Re: [hwloc-users] Reporting an operating system warning

2017-01-03 Thread Johannes Goller
Definitely!

I am using version 3.0 (the latest) of this SKD:
http://developer.amd.com/tools-and-sdks/opencl-zone/amd-accelerated-parallel-processing-app-sdk/

This, as far as I can tell, is a standard way to get opencl.so, which I
need to link to in order to use OpenCL.

What I made was a command line tool that shows hardware information
(similar to the "clinfo" command line tool) -- it's the show-info.cpp from
here: https://github.com/jogojapan/jogotools/tree/master/opencl

The makefile shows how it's linked to opencl.so from the SDK. At what point
exactly hwloc comes in I don't know.

2017-01-03 18:07 GMT+09:00 Brice Goglin :

> Only the first warning is reported, and the detection might be in a
> different order when we force HWLOC_COMPONENTS. So I guess both L1i and L2
> report wrong cache sharing in this processor. You might try upgrading the
> BIOS if you care, or just ignore those warnings.
>
> The kernel bug about dual-core being reported as single-core hyperthreaded
> is another story. If you use the CPUs with the AMD OpenCL SDK, you might
> see half the performance missing (because 4 cores are considered hardware
> threads). No problem with GPUs.
>
> I didn't know AMD used hwloc yet, I'll check with them to better
> understand what they do with it. Can you give more details about the AMD
> software you're using (name, version, downloaded from?)
>
> Brice
>
>
>
>
>
>
> Le 03/01/2017 09:47, Johannes Goller a écrit :
>
> Thanks again!
>
> Yes, I did get warnings, but -- now that I am taking a closer look -- I
> can see that the warning is different. It complains about L2 cache:
>
> 
> 
>
> * hwloc 1.11.0 has encountered what looks like an error from the operating
> system.
>
> *
>
>
> * L2 (P#0 cpuset 0x0003) intersects with Group0 (P#0 cpuset
> 0x0081) without inclusion!
>
> * Error occurred in topology.c line 983
>
>
> *
>
>
> * The following FAQ entry in the hwloc documentation may
> help:
>
> *   What should I do when hwloc reports "operating system"
> warnings?
>
> * Otherwise please report this error message to the hwloc user's mailing
> list,
>
> * along with the output+tarball generated by the hwloc-gather-topology
> script.
>
> 
> 
>
>
> 2017-01-03 17:23 GMT+09:00 Brice Goglin :
>
>> Thanks
>>
>> Surprisingly, I don't see any L1i in the XML output either. Did you get
>> warnings during this run "HWLOC_COMPONENTS=x86 lstopo foo.xml" ?
>>
>> Indeed, you (very likely) don't care about that warning in the AMD SDK.
>> Pass HWLOC_HIDE_ERRORS=1 in the environment to silence it.
>>
>> Brice
>>
>>
>>
>> Le 03/01/2017 07:59, Johannes Goller a écrit :
>>
>> Hi Brice,
>>
>> thank you very much for looking into this!
>>
>> I am attaching the generated foo.xml.
>>
>> I actually came across this error message when trying to play with
>> OpenCL, using the AMD SDK API. My main interest is in getting that to work
>> on the GPU, and that might still work on my current kernel (4.8), even if I
>> get warnings like the one reported.
>>
>>
>> johannes.
>>
>> 2017-01-03 15:15 GMT+09:00 Brice Goglin < 
>> brice.gog...@inria.fr>:
>>
>>> Hello Johannes
>>>
>>> I think there are two bugs here.
>>>
>>> First one is that each "dual-core compute unit" is reported as a single
>>> core with two hardware threads. That's a kernel bug that appeared in 4.6.
>>> There's a fix at 
>>> https://lkml.org/lkml/2016/11/29/852 but I don't think it has been
>>> applied yet.
>>>
>>> The second bug is a conflict between dual-core compute unit sharing and
>>> L1i. I am not sure which one is actually buggy. Can you run
>>> "HWLOC_COMPONENTS=x86 lstopo foo.xml" and send the generated foo.xml? (this
>>> is our raw detection that works around the kernel detection).
>>>
>>> Trying a Linux kernel <= 4.5 may help in the meantime.
>>>
>>> thanks
>>> Brice
>>>
>>>
>>>
>>>
>>> Le 03/01/2017 05:29, Johannes Goller a écrit :
>>>
>>> As requested on https://www.open-mpi.org/
>>> projects/hwloc/doc/v1.10.1/a00028.php ("What should I do when hwloc
>>> reports 'operating system' warnings?"), I am reporting the warning/error I
>>> received as follows
>>>
>>> 
>>> 
>>> * hwloc 1.11.0 has encountered what looks like an error from the
>>> operating system.
>>> *
>>> * L1i (cpuset 0x0003) intersects with Core (P#0 cpuset 0x0081)
>>> without inclusion!
>>> * Error occurred in topology.c line 983
>>> *
>>> * The following FAQ entry in the hwloc documentation may help:
>>> *   What should I do when hwloc reports "operating system" warnings?
>>> * Otherwise please report this error message to the hwloc user's mailing
>>> list,
>>> * along with the output+tarball generated by the 

Re: [hwloc-users] Reporting an operating system warning

2017-01-03 Thread Johannes Goller
Thanks again!

Yes, I did get warnings, but -- now that I am taking a closer look -- I can
see that the warning is different. It complains about L2 cache:



* hwloc 1.11.0 has encountered what looks like an error from the operating
system.

*

* L2 (P#0 cpuset 0x0003) intersects with Group0 (P#0 cpuset 0x0081)
without
inclusion!

* Error occurred in topology.c line
983

*

* The following FAQ entry in the hwloc documentation may
help:

*   What should I do when hwloc reports "operating system"
warnings?

* Otherwise please report this error message to the hwloc user's mailing
list,

* along with the output+tarball generated by the hwloc-gather-topology
script.




2017-01-03 17:23 GMT+09:00 Brice Goglin :

> Thanks
>
> Surprisingly, I don't see any L1i in the XML output either. Did you get
> warnings during this run "HWLOC_COMPONENTS=x86 lstopo foo.xml" ?
>
> Indeed, you (very likely) don't care about that warning in the AMD SDK.
> Pass HWLOC_HIDE_ERRORS=1 in the environment to silence it.
>
> Brice
>
>
>
> Le 03/01/2017 07:59, Johannes Goller a écrit :
>
> Hi Brice,
>
> thank you very much for looking into this!
>
> I am attaching the generated foo.xml.
>
> I actually came across this error message when trying to play with OpenCL,
> using the AMD SDK API. My main interest is in getting that to work on the
> GPU, and that might still work on my current kernel (4.8), even if I get
> warnings like the one reported.
>
>
> johannes.
>
> 2017-01-03 15:15 GMT+09:00 Brice Goglin :
>
>> Hello Johannes
>>
>> I think there are two bugs here.
>>
>> First one is that each "dual-core compute unit" is reported as a single
>> core with two hardware threads. That's a kernel bug that appeared in 4.6.
>> There's a fix at https://lkml.org/lkml/2016/11/29/852 but I don't think
>> it has been applied yet.
>>
>> The second bug is a conflict between dual-core compute unit sharing and
>> L1i. I am not sure which one is actually buggy. Can you run
>> "HWLOC_COMPONENTS=x86 lstopo foo.xml" and send the generated foo.xml? (this
>> is our raw detection that works around the kernel detection).
>>
>> Trying a Linux kernel <= 4.5 may help in the meantime.
>>
>> thanks
>> Brice
>>
>>
>>
>>
>> Le 03/01/2017 05:29, Johannes Goller a écrit :
>>
>> As requested on
>> 
>> https://www.open-mpi.org/projects/hwloc/doc/v1.10.1/a00028.php ("What
>> should I do when hwloc reports 'operating system' warnings?"), I am
>> reporting the warning/error I received as follows
>>
>> 
>> 
>> * hwloc 1.11.0 has encountered what looks like an error from the
>> operating system.
>> *
>> * L1i (cpuset 0x0003) intersects with Core (P#0 cpuset 0x0081)
>> without inclusion!
>> * Error occurred in topology.c line 983
>> *
>> * The following FAQ entry in the hwloc documentation may help:
>> *   What should I do when hwloc reports "operating system" warnings?
>> * Otherwise please report this error message to the hwloc user's mailing
>> list,
>> * along with the output+tarball generated by the hwloc-gather-topology
>> script.
>> 
>> 
>>
>> Please find the tarball attached.
>>
>>
>>
>> regards,
>>
>> Johannes Goller.
>>
>>
>>
>> ___
>> hwloc-users mailing 
>> listhwloc-us...@lists.open-mpi.orghttps://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
>>
>> ___ hwloc-users mailing list
>> hwloc-users@lists.open-mpi.org https://rfd.newmexicoconsortiu
>> m.org/mailman/listinfo/hwloc-users
>
> ___
> hwloc-users mailing 
> listhwloc-us...@lists.open-mpi.orghttps://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
>
>
> ___
> hwloc-users mailing list
> hwloc-users@lists.open-mpi.org
> https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
>
___
hwloc-users mailing list
hwloc-users@lists.open-mpi.org
https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users

Re: [hwloc-users] Reporting an operating system warning

2017-01-03 Thread Brice Goglin
Thanks

Surprisingly, I don't see any L1i in the XML output either. Did you get
warnings during this run "HWLOC_COMPONENTS=x86 lstopo foo.xml" ?

Indeed, you (very likely) don't care about that warning in the AMD SDK.
Pass HWLOC_HIDE_ERRORS=1 in the environment to silence it.

Brice


Le 03/01/2017 07:59, Johannes Goller a écrit :
> Hi Brice,
>
> thank you very much for looking into this!
>
> I am attaching the generated foo.xml.
>
> I actually came across this error message when trying to play with
> OpenCL, using the AMD SDK API. My main interest is in getting that to
> work on the GPU, and that might still work on my current kernel (4.8),
> even if I get warnings like the one reported.
>
>
> johannes.
>
> 2017-01-03 15:15 GMT+09:00 Brice Goglin  >:
>
> Hello Johannes
>
> I think there are two bugs here.
>
> First one is that each "dual-core compute unit" is reported as a
> single core with two hardware threads. That's a kernel bug that
> appeared in 4.6. There's a fix at
> https://lkml.org/lkml/2016/11/29/852
>  but I don't think it has
> been applied yet.
>
> The second bug is a conflict between dual-core compute unit
> sharing and L1i. I am not sure which one is actually buggy. Can
> you run "HWLOC_COMPONENTS=x86 lstopo foo.xml" and send the
> generated foo.xml? (this is our raw detection that works around
> the kernel detection).
>
> Trying a Linux kernel <= 4.5 may help in the meantime.
>
> thanks
> Brice
>
>
>
>
> Le 03/01/2017 05:29, Johannes Goller a écrit :
>> As requested on
>> https://www.open-mpi.org/projects/hwloc/doc/v1.10.1/a00028.php
>> 
>> ("What should I do when hwloc reports 'operating system'
>> warnings?"), I am reporting the warning/error I received as follows
>>
>> 
>> 
>> * hwloc 1.11.0 has encountered what looks like an error from the
>> operating system.
>> *
>> * L1i (cpuset 0x0003) intersects with Core (P#0 cpuset
>> 0x0081) without inclusion!
>> * Error occurred in topology.c line 983
>> *
>> * The following FAQ entry in the hwloc documentation may help:
>> *   What should I do when hwloc reports "operating system" warnings?
>> * Otherwise please report this error message to the hwloc user's
>> mailing list,
>> * along with the output+tarball generated by the
>> hwloc-gather-topology script.
>> 
>> 
>>
>> Please find the tarball attached.
>>
>>
>>
>> regards,
>>
>> Johannes Goller.
>>
>>
>>
>> ___
>> hwloc-users mailing list
>> hwloc-users@lists.open-mpi.org
>> 
>> https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
>> 
> ___ hwloc-users
> mailing list hwloc-users@lists.open-mpi.org
> 
> https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
>  
>
> ___
> hwloc-users mailing list
> hwloc-users@lists.open-mpi.org
> https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
___
hwloc-users mailing list
hwloc-users@lists.open-mpi.org
https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users

Re: [hwloc-users] Reporting an operating system warning

2017-01-02 Thread Johannes Goller
Hi Brice,

thank you very much for looking into this!

I am attaching the generated foo.xml.

I actually came across this error message when trying to play with OpenCL,
using the AMD SDK API. My main interest is in getting that to work on the
GPU, and that might still work on my current kernel (4.8), even if I get
warnings like the one reported.


johannes.

2017-01-03 15:15 GMT+09:00 Brice Goglin :

> Hello Johannes
>
> I think there are two bugs here.
>
> First one is that each "dual-core compute unit" is reported as a single
> core with two hardware threads. That's a kernel bug that appeared in 4.6.
> There's a fix at https://lkml.org/lkml/2016/11/29/852 but I don't think
> it has been applied yet.
>
> The second bug is a conflict between dual-core compute unit sharing and
> L1i. I am not sure which one is actually buggy. Can you run
> "HWLOC_COMPONENTS=x86 lstopo foo.xml" and send the generated foo.xml? (this
> is our raw detection that works around the kernel detection).
>
> Trying a Linux kernel <= 4.5 may help in the meantime.
>
> thanks
> Brice
>
>
>
>
> Le 03/01/2017 05:29, Johannes Goller a écrit :
>
> As requested on https://www.open-mpi.org/projects/hwloc/doc/v1.10.1/
> a00028.php ("What should I do when hwloc reports 'operating system'
> warnings?"), I am reporting the warning/error I received as follows
>
> 
> 
> * hwloc 1.11.0 has encountered what looks like an error from the operating
> system.
> *
> * L1i (cpuset 0x0003) intersects with Core (P#0 cpuset 0x0081)
> without inclusion!
> * Error occurred in topology.c line 983
> *
> * The following FAQ entry in the hwloc documentation may help:
> *   What should I do when hwloc reports "operating system" warnings?
> * Otherwise please report this error message to the hwloc user's mailing
> list,
> * along with the output+tarball generated by the hwloc-gather-topology
> script.
> 
> 
>
> Please find the tarball attached.
>
>
>
> regards,
>
> Johannes Goller.
>
>
>
> ___
> hwloc-users mailing 
> listhwloc-us...@lists.open-mpi.orghttps://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
>
>
>
> ___
> hwloc-users mailing list
> hwloc-users@lists.open-mpi.org
> https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users
>



  









  






  

  

  


  

  

  
  

  

  


  

  

  
  

  

  


  

  

  
  

  

  


  

  

  

  


  

  
  
  

  
  

  

  

  
  

  
  
  
  
  
  

  
  

  

___
hwloc-users mailing list
hwloc-users@lists.open-mpi.org
https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users

Re: [hwloc-users] Reporting an operating system warning

2017-01-02 Thread Brice Goglin
Hello Johannes

I think there are two bugs here.

First one is that each "dual-core compute unit" is reported as a single
core with two hardware threads. That's a kernel bug that appeared in
4.6. There's a fix at https://lkml.org/lkml/2016/11/29/852 but I don't
think it has been applied yet.

The second bug is a conflict between dual-core compute unit sharing and
L1i. I am not sure which one is actually buggy. Can you run
"HWLOC_COMPONENTS=x86 lstopo foo.xml" and send the generated foo.xml?
(this is our raw detection that works around the kernel detection).

Trying a Linux kernel <= 4.5 may help in the meantime.

thanks
Brice



Le 03/01/2017 05:29, Johannes Goller a écrit :
> As requested on
> https://www.open-mpi.org/projects/hwloc/doc/v1.10.1/a00028.php ("What
> should I do when hwloc reports 'operating system' warnings?"), I am
> reporting the warning/error I received as follows
>
> 
> * hwloc 1.11.0 has encountered what looks like an error from the
> operating system.
> *
> * L1i (cpuset 0x0003) intersects with Core (P#0 cpuset 0x0081)
> without inclusion!
> * Error occurred in topology.c line 983
> *
> * The following FAQ entry in the hwloc documentation may help:
> *   What should I do when hwloc reports "operating system" warnings?
> * Otherwise please report this error message to the hwloc user's
> mailing list,
> * along with the output+tarball generated by the hwloc-gather-topology
> script.
> 
>
> Please find the tarball attached.
>
>
>
> regards,
>
> Johannes Goller.
>
>
>
> ___
> hwloc-users mailing list
> hwloc-users@lists.open-mpi.org
> https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users

___
hwloc-users mailing list
hwloc-users@lists.open-mpi.org
https://rfd.newmexicoconsortium.org/mailman/listinfo/hwloc-users