works for
> > you too?
> >
> >
>
> Yes, it works just fine:
>

Ok thanks for testing so quickly, I just pushed this part to SVN.

Brice

> numa distance indexes: 0 1 4 5 8 9 12 13
> os node 0 has cpuset 0xffffffff
> os node 1 has cpuset 0xffffffff,0x0
> os node 4 has cpuset 0xffffffff,,0x0
> os node 5 has cpuset 0xffffffff,,,0x0
> os node 8 has cpuset 0xffffffff,,,,0x0
> os node 9 has cpuset 0xffffffff,,,,,0x0
> os node 12 has cpuset 0xffffffff,,,,,,0x0
> os node 13 has cpuset 0xffffffff,,,,,,,0x0
> node distance matrix:
> 0 1 4 5 8 9 12 13
> 0 10 20 40 40 40 40 40 40
> 1 20 10 40 40 40 40 40 40
> 4 40 40 10 20 40 40 40 40
> 5 40 40 20 10 40 40 40 40
> 8 40 40 40 40 10 20 40 40
> 9 40 40 40 40 20 10 40 40
> 12 40 40 40 40 40 40 10 20
> 13 40 40 40 40 40 40 20 10
> trying to group NUMANode objects into misc objects according to physical
> distances
> found minimal distance 20 between objects
> object 1 is minimally connected to 0
> found transitive graph with 2 objects with minimal distance 20
> object 3 is minimally connected to 2
> found transitive graph with 2 objects with minimal distance 20
> object 5 is minimally connected to 4
> found transitive graph with 2 objects with minimal distance 20
> object 7 is minimally connected to 6
> found transitive graph with 2 objects with minimal distance 20
> adding misc object with 2 objects and cpuset 0xffffffff,0xffffffff
> adding misc object with 2 objects and cpuset 0xffffffff,0xffffffff,,0x0
> adding misc object with 2 objects and cpuset 0xffffffff,0xffffffff,,,,0x0
> adding misc object with 2 objects and cpuset 0xffffffff,0xffffffff,,,,,,0x0
> group distances:
> 15 40 40 40
> 40 15 40 40
> 40 40 15 40
> 40 40 40 15
> trying to group Group objects into misc objects according to physical
> distances
> found minimal distance 40 between objects
> object 1 is minimally connected to 0
> object 2 is minimally connected to 0
> object 3 is minimally connected to 0
> found transitive graph with 4 objects with minimal distance 40
>
>
>