You may also hit the 'f' key from the graphical X11 output to toggle
factorizing of cores and collapsing of PCI devices (those shortcuts are
shown in the terminal text output while the graphical window is running).
Brice
Le 25/06/2024 à 21:34, Thompson, Matt (GSFC-610.1)[SCIENCE SYSTEMS AND
Hello
I don't have access to a MI300A but I worked with AMD several month ago
to solve a very similar issue. It was caused by a buggy APCI HMAT in the
BIOS.
Try setting HWLOC_USE_NUMA_DISTANCES=0 in the environment to disable the
hwloc code that uses this HMAT info. If the warning goes away
Le 24/11/2023 à 08:51, John Hearns a écrit :
Good question. Maybe not an answer referring to hwloc.
When managing a large NUMA machine, SGI UV, I ran the OS processes in
a boot cpuset which was restricted to (AFAIR) the first 8 Cpus.
On Intel architecures with E and P cores could we think of r
Le 23/11/2023 à 19:29, Jirka Hladky a écrit :
Hi Brice,
I have a question about the hwloc's support for Intel's hybrid
architectures, like in Alder Lake CPUs:
https://en.wikipedia.org/wiki/Alder_Lake
There are P (performance) and E (efficiency) cores. Is hwloc able to
detect which core is wh
ell
account for sundry testing, mostly of build procedures.
Is there anything I could do to get hwloc to work?
Regards,
Max
---
On Wed, Aug 02, 2023 at 03:12:27PM +0200, Brice Goglin wrote:
Hello
There's something wrong in this machine. It exposes 4 cores (number 0 to 3)
and no NUMA nod
PU L#0 (P#0)
PU L#1 (P#1)
HostBridge
PCI 00:03.0 (Other)
Block(Disk) "sda"
PCI 00:04.0 (Ethernet)
Net "ens4"
PCI 00:05.0 (Other)
(from which I conclude my build procedure is correct).
At the suggestion of Brice Goglin (in response t
Hello
If I understand correctly, your allocations are going to swap as soon as
you use more than half of your available local RAM? I don't see any
reason for this unless some additional limit is preventing you from
using more. Is there any chance your job was allocated with less memory
(for i
Le 02/03/2022 à 12:31, Mike a écrit :
Hello,
Can you display both mask before set_area_membind and after
get_area_membind and send the entire output of all processes and
threads? If you can prefix the line with the PID, it'd help a lot :)
What do you mean with output of all process
Le 02/03/2022 à 11:38, Mike a écrit :
Hello,
If you print the set that is built before calling
set_area_membind, you should only see 4 bits in there, right?
(since threadcount=4 in your code)
I'd say 0xf for rank0, 0xf0 for rank1, etc.
set_area_membind() will translate that
what threadcount means in your
code. Are you calling the allocate function multiple times with many
different ranks? (MPI ranks?)
Brice
Mike
Am Mi., 2. März 2022 um 09:53 Uhr schrieb Brice Goglin
mailto:brice.gog...@inria.fr>>:
Le 02/03/2022 à 09:39, Mike a écrit :
Le 02/03/2022 à 09:39, Mike a écrit :
Hello,
Please run "lstopo -.synthetic" to compress the output a lot. I
will be able to reuse it from here and understand your binding mask.
Package:2 [NUMANode(memory=270369247232)] L3Cache:8(size=33554432)
L2Cache:8(size=524288) L1dCache:1(size=32
Le 01/03/2022 à 17:34, Mike a écrit :
Hello,
Usually you would rather allocate and bind at the same time so
that the memory doesn't need to be migrated when bound. However,
if you do not touch the memory after allocation, pages are not
actually physically allocated, hence there'
Le 01/03/2022 à 15:17, Mike a écrit :
Dear list,
I have a program that utilizes Openmpi + multithreading and I want the
freedom to decide on which hardware cores my threads should run. By
using hwloc_set_cpubind() that already works, so now I also want to
bind memory to the hardware cores.
.open-mpi.org; Brice Goglin
*Subject:* RE: [OMPI users] hwloc error
Hello Brice
Thanks for your reply.
I forgot to mention that my machine is a windows one and not Linux.
I did download the new version of hwloc.
Could you brief me the steps for installing it? Are the steps similar
to this?
Le 23/03/2021 à 08:08, Brice Goglin a écrit :
> Le 23/03/2021 à 02:28, ro...@uberware.net a écrit :
>> Hi. I'm trying to build hwloc on OS-X Big Sur on an M1. Ultimate plan is
>> to build it as a universal binary. Right now, I cannot even get the git
>> master to a
Le 23/03/2021 à 02:28, ro...@uberware.net a écrit :
> Hi. I'm trying to build hwloc on OS-X Big Sur on an M1. Ultimate plan is
> to build it as a universal binary. Right now, I cannot even get the git
> master to autogen. This is what I get:
>
> robin@Robins-Mac-mini hwloc % ./autogen.sh
> autoreco
Hello
The "latest_snapshot.txt" files on the website were broken (for years).
Things are now fixed and improved. And they are also explicitly
documented on the main web page.
If you want the version string of the latest release or release
candidate, read
https://www.open-mpi.org/software/hwloc/cu
Hello Kevin
There is some very experimental support for Cray networks as well as
Intel OmniPath. But the entire subproject has been unmaintained for a
while and I don't expect anybody to revive it anytime soon unfortunately.
Brice
Le 16/02/2021 à 17:00, ke...@continuum-dynamics.com a écrit :
>
Hello
I am not sure we ever tested this because offlining cpu0 was impossible in
Linux until recently. I knew things would change because arm kernel devs were
modifying Linux to allow it. Looks like it matters to x86 too, now. I'll take a
look.
Brice
Le 5 février 2021 20:43:33 GMT+01:00, "Clay
Le 02/10/2020 à 01:59, Jirka Hladky a écrit :
>
> I'll see if I can make things case-insensitive in the tools (not
> in the C API).
>
> Yes, it would be a nice improvement. Currently, there is a mismatch
> between different commands. hwloc-info supports both bandwidth and
> Bandwidth, bu
good performance
measurement tool isn't easy. I see people sending patches for adding
some assembly because this corner case on this processor isn't well
optimized by GCC :/ I am not sure we want to put this inside hwloc.
Brice
>
>
> On Thu, Oct 1, 2020 at 7:28 PM Brice Goglin
ist to
> hwloc-info -h?
I could add the default ones, but I'll need to specify that additional
user-given attributes may exist.
Thanks for the feedback.
Brice
>
> hwloc-info --best-memattr bandwidth
> hwloc-info --best-memattr latency
>
> Thanks a lot!
> Jirka
>
If you don't care about the overhead, tell python to use the output of
shell command "hwloc-calc -N pu all".
Brice
Le 31/08/2020 à 18:38, Brock Palen a écrit :
> Thanks,
>
> yeah I was looking for an API that would take into consideration most
> cases, like I find with hwloc-bind --get where I
Le 31/08/2020 à 18:19, Guy Streeter a écrit :
> As I said, cgroups doesn't limit the group to a number of cores, it
> limits processing time, either as an absolute amount or as a share of
> what is available.
> A docker process can be restricted to a set of cores, but that is done
> with cpu affin
Hello
Do you know which lstopo is correct here? Do you have a way to know if
the IB interface is indeed connected to first NUMA node of 2nd package,
or to 2nd NUMA node of 1st package? Benchmarking IB bandwidth when
memory/cores are in NUMA node #1 vs #2 would be nice.
The warning/fixup was added
07/2020 à 19:32, Jon Dart a écrit :
> That was it - the older DLL was in the path. Thanks for looking into it.
>
> --Jon
>
> On 7/22/2020 6:02 AM, Brice Goglin wrote:
>>
>> Hello Jon
>>
>> Sorry the delay. I finally got some time to look at this. I can only
&g
Le 01/07/2020 à 15:55, Jon Dart a écrit :
> On 6/30/2020 4:00 PM, Brice Goglin wrote:
>>
>> Hello
>>
>> We don't have many windows-specific changes in 2.1 except some late
>> MSVC-related changes added after rc1. Can you try 2.1.0rc1 instead of
>> 2.1.
Hello
It looks your hardware and/or OS is reporting buggy information. We'd
need more details to debug this. Can you open an githab issue at
https://github.com/open-mpi/hwloc/issues/new ? This page lists what
information you need to provide for debugging.
It looks like you're using hwloc inside a
Hello
We don't have many windows-specific changes in 2.1 except some late
MSVC-related changes added after rc1. Can you try 2.1.0rc1 instead of
2.1.0? It's not visible on the download page but it's actually
available, for instance at
https://download.open-mpi.org/release/hwloc/v2.1/hwloc-win64-bui
Oh sure, I thought we fixed this a while ago. I pushed it to master. Do
you need in 2.2 only or also earlier stable series?
Brice
Le 29/05/2020 à 05:32, Balaji, Pavan via hwloc-users a écrit :
> Hello,
>
> We are maintaining this patch for hwloc internally in mpich. Can this be
> upstreamed?
>
Hello Stephen
There's no equivalent in hwloc 2.x unfortunately, even with netloc.
"custom" caused too many issues for core maintenance (mostly because of
cpusets being different between machines) while use cases were very rare.
Brice
Le 12/05/2020 à 08:01, Herbein, Stephen via hwloc-users a é
Hello
Several people asked recently how hwloc exposes heterogeneous memory and
how to recognize which NUMA nodes is which kind of memory. Short answer
is that it's currently ugly but we're working on it for hwloc 2.3. I put
all details in this wiki page :
https://github.com/open-mpi/hwloc/wiki/He
Hello Liam
dmidecode is usually reserved to root only because it uses SMBIOS or
whatever hardware/ACPI/... tables. Those tables are read by the Linux
kernel and exported to non-root users in sysfs:
$ cat /sys/bus/pci/devices/:ae:0c.6/numa_node
1
However this file isn't that good because som
Oops wrong list, sorry :)
Le 15/11/2019 à 10:49, Brice Goglin a écrit :
> Hello
>
> We have a platform with an old MLX4 partition and another OPA partition.
> We want a single OMPI installation working for both kinds of nodes. When
> we enable UCX in OMPI for MLX4, UCX ends up be
Hello
We have a platform with an old MLX4 partition and another OPA partition.
We want a single OMPI installation working for both kinds of nodes. When
we enable UCX in OMPI for MLX4, UCX ends up being used on the OPA
partition too, and the performance is poor (3GB/s instead of 10). The
problem se
symbols and avoid
> altogether the prefixed ones?
>
> Thank you,
>
> Sam
>
>> On Oct 10, 2019, at 10:30 AM, Brice Goglin > <mailto:brice.gog...@inria.fr>> wrote:
>>
>> Le 10/10/2019 à 17:38, Gutierrez, Samuel K. via hwloc-users a écrit :
>>> Good mor
Le 10/10/2019 à 17:38, Gutierrez, Samuel K. via hwloc-users a écrit :
> Good morning,
>
> I have a question about expected name mangling behavior when using
> HWLOC_SET_SYMBOL_PREFIX in hwloc v2.1.0 (and perhaps other versions).
>
> Say, for example, I do the following in a project embedding hwloc
Hello
Indeed we would like to expose this kind of info but Netloc is
unfornately undermanpowered these days. The code in git master is
outdated. We have a big rework in a branch but it still needs quite a
lot of polishing before being merged
The API is still mostly-scotch-oriented (i.e. for pro
Hello
It may be similar to https://github.com/open-mpi/hwloc/issues/290 but we
weren't able to find the exact issue unfortunately :/
Setting HWLOC_COMPONENTS=-x86 in the environment would disable that code
path, causing the topology to be possibly not as precise.
Brice
Le 08/07/2019 à 20:43,
Thanks Pavan, I am pushing this.
Brice
Le 25/05/2019 à 08:19, Balaji, Pavan via hwloc-users a écrit :
> Folks,
>
> We ran into an issue with the hwloc integration for MPICH. On Mac OS, hwloc
> detects that OpenCL is available, but the corresponding LDFLAGS are not
> exported upstream to MPIC
alaji, Pavan via hwloc-users a écrit :
> Brice, all,
>
> Any update on this? Are you guys planning on fixing these?
>
> -- Pavan
>
>> On Feb 25, 2019, at 7:33 AM, Balaji, Pavan via hwloc-users
>> wrote:
>>
>> Hi Brice,
>>
>>> On Feb 25, 2019
Hello Pavan,
Are you sure you're not passing -Wstack-usage? My Ubuntu 18.04 with
latest gcc-7 (7.3.0-27ubuntu1~18.04) doesn't show any of those warnings.
It looks like all these warnings are caused by C99 variable-length
arrays (except 2 that I don't understand). I know the kernel devs
stopped us
parent_hugepage/enabled
> [always] madvise never
>
> is set already, so I'm not really sure what should go in there to disable it.
>
> JB
>
> -Original Message-
> From: Brice Goglin
> Sent: 29 January 2019 15:29
> To: Biddiscombe, John A. ; Hardware local
it.
>
> Problem seems to be solved for now. Thank you very much for your insights and
> suggestions/help.
>
> JB
>
> -Original Message-
> From: Brice Goglin
> Sent: 29 January 2019 10:35
> To: Biddiscombe, John A. ; Hardware locality user list
>
> Subj
0 0 0 0 0 0 0 0 0
> 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
> 0 0 0 0 0
>
> On the 8 numa node machine it sometimes gives the right answer even with 512
> pages.
>
> Still baffled
>
> JB
>
> -Original Message-
> Fro
1-1-1-1-1
> 1-1-1-1-1-1-1-1-1-1-1-1-1-1-1-1-1-1-1-1-1-1-1-1-1-1-1-1-1-1-1-1-1-1-1-1-1-1-1-1-
> which is correct because the '-' is a negative status. I will run again and
> see if it's -14 or -2
>
> JB
>
>
> -Original Message-
> From: Brice
y next since I can
> see the memory contents hold the correct CPU ID of the thread that touched
> the memory, so either the syscall is wrong, or the kernel is doing something
> else. I welcome any suggestions on what might be wrong.
>
> Thanks for trying to help.
>
> JB
>
>
Le 25/01/2019 à 23:16, Biddiscombe, John A. a écrit :
>> move_pages() returning 0 with -14 in the status array? As opposed to
>> move_pages() returning -1 with errno set to 14, which would definitely be a
>> bug in hwloc.
> I think it was move_pages returning zero with -14 in the status array, an
Le 25/01/2019 à 14:17, Biddiscombe, John A. a écrit :
> Dear List/Brice
>
> I experimented with disabling the memory touch on threads except for
> N=1,2,3,4 etc and found a problem in hwloc, which is that the function
> hwloc_get_area_memlocation was returning '0' when the status of the memory
Le 21/01/2019 à 17:08, Biddiscombe, John A. a écrit :
> Dear list,
>
> I'm allocating a matrix of size (say) 2048*2048 on a node with 2 numa domains
> and initializing the matrix by using 2 threads, one pinned on each numa
> domain - with the idea that I can create tiles of memory bound to each
Hello
That's not how current operating systems work, hence hwloc cannot do it.
Usually you can bind a process virtual memory to a specific part of the
physical memory (a NUMA node is basically a big static range), but the
reverse isn't allowed by any OS I know.
If you can tweak the hardware, you
> team if you want them to upgrade :-)
>
> Jeff
>
> On Thu, Sep 13, 2018 at 8:42 AM, Brice Goglin <mailto:brice.gog...@inria.fr>> wrote:
>
> This is actually just a warning. Usually it causes the topology to
> be wrong (like a missing object), but it shouldn'
This is actually just a warning. Usually it causes the topology to be
wrong (like a missing object), but it shouldn't prevent the program from
working. Are you sure your programs are failing because of hwloc? Do you
have a way to run lstopo on that node?
By the way, you shouldn't use hwloc 2.0.0rc
Hello
The only public portability layer we have for PIDs is hwloc_pid_t when
passed to things like set_proc_cpubind(). But we don't have a portable
getpid() or printf(). You'll have to use getpid() and printf("%ld",
(long)pid) on Unix.
On Windows, hwloc_pid_t is a HANDLE, you don't want to print
This was also addressed offline while the mailing was (again) broken.
Some symbols weren't renamed in old releases. This was fixed a couple
months ago. It will be in 2.0.2 and 1.11.11 (to be released on Monday
Sept 3rd).
Brice
Le 30/08/2018 à 06:31, Junchao Zhang a écrit :
> Hi,
> My progra
Hello
If you bind a thread to a newset that contains 4 PUs (4 bits), the
operating system scheduler is free to run that thread on any of these
PUs. It means it may run on it on one PU, then migrate it to the other
PU, then migrate it back, etc. If these PUs do not share all caches, you
will see a
This question was addressed offline while the mailing lists were offline.
We had things like hwloc_bitmap_set_ith_ulong() and
hwloc_bitmap_from_ith_ulong() for packing/unpacking but they weren't
very convenient unless you know multiple ulongs are actually needed to
store the bitmap.
We added new
Hello Hartmut
The mailing list address changed a while ago, there's an additional
"lists." in the domaine name.
Regarding your question, I would assume you are running in a cgroup with
the second NUMA node disallowed (while all the corresponding cores are
allowed). lstopo with --whole-system woul
Le 04/04/2018 à 16:49, Madhu, Kavitha Tiptur a écrit :
>
> — I tried building older netloc with hwloc 2.0 and it throws compiler errors.
> Note that netloc was cloned from it’s git repo.
My guess is that the "map" part that joins netloc's info about the
fabric with hwloc's info about the nodes do
à 01:36, Balaji, Pavan a écrit :
> Brice,
>
> We want to use both hwloc and netloc in mpich. What are our options here?
> Move back to hwloc-1.x? That’d be a bummer because we already invested a lot
> of effort to migrate to hwloc-2.x.
>
> — Pavan
>
> Sent from my iP
embedded mode?
>
>
>> On Mar 30, 2018, at 1:34 PM, Brice Goglin wrote:
>>
>> Hello
>>
>> In 2.0, netloc is still highly experimental. Hopefully, a large rework
>> will be merged in git master next month for being released in hwloc 2.1.
>>
>>
Hello
In 2.0, netloc is still highly experimental. Hopefully, a large rework
will be merged in git master next month for being released in hwloc 2.1.
Most of the API from the old standalone netloc was made private when
integrated in hwloc because there wasn't any actual user. The API was
quite la
Hello
In case you missed the announce yesterday, hwloc 2.0.1rc1 changes the
library soname from 12:0:0 to 15:0:0. On Linux, it means that we'll now
build libhwloc.so.15 instead of libhwloc.so.12. That means any
application built for hwloc 2.0.0 will need to be recompiled against 2.0.1.
I should h
to objects at the depth or above in Hydra previously. As
>> you pointed out, the functionality makes no sense with NUMA/IO objects
>> possibly being at different depths or for objects.
>>
>>> On Mar 14, 2018, at 3:00 PM, Brice Goglin wrote:
>>>
>>>
Hello
I can fix the documentation to say that the function always suceeds and
returns the virtual depth for NUMA/IO/Misc.
I don't understand your third sentence. If by "actual depth", you mean
the depth of a (normal) parent where NUMA are attached (for instance the
depth of Package if NUMAs are a
Hello
As you may know, hwloc only discovers KNL MCDRAM Cache details if
hwloc-dump-hwdata ran as root earlier. There's an issue with that tool
in 2.0, which was supposed to be a feature: we fixed the matching of
SMBIOS strings, and now it appears some vendors don't match anymore
because they didn'
ase let me know.
Brice
Le 05/02/2018 à 23:19, Madhu, Kavitha Tiptur a écrit :
> Hi
>
> Thanks for the response. Could you also confirm if hwloc topology
> object would have only machine node?
>
> Thanks,
> Kavitha
>
>
>
>> On Feb 5, 2018, at 4:14 PM, B
Hello,
Oops, sorry, this sentence is obsolete, I am removing it from the doc
right now.
We don't support the assembly of multiple machines in a single hwloc
topology anymore. For the record, this feature was a very small corner
case and it had important limitations (you couldn't bind things or us
Hello
I need people running Mac OS to test some patches before releasing them
in 2.0rc2 (which is likely delayed to Monday).
Just build this tarball, run lstopo, and report any difference with
older lstopo outputs:
https://ci.inria.fr/hwloc/job/zbgoglin-0-tarball/lastSuccessfulBuild/artifact/hwl
It looks like our Mac OS X backend doesn't properly handle processors
that support hyperthreading without actually having hyperthreads enabled
in hardware. Your processor has 4-core without HT but it's based on a
processor with up to 8 cores and 16 threads. Our current code uses the
latter and ther
It is actually easy to fix, we just need to move hwloc's #include before
what base64.c actually #include's. That'll be fixed in rc2 too.
Brice
Le 25/01/2018 à 10:56, Brice Goglin a écrit :
> Like the error below?
>
> This code hasn't changed recently. Did yo
Like the error below?
This code hasn't changed recently. Did you ever build with these flags
before?
I am not sure I'll have time to fix yet another header crazyness before rc2.
Brice
CC base64.lo
In file included from
/builds/hwloc-master-20180124.2347.gitf53fe3a/include/private/priv
The output of sysctl -a would also help.
Brice
Le 25/01/2018 à 07:34, Brice Goglin a écrit :
> Hello
>
> I don't see anything obvious. Can you rebuild with --enable-debug and
> report the full lstopo output?
>
> Brice
>
>
>
> Le 25/01/2018 à 07:14,
f9
> https://github.com/pmodels/hwloc/commit/9bf3ff256511ea4092928438f5718904875e65e1
>
> The first one is definitely not usable as-is, since that breaks standalone
> builds. But I'm interested in hearing about any better solution that you
> might have.
>
> Thanks,
>
Thanks, I am fixing this for rc2 tomorrow.
Brice
Le 24/01/2018 à 22:59, Balaji, Pavan a écrit :
> Folks,
>
> I'm seeing these warnings on the mac os when building hwloc-2.0rc1 with clang:
>
> 8<
> CC lstopo-lstopo.o
> lstopo.c: In function 'usage':
> lstopo.c:425:7: warning: "CAI
system that seq faults, and 1.6.6 on the one that succeeds.
> And that the first looks to be the standard OFED release and the 1.6.6
> version a mellanox release of OFED.
>
> Craig.
>
> On Tue, 23 Jan 2018 at 17:10 Brice Goglin <mailto:brice.gog...@inria.fr>> wrote:
&
Hello
I didn't know you use submodule. I just pushed tag "hwloc-2.0.0rc1" and
I'll try to remember pushing one for each future rc. If I don't, please
remind me.
I am not going to push all the previous ones because there are just too
many of them. If you need some specific ones, please let me know
Hello,
If the output isn't too big, could you put the files gathered by
netloc_ib_gather_raw online so that we look at them and try to reproduce
the crash?
Thanks
Brice
Le 23/01/2018 à 03:54, Craig West a écrit :
> Hi,
>
> I can't find the version requirements for netloc. I've tried it on an
Le 29/12/2017 à 23:15, Bill Broadley a écrit :
>
>
> Very interesting, I was running parallel finite element code and was seeing
> great performance compared to Intel in most cases, but on larger runs it was
> 20x
> slower. This would explain it.
>
> Do you know which commit, or anything else t
Hello
Make sure you use a very recent Linux kernel. There was a bug regarding L3
caches on 24-core Epyc processors which has been fixed in 4.14 and backported
in 4.13.x (and maybe in distro kernels too).
However, that would likely not cause huge performance difference unless your
application hea
Hello
This sound very interesting, likely what we want for
https://github.com/open-mpi/hwloc/issues/54
I couldn't test it yet (dnf on Fedora25 says nothing provides
python2-hwloc even that package is installed in version 2.3.1). Do you
plan to put python-hwloc (and maybe pianofish too) in pypi? I
thanks.
Brice
>
> Thank you,
>
> David
>
> On 29/11/2017 13:35, Brice Goglin wrote:
>> Hello
>>
>> We only add hwloc Group objects when necessary. On your system, each
>> processor group contains a single NUMA node, so these Groups would not
>> real
Hello
We only add hwloc Group objects when necessary. On your system, each
processor group contains a single NUMA node, so these Groups would not
really bring additional information about the hierarchy of resources.
If you had a bigger system with, let's say, 4 NUMA nodes, with 2 of them
in each p
Brice
Le 28/10/2017 09:31, Brice Goglin a écrit :
> Hello,
> The Linux kernel reports incorrect L3 information.
> Unfortunately, your old kernel seems to already contain patches for
> supporting the L3 on this hardware. I found two candidate patches for
> further fixing this, one is
m/open-mpi/hwloc/pull/277
Make all depths *signed* ints
https://github.com/open-mpi/hwloc/pull/276
Remove the "System" object type
https://github.com/open-mpi/hwloc/pull/275
Move local_memory to NUMA node specific attrs
https://github.com/open-mpi/hwloc/pull/274
Brice
Le 26/
I hoped for.
>
> Thanks again
>
> JB
>
>
> From: hwloc-users [hwloc-users-boun...@lists.open-mpi.org] on behalf of Brice
> Goglin [brice.gog...@inria.fr]
> Sent: 13 November 2017 15:32
> To: Hardware locality user list
> Subject
t;
> aha. thanks. I knew I'd seen a function for that, but couldn't remember what
> it was.
>
> Cheers
>
> JB
> ____
> From: hwloc-users [hwloc-users-boun...@lists.open-mpi.org] on behalf of Brice
> Goglin [bri
-users [hwloc-users-boun...@lists.open-mpi.org] on behalf of
> Samuel Thibault [samuel.thiba...@inria.fr]
> Sent: 12 November 2017 10:48
> To: Hardware locality user list
> Subject: Re: [hwloc-users] question about hwloc_set_area_membind_nodeset
>
> Brice Goglin, on dim. 12 nov. 201
Le 12/11/2017 00:14, Biddiscombe, John A. a écrit :
> I'm allocating some large matrices, from 10k squared elements up to
> 40k squared per node.
> I'm also using membind to place pages of the matrix memory across numa
> nodes so that the matrix might be bound according to the kind of
> pattern a
Hello
It should have been 0x00010b03 but I forgot to increase it unfortunately
(and again in 1.11.6).
I need to add this to my release-TODO-list.
The upcoming 1.11.9 will have the proper HWLOC_API_VERSION (0x00010b06
unless we had something) so that people can at least check for these
features in
Hello,
The Linux kernel reports incorrect L3 information.
Unfortunately, your old kernel seems to already contain patches for
supporting the L3 on this hardware. I found two candidate patches for
further fixing this, one is in 4.10 (cleanup of the above patch) and the
other will only be in 4.14.
I
Hello
I finally merged the new memory model in master (mainly for properly
supporting KNL-like heterogeneous memory). This was the main and last
big change for hwloc 2.0. I still need to fix some caveats (and lstopo
needs to better display NUMA nodes) but that part of the API should be
ready.
Now
Hello
On Linux, the PCI linkspeed requires root privileges unfortunately
(except for the uplink above NVIDIA GPUs where we have another way to
find it).
The only way to workaround this is to dump the topology as XML as root
and then reload it at runtime (e.g. with HWLOC_XMLFILE) :/
Brice
Le 13
:01:00.0 3D controller: NVIDIA Corporation GP100GL (rev a1)
>
> But the only devices returned by hwloc are named "cardX" (same as what
> lstopo shows) and have osdev.type of HWLOC_OBJ_OSDEV_GPU and we see no
> devices of type HWLOC_OBJ_OSDEV_COPROC
>
> Sorry, I&#x
Le 07/07/2017 21:51, David Solt a écrit :
> Oh, Geoff Paulsen will be there at Open MPI meeting and he can help
> with the discussion. We tried searching for
>
> // Iterate over each osdevice and identify the GPU's on each socket.
> while ((obj = hwloc_get_next_osdev(machine_topology, obj)) !
Le 07/07/2017 20:38, David Solt a écrit :
> We are using the hwloc api to identify GPUs on our cluster. While we
> are able to "discover" the GPUs, other information about them does not
> appear to be getting filled in. See below for example:
> (gdb) p *obj->attr
> $20 = {
> cache = {
> size
(P#0) + L3 L#0 (6144KB)
> L2 L#0 (2048KB) + L1i L#0 (64KB)
> ...
>
> These nodes are the only one in our entire cluster to cause zombie
> processes using torque/moab. I have a feeling that they are related.
> We use hwloc/1.10.0.
>
> Not sure if this helps at all, but yo
Le 30/06/2017 22:08, fabricio a écrit :
> Em 30-06-2017 16:21, Brice Goglin escreveu:
>> Yes, it's possible but very easy. Before we go that way:
>> Can you also pass HWLOC_COMPONENTS_VERBOSE=1 in the environment and send
&
Hello
We've seen this issue many times (it's specific to 12-core opterons),
but I am surprised it still occurs with such a recent kernel. AMD was
supposed to fix the kernel in early 2016 but I forgot checking whether
something was actually pushed.
Anyway, you can likely ignore the issue as docume
Thanks a lot for the input.
I opened https://github.com/open-mpi/hwloc/issues/243
I have access to a T5 but this will need investigation to actually find
where to get the info from.
Feel free to comment the issue if you find more. I am going to modify
Pg.pm to better understand where Caches come fr
1 - 100 of 464 matches
Mail list logo
